RFR: 8209346: Refactor SATBMarkQueue filter configuration

[Kim Barrett]

> On Aug 13, 2018, at 6:40 AM, Roman Kennke <rkennke at redhat.com> wrote:
> 
> Oops no wait. This was intended to go to another thread.
> 
> Doesn't this require a change in g1_globals.hpp for the changed global?
> 
> While changing the global anyway, could we make it non-G1 too?

I didn’t change the global at all.  Making it no longer G1-specific (e.g. renaming it)
requires going through the CSR process.  It’s also not necessary for the purpose of
making the SATB code GC-agnostic and moving it to shared; the Shenandoah qset
initialization can pass in it’s preferred value.

I changed the internal handling of the configuration value, translating it from a
percentage to an index for direct comparison with the post-filtered buffer index,
rather than the old code that used that index to compute percent full and compared
that to the CLA option.  But the form of the configuration value (a percent full
threshold) hasn’t changed.

> Roman
> 
> 
> Am 13. August 2018 04:12:05 MESZ schrieb Kim Barrett <kim.barrett at oracle.com>:
>> Please review this change to how SATBMarkQueue determines whether a
>> post-filtered buffer should be enqueued or continue to be used.
>> 
>> […]
>> CR:
>> https://bugs.openjdk.java.net/browse/JDK-8209346
>> 
>> Webrev:
>> http://cr.openjdk.java.net/~kbarrett/8209346/open.00/
>> 
>> Testing:
>> Mach5 tier1-3, hs-tier4-5, in conjunction with JDK-8209345 and
>> JDK-8209347.
>> Locally tier1 and hotspot_gc without those other changes.