RFR: Remove CSetThreshold handling from heuristics

[Aleksey Shipilev]

On 04/05/2018 03:57 PM, Roman Kennke wrote:
> Am 05.04.2018 um 15:50 schrieb Aleksey Shipilev:
>> Ping?
>>
>> On 04/04/2018 07:57 PM, Aleksey Shipilev wrote:
>>> http://cr.openjdk.java.net/~shade/shenandoah/purge-cset-threshold/webrev.01
>>>
>>> We did this thing before separate UR phase was introduced. The role for this threshold was to avoid
>>> double-cycle when heuristics thinks the free space hoarded in cset is not available. However, it is
>>> not fully reliable: the fact that we "appear" to have more free space than we actually have risks to
>>> run into alloc-failure during late mark. Also, since separate UR phase is enabled, we don't even
>>> care about double-cycle issue anymore: CM-with-UR cycles _are_ back-to-back.
>>>
>>> Removing this simplifies the code.
>>>
>>> Testing: hotspot_gc_shenandoah, SPECjbb (where the original double-cycle was observed)
>>>
>>> Thanks,
>>> -Aleksey
>>>
>>
> 
> Have you checked that the double-cycle issue does not occur? E.g. in
> visualizer?

Yes, I did ran SPECjbb under Visualizer, and saw no double-cycles. Well, all cycles were
CM-without-UR, and CM-with-UR happened only on small heaps where cycles are back-to-back anyway.

-Aleksey