RFR: Trim the TLAB sizes to avoid wasteful retirement under TLAB races

[Aleksey Shipilev]

On 09/21/2017 12:15 PM, Roman Kennke wrote:
> Am 21.09.2017 um 09:31 schrieb Aleksey Shipilev:
>> On 09/20/2017 09:17 PM, Aleksey Shipilev wrote:
>>> http://cr.openjdk.java.net/~shade/shenandoah/tlab-race-trim/webrev.01/
>>>
>>> This implements TLAB trimming to avoid wasteful retirement. The rationale is given in the comment
>>> right in the code. It does not seem to make allocation significantly more costly. To make it less
>>> overheady, changed the counter update code to be called on new region issuance, not on TLAB
>>> issuance.
>> Added more discussion:
>>    http://cr.openjdk.java.net/~shade/shenandoah/tlab-race-trim/webrev.02/
>>
>> -Aleksey
>>
> The number 8 is hardwired. Does this make sense? Should we allow this to be configured, if only for
> our own experimentation?

Well, it could be experimental -- which requires winding up argument checking logic and tests -- but
I would really like to move on already :) We can turn it experimental if there is interest later,
but I don't see any interest right now.

-Aleksey