[master] RFR: 8304341: [Lilliput] Use fixed-size lock-stack [v2]

[Aleksey Shipilev]

On Fri, 17 Mar 2023 08:31:09 GMT, Roman Kennke <rkennke at openjdk.org> wrote:

>> Until now, we used to have a variable-sized lock-stack: when pushing an object to it and capacity is exceeded, it would re-allocate a new stack and use that. However, experiments show that the lock-stack very rarely exceeds 5 slots (I have not yet found a workload that does actually exceed it). It makes sense to make the lock-stack a fixed-size array: it makes addressing the lock-stack simpler and more efficient and it increases the likelyhood of the lock-stack being in CPU cache. If the lock-stack is ever exceeded, we would not do stack-locking at all, but instead inflate the monitor and use that.
>> 
>> This is already integrated in the related upstream PR: https://github.com/openjdk/jdk/pull/10907
>> 
>> Testing:
>>  - [x] tier1
>>  - [ ] tier2
>
> Roman Kennke has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Set condition flags correctly after fast-lock call on aarch64

I guess there is nothing actually to review here, and the comments should be added to https://github.com/openjdk/jdk/pull/10907/. I skimmed through the changes and there are no evident problems with it.

-------------

Marked as reviewed by shade (Committer).

PR Review: https://git.openjdk.org/lilliput/pull/81#pullrequestreview-1354390531