[lilliput-jdk17u:lilliput] RFR: 8310156: [Lilliput/JDK17] Specialize full-GC loops [v4]

[Aleksey Shipilev]

On Tue, 20 Jun 2023 10:35:58 GMT, Roman Kennke <rkennke at openjdk.org> wrote:

>> We already ported most of [JDK-8305896](https://bugs.openjdk.org/browse/JDK-8305896) to Lilliput/JDK17. What's missing is the specialization of the full-GC loops so that performance on the legacy path is not impacted and performance on the alt-GC-forwarding path is minimized.
>> 
>> This corresponds to: https://github.com/openjdk/jdk/pull/13582/commits/6bbd29525b2864be37d96750ddc78f73d8eb0b65
>> 
>> However, the changes differ significantly:
>> - G1 serial compaction is different
>> - There is no humongous compaction in JDK17 G1
>> - Serial GC has seen some refactors of the relevant loops between JDK17 and JDK22
>> 
>> I see failing tests with Shenandoah, but they are pre-existing and should be investigated separately.
>> 
>> Testing:
>>  - [x] hotspot_gc
>>  - [x] hotspot_gc -UCOH
>>  - [x] tier1 -XX:+UseG1GC
>>  - [x] tier1 -XX:+UseSerialGC
>>  - [x] tier1 -XX:+UseShenandoahGC (not clean, but no regressions, either)
>>  - [x] tier1 -XX:+UseG1GC -UCOH
>>  - [x] tier1 -XX:+UseSerialGC -UCOH
>>  - [x] tier1 -XX:+UseShenandoahGC -UCOH (not clean, but no regressions, either)
>
> Roman Kennke has updated the pull request incrementally with one additional commit since the last revision:
> 
>   One more

Looks good for Lilliput JDK 17!

We would need to trawl through the difference against upstream, as all PRs land, and that might produce some adjustments to the code introduced in this PR.

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

Marked as reviewed by shade (Reviewer).

PR Review: https://git.openjdk.org/lilliput-jdk17u/pull/32#pullrequestreview-1487864645