RFR: 8256265: G1: Improve parallelism in regions that failed evacuation [v6]

[Albert Mingkun Yang]

On Wed, 14 Sep 2022 10:55:50 GMT, Thomas Schatzl <tschatzl at openjdk.org> wrote:

>> Hi all,
>> 
>>   can I have reviews on this change that makes the evacuation failure remove forwards phase split the work across threads within regions?
>> 
>> This work is a continuation of PR#7047 from @Hamlin-Li and latest improvements from @albertnetymk in that thread. This is reflected in the first commit, I added a fair amount of changes to fix bugs and streamline the code.  
>> 
>> Testing: tier1-8 with G1EvacuationFailureALot enabled, tier1-5 as is
>
> Thomas Schatzl has updated the pull request incrementally with one additional commit since the last revision:
> 
>   sjohanss, ayang review 4

Just some minor comments.

src/hotspot/share/gc/g1/g1EvacFailure.cpp line 209:

> 207: }
> 208: 
> 209: void G1RemoveSelfForwardsTask::initialize() {

Any particular reason why this is not part of the constructor?

src/hotspot/share/gc/g1/g1YoungGCPostEvacuateTasks.cpp line 117:

> 115: 
> 116:     double chunks_per_thread = (double)G1CollectedHeap::get_chunks_per_region() / G1RestoreRetainedRegionChunksPerWorker;
> 117:     return chunks_per_thread * _evac_failure_regions->num_regions_failed_evacuation();

The calculation is correct, but the var name is wrong.

`chunks/region  /   chunks/worker ==  workers/region` then `workers/region  *  region == workers`. This method returns the optimal number of workers.

(PS: I still think doing multiplication is easier to reason, but that can be subjective.)

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

Marked as reviewed by ayang (Reviewer).

PR: https://git.openjdk.org/jdk/pull/9980