RFR: 8245022: ParallelGC abuses StarTask to also include partial objarray scan tasks

Kim Barrett kim.barrett at oracle.com
Wed May 20 12:50:09 UTC 2020


> On May 20, 2020, at 4:45 AM, Thomas Schatzl <thomas.schatzl at oracle.com> wrote:
> 
> Hi,
> 
> On 20.05.20 10:11, stefan.johansson at oracle.com wrote:
>> Hi Kim,
>> On 2020-05-20 00:41, Kim Barrett wrote:
>>> Please review this change to ParallelGC to also use the recently added
>>> ScannerTask (JDK-8244684), eliminating some casts between oop and oop*
>>> and use of some misaligned oop* values.
>>> 
>>> Also renamed G1's ScannerTasksQueue[Set] to G1ScannerTasksQueue[Set].
>>> I should have used those names in JDK-8244684.  This is consistent
>>> with the new PSScannerTasksQueue[Set] now being used by ParallelGC.
>>> The types used by G1 and Parallel are currently the same, but don't
>>> need to be.
>>> 
>>> Also made a small simplification, reordering the template parameters
>>> for PSPromotionManager::copy_and_push_safe_barrier.  By making the
>>> promote_immediately flag first and the reference type second we can
>>> explicitly specify the non-deduced flag and allow the reference type
>>> to be deduced, rather than having to specify both explicitly in calls.
>> Nice!
>>> 
>>> CR:
>>> https://bugs.openjdk.java.net/browse/JDK-8245022
>>> 
>>> Webrev:
>>> https://cr.openjdk.java.net/~kbarrett/8245022/open.00/
>>> 
>> Looks good, one very small nit is the extra spaces left here:
>> share/gc/parallel/psPromotionManager.hpp
>> ---
>>  114  protected:
>>  115   static PSScannerTasksQueueSet* stack_array_depth()   { return _stack_array_depth; }
>> They were pre-existing, but since we are touching that row, we can remove those spaces that seems to be left from some old alignment.
>> ---
> 
> agree. Looks good.
> 
> Thomas

Thanks.



More information about the hotspot-gc-dev mailing list