RFR(S) 8220444: Shenandoah should use parallel version of WeakProcessor in root processor for weak roots

Roman Kennke rkennke at redhat.com
Mon Mar 11 17:59:38 UTC 2019


Ok good.

I remember I experimented with this stuff back in October or so, and at 
some point it probably touches the whole 'concurrent stuff processing' 
which we cannot do yet. But this is a different story.

Please ensure the bug has Affects: 12 if it also affects 12, so that we 
get backports right. (I've asked to get 11-shenandoah and 8-shenandoah 
version tags, as soon as we have those, we might also put 11-shenandoah 
in affects-field )

Reviewed.

Thanks!
Roman

> On 3/11/19 1:44 PM, Roman Kennke wrote:
>> Very nice!
>>
>> Does the weakprocessor only handle the jni weak oops? I seem to 
>> remember it covers more 'weak roots' too?
> 
> It covers more than jni weak oops. I intend to breakup "JNI Weak Roots" 
> in followup RFE, if Aleksey has not objection.
> 
> Thanks,
> 
> -Zhengyu
> 
>>
>> Roman
>>
>>> Shenandoah still uses serial version, should switch to parallel version.
>>>
>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8220444
>>> Webrev: http://cr.openjdk.java.net/~zgu/JDK-8220444/webrev.00/
>>>
>>> Test:
>>>    hotspot_gc_shenandoah (fastdebug and release)
>>>
>>>    Benchmark:
>>>    Compiler.compiler
>>>    Baseline: http://cr.openjdk.java.net/~zgu/JDK-8220444/serial.log
>>>    After Patch: http://cr.openjdk.java.net/~zgu/JDK-8220444/par.log
>>>
>>>
>>> Thanks,
>>>
>>> -Zhengyu


More information about the hotspot-gc-dev mailing list