decoupling interpreter buffers from the JIT's value processing
Paul Sandoz
paul.sandoz at oracle.com
Fri Jul 7 16:41:23 UTC 2017
Thanks, i added that to the wiki page, primarily so I don’t forget:
https://wiki.openjdk.java.net/display/valhalla/Minimal+Value+Types+and+Valhalla+value+types+command+line+options <https://wiki.openjdk.java.net/display/valhalla/Minimal+Value+Types+and+Valhalla+value+types+command+line+options>
Paul.
> On 7 Jul 2017, at 08:52, Vladimir Ivanov <vladimir.x.ivanov at oracle.com> wrote:
>
>> How do you determine if there are allocations or not? some specific hotspot flags?
>
> Yes, diagnostic EA output and eyeballing generated code (w/ -XX:-UseTLAB):
> -XX:+PrintEliminateAllocations -XX:+PrintEscapeAnalysis
>
> Best regards,
> Vladimir Ivanov
>
>>> On 7 Jul 2017, at 05:19, Vladimir Ivanov <vladimir.x.ivanov at oracle.com> wrote:
>>>
>>>> Here is a first shot:
>>>> http://cr.openjdk.java.net/~roland/8183137/webrev.00/
>>>> I don't see the allocations in your test anymore. Can you give it a try?
>>>
>>> Thanks, Roland! I confirm that there are no allocations left in the loop test I wrote (VectorTest::sumArrayL2). I'll experiment with other code shapes and let you know about the results.
>>>
>>> Best regards,
>>> Vladimir Ivanov
More information about the valhalla-dev
mailing list