Metal renderer threading issues [macOS]

[Alexey Ushakov]

Not sure about this particular case but in general the change that we have in our fork allows to have a one CVD thread per display (not per MTLLayer as we have in OpenJDK mainstream). But even this optimization is not enough for my opinion. We are currently working on the change that performs updates without any display link thread (similar to what we have in OGL). The display link approach introduced in the Metal pipeline was kind of a workaround for the missing content artifacts due some flaws in synchronization. I think that after fixing all the synchronization issues we’ll be able to get rid of display link threads completely.

Best Regards,
Alexey

> On Oct 14, 2023, at 5:11 PM, Alan Snyder <javalists at cbfiddle.com> wrote:
> 
> 
>>> 
>>> On Oct 10, 2023, at 3:32 AM, Alexey Ushakov <alexey.ushakov at jetbrains.com> wrote:
>>> 
>>> We actually changed this logic in JetBrains Runtime and invoke the CVD api only from one thread but still I see multiple display link threads on M2 machines in some rare cases. So it maybe OS related issue.
>> 
> 
> Did the change to use a single thread make any difference?
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/client-libs-dev/attachments/20231017/9844b3f8/attachment.htm>