[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openjdk-2d-dev
Subject:    Re: RFR: 8304825: MacOS metal pipeline - window isn't painted if created during display sleep
From:       Alexey Ushakov <avu () openjdk ! org>
Date:       2023-03-30 8:55:18
Message-ID: 6co54WQ1FA_lD9KZhw9eWhOh_0Tty6gyXx2SYthOA28=.ab716fe0-6e1c-43d5-b5ee-4659fb2f3f38 () github ! com
[Download RAW message or body]

On Thu, 30 Mar 2023 06:53:46 GMT, Ajit Ghaisas <aghaisas@openjdk.org> wrote:

> > Use one display link thread per MTLContext. Adjust the refresh rate with the \
> > current display.
> 
> I tested the test program of this bug and SwingSet2 - with and without external \
>                 monitor attached.
> - This patch fixes the reported issue. Also, I did not see any regression in \
> SwingSet2 while moving the test window between external monitors. All regression \
>                 tests also ran successfully with this patch.
> - This patch has a negative impact on rendering performance.
> SwingMark tests show very poor numbers on x64 based macs. There is a performance \
> decrease on M1 macs as well. I think, this has to do with removal of a dedicated \
> `blitCommandQueue`.  
> We need to address the performance degradation that this patch causes or look for a \
> simpler solution.

@aghaisas could you return back the blit command queue in my patch and see if it \
resolves the performance issue with SwingMark? I'm not completely sure that it's the \
main reason for the performance degradation. Another possible reason is the increased \
amount of start/stop displaylink  calls because of changes in the logic that manages \
them.

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

PR Comment: https://git.openjdk.org/jdk/pull/13230#issuecomment-1489936273


[prev in list] [next in list] [prev in thread] [next in thread]