[Bug 106175] amdgpu.dc=1 shows performance issues with Xorg compositors when moving windows

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Thu Jan 10 19:04:49 UTC 2019


https://bugs.freedesktop.org/show_bug.cgi?id=106175

--- Comment #81 from Nicholas Kazlauskas <nicholas.kazlauskas at amd.com> ---
(In reply to tempel.julian from comment #80)
> It seems like the issue is actually not 100% resolved (linux 5.0-rc1).
> The moving of windows is free of stutter now, but moving of windows can
> still negatively affect performance of other windows as long as fullscreen
> vsync is enabled (not necessarily via compositor, can also be done via
> TearFree without a compositor).
> Again, this is best seen on https://www.vsynctester.com/ . This seems to
> happen mostly when mouse clicks occur, but sometimes also apart from this.
> It can also happen when just moving the mouse cursor repeatedly on top of
> shell elements which trigger pop ups, like the system tray of KDE Plasma.
> As expected, setting amdgpu.dc=0 completely "fixes" the issue.

I'm not sure how much of this is actually amdgpu or Plasma.

I can reproduce what you're reporting - red lines and spikes in the graph on
vsynctester.com. This happens whenever I do something like open the dock or
volume widgets in the tray on Plasma with the compositor tearing prevention set
to automatic.

However, moving the cursor or moving windows doesn't really seem to affect this
and you can verify that in Plasma and other compositors.

As for the difference between dc=1 and dc=0, that might just be a difference in
behavior on the DRM level for atomic vs non-atomic drivers. Or a difference in
userspace if they make a distinction there. It affects more than just amdgpu at
least.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20190110/844ddfb8/attachment-0001.html>


More information about the dri-devel mailing list