temporary garbled window content when resizing due to unflushed VRAM content

[aufkrawall]

Not really a bug report, rather a cosmetic suggestion:
The RadeonSI driver doesn't flush previous window content which causes ugly artifacts for a brief moment when resizing.
As a workaround, mpv can be started with AMD_DEBUG=zerovram. Though it'd be more elegant if mpv flushed content of previous frames by itself in such cases.

Otherwise we should suggest to add mpv to the compatibility workaround list of Mesa to enable flushing by default there.

[ghost]

Yeah, I noticed this, e.g. on startup. It happens with EGL, but didn't happen with GLX.
Is it even possible for the application to fix this? VRAM doesn't exist.

[aufkrawall]

It'd be weird if not (but dunno), as afaik the driver behaves that way to strictly meet the OGL spec.
Though lots of applications show these artifacts with window operations.

As mpv often isn't started via XDG .desktop entry, one can also set this up in e.g. ~/.drirc

<driconf>
    <device driver="radeonsi">
        <application name="mpv" executable="mpv">
            <option name="radeonsi_zerovram" value="true" />
        </application>
</driconf>

[ghost]

The OGL spec. doesn't mention "vram" and doesn't deal with X11.
Isn't this more of a X11 thing, where window contents remain undefined if the application doesn't redraw quickly enough?
Do you have anything informed to say, or are you just guessing? (Not that I'm not guessing either.)

[aufkrawall]

The workaround flag mentioning VRAM might be misleading and thus also the title of my ticket (yeah, I'm guessing badly).
Perhaps it's the same backbuffer issue with eglSwapBuffers in mpv like with glXSwapBuffers in Blender:
https://bugs.freedesktop.org/show_bug.cgi?id=110448#c8

[ghost]

That seems very unlikely and I'm just going to say this is not the case, unless someone can disprove it with apitrace.

[aufkrawall]

@wm4 It's gone for me when starting mpv, but it still occurs when switching into fullscreen the first few times during playback. The radeonsi_zerovram flag "fixes" that as well.
Definitely driver specific, never seen it with Intel i965.

[ghost]

At least it removes the startup annoyance.
Any chance you could report this to Mesa? Maybe with API traces before and after this commit.
Edit: actually I guess the blender issue is exactly the same problem.

[aufkrawall]

Any chance you could report this to Mesa? Maybe with API traces before and after this commit.

Likely this weekend.

[aufkrawall]

@wm4 mesa report: https://gitlab.freedesktop.org/mesa/mesa/issues/2085

[ghost]

Thanks.

[aufkrawall]

Seems it is not going anywhere. Perhaps a simple MR to put mpv on that exception list would just do the trick for the time being?