Artifacts in "The Saboteur" game.

[rox4d]

Wine d9d and dxvk are fine here, with nine on rx560 or hd7770 it looks like this screenshot.

Trace :
https://drive.google.com/file/d/1QM3g29m5Xs5R-cfQpIHNUrrXnGcvjke1/view?usp=sharing

[axeldavy]

Thank you, I can reproduce the issue

[rox4d]

a one more thing which was working with r600

[axeldavy]

Well, I could not figure out how to fix the issue.
Maybe it will be fixed one day by a random change in the driver.

[rox4d]

As a driver's developers don't care about nine, one can only hope.

[dungeon007]

On radeon driver flicker once then dissapear, meanwhile on amdgpu this looks like cooking... you can try on HD7770 with radeon to observe a difference :D Not sure of r600 as i dont have such hw, could be that is like this there too, still there just appear less with radeon...
Seems does not happen with AMD_DEBUG=nohyperz, kind of z-fighting from hz engine on a reflective material or maybe appear or get avoided by the luck on gl/vk...
Trace also needs dynamic_texture_workaround=true var on older than 21.1 mesa, to not show corruption at begining.

[rox4d]

Thank for digging it, i can't see any difference between radeon and amdgpu drm on SI (R7 450 currently) here, but with AMD_DEBUG=nohyperz it looks correctly in both cases (it also helps with rx560).

[rox4d]

Lets try to inform radeonsi developers https://gitlab.freedesktop.org/mesa/mesa/-/issues/5690

[mirh]

Blip blop
I'm not a bot, but it seems like they have a fix incoming for this issue in the last link.

[lorn10]

Great news. This radeonsi related bug here will be resolved with Mesa MR !22644. 😃

[rox4d]

Are you sure? i.e. your test it and it fixes issue?

[lorn10]

Yes, you can check the comment in Mesa bug 8716 here.

As said, we will see it when Mesa MR !22644 lands. 👍

And this here was in the end not a HyperZ problem (although disabling it helped). It was a problem of the radeonsi driver in conjunction with an old functionality called VS_WINDOW_SPACE_POSITION. That one seems to be missing in radeonsi and the emulation of that feature fixes the problem.

[rox4d]

Yes,

?
In pointed link i see a fixed branch https://gitlab.freedesktop.org/axeldavy/mesa/-/commits/unsupported_features_for_merge
(with env var force_features_emulation=true) but pointed MR are not about it.

[lorn10]

When you build Mesa with MR !22644 then you can verify oneself if this radeonsi bug here is resolved or not. 👍

[rox4d]

So why you does`t do it before writing?

[axeldavy]

It is not fixed by the MR.

[lorn10]

So why you does`t do it before writing?

I have currently no radeonsi hardware, only r600, crocus and old old r300 stuff.

It is not fixed by the MR.

😞 So the !22644 is only one piece. Whatever, sorry for the noise.