Artefacts in Vulkan game (RX 6800XT)

[Riptils]

• Sway Version:
  sway version 1.10-dev-f293418d (Dec 11 2024, branch 'master')

• Debug Log:
  swaydebug-verbose.txt
  I had some photos and videos : https://photos.app.goo.gl/G5e7azBdymTK7g7X7

• Configuration File:

  • My personal config is on my gitlab (https://gitlab.com/Riptils/dotfiles)
    I'm using mesa driver for my rx 6800xt

• Description:
  When I want to use my game (beamng) with Vulkan (natively or with wine dxvk) it creates artifacts regardless of the graphics level. I have the same thing with assetto corsa but I don't have the time to reproduce it.
  PS: I find that the programs are quite slow compared to the windows installation I had done beforehand, maybe there's a problem with my gpu?

[emersion]

Can't find anything relevant in the logs.

[auroraanna]

does this occur on the same games but on other compositors like KDE Plasma's?

[Riptils]

I just tried on kde with vulkan etc. but I don't have the artifacts. On the other hand, I've noticed that on beamng I've got some sort of micro flash or the screen goes black with the game when the refresh rate drops too low.

[fenveireth]

I have this card, and similar artefacts on several other games (consistent repro with Deep Rock Galactic + proton vkd3d)

They disappear if the game is started with RADV_DEBUG=nodisplaydcc, or if sway is started with either WLR_RENDER_NO_EXPLICIT_SYNC=1 or WLR_SCENE_DISABLE_DIRECT_SCANOUT=1

[Nefsen402]

They disappear if the game is started with RADV_DEBUG=nodisplaydcc, or if sway is started with either WLR_RENDER_NO_EXPLICIT_SYNC=1 or WLR_SCENE_DISABLE_DIRECT_SCANOUT=1

This is almost certainty a RADV/amdgpu bug. KDE doesn't get affected the same way because it probably doesn't enter direct scanout in as many cases as sway does.

[rpigott]

What is your wlroots version? I have an RX 6750XT with similar symptoms on scanout, including disappearing with RADV_DEBUG=nodcc, so I initially suspected a radv bug, but after discussion in #radeon I ended up bisecting the artifacts to https://gitlab.freedesktop.org/wlroots/wlroots/-/commit/9e71c88467fc41d6330bbd4b99f82f3df49ed4ac.

Haven't had the time to further investigate.

[Nefsen402]

Maybe explicit sync has a hand in this? Maybe fences aren't being passed properly and the display is trying to scan out an unfinished buffer? But that commit should have nothing to do with that.