Linux Bloodbourne Missing Geometery

[mxli2025]

above is what it looks like on first load

below is after running around abit quitting to mainmenu and reloading

Loading bug maybe ? as it can display it

LCD Deck
Mint 22
no mods

[gurrgur]

I have the same issue on Arch Linux / RX6700XT / RADV. I can also confirm that on second load most of the time, the geometry data loads properly.

I did two traces in Renderdoc (first with missing meshes and second with no missing meshes) and noticed that some vs input vertex buffers and index buffers are filled with zeros only on the trace with missing meshes.

My (uneducated) theory on this is that when the game loads in new assets from disk, due to some unknown reason (perhaps sync issue like missing fence / barrier?) the meshes are only partly loaded into GPU memory. On the second load, the data is already cashed in system memory so that the upload to GPU is quick enough to complete in time. Rdoc also shows, that a compute shader acesses these buffer objects before they are passed to the VS stage (which is part of a depth pass IIRC).

Also, I'd like to add that when moving from one area to another, the geometry disappears again. For example, load into central yarnham and climb the ladder. After some steps on the ladder, the screen will turn grey for a short amount of time and after that the meshes will be gone.

[gurrgur]

Oh and this is a RADV only issue. amdvlk has no missing meshes (though it has waaaaaaay worse perf).

And finally, sometimes it looks like some of the zero filled vertex buffers are instead filled with garbage data leading to flicker. Seems kind of random and apparently depends on timing.

[LNDF]

Same issue. Fedora 40 using RADV.

It might be nice to see what the validation layers print out.

[mxli2025]

issue #976 , seems to be similar. Windows on a nVidia card.

[grassponderer]

Same here. Arch, 6900XT using vulkan-radeon, tested on linux, linux-zen, and linux-cachyos, no difference.

When I tried amdvlk, got to the first loading screen before the area loads in, then shadps4 crashes. I don't know if it's helpful, but I noticed that I get the following error spammed repeatedly:
[Render.Vulkan] <Error> tile_manager.cpp:TryDetile:391: Unsupported tiled image: R32Sfloat (Depth_MacroTiled)
Don't have enough background as to whether or not this might point to the cause of the issue though.

[Hermiten]

Can you test with the last update ?
This commit #1146 can maybe fix somethings

[R1chterScale]

Can you test with the last update ? This commit #1146 can maybe fix somethings

No dice

[HermitsA]

As a compromise, using the amdgpu-pro drivers fix this. You can have both the mesa and proprietary drivers on the same machine, defaulting to mesa normally, and enabled by using "progl [command]". You can also try the amdvlk drivers, altough performance was horrible last time i tested.

Tested on the 3 different vulkan implementations (RADV, AMDVLK, VULKAN-AMDGPU-PRO) on a 6700xt with archlinux

[Mast3rwaf1z]

I have the same issues, amdgpu-pro works, but the game stutters, i just tried running through nightmare of mensis with radv, no textures but the game is much more smooth with RADV.

[grassponderer]

Just a quick update, now it seems to work for me using this
https://github.com/ngoquang2708/shadPS4/actions/runs/11869943285
and AMDVLK. Seems to be related to #1193 since when using amdvlk these error don't get spammed.

[Malumen]

Just a quick update, now it seems to work for me using this https://github.com/ngoquang2708/shadPS4/actions/runs/11869943285 and AMDVLK. Seems to be related to #1193 since when using amdvlk these error don't get spammed.

how do we enable AMDvlk?

[grassponderer]

how do we enable AMDvlk?

Assuming you use arch, the safest way without messing up everything else that uses vulkan-radeon is probably https://github.com/Frogging-Family/amdvlk-opt

[Channeira]

Did someone report the RADV specific issues to mesa already?

[dunestorm333]

Can confirm, this is still a problem on the latest shadPS4 build :(

Running Fedora 41 (Linux 6.11.10-300.fc41.x86_64)
Mesa 24.2.7

AMD 6800XT

[raphaelthegreat]

Issues on the main repo should be about main branch only, please test that

[Missake212]

Does this still happen on latest main ?

[R1chterScale]

Does this still happen on latest main ?

No, seems good now, might be interesting to bisect the change to report to the Mesa devs

[rharish101]

Can confirm that it's fixed for me on RADV (on a 7900 XTX), although I only tested this in Iosefka's clinic.