Not so fast.
Game runs fine so far, but every 5-10 seconds the game stutters immensely.
Log from Proton attached.
Mangohud didn't show any spikes, only drops in FPS.
steam-900040.log

Wow, I am such a dumbass.
AMDVLK was active and I thought I used Mesa all time.
Must have forgotten about it after some tinkering one day.

Not perfect, but with Mesa the stutters are way less, next to nothing.
Thank you very much @kisak-valve .

The game stutters even on Windows, as reported by some reviews. Even the the Steam Community hub has lots of discussions on performance problems.

Yes and no.
The code itself is not top notch and way too resource hungry,
but it can be played smoothly.
But with AMDVLK it was next to unplayable, with Mesa it runs
pretty smooth, like 99,9%.

@crone66 hi, there's an access violation happening somewhere that we are unable to reproduce locally. It's even hard to tell where does it happen. Address suggests it's some unix .so.

Can you reproduce the crash but before the process actually dies provide the contents of /proc/$PID/maps + the log from that session? That would help to narrow things down using mapped addresses.

If that proves too hard you can download Proton build from https://www.codeweavers.com/xfer/ahiler/log-so-addr/ (password: log-so-addr), extract the contained directory in $HOME/.steam/root/compatiblitytools.d/, restart Steam, go to game's properties -> compatibility and select log-so-addr. After that if you can capture log with PROTON_LOG=+module %command% that would be great.

@ivyl Hi, I used your proton build, the log can be found here: https://gist.github.com/crone66/18a70e451f0508db0c520df0e6e6a7ca

I hope it can help you to solve the issue.

In the meantime I used ge-proton7-1 as a workaround via protonup-qt. It would be great if previous proton version aren't removed from steam immediately after an update.

@crone66 thank you for the logs! Can you capture another one using 7.0-2 with PROTON_LOG=+vulkan? It looks like it crashes somewhere in the Vulkan driver a few calls down from winevulkan.dll that's used by DXVK's DirectX11 implementation.

Also I have two build for you to try, one which reverts the few winevulkan changes we had since 7.0-1 and one that downgrades DXVK to the version that was used in 7.0-1. Those are available in the same place: https://www.codeweavers.com/xfer/ahiler/log-so-addr/ (password: log-so-addr). Is any one of those helping with your crash?

Just making sure, you are launching the game in the game mode, not the desktop mode, right? I've tried to reproduce this by starting the game with 7.0-1b + a fresh prefix and then with 7.0-2 but I had no luck.

As of 7.0-1, you can find it in my personal archive https://f002.backblazeb2.com/file/ivyl-proton-archive/index.html. Sadly currently there's no easy way of keeping it available in the selection UI but I'm considering keeping it on a beta branch of 7.0.

@ivyl I was not 100% if I created the log in game mode, therefore I created a new one with your log-so-addr build with +module: https://gist.github.com/crone66/02875c0b464c8961dc15eff2a45f783d

I also created the requested log with the default proton7.0-2 build with +vulkan: https://gist.github.com/crone66/ebdf6e6f88805f9c1f15f7afe79aef3c

Sadly I wasn't able to create the log for the other two proton builds because the download speed is just around 6KB/s for some reason. I will try again tomorrow.

@crone66 hi, that was helpful. Thanks for your patience! We really have no other way debugging this than asking you for help as it cannot be reproduced locally.

Looks like it crashes during vkCreateGraphicsPipelines():

6843.061:0138:0170:trace:vulkan:wine_vkCreateGraphicsPipelines 0x7effe7e0, 0x7ed83a60, 1, 0x64df1b0, (nil), 0x64dec88
6843.061:0138:0170:trace:seh:dispatch_exception code=c0000005 flags=0 addr=00007F21DD3F1786 ip=00007F21DD3F1786 tid=0170
6843.061:0138:0170:warn:seh:dispatch_exception EXCEPTION_ACCESS_VIOLATION exception (code=c0000005) raised

I've added another build to the same place with a revert of https://github.com/ValveSoftware/wine/commit/1cd7b16f12e5af122623fc3af2de124f4f63742e which touched the offending function, but it still feels like a long shot. The change should be fairly safe and it seems to not be causing problems elsewhere...

@ivyl no problem I'm glad that I can help you to solve the problem. I tested your builds the results are as follows.

Revert-direct-call-functions: works
Revert-dxvk-bump: crashs
Revert-winevulkan-changes: works

For both working builds my old savegames from my proton 7.0-1/ge-proton7.0-1 sessions are missing for some reason but I was able to create new savegames.

The game received an update today which doesn't seem to have any effects regarding the crash behavior I tested your builds before and after the update.

@crone66 ok, so now we know which exact commit caused the problem for you which is great news. Bad news is that it's widely used, doesn't cause any pains elsewhere and it's kinda important because some drivers don't like having limited stack space they would have without it and those would just crash.

The crash happens somewhere withing the Vulkan driver stack (one of the layers or mesa) and debugging that remotely via back-and-forth would be hard, but there's a few more things we can try out :-)

You can try moving $HOME/.steam/root/steamapps/compatdata/900040 so you have saves backed up somewhere else and try to start the game with 7.0-2. I doubt this will affect anything but having a fresh prefix is a good first step.

Then we can look into shader caches, as the call in questions may be affected by them. Can you move $HOME/.steam/root/steamapps/shadercache/900040 somewhere and try with 7.0-2?

The last thing to try out would be moving $HOME/.cache/mesa_shader_cache.

If any of those steps helps it would be great if you could pack the directory that caused the problems (the copied backup) and share it with me in hope that I be able to repro locally.

@ivyl deleting the shader cache did the trick after that the game didn't want to start at all but rebooting the steam deck solved that. The first exit also didn't work and I had force quit the game but every other try so far worked flawlessly.

The backed up shadercache can be found here:
https://1drv.ms/u/s!AkEMgSEYjPAAt3syjj9dpHQal7he?e=Et5ASL

I hope that helps you to figure out what was wrong.
Thank you for all your work on trying to solve the issue.

@crone66 Would be to able to confirm that it doesn't crash with the following steps?

1. remove existing steamapps/shadercache/900040 dir
2. unzip your backup archive
3. remove steamapps/shadercache/900040/DXVK_state_cache
4. launch the game

Thanks!

@hakzsam I can confirm that deleting the directory fixed the problem for my backup.

@crone66 Thanks for confirming. We actually don't know how this happened, your DXVK cache seems completely corrupted... Is ELEX II the only game that is affected on your side?

@hakzsam I tested multiple games that I've already started on the steamdeck before the proton 7.0-2 update. All of them run without issues, therefore only Elex 2 seems to be affected.

@crone66 Good to know. Are you able to reproduce this issue with Elex 2 after removing the DXVK cache ie. does it always happen?

Additional tests:
I just tried it with the new Proton Next 7.0-5 and the issue persists, it also occurs with the GE-Proton7-41 version.

No issues under the DX11 version of the game on linux, aside from the horrible fps that also plagues the windows version of the game.

Just tested with Proton Next 7.0-6, Proton Experimental, Proton Hotfix and GEproton7-47 ... issue persists and seems to be getting worse.
System changes below:
Driver/LLVM version: PPA 22.2.5-kisak-mesa
Linux Mint 21.1, Kernel version: 5.15.0-58

Just tested with proton 8.0-2 and experimental ... issue persists.

System changes below:
Driver/LLVM version: stock mesa 22.2.5
Linux Mint 21.1, Kernel version: 5.15.0-71

@Xenophobe-01 can you try with updated drivers? Mesa 22.2.5 is pretty old and not getting any bug fixes anymore.

@mbriar ... I just tested with ppa:kisak/turtle 22.3.7 and with ppa:kisak/kisak-mesa 23.0.3 ... same result with both of the fireball storm in DX12 mode but great fps, no firestorm in DX11 mode with trash fps.

edit: I also deleted the shader cache in /steamapps/shadercache/900040 to let it regenerate a clean cache, as well as the proton prefix so it was freshly generated too.

Same issue here with the "fireballs" using mesa 23.0.3, though it's happened on earlier versions too. I've also tried different proton versions like 7.0, 8 and experimental to no avail.
Info:
OpenGL vendor string: AMD
OpenGL renderer string: AMD Radeon RX 5700 XT (navi10, LLVM 15.0.7, DRM 3.52, 6.3.0-x64v1-xanmod1)
OpenGL core profile version string: 4.6 (Core Profile) Mesa 23.0.3 - kisak-mesa PPA
OpenGL core profile shading language version string: 4.60
OpenGL core profile context flags: (none)
OpenGL core profile profile mask: core profile

UPDATE: Also happens on mesa 23.2.0 (development version) and 22.2.5.

Are any special settings needed for this to happen? I've been unable to reproduce on my rdna3 with either mesa 23.0.2 or git from yesterday. Tested Proton 7, 8 and BE while poking at a few ingame settings.

Installing on Steam Deck to check it out there too.

@Blisto91 no special settings I'm aware of. fullscreen only,1080p, vsync on, 60fps, everything in-game on high setting.

Are any special settings needed for this to happen? I've been unable to reproduce on my rdna3 with either mesa 23.0.2 or git from yesterday. Tested Proton 7, 8 and BE while poking at a few ingame settings.

Installing on Steam Deck to check it out there too.

Thanks for checking! I have everything on the highest settings with resolution 1440p@144hz (also tried 1080p@60hz) and fullscreen only. I've also tried with the lowest settings, and the issue still remains unfortunately.
For me it seems a bit random, sometimes it happens almost immediately upon loading the save, and other times it takes a couple of minutes. I've found that changing the render range option makes the fireballs go away, but only temporarily.

Strange. I've tried a bit on Deck now too. Played the cutscene a bunch of times and ran around in the camp for a while but still haven't triggered it.

@Blisto91 Linux Mint is still using X11, I'm not sure if the deck is on Wayland or not ... is your pc distro on X11 or Wayland?

Also, the fireball storm glitch only seems to trigger when entering into conversation with an npc, Caja or even Adam Dawkins has triggered it.

I think it's pretty much impossible that x11 or Wayland affects a bug such as this.

11 May 2023, 18:54 by @.***:

@Blisto91 https://github.com/Blisto91> Linux Mint is still using X11, I'm not sure if the deck is on Wayland or not ... is your pc distro on X11 or Wayland?

Also, the fireball storm glitch only seems to trigger when entering into conversation with an npc, Caja or even Adam Dawkins has triggered it.

—
Reply to this email directly, > view it on GitHub https://github.com/ValveSoftware/Proton/issues/5670#issuecomment-1544340739> , or > unsubscribe https://github.com/notifications/unsubscribe-auth/A3G6J6LGEJMMMI4Z4PJ3XPLXFUKVBANCNFSM5QSCWZJQ> .
You are receiving this because you were mentioned.> Message ID: > <ValveSoftware/Proton/issues/5670/1544340739> @> github> .> com>

I can quickly boot in to X11, but i'm also 99.99% sure that won't do anything as mbriar pointed out.

The only thing i see in common here, besides it being AMD Rdna 1 & 2 GPU's so far, is that you both are using Kisak ppa drivers. I'm on a Arch based distro myself, but i'm not really thinking it should affect it.

I can quickly boot in to X11, but i'm also 99.99% sure that won't do anything as mbriar pointed out.

The only thing i see in common here, besides it being AMD Rdna 1 & 2 GPU's so far, is that you both are using Kisak ppa drivers. I'm on a Arch based distro myself, but i'm not really thinking it should affect it.

also does it with the stock mesa driver 22.2.5 which is the latest on mint. HomerSP said it happens with mesa 23.2.0 (development version) too.

Ye realized afterwards

I tried this now as well and can't reproduce it either, played for an hour and ran the cutscene from the video 6 times. 6700XT, both a week old mesa-git and even amdgpu-pro work fine (and on x11).

I just tried it with a Manjaro KDE live USB (fully updated, mesa 23.0.2 ) and the exact same thing happened ... no idea what's causing it.