Far Cry 3 - waterfalls show corruption with MSAA
Can you try dxgi.nvapiHack = True?
We actually set it to false by default for this game because it requires a clear value of 0.5 to map to different integer values depending on the GPU vendor, which othersie can cause lighting issues with these rectangular shapes. This used to work in the past, but it's possible that Nvidia's Vulkan driver behaviour changed or that somethig else isn't quite consistent with what the game expects.
If that doesn't work, well, not sure what to do. It's really a game issue since they hard-coded weird behaviour into their game that may not even hold true on future hardware.
That apitrace doesn't work.
dxgi.nvapiHack = True doesn't help with the waterfalls and additionally introduces semi-transparent rectangles floating around in some parts of the scene.
No idea why the apitrace doesn't work. I made sure all overlays were disabled and the game seemed to have terminated normally after alt + f4.
Probably not fixable then.
Looks fine here with nvidia 515.49.14 drivers and proton-experimental (not sure what version of DXVK experimental exactly uses, but should be recent enough i think)
MSAA enabled? Looks like the issue only occurs with it enabled.
Might be a different issue than the clear value thing if it only happens with MSAA.
Can you post validation layer output?
I'll see to it.
Game just freezes for me when run in Proton. It works on Windows with Vulkan SDK installed and VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation, but no idea where I'd find any output of it.
The output is in stdout. If you set PROTON_LOG=1, Proton/Steam will redirect that into a file called steam-APPID_OF_YOUR_GAME.txt in your home directory. That should have the validation messages.
Confirming the MSAA breaks waterfalls on linux also.
steam-220240.log
i think i got the validation layers enabled, let me know if its wrong.
Used command in steam VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation PROTON_LOG=1 %command%
The output is in stdout. If you set PROTON_LOG=1, Proton/Steam will redirect that into a file called steam-APPID_OF_YOUR_GAME.txt in your home directory. That should have the validation messages.
Thanks! I got the game running by renaming farcry3_d3d11.exe to FC3Updater.exe, as Ubisoft Connect starts the latter first and something with that goes wrong with the regular Ubisoft Connect game version manually added to be run via Proton in Steam.
xpander69 beat me to it, but I'll try regardless now. I'm sure the relevant information regarding VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation and stdout can be found when following the links to Khronos docs, but I suppose it could be made a bit more obvious in the DXVK wiki. As unlike pressing a button in RenderDoc, it's not really self-explanatory.
xpander69 beat me to it, but I'll try regardless now.
His logs don't contain any validation messages. He likely didn't have 32 bit validation layers installed.
but I suppose it could be made a bit more obvious in the DXVK wiki
Usually we prefer an apitrace which we can just run it with validation layers ourselves. But we already tried that here and it didnt work so there's that.
No i didnt beat you do it. Game doesnt launch for me.. I didnt have 32bit validation layers. So my log is useless
Indeed lib32-vulkan-validation-layers also isn't included in Arch vulkan-devel meta package, how treacherous.
But with it installed, the game crashes on launch with VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation. Oh my...
Yeah tried to replace executables also but still crashes right after splash screen coming up with the good old
Unhandled exception: page fault on read access to 0x00000158 in 32-bit code (0xf26cf7ba).
I think its doomed :(
I guess I can try diverting stdout to a log file on Windows, I had the game running with VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation there (at least that's my conclusion after it ran with only ~15fps). If not something else breaks, that is... :hurtrealbad:
I recorded a video about the effect/bug, just in case maybe some other games have had same issue and it rings a bell to someone:
https://www.youtube.com/watch?v=PknyM_nOof0
Looks like the linked apitrace version is outdated? I can replay a capture with version 11.1 (DXVK wiki links 10.0). Uploading...
Thanks. I updated the links.
VUID-vkCmdDrawIndexed-None-02699(ERROR / SPEC): msgNum: -1539028524 - Validation Error: [ VUID-vkCmdDrawIndexed-None-02699 ] Object 0: handle = 0x20ffe40000018bb6, type = VK_OBJECT_TYPE_DESCRIPTOR_SET; | MessageID = 0xa44449d4 | Descriptor set VkDescriptorSet 0x20ffe40000018bb6[] encountered the following validation error at vkCmdDrawIndexed time: Descriptor in binding #9 index 0 requires bound image to have VK_SAMPLE_COUNT_1_BIT but got VK_SAMPLE_COUNT_4_BIT. The Vulkan spec states: Descriptors in each bound descriptor set, specified via vkCmdBindDescriptorSets, must be valid if they are statically used by the VkPipeline bound to the pipeline bind point used by this command (https://www.khronos.org/registry/vulkan/specs/1.3-extensions/html/vkspec.html#VUID-vkCmdDrawIndexed-None-02699)
So yeah, it's that good old MSAA bug again. Unfortunately the only way to fix this is to disable MSAA, this cannot be worked around in DXVK in any reasonable way.
Renders fine on AMD fwiw.
Well, was worth a try. :) Thanks anyway!
Out of pure curiosity, does it work if you use the D3D9 renderer?
With D3D9, the MSAA and AO options aren't available/get grayed out.
On the bright side, the trace uncovered an actual bug, see 5117210c93d7736f36719cf62d1780c4ae62ebfa. This won't affect these rendering issues though.
Proton versions
proton experimentalx1 2022-08
Launch options
PROTON_LOG=1,x2 2022-08PROTON_LOG=1x1 2022-08
D3D11:
DXVK (missing parts in rectangular shape that change with camera perspective):
RTX 3060, Windows 11, driver 516.94, DXVK 17959640c3bfe96a96a600ab14b96658f6acf4d6 , latest game version via Ubisoft Connect, maximum visual details
If it's a regression, it's not a too recent one, as this occurred already months ago with older DXVK and drivers on Linux.
Logs:
farcry3_d3d11_d3d11.log
farcry3_d3d11_dxgi.log
apitrace:
https://drive.google.com/file/d/1VbwMO7_36qzpXXZxRr_ZQDk6QOBq7b1A/view?usp=sharing