CME
CME
There are two issues here, one is depth bias on nvidia and one is some fixed function fun, i have an nsight trace locally here where it's reproducible.
Yeah i just got myself confused back then, there is also a depth bias problem on nvidia, but on AMD it's only some fixed function stuff.
https://imgsli.com/MTA1MTE2/0/2 dxvk vs wined3d vs dxvk-nv nsight: https://drive.google.com/file/d/1YlJxu5pcCYZ0HTLsuMqeUazXpW7oYj0A/view?usp=sharing
pic1: The game is using a cube map with a different texture on each face to render the terrain and it relies on the cube map to be seamful, similar...
Yeah sorry i forgot to comment here, the shadow issue happens due to the d24s8 -> d32s8 fixup that is necessary on AMD. I can also repro it on NV...
The issue is due to depth bias, changing the RValue from 1
can you please try this patch? ```diff --git a/src/d3d9/d3d9_util.h b/src/d3d9/d3d9_util.h index 50be2daf..58e923b3 100644 --- a/src/d3d9/d3d9_util.h +++ b/src/d3d9/d3d9_util.h @@ -182,10 +182,9 @@ namespace dxvk { case VK_FORMAT_D16_UNORM: return float(1
This is also due to depth bias, use my old patch as a workaround until there is a proper fix: https://github.com/doitsujin/dxvk/issues/1329#issuecomment-574419167
I can repro it on my 6900xt with both radv and amdvlk but not on my 1060, radv_disable_aniso_single_level=true doesn't help.
fwiw metald3d logs a couple of error messages like this, taking a quick look at the renderdoc it seems like this is the issue: ``` Failed to create pipeline state:...