search

daniel-schuermann / mesa

Mesa 3D graphics library (mirror; no pull requests here please)
http://mesa3d.org
135 stars 3 forks source link

Trine 4 - corruption of certain effects #159

Closed aufkrawall closed 4 years ago

aufkrawall commented 4 years ago

LLVM: Screenshot_20191022_094143

ACO: Screenshot_20191022_094244

RenderDoc capture: https://drive.google.com/open?id=1QbPxL2vU7XEOYgfUjj3vZHhMlOjf5Bkz

latest mesa-git mainline master RX 570

daniel-schuermann commented 4 years ago

@aufkrawall Unfortunately, I cannot reproduce the issue with the renderdoc capture.

aufkrawall commented 4 years ago

@daniel-schuermann I see what you mean, the weird "lensflares" aren't showing here either in RD. However, the partially missing shadows/SSAO in the trees in the background is visible (it's not random, but also part of the corruption).

Edit: The SSAO in the grass in the front is not right either.

daniel-schuermann commented 4 years ago

Can you give me a picture from the renderdoc opened with aco pointing out what exactly looks different? I'm not seeing it :/

aufkrawall commented 4 years ago

In the grass in the front, the SSAO (at least it's my best guess that it's the SSAO) has weird patterns which are not there on the LLVM screenshot: Screenshot_20191022_194139 Those patterns also look like partially "detached" in motion when playing the actual game.

Also, you can notice that especially the trees on the right side of the background look too bright with ACO. Probably there is something wrong with the SSAO as well.

daniel-schuermann commented 4 years ago

But when opening the capture with LLVM, it looks exactly the same which makes it extremely hard to find.

aufkrawall commented 4 years ago

That's a bummer. :( I suppose it wouldn't make sense to create more captures, the issue basically is always there and always looks the same.

aufkrawall commented 4 years ago

So the issue is that the corruption is "hardcoded" into the capture with ACO? When I let radv-llvm render the game and capture it, there is no corruption in the capture (despite of using ACO in qrenderdoc).

aufkrawall commented 4 years ago

Good news: I can't reproduce this anymore with mesa 20.1.4 and RX 480. Though amdvlk-pro is still 10% faster.