daniel-schuermann / mesa

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

Monster Hunter World: Graphical corruption and performance hit not present with RADV_PERFTEST=llvm #92

Closed ghost closed 5 years ago

ghost commented 5 years ago

Describe the bug Monster Hunter World has severe graphical problems and performance decrease not present with RADV_PERFTEST=llvm

Screenshots ACO: aco

RADV_PERFTEST=llvm: llvm

System information:

Additional context Pretty sure I ran MHW fine with earlier versions of mesa-aco-git.

Johnnynator commented 5 years ago

Haven't checked the latest commit, but even with e.g https://github.com/daniel-schuermann/mesa/commit/a60c307606237ddc9bbc35174a1274506f7974da and some previous commit the game was looking and performing fine, until ~30-60 minutes in where the frame rate dropped sub 1fps and amdgpu (Can make a proper bug report once I have time for it)

daniel-schuermann commented 5 years ago

Thx for the report! Could you test with RADV_PERFTEST=llvmvs and see if the problem persists (although it looks like pixel shader) ? Either way, would you be able to provide a renderdoc capture of the issue?

ghost commented 5 years ago

It does still happen with RADV_PERFTEST=llvmvs

Johnnynator commented 5 years ago

Here is a renderdoc capture: https://void.johnnynator.dev/Monsterhunterworld-renderdoc.tar.gz

daniel-schuermann commented 5 years ago

Can you confirm that this issue is fixed with b2bb91dcccff6053f1f2f223f4e00da957190421 ?

Johnnynator commented 5 years ago

Yep this fixed the issue.

pendingchaos commented 5 years ago

How is performance after that fix?

Johnnynator commented 5 years ago

How is performance after that fix?

Didn't do any exact measurements. Roughly in the position of OP's screenshot it is roughly 1-2 fps more with ACO (~42fps max instead of ~40fps max).

pendingchaos commented 5 years ago

ok

ghost commented 5 years ago

Also confirming it's fixed for me. Thanks genius (not sarcastic).

I am getting same fps with mesa ACO or RADV_PERFTEST=llvm, in a test spot standing in same exact location, however I'm probably CPU bound.