Created attachment 67819 [details] complete log file 16:02 <vlj> weirdo, adds on the bug that the shader uses too much temp (more than 128, max r600 temp) because of indirect addressing 16:03 <vlj> weirdo, glsl-to-tgsi does not allocate register when there is indirect addressing
This seems like an issue in the r600 back-end. The glsl-compiler component is the common front-end code. Perhaps Marek can comment...
Please test latest Mesa master. TGSI recently got support for array declarations of temporaries, making register allocation possible with indirect addressing. That said, I'm not sure if the r600g shader backend takes advantage of it.
When running as: -- % MESA_DEBUG=1 LIBGL_DEBUG=verbose vblank_mode=0 WINEDEBUG=fps R600_DEBUG=sb R600_LLVM=1 wine Falcon\ BMS.exe -window -- in BMS 4.32u6 there are no more shader miscompilation errors -or- CP stalls when looking into certain cockpit parts. As for the latter, it was present from 9.1 till very recently. HDR doesn't render but this is minor, probably a missing extension or a Wine quirk. I'm going to pester Stefan and Henri from Codeweavers about it before filing further bugs. This is CAYMAN 6970. Thanks to all involved. -sh
Created attachment 84979 [details] CP stall when looking down and to the right in the cockpit
The CP stall is still there, just different way of reproducing. One way or another, involves cockpit view manipulation. Do you want me to create a separate bug report? Attaching dmesg fragment.
Freeze seems R600_HYPERZ-related.
The freeze is tracked separately in #68799. I'm marking the issue resolved.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.