06:41 MrCooper: kisak: FWIW, it's not directly related to DRI2, Xwayland never supported that
06:42 MrCooper: rather looks like Mesa requires explicit sync for DRI3 now
08:50 MrCooper: kisak: actually it's probably because Mesa dropped support for the wl_drm Wayland protocol, which older Xwayland required for HW acceleration; when Xwayland falls back to software rendering, it doesn't expose the DRI3 extension, resulting in the reported failure
19:14 zf: do piglit tests get run on i386?
19:14 zf: for llvmpipe specifically, if that matters
19:44 karolherbst: mareko: what's the best way to figure out why I get "radeonsi: error: LLVM triggered Diagnostic Handler: couldn't allocate output register for constraint 'v'"? I'm looking into subgroup stuff and I think LLVM doens't like 8 bit int subgroup operations, but it's not entirely clear to me where the constraint is coming from
20:28 pendingchaos: karolherbst: are you running with AMD_DEBUG=checkir?
20:28 pendingchaos: might be ac_build_optimization_barrier()
20:30 karolherbst: pendingchaos: I think that's done by default in debug builds? But I also thought it's ac_build_optimization_barrier but disabling it didn't seem to make much of a difference? maybe I messed up, let me check again
20:30 karolherbst: oh huh...
20:30 pendingchaos: doesn't look like it's on by default
20:31 karolherbst: okay so it is ac_build_optimization_barrier
20:31 karolherbst: well I'm getting it without checkir
20:31 karolherbst: then
20:31 karolherbst: but with ac_build_optimization_barrier disabled and checkir I don't get any errors anymroe
20:32 karolherbst: it's a bit weird because radv using llvm doesn't run into this issue, but maybe some subgroup lowering is different
20:35 mareko: karolherbst: ac_build_optimization_barrier should do the same thing for i8 that it does for i1
20:41 karolherbst: ahh okay, that's easy enough to fix then
20:45 karolherbst: yep, that seems to work, thanks!