21:17 mareko: the switch to PACKED packets seems like quite a rewrite
21:21 mareko: karolherbst: I think a lot the bugs you're seeing are due to MESA_SHADER_KERNEL not being handled
21:23 mareko: the current radeonsi approach is to buffer register writes in the driver, and then set all buffered registers with a single packet before a draw/dispatch
21:50 karolherbst: mhhh,, possibly
21:50 karolherbst: but I also don't see many bugs
21:50 karolherbst: I have like 15 remaining fails and they are not related to that
21:51 karolherbst: but yeah.. like that darktable issue _could_ be caused by that or something...
21:51 karolherbst: I'll think about how to ditch SHADER_KERNEL, but it's a non trivial change for a bunch of drivers