15:37 gfxstrand: MrCooper_: It's because the semaphore wait gets inserted in the 3D queue, blocking on the previous rendering and then queue ordering rules kick in. If panvk were better about not doing unnecessary VS<->FS sync and we were better about only etting a few flags on our subits, it might still be avoidable.
15:38 gfxstrand: But also, the entire synchronization is unnecessary since only the compositor needs to wait on that frame, not further rendering, unless the client explicitly requests it.
16:29 Leftmost: I'm trying to test rusticl modified to enable SPV_KHR_untyped_pointers. With support enabled, SPIRV-LLVM-Translator generates function bodies with OpUntypedVariableKHR, for which trunk doesn't implement support. If I treat it like a variable and pass it to vtn_handle_variables() (which is also what vtn_handle_variable_or_type_instruction() does), I don't get the unsupported op error, but my entire system grinds to a halt and I have
16:29 Leftmost: to hard reset. Looking for suggestions on what could be happening or how I could debug this. (Ping cmarcelo as the implementer of SPV_KHR_untyped_pointers.)
16:33 Leftmost: Not at all familiar with the VTN code, so even just a way of dumping the processed instructions would be a good start.
18:02 zmike: mareko: great news - I think I finally figured out and fixed those crashes in viewperf
19:07 jenatali: zmike: The pipe_surface stuff finally bubbled up to the top of my list because it turns out there's actually a bug that needs to be fixed :)
19:07 zmike: it found a nonzero number of bugs in many drivers
19:08 jenatali: No this is a bug in the util replacement specifically
19:08 jenatali: Draw -> blit -> draw recursion breaks that pretty badly...
19:09 zmike: good find
22:02 jenatali: zmike: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36945
22:03 zmike: nature is healing