06:03marysaka[d]: oh that's why I see
15:21karolherbst[d]: will be doing some rusticl on zink on nvk testing today, wish me luck
17:04karolherbst[d]: `Device Name zink Vulkan 1.3(Quadro RTX 6000 (NVK TU102) (MESA_NVK))`
17:04karolherbst[d]: let's see how good this nvk driver really is 😛
17:10karolherbst[d]: the one thing standing out: running the CL CTS in parallel is super slow, but that's more of a "context creation is expensive" kinda situation
18:27karolherbst[d]: mhh..
18:27karolherbst[d]: nak chokes on `ufind_msb` having a 64 bit source
18:27karolherbst[d]: but anyway
18:27karolherbst[d]: `Pass 2595 Fails 84 Crashes 23 Timeouts 0`
18:28karolherbst[d]: something up with images...
18:29karolherbst[d]: quite a lot of annoying fails.. this is gonna be interesting
21:20dadschoorse[d]: karolherbst[d]: find_msb/lsb are limited to 32bit in Vulkan
21:21karolherbst[d]: mhhh...
21:21karolherbst[d]: not in nir tho
21:21karolherbst[d]: but maybe I also have to fix zink
21:22dadschoorse[d]: in nir it's not because radv needs the 64bit ops for OpGroupNonUniformBallotFindMSB/LSB