12:02 pmoreau: karolherbst: Sounds good for the SPIR-V extensions. And not too bad for the 2.2 CTS! What are the main things missing/failing?
12:02 karolherbst: images and random stufff
12:02 karolherbst: right now I try to get all the SVM stuff to pass though
12:04 karolherbst: pmoreau: here is my runner btw: https://gist.github.com/karolherbst/c30ff1352bffdf224a785564fec62b28
12:05 karolherbst: ohh right, conversions is a big thing as well
12:05 karolherbst: but simply because there are endless combinations :p
12:36 karolherbst: pmoreau: btw, mind reviewing the SVM bits?
13:03 karolherbst: pmoreau: ohh, and generic pointers is probably the biggest feature missing from the 2.0 stuff (besides SVM)
20:14 Lyude: a-ha!
20:15 Lyude: my guess on whythe outp crc source was broken was 100% right :), and I've got it spitting out CRCs now
20:16 Lyude: as it turns out, and this sort of makes sense, rg and sf crc outputs can just be turned on immediately whenever. outp CRCs however, have to be configured and ready when the outp is first acquired in hw
20:17 Lyude: so now that I completed the next step; moving CRC enable/disable over to be part of the atomic state; things just work
20:23 Lyude: I guess the next step is to confirm we're truly missing CRC events on context switches (I'm 95% sure of this, but I'd like to write a test to check instead of just relying on observing the crc audit timestamps from the GPU since I realized it's possible we might just be seeing a repeated CRC, not a dropped CRC when switching contexts) and if so, teaching nouveau how to figure out how many vblanks we've
20:23 Lyude: missed when switching contexts/sources based off the GPU crc audit timestamps and vblank interval
21:02 Lyude: ....mhh, seems outp still gives some trouble with mst though but maybe that's not unexpected
21:06 Lyude: hm, could also have something to do with the fact that we perform link training only after programming the core channel
23:18 bylaws: karolherbst sorry it's a bit late but kmscube runs
23:18 bylaws: mesa 19.2.6
23:18 bylaws: with tegra
23:18 karolherbst: bylaws: okay, so it's the same issue
23:19 bylaws: How hard I it to fix?
23:19 bylaws: is*
23:38 bylaws: I assume we couldn't just work around it?
23:38 bylaws: karolherbst:
23:39 karolherbst: ohh, I have a workaround, but tagr_ would like to fix it properly
23:39 karolherbst: bylaws: you could use that to workaround it for now: https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2960
23:41 bylaws: oh thanks
23:41 bylaws: I'll take a look