04:29 airlied: demarchi: fyi I've left tip unmerged after xe, I just ran out of time, sorry to anyone else, if someone with xe knowledge wants to fix it that would be nice
08:03 shadeslayer: cmarcelo: we could have a nir_type_float | bfloat16 like we do for float16
11:41 K900: Hey folks, if I may, can I get some eyes on https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31074
13:04 demarchi: airlied: let me take a look
13:20 demarchi: airlied: merged, thanks
13:26 demarchi: rodrigovivi: it was due to some suspend/resume patches applied in drm-fixes and not others... drm-next and drm-tip now makes sense to me, but I'm not sure about the missing patches in drm-fixes. May be worth taking another look if we didn't miss something
16:33 agd5f: airlied, sima: Al Viro's fixes for prim fd handling kind of fell off my radar last week. I pull them in this week. Are you ok with a PR for them for 6.12, or would you rather wait for 6.13?
16:34 sima: I think 6.12 is fine, they don't look scary ...
16:35 agd5f: sounds good
16:35 sima: since they're fixes kinda should 6.11 but 6.12 should be ok too
18:46 spkry: hello, i might not be the first one to ask and i'm sorry but i do want to know how is a specific merge commit going, and i'm addressing idr and dj-death specifically
18:46 spkry: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31042
18:58 spkry: why are people still using mesa
18:58 spkry: i mean irc*
18:58 psykose: it good
18:59 spkry: for the most part it lacks encryption
18:59 spkry: also images
18:59 zmike: images should only be used to verify test results
19:00 psykose: images are only for sharing memes and this is a 100% serious results-oriented channel
19:00 spkry: sure it is
19:01 spkry: the MOTD itself is some kind of inside joke isn't it
19:02 zmike: no, it's incredibly literal
19:03 spkry: mhm
19:08 spkry: what's a bar anyway
19:53 abhinav__: vsyrjala jani Hi, I had a question on intel_dp_create_fake_mst_encoders(). May I know why its called "fake_mst"? IIUC, the encoder is being created for the port but the connector will be created later during MST hotplug and attached to this encoder. so encoder is still real hardware just not attached to any connector?
19:54 abhinav__: so its more of creating encoders in advance and then using it when connector is created
19:56 abhinav__: I am seeing a similar code in AMD too dm_dp_create_fake_mst_encoders(). so wanted to understand if i am missing some concept here which is generic across drm drivers
20:09 Lyude: Huh, did something happen to include/drm/drm_fbdev_generic.h? I noticed it's gone from the git tree, but unfortunately git/ddg don't seem to have any useful context
20:15 jannau: Lyude: aae4682e5d66 ("drm/fbdev-generic: Convert to fbdev-ttm")
20:36 Lyude: jannau: thank you!
21:59 airlied: abhinav__: they are fake because there is no actual encoder hw for them, we just have to create them to satisfy the constraints of the kms api
22:07 abhinav__: airlied ack, i guess it depends on each driver's hw modelling but lets say there was DSC over DP MST, then we would have an actual encoder hw right .... probably this part is not clear to me or maybe DSC case is handled differently
22:10 airlied: abhinav__: there is actual encoder hw, but the encoder object doesn't correspond to it
22:11 airlied: but it might have evolved that is actually does in some ways now and we could drop the fake :-)
22:14 abhinav__: airlied ack got it, i was going to create encoder objects corresponding to the encoder hw and then when MST connectors are created tie them up with these encoder objects. so i was referring current MST implementations, but didnt follow the fake encoder part. it seems like to me, its just implementation specific that encoder object does not
22:14 abhinav__: correspond to the encoder hw in the MST case but no harm if it does?
22:18 airlied: abhinav__: yup I think so, I'm kinda a bit vague on the modern details there, historically crtcs/encoders/connectors has certain requirements on each other about what connectors could use what crtcs etc and MST upset all of that, so adding fake ones was the only way to fix it and keep API operational
22:18 airlied: I think atomic has made some of the original reasons moot at this point
22:19 abhinav__: airlied got it, thank you so much