#
✕ nick:chadmed
#asahi-dev · 2026-01-02 12:10
chadmedi think this probably needs some work in kwin lol
#asahi-dev · 2026-01-02 12:10
chadmedoh hm no now kwin just falls back to compositing if there are more than two surfaces
#asahi-dev · 2026-01-02 12:08
chadmedyup lets fucking go three surfaces blending with immutable zpos
#asahi-dev · 2026-01-02 11:59
chadmedthis smells like maybe a kwin bug. aiui there should never be two planes at the same zpos
#asahi-dev · 2026-01-02 11:57
chadmedyeah 15" m2 air
#asahi-dev · 2026-01-02 11:55
chadmedh.264 playback is sitting at ~1 W with this though so thats nice
#asahi-dev · 2026-01-02 11:53
chadmedugh ffs yeah kwin is trying to put two planes at the same zpos
#asahi-dev · 2026-01-02 11:51
chadmedor maybe this really is just busted
#asahi-dev · 2026-01-02 11:51
chadmedjannau: kwin just does it? only issue ive seen is occasional hitching/wrong fb being scanned out but afaict we are relying on implicit sync and idk if kwin likes that
#asahi-dev · 2026-01-02 11:49
chadmedi think we might need proper sync for this to work though... it quite often wants to put old framebuffers on the screen
#asahi-dev · 2026-01-02 11:34
chadmedwe can actually scan out all 4 surfaces at once!
#asahi-dev · 2026-01-02 11:34
chadmedholy shit yes
#asahi-dev · 2026-01-02 09:28
chadmedhm turning on fb damage clips breaks kwins ability to turn off the builtin display
#asahi-dev · 2026-01-01 10:55
chadmedrefer to the latest progress report for a foolproof method for debugging allocation errors :P
#asahi-dev · 2026-01-01 10:55
chadmedabsolutely indispensible especially given how useless clangd still seems to be with the kernel tree
#asahi-dev · 2026-01-01 10:54
chadmedelixir my beloved
#asahi-dev · 2026-01-01 10:52
chadmedi will clean up the commit messages and explain stuff better at some point
#asahi-dev · 2026-01-01 10:51
chadmedleio: it was educational for me too. i knew nothing about drm/kms two weeks ago. now i know very very slightly more than nothing!
#asahi-dev · 2026-01-01 07:11
chadmedbasically just the stuff for YUV scanout and some fixes for things we were doing wrong
#asahi-dev · 2026-01-01 07:10
chadmedive pulled some of the more experimental stuff out of my dcp dev branch and have a branch with just what can be merged now http://github.com/chadmed/linux/tree/dcp/mergable
#asahi-dev · 2025-12-31 14:27
chadmedonly very very old stage1 m1n1s need updating. if your install is ~recent (last 18 months or so) you should be fine
#asahi-dev · 2025-12-31 06:57
chadmedflying reeeaaaalll close to the sun using a 14 year old secondhand disk with dvr history and no backups though....
#asahi-dev · 2025-12-31 06:56
chadmedthat sucks :(
#asahi-dev · 2025-12-30 12:32
chadmedfun :p
#asahi-dev · 2025-12-30 12:30
chadmedyep, try setting DRM_FORMAT_MOD_APPLE_GPU_TILED and DRM_FORMAT_MOD_APPLE_GPU_TILED_COMPRESSED as valid format motifiers in the dcp driver and see what happens
#asahi-dev · 2025-12-30 12:29
chadmedscanning out nonlinear fbs is just a nice to have power optimisation that i suspect might also help with our dismal swap timing
#asahi-dev · 2025-12-30 12:29
chadmedif thats too daunting then lets work on finding the iosurface transfer functions and colourspaces
#asahi-dev · 2025-12-30 12:28
chadmedbut looking at IOMFB, there is some flag we need to set _somewhere_ to make DCP scan out interchange or AGX format framebuffers. i havent been able to find this flag at all and id really like for us to be able to scan out AGX format framebuffers if possible
#asahi-dev · 2025-12-30 12:26
chadmedAGX supposedly can write "interchange" format textures but cannot read them so we never bothered looking at them
#asahi-dev · 2025-12-30 12:26
chadmedmore RE if youre happy to do that i suppose. if you really want a challenge, youll notice that the framebuffer formats used by macos are quite odd. that is because they are "apple interchange/interchange compressed" which we dont know much about yet
#asahi-dev · 2025-12-30 12:25
chadmednice
#asahi-dev · 2025-12-30 12:24
chadmedtry using 'r01l' for A/XARGB2101010 instead
#asahi-dev · 2025-12-30 12:24
chadmedi have a hard time believing 'r03w' actually "looks fine" though :P
#asahi-dev · 2025-12-30 12:23
chadmedyeah if we sort the fake edids out then i dont see why not
#asahi-dev · 2025-12-30 12:23
chadmedffmpeg does the right thing so its not terribly difficult to make known-good source files to play withf
#asahi-dev · 2025-12-30 12:21
chadmedso as long as your mp4 file has all the proper tags and colourspace metadata, quicktime will tell the compositor to tell dcp to do what you want it to
#asahi-dev · 2025-12-30 12:21
chadmede.g. an apple tv will hw decode and direct scanout h.265 video tagged with 'hvc1' but not 'hevc' or 'hev1' which are for some reason common nonstandard tags with.... "alternatively sourced" media
#asahi-dev · 2025-12-30 12:20
chadmedapples multimedia handling is in general very very good when the media is spec-conformant
#asahi-dev · 2025-12-30 12:19
chadmedthats how i found BT601 and BT709
#asahi-dev · 2025-12-30 12:19
chadmedget some source video that explicitly uses each and then play it back in quicktime while tracing
#asahi-dev · 2025-12-30 12:18
chadmedi havent tried plugging in my dolby vision oled yet but thats another factor we have to consider since DV uses completely different colour spaces again lol
#asahi-dev · 2025-12-30 12:17
chadmedso im not terribly keen on enabling hdr until we have some more control over colour modes and what dcp is doing to blend surfaces. at a minimum we need to figure out wiht more certainty what all of the surface colourspaces and transfer functions are
#asahi-dev · 2025-12-30 12:16
chadmedwhen i plug the exact same monitor in via hdmi, it looks like total garbage because dcp thinks limited YCbCr is the "best" colour mode for that display
#asahi-dev · 2025-12-30 12:15
chadmedso yeah it looks fine when everything lines up. my monitor plugged in to displayport with full range 10-bit RGB, PQ transfer func and BT.2020 colour space of course looks fantastic when kwin sets the surface framebuffer to ARGB2101010 and uses the PQ transfer function with BT.2020 colour space
#asahi-dev · 2025-12-30 12:13
chadmedproblem 3 is that without flaky gross hacks that sometimes cause locks and hangs, we cannot switch between the correct surface primaries and transfer functions when enabling/disabling HDR at runtime
#asahi-dev · 2025-12-30 12:12
chadmedtrying to output an RGB BT.2020 surface on to a limited range YCbCr BT.2020YCC display looks like crap
#asahi-dev · 2025-12-30 12:12
chadmedproblem 2 is that DCP's mode selection logic is stupid, and for some displays will select YCC BT.2020 primaries
#asahi-dev · 2025-12-30 12:12
chadmedproblem 1 is that we dont know that "DCP_XFER_FUNC_HDR" actually corresponds to PQ, HLG, or HDR10 which all use different curves
#asahi-dev · 2025-12-30 12:11
chadmedoliver: kwin can currently only put the connector into RGB BT.2020 with PQ transfer function
#asahi-dev · 2025-12-30 11:16
chadmedbut im back at work after this week :(

plus ›