#asahi-dev · 2026-03-24 10:08
chadmedkwin bypasses the pipeline unless theres hardware compositing happening anyway, so its not going to let us get rid of some of the jank we need to do to figure out which colourspace and xfer function to use for hdr modes unfortunately#asahi-dev · 2026-03-24 10:04
chadmedoh neat, might test with gstreamer when this hits ::gentoo #asahi-dev · 2026-03-24 10:00
chadmedverbose log implies it succeeds with sw decode up until the point it tries to export a frame but that could just be the logs being nondescriptive#asahi-dev · 2026-03-24 09:51
chadmed(and by "pipeline" i mean a single stage for the transfer function since theres no upstream colourspace stage yet)#asahi-dev · 2026-03-24 09:51
chadmedi got colour pipelines "working" but its a massive pain to test since mpv does not like exporting wayland dmabufs to dcp even though we support the formats#asahi-dev · 2026-03-23 11:08
chadmeddcp doesnt pick up any >=10bpc full range ycc colour modes on any of my tvs so i cant even concretely confirm if it's due to the limited range. i _assume_ it is given that everything looks washed out though#asahi-dev · 2026-03-23 11:05
chadmedunless i can find proof that the macos compositor isnt doing some weird stuff to compensate for ycc hdr modes and we're just holding dcp wrong#asahi-dev · 2026-03-23 11:05
chadmedwe may have to just not support hdr on the hdmi port#asahi-dev · 2026-03-20 23:50
chadmedpreferably dumping the IOSurface and IOMFBSwapRequest structs for each swap#asahi-dev · 2026-03-20 23:49
chadmedif anyone has a machine with an HDMI port still running 13.5, _and_ a monitor that advertises DisplayHDR over HDMI using a Y'CbCr colour mode i would very much appreciate a dcp/dcpext trace#asahi-dev · 2026-03-16 08:46
chadmedshould knock us down by a couple of patches :P #asahi-dev · 2026-03-16 08:46
chadmedill have to prep a new audio branch based on what got merged#asahi-dev · 2026-03-15 12:44
chadmedespecially as stuff (dt bindings) churn upstream#asahi-dev · 2026-03-15 12:43
chadmedplease try to keep m1n1 up to date unless you are running really old kernels for a specific reason#asahi-dev · 2026-03-15 06:02
chadmedmacos does the right thing. perhaps we are missing surface colourspace or transfer function values#asahi-dev · 2026-03-15 04:54
chadmeddisplayport is fine though, same monitor looks great there because it has the bandwidth for an RGB colour mode#asahi-dev · 2026-03-15 04:54
chadmedhmm not looking good for HDR on the HDMI port. DCP really doesnt like tone mapping RGB BT.2020 surfaces to a BT.2020 YCC display. looks absolutely awful #asahi-dev · 2026-03-14 22:45
chadmedohh so thats where broadcom got their inspiration for the raspi socs#asahi-dev · 2026-03-14 22:42
chadmedi think even the 2600 could do more, didnt they port space invaders to it?#asahi-dev · 2026-03-14 22:42
chadmedwell you can have a background and two sprites :P #asahi-dev · 2026-03-14 22:18
chadmedwe can swap 3 if we're very careful about what's on each of those three surfaces, and which three surfaces the content is on#asahi-dev · 2026-03-14 22:16
chadmednicolas17: it's 4 on paper and we've only seen macos use 3 at a time, and only ever actively swap/page flip two of those#asahi-dev · 2026-03-14 04:19
chadmedis there an interlock? does SIL need to report that it's active for the mic/webcam to come up? #asahi-dev · 2026-03-12 12:20
chadmedactually if i can sort out forcing a modeset when hdr is toggled on/off and fixing the colour mode selection logic to be less silly with oddball monitors (dcp puts mine into some fucked 12-bit ycc mode when enabling hdr) we can probably push some version of hdr support for external monitors behind a module parameter#asahi-dev · 2026-03-12 12:17
chadmedi have a couple of dp 1.4 dongles and my monitors get full 165hz (and hdr works pretty well too but we're not talking about that until i look at drm colour pipelines teehee)#asahi-dev · 2026-03-12 12:15
chadmedyeah#asahi-dev · 2026-03-11 08:51
chadmedit wouldnt make sense for it to avoid the cluster at all if it wasnt at least partially shutting it off #asahi-dev · 2026-03-11 08:51
chadmedhmm, i do know that it will try to avoid migrating tasks across clusters if there is still capacity in an already-active cluster but im not 100% sure if it shuts the unused cluster down or just puts all the cores into wfi like we do #asahi-dev · 2026-03-11 08:42
chadmeddoes macos not do this to shut off unused clusters#asahi-dev · 2026-03-11 08:12
chadmedso this is really only useful for platform suspend and not opportunistically sleeping clusters without some guard logic to ensure at least one core is alive#asahi-dev · 2026-03-11 08:10
chadmedlmao right so if all but one core is off, and that core then happens to go into wfi without another core waking up, we're boned #asahi-dev · 2026-03-11 08:09
chadmede.g. if all the pcores shut off and the soc shuts off the uncore cluster parts, that is a nice power saving, and whatever we leave behind in m1n1 can spin them back up enough for linux to restore state#asahi-dev · 2026-03-11 08:08
chadmedshutting down only the secondaries might be fine for a cpuidle hack, hard to say #asahi-dev · 2026-03-11 08:08
chadmedmaybe, but if we want to support s2ram then linux will try to shut them all down anyway#asahi-dev · 2026-03-10 08:50
chadmedSEPFW and DCP firmware say hello :P #asahi-dev · 2026-03-10 08:50
chadmedm1n1 is what, ~600 kB? no ones gonna miss that, lets just reserved-memory the whole image :P #asahi-dev · 2026-03-10 08:48
chadmedyeah#asahi-dev · 2026-03-10 08:48
chadmedshould we have printfs in that hot path#asahi-dev · 2026-03-10 08:38
chadmedas in, coming out of the state seems standard enough but do we know how to actually get into it properly#asahi-dev · 2026-03-10 08:36
chadmedbut if we dont actually understand how it works fully theres no point #asahi-dev · 2026-03-10 08:36
chadmedyeah i was looking into what would have to change in m1n1, i dont think its _that_ much (put a trampoline at the entry point, reserve the range, resume table for the cpuidle driver to write to?)#asahi-dev · 2026-03-10 08:29
chadmedcpuidle-apple already does deep wfi it seems#asahi-dev · 2026-03-10 08:29
chadmedhuh i thought the busted sleep state was deep wfi, is there a deeper state we're missing?#asahi-dev · 2026-03-07 03:53
chadmedspecifically 14" mbp users#asahi-dev · 2026-03-07 03:53
chadmedpushed some changes to asahi-audio, testing welcome#asahi-dev · 2026-03-06 11:44
chadmedthe screen looks fine, the cursor just falls back to software when the framebuffer cant be created#asahi-dev · 2026-03-06 11:43
chadmedno the output of wayland-info #asahi-dev · 2026-03-06 11:22
chadmedthe output on my m2 max studio is quite... interesting#asahi-dev · 2026-03-06 10:12
chadmedthe compositor should be trying to offload them to a plane if it can#asahi-dev · 2026-03-06 10:12
chadmedwayland surfaces being interchange isnt pathological yeah