#asahi-dev · 2026-08-31 09:26
chadmedeh if it doesnt work ill swap it, just sent the format mod patch and cced you anyway #asahi-dev · 2026-08-31 09:26
chadmedhm, AIL initialises compression buffers the other way around#asahi-dev · 2026-08-31 07:31
chadmedsofus: can you please sanity-check my dcp/interchange branch with avd? i think it should be correct now#asahi-dev · 2026-08-30 12:06
chadmedthis has gone so well for fairydust ;)#asahi-dev · 2026-08-30 12:00
chadmedthats not too bad, i think ive ended up sending like double that just for audio#asahi-dev · 2026-08-30 11:56
chadmednice#asahi-dev · 2026-08-30 11:55
chadmedare you going to send the latter without the rfc tag? #asahi-dev · 2026-08-30 11:53
chadmedmesa notwithstanding#asahi-dev · 2026-08-30 11:53
chadmedhopefully that goes in for 7.4 and we can get interchange happening before the end of the year#asahi-dev · 2026-08-30 11:52
chadmedi hope to have the format modifier commit out shortly after we get an -rc1 for 7.3 #asahi-dev · 2026-08-30 11:51
chadmedezpz#asahi-dev · 2026-08-30 11:45
chadmedill fix up the dcp driver up tomorrow at some point#asahi-dev · 2026-08-30 11:44
chadmedeyyyyy lets gooo#asahi-dev · 2026-08-29 13:20
chadmedif a client doesnt request a specific modifier it probably doesnt know about them at all and so we shouldnt assume interchange use is indicated. agx can deal with its own format fine though so we're allowed to use that if the requested buffer can be tiled and compressed#asahi-dev · 2026-08-29 13:20
chadmedmesa currently only allows interchange if the drm client explicitly requests it to prevent it being used for agx-local framebuffers it shouldnt be used for#asahi-dev · 2026-08-29 13:19
chadmedok maybe we dont need to change mesa then#asahi-dev · 2026-08-29 13:17
chadmedi recommend the overlay plane with interchange support so that you dont smush the tty #asahi-dev · 2026-08-29 13:17
chadmedreplace plane-id with the drm object id of whichever surface you want to put the video on#asahi-dev · 2026-08-29 13:17
chadmedkmssink driver-name=apple skip-vsync=true plane-id=40#asahi-dev · 2026-08-29 13:15
chadmedits just apple#asahi-dev · 2026-08-29 13:14
chadmedyou need to disable vsync, we dont have fences#asahi-dev · 2026-08-29 13:14
chadmedim not 100% sure how v4l2 format and drm format modifier interop is supposed to work at this point, maybe we need to allow interchange even when its not explicitly requested by a drm client#asahi-dev · 2026-08-29 13:13
chadmedyou might want to try gstreamer's kmssink#asahi-dev · 2026-08-29 13:12
chadmedhave you applied both the kernel and mesa branches#asahi-dev · 2026-08-29 12:32
chadmedsven: ack#asahi-dev · 2026-08-29 12:29
chadmedyou dont get to submit kernel code if you did binary re, please familiarise yourself with https://asahilinux.org/copyright/#asahi-dev · 2026-08-29 12:03
chadmedsven: should https://github.com/chadmed/linux/commit/1442db420957d3d9985ac6b029d945a445f95f7c go through your tree or arm?#asahi-dev · 2026-08-28 13:37
chadmedvalue.value.i = 0x30313050 for both vos#asahi-dev · 2026-08-28 13:18
chadmedit's not the libva translation layer, it happens with -hwdec=v4l2request too #asahi-dev · 2026-08-28 13:17
chadmedi suspect mpv's shaders given ff and gstreamer seem to work fine?#asahi-dev · 2026-08-28 13:16
chadmedits not like it knows the buffer is nv16, the playback stats still say nv12 #asahi-dev · 2026-08-28 13:16
chadmedyeah i wonder why dmabuf-wayland is doing the right thing then#asahi-dev · 2026-08-28 13:05
chadmedaah i cant grab an apothecary diaries or hellsing screenshot too, theyre both AV1 :P #asahi-dev · 2026-08-28 13:01
chadmedit does the same thing for P010 #asahi-dev · 2026-08-28 12:46
chadmedthe washout seems to be frame-by-frame and related to the blocking artefacts ive seen with h264 on certain videos#asahi-dev · 2026-08-28 12:43
chadmedyeah#asahi-dev · 2026-08-28 12:42
chadmedkwin does not advertise a scanout tranche for dcp yet, but it will hopefully do so Soon#asahi-dev · 2026-08-28 12:41
chadmednot yet :P#asahi-dev · 2026-08-28 12:39
chadmedbut that said, the dmabuf-wayland vo also exports an nv12 buffer#asahi-dev · 2026-08-28 12:39
chadmedi dont think the gpu's yuv handling has been thoroughly tested thus far#asahi-dev · 2026-08-28 12:34
chadmedthe problem becomes more obvious with low entropy images ^ #asahi-dev · 2026-08-28 12:27
chadmedthe "colours being wrong" appears to be pixel format related. the chroma is offset from the luma rather than any one colour being "wrong"#asahi-dev · 2026-08-25 13:43
chadmedit can handle an arbitrary number of different cores with different perf#asahi-dev · 2026-08-25 13:43
chadmedin the linux world we just tell the scheduler how fast the core is with a meaningless dmips/mhz metric, how much energy it uses at each pstate, and it does the rest#asahi-dev · 2026-08-25 13:43
chadmedhardware scheduling is more of a gpu thing, though intel had to bring it back for their heterogeneous socs initially because the windows scheduler is so shit#asahi-dev · 2026-08-25 13:42
chadmedthe scheduler is very cognisant of it#asahi-dev · 2026-08-25 13:41
chadmedof course in the real world its not so cut and dry. the pcores at a lower pstate might have better perf/w than the mid cores at a higher pstate #asahi-dev · 2026-08-25 13:40
chadmedif something is juuust too demanding for the ecores the scheduler can put it on the mid cores and avoid waking up the pcores if it doesnt need to#asahi-dev · 2026-08-25 13:40
chadmedit wont really, it will allow the scheduler to make different energy efficiency tradeoffs though#asahi-dev · 2026-08-25 13:39
chadmedyo dawg i heard you like core types