#
✕ nick:chadmed
#asahi-dev · 2026-06-25 11:23
chadmedhow much of the instruction stream do you need? just one frame?
#asahi-dev · 2026-06-25 11:21
chadmedone sec
#asahi-dev · 2026-06-25 11:13
chadmedi will do t6k tomorrow morning aus/brisbane time
#asahi-dev · 2026-06-25 11:13
chadmedsofus: from t8103 https://paste.debian.net/hidden/e6057765
#asahi-dev · 2026-06-24 11:56
chadmedno none at all, just the dev_info we added and the errors about the fifo slots
#asahi-dev · 2026-06-24 11:39
chadmed(not that it matters, the output is still garbage/uninitialised green)
#asahi-dev · 2026-06-24 11:39
chadmedare we not clearing the vpdone interrupt properly
#asahi-dev · 2026-06-24 11:38
chadmedsofus: we now get 0x102s but still invalid slots, which eventually become no free slots
#asahi-dev · 2026-06-24 11:04
chadmedits been 10 years and still nothing properly supports v4l2req sadly
#asahi-dev · 2026-06-24 11:03
chadmedif we stick with v4l2req we will probably have to adopt bootlin's libva shim too
#asahi-dev · 2026-06-24 11:02
chadmedjannau: no stress, as long as youre happy with v2
#asahi-dev · 2026-06-24 10:42
chadmedhttps://paste.debian.net/hidden/2ad1ab02
#asahi-dev · 2026-06-24 10:40
chadmedsame thing
#asahi-dev · 2026-06-24 10:39
chadmedfolded with zero pressure
#asahi-dev · 2026-06-24 10:39
chadmedactually nah ill do it now
#asahi-dev · 2026-06-24 10:39
chadmedyeah sure, will do tomorrow or friday
#asahi-dev · 2026-06-24 10:36
chadmedhm, got a bunch of 0x100s, then a no free slots with -ENOMEM after an 0x1000
#asahi-dev · 2026-06-24 10:31
chadmedi have USE="soc" so the v4l2 request api patches are applied
#asahi-dev · 2026-06-24 10:27
chadmedsorry not t6k, t6021
#asahi-dev · 2026-06-24 10:27
chadmedsame video worked on t6k and t8112
#asahi-dev · 2026-06-24 10:27
chadmedyeah via mpv
#asahi-dev · 2026-06-24 10:24
chadmednah, just no assigned and invalid
#asahi-dev · 2026-06-24 10:23
chadmedit also says no assigned vp slots
#asahi-dev · 2026-06-24 10:23
chadmedno not my fault
#asahi-dev · 2026-06-24 10:21
chadmedoh hang on this may be my fault one sec
#asahi-dev · 2026-06-24 10:19
chadmedyep invalid fifo slot: 255 for every frame
#asahi-dev · 2026-06-24 10:14
chadmedyeah
#asahi-dev · 2026-06-24 10:07
chadmedhang on i have the adt around here somewhere
#asahi-dev · 2026-06-24 10:07
chadmedsofus: wont probe on t8103, the mailbox region cannot be reserved?
#asahi-dev · 2026-06-24 09:53
chadmedi would have liked to test direct scanout but its 8pm, im not hacking janky colourop support into dcp now :P
#asahi-dev · 2026-06-24 09:53
chadmedso avd->gpu->dcp
#asahi-dev · 2026-06-24 09:52
chadmedand thats with gpu compositing/tonemapping
#asahi-dev · 2026-06-24 09:52
chadmed14 W above background down to ~1.6 W above background for 4K60
#asahi-dev · 2026-06-24 09:51
chadmedholy fuck
#asahi-dev · 2026-06-24 09:47
chadmedit's just https://www.youtube.com/watch?v=njX2bu-_Vw4
#asahi-dev · 2026-06-24 09:47
chadmedmaybe interchange buffers will help a bit, but it _should_ be zerocopy with `-hwdec=v4l2-request -vo=dmabuf-wayland`
#asahi-dev · 2026-06-24 09:46
chadmednope
#asahi-dev · 2026-06-24 09:46
chadmedthis 1080p video is choking
#asahi-dev · 2026-06-24 09:43
chadmedmmm, we're not using all four vp slices are we
#asahi-dev · 2026-06-24 09:41
chadmednup, no 4k avc anymore. ill just download the vp9 and reenc using ffmpeg
#asahi-dev · 2026-06-24 09:36
chadmedyeah
#asahi-dev · 2026-06-24 09:34
chadmedoh you can only get 4k from youtube in vp9
#asahi-dev · 2026-06-24 09:30
chadmedyeah let me download one
#asahi-dev · 2026-06-24 09:29
chadmed`mpv -hwdec=v4l2request_copy -vo=dmabuf-wayland ~/Videos/E26.mp4` results in ~700 mW above background
#asahi-dev · 2026-06-24 09:28
chadmedok so while i wait for a kernel build on t8103, on t8112 `mpv ~/Videos/E26.mp4` results in ~1.4 W above background (1280x720 dbza ep 26 from archive dot org)
#asahi-dev · 2026-06-24 08:51
chadmedhad some life stuff to deal with last night, let me fire up this machine now
#asahi-dev · 2026-06-24 08:41
chadmedchaos_princess: so most of our hardware :P
#asahi-dev · 2026-06-24 08:38
chadmedi mean tbh i think dealing with the quirks in the driver _is_ fine in theory and wont get us NACKed or anything, but if we get 7 more AVD variants for 3 more socs.... it's gonna be a bit silly
#asahi-dev · 2026-06-24 08:37
chadmedyeah, and we're not going to have a 3-line driver with static void decode_frame(void *enormous_command_buffer) a la vendor drivers
#asahi-dev · 2026-06-23 12:36
chadmedit will be mentioned in the progress report though dont worry

plus ›