#
✕ nick:sofus
#asahi-dev · 2026-09-24 09:32
sofusis userspace supposed to handle the size calculation? In particular the offset to the second plane
#asahi-dev · 2026-09-24 09:28
sofusyes! after changing tile_bytes calculation, using 10/20 for bpp and changing the dcp format to 'p420' it works!
#asahi-dev · 2026-09-24 09:12
sofusit could actually be the same format as the one i found
#asahi-dev · 2026-09-24 09:11
sofusyeah p010 is wrong for interchange i think, or at least avd seems to output something else. We should have 10 bpp (ie. packed)
#asahi-dev · 2026-09-24 08:47
sofusthanks!
#asahi-dev · 2026-09-24 08:39
sofusi think im using the wrong format
#asahi-dev · 2026-09-24 08:36
sofusi have only tested nv12, but yes
#asahi-dev · 2026-09-24 08:20
sofusim talking about yuv surfaces, im getting Decomp fail messages
#asahi-dev · 2026-09-24 08:09
sofusi meant everything, because something is wrong
#asahi-dev · 2026-09-24 08:07
sofusie tile size and so on
#asahi-dev · 2026-09-24 08:06
sofusdoes anyone have the data for a 10b interchange surface?
#asahi-dev · 2026-09-22 18:08
sofusmulti core was easy to add, i cant test that it works tho
#asahi-dev · 2026-09-22 16:41
sofusit does not work with the old driver, both should be updated at the same time
#asahi-dev · 2026-09-22 14:57
sofusi wonder how the dart handles a reset when the iommus are merged
#asahi-dev · 2026-09-22 14:07
sofusthe code can be seen here: https://github.com/sofus13/avd-fw/tree/avd-next-fw and kernel: https://github.com/sofus13/linux/tree/b4/avd
#asahi-dev · 2026-09-22 14:05
sofusit will not, the cm3 talks to the hw while the driver handles load balancing and scheduling
#asahi-dev · 2026-09-22 13:59
sofusi dont fully get your question. I esentially moved some of the kernel code to the cm3, the only difference is esentially *where* things are done
#asahi-dev · 2026-09-22 12:52
sofusno, each block has their own
#asahi-dev · 2026-09-22 10:46
sofuscould avd be tested? ie just installing ffmpeg and running it
#asahi-dev · 2026-09-22 10:39
sofusbtw, does the dart on m4 work?
#asahi-dev · 2026-09-22 10:35
sofus(we dont even know if there is a performance benefit)
#asahi-dev · 2026-09-22 10:29
sofusbut not even apple seems to do it
#asahi-dev · 2026-09-22 10:28
sofusno job dispatch tho, but its possible to add it
#asahi-dev · 2026-09-22 10:21
sofusyeah will link something soon. The blob is submitted to the cm3 and it handles everything from there. The instructions are still prepared by the AP
#asahi-dev · 2026-09-22 10:09
sofusthe avd driver submits jobs to the cm3 now :) 9 files changed, 160 insertions(+), 432 deletions(-)
#asahi-dev · 2026-09-21 15:37
sofusintegralpilot: what does NET stand for?
#asahi-dev · 2026-09-21 15:37
sofus(only easy because they did all the hard work)
#asahi-dev · 2026-09-21 15:36
sofusmulti core should be pretty easy to add, i might even do it for v2
#asahi-dev · 2026-09-21 15:34
sofuscool thanks! Videos is enough for me, but thanks!
#asahi-dev · 2026-09-21 15:29
sofusthey might also have that, though im not sure. I hope slides/recording are published, im sure i could learn something
#asahi-dev · 2026-09-21 15:16
sofusthey also have that. They combine them so jobs dont depend on where they where allocated
#asahi-dev · 2026-09-21 15:12
sofusi see, thanks. I took a look at the patch, i would imagine we could do the exact same
#asahi-dev · 2026-09-21 15:06
sofusahh i see, i thought the dual core thing was one "device". But its the exact same as avd. I wonder why its not userspace that should handle that
#asahi-dev · 2026-09-21 14:36
sofusjannau: not sure, sounds like the talk is mostly about their multi core thing, i dont think avd has something similar. Im interested in the multi job thing, but im not sure if v4l2 already supports that
#asahi-dev · 2026-09-21 08:58
sofusJPeisach: please take your time, i will likely begin v2 in a week or two
#asahi-dev · 2026-09-20 07:43
sofusim very happy avd is so 'simple'. avd's piodma is only used to transfer to/from dart to the cm3. The same could basicly be implemented with a simple memcpy_toio
#asahi-dev · 2026-09-20 07:34
sofusoh ok i guess its used kinda different, having the same mappings should be ok for avd
#asahi-dev · 2026-09-20 07:32
sofusthanks! Out of curiosity, why does the dcp driver not do that?
#asahi-dev · 2026-09-20 07:18
sofushaving multiple entries in the iommus would not work right?
#asahi-dev · 2026-09-20 07:15
sofusi assume the dcp's piodma is like the avd's. Is there a simpler way to do that or do i need the subdevice?
#asahi-dev · 2026-09-19 13:32
sofusi think i confused myself and wrote bad dt-bindings. The docs says to try to make the bindings complete, but this is related things like 'has an interrupt' and not 'has 10 interrupts that do this' right?
#asahi-dev · 2026-09-19 07:17
sofusonly t6030 i think.
#asahi-dev · 2026-09-19 06:50
sofusintegralpilot: sorry, assume a todler wrote them. Apparently i suck at writing dt's
#asahi-dev · 2026-09-18 17:05
sofusJPeisach: the actual test setup is a little involved. You would install fluster and gstreamer and do the tests listed
#asahi-dev · 2026-09-18 16:32
sofusyeah
#asahi-dev · 2026-09-18 16:31
sofusJPeisach: iirc the device trees where only visual changes
#asahi-dev · 2026-09-18 15:28
sofusi did change some things so not exactly. For convenience maybe reverting the isp branch and then applying my series would be the easiest
#asahi-dev · 2026-09-18 15:24
sofusbut i should have tested and included that...
#asahi-dev · 2026-09-18 15:23
sofusJPeisach: i am only aware of https://github.com/AsahiLinux/linux/commit/3e3bd38fb1163061976866b13688f163c6f38277 being needed
#asahi-dev · 2026-09-18 14:08
sofusthanks, i will take a look

plus ›