#asahi-dev · 2026-08-09 11:10
agmitron[m]Device: Macbook Pro 16” M2 Max#asahi-dev · 2026-08-09 11:10
agmitron[m]Is there a place I can report my observations about using PMP?#asahi-dev · 2026-08-09 11:10
agmitron[m]Hi. Are PMP usage reports welcome?#asahi-dev · 2026-08-09 10:21
svenno idea if it’s a new variant or just the regular „other“ one used everywhere else #asahi-dev · 2026-08-09 10:20
svenfrom a quick glance I think at least the mailbox has changed #asahi-dev · 2026-08-09 10:19
chaos_princessdon't have any of those#asahi-dev · 2026-08-09 10:18
svenm2 I can’t guess, I need a trace there #asahi-dev · 2026-08-09 10:17
svennice!#asahi-dev · 2026-08-09 10:16
chaos_princesst6000 works#asahi-dev · 2026-08-09 09:55
nickchanthe thing is I have heard of *both* A8 and A11 and I do not know which is which#asahi-dev · 2026-08-09 09:54
nickchanthat would match with metal support#asahi-dev · 2026-08-09 09:53
mmediouni[m]nickchan: wasn't there AGX0? realworldtech claims that the Apple-designed shader core shipped in the A8: https://www.realworldtech.com/apple-custom-gpu/#asahi-dev · 2026-08-09 09:45
nickchanwould make adding a11 nvme less ugly code wise#asahi-dev · 2026-08-09 09:45
nickchanspeaking of nvme probably could simplify m1n1's code by using a queue size of 1#asahi-dev · 2026-08-09 09:39
nickchanA11 is the first AGX, previous is powerVR I think#asahi-dev · 2026-08-09 09:38
nickchandon't worry about pre M-series AGX too much, it would be... some time at least#asahi-dev · 2026-08-09 09:38
svenhttps://github.com/AsahiLinux/linux/commits/b4/cio-reset/ assuming the poll works now that's the reset controller series#asahi-dev · 2026-08-09 08:48
svencould be the same thing, could be something new. guess i'll just stick to M1 for now#asahi-dev · 2026-08-09 08:48
svenhm... M2 has iop,mxwrap-acio now instead of iop,m3wrap-v2-acio#asahi-dev · 2026-08-09 08:29
svenand then dt-bindings :(#asahi-dev · 2026-08-09 08:28
sveni'll do M1/M2 for now since atcphy is the same there#asahi-dev · 2026-08-09 08:28
chaos_princessthat package has m3 too if you are interested#asahi-dev · 2026-08-09 08:27
svenyeah, nvm, i already have them#asahi-dev · 2026-08-09 08:27
svenah wait.. maybe they're #asahi-dev · 2026-08-09 08:27
svenanyone got all m2 adt templates for me?#asahi-dev · 2026-08-09 08:23
sofuschadmed: either im dumb or there was no change to the submissions/instructions, maybe there is an enable_interchange register or something#asahi-dev · 2026-08-09 08:19
svennow do i want to look into how m2 does this or do i want to write dt-bindings? :(#asahi-dev · 2026-08-09 08:17
svenchaos_princess: maybe fixed, i was confused by readl_poll_timeout again. if this doesn't work i'll just add a udelay to the reset controller#asahi-dev · 2026-08-09 07:45
svenbut good to know the rest works then #asahi-dev · 2026-08-09 07:45
svenokay, I guess the „wait for the reset thingy to finish“ poll is just broken then #asahi-dev · 2026-08-09 06:25
chadmedwas just curious if we knew how to make avd do it#asahi-dev · 2026-08-09 06:25
chadmedyeah but not important until the gpu and dcp are sorted out#asahi-dev · 2026-08-09 06:16
sofusi suppose we could just add our own and add special handling in the vaapi-wrapper#asahi-dev · 2026-08-09 06:07
chadmedoh god what a mess, looks like each vendor just creates their own pixfmts lol#asahi-dev · 2026-08-09 06:03
sofusim not sure, but i dont think so#asahi-dev · 2026-08-09 06:03
chadmedim not entirely sure. does v4l2 support format modifiers in the same way drm does?#asahi-dev · 2026-08-09 06:00
sofushow would it work from a driver perspective? would it just be adding a new pixel format?#asahi-dev · 2026-08-09 05:56
chadmedthere may be cases where the gpu cant read interchange (i dont think anyone tested G13 but potentially pre-M-series AGX too) but that doesnt matter for direct scanout since all the dcp variants we support can read them#asahi-dev · 2026-08-09 05:53
chadmedto jog your memory: https://oftc.catirclogs.org/asahi-dev/2026-06-22#35437355;#asahi-dev · 2026-08-09 05:53
chadmedim just thinking _potentially_ some of the artefacts/slowness could be related to copying big linear buffers to the gpu. both the gpu and dcp can read interchange #asahi-dev · 2026-08-09 05:52
chadmedno rush, still need to get a bunch of other dcp-related stuff merged first#asahi-dev · 2026-08-09 05:51
sofusi dont think i ever traced an actual video playing... will do later#asahi-dev · 2026-08-09 05:50
chadmedit should be using them already as the surface they end up scanned out on is an interchange surface#asahi-dev · 2026-08-09 05:49
sofuschadmed: no, do you know if i need to to anything special to make macos use them? (for tracing)#asahi-dev · 2026-08-09 05:38
LarstiQ(the peanut gallery appreciates the context and progress)#asahi-dev · 2026-08-09 00:44
chadmedsofus: do we know how to make AVD export interchange compressed buffers?#asahi-dev · 2026-08-08 22:42
chaos_princessok, tb networking is probably not working correctly, with two linux machines at both sides they don't see each other either#asahi-dev · 2026-08-08 22:35
chaos_princessand, with the else branch commented out - tb links up#asahi-dev · 2026-08-08 22:29
chaos_princessi added a udelay(100) for debug purposes after the write to APPLE_CIO_M3_CTRL_START, and the serror is gone#asahi-dev · 2026-08-08 22:20
svenif that points to the mbox MMIO the power on sequence is wrong