#asahi · 2025-08-04 22:02
linkmauveI don’t know if they did for this particular feature, but they do contribute some upstream kernel review AFAIK.#asahi · 2025-08-04 21:58
tpw_rulesdid rockchip contribute at all?#asahi · 2025-08-04 21:57
tpw_rulesi thought it was still all shitty blobs, interesting that it has been upstreamed#asahi · 2025-08-04 21:57
tpw_rulesoh interesting, that must be a new thing#asahi · 2025-08-04 21:56
linkmauveThe kernel API is just a bunch of ioctl()s for submitting work.#asahi · 2025-08-04 21:56
linkmauvetpw_rules, the userland API? It’s Teflon, part of Mesa, which implements the tflite API AFAIK.#asahi · 2025-08-04 21:55
tpw_ruleslinkmauve: what is the api for the NPU? do you have docs?#asahi · 2025-08-04 14:03
linkmauveWell, targeting 6.18 rather than 6.17 for this one.#asahi · 2025-08-04 14:03
linkmauveThey recently upstreamed support for the NPU in both the kernel and Mesa.#asahi · 2025-08-04 14:03
linkmauveMainline is still missing encoding video using V4L2 stateless, and the JPEG decoder, but aside from that it provides everything I need for my rk3588 board.#asahi · 2025-08-04 14:02
chaos_princessi wish the frankenkernel was useful as more than hw reference, on my rk3588 board it can't do nvme, and upstream kernel was missing a bunch of stuff the last time i tried it#asahi · 2025-08-04 14:01
linkmauveYeah, following the TRM in addition to the existing kernel or userland driver is usually useful.#asahi · 2025-08-04 14:00
jannauthe frankenkernel is at least some HW documentation although certainly partly misleading/wrong#asahi · 2025-08-04 14:00
chaos_princesslol#asahi · 2025-08-04 14:00
linkmauveI’m partial to at least having example code available, even if the quality is terrible. Also fewer firmware required (if possible none) is nicer usually.#asahi · 2025-08-04 14:00
chadmedive not encountered a single BSP that isnt worse than simply having nothing so far#asahi · 2025-08-04 13:59
chaos_princessmacos gets updates at least i guess#asahi · 2025-08-04 13:58
chaos_princessand idk what is worse, the apple approach of "here is macos, you are on your own", or the sbc vendors' "here is some linux frankenkernel, go and suffer"#asahi · 2025-08-04 13:58
chadmedcompiling firefox on this m2 macbook air takes less time than doing so on my 5600x machine#asahi · 2025-08-04 13:57
linkmauveIf Apple sbcs were a bit cheaper I’d probably get one, but even second hand they still cost at least 400€ here.#asahi · 2025-08-04 13:57
chadmedthese machines arent shit, and can do more than run a digital photo frame for starters#asahi · 2025-08-04 13:57
linkmauveI happen to only have AllWinner, Rockchip, Qualcomm and Nintendo SoCs available here.#asahi · 2025-08-04 13:56
chaos_princessthe fact that the performance does not suck for starters#asahi · 2025-08-04 13:56
linkmauveThere are probably many, but which one do you want to point?#asahi · 2025-08-04 13:55
chaos_princessok, but there is a difference between a m1 and a raspi or a rk3588#asahi · 2025-08-04 13:55
linkmauvechaos_princess, well, you’re in a channel dedicated to Apple’s ARM sbcs. :p#asahi · 2025-08-04 13:54
chaos_princessi've stopped wasting my time with arm sbcs by that point, so didn't watch that space#asahi · 2025-08-04 13:54
linkmauveDo you know anyone still using a Firefox version older than two years?#asahi · 2025-08-04 13:54
chaos_princess2 years ago - yea, that is relatively recently :P#asahi · 2025-08-04 13:54
chadmedwhich is funny because motherboard vendors are increasingly pushing out utterly useless and abysmally dogshit smbios implementations that are themselves effectively useless#asahi · 2025-08-04 13:53
linkmauvechaos_princess, according to https://bugzilla.mozilla.org/show_bug.cgi?id=1833354 it seems to also support the stateful V4L2 API.#asahi · 2025-08-04 13:53
chadmedi had to teach kfsolid how to deal with devices that dont have dmi because it never occurred to anyone that some devices with plasma simply may not implement smbios#asahi · 2025-08-04 13:52
chadmedno one did it because no one cares. like, we really are the first movers for 99% of this stuff#asahi · 2025-08-04 13:52
chaos_princessnope#asahi · 2025-08-04 13:52
aruizchadmed, the easiest way to compile an EFI binary on linux using GCC is with mingw#asahi · 2025-08-04 13:51
linkmauvechaos_princess, and no one added support for Vulkan video or V4L2 in all these years?#asahi · 2025-08-04 13:51
aruizchadmed, and the toolchain is annoying#asahi · 2025-08-04 13:51
aruizchadmed, it is ironic and funny, yes#asahi · 2025-08-04 13:51
chaos_princesslinkmauve: eh, maybe? the issue with v4l is pretty simple, firefox only supports vaapi#asahi · 2025-08-04 13:51
chadmedaruiz: i know its not anything of consequence lmfao, i literally mean its kind of funny in the sense that i get a laugh out of thinking about it#asahi · 2025-08-04 13:50
aruizand uboot's UEFI implementation is actually lean and neglectable unlike EDKII#asahi · 2025-08-04 13:50
linkmauveIt would map very badly to Vulkan video from my understanding though.#asahi · 2025-08-04 13:50
aruizchadmed, a windows PE is just a struct, I am not too bothered by that, UKI is rather useful, at Red Hat we are building a Chromebook like A/B boot schema using UKI, similar to the Android A/B boot protocol too#asahi · 2025-08-04 13:50
linkmauvechaos_princess, oh hmm, would the stateful API work better for you?#asahi · 2025-08-04 13:50
linkmauveI’m currently developing against Rockchip (actually Hantro) and AllWinner SoCs which both already expose some of their decoders to V4L2.#asahi · 2025-08-04 13:49
chaos_princessand our hw is not very stateless#asahi · 2025-08-04 13:49
chaos_princessneat. #asahi · 2025-08-04 13:49
linkmauvechaos_princess, chadmed, I happen to be working on a Vulkan video driver which delegates to V4L2 stateless atm. :)#asahi · 2025-08-04 13:48
chaos_princessand by vaapi or vulkan video i mean just vaapi, as this is what everything supports#asahi · 2025-08-04 13:48
aruizthe guests use grub