#asahi-dev · 2026-01-28 11:49
integralpilotokay i've figured out for OpenGL 1.0, we only need to decode about 14 of the instructions (if i'm understanding what i'm reading correctly). i'm working on those 14 now.#asahi-dev · 2026-01-28 11:46
jannauyou could let the kernel handle the initramfs decompression. see m1n1_initramfs in run_guest_kernel.sh#asahi-dev · 2026-01-28 11:41
nickchanbut for the initramfs that don't work right i think#asahi-dev · 2026-01-28 11:41
nickchanokay xz -T1 --lzma2=preset=6e,lc=3,pb=2,lp=0 can work for the kernel#asahi-dev · 2026-01-28 11:26
nickchannow i assume the last one is lp = 0#asahi-dev · 2026-01-28 11:25
nickchansooo lc must be 3 and 0 at the same time#asahi-dev · 2026-01-28 11:25
nickchan> The LZMA2 property byte must indicate the LZMA properties lc = 3, pb = 2, lc = 0#asahi-dev · 2026-01-28 11:24
jannauI could imagine that xzutils now defaults to multi-threaded compression. try disable that#asahi-dev · 2026-01-28 11:23
integralpilotokay so there's a new mystery byte in the mov instruction, will experiment and try to find what it means#asahi-dev · 2026-01-28 11:22
jannaurestrictions from minlzma: https://github.com/ionescu007/minlzma?tab=readme-ov-file#limitations-and-restrictions#asahi-dev · 2026-01-28 11:18
jannaunickchan: xz support might have bitrotted since it's not that useful with usb device support as speedups from the smaller size are eaten by slower compress/decompress time. I'd look at marcan's streams when it was added (or just the commit messages)#asahi-dev · 2026-01-28 11:17
sveni just have EARLY_DOCKCHANNEL_UART_BASE and if you define both it'll just push stuff to both uarts#asahi-dev · 2026-01-28 11:16
svenit's really just a few changes#asahi-dev · 2026-01-28 11:16
svenyup#asahi-dev · 2026-01-28 11:15
jannauthat should be already in place for the usb device support. just missing a driver for the dockchannel uart and possibly special handling for EARLY_UART_BASE#asahi-dev · 2026-01-28 11:12
sveni have local changes for all that already#asahi-dev · 2026-01-28 11:12
svenwell, yeah#asahi-dev · 2026-01-28 11:12
nickchanwould need to modify m1n1 to support multiple uart ports too#asahi-dev · 2026-01-28 11:11
noopwafelsven: <3#asahi-dev · 2026-01-28 11:11
nickchanthe Kanzi In System protocol#asahi-dev · 2026-01-28 11:11
svens/machine/mac/#asahi-dev · 2026-01-28 11:11
svenshould make it easier for people to get hardware uart then and drop the requirement for special hardware or a second machine #asahi-dev · 2026-01-28 11:10
svenmy usb2 sniffer only arrives on saturday and i'll already be at FOSDEM then. but hopefully next week i can just figure out that debugusb / dockchannel UART protocol#asahi-dev · 2026-01-28 11:09
noopwafelI thought about wrapping something in the samsung console debug define but ofc gregkh got rid of that too, and that patch seems the simplest path forward..#asahi-dev · 2026-01-28 11:07
integralpilot:D#asahi-dev · 2026-01-28 11:07
nickchan:)#asahi-dev · 2026-01-28 11:07
noopwafelhttps://noopwafel.net/asahi/0001-tty-serial-samsung_tty-avoid-dev_dbg-deadlock.patch <- also, siiiigh#asahi-dev · 2026-01-28 11:06
svenstill, m3 gpu is gonna be orders of magnitude more complex than getting the rest of drivers to work on m3 and writing device trees#asahi-dev · 2026-01-28 11:06
noopwafelhonestly I didn't look into the ISA side much because the fw is a more immediate .. fun task, but that tree at least has a good start on the ISA#asahi-dev · 2026-01-28 11:06
integralpilotsven: oh i was working off the main applegpu upstream, i'll take a look at that, thanks!#asahi-dev · 2026-01-28 11:05
nickchanerm, I mean xz -ze6T0 initramfs#asahi-dev · 2026-01-28 11:05
svenhttps://github.com/TellowKrinkle/applegpu/tree/M3#asahi-dev · 2026-01-28 11:05
nickchanneither does xz -zeT0 initramfs#asahi-dev · 2026-01-28 11:05
nickchanIn particular Linux's Image.xz does not work#asahi-dev · 2026-01-28 11:05
noopwafelyou did check tellowkrinkle's branch right? (for the m3 isa)#asahi-dev · 2026-01-28 11:04
integralpilot(about some things the same, not about xz not working)#asahi-dev · 2026-01-28 11:04
integralpilotoh that's good#asahi-dev · 2026-01-28 11:04
nickchanhaven't found any examples of m1n1's xz ever working#asahi-dev · 2026-01-28 11:03
noopwafelthere's still a lot of things which are the same#asahi-dev · 2026-01-28 11:03
svenand we don't want to rerwrite the entire userspace, we want to adapt the current one to support both the old and new instruction sets#asahi-dev · 2026-01-28 11:02
svenyeah, we've know that. that's why it's such a big task#asahi-dev · 2026-01-28 11:01
integralpilottrying to do bitops like M1 causes a GPU page fault, and trying to do many other things like M1 result in outputs that aren't right in any way#asahi-dev · 2026-01-28 11:00
integralpilotoh no after doing some looking into it, it seems the M3 GPU ISA has changed signficantly, to the point we might need to make a new compiler rather than modifying honeykrisp#asahi-dev · 2026-01-28 10:36
fl0_idcries in no cpufreq nickchan ;)#asahi-dev · 2026-01-28 10:30
integralpilot:D#asahi-dev · 2026-01-28 10:05
noopwafeloof :-)#asahi-dev · 2026-01-28 09:50
nickchanbut is not right at all#asahi-dev · 2026-01-28 09:50
nickchanso it *worked*#asahi-dev · 2026-01-28 09:49
nickchanand the two interface happens to be compatible enough if size/pa is not out of range for SART0