#asahi-dev · 2026-08-08 12:09
mmediouni[m]Arm has two important and conflicting roles for CPUs: being a standards body for an ISA, and providing first party cores. the second has the advantage that architecture licensees cannot license out their cores (hence why they got really mad at the nuvia situation) or it'd have a much harder time surviving#asahi-dev · 2026-08-08 12:07
NickChan[m]another solution would be to buy ARM but regulators wojld probably stop that#asahi-dev · 2026-08-08 12:07
chaos_princessexcept apple's license is closer to "i can do whatever i want"#asahi-dev · 2026-08-08 12:05
Sobek[m]Well, if that’s a reasonable assumption la e to try out RISC V, probably a good idea. If you want to bolt a bunch of custom vendor instructions, it might be easier than ARM, and avoid IP license shenanigans.#asahi-dev · 2026-08-08 12:04
Sobek[m]:o#asahi-dev · 2026-08-08 12:03
mmediouni[m]not RV64#asahi-dev · 2026-08-08 12:03
mmediouni[m]they're RV32 cores too#asahi-dev · 2026-08-08 12:03
chaos_princessdid not expect apple to do riscv instead of yet-another-chinook or sth#asahi-dev · 2026-08-08 12:02
chaos_princesscurious#asahi-dev · 2026-08-08 12:02
mmediouni[m]on the a19#asahi-dev · 2026-08-08 12:02
mmediouni[m]AVE has an arm64 primary coprocessor and a bunch of RISC-V secondaries#asahi-dev · 2026-08-08 12:01
chaos_princesswhich part is riscv now?#asahi-dev · 2026-08-08 12:01
mmediouni[m](for video)#asahi-dev · 2026-08-08 12:00
mmediouni[m]the M33 adds a custom instruction interface but apple chose to RISC-Vify the thing instead of using that#asahi-dev · 2026-08-08 11:58
chaos_princessNickChan[m]: atc has a m3#asahi-dev · 2026-08-08 11:58
mmediouni[m]but the private peripheral interface is APB only on the cortex-m3#asahi-dev · 2026-08-08 11:57
mmediouni[m]the primary interfaces can be AHB-Lite (and then bridged over to AXI or whatever outside the core)#asahi-dev · 2026-08-08 11:57
NickChan[m]there's also a "APB_RESET_N" in m1n1's atc python #asahi-dev · 2026-08-08 11:57
chaos_princessare they not on AXI?#asahi-dev · 2026-08-08 11:56
mmediouni[m]but not too relevant of a question especially in this context lol#asahi-dev · 2026-08-08 11:56
mmediouni[m]the Cortex-M private peripheral interface is APB only, so wouldn't be surprised if there's some APB here and there - especially for the Ms in the video decode part?#asahi-dev · 2026-08-08 11:53
chaos_princess(but it usually catches stupid mistakes in, like, error or teardown paths, as i actually test the normal ones)#asahi-dev · 2026-08-08 11:51
NickChan[m]"APB" do exist on apple hw i think but probably not stand for the same thing#asahi-dev · 2026-08-08 11:51
svenyeah, and it burns a lot of tokens if you try to run it yourself #asahi-dev · 2026-08-08 11:51
chaos_princessit is more that the way this specific series is structured#asahi-dev · 2026-08-08 11:51
svenfair enough #asahi-dev · 2026-08-08 11:51
chaos_princessjust annoying that i have to send it to ml to do that#asahi-dev · 2026-08-08 11:50
chaos_princessmeh, it catches stupid mistakes too#asahi-dev · 2026-08-08 11:50
sventhere’s no magic refusal uuid for google‘s models, is there? Just put that in the signature otherwise 😁#asahi-dev · 2026-08-08 11:49
svenlol#asahi-dev · 2026-08-08 11:48
chaos_princessyeah, mate, APB is a thing that exists on this hw, sure#asahi-dev · 2026-08-08 11:48
chaos_princess>> Apple Silicon APB peripherals commonly enforce strict 32-bit access widths.#asahi-dev · 2026-08-08 11:47
chaos_princessthe 32-bit thing is somewhat more legitimate as i386 builders exist and you don't want them to complain#asahi-dev · 2026-08-08 11:46
jannauat least for that we can depend on 64BIT#asahi-dev · 2026-08-08 11:46
JamesCalligeros[m]I copped it from a real person because the smc hwmon driver was not 32-bit friendly :p#asahi-dev · 2026-08-08 11:45
chaos_princessyea, i don't care tbh, just funny/annoying#asahi-dev · 2026-08-08 11:45
jannauchaos_princess: sigh, sorry. tell the clanker that ARCH_APPLE is little-endian only and COMPILE_TEST is not hardware#asahi-dev · 2026-08-08 11:43
jannauand released https://github.com/AsahiLinux/m1n1/releases/tag/v1.6.1#asahi-dev · 2026-08-08 11:43
chaos_princessjannau: one downside of your suggestion of using memcpy is that i get like 4 mails from the clanker every time that the driver won't work on BE hardware#asahi-dev · 2026-08-08 11:39
chaos_princessit has a weird tb setup by default, with all tb interfaces under a bridge#asahi-dev · 2026-08-08 11:39
fl0_idchaos_princess right. mmmh, or just network sheningans. why would macos <-> matter#asahi-dev · 2026-08-08 11:38
jannaum1n1 v1.6.1 tagged#asahi-dev · 2026-08-08 11:38
chaos_princessmaybe the fact that it is a macos<->linux connection plays into it somehow#asahi-dev · 2026-08-08 11:37
chaos_princessfl0_id: it is probably doing normal wifi/ethernet, i just have no idea how it decided to route to a ip that is assigned to a tb interface via a wifi interface#asahi-dev · 2026-08-08 11:37
sofusyeah KO is fail#asahi-dev · 2026-08-08 11:36
fl0_idjannau very possible, no idea#asahi-dev · 2026-08-08 11:35
jannauisn't that just the short form for frame data mismatch#asahi-dev · 2026-08-08 11:35
fl0_idchaos_princess so you basically invented tb over wifi? put that in the blog post...#asahi-dev · 2026-08-08 11:34
jannauafaik netflix uses film-grain. but it's not fully required. the standard is explicitly written to allow HW decoders + filmgrain generation via GPU shaders#asahi-dev · 2026-08-08 11:34
fl0_idI love how for film grain it says |KO|