#asahi-gpu · 2026-11-17 06:16
SheJessicamy photo look pls https://l24.im/03ihV#asahi-gpu · 2026-11-07 19:29
zapritthank you#asahi-gpu · 2026-11-07 19:29
zapritin that case I shall wait patiently#asahi-gpu · 2026-11-07 19:26
fl0_idzaprit afaik there are no dev branches public, probably also because it's not rly workable yet etc#asahi-gpu · 2026-11-07 19:18
murdakahCool cool#asahi-gpu · 2026-11-07 19:18
chaos_princessmurdakah: let's go just plain asahi#asahi-gpu · 2026-11-07 19:18
murdakahAlso is there a better channel for this discussion?#asahi-gpu · 2026-11-07 19:17
murdakahI never thought about those needing co processors too. Are those microphone coprocessors on the M silicon or somewhere separate and closer to the mics? I'm a bit lost in the acronyms, is that always on processor or apple audio processor?#asahi-gpu · 2026-11-07 19:10
chaos_princessThose mics have a coproc called "LEAP", that runs some apple firmware that does decimation/pdm conversion and a few other things. Theoretically we can even upload our own fw, but, a part of the reason why mics took so long was that the previous guy was trying that, and i just went and used aop "macos-way" instead #asahi-gpu · 2026-11-07 19:05
murdakahAh cool thanks, yeah sorry I know its off topic. #asahi-gpu · 2026-11-07 19:04
chaos_princessnot quite, (also, not gpu, but w/e) - decimation is more about converting from pdm to pcm data, the normal output from those mics is extremely high frequency, but either 1 or 0, and we want human-hearable frequencies and a few more bits of amplitude#asahi-gpu · 2026-11-07 19:03
murdakahOh btw @chaos_princess, I was reading the section on decimation for the M3, is just referring to audio filtering?#asahi-gpu · 2026-11-07 19:01
chaos_princessthe link above is to the correct timestamp on the stream btw#asahi-gpu · 2026-11-07 19:01
murdakahAh thats handy, I was just about to start scrolling through it#asahi-gpu · 2026-11-07 19:00
chaos_princessxdc channel now has the talks cut up instead of one stream, you'd want this https://www.youtube.com/watch?v=nkRMGFN8lNA#asahi-gpu · 2026-11-07 18:59
murdakahthx#asahi-gpu · 2026-11-07 18:53
zaprithttps://www.youtube.com/live/TPwe53kCJuM?t=2082#asahi-gpu · 2026-11-07 18:48
murdakahCould you share the talk?#asahi-gpu · 2026-11-07 18:32
zapritAfter reading the 7.3 blog post and watching the XDC talk about M3 GPU reverse engineering, I've been inspired to poke around and see what I can learn. However after a long search around github and a very short search around the freedesktop gitlab I've not managed to find any dev branches related to the M3 GPU. Does anyone have any useful pointers to where development would be taking place?#asahi-gpu · 2026-11-07 07:38
sobek[m]👏 Bravo !!!#asahi-gpu · 2026-11-07 05:17
chadmedhttps://asahilinux.org/2026/11/progress-report-7-3/#asahi-gpu · 2026-11-03 09:31
leioI don't think people who could criticize it more heavily are really even looking at it, as to not get tainted by it.#asahi-gpu · 2026-11-03 02:32
GhostFace[m]yeah honestly i just saw that everything got grouped into opcodes without crashing agx_disasm and thought "hey, 100% recognition!"... which obviously isn't real reversing when you look at the raw numbers. definitely learned my lesson on overclaiming with llm stuff here. thanks for actually taking a look though!#asahi-gpu · 2026-11-03 02:32
GhostFace[m]haha fair catch, caught me red handed :P#asahi-gpu · 2026-11-02 17:54
svenwhile you probably could use an LLM to do proper clean room reversing it takes a lot more than just adding "do this cleanroom and make no mistakes" to the prompt#asahi-gpu · 2026-11-02 17:53
svensame reason why I just cannot take any of these "100% clean room" claims serious#asahi-gpu · 2026-11-02 17:53
sven"I'm sorry, you are right. my earlier claim of 100% instruction recognition was an overstatement. Now here's what this instruction actually means: ..." :P#asahi-gpu · 2026-11-02 17:45
tellowkrinkleWas a bit curious on what a LLM disassembler looks like so I took a peek, and "fballot_ext8 279288837046789" isn't what I would have called "100% instruction recognition". How do you even know that's what it does if you don't know any of its arguments or outputs?#asahi-gpu · 2026-11-02 12:43
GhostFace[m]Fair point, and thanks for pointing that out. Yes, LLM assistance was indeed used for code scaffolding, script generation, and drafting the reports in this exploration. I have just added an explicit AI / LLM Usage Disclosure section to the README to make this completely transparent to everyone. Appreciate the feedback!#asahi-gpu · 2026-11-02 12:38
leioSeems like an acknowledgment of LLM use; could the repo at least explicitly state so anywhere at all then?#asahi-gpu · 2026-11-02 12:35
GhostFace[m] serve as a useful reference for the Asahi developers to help accelerate M3 bring-up. Thanks again for looking into it and for all your incredible work!#asahi-gpu · 2026-11-02 12:35
GhostFace[m]<GhostFace[m]> "Hi everyone! 👋..." <- Totally understood, and I completely respect the team's decision regarding upstream acceptance. I just wanted to share the findings, decodings, and test cases in case they can#asahi-gpu · 2026-11-01 02:24
JamesCalligeros[m]https://asahilinux.org/llm-policy#asahi-gpu · 2026-11-01 02:18
GhostFace[m]Hi everyone! 👋... (full message at <https://matrix.org/oftc/media/v1/media/download/AeY0wfn2N5wt9K9zZlmeDIVMW6dtDzrtJmXAHgK2S-KRW_4Web_5cWCWkuX7KtY5WAXCPyS3nJXlMeNYiuIOCtBCeh3P-BtQAG1hdHJpeC5vcmcvUHVrdWZaRWZyU0h0YVVRQ0JOYnVqSERT>)#asahi-gpu · 2026-10-15 22:55
j_rustaceanhi I'm a beginner rustacean looking to contribute to the asahi-linux gpu drivers on the m3 mac. where and how do i get started :). Kindly Advise#asahi-gpu · 2026-10-07 22:14
MatveyAleksandrov[m]1hi! i have an M3 Pro 14 MBP (MRX33). if anyone needs help with testing, feel free to ping me#asahi-gpu · 2026-10-01 22:02
MichaelReeves[m]sure! what do you need me to do?#asahi-gpu · 2026-10-01 15:39
sll00I have some metal code which would answer all my questions, if someone has an M3 and a few minutes time to spare. #asahi-gpu · 2026-10-01 15:27
sll00sorry, not SFU, not sure what to call it on silicon really. TA and vertex I guess#asahi-gpu · 2026-10-01 15:15
sll00does anyone know how much the triangle coefficient setup SFU differs between M2 and M3?#asahi-gpu · 2026-10-01 14:53
sobek[m](Though this has nothing to do with Asahi, unless we can show there's a performance regression on the firmware side ?)#asahi-gpu · 2026-10-01 14:52
TheLinkso not a fan either#asahi-gpu · 2026-10-01 14:52
sobek[m]Same for me !#asahi-gpu · 2026-10-01 14:51
TheLinkon macos the fullscreen window menu bar animation is slow and not smooth for me#asahi-gpu · 2026-10-01 05:49
ffxtested on m1pro#asahi-gpu · 2026-10-01 05:49
ffxhi, the current asahi remix 45 makes smooth animation in the browsers even worse. gpu power management seems to be even more aggressive. I think a butter smooth scrolling should be the normal at any power profile. now the scrolling even stutters at the high performance profile.#asahi-gpu · 2026-09-27 17:56
myriaNot sure if it's there though#asahi-gpu · 2026-09-27 17:55
myriamaybe try here? https://dougallj.github.io/applegpu/docs.html#asahi-gpu · 2026-09-27 17:53
janoalyssa: happy to try to help with that if we know how the instruction works#asahi-gpu · 2026-09-27 16:59
alyssaI thought dougallj already r/e'd those instructions, just needs wiring up (which is seriously nontrivial)