#
✕ nick:JamesCalligeros[m]
#asahi-dev · 2026-09-04 09:42
JamesCalligeros[m]There are probably more neos kicking about than m4 pros
#asahi-dev · 2026-09-04 09:22
JamesCalligeros[m]Havent tested it in a while but it should be possible.
#asahi-dev · 2026-09-03 03:08
JamesCalligeros[m]read the m1n1 user/dev guide again
#asahi-dev · 2026-09-03 03:08
JamesCalligeros[m]youre not supposed to run m1n1 like that at all, and please stop dumping terminal output like that
#asahi-dev · 2026-09-02 04:51
JamesCalligeros[m]what did the llm find (broad strokes), if you dont mind me asking
#asahi-dev · 2026-09-02 04:46
JamesCalligeros[m]theres a reason that e.g. the wine folks forbade anyone from looking at the nt 5.3 source leak even if they werent planning on contributing code
#asahi-dev · 2026-09-02 04:46
JamesCalligeros[m]two-teaming works when the stuff team A is looking at isnt outright illegal to look at
#asahi-dev · 2026-09-02 04:45
JamesCalligeros[m]and even given all of this, we still get LLM-assisted contribution attempts from time to time, and every single one of them is majorly incorrect
#asahi-dev · 2026-09-02 04:44
JamesCalligeros[m]so even if you were to "manually" rewrite its findings or whatever its immaterial. the findings themselves are tainted and cannot be used
#asahi-dev · 2026-09-02 04:44
JamesCalligeros[m]in fact it is our understanding that some attempted uses of LLMs for this purpose have already done exactly that
#asahi-dev · 2026-09-02 04:43
JamesCalligeros[m]we know there is a great volume of this stuff available very easily, on sites commonly scraped by LLM crawler bots (e.g. github) and that it is all but certain that it has been picked up into training sets
#asahi-dev · 2026-09-02 04:42
JamesCalligeros[m]the issue is, as explained in our policy document, that we cannot guarantee that the LLM will not reference training material the we absolutely must not ever reference. this includes things like leaked apple documentation, leaked apple internal code, etc
#asahi-dev · 2026-09-02 04:38
JamesCalligeros[m]and fwiw no one who has tried to unleash agentic coding tools on "clean room documentation" has managed to succeed by any metric whatsoever as of yet. all they manage to do is produce marginally-less-tainted-but-tainted-nonetheless asm2diagram or asm2english or asm2pseudocode
#asahi-dev · 2026-09-02 04:37
JamesCalligeros[m]we cannot risk it, not least because apple are extremely unhappy with certain LLM companies right now
#asahi-dev · 2026-09-02 04:36
JamesCalligeros[m]what part of "absolutely forbidden" is unclear lol
#asahi-dev · 2026-09-01 22:30
JamesCalligeros[m]question now is can i have the audio branch cleaned up and submitted on list for 7.6
#asahi-dev · 2026-09-01 22:29
JamesCalligeros[m]yay shared gpio went in!
#asahi-dev · 2026-09-01 12:53
JamesCalligeros[m]git format-patch fairydust..asahi-wip yields no diff?
#asahi-dev · 2026-09-01 12:05
JamesCalligeros[m]its getting late here
#asahi-dev · 2026-09-01 12:05
JamesCalligeros[m]sure, i can look tomorrow
#asahi-dev · 2026-09-01 11:58
JamesCalligeros[m]does the hexdump include the compression buffer or is that just the image data region
#asahi-dev · 2026-09-01 11:49
JamesCalligeros[m]im not sure. did setting that flag change nothing for 8 bit images?
#asahi-dev · 2026-09-01 11:46
JamesCalligeros[m]try decoding one using the reverse of what's in AIL :P
#asahi-dev · 2026-09-01 11:45
JamesCalligeros[m]i would be surprised but not that surprised if avd can spit out AGX tiled/compressed buffers
#asahi-dev · 2026-09-01 11:45
JamesCalligeros[m]im not familiar with how that looks in hex but i was thinking the same thing
#asahi-dev · 2026-09-01 11:35
JamesCalligeros[m]can you dump 16 bits of luma? unless the values just repeat those 8
#asahi-dev · 2026-09-01 11:12
JamesCalligeros[m]p010 = y:x, cr:x, cb:x [10:6, 10:6, 10:6]
#asahi-dev · 2026-09-01 11:12
JamesCalligeros[m]p012 = y:x, cr:x, cb:x [12:4, 12:4, 12:4]
#asahi-dev · 2026-09-01 11:11
JamesCalligeros[m]yeah
#asahi-dev · 2026-09-01 11:08
JamesCalligeros[m]but i think thats the reverse of the issue irght
#asahi-dev · 2026-09-01 11:08
JamesCalligeros[m]so if whatever is expecting p012 is getting p010 you will read into 2 padding bits
#asahi-dev · 2026-09-01 11:06
JamesCalligeros[m]p012 has 4 padding bits on the chroma plane, p010 has 6
#asahi-dev · 2026-09-01 10:59
JamesCalligeros[m]it's not p012?
#asahi-dev · 2026-09-01 10:55
JamesCalligeros[m]nv24? that's the 4:4:4 one
#asahi-dev · 2026-08-31 21:11
JamesCalligeros[m]B*
#asahi-dev · 2026-08-31 21:11
JamesCalligeros[m]sofus: it should be 128b aligned aiui
#asahi-dev · 2026-08-31 09:39
JamesCalligeros[m]I dont expect a v2 to be required but ofc i will fix it if it is
#asahi-dev · 2026-08-31 09:39
JamesCalligeros[m]Lol sashiko caught it too
#asahi-dev · 2026-08-30 11:13
JamesCalligeros[m]the only thing i can see that we dont set is unk_1
#asahi-dev · 2026-08-30 11:08
JamesCalligeros[m]actually no the UV is going to be 2 bytes isnt it
#asahi-dev · 2026-08-30 11:01
JamesCalligeros[m]dont mind me it's getting late :P
#asahi-dev · 2026-08-30 11:01
JamesCalligeros[m]oh yes
#asahi-dev · 2026-08-30 11:00
JamesCalligeros[m]it looks like the tile_bytes field of the compression info is the same
#asahi-dev · 2026-08-30 10:59
JamesCalligeros[m](youll need to guard that behind an if (yuv) otherwise RGB images will probably break
#asahi-dev · 2026-08-30 10:59
JamesCalligeros[m]did you try making tile_size just tile_dim * tile_dim instead of tile_dim * tile_dim * bytes per pix?
#asahi-dev · 2026-08-30 10:57
JamesCalligeros[m]what did you make .data_offset?
#asahi-dev · 2026-08-30 10:53
JamesCalligeros[m]oh yeah so we do, that wont work for planar formats lol
#asahi-dev · 2026-08-30 10:53
JamesCalligeros[m]that looks fine
#asahi-dev · 2026-08-30 10:52
JamesCalligeros[m]it is for the first plane
#asahi-dev · 2026-08-30 10:51
JamesCalligeros[m]the only thing i can see that is sussy is that tile_size is calculated without reference to the bit depth of the image here

plus ›