14:31 <alyssa> lina: uh oh

14:31 <alyssa> I think I'm figuring out what the 02345 buffer is...

14:44 <alyssa> OK. I think I have a more or less complete picture of bindless texture handles then

14:44 <alyssa> still mystified by bindless samplers.

14:45 <alyssa> namely

14:45 <alyssa> add the register (if present) to the 64-bit uniform (must be present) and that's the address of the texture descriptor in memory, bam, done.

14:46 <alyssa> Metal's texture handles are 32-bit offsets into the heap or something like that

14:46 <alyssa> byte offsets

14:46 <alyssa> multiples of 24

14:46 <alyssa> as for the samplers though

14:46 <alyssa> the sampler register is a 16-bit index, not a byte offset (!)

14:47 <alyssa> but it isn't at all clear to me where the sampler descriptors themselves live. presumably this is indexing into a global heap but I can't figure out where.

15:15 <alyssa> i feel like i'm getting warmer but this isn't clicking yet

15:17 <alyssa> either my Metal code is wrong or something is horribly broken in my decoder ..

15:29 <alyssa> I'm clearly missing something dumb

15:59 <alyssa> but, if changing the sampler descriptor is not changing memory as far as I can see, IDK what to make of that other than not mapping memory hard enough

16:06 <alyssa> IOGPU Compute word 38: # of bindless samplers + 1

16:06 <alyssa> IOGPU Compute word 39: # of bindless samplers

16:06 <alyssa> ??

16:06 <alyssa> (seeing as they each bump up when I create a bindless sampler even if it's not actually used)

16:06 <alyssa> and corresponding index in the 02345 buffer gets the second word replaced by 0x780 or 0x781

16:07 <alyssa> but that, doesn't tell me where the actual samplers live!

16:07 <alyssa> 02345 is the only interesting thing here but, I'm printing it and it's not telling me what I need

16:10 <alyssa> 02345 is clearly a table of 8-byte records of some kind

16:10 <alyssa> relating to the bindless samplers

16:10 <alyssa> but .. I'm not seeing where the cntents go

16:10 <alyssa> there'sjust, not enough entropy here

16:11 <alyssa> unless.

16:12 <alyssa> no this would be pathological even by AGX standards..

16:13 <alyssa> no it's not that either.

16:20 <alyssa> at a loss here, where is this missing memory

16:22 <alyssa> managed to provoke 0x783

18:42 <eric_engestrom> marcan (lina, alyssa): for the manjaro/alarm snafu with enabling asahi in the can't-work-upstream package, this reminds me that I was surprised by the choice to rename the mesa package in the asahi repo instead of naming it `mesa` so that it would shadow the one from alarm; perhaps this should still be done now to avoid this kind of issues in the future? (although I can't think of something else they could do)

23:31 <jannau> eric_engestrom: I suppose the idea was to make it optional and easy to revert to the upstream version in the case we screw up