07:47 asdqueerfromeu[d]: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30561 :frog_gears:
09:53 karolherbst[d]: notthatclippy[d]: do you know if GSP reacts to the low power GPIO on its own?
09:54 karolherbst[d]: https://nvidia.github.io/open-gpu-doc/DCB/DCB-4.x-Specification.html " NVVDD PSI" I think it was laptop firmwares use to indicate they run on battery? maybe there was another?
09:55 karolherbst[d]: maybe "111 HW Only Slowdown Enable" as well...
09:55 karolherbst[d]: anyway, the question is more if GSP handles that for us or if the kernel driver has to
09:56 karolherbst[d]: and if GSP handles it, there might even be something we need to setup first I guess?
10:04 notthatclippy[d]: karolherbst[d]: Sorry, still OOTO, please ask again on Monday? But from memory, it needs the kernel in the loop
10:10 karolherbst[d]: okay
10:10 karolherbst[d]: good to know
10:10 karolherbst[d]: and sorry for the ping, I forgot :ferrisUpsideDown:
12:33 triang3l[d]: gfxstrand[d]: On R600 the only way to pass a sampler is via one of the 18 per-stage sampler slots, so an embedded sampler would still have to count towards the per-stage limit
12:36 triang3l[d]: I just treat immutable samplers as regular sampler descriptors so they don't need a separate binding path, and I write them to the set in vkAllocateDescriptorSets
12:40 triang3l[d]: (though I'll be taking some advantage of immutable samplers later when I implement unnormalized coordinates… because on Evergreen they are toggled in fetch instructions, not in samplers, so for non-immutable samplers I'll have to patch shader binaries :blobcatnotlikethis:)
13:33 gfxstrand[d]: triang3l[d]: Weird. I knew there was a reason it's part of the shader source in GL. I didn't know it was still an issue for hardware that recent, though. Then again, r600's binding model is ancient. 🤷🏻‍♀️
13:38 gfxstrand[d]: On Intel we had to emulate it pre-SNB. I'm pretty sure vc4 needs emulation as well.
14:04 dadschoorse[d]: amd still has a in-shader toggle today (don't ask me why when it's possible to modify the sampler too)
14:35 triang3l[d]: gfxstrand[d]: Late TeraScale (Cayman) added the switch for forcing all coordinate components to be treated as unnormalized to the sampler (replacing the descriptor bits previously used for chroma keying — wondering what needed it, compositors?), I don't know what API demanded that though. OpenGL only had rectangle textures if I recall correctly, and Direct3D didn't have sampling with unnormalized
14:35 triang3l[d]: coordinates at all
22:08 gfxstrand[d]: Of course I'm finding bugs in the EXT_descriptor_buffer tests...