10:09 <alyssa> Retr0id: going to guess we run out of registers

10:10 <alyssa> that blows up an assert on debug builds but asserts are compiled out of release builds

10:22 <Retr0id> ahh

10:53 <alyssa> Retr0id: https://gitlab.freedesktop.org/asahi/mesa/-/merge_requests/199

10:54 <Retr0id> 🙏

11:04 <alyssa> Retr0id: incidentally that shader toy seems to work on my dev build

11:04 <alyssa> using exactly the whole reg file no spill

11:04 <alyssa> 16.9fps on m1

11:04 <Retr0id> nice

11:05 <alyssa> I did a pile of scheduling work which I guess fixed stuff

11:08 <alyssa> Retr0id: 60fps in chromium

11:08 <alyssa> firefox's webgl has always been.. special

11:09 <alyssa> ait different resolutio

11:10 <alyssa> chromium needs `--use-gl=egl --in-process-gpu` for webgl2 tho

12:10 <ydalton> a question for the agxv devs: i see in agxv_instance.c a function called agxv_CreateInstance() is defined which is presumably what is called when you call vkCreateInstance in a vulkan application. my question is, how does the vulkan driver know to call that function? is there a struct that lays out function pointers to driver specific functions? similar to the file_operations struct in the linux kernel. thanks!

12:14 <alyssa> ydalton: https://indico.freedesktop.org/event/2/contributions/57/attachments/57/145/2022-10-04%20XDC%202022.pdf

12:32 <ydalton> alyssa: i think that answers my question, thanks!

12:38 <ydalton> if i understood correctly, vk_entrypoints_gen.py takes a driver prefix (agxv in this example) and it creates two C files and that's where the mapping happens, is my understanding correct?

13:59 <ella-0> yup that's correct

15:06 <i509vcb> I assume the fedora spin has a new enough abi for agxv?

15:06 <i509vcb> My arch install imploded so I moved :)

15:07 <i509vcb> (Or I'll figure out in a few hours)

15:26 <ydalton> ella-0: thanks!