search

sultim-t / RayTracedGL1

MIT License
128 stars 29 forks source link

Fix ssbo binding count mismatch (AMD RDNA fix) #10

Closed kd-11 closed 2 years ago

kd-11 commented 2 years ago

The mismatch caused the pool to not have enough ssbo descriptors to allocate for the second frame. Allows RDNA2 to launch albeit with some minor corruption that I have not looked into. image

Issue: https://github.com/sultim-t/prboom-plus-rt/issues/8

kd-11 commented 2 years ago

PS. I was unsure which branch to commit to as there is no branch named "main" or "master" so I just used the default, let me know if I need to change it.

sultim-t commented 2 years ago

Woah! That's unbelievably cool! I will compile and post the new version of PrBoom mentioning you :)

kd-11 commented 2 years ago

Its still a bit crashy on AMD, I haven't had too much time to look into it, I'll do a deeper dive tomorrow and try to knock down a few big issues.

sultim-t commented 2 years ago

Maybe just call it AMD Preview version, heh... Or maybe I should better hold the release, since it still crashes? What do you think? How frequently the crashes occur?

kd-11 commented 2 years ago

Preview version is fine, though I do not know how unstable it is. Sometimes it crashes in the menus, sometimes after 4 levels, sometimes after only 1. Maybe its good to have something out there to gather feedback, but I also think I can fix a lot more problems in the next day or two. I say we wait another day so I have some time to poke at it. As for the graphics glitch, I haven't investigated yet, but I'm 99% sure its just NaN poisoning, and I can confirm disabling bloom isolates it to random pixels in the top region of the screen.

sultim-t commented 2 years ago

That's extremely nice, thank you! I will hold it for now, there's no time pressure anyway (and we are not paid 😅), making it work on AMD is a huge task!