back again with some more gpu stuff !
here is the radeon R5 340X working with opengl on haiku :3
That’s Insane!!! do you have a link to your driver (maybe a github repo?) and does it support polaris?
I just hope that all these AI-generated drivers will be truly polished and thoroughly tested.
And I’m eagerly awaiting the drivers for Intel Xe—my 11th Gen Core i7 can’t wait.
The reason I haven’t put it on GitHub yet is because me (and some friends of mine) are putting probably far too much effort into declankering it lol
4 Likes
Polaris is GCN2 iirc? I don’t have a GCN2 card sadly
edit: its GCN4, woops, also dont have a GCN4 card
I’m curious. What’s your declankerization process?
manually going through the code and refactoring/rewriting basically all of it lol
3 Likes
Whats the newset gpu this supports?
right now it only really works on the mullins iGPU in a thin client i have, and on the oland r5 340x i had laying around
i do have another thin client with a vega 8 though, could try adding that but that would maybe be a bit on the harder side of things lol
2 Likes
I have a laptop with a Vega 8 iGPU. But it’s not a Vega 8, it’s an RX Vega 8—a Radeon RX Vega 8 in an AMD Ryzen 7730U processor
the thing i have has a Ryzen Embedded V1756B cpu in it, might be similar enough graphics wise?
i don’t have anything to test integrated display output though, they normally work a bit differently, havent looked much at the vega 8 yet though, just got it
from the bit i know though, i think they use atombios less, so its probably gonna be harder
Nice! if you post it on gh i can get a r5 340 for like 5 bucks and get it running on my optiplex 780 lol
also, what AI models did you use, as I have a couple chipsets i would like to speedrun writing drivers for! 
theres a huge amount of different oland cards that are kind just rebrands, lol
1 Like
as for the ai models thing, i could say i used codex and such but the thing is, i dont want to encourage people vibecoding drivers, especially if they are for anything that could even slightly cause things such as data loss/corruption or other such damage
drivers are a weird thing, in that they can be both shockingly easy/simple, and surprisingly complex at the same time/in different areas. ill assume you know this, but ai models cant really thing, they have no strong concept of code quality, or of exactly what needs to be checked and what doesnt need to be checked, etc, the only reason a model was anything close to useful for this was because it could look at what amdgpu did and basically do that too. ive said this before in my previous thread, im pretty sure if amd released actual gpu register documentation and such that a halfways decent programmer, assuming knowledge on the relevant parts in the target kernel, could probably get a driver like this done quicker than someone using ai for it if you include the time needed to clean the code up.
and theres a lot of code to cleanup here, if there wasnt much, id have put it on github already. perhaps sadly, i have too much self respect to put purely clankered stuff on the internet like this.
5 Likes
Done. 
I’ll digress from that side topic for now.
But this AMDGPU-variant driver is quite amazing to see as I notice use of R!b6’s default Mesa and not the newer graphics architecture.
Declank away. I agree with properly vetting these things…
I have an R9 380 Tonga and an rx550 and rx580(two of them) the others are very old Radeon hd 5770 and Radeon hd 5670. At least rx550 and 580 maybe can be supported?
Why are you asking me this? I don’t know
I don’t have any Polaris cards, when I’m comfy releasing things you can try adding support for it yourself, I’d absolutely appreciate it :3
1 Like
I would love to!! Could you maybe send a link to the non-cleaned up source code so I can start experimenting! : )
1 Like