RADV: A Community Open-Source Effort To Get Vulkan Working On Radeon

[bridgman]

Wtf? AMD said multiple times (or maybe it was just bridgman) that they would open source it after about a year.

Yeah, there were some raised eyebrows about the "no communication" part - Alex, Jammy and I have all talked about it.

That said, I think this is a great idea. Our Vulkan driver currently lives entirely outside of Mesa with a completely different hardware layer (proprietary Linux drivers code-share with other OSes rather than other vendors) so just open sourcing it as-is isn't enough anyways.

We need to get to the point where we are sharing HW layer code between the open Vulkan driver and Mesa GL in order to have a sustainable open source stack, so having a community driver "pulling" and a proprietary driver "pushing" seems like pretty much the ideal way to work out what the final stack should look like.

[airlied]

Meaning what? If you go for nvidia, you're not getting FOSS Vulkan support at all, if ever. If you go for Intel, to my knowledge, only Broadwell supports it.


But here's what I don't get - why not nouveau? They need all the help they can get. Instead, you're focusing on giving Vulkan support to GPUs that already have it. Meanwhile, there are other AMD GPUs that are not getting enough Vulkan support (such as Hawaii) and you chose not to support those yet. The priorities just don't make sense to me. I don't mind that you're supporting AMD for Vulkan drivers, but why not support the ones that need it most? When you bring a 3 children to an ice cream parlor, why would you buy only one kid 3 desserts, who already had one before arriving?

I appreciate your efforts, I just implore you reconsider which chips you support first. We need people like you to help out with Vulkan.

For Intel, haswell, broadwell and skylake should all have vulkan at various stages.

Why not nouveau? I don't have the hw knowledge there at a level I could tackle this sort of thing, nvidia hw is quite different and I could spend a lot of time on it.

I also just started on the card that was in my desktop at the time, which was a Tonga, I really didn't spend time on a decision before hand. I don't think bringing up basic support on the older GPUs will take that long, in a possibly reduced featureset. I'd rather do that once I had a bit better support for VI, and more experience with vulkan.

Dave.

[haagch]

Ivy Bridge and Haswell partly work, Broadwell and Skylake should be ready, at least for the features intel has implemented so far. E.g. afaik no tessellation shaders at all.

Hawaii not getting enough Vulkan support? It's 2nd gen GCN ("1.1"), so at least it has some Vulkan support. pitcairn, tahiti, cape verde, oland got ZERO Vulkan support so far.

[entropy]

Thanks for your excellent work Dave!
This is all very appreciated!

[bridgman]

Hawaii not getting enough Vulkan support? It's 2nd gen GCN ("1.1"), so at least it has some Vulkan support. pitcairn, tahiti, cape verde, oland got ZERO Vulkan support so far.

I know you get the distinction but not everyone will - Pitcairn, Tahiti, Cape Verde and Oland are supported in the Vulkan driver today, what's missing is user-ready amdgpu kernel driver support.

[Awesomeness]

Wtf? AMD said multiple times (or maybe it was just bridgman) that they would open source it after about a year.

I don't trust such announcements until I see the actual code release.

[raytracer]

This is truly exciting time in the mesa world. I've been waiting to see how Gallium based drivers would tackle Vulkan support and this is certainly an interesting path. Wonder if there are other pieces outside of NIR->LLMV IR that can be shared with other hardware (not forgetting SPIR-V parser, SPIR-V->NIR work done by intel). Huge thanks to Dave and Bass for spread-heading this effort!

[bridgman]

I don't trust such announcements until I see the actual code release.

Nor should you... but "not trusting" is not the same as saying "no communication".

[geearf]

We need to get to the point where we are sharing HW layer code between the open Vulkan driver and Mesa GL in order to have a sustainable open source stack, so having a community driver "pulling" and a proprietary driver "pushing" seems like pretty much the ideal way to work out what the final stack should look like.

I was thinking about that when I saw Dave's blog, and so I am happy to see someone from AMD agreeing

As for the rest, thank you Dave for sharing whatever you're working on, it's always appreciated!

[microcode]

Meaning what? If you go for nvidia, you're not getting FOSS Vulkan support at all, if ever. If you go for Intel, to my knowledge, only Broadwell supports it.


But here's what I don't get - why not nouveau? They need all the help they can get. Instead, you're focusing on giving Vulkan support to GPUs that already have it. Meanwhile, there are other AMD GPUs that are not getting enough Vulkan support (such as Hawaii) and you chose not to support those yet. The priorities just don't make sense to me. I don't mind that you're supporting AMD for Vulkan drivers, but why not support the ones that need it most? When you bring a 3 children to an ice cream parlor, why would you buy only one kid 3 desserts, who already had one before arriving?

I appreciate your efforts, I just implore you reconsider which chips you support first. We need people like you to help out with Vulkan.

The GPUs need to be fast enough that games are regularly CPU-bound, for it to matter. Seems reasonable to implement Vulkan on recent-gen GPUs in my mind.I also suspect it was just what was in airlied's workstation. He doesn't work for you, I don't see why he should strive so much to make his work initially useful for you.