Porting Of Changes/Fixes From AMDVLK To RADV Vulkan Driver Begins

[pal666]

RADV might be developed by 2 people

it might, if you live in some alternative universe

[Awesomeness]

RADV might be developed by 2 people, but Dave and Bas put so much work into it. Thanks guys

No disrespect against these two but if you cared to look at radv copyright headers, eg. https://github.com/mesa3d/mesa/blob/...dv_meta_copy.c, you'd see that radv is a port of Intel's Mesa Vulkan driver.

[airlied]

No disrespect against these two but if you cared to look at radv copyright headers, eg. https://github.com/mesa3d/mesa/blob/...dv_meta_copy.c, you'd see that radv is a port of Intel's Mesa Vulkan driver.

You can't really "port" another hw vendors vulkan driver. anv was a template for radv, we removed all the intel code and filled all the blanks in with amd code.

In saying that, myself and Bas are just adding the final vulkan step on top a bunch of other work, the AMD addrlib, llvm compiler, the Mesa SPIR-V->NIR, all of NIR, lots of refactored code out of radeonsi.

However a lot of stuff is two way, we did a lot of the NIR->LLVM Validation that allows ARB_gl_spirv to be a thing on radeonsi, and allows for NIR to be used instead of TGSI in the future, we've added features and fixed bugs in the other bits.

It's hard to say how many "radv" developers there are, features developed for other drivers help radv, features developed for radv help other drivers. I suppose you can say there are two radv maintainers, a number of part-time direct to radv contributors, and a number of indirect contributors.

Dave.

[Sethox]

I thought that the point of open source was the availability to help to develop the code to evolve, a "crowd resource" kind of thing. Of course you have to have some notoriety to bring "valid" code etc etc since some projects are more important than others (professional is the word I am looking for).

Or we all be saying that Linus Torvalds is the guy to credit to ALL of this thanks to git and the platform (Linux) if we are following the logic of who started the project.


Back on track, I still thank for this news and RADV. It's a project worth reading and following !

[shmerl]

Great! Hopefully drivers can benefit each other in both directions.

[nhaehnle]

I believe both closed and open generate LLVM IR from SPIR-V, however the usage from that point on is different - the closed driver generates proprietary IR from LLVM IR then uses our proprietary shader compiler to generate HW instructions, while the open driver uses the LLVM IR direct-to-ISA back end also used in radeonsi and HCC/ROCm (and radv IIRC).

The closed source shader compiler doesn't use LLVM IR at any point. It goes straight from SPIR-V to AMD IL, which is then translated to HW instructions.

[oooverclocker]

Among his questions were how the AMDVLK development process will be handled at AMD, how AMD will be engaging with the open-source community, the development process for new Vulkan functionality, etc.

airlied
Many thanks for taking the initiative for a constructive communication about the future of the projects before acting!

@AMD-developers
Please take it very seriously to find the best possible way from the perspective of your customer who buys a PC with your GPU and installs a consumer distribution! No matter whether the result is a close collaboration between RADV and AMDVLK that respects the structures of both drivers or a hybrid driver fitting in all environments.
The ability to run a Linux machine with AMD hardware initially without losing performance compared to Windows is very important!

[darkbasic]

It's hard to say how many "radv" developers there are, features developed for other drivers help radv, features developed for radv help other drivers. I suppose you can say there are two radv maintainers, a number of part-time direct to radv contributors, and a number of indirect contributors.

Dave.

Please don't stop developing radv unless AMD will prove to develop their Vulkan driver in the open: we don't need another "Android".

[Awesomeness]

You can't really "port" another hw vendors vulkan driver. anv was a template for radv, we removed all the intel code

The copyright headers don't reflect that. Maybe you should adapt those to the reality.

[cruelj]

I'm really curious how AMD will convince many people to use AMDVLK, because if nobody uses it, why have it?
I assume most distros will only pack RADV and it's not like back in the day anymore, when the first thing to do after setting up a new system was to install the proprietary driver...
I'm pretty sure nobody of the open source guys at AMD is happy about this situation and i would love to know what the higher-ups who made the decision to cause this situation plan to solve it, if they even know or care...