RADV Vulkan Driver Has Patches For Radeon RX Vega Support

[ObiWan]

From what I recall, AMD is too committed at this point to change their course. It's pretty much a race to see who finishes first - Dave's open-source drivers or AMD open-sourcing their drivers.

So AMD lawyers vs independent developer.

My bet would be on the developer

[agd5f]

Can AMD scrap the open sourcing of their Vulkan driver and put these resources to something else. And give dave a big sum of moneyz. Or beer. Or cheeseburgers. or whatever he wants for his awesome work.

As I've said a number of times, there is not a Linux vulkan team at AMD. There is just a vulkan team. We still have to support Vulkan on other OSes so not supporting Linux would not free up a bunch of additional resources to work on something else.

[humbug]

As I've said a number of times, there is not a Linux vulkan team at AMD. There is just a vulkan team. We still have to support Vulkan on other OSes so not supporting Linux would not free up a bunch of additional resources to work on something else.

Makes sense, since you have to maintain Vulkan across win7, win8, win10.

Question- once the Vulkan component of amdgpu-pro is open sourced do you expect that the linux vulkan driver will start diverging more and more from the windows driver or will they be stay similar with future updates too?

[agd5f]

Makes sense, since you have to maintain Vulkan across win7, win8, win10.

Question- once the Vulkan component of amdgpu-pro is open sourced do you expect that the linux vulkan driver will start diverging more and more from the windows driver or will they be stay similar with future updates too?

The idea is to keep the code in sync. We don't want to maintain two code bases.

[Drago]

The idea is to keep the code in sync. We don't want to maintain two code bases.

What is the current status of AMD Vulkan release? What is holding it up? Is it still code review/lawyers, or something more complicated and the process stalled.

[agd5f]

What is the current status of AMD Vulkan release? What is holding it up? Is it still code review/lawyers, or something more complicated and the process stalled.

Nothing is stalled. Still code review and carefully splitting out the things we can open source and the things we can't (e.g., stuff licensed from a 3rd party, etc.); then changes making it all fit together nicely so we can properly integrate contributions from the community and maintain those different pieces going forward in a reasonable manner. While doing all of this we also have to add support for new asics, etc.

[Drago]

Nothing is stalled. Still code review and carefully splitting out the things we can open source and the things we can't (e.g., stuff licensed from a 3rd party, etc.); then changes making it all fit together nicely so we can properly integrate contributions from the community and maintain those different pieces going forward in a reasonable manner. While doing all of this we also have to add support for new asics, etc.

Which ISA backend you are going to use for Vulkan? I can imagine it will not be part of Mesa, so no RadeonSI. Are you going to introduce another back-end?

[bridgman]

AFAIK the ISA backend you are talking about is part of LLVM, not Mesa, and is already being used in a number of different projects:

http://llvm.org/docs/AMDGPUUsage.html

[Drago]

AFAIK the ISA backend you are talking about is part of LLVM, not Mesa, and is already being used in a number of different projects:

http://llvm.org/docs/AMDGPUUsage.html

Okay so AMD Vulkan will emit LLVM IR, then Mesa and it will use the same LLVM backend already installed in the libllvm. Understood.

[ermo]

As I've said a number of times, there is not a Linux vulkan team at AMD. There is just a vulkan team. We still have to support Vulkan on other OSes so not supporting Linux would not free up a bunch of additional resources to work on something else.

Nothing is stalled. Still code review and carefully splitting out the things we can open source and the things we can't (e.g., stuff licensed from a 3rd party, etc.); then changes making it all fit together nicely so we can properly integrate contributions from the community and maintain those different pieces going forward in a reasonable manner. While doing all of this we also have to add support for new asics, etc.

Sounds like "fun". Here's hoping that all the extra work imposed by the Linux kernel devs' way of doing things will actually make the code base both cleaner and easier to maintain going forward.

What's your take? Is the above just a pipe dream?