Open-Source RADV Vulkan Driver Is Seeing Work To Allow Building It On Windows

[phoronix]

Phoronix: Open-Source RADV Vulkan Driver Is Seeing Work To Allow Building It On Windows

An independent party has slowly begun merging patches into mainline Mesa for allowing the open-source Radeon Vulkan driver "RADV" to build on Microsoft Windows...

http://www.phoronix.com/scan.php?page=news_item&px=RADV-Windows-WIP

[QwertyChouskie]

Could this pave the way for the RadeonSI OGL driver on Windows? Could be a huge win for Minecraft Java Edition players...

[Xaero_Vincent]

So this is just for cross-compiling RADV with Windows compilers for Linux use?

[FireBurn]

My theory is it's for virtualized Windows running under a Linux hypervisor. Vulkan windows games could be run whilst sending all the commands directly down to the DRM kernel layer in Linux

[QwertyChouskie]

My theory is it's for virtualized Windows running under a Linux hypervisor. Vulkan windows games could be run whilst sending all the commands directly down to the DRM kernel layer in Linux

Definitely possible. That being said, the whole point of the AMDGPU kernel driver is that it provides the same API and both Windows and Linux, allowing the same user-space driver to run on top of it. AMD already does that with AMDVLK, there's probably no good reason why this wouldn't also be possible with RADV, since they both ultimately interface with the AMDGPU API.

[N0dens]

My theory is it's for virtualized Windows running under a Linux hypervisor. Vulkan windows games could be run whilst sending all the commands directly down to the DRM kernel layer in Linux

Isn't AMD expected to enable sr-iov soon on consumer cards? We all know the chips are capable of doing it but in spite of many requests AMD has until now reserved these features for the Enterprise segment. Intel is already moving in that direction with Iris Xe. I know a lot of people (including me) are hoping an announcement from AMD with the release of their new architecture. If this happen then there would be no reason to make RADV run on Windows to enable virtualization. There might be other reasons to do it ... just not this one.

[bridgman]

Definitely possible. That being said, the whole point of the AMDGPU kernel driver is that it provides the same API and both Windows and Linux, allowing the same user-space driver to run on top of it. AMD already does that with AMDVLK, there's probably no good reason why this wouldn't also be possible with RADV, since they both ultimately interface with the AMDGPU API.

Not exactly... the amdgpu kernel driver added some functionality so that it could support the closed source userspace drivers, but it does not implement the same API as a Windows kernel driver... it uses standard drm interfacing. The userspace drivers still have different back ends for Linux and Windows.

Isn't AMD expected to enable sr-iov soon on consumer cards? We all know the chips are capable of doing it but in spite of many requests AMD has until now reserved these features for the Enterprise segment. Intel is already moving in that direction with Iris Xe. I know a lot of people (including me) are hoping an announcement from AMD with the release of their new architecture. If this happen then there would be no reason to make RADV run on Windows to enable virtualization. There might be other reasons to do it ... just not this one.

No plans to enable SR-IOV on consumer cards to my knowledge.

[N0dens]

No plans to enable SR-IOV on consumer cards to my knowledge.

That's very disappointing. Not even with the Radeon Pro cards ?
It sounds like Intel is the one one who want to allow end users to use virtualization outside of a data center. Let's hope they manage to improve their DG1 card.

[microcode]

Gotta say, the years go by and I think I speak for quite a few people when I say I've seen the AMD not paying quiiiite enough attention to software story play out a few times now. AMD should have a budget for grants to people doing interesting things with their products, because they are in a unique position as the only major desktop GPU manufacturer who publishes enough information for third parties to program their GPUs (assuming Intel can't realistically catch up in one or two gens). If the new hotness is discovered, it will probably be discovered on an AMD GPU, and probably sweetening the pot for devs and corps advancing the art would increase the odds of that.

[Guest]

Hm, I wonder if some day we can use RADV + DXVK/VKD3D etc. on Windows, in place of AMD's existing 3D driver stack. Atleast for games. It would be interesting to see how well it performs compares to that.