Valve Developed An Intel Linux Vulkan GPU Driver

[-MacNuke-]

I don't know for sure but is Gallium3D even low enough level to implement Vulkan? You can probably throw the state tracker code out as there won't be much state to track..

My thoughts exactly...

Gallium3D is a state tracker and Vulkan is a stateless API...

Vulkan is more or less a Virtual Machine and the drivers have to execute the given SPIR-V code. I can not see any usefulness in Gallium3D or Mesa.

A "simple" kernel driver with all the Userland-Tools should be enough. All hard work is done in userland by the application.

[sarmad]

Looks like Linux will just skip on OpenGL and jump straight to Vulcan, which is the right thing to do.

Well done Valve, well done.

[duby229]

I don't know for sure but is Gallium3D even low enough level to implement Vulkan? You can probably throw the state tracker code out as there won't be much state to track...

If that is the case, that may be why they could have chosen intel's driver stack over other cards. Perhaps intel's drivers are more ogl abstract than one may first assume?

There probably -will- be a vulkan state tracker for gallium3d. It doesn't have to deal with SPIR-V. The hardware driver does though, and that will probably be handled through LLVM.

[starheap]

My thoughts exactly...

Gallium3D is a state tracker and Vulkan is a stateless API...

Vulkan is more or less a Virtual Machine and the drivers have to execute the given SPIR-V code. I can not see any usefulness in Gallium3D or Mesa.

A "simple" kernel driver with all the Userland-Tools should be enough. All hard work is done in userland by the application.

yup vs Gallium3D takes in TGSI ir, perhaps something could be used from Gallium3D if it was straight forward to do SPIR-V -> TGSI.
I found this image interesting http://upload.wikimedia.org/wikipedi...ple_matrix.svg

[justmy2cents]

Gallium3D is a state tracker and Vulkan is a stateless API...

[joke] sure it is, just add states to Vulkan implementation it's not like state setting was ever a problem in GL or there ever was a reason for removing that [/joke]

[Kemosabe]

If that is the case, that may be why they could have chosen intel's driver stack over other cards. Perhaps intel's drivers are more ogl abstract than one may first assume?

Perhaps you're right the gallium way is not adequate for Vulkan but i think they chose intel for a much simpler reason:
It's the only official opensource driver out there...

[-MacNuke-]

There probably -will- be a vulkan state tracker for gallium3d. It doesn't have to deal with SPIR-V. The hardware driver does though, and that will probably be handled through LLVM.

So, Gallium3D will not be used at all

For the drivers there is nothing more than the SPIR-V code to execute.

[Blahblah]

Wow; Valve really does not let anything get in their way, do they?

[justmy2cents]

And hopefully wine devs will replace their OGL wrapper with Vulkan

it should probably be easier and much faster than now. implementing high level engine with different high level engine like they did and with so little losses like they succeeded... i call that ultimate heroism just a bit short of jumping on grenade in order to save people

[duby229]

So, Gallium3D will not be used at all

For the drivers there is nothing more than the SPIR-V code to execute.

exactly. I think that's what makes it nice.