AMD Pushes Out Their First Post-Release Update Of AMDVLK/XGL

[phoronix]

Phoronix: AMD Pushes Out Their First Post-Release Update Of AMDVLK/XGL

AMD developers working on the newly open-sourced AMDVLK Vulkan driver have pushed out their first post-release code update synced against the latest changes in their internal AMD driver tree...

http://www.phoronix.com/scan.php?page=news_item&px=AMDVLK-XGL-First-Code-Update

[tessio]

The official OpenGL driver is open source too, or only the Vulkan one?

[darkbasic]

Only Vulkan.

[humbug]

The official OpenGL driver is open source too, or only the Vulkan one?

As far as openGL is concerned the AMD devs have been contributing to the open source driver for years. They are already credited with getting it into the fine shape it is today. And it already is the driver which they 'officially' recommend for gamers.

the other closed source openGL driver is slow and they recommend it only for workstation users. No plans to open source this as far as I know. Nobody is really clamouring for it either.

[pal666]

The official OpenGL driver is open source too, or only the Vulkan one?

official opengl driver is mesa

[agaman]

I don't think this big undocumented code drops are good if they want help from the community...

From the response to ailied in the email:

- Will llvm master eventually be shippable? Will new llvm features be developed in the
For changes in LLVM, normally/eventually we will contribute to master as long as the changes could be accepted by the reviewers.

Seems from this that will be a normal thing to require a non-upstream patched LLVM to build AMDVLK.

[L_A_G]

Definitely nice to see them immediately get going on updating AMDVLK and I do hope the lack of a change log was just a slip-up on their part and not something intentional.

Still, now that AMDGPU has been mainlined I hope they can dedicate more resources into improving AMDVLK and maybe even getting it on par with AMDGPU.

[DanL]

Does anyone else think of this when they hear 'XGL' or have I just been using Linux way too long?

https://en.wikipedia.org/wiki/Xgl

[Michael]

Does anyone else think of this when they hear 'XGL' or have I just been using Linux way too long?
https://en.wikipedia.org/wiki/Xgl

Yep what I thought of when first hearing AMD mention it.

[oooverclocker]

I am very happy that AMD has released this open source code. But now we are facing several problems. Imagine you were a Feral Interactive developer and you would like to make an improvement. Or imagine you would work with VALVE... - Where would you post your changes? Which driver would you package in your distribution?
Don't underestimate this.

I have read about people who want to "cannibalize" AMDVLK because RADV is more community driven in their opinion. I can't share these thoughts. AMD and AMD developers are equal members of our community like everyone else who is helpful. And it's very helpful to have their code as well as their long term contribution to maintain and improve it. We have the goal that users of Linux distributions can flawlessly use hardware with maximum performance and minimum effort and that game developers happily write games for Linux and improve the drivers as well. Please realize that RADV might become as powerful as AMDVLK but without the compiler optimizations, it seems to be impossible to ever overtake the Windows performance and the performance of the proprietary AMDGPU-Pro-driver! Instead changes will also contribute indirectly to the Windows drivers.

I see two concerns about AMDVLK from the Linux perspective:
1. The driver should not fix defective code in games. The way to go mus be to communicate with developers of popular games when bugs are detected. When the specification is unclear the specification must be clarified.
2. As airlied already mentioned in his questions, we must be very interested about the different compilers. When the performance of AMDVLK + LLVM remains slower than the version that AMD uses internally, we can't accept this state because that would mean Windows and AMDGPU-Pro remains faster. We want a fair competition with other operating systems.

My personal idea how a compromise can look like is that AMD agrees to open their compiler knowledge as much as necessary to get a difference of less than +/- 1% in performance between their proprietary and the completely open source driver until 2019. Meanwhile although I am quite dogmatic, I think concessions can be made about code that's useless for Linux users, just making the driver compatible with AMDGPU-Pro/Radeon Software. Wasting resources to maintain two drivers would be a worse option but when this is necessary, we must accept it.

These are my thoughts and suggestions for the Mesa people to discuss with AMD.