There's Experimental Work On A Vulkan Renderer For KDE's KWin

[phoronix]

Phoronix: There's Experimental Work On A Vulkan Renderer For KDE's KWin

There is an experimental branch of KDE's KWin window manager / compositor with support for Vulkan compositing...

http://www.phoronix.com/scan.php?page=news_item&px=KDE-KWin-On-Vulkan-Experimental

[schmidtbag]

Though working on a Vulkan renderer is an inevitable goal, I personally would rather have more attention drawn to implementing Wayland support. Desktop compositing isn't really demanding enough to warrant Vulkan (even Intel GMA GPUs can handle some compositing effects), whereas Wayland support still has functionality issues.

[torturedutopian]

Dunno, might be useful. Indeed, Intel GMA GPUs handle kwin animations / compositing beautifully, but it's not perfectly smooth and happens to be jerky with some other chips like NVIDIA ones (under GNOME too). Here Vulkan might make a difference ?

Wonder if this Vulkan renderer runs under an X context ?

[lunarcloud]

There very well may be platforms in future where the Vulkan support is better than the OpenGL support or even where there is *only* Vulkan support. Those two situations make doing this worthwhile alone.

[zanny]

Though working on a Vulkan renderer is an inevitable goal, I personally would rather have more attention drawn to implementing Wayland support. Desktop compositing isn't really demanding enough to warrant Vulkan (even Intel GMA GPUs can handle some compositing effects), whereas Wayland support still has functionality issues.

By the time we see any platforms considering going Vulkan-only we will almost certainly have an OpenGL in Vulkan implementation to run all the GL software out there on. GL Overload might be dead but all it takes is the market pressure to make it happen and you'd be surprised at the turnaround time.

[M@GOid]

The people behind the Nintendo Switch hack to launch Linux, now are able to run KDE with touchscreen controls:

https://twitter.com/fail0verflow/status/964954316892119040/video/1

[AsuMagic]

Though working on a Vulkan renderer is an inevitable goal, I personally would rather have more attention drawn to implementing Wayland support. Desktop compositing isn't really demanding enough to warrant Vulkan (even Intel GMA GPUs can handle some compositing effects), whereas Wayland support still has functionality issues.

Vulkan improves CPU bound scenarios, too. It doesn't seem to be any better in GPU bound scenarios.
So, Vulkan compositing is at least definitively worth considering.

[schmidtbag]

lunarcloud, zanny and AsuMagic
I understand that adopting Vulkan will be slow, and to clarify, I am in-favor of its implementation, eventually. I don't think we will see DX12/Vulkan-only hardware for at least a decade (it would be foolish to make hardware incompatible with older APIs), and the hardware demands for Kwin compositing is too minimal for there to be a need for Vulkan support. So at the moment, implementing it serves no purpose other than "this might be useful one day".

As for using Vulkan for performance reasons - remember, Vulkan doesn't necessarily make GPUs calculate faster, it just makes the CPU and PCIe bus not work as hard. In other words, if your GPU is slow enough to encounter stuttering issues, Vulkan isn't going to fix that. In my experience, Xrender has often been fast enough to make me think "wait a minute, is GPU acceleration working?" even on a Cortex A9, so I don't think CPU usage is a major concern.

I suppose one major benefit of adding Vulkan support is for ARM devices, where the added efficiency of Vulkan would improve battery life. But... Vulkan support on ARM is in rough shape. I think it'd take less time to write a Vulkan compositor for KDE than it would take to write Vulkan drivers for ARM products.

EDIT:
To reiterate, I'm not saying to not do Vulkan rendering, but maybe just hold it off until KDE's Wayland implementation is in better shape.

[FishPls]

Great stuff IMO.

[shmerl]

KWin doesn't support threaded rendering as would be needed by Vulkan

While it's not exactly correct (Vulkan allows, but doesn't necessarily require threaded rendering), did this work actually implement parallelized rendering for KWin? Something Servo like would really be cool!

One place where it can be very useful is Plasma Mobile. Implementing Vulkan driver for mobile chips should be way easier, than OpenGL one (consider Etnaviv and Librem 5). So it's a major plus for KWin to run on Vulkan in such cases, since it will enable Plasma Mobile on more devices sooner potentially.