Zink Lands Async Pipeline Precompiles For Better Performance, Less Game Stuttering

[phoronix]

Phoronix: Zink Lands Async Pipeline Precompiles For Better Performance, Less Game Stuttering

Mesa's Zink driver implementing the OpenGL API atop Vulkan continues advancing at a rapid pace and today the latest major addition landed: async pipeline precompiles...

https://www.phoronix.com/news/Zink-Async-Pipeline-Pre

[uid313]

I remember hearing that "Zink was done", like was complete and there would be nothing more coming to it, but it keeps getting better and better and more improvements all the time.

What is next for Zink?

[arun54321]

Ok. can we have some benchmarks?

[MauganRa]

I remember hearing that "Zink was done", like was complete and there would be nothing more coming to it, but it keeps getting better and better and more improvements all the time.

What is next for Zink?

He probably meant that it left the experimental phase for good. OpenGL compatibility was achieved, weird bugs got trimmed down to a minimum, and performance was acceptable, at least for the time being. Of course, optimization and bugfixing never stops.

[uid313]

He probably meant that it left the experimental phase for good. OpenGL compatibility was achieved, weird bugs got trimmed down to a minimum, and performance was acceptable, at least for the time being. Of course, optimization and bugfixing never stops.

I don't know, after the easy optimzations have been done it gets harder to optimize much more, then after more smaller optimizations have done, then there is not much more room for further optimization, maybe none when its pretty optimized overall.

After it has matured and lots of bugs have been fixed without any new features having been added, its not going to be much bugfixes either.

[Amaranth]

I don't know, after the easy optimzations have been done it gets harder to optimize much more, then after more smaller optimizations have done, then there is not much more room for further optimization, maybe none when its pretty optimized overall.

After it has matured and lots of bugs have been fixed without any new features having been added, its not going to be much bugfixes either.

This assumes both OpenGL and Vulkan never gain any new features. A new OpenGL feature obviously requires additional work on Zink to support it and new Vulkan features might make it possible to do something in Zink more efficiently (like the change talked about here). Then you also have general maintenance and cleanup work happening in Gallium as a whole that Zink needs to keep up with. It'll never really be "done", there is always something else you can improve.

[theriddick]

Maybe using Vulkan more to help OGL threading work better would be neat! If there is still room for that. I think that was a big issue with OGL to begin with.

[marlock]

Was pipeline precompilation even a thing on normal OpenGL drivers?

[jsnjkl]

He probably meant that it left the experimental phase for good. OpenGL compatibility was achieved, weird bugs got trimmed down to a minimum, and performance was acceptable, at least for the time being. Of course, optimization and bugfixing never stops.

Do you have quantitative data when you are stating something is improving?