It Turns Out

…that this year is a lot busier than expected. Blog posts will probably come in small clusters here and there rather than with any sort of regular cadence.

But now I’m here. You’re here. Let’s get cozy for a few minutes.

NVK O’clock

I’m sure you’ve seen some news, you’ve been trawling the gitlab MRs, you’re on the #nouveau channels. You’re one of my readers, so we both know you must be an expert.

Zink on NVK is happening.

Those of you who remember the zink XDC talk know that this work has been ongoing for a while, but now I can finally reveal the real life twist that only a small number of need-to-know community members have been keeping under wraps for years: I still haven’t been to XDC yet.

Let me explain.

I’m sure everyone recalls the point in the presentation where “I” talked about progress made towards Zink on NVK. A lot of people laughed it off; oh sure, you said, that’s just the usual sort of joke we expect. But what if I told you it wasn’t a joke? That all of it was 100% accurate, it just hadn’t happened yet?

I know what you’re thinking now, and you’re absolutely correct. The me that attended XDC was actually time traveling from the future. A future in which Zink on NVK is very much finished. Since then, I’ve been slowly and quietly “backporting” the patches my future self wrote and slipping them into git.

Let’s look at an example.

The Great Gaming Bug Of ‘24

20 Feb 2024 was a landmark day in my future-journal for a number of reasons, not the least due to the alarming effects of planetary alignment that you’re all no doubt monitoring. For the purposes of the current blog post that I’m now writing, however, it was monumental for a different reason. This was the day that noted zinkologist and current record-holder for Most Tests Fixed With One Line Of Code, Faith Ekstrand (@gfxstrand), would delve into debugging the most serious known issue in zink+nvk:


Yup, it’s another clusterfuck.

Now let me say that I had the debug session noted down in my journal, but I didn’t add details. If you haven’t been in #nouveau for a live debug session, it’s worth scheduling time around it. Get some popcorn ready. Put on your safety glasses and set up your regulation-size splatterguard, all the usual, and then…

Well, if I had to describe the scene, it’s like watching someone feed a log into a wood chipper. All the potential issues investigated one-by-one and eliminated into the pile of growing sawdust.

Anyway, it turns out that NVK (currently) does not expose a BAR memory type with host-visible and device-local properties, and zink has no handling for persistently mapped buffers in this scenario. I carefully cherry-picked the appropriate patch from my futurelog and rammed it through CI late at night when nobody would notice.

As a result, all GL games now work on NVK. No hyperbole. They just work.

Stay tuned for future updates backported from a time when I’m not struggling to find spare seconds under the watchful gaze of Big Triangle.

Written on February 21, 2024