Zink OpenGL-On-Vulkan Can Finally Render glxgears With Great Speed

[cmsigler]

Where do babies come from?

How is babby formed

[arQon]

On a serious note, this is great progress. Rather than the random off-topic questions, I was expecting this thread to be more filled with "hurr durr glxgears isn't a benchmark!" while totally missing the point that if you can't run glxgears quickly, there's an underlying issue.

Honestly, in this particular case there really kinda isn't. Using GL_QUADS is frankly a pathological case, and that was true even 20 years ago.

I mean, yeah, it's nice that it got fixed, sure - but between that and the display lists this sure as hell would have had a "not my problem, you're holding it wrong, wontfix" ending instead if he'd had the attitude of a RedHat staffer. (Okay, fine: RH would have just ignored it for 6 months and then autoclosed it, but same difference).

Sorry for the OT post: I don't think I have any support questions right now.

[schmidtbag]

Honestly, in this particular case there really kinda isn't. Using GL_QUADS is frankly a pathological case, and that was true even 20 years ago.

I mean, yeah, it's nice that it got fixed, sure - but between that and the display lists this sure as hell would have had a "not my problem, you're holding it wrong, wontfix" ending instead if he'd had the attitude of a RedHat staffer. (Okay, fine: RH would have just ignored it for 6 months and then autoclosed it, but same difference).

That's the thing though: it got fixed, which is a critical detail. If there was some inherent reason why the drivers couldn't run glxgears much faster, that would just be a case of "not my problem; won't fix" because sometimes, sacrificing performance of legacy products is worth it for better performance in newer products. Think of it like DXVK, which runs like crap on a 4-thread CPU but on 8+ threads, it even has the potential to outperform native DX11. Attempting to optimize DXVK for low-end CPUs could threaten the performance of more modern systems, and the fact of the matter is, the nature of the driver will result in extra overhead (compared to a native driver), so low-end CPUs will never perform good enough anyway.
Anyway, in the case of Zink and glxgears, it resulted in a major improvement, yet ostensibly, there was no regression or loss of features. That fact alone means there was a bottleneck in the driver, which could mean other programs are affected too. Obviously with glxgears being so basic, that bottleneck is likely insignificant in modern AAA titles, but the point is that's potential extra performance that just became unlocked. Every little bit counts.

Just as a side note, the "you're holding it wrong" statement is in and of itself a problem, as the context of that saying was being dismissive of a design flaw. I get what you were trying to say though.

Sorry for the OT post: I don't think I have any support questions right now.