Mesa 24.1 Zink Lands "Super Fast" Merge Request To Optimize IO

[Veto]

Somewhat off-topic, but this versioning scheme where software like Mesa and XWayald by the year is really interesting. I think I'm a big fan, though my brain likes to say "24.1 was the explicit sync patch. Oh wait, was that for XWayland or Mesa??"

I think that latter confusion is a me-problem and I can just adapt. A really cool thing is that the year-versioning leaves no room for doubt when you wonder what major version you're running on. If you're on Mesa 24.1 in the year 2028, you know you are quite behind.

The only minor catch I see is that it may be less obvious where a major change or improvement lands (for example, comparing Plasma 5.27, a major change over its predecessor, vs Plasma 6.0, an extremely major change over its predecessor).

Semantic versioning is bullshit in most cases. It only really makes sense for libraries that are completely adhering to the backwards compatibility promise within a major version. In most other cases it is just an arbitrary number - and what constitutes a minor change for some is a big one for others. For a complex project like Mesa, which always have big changes all around - but actually tries to uphold a dependable API it only really makes sense with a date based versioning. Even for the Linux kernel, which has a strict backwards compatibility policy, the major version is just an arbitrary extra digit.

[Quackdoc]

Zink has been great on android, always glad to see perf increases

[ezst036]

These kind of optimizations enable so many new apps. Imagine how much old good apps we can now write in inefficient web programming languages by IQ 80 morons using React and Electron.

The problem with this mentality of yours is it that takes the 80 IQ guys to get to writing GUIs that normal people can actually use, instead of the same old unusuable CLIs that have languished the Linux ecosystem for a long time.

The sad fact is that there is a certain segment of users who identify as CLI users and they'll never lift a finger to help anybody out, to the point of being a form of nasty elitism. If you don't want to use CLI too? You don't want to RTFM? You don't want to suffer through the MAN pages? Well, they don't, can't, and will not help you. You're simply beneath them.

It's 2024 now, and pretty much anything you want to do in Linux these days is just a click click click away with some GUI, but I still often run into people who haven't kept up on Linux and the first thing out of their mouth is about command lines and how they don't want to get stuck using command lines.

Yeah it's partly their fault for not trying something more recent before spouting words that make them stupid, but realistically the nasty CLI elitists foisted this upon us with their conservative refusal to change and see the utility of GUIs, even if they wouldn't be forced to use the GUIs themselves.

[zcansi]

I'm pretty sure that was a joke. It's the Zink code that will need fixes, not the test suite.

In the issue tracker it's pretty clear that it's actually the tests which are incorrect. Those tests unintentionally rely on something that is not actually guaranteed by the spec.

[ssokolow]

The problem with this mentality of yours is it that takes the 80 IQ guys to get to writing GUIs that normal people can actually use, instead of the same old unusuable CLIs that have languished the Linux ecosystem for a long time.

Not really. I'm a big UI/UX-head (I'm currently wavering over the temptation to write a PyQt-based "API client" for the automatic1111 distro of Stable Diffusion that beats the pants off Easy Diffusion on UI fit and finish)) and I'd choose a native GUI toolkit over Electron-based tech any day, purely on the amount of accumulated polish and improvement that's gone into the building blocks.

It's not necessarily elitism, because it's also that, since about 2001, we've lost our appreciation for what a giant gulf there is between "everything I tested works well" and "everything works well" and jumped on reinventing the same edge-case bugs time-after-time in pursuit of web-tech's no-deploy deployment.

Give the Macintosh Human Interface Guidelines a read. That's from 1992 (the 1995 revision apparently used to be up on Apple's website for download and there are PDFs floating around still) and it's shameful how much we've backslid because stuff got designed for good, still-applicable reasions, baked into the toolkit as a default, forgotten, and then abandoned in the switch to web-tech.

Joel Spolsky's Things You Should Never Do, Part I (about rewrites) is infamous.

This X post about SwiftUI glitching out used to host a reply thread back when it was Twitter which showed off all the different bugs that had crept back in when Apple reinvented native GUIs.

In Loving Memory of Square Checkbox laments how fully even Apple have lost the plot.

On the side of the "everyone copied Windows 8's flat design" hell we're living in, UX specialists like the Nielsen-Norman Group have written things like Flat UI Elements Attract Less Attention and Cause Uncertainty. (Makes sense to me. It's a design style from print, where nothing will happen no matter how many times you tap or hold on anything.)

We're living in the UI post-apocalypse, where designers who need to justify their salaries break things and it's OK because it "looks cooler".

I'm reminded of the rant at the end of Cathode Ray Dude's Quick Start, Episode 7: Crimes... And Felonies. about executives un-solving solved problems because it's the only way to make products change to justify people buying new ones, or the one at the end of Why would you scan things with a... mouse? where he talks about how nobody makes anything new anymore and hasn't since 2010. It's all just bait to get bought out by a multinational, and it's gotten to the point where security researchers despair because they know that, no matter how many layers of clever obfuscation they have to break through, the payload will always turn out to be a cryptominer.

[shmerl]

What does IO mean for OpenGL?

[airminer]

What does IO mean for OpenGL?

It means memory reads and writes by shaders.

This patch series ports some new code from RadeonSI that optimizes away unnecessary memory reads and writes by moving code between shader pipeline stages during shader linking.

[ssokolow]

Michael Poke, poke. Unapproved.

[ezst036]

Not really. I'm a big UI/UX-head (I'm currently wavering over the temptation to write a PyQt-based "API client" for the automatic1111 distro of Stable Diffusion that beats the pants off Easy Diffusion on UI fit and finish)) and I'd choose a native GUI toolkit over Electron-based tech any day, purely on the amount of accumulated polish and improvement that's gone into the building blocks.

Alright, that's good to know. I spoke too soon in this instant and the problem was PEBKAC on this end.

It's not necessarily elitism, because it's also that, since about 2001, we've lost our appreciation for what a giant gulf there is between "everything I tested works well" and "everything works well" and jumped on reinventing the same edge-case bugs time-after-time in pursuit of web-tech's no-deploy deployment.

I can see some truth to this, and since posting the thought dawned on me that review and how-to websites bear some of the guilt here also.

When going to try to find instructions on how to do something, it is still seemingly the majority of the case that you're given "drop to terminal and type this in". Which is actually counter productive. It would've been better if the how-to wouldn't have been written at all. It perpetuates the myth that is now well over a decade old that none of these things can be done with point and click.

Well why would anybody know that there are slick UX for these things. Nobody ever shows them off. Speaking of slick point and click UX, don't forget to type in apt-get!

(That bolded sentence is supposed to look bizarre, it highlights the problem. Why did you tell me to go to terminal, I could've done it in three clicks! SMH)

We're living in the UI post-apocalypse, where designers who need to justify their salaries break things and it's OK because it "looks cooler".

I know the type, but that's only the case where where UIs have actually been designed in the first place.

I'm reminded of the rant at the end of Cathode Ray Dude's Quick Start, Episode 7: Crimes... And Felonies. about executives un-solving solved problems because it's the only way to make products change to justify people buying new ones, or the one at the end of Why would you scan things with a... mouse? where he talks about how nobody makes anything new anymore and hasn't since 2010. It's all just bait to get bought out by a multinational, and it's gotten to the point where security researchers despair because they know that, no matter how many layers of clever obfuscation they have to break through, the payload will always turn out to be a cryptominer.

Much of the Linux world though is still lacking corporations - meaning, no executives. Though I can imagine that some of the non-profits have a similar mindset but for different reasons.

[ssokolow]

When going to try to find instructions on how to do something, it is still seemingly the majority of the case that you're given "drop to terminal and type this in". Which is actually counter productive. It would've been better if the how-to wouldn't have been written at all. It perpetuates the myth that is now well over a decade old that none of these things can be done with point and click.

Well why would anybody know that there are slick UX for these things. Nobody ever shows them off. Speaking of slick point and click UX, don't forget to type in apt-get!

(That bolded sentence is supposed to look bizarre, it highlights the problem. Why did you tell me to go to terminal, I could've done it in three clicks! SMH)

I can understand why they do it though. KDE, GNOME, Xfce, etc. all have different UIs, which might vary between major versions of the DE, but, if you're on any Debian-family distro, the same apt-get command should do the trick and the relevant subcommand has probably remained unchanged since apt-get was introduced in 1998... and the author also doesn't need to take and host screenshots.

Much of the Linux world though is still lacking corporations - meaning, no executives. Though I can imagine that some of the non-profits have a similar mindset but for different reasons.

On the Linux side, you don't have paying employers riding herd on people to counter the natural urge that it's more fun to rewrite and redesign than it polish out the bugs.