AMD Zen 3 Scheduler Model Finally Added To LLVM/Clang

[phoronix]

Phoronix: AMD Zen 3 Scheduler Model Finally Added To LLVM/Clang

While last minute AMD Zen 3 "znver3" improvements managed to make it for GCC 11 that was recently released, the recent debut of LLVM 12.0 wasn't so lucky on the Zen 3 support front. There was the very basic enablement that landed in LLVM 12 but now the more complete support isn't expected until LLVM 13 this autumn...

https://www.phoronix.com/scan.php?page=news_item&px=AMD-Zen-3-Scheduler-Model-LLVM

[ms178]

From the commit message: "I haven't run much benchmarks with this...". Is this AMD calling for help from Michael?

Maybe a LLVM-git benchmark round on a couple of Ryzen 5000 / EPYC processors could shed some light on this matter?

[rene]

From the commit message: "I haven't run much benchmarks with this...". Is this AMD calling for help from Michael?

Maybe a LLVM-git benchmark round on a couple of Ryzen 5000 / EPYC processors could shed some light on this matter?

I found this tuned instruction scheduling usually not having too much of an performance impact.

[gigi]

probably we should call AMD as ltm (last minute)

[sdack]

... It's too bad though that this didn't land months ago in ensuring good out-of-the-box compiler support ...

How much more of this self-entitled nonsense do we need to read? Apart from the world being plagued by a virus does Michael want to insist there was a problem and that the compilers did not provide good support.

How about providing us with good and fair articles, Michael? Or send the compiler patches yourself ...

[Guest]

From the commit message: "I haven't run much benchmarks with this...". Is this AMD calling for help from Michael?

Maybe a LLVM-git benchmark round on a couple of Ryzen 5000 / EPYC processors could shed some light on this matter?

Sure, if AMD pays him for it. They seem to be doing quite well financially. No excuses there.

[Guest]

How much more of this self-entitled nonsense do we need to read? Apart from the world being plagued by a virus does Michael want to insist there was a problem and that the compilers did not provide good support.

How about providing us with good and fair articles, Michael? Or send the compiler patches yourself ...

Oh look, triggered AMD fanboy. Sure, pay Michael 150,000 USD per year, give him access to the hardware and docs, and he or someone else will gladly write those patches for you. Until then, stop being triggered on AMDs behalf, they have PR to do that job.

[Guest]

Ah, shitty AMD Linux support as usual. Time to permanently use Intel only, goodbye AMD!

[sdack]

Oh look, triggered AMD fanboy. Sure, pay Michael 150,000 USD per year, give him access to the hardware and docs, and he or someone else will gladly write those patches for you. Until then, stop being triggered on AMDs behalf, they have PR to do that job.

I am not a fanboy, but you seem to be one now that you only want to buy Intel. Good luck with your choice.

GCC and CLANG are open-source and they are based on voluntary work. Nobody has to make contributions. Nor is there any guarantee given by these projects in regards to what their software does or does not do or provide. To complain about open source projects is generally frowned upon and the consent here is that if you do not like an open source project then you can start by getting active yourself, by making your own contributions, to make your own fork and version, or to start an entirely new project.

And as already pointed out are we fighting off a pandemic. People are more often at home and take care of themselves and their families. Many businesses have slowed down. Just take a look at how the hardware market has cramped up to the point where several products are just not available. To then complain how a company would not provide a set of patches fast enough is not more than a pile of self-entitlement nonsense.

Why should I pay Micheal money? He cannot even write objective articles. He knows enough about the open source community to know better and his articles come across as the writings of a drunk. Where are then the benchmarks that could show the difference and importance between znver2 and znver3? There are not any. This says enough.