GNU C Library 2.33 Released With HWCAPS To Load Optimized Libraries For Modern CPUs

[phoronix]

Phoronix: GNU C Library 2.33 Released With HWCAPS To Load Optimized Libraries For Modern CPUs

The GNU C Library 2.33 release is out today as expected. Exciting with this libc update is HWCAPS in making it easier to load optimized libraries for modern CPUs...

http://www.phoronix.com/scan.php?page=news_item&px=GNU-C-Library-2.33

[Peter Fodrek]

Congratulations . Are there any performance benchmark plans for
1. Zen
2. Zen+
3. Zen2
4. Zen3

comparision while running glibc 2.33 vs 2.32

[ms178]

By the way, there was another Ryzen-specific patchset but it didn't make it into 2.33, it is reviewed though and okay'd for 2.34.

[cbxbiker61]

As far as I know, there doesn't exist a Linux distribution enabling such recompilation seamlessly yet.

I have my own private distribution and I've set my local build server to build x86_64, x86_64v2 and x86_64v3. Each of my client cpu's run a script to determine the highest level of support and install the appropriate version. I rebuild approximately a third of my packages every month and this month is the first month that I'll have multiple x86_64 versions.

[Peter Fodrek]

I am not sure what you mean exactly, so, just a note to clarify things: In my opinion, the effect of loading a CPU-optimized glibc.so library alone won't be that big, It will take some time for other libraries (outside of glibc, such as libpng.so and others) to get compiled for multiple CPU architectures. A proper "comparison while running glibc 2.33 vs 2.32" would mean recompiling all libraries in the system. As far as I know, there doesn't exist a Linux distribution enabling such recompilation seamlessly yet.

---

There is virtually no difference between Zen and Zen+, except for 5-10% higher frequencies in Zen+ (the CPU and the DDR4 interface are clocked higher).

Really? Intel MKL shows huge difference

Nov 18th, 2019
MATLAB is a popular math computing environment in use by engineering firms, universities, and other research institutes. Some of its operations can be made to leverage Intel MKL (Math Kernel Library), which is poorly optimized for, and notoriously slow on AMD Ryzen processors. Reddit user Nedflanders1976 devised a way to restore anywhere between 20 to 300 percent performance on Ryzen and Ryzen Threadripper processors, by forcing MATLAB to use advanced instruction-sets such as AVX2.

https://www.techpowerup.com/261241/matlab-mkl-codepath-tweak-boosts-amd-ryzen-mkl-performance-significantly?cp=3

HWCAP may do something similar then forcing use of AXV2. Or not?

25 March 2020

Stemming from Glibc semantics that effectively "cripple AMD" in just checking for Intel CPUs while AMD CPUs with Glibc are not even taking advantage of Haswell era CPU features,

https://www.phoronix.com/scan.php?page=news_item&px=GNU-libc-Platform-Optimize-Zen

[Danny3]

I hope this could bring some Zen 2/3 improvements and optimizations!

[Peter Fodrek]

To show a "huge difference" between Zen and Zen+ is physically impossible.

But I mention huge difference between Zen with glibc 2.33 compared to glibc 2.32 maybe much higher then Zen+ compared to Zen with same glibc version

[f0rmat]

Michael Possible grammar mistake

See that linked article from last week for more details but Glibc 2.33 but basically it allows the dynamic linker to load optimized versions of libraries within a glibc-hwcaps directory on the library search path.

[f0rmat]

Really? Intel MKL shows huge difference

https://www.techpowerup.com/261241/matlab-mkl-codepath-tweak-boosts-amd-ryzen-mkl-performance-significantly?cp=3

HWCAP may do something similar then forcing use of AXV2. Or not?

Thanks for the link. That was very interesting.

[f0rmat]

I am not sure what you mean exactly, so, just a note to clarify things: In my opinion, the effect of loading a CPU-optimized glibc.so library alone won't be that big, It will take some time for other libraries (outside of glibc, such as libpng.so and others) to get compiled for multiple CPU architectures. A proper "comparison while running glibc 2.33 vs 2.32" would mean recompiling all libraries in the system. As far as I know, there doesn't exist a Linux distribution enabling such recompilation seamlessly yet.

---

There is virtually no difference between Zen and Zen+, except for 5-10% higher frequencies in Zen+ (the CPU and the DDR4 interface are clocked higher).

I may be showing a huge amount of ignorance here (so please correct me if i am wrong), but does not Gentoo fairly seamlessly recompile libraries when you update it using portage? I am not a Gentoo user, so if someone could explain this to me, I would appreciate it.