An Intro to AVX-512 Assembly Programming - Comments

[Mark44]

Greg Bernhardt submitted a new blog post

An Intro to AVX-512 Assembly Programming

Continue reading the Original Blog Post.

 

[berkeman]

Very cool subject. Thanks!

 

[jedishrfu]

Very cool, will this technology be integrated into Machine Learning applications as it seems its geared for matrix operations doing parallel adds like that.

Also I found some videos on Youtube covering the topic in more detail for those of us who are reading challenged.

 

[Mark44]

Very cool, will this technology be integrated into Machine Learning applications as it seems its geared for matrix operations doing parallel adds like that.

No idea -- I don't know much of anything about machine learning or neural network algorithms or the like. If they involve working with matrices, then this stuff is applicable.

Also I found some videos on Youtube covering the topic in more detail for those of us who are reading challenged.

This article is the first in a series of five or more. What I covered in the first article is a bit of background history and a very simple example, for which lots of documentation and examples already exist. In the next few articles I plan to talk about other AVX-512 instructions for which 1) the Intel documentation is very minimal, and 2) there are no usage examples online (if there are any, I couldn't find them after long searches), let alone videos.

"Real programmers don't need no stinkin' videos!"

 

[jedishrfu]

So true, REAL programmers don't read either until they have a problem with their code, if ever.

 

[Mark44]

In the next installment, I'll look at how you can go through a fairly large array of signed numbers, separating out the positive and negative values, and generating totals for each, without using any kind of decision control structures (i.e., no if statements).

Coming Real Soon...

 

[Mark44]

So true, REAL programmers don't read either until they have a problem with their code, if ever.

But if you're working with an API that you aren't intimately familiar with, you need to have the API docs very handy.

 

[jedishrfu]

I usually start with example code in Netbeans where little api help tips pop up. Only when I hit a bigger issue do I need to delve into the reference docs. It’s coding on the edge of a precipice.

 

[Mark44]

where little api help tips pop up

LOL! There are zero API help tips that pop up in my VS IDE. Maybe the Intel C/C++ compiler has them, or maybe not - don't know 'cause I don't have that compiler. I'm reasonably sure they don't pop up in the Gnu compiler, either. That's why I keep the PDF of the Intel Software Developer's Manual open when I'm writing that code.

 

[Mark44]

will this technology be integrated into Machine Learning applications

It occurred to me later that one of the parts of AVX-512 is VNNI, or Vector Neural Network Instructions. It isn't here now, but is slated to be released with the Ice Lake microarchitecture, the 10th gen Core architecture. The release is slated for sometime this year or next.

There's also another part, GFNI, Galois Field New Instructions, that are also tied to the Ice Lake microarchitecture.
At this point, there are 18 separate subsets of the AVX-512 instruction set, not all of which have been released just yet. My new computer, with its Xeon Silver 10-core processor, supports 5 of those subsets.

 

[jedishrfu]

I’m a java developer so that’s why we see those tips in Netbeans as part of the javadoc builtin comments.