Gabriele Svelto (gabriele.svelto.delete@this.gmail.com) on January 11, 2014 3:17 pm wrote:
> Nicolas Capens (nicolas.capens.delete@this.gmail.com) on January 11, 2014 1:20 pm wrote:
> > Trust me, I know how complex they are but I also know what
> > isn't complex to an experienced compiler engineer.
>
> Why should I trust you? You brought exactly zero proof of your theory, not even a patent
> or some example present in any previous architecture and you continuously hand-wave
> away arguments with actual reference without providing any reference yourself.

Proof of a theory makes it a fact. Since KNL has not launched yet, nobody has proof of any of their theories. As for a patent, first of all I'm a patent author myself and like any engineer I don't read patents other than my own to avoid triple damages. Secondly patents aren't published until 18 months after filing so you may not find one right away. Thirdly there might be an old patent for a purpose that doesn't seem related at first but still counts as prior art which makes Intel's implementation not patentable. Also there might be research papers which described some important parts of it before, such as this one: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.9.847&rep=rep1&type=pdf