Using cutting edge vector and parallel code, Codex is taking software optimization to new levels, increasing software performance in orders of magnitude (quite like using rocket fuel in your SUV !)

Codex News

Codex sponsors kyuba!

We at Codex are very proud to sponsor an open source project like kyuba/dev9 which when finished -which won't be far off in the future actually-, it will revolutionize Linux (and probably other OSes as well)!!

Kyuba is a rewrite of einit, made again by the same dedicated and knowledgeable coder, Magnus Deininger. Anyone who has used einit before will tell you that it has changed their idea of boot times and system efficiency. A beta einit version was able to reduce boot times more than 100%, from 55s down to 20s on a powerbook G4 and even down to 7 SECONDS on an Athlon X2!!!!

<

libfreevec 1.0.4 released!

Minor but important upgrade of libfreevec, get it here!

Talk in Software Optimisations, SIMD and Cell!

This Friday (Jun 6, 2008), I gave a talk (in Greek) at the University of Peiraeus about "Software Optimisations, SIMD and the Cell BE".

The flyer is here and the official page of the event is here.

Live streaming of the talk (in Greek) was provided, and soon the videos will be available online (I'll post the URLs when ready).

You can find the slides at the event page or here

preliminary AltiVec optimizations in Eigen2

The post on the Eigen2 mailing list says all.

In my continuing effort of AltiVec optimizations in core libraries/components of Linux, Eigen2 was relatively easy optimized for AltiVec. Eigen is used in KSpread2 (part of KOffice the KDE Office suite) soon to be released. This would mean that all kinds of math would run much faster on KSpread.

UPDATE: With some extra altivec hacking, I managed to get down from 11s for 4x4 matrices down to 4s (scalar does 14s, 40M

Is it time yet for another libc?

Well, at least for ppc it is. Glibc ppc support is very minimal, and anyway, most functions in glibc are slow (reference implementations doing per-byte processing). Where there are any optimizations, they exist only for x86/x86_64. PowerPC optimizations exist only for very few functions. While these are the most common ones, still, ppc units such as AltiVec are left unused.
I intend to (try) to change that.

But more about this later.


< first   < previous   1  2  3  4  5  6  7   next >   last >