Codex News
Fast 4x4 Matrix inversion routine for AltiVec
I just finished this routine, and I'm really proud of it:
http://www.freevec.org/function/inverse_matrix_4x4_using_partitioning
Apart from beating the scalar version, I actually managed to beat the matrix inversion routine by the cellperformance.com guys who are really good at this stuff!
Anyway, things are progressing steadily and soon I'll be able to just incorporate all these functions into bigger projects like Mesa 3D and Blender. Most of the code is being s
Shortly after libfreevec 1.0.3 was released, we are pleased to announce the release of benchsuite 1.0.2, the software used to test and benchmark libfreevec. Though it was used to develop and test libfreevec, it's quite generic and should be easily adapted to pretty much any type of C functions in a library or standalone.
The data is saved in a SQLite table, so that it can be easily modified for any use.
The pretty Google graphs in http://freevec.org/functions have bee
After long hours of testing and debugging, CODEX releases libfreevec 1.0.3, get it here!
For benchmarks, check http://www.freevec.org/functions
Ok, it's official, I started working on SIMDizing 3D code:
http://www.powerdeveloper.org/forums/viewtopic.php?p=10160#10160
First stop is this library SIMDx86, which is ok, but apparently it has some limitations that will force to fork it or in any case change it in a fundamental way to be able to reuse it in other projects. Instead of copying and pasting the same code in many apps, I'll just provide the optimizations in this (or in a derivative) library and just change the
Well, the holidays are over, not that we really got in the spirit of the holidays anyway, as we were all sick here, my wife, the baby and myself. But at least, we don't have any excuse to stay away from work :-(
Well, libfreevec is almost ready for its first beta release 1.0rc1. Only memmove() is left for testing/fixing and we're ready for release. After that, the plan is to focus on the Blender port to Cell, port libfreevec to 64-bit (G5/Cell/Power6, it works already with adequate results but no 64-bit optimization has bee
