Eigen 2.0 beta1 released
22 Sep 2008 5:11pm GMT We just released the beta version of Eigen 2.0. It has matured a lot over recent weeks as quite a number of apps have been ported to it: Avogadro, KOffice (mainly Krita and a little KSpread), KGLLib, SolidKreator, KGLEngine. I have started working on a Mandelbrot fractal wallpaper-plugin for Plasma, using Eigen 2.0 for portable vectorization. We also have people on the mailing list doing very interesting things with Eigen 2.0: one is an engineer using it to design airplane engines, a few others...
(off-topic): Dimensions
26 Aug 2008 10:35pm GMT Here's something unrelated to KDE or Free Software, but which I still allow myself to blog about because every nerd here will agree it'd insanely cool.A team led by famous mathematician Etienne Ghys has produced a series of 9 short videos explaining some very cool geometry and topology. If you ever wanted to get a feel of what 4D geometry is like, this is what you've been looking for. The project is called Dimensions and is translated into several languages already.The videos are Creative...
Eigen2 alpha6
12 Aug 2008 2:03pm GMT It's been some time since I last updated you on Eigen2...We just released alpha6. At this point, Eigen2 can do all what Eigen1 did (and much, much more) so everyone is encouraged to use it for new code, and to port existing code to it.We plan to release Eigen 2.0 during this fall, in time for the releases of the software projects which plan to use it (tentatively: KDE 4.2, Avogadro 1.0, KOffice 2.0, OpenBabel 3.0).Eigen2 is a c++ pure-template-library (all the code in headers) for matrix/vector...
Eigen2 as a vectorization library
1 Jul 2008 4:17pm GMT A big issue for any math-oriented software is vectorization i.e. how to make use of instruction sets such as SSE or AltiVec. Unfortunately this seems to be very hard for the compiler to do automatically, and even with Intel's own compiler, with complicated C++ code such as ours, the benefits were small (like +25% with ICC and +5% with GCC).Yet, the theoretical benefit of vectorization is huge: with SSE, computations of floats can become 4x faster, and more in certain cases. So what math...
benchmarking virtual function calls in c++
20 Jun 2008 5:14pm GMT This is a followup to my two previous blog entries...In order to measure the speed overhead of virtual function calls, I wrote a simple 3-float vector class with a dot-product method. One version is done without virtual functions, all in a single Vector class. The other version uses a pure virtual base class VectorBase implementing the dot product, and the vector coordinates accessors are virtual. Aside from that, there is no difference in the code.The benchmark is here, and archive is here.The...