I set about to wrap the excellent Eigen C++ template class.
To my surprise the whole endeavour was quite straight forward.
The code consist mostly of simple boiler plate code and tries
to follow numpy call signatures where it is natural.
The Eigen library is at: http://eigen.tuxfamily.org
For now only the dynamic Vector and Matrix
(integer,double & complex variant) are wrapped.
Initial tests show a impressive speed on my workstation with
up to 10x speed gain compared to numpy. This was tested on a
modern CPU (AMD Athlon 64) and the latest MinGW compiler (4.4)
with flags for the SSE2 added. Your mileage may wary depending
on your compiler.
Comments and modification are welcomed.
Consider this a proof of concept for now.
Link to source: http://pyscite.googlecode.com/files/Eigen-0.1a.zip
Runar