This week I got a simple Linbox application written that calls the existing Block-Wiedemann implementation block-coppersmith-domain. This implementation is one of two in Linbox, and seems to be more amenable to parallelism. I'm now working on generating some test matrices that I can run through this code to ensure that I continue to generate the right results when I start adding OpenMP #pragmas.
I've also written a reference implementation of (non-blocked) Wiedemann's algorithm in Python along with some simple finite field libraries to accompany it. I hope to add a Block-Wiedemann implementation as well that I can use to try out algorithmic changes before I change the much more complicated Linbox version.
No comments:
Post a Comment