Large Scale Lanczos

This is an important algorithm for finding eigenvalues of large matrices eg., for latent semantic indexing in search. I’m currently porting this to GPU for a colleague’s research into topological exitons in FQHE (utilizing diagonalization of the Hamiltonian); matrices involved have side N > 1e6. The Lanczos algorithm is relatively simple, although it does require restart/re-orthogonalization throughout. The implementation uses, among others, the sgemv routine from Intel MKL and/or cuBLAS, initial results promising; 3x speedup using GPU, over 8 CPU threads.


Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s