The purpose of this software library is to perform fast massively parallel eigenvalue computations of large quantities of comparably small non-singular submatrices within a much larger positive-symmetric matrix in CUDA; notably, this library has been developed to assist in the study of Reduced Isometry Property (RIP) sensing matrices in the field of Compressive Sensing. To this end, several Matlab front-ends are included, as well as standard command-line C examples. This library was written entirely by Brian Tuomanen.
This work was supported by the National Science Foundation. (NSF ATD 1321779)
Please see ParaLanczos.pdf for further information.