SVD computation on the Connection Machine
L. Magnus Ewerbring, Franklin T. Luk, A.H. Ruttenberg · 1988
The computation of the singular-value decomposition of a matrix on the Connection Machine is presented. The machine architecture is described, and it is explained why a Jacobi-type approach makes very efficient use of the available massive parallelism. Different implementation strategies are discussed for the case in which the problem size exceeds the machine size and vice versa. Encouraging timing data showing linear-time behavior of the algorithm are presented.>