
248
F. Lorenzelli and K. Yao
3.1 SVD UP/DOWNDATING IN PARALLEL
The parallel algorithm that we present here is based on the scheme proposed in [8], and we
assume that the reader is familiar with its operation. In particular, the E matrix is stored
in a triangular array, composed of
O(n 2)
locally connected processing elements, each of
which has access to four memory cells (entries of ~). All the processors are capable of both
column and row rotations, and the processors on the main diagonal additionally perform
2 • 2 SVD's. The particular array for the updating of the V matrix is not of