Wed Aug 9 13:09:37 CEST 2006
removing double precision?
what about keeping double precision matrices inside PF? the only place
where this is important is for ill conditioned 'cancellation' problems
which probably only occur as intermediate states in some algorithms,
and not as source data or end product.
currently it's only the LU decomp that requires double precision,
which is only used in the matrix:inverse word atm. in the future the
SVD might show up somewhere, and some levinson / shur like algorithms.