From: Nico Galoppo (ngaloppo_at_[hidden])
Date: 2008-03-05 17:55:54
I know I am going to touch all (i,j), therefore precomputation with
gemm is going to give me the best performance that I can think of. I
was just hoping there'd be a cleaner way to 'flatten' a rxr matrix
into an r^2 vectors, perhaps using bindings/traits. Right now, my code
assumes a lot about the underlying layout.
On Wed, Mar 5, 2008 at 5:44 PM, Gunter Winkler <guwi17_at_[hidden]> wrote:
> Nico Galoppo wrote:
> > Paul,
> > thanks for the tip. Unfortunately, the matrices A and B that I'm
> > dealing with are not monomial. As a matter of fact, A itself is
> > constructed as a low-rank multiplication: A = U1 * U2' with v1 and v2
> > both r-by-3 matrices.
> IMO the best solution would be to provide a custom operator like
> matrix_matrix_prod which knows how to compute element (i,j) of the result.
> ublas mailing list
-- Nico Galoppo :: http://www.ngaloppo.org