Re: full/sparse/banded/triangular matrices...

[David Bateman]

According to Paul Kienzle <address@hidden> (on 12/01/04):
> LAPACK also has tridiagonal functions.  In the spline code we are
> doing a couple of rotations outside the solver so that the cyclic
> case doesn't need a full matrix solve.  I doubt you want to support
> a mostly banded matrix class though.

Isn't tridiagonal just a special case of a banded matrix? Does it need
special treatment relative to a banded matrix? Your cyclic solver would
though, humm...

> Another matrix attribute to consider is a transpose marker, but like
> triangular matrices, this only gives a small speedup compared to the
> cost of the op.

Yeah, I think I place that one in the nice but not worth it basket...

D.

-- 
David Bateman                                address@hidden
Motorola CRM                                 +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 1 69 35 77 01 (Fax) 
91193 Gif-Sur-Yvette FRANCE

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary