From: Ian McCulloch (ianmcc_at_[hidden])
Date: 2002-04-10 08:49:03
On Wed, 10 Apr 2002, Toon Knapen wrote:
> On Wednesday 10 April 2002 08:51, you wrote:
> > P.S.: To give an impression, how the test integration works, I'll include a
> > sample implementation and a cutout of the modified CLAPACK's blaswrap.h.
>
> and I suppose gmm is implemented using ublas routines. So this also compares
> the speed of ublas with BLAS (I'v always been thinking in the other
> direction, that you made binding for BLAS to be able to call BLAS using ublas
> matrices)
What is the point of reimplementing BLAS? Do you really expect to end up
with something better than, say, the ATLAS implementation? What extra
functionality does ublas give?