Hi - yes, we would ideally like to switch to more optimised linear
algebra routines such as those that you mention. However, we would
need a C++ wrapper to these that looked identical to our current calls
to newmat - otherwise we would need a lot of programming manpower to
convert all our code. So - it's on our todo list....
Cheers.
On 17 Apr 2008, at 16:04, Blaise Frederick wrote:
> Cool - I had just made the same change and was testing it - glad to
> see that's it!
>
> As an aside, in general profiling, it seems that pretty much all of
> the programs that take a lot more
> time than I'd like (film_gls - I'm looking at you!) seem to spend a
> huge amount of time (over 85%) in
> NEWMAT, especially in the matrix multiplication routine. This isn't
> surprising, given that fMRI
> processing is basically just a lot of linear algebra. Have you
> given any thought to recoding some of
> the routines in NEWMAT (especially matrix multiplication and SVD) to
> use LAPACK and BLAS? That
> would get you vector performance for free, since pretty much every
> OS has hand-optimized versions
> of these libraries available.
>
> Blaise
>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|