Re: [boost] Eigen vectors?

From: Thomas Klimpel (Thomas.Klimpel_at_[hidden])
Date: 2008-03-10 13:06:04

Hello David,

> Ok, I got the bindings out of the sandbox and put them in my
> .../boost_1_34_1\boost\numeric folder.  I downloaded TNT from here:
> http://math.nist.gov/tnt/tnt126.zip, but it is just 21 .h files and I
dont
> see really any functions that I am looking for.

I don't know TNT in detail, but I guess you would have to also download
JAMA (which is available from the TNT webpage), if you need a solver for
Eigen vectors.

1) Am I right that you can use atlas OR tnt?

No, these are complementary. You can use tnt with bindings+atlas, or
with JAMA. You can use uBlas with bindings+atlas. 

ATLAS and LAPACK are not exactly the same thing. ATLAS provides a
complete BLAS implementation, but contains only a subset of the routines
provided by LAPACK. I don't know ATLAS in detail, but I don't think that
it has a solver for Eigen vectors included directly (I used
http://www.netlib.org/lapack/lug/node142.html to dispatch the cryptic
Fortran 6 letter names in clapack.h from ATLAS, and they don't seem to
match with a solver for Eigen vectors). So you need LAPACK (Fortran) or
CLAPACK if you need access to a solver for Eigen vectors. The bindings
also work with LAPACK and CLAPACK, so this should be no problem.

The JAMA library (from TNT) on the other hand probably contains a solver
for eigen vectors.

> 2) Should I be using TNT instead of lapack?

I personally prefer uBlas + bindings + lapack, but as I already pointed
out, I don't know TNT in detail. 

> 3) do the TNT .h files have to be in a particular location?

I guess the bindings expect them to be in a subfolder tnt relative to
your include path.

> Sorry for the confusion, I've just never used anything with "bindings"
> before, so I don't know how it works!

Maybe you should have a look at
http://lists.boost.org/MailArchives/ublas  (ublas_at_[hidden]).

Regards,
Thomas