Re: GSL, blas and Mathematica

["Robert G. Brown" <rgb at phy dot duke dot edu>]

On Fri, 18 Oct 2002, Alberto [iso-8859-1] Simões/EPL wrote:

> 
> Hi.
> 
> I'm developing a mathematica alike program for linux (GSL) named 
> numexp (numexp.sf.net). At the moment, it is only a numeric processor
> (not algebraic functionalities at all (almost)).
> 
> We are using GSL to implement a bunch of functions. I've read something
> about blas, but I don't have certain I understood it... is it realted
> to algebraic processment? If not, where can I get good documentation
> on how to implement this hard stuff?

BLAS stands for Basic Linear Algebra System -- it is an API for
low-level routines that perform "atomic" linear algebra functions, from
which more complex linear algebra operations can be build.  In order to
proceed optimally efficiently, a lot of those operations need to be
tuned (both blocksize and algorithm) to the operational hardware on the
system so that things like cache hits and fetch times can be optimized.
This is the basis for the ATLAS (automatically tuned linear algebra
system) which basically builds a special BLAS library that can be used
in LAPACK (a standard library for higher order linear algebra calls) or
GSL (which already lets you specify a blas library IIRC) and other
things.

A good place to learn about this sort of stuff online is at the oak
ridge netlib repository.  Jack Dongarra (onetime author of PVM) is also
one of the co-parents of ATLAS (written mostly by his student/colleague
Clint Whaley), and there is a lot of documentation there on BLAS and
linear algebra libraries in general.  This group is Very Bright and has
been for some time.

Netlib (netlib.org, follow link to browse the repository) has a blas
link with all sorts of stuff -- references, fortran sources, and more.
Ditto atlas.  Ditto, in fact, a ton of stuff -- netlib has been around
long enough that I've accessed it with ftp, gopher, its own custom
interface, and the web.  An unfortunate bias towards fortran libraries,
perhaps, but still a lovely place to look for certain things.

I for one would love to see a clone of mathematica good enough to
completely replace it, built on top of the GSL.  Good luck.  I think a
lot of mathematica (and formac, maple, and the various algebraic
manipulation languages over the ages) comes down to token parsing and
lexical analysis, though, more than application of numerical algorithms
-- they are in a sense the "easy" part.

   rgb

> 
> Thanks
> Alberto
> -- 
> Departamento de Informatica - Universidade do Minho
> 
> "The C Programming Language -- A language which combines the
> flexibility of assembly language with the power of assembly language."
> 

Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb@phy.duke.edu