[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: C++ libraries for scientific computing anyone?



On Wed, Mar 22, 2000 at 04:44:41AM +0000, Stephen J. Turnbull wrote:
> >>>>> "Jose" == Jose L Gomez Dans <j.l.gomez-dans [at] sheffield.ac.uk> writes:
> 
>     Jose> 	However, rather than go through all that, I was
>     Jose> wondering whether someone uses a C++ library for numerical
>     Jose> work.
> 
> All of the econ stats/simulation packages I know of have UIs written
> in C or C++ (or Perl or Python ;-) and link to tried and true robust
> FORTRAN libraries.

	That seems to be the case: since you can compile your fortran
libs with anything and call them from anywhere (I don't know about ADA
:D), the usual take is: OK, you just call the fortran compiled library
from within C/C++, and yes, you can write a wrapper for it. However,
when you need to call functions that need about 20 parameters in
fortran, and you realise that they'll need around 40 in C... erm... If I
could only find the wrappers!

> Have you looked at what Octave and R use?

	Yes, they do use LAPACK. In fact, you can use Octave's library
to link your programs to. But then you have to use Octave's classes and
types, which aren't really standard.

	The best solution I have found so far is the Scalar, Vector and
Template Library. However, the real winner seems to be Fortran 90. I am
yet to download the free personal version of the compiler for Linux, but
I have used it a bit in the Uni's UltraSparc cluster, and it does loads
of nice things (spread code over several processors automagically, has
array-type operations built-in by default...).

	Thanks for the tip, anyway. 
	José

-- 
José L Gómez Dans			PhD student
					Radar & Communications Group
					Department of Electronic Engineering
					University of Sheffield UK

PGP signature