Scientific computation [was Conference-DVDs arrived]

[François Chaplais]

Le 27 févr. 2010 à 10:09, Pierre Sahores a écrit :

> Hi Jacque and All,
> 
> As you, for sure, know too, there are lots of world-class companies that are using Rev but, for yet, they don't want to communicate about this ! Why ?
> 
> Revolution is the environment of development which allows scientists and researchers of very high level to elaborate their own workflow without having to pass by caudines forks of their IT departments. Every time, that a physicist, a bilogiste, a chemist, a doctor or a psychologist cannot be any more listened as high as his waits(expectations) by a hard line computer specialist, he puts his co-workers in the work and, sometimes an outside consultant, so that a solution is found, thanks to Revolution.
> 
> Hundred years ago, no safety(salute) outside the horse-drawn carriage and tomorrow our cars will be electric and tomorrow even, as you say it so well, the programming will have become an art and the right way to go with general dev tasks. We, of Edimburg up to all the community of Rev developers, us all share the adventure of the precursors and the discoverers and it is so much the better. 2020 is not so far:-D
> 
> As an example, EADS Astrium Transportation (Ariane rockets) is using Rev. Telling more is not allowed (NDA)...
> 
> Friendly yours,
> 
> Pierre
> 
I have worked with Astrium (at the time Aerospatiale in the past) and their main computation workload was done on SUN workstations (most likely windows PCs by now) using Matlab. Of course, as they went closer to the actual implementation of, say, GNC routines, they would use tools that were closer to the embedded processor...

I guess you know the price of a full professional Matlab license (plus some extra toolboxes)...
Still, many people use Matlab for engineering computations because Matlab does some very good job at providing numerical solutions.
Freeware alternatives (like scilab) have all started (IMHO) by reusing Fortran Libraries and hooking them to a syntax engine (the original Blaise at INRIA was based on Lisp, I believe). As a consequence, the implementation as a package is, to say the least, unpleasant to use.

Now, on the other side, Rev has a very nice language, but non existent support for, let us say, C style arrays of fixed size double precision floats. What I mean is that variable typing somehow is not really supported in transcript. In many cases this is a blessing, but when you go for pure computing performances, it's too bad.

I will take an example. Go to
http://www.macresearch.org/performance_tutorial_part_i_introducing_accelerate
and you will see that MacOS X included vector optimized of BLAS and LAPACK (public domain numerical package for linear computations, for those who are not in this field). Now, because of their history, parameters must be passed using the FORTRAN conventions. I think this can be done in the rev externals. However, I do no know if it is possible to pass, say, a double float array from runrev.

If this could be done, this could be very beneficial to the scientific (in the broad sense) community because the core of matlab's routines could be ported to revolution, with the benefit of having the wonderful rev environment.

If, in addition to this, externals based on the accelerate framework could be used in the now free revmedia, this would be a tremendous opportunity for the scientific community (especially in college) to drop the pricey matlab in favor of revmedia.

Unfortunately, I am not geeky enough to do this by myself...

cheers
	François


François Chaplais
MINES ParisTech
Centre Automatique et systèmes
Ecole des Mines de Paris
http://cas.ensmp.fr/~chaplais/en/