[Scilab-Dev] Howto use default installed maths lib under windows + ACML

[Allan CORNET]

Hi,

If you use blasplus.lib to link to your project , you will have any problems

(licence, choice of the library between mkl or atlas or others).

ACML seems to have some troubles to work correctly
(http://bugzilla.scilab.org/show_bug.cgi?id=4523)

Allan

N.B: please notice that blas (atlas, mkl) library for Scilab it is a subset
of the full blas library.
(we did not embedded all blas functions)

To know exported symbols:  dumpbin /exports blasplus.dll


-----Message d'origine-----
De : Vincent [mailto:vincent.lejeune at institutoptique.fr] 
Envoyé : mercredi 12 août 2009 16:54
À : dev at lists.scilab.org
Objet : [Scilab-Dev] Howto use default installed maths lib under windows +
ACML

Hi,

I use some routine from Lapack and Blas for my project ; however, these 
routine have several "version", the raw ones from netlib, the one from
ATLAS, 
the one from the Intel MKL.

Until now, I have mostly worked under Linux, so I only had to link my
project 
against libatlas with a single command flag. However, the window version of 
scilab is shipped with netlib lapack, and the MKL can be installed later.
I'd like my project to call functions in the library installed in the user 
system, without knowing if it is MKL or netlib version. In addition, I don't

know if I have the right to link against MKL, as it is a proprietary library

and I don't have the licence...But I think I have the right to link against
a 
"agnostic" library that can wrapp call to MKL function.
Is there any way to do so ?

On a different topic, I've found that AMD provide their own version of MKL, 
it's called ACML. Contrary to MKL, it's free and redistribuable,as long as 
their "redistribution agreement" is signed. Is there any plan  to use ACML
as 
a "backend" for matrix computation under SciLab ? I don't know how well it 
does perform against MKL or ATLAS, but it can be easier to use than ATLAS(ie

no config/compile/tune step).

Regards,
Vincent.