[SciPy-user] testing SciPy without ATLAS ;-(

[Pearu Peterson]

On Tue, 28 May 2002, Pearu Peterson wrote:

> 
> On Tue, 28 May 2002, Gerard Vermeulen wrote:
> 
> > On Tuesday 28 May 2002 00:02, Pearu Peterson wrote:
> <snip>
> > > So, what are the suggestions if ATLAS is not available:
> > > 1) to require the standard BLAS and the standard LAPACK?
> > > 2) or to require the standard LAPACK with the included but incomplete
> > > BLAS?
> > >
> > > I would prefer (1) because linalg claims to have wrappers to the BLAS
> > > routines. However, if it is common that distributors provide only
> > > incomplete BLAS, we can consider (2) or let users to complain to
> > > distributors.
> > >
> > > What do you think?
> > >
> > Or fixing up BLAS in the way you suggested masking the "post-update
> > LAPACK routine". Maybe getinfo("blas") can peek into BLAS (using nm) to
> > see what is available and provide a list of extra fortran sources to
> > be compiled and are included with SciPy. If you like I can start to look
> > for them, tonight.
> 
> Unfortunately it is not that easy. This information is needed already 
> in interface_gen.py that generates the *.pyf files. It is always possible
> to add additional hooks but I would prefer that these hooks would be
> as simple as possible.

Actually we can (i) factor out BLAS routines that are not included in
LAPACK, (ii) put them into a separate .pyf file, (iii) include this file
to generic_fblas.pyf, and (iv) let users comment out the corresponding
include statement if they have an incomplete BLAS. 
How this sounds? End users need to learn only the last step, of course.

Pearu