[GRASS-dev] Re: [GRASS-PSC] BLAS/LAPACK

[Brad Douglas]

On Thu, 2007-09-06 at 15:24 +1200, Hamish wrote:
> Brad Douglas wrote:
> > Both Soren and myself would like to replace the BLAS/LAPACK (Fortran
> > code) libraries with ATLAS, a tuned C version.  This also gets us away
> > from the Fortran issues suffered by gcc4.
> >
> > If there are no objections, I would like to go ahead and modify
> > configure.in and include/ to reflect this. 
> ..
> > One thing we have not discussed yet is possibly merging lib/gmath and
> > lib/gpde.  Right now, we should probably leave it alone.  I haven't
> > checked to see what all uses lib/gpde.
> 
> Hi,
> 
> IMO important technical decisions such as this need to happen over on
> the grass-dev mailing list to allow contributions from all devels and
> a single source for future thread archive searches.

I've already discussed most of this without objection on the dev list.
cc'ing this email.

> and FWIW, fortran90 support in the latest gcc4 releases seems to be much
> improved over earlier versions. (but AFAIK grass has ever only linked to
> f77 stuff)

That requires that BLAS/LAPACK be ported to f90.  BTW, I have created a
work-around that allows gcc4 to still use f77.

Regardless, BLAS/LAPACK needs better definition in autoconf.  I have
autoconf code that allows us to pick and close the BLAS/LAPACK
implementation that best suits our needs.  That could be traditional f77
BLAS/LAPACK, ATLAS, SciLAPACK, etc.

This gives the person compiling the ability to "tune" to their
application (cluster, SMP, SMP cluster).

Unfortunately, I haven't been able to test it (autoreconf issues).


-- 
73, de Brad KB8UYR/6 <rez touchofmadness com>