Re: gslclapack

[Gerard Jungman <jungman at lanl dot gov>]

On Wed, 2006-03-29 at 11:44 -0600, Linas Vepstas wrote:
> I found this conversation a tad too abstract.
> 
> On Tue, Mar 28, 2006 at 05:30:07PM -0700, Gerard Jungman wrote:
> >     I am hoping that the clapack standard
> >     says something about this situation; after all, it
> >     ought to be designed to work well with cblas.
> 
> I don't udnerstand. A google of clapack reveals things like 
> http://www.netlib.org/clapack/
> 
> which seems no better than calling the fortran libs directly 
> from c code. And the problem with this is, of course, that 

Sorry. I was not precise enough. There is a thing called "clapack",
which is disgusting (I think it is basically f2c run on the lapack
source, followed by twiddling). What I mean by a "C lapack" is
whatever the new development lapack cycle is going to produce.
Currently, Dongarra et al are supposed to be working on improved
versions of lapack and sca-lapack. This new work includes
standardized wrappers for a set of languages, like C, python, f95, etc.
They propose to implement in f77, as before, which is understandable.

The (funded) proposal is available at
http://www.cs.berkeley.edu/~demmel/Sca-LAPACK-Proposal.pdf

The announcement related to this can be found by 
searching for "New Release of LAPACK and ScaLAPACK planned"
on google-groups.

They state explicitly that they want to use "better interfaces
using features of up-to-date programming langauges" to "hide
some of the complex distributed data structures and other features".
These are exactly the kind of interface improvements we want,
so it would be great if they could sort it all out and
produce something attractive.


> > If somebody dropped a working lapack interface on the table,
> > I would certainly want to put it in GSL. It's a clear win.
> 
> I presume you mean "a C wrapper that solves the strange 
> calling convention, documentiation and workspace problems". ?

Exactly, as per the above.

Linas, thanks for the chance to clarify what I meant!

-- 
Gerard Jungman <jungman@lanl.gov>
Los Alamos National Laboratory