[SciPy-user] Re: [Numpy-discussion] Sparse matrices
eric at enthought.com
Thu Aug 1 15:08:01 CDT 2002
I have never used PETSc, but have looked at it before and just took
another look. It looks like a very comprehensive and powerful library.
One of the main issues of adding something like PETSc to SciPy would be
the huge build complexity that comes with it. It is a very large
library with Unix centric build tools that require cygwin on windows.
Trying to cram this into distutils would be, ahem..., unpleasant.
The second issue is its emphasis on parallel solvers and the complexity
that this adds, both for building software and for developing it.
Parallel libraries typically have multiple extra parameters for
communicators, process_id, etc. to many of their functions. Also,
usually multiple calls are needed to scatter/assemble vectors, matrices,
etc. for parallel algorithms. This stuff is all necessary for parallel
stuff, but being burdened with it in a serial library like SciPy adds to
SuperLU has a fairly large user base, is small relative to PETSc, and
has a good reputation for performance. It, or something like it, fits
better into the modular approach that SciPy has.
That said, PETSc has probably conquered a number of the hairy parallel
issues that show up in vector/matrix solvers. Perhaps it would form a
solid foundation for a parallel version of SciPy that works with pyMPI?
(Scalapack has proven difficult to wrap in the past). I'm glad to hear
that there is an effort underway to make it Python accessible. Is there
a list to follow this development? I'd be interested in watching it.
> -----Original Message-----
> From: scipy-user-admin at scipy.net [mailto:scipy-user-admin at scipy.net]
> Behalf Of Paul LE TEXIER
> Sent: Friday, August 02, 2002 3:07 AM
> To: scipy-user at scipy.net
> Cc: Paul.letexier at cea.fr
> Subject: [SciPy-user] Re: [Numpy-discussion] Sparse matrices
> May be its stupid, but why don't you use PETSc ?
> I think there is a lot of VERY good thinks in this framework,
> not only for hight performance computations !
> In a near future (ask people from ANL/PETSc) we will have
> a PETSc componant (cf. cca-forum.org Common Componant Architecture)
> using SIDL (Scientific Interface Description Language).
> So many PETSc should be available (even Python !).
> Have a lice day.
> eric jones a écrit :
> > No one has worked on the sparse stuff as far as I'm aware, so the
> > package in scipy is more a place holder than an implementation. I
> > thought there were some licensing issues, but, looking at the
> > docs, the license looks fine. Maybe the license issue was with a
> > different sparse matrix package Travis O. had used. I know that he
> > wrapped more than one in the past.
> > eric
> > > -----Original Message-----
> > > From: scipy-user-admin at scipy.net
[mailto:scipy-user-admin at scipy.net]
> > On
> > > Behalf Of Jochen Küpper
> > > Sent: Thursday, August 01, 2002 12:30 AM
> > > To: scipy-user
> > > Cc: Victor S. Miller
> > > Subject: [SciPy-user] Re: [Numpy-discussion] Sparse matrices
> > >
> > > There seems to be a problem with the sparse module with respect to
> > > files missing in the snapshot:
> > >
> > > On Thu, 01 Aug 2002 00:02:43 -0400 Victor S Miller wrote:
> > >
> > > Victor> Jochen, I downloaded the source built today (7/31/02).
> > > what's
> > > Victor> in the sparse directory. The SuperLU directory just has
> > file
> > > Victor> make.inc. I guess I'll have to supplement it with the
> > >
> > > Victor> /usr/local/coop/src/SciPy-0.2.0_alpha_105.3699/sparse:
> > > Victor> total 3
> > > Victor> drwxr-xr-x 26 victor 1024 Jul 31 17:02 ..
> > > Victor> drwxr-xr-x 3 victor 512 Jul 31 04:59 .
> > > Victor> drwxr-xr-x 2 victor 512 Jul 31 04:59 SuperLU
> > >
> > >
> > > Greetings,
> > > Jochen
> > > --
> > > Einigkeit und Recht und Freiheit http://www.Jochen-
> > > Kuepper.de
> > > Liberté, Égalité, Fraternité GnuPG key:
> > > Sex, drugs and rock-n-roll
> > >
> > > _______________________________________________
> > > SciPy-user mailing list
> > > SciPy-user at scipy.net
> > > http://www.scipy.net/mailman/listinfo/scipy-user
> SciPy-user mailing list
> SciPy-user at scipy.net
More information about the SciPy-user