ScaLAPACK Archives

[Scalapack] next ScaLAPACK - missing PDSYGVD, PDSYEVR, etc.

Julien,

See my comments below. Also, I did not get a response to the very last question 
in
my initial e-mail.

I have a distributed matrix, global dimensions N-by-G (G >> N), and each 
processor
will have an equal size block of N (LOCr), but not G (LOCc). It does not look
like ScaLAPACK is set-up to handle this scenario even if this matrix were 
distributed in a 2D block-cyclic manner.

Nichols A. Romero, Ph.D.
Argonne Leadership Computing Facility
Argonne National Laboratory
Building 240 Room 2-127
9700 South Cass Avenue
Argonne, IL 60490
(630) 252-3441


----- "Julien Langou" <julien.langou@Domain.Removed> wrote:

PDSYEVR is there. We did not release it yet but it works fine. Do you
want 
it? We have distributed to few individuals already. The code has been

written by Christof Voemel. We are missing time to add some TESTING
and 
then we can release. People have tried and were satisfied. The
interface 
is about the same as other PDSEYVx.
I checked the ScaLAPACK mail archives and know where I can get the code.


For the generalized symmetric eigenvalue problem PDSYGVD and PDSYGVR,
(I 
know it sounds a joke) but it's not too hard to do oneself ...
    PDPOTRF
    PDSYNGST
    PDSYEV (which ever PDSYEV, PDSYEVX, PDSYEVD, PDSYEVR)
    PDTRSM
done. Look at PDSYGVX. This is in the todo list since a while sorry
about 
that.
Yes, I saw this as well. Just wanted to make sure it was that simple.


Can you describe the type of application that would need these
routines? 
That might motivate sme of us to spend the time writing the drivers.
The applications solves the KS-DFT equations. We have iterative
eigensolver that uses ScaLAPACK for subspace diagonalization and
parallel dense linear algebra. 
https://wiki.fysik.dtu.dk/gpaw/

This is our home grown Python interface to ScaLAPACK:
https://trac.fysik.dtu.dk/projects/gpaw/browser/trunk/c/blas.c

Best wishes,
Julien

On Mon, 11 Jan 2010, naromero@Domain.Removed wrote:

Hi,

This question must get asked a lot and I do not wish to be
annoying:

When is the next release of ScaLAPACK? And will it include:

PDSYGVD, D&C for generalized eigenvalue problem
PDSYEVR, MR^3 for standard eigenvalue problem
PDSYGVR, MR^3 for generalized eigenvalue problem

Also, I know that ScaLAPACK uses the 2D block-cyclic distribution,
but is
there anyway to get irregular distributions like those supported in
SRUMMA?
http://hpc.pnl.gov/projects/srumma/cf06.pdf

I frequently get away with 2D block distribution for PDGEMM
operations (details
can be provided). But having irregular distributions, would seem to
be currently impossible.

Nichols A. Romero, Ph.D.
Argonne Leadership Computing Facility
Argonne National Laboratory
Building 240 Room 2-127
9700 South Cass Avenue
Argonne, IL 60490
(630) 252-3441

_______________________________________________
Scalapack mailing list
Scalapack@Domain.Removed
http://lists.eecs.utk.edu/mailman/listinfo/scalapack


<Prev in Thread] Current Thread [Next in Thread>


For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or