On 12/9/2015 03:52, Bryan Jurish wrote:
morning all,
I've just released a new module PDL::SVDSLEPc
<https://metacpan.org/release/PDL-SVDSLEPc> to CPAN: it's basically
just a thin wrapper around the (reduced) SVD solver routines from the
SLEPc c library ("Scalable Library for Eigenvalue Problem
Computations", http://slepc.upv.es/). Preliminary tests with the
default configuration run about 4-5x faster than PDL::SVDLIBC
<https://metacpan.org/release/PDL-SVDLIBC> on my machine here (I
believe it's using ARPACK under the hood, which was mentioned on this
list a few years back, but my searches have failed to turn up any
PDL-ARPACK interface code).
It would be nice to have a more general interface to all the SLEPc
functions, but doing that the "right" way would seem to involve a
low-level PDL interface to the underlying PETSc matrix representation
classes (http://www.mcs.anl.gov/petsc/; these guys are using OpenMPI
for parallel computation), and I'm running low on tuits at the
moment. Just something on my wish list for future PDL incarnations :-)
I took a quick look at the types uses by PETSc and it appears
that the plan for PDL Next Gen (PDLng) to support arrays of
structs and a more complete set of C data types should
make bindings simpler. Improved support for linear algebra
and other computational libraries has long been on the
PDL "wish list".
Thanks for the announcement Bryan.
--Chris
------------------------------------------------------------------------------
_______________________________________________
pdl-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/pdl-general
