[Ipopt-tickets] [Ipopt] #153: linearsolverloader pardiso interface out of date
Ipopt
coin-trac at coin-or.org
Thu Apr 7 16:29:42 EDT 2011
#153: linearsolverloader pardiso interface out of date
---------------------+------------------------------------------------------
Reporter: jdh | Owner: ipopt-team
Type: defect | Status: new
Priority: normal | Component: Ipopt
Version: 3.9 | Severity: normal
Resolution: | Keywords:
---------------------+------------------------------------------------------
Comment (by jdh):
I've now got a prototype version of linearsolverloader that copes with
this by detecting the presence of the new interface and providing wrappers
to it for the old interface. I remain convinced that we really should do
this the other way around for future proofing reasons (presumably at some
stage IPOPT will support the iterative method part of PARDISO). I'm still
waiting for the legal dept to greenlight code contribution to IPOPT, so
can't post it here yet.
This has highlighted another problem however - at present if no --with-
pardiso option is used then regardless of OMP_NUM_THREADS the pardiso
number of threads is set to 1 as HAVE_PARDISO_PARALLEL is not set. This
leads to the confusing situation of no speedup up on wall clock time, but
an n-fold increase in CPU time where n is the number of CPUs. Again, this
is probably best dealt with by defaulting to assuming a parallel Pardiso
unless a non-parallel version is supplied. A simple boolean function could
be added to linearsolverloader to say yes or no to the parallel version
being present.
Jonathan.
--
Ticket URL: <https://projects.coin-or.org/Ipopt/ticket/153#comment:1>
Ipopt <http://projects.coin-or.org/Ipopt>
Interior-point optimizer for nonlinear programs.
More information about the Ipopt-tickets
mailing list