On Tue, 29 Sep 2009, Derek Gaston wrote: > Is there something specific I can do to recreate the bug to help test?
Your ICES/CFDLab login is still active? If so, copy ~roystgnr/libmesh/svn/src/apps/adjoints_test_poisson/application_code/ and try rebuilding and re-running heat-dbg therein. You may need "-pc_type jacobi" to work around a PETSC2 bug. Compare a serial run to a 2 or 4 processor parallel run. The latter will cut a partition boundary across a hanging node, and will leave you with a discrepancy at that node (try ~roystgnr/bin/subtractgmv.py out.gmv.0002.serial out.gmv.0002.parallel to see it better) and a huge error estimate (gmv -i error.gmv.0000.parallel) on the adjoining cells. I'm not optimistic - I managed to replicate the error on lonestar, so it's not ICES-specific. Now I just need to figure out why my home computer is magically immune... --- Roy > Derek > > On Sep 28, 2009, at 8:44 PM, Roy Stogner wrote: > >> >> On Mon, 28 Sep 2009, Roy Stogner wrote: >> >>> Vikram couldn't get this workaround to work, and trying it again >>> neither can I; I must have been looking at the wrong output files >>> before. This must not be a send_list problem after all. >>> >>> It's not a ghosted vectors problem, either; or at least rebuilding >>> with --disable-ghosted gives the same result. I'm very busy this >>> week, but I'm still looking when I get the time. >> >> Bug update: I was "looking at the wrong output files" when I looked at >> the files produced by my home computer, which doesn't exhibit this >> bug, workaround or not. It is still occurring in the ICES environment >> that Vikram and I both use. >> >> I'm using the libMesh SVN head on both, same application code, same >> "configure --enable-everything", PETSc 2.3.3 on both... >> >> I've got a few other systems I can try this out on conveniently; at >> least we'll know whether the bug's appearance is the exception or the >> rule. >> --- >> Roy >> >> ------------------------------------------------------------------------------ >> Come build with us! The BlackBerry® Developer Conference in SF, CA >> is the only developer event you need to attend this year. Jumpstart your >> developing skills, take BlackBerry mobile applications to market and stay >> ahead of the curve. Join us from November 9-12, 2009. Register now! >> http://p.sf.net/sfu/devconf >> _______________________________________________ >> Libmesh-devel mailing list >> [email protected] >> https://lists.sourceforge.net/lists/listinfo/libmesh-devel > ------------------------------------------------------------------------------ Come build with us! The BlackBerry® Developer Conference in SF, CA is the only developer event you need to attend this year. Jumpstart your developing skills, take BlackBerry mobile applications to market and stay ahead of the curve. Join us from November 9-12, 2009. Register now! http://p.sf.net/sfu/devconf _______________________________________________ Libmesh-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-devel
