Copying this to the list since it involves a coding error that is very easy to make...
>> I think the following line is problematic: >> >> // Solucion actual del problema no lineal --> V >> AutoPtr<NumericVector<Number> > V (system.solution); >> >> This constructs an AutoPtr "V" from the existing AutoPtr >> system.solution using the copy constructor. When that happens, >> ownership of the system solution vector is released to V, and when V >> goes out of scope, the vector will be destroyed... I think this is >> the main problem with the code, but there were a couple of other >> things: >> >> .) Make sure to close V after the call to V->add(Uold) and Uold after >> the call to Uold.add(*V). These *might* not be bugs in serial code, >> but they definitely will be in parallel. > I have replaced > AutoPtr<NumericVector<Number> > V (system.solution); > by > AutoPtr<NumericVector<Number> > V (system.solution->clone()); > > and it works! but still left a lot of things to do, for instance: to check > my formulation and, as you have mentionned before, to replace some sentences > like Uold.add(*V), that could be a source of future bugs. How should I add a > vector in order to avoid problems with parallel codes? Uold.add(*V) is fine, you just need to be sure and call Uold.close() so that the addition can be properly communicated in parallel. -- John ------------------------------------------------------------------------------ Got Input? Slashdot Needs You. Take our quick survey online. Come on, we don't ask for help often. Plus, you'll get a chance to win $100 to spend on ThinkGeek. http://p.sf.net/sfu/slashdot-survey _______________________________________________ Libmesh-users mailing list Libmesh-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-users
