>> I think a >> VecDuplicate followed by a VecCopy should do the trick, and I'll try >> that out on the examples and on our few unit tests today if I find >> time, but this is why I wanted other people looking at the problem; >> I'm not familiar enough with PETSc's APIs > > Nor am I, as you see. That's just the problem. It would actually be > nice if additional people would look at the code as well. On the > other hand, having at least you and me looking at the PETSc part of it > (and hopefully you and Ben on the "other" part) is at least better > than only one person alone.
Indeed. I've just started catching up on some stuff, and am happy to see the direction this is taking... I'd still like to ultimately change the PetscVector() to use VecCreateGhostWithArray() (or whatever the proper name is). In that approach we would allocate our own buffer (preferably in the NumericVector base class) and this bypass a lot of the Petsc API, especially in accessing individual elements. This is a big step to help make that happen, though. Once we get away from the notion of localizing a parallel vector to a serial one as standard practice the rest is a detail which will fall in to place. -Ben ------------------------------------------------------------------------------ This SF.net email is sponsored by: SourcForge Community SourceForge wants to tell your story. http://p.sf.net/sfu/sf-spreadtheword _______________________________________________ Libmesh-devel mailing list Libmesh-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-devel