On Fri, Jun 20, 2008 at 8:31 AM, Jed Brown <[EMAIL PROTECTED]> wrote: > On Tue, Jun 17, 2008 at 15:22, John Peterson <[EMAIL PROTECTED]> wrote: >> On Tue, Jun 17, 2008 at 8:09 AM, Jan Biermann <[EMAIL PROTECTED]> wrote: >>> Hi, >>> I would like to use ILU preconditioning in parallel. Here, Petsc provides an >>> interface to BlockSolve95 but requires the matrix format MPIBAIJ. So far >>> MPIAIJ is standard in libmesh. Do you think I can just change the matrix >>> format or does that cause any problems (with parmetis or whatever)? >> >> I've often wondered this myself...unfortunately I don't know enough >> about the different Petsc matrix types to say for sure. If you do get >> something working, for example a user-selectable PETSc matrix type, we >> would definitely be interested in getting it in the library. > > I'm not sure what you mean and I don't have source code in front of > me, but what is wrong with this? > > MatCreate() > MatSetType() // default > MatSetFromOptions() > MatMPIAIJSetPreallocation() > MatMPIBAIJSetPreallocation() > MatSeqAIJSetPreallocation() > MatSeqBAIJSetPreallocation() > // you can call all of these and they will be a no-op if the type does > not match.
That's good to know...I didn't mean to say it was difficult to do, just that we don't currently have the code for it. Since you can call all of the routines and it will just drop the ones it's not using, that's one way to go. On the PETSc side I wish it could be cleaner...why have two totally separate function names instead of just letting the arguments' types determine the behavior? -- John ------------------------------------------------------------------------- Check out the new SourceForge.net Marketplace. It's the best place to buy or sell services for just about anything Open Source. http://sourceforge.net/services/buy/index.php _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
