On 01/11/13 06:01, Matthew Knepley wrote:
On Wed, Oct 30, 2013 at 9:19 PM, Rhodri Davies <[email protected] <mailto:[email protected]>> wrote:Dear all, Would you mind giving us an update on your thoughts re: this question? We're very keen to be able to use such functionally and it would be good to know if you're looking into this / have any suggestions? I looked at the code. It does not look hard to make a MPIBAIJ version. Since the scatter context for BAIJ is already exploded, most of the MPIAIJ code goes straight through, it just a matter of handling the blocking in a few places. Here is what would be really helpful for us. Write a nice test. The code would probably only take an hour of being careful. However, the test assembly would be more. If you give us a nice test example for it, writing it could happen pretty quickly. Thanks, Matt
That would be great. I had a look at the existing tests and found src/mat/examples/tests/ex140.c which already claims to test MatZeroRowsColumns with MPIBAIJ - except it isn't being run in parallel. Would running this test in parallel be a sufficient test to work with? It currently fails with the expected message "No support for this operation for this object type!". I'd be happy to write another test but it would basically be creating another matrix and testing it does the right thing with the matrix and the rhs. Let me know if there's anything else we could help out with
Cheers Stephan
From: Stephan Kramer <[email protected]> Subject: Fwd: [petsc-users] MatZeroRowsColumns for mpibaij Date: 31 October 2013 13:16:05 GMT+11:00 To: Rhodri Davies <[email protected]> -------- Original Message -------- Subject: [petsc-users] MatZeroRowsColumns for mpibaij Date: Wed, 19 Jun 2013 13:22:31 +0100 From: Stephan Kramer <[email protected]> To: PETSc users list <[email protected]> Dear all, We have found the MatZeroRowsColumns() routine to be very useful for lifting boundary conditions. Unfortunately we also found that it isn't implemented for matrix type mpibaij and we're quite keen on keeping the baij structure as it seems to give us significant savings for DG problems (in particular the blocked sor). I had a quick look at the implementation of MatZeroRowsColums_MPIAIJ but couldn't really estimate how hard it would be to do something similar for mpibaij (an implementation for seqbaij exists already). Is this something that's likely to be looked at in the future? Stephan
