[petsc-dev] Parallel Gauss - Seidel

Barry Smith bsmith at mcs.anl.gov
Mon Jun 11 11:04:47 CDT 2012


   Currently we have MatGetRowIJ((Mat mat,PetscInt shift,PetscBool  symmetric,PetscBool  inodecompressed,PetscInt *n,PetscInt *ia[],PetscInt* ja[],PetscBool  *done)

which does the compression based on inode information. We could add another flag that indicated we wanted compression based on the block size for AIJ matrices.

    Barry

On Jun 11, 2012, at 11:01 AM, Mark F. Adams wrote:

> As far as how to proceed, now that we are insisting that the block size is set early in matrix creation is there a way to get a vertex graph from a blocked matrix?  The MIS code has to manually collapse a blocked matrix, which is ugly.  Perhaps I could move this code into MATAIJ at least.
> 
> There is not hurry to do this so I'd like to look at doing this better than MIS.  We can talk about this when I'm in town Wed and Thurs (I will also be coming in July).    Maybe we could add a getGraph method to Mat ...
> 
> Should I work in SOR?  I could add a struct* to PC_SOR that would hold all my junk and a setup method would have to be called.
> 
> Mark




More information about the petsc-dev mailing list