VecGhost trouble

Barry Smith bsmith at mcs.anl.gov
Wed May 20 12:34:23 CDT 2009


    Jed,

      These are likely there because MatGetVecs() was introduced  
pretty recently and those are places that "never got fixed" when  
MatGetVecs() was introduced.

      Please feel free to fix and push.

     Thanks

    Barry

On May 20, 2009, at 11:49 AM, Jed Brown wrote:

> I have some heavy use of VecGhost with MatShell (so the ghosts are
> required to define MatMult).  While exercising this framework, I have
> noticed some operations that do not work correctly, in particular
>
> * MatComputeExplicitOperator uses VecCreateMPI rather than MatGetVecs
>
> * same pattern in MatNullSpaceTest (if there are no vectors to
> duplicate in the null space, i.e. just the constant)
>
> I think these cases can just use MatGetVecs which I can use to produce
> the correct vectors.  Any problems with this?  (Any reason the current
> code calls VecCreateMPI directly?)
>
> Jed
>




More information about the petsc-dev mailing list