[petsc-dev] Expensive VecDuplicate_MPI_DA

[Jed Brown]

"Smith, Barry F." <bsmith at mcs.anl.gov> writes:

>   Ahh yes. Since the layout can get large we wanted to avoid each vector having its own copy. 
>
>    But it seems sloppy since DMCreateGlobalVector_DA() does keep a new copy for each vector.
>
>    We could revisit the entire VecDuplicate/DMCreateGlobalVector() infrastructure for all the vector types including the DM ones to 
>
> 1) prevent collective calls 
> 2) reference the layout instead of making new ones. 
>
>
>   Barry
>
> It looks like VecDuplicate_MPI() does the right thing.

Yes, we just don't have an analogous interface for DM.