dolfin team mailing list archive

[Niclas Jansson <njansson@xxxxxxxxxx>]

Anders Logg <logg@xxxxxxxxx> writes:

> On Tue, Feb 23, 2010 at 11:44:44PM +0000, Garth N. Wells wrote:
>>
>>
>> Anders Logg wrote:
>> > On Tue, Feb 23, 2010 at 10:38:38PM +0000, Garth N. Wells wrote:
>> >>
>> >> Marie Rognes wrote:
>> >>> Am I the only one getting the following error with a fresh dolfin?
>> >>>
>> >> I've commented out the offending function. I don't see why it was added.
>> >>
>> >> Garth
>> >
>> > It's in the commit log. :-)
>> >
>>
>> Where's the implementation? I don't see why it was added without an
>> implementation.
>
> The .cpp file was missing. It's there now.
>
>> I also think that this isn't a good idea from the la standpoint,
>> especially in parallel.
>
> It works through ident() which works fine in parallel. The problem is
> getrow() that doesn't work in parallel. There might be other ways to
> figure out which rows are zero.
>

Hmm, getrow() works in parallel but is restricted to local rows. There
is a workaround that I've used in unicorn, use submatrices in PETSc to
create ghosted copies of the off processor rows that are needed. 

Of course, it's superior to change data layout/ algorithms such that the
program never depends on off processor rows.


Niclas