about Jed's changes to MatGetSubMatrix

[Lisandro Dalcin]

On Thu, Jun 4, 2009 at 8:16 PM, Jed Brown <jed at 59a2.org> wrote:
> Lisandro Dalcin wrote:
>> I like a lot the new API, though I would like to comment on the change
>> below and previous behaviour of MatGetSubMatrix() in 3.0.:
>>
>> Author: Jed Brown <jed at 59A2.org>  2009-05-27 15:15:56
>> Committer: Jed Brown <jed at 59A2.org>  2009-05-27 15:15:56
>> Parent: 13687:1da19bea6533 (Update ex19 output for minlambda)
>> Child:  13689:7903ebaee227 (documentation update for the
>> PetscComposedDataGet* routines)
>>
>>     Compose gathered IS to amortize cost of ISAllGather with
>> MatGetSubMatrix (MPIAIJ,MPIBAIJ)
>>
>>
>> It is really required to make the call fail if MAT_REUSE_MATRIX is
>> requested, but the submat is not the same that the original one but
>> anyway have appropriate layout and nonzero pattern?
>>
>> See this bit of Python code, it currently fails, but perhaps it should not?
>>
>> S1 = A.getSubMatrix(isrows,iscols) # uses MAT_INITIAL_MATRIX
>> S2 = S1.duplicate()
>> A.getSubMatrix(isrows,iscols,S2) # uses MAT_REUSE_MATRIX
>
> Did this really work in the past (in parallel)?  The MPIXAIJ
> implementation uses a sequential matrix to extract values.  It is
> composed with the object under the name "SubMatrix", and errors if that
> doesn't exist, so I'll be very confused if it did.
>

No, it did not worked. I'm not saying your implementation was
sub-optimal :-) ... Just polling to figure out if it could / is worth
to improve.

> We could create a new one if it doesn't exist, even with
> MAT_REUSE_MATRIX, but that feels wrong to me.

Why do you have such feeling?

> Implementation is
> trivial, but I think it's likely to be confusing and it becomes more
> difficult to give meaningful error messages.  What do you think?
>

Why do you think it is confusing? As long as the reuse submat has a
proper parallel layout (after a fast look, it seems you do not even
have constraints on nonzero pattern), the call would work.



-- 
Lisandro Dalcín
---------------
Centro Internacional de Métodos Computacionales en Ingeniería (CIMEC)
Instituto de Desarrollo Tecnológico para la Industria Química (INTEC)
Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET)
PTLC - Güemes 3450, (3000) Santa Fe, Argentina
Tel/Fax: +54-(0)342-451.1594