[petsc-users] Convert mat SEQAIJ to MPIAIJ

[Jed Brown]

Matt is giving you some misguided advice. All the submatrices in a MatNest
have to share the same communicator. This restriction could be lifted in
the future, but that's how it is now.

Why do you want to concatenate these sequential blocks?

Using MatSetValues() is the standard answer and what I would recommend. You
may also be able to create a new MPIAIJ matrix with the correct local
sizes, then

MatMPIAIJGetSeqAIJ(Ampi,&Aseq,&Bseq,&colmap);
MatCopy(myblock,Aseq,DIFFERENT_NONZERO_PATTERN);

you'll have to do a dummy assembly if you go this route.

On Tue, Sep 4, 2012 at 9:23 AM, Thomas Witkowski <
thomas.witkowski at tu-dresden.de> wrote:

>      Mat seqMat;
>     MatCreateSeqAIJ(PETSC_COMM_SELF, 10, 10, 0, PETSC_NULL, &seqMat);
>
>     Mat nestMat;
>     MatCreateNest(PETSC_COMM_WORLD, 1, PETSC_NULL, 1, PETSC_NULL, &seqMat,
> &nestMat);
>
> Results in the following error message:
>
> [0]PETSC ERROR: PetscSplitOwnership() line 93 in
> /home/thomas/software/petsc-3.3-p0/src/sys/utils/psplit.c Sum of local
> lengths 20 does not equal global length 10, my local length 10
>   likely a call to VecSetSizes() or MatSetSizes() is wrong.
> See http://www.mcs.anl.gov/petsc/documentation/faq.html#split
> [1]PETSC ERROR: PetscSplitOwnership() line 93 in
> /home/thomas/software/petsc-3.3-p0/src/sys/utils/psplit.c Sum of local
> lengths 20 does not equal global length 10, my local length 10
>   likely a call to VecSetSizes() or MatSetSizes() is wrong.
>
>
> Thomas
>
>
> Am 04.09.2012 17:36, schrieb Matthew Knepley:
>
> On Tue, Sep 4, 2012 at 10:28 AM, Thomas Witkowski <
> thomas.witkowski at tu-dresden.de> wrote:
>
>>  Am 04.09.2012 17:20, schrieb Matthew Knepley:
>>
>> On Tue, Sep 4, 2012 at 9:52 AM, Thomas Witkowski <
>> thomas.witkowski at tu-dresden.de> wrote:
>>
>>>
>>>
>>> http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/Mat/MATNEST.html
>>>
>>>   As I wrote in my initial question, each rank contains one and only
>>> one seqaij matrix, which all should be joined to one global matrix such
>>> that each local matrix is the corresponding diagonal block of the mpiaij
>>> matrix. I think, this does not work with nested matrices?
>>>
>>
>>  Why does this not work? I really think you are making this harder than
>> it has to be.
>>
>>   Mh, maybe I have an incomplete view of the possibilities how to use
>> nested matrices.
>>
>> To become more specific: In the case of two mpi tasks, each containing
>> one seqaij matrix, how to call MatCreateNest? Is this correct:
>>
>> Mat A;
>> MatCreaeteNest(PETSC_COMM_WORLD, 2, PETSC_NULL, 2, PETSC_NULL, V ,&A);
>>
>                                                                  ^^^ This
> should be 1.
>
>     Matt
>
>   and V is defined on rank 0 as
>>
>> Mat V[2] = {seqMat, PETSC_NULL} ;
>>
>> and and rank 1 as
>>
>> Mat V[2] = {PETSC_NULL, seqMat};
>>
>>
>> Thomas
>>
>>     Matt
>>
>>
>>>  Thomas
>>>
>>>
>>>
>>>>
>>>>  Hong
>>>>
>>>>>
>>>>> Thomas :
>>>>>
>>>>>> In my FETI-DP code, each rank creates a SEQAIJ matrix that represents
>>>>>> the discretization of the interior domain. Just for debugging, I would like
>>>>>> to join these sequential matrices to one global MPIAIJ matrix. This matrix
>>>>>> has no off diagonal nnzs and should be stored corresponding to the ranks
>>>>>> unknowns, thus, first all rows of the first rank and so on. What's the most
>>>>>> efficient way to do this? Is it possible to create this parallel matrix
>>>>>> just as a view of the sequential ones, so without copying the data? Thanks
>>>>>> for any advise.
>>>>>
>>>>>
>>>>>
>>>>> http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/Mat/MatCreateMPIAIJConcatenateSeqAIJ.html
>>>>>
>>>>>  Note: entries in seqaij matrices are copied into a mpiaij matrix
>>>>> without
>>>>> inter-processor communication. Use petsc-3.3 for this function.
>>>>>
>>>>>
>>>>>  The function does not do what I expect. For example, if we have two
>>>>> mpi task and each contains one local square matrix with n rows, I want to
>>>>> create a global square matrix with 2n rows. This function create a
>>>>> non-square matrix of size 2n x n.
>>>>>
>>>>> Thomas
>>>>>
>>>>
>>>>
>>
>>
>>  --
>> What most experimenters take for granted before they begin their
>> experiments is infinitely more interesting than any results to which their
>> experiments lead.
>> -- Norbert Wiener
>>
>>
>>
>
>
>  --
> What most experimenters take for granted before they begin their
> experiments is infinitely more interesting than any results to which their
> experiments lead.
> -- Norbert Wiener
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120905/748f11ed/attachment-0001.html>