[Trilinos-Users] Error in AZ_extract_comm_info

Dorian Krause dorian.krause at lu.unisi.ch
Fri Sep 18 06:54:35 MDT 2009


(* sorry, forgot to send to the list *)

Hi Mike,

thanks for your answer. You are right I didn't specify any Column map in
my code. Is this necessary?

In any case, the attached code example reads the MatrixMarket file of
the matrix and solves the system. You can download the matrix file from
here:

http://cip.physik.uni-bonn.de/~dkrause/Lambda.mm

If I get it correct, EpetraExt should take of the correct distribution
of the matrix?! The test is designed to run with 4 processes

I did take a look at this and I found the the following (maybe you know
this already, but just in case ...):

The problem is the routine AZ_extract_comm_info which is called from the
aztec routine AZ_set_MATFREE in AztecOO::SetUserOperator (line ~1626) in
my version. Looking at what happens there is that the array vec_data is
exchanged using the routine Epetra_Aztec_comm_wrapper. If I look at the
vec_data after the exchange on processor 2 it looks like

2 2 2 .... 2 2 0 0 0 ... 0 0 2 2 2 ... 2 2 2

So the zeros (from processor 0) are in fact not at the end. However I
was not yet able to figure out why.

Thanks for your help!

Dorian

Heroux, Michael A wrote:
> Dorian,
>
> AztecOO constrains the column map of the matrix such that column IDs 
> associated with rows on other processors must be ordered after columns 
> IDs that match row IDs on the same processor.  Epetra will 
> automatically organize the column map this way if allowed to do so. 
>  Are you some how specifying the column map?  I don’t see this 
> happening in your code, so maybe there is something going on with your 
> matrix.
>
> If you want to put your matrix somewhere for me to fetch, please do 
> so, and send your test code.  I can look at it.
>
> Mike
>
>
> On 9/17/09 4:16 PM, "Dorian Krause" <dorian.krause at lu.unisi.ch> wrote:
>
>     Hi again,
>
>     I was able to reproduce this problem using the ex1.cpp from the
>     aztecoo
>     examples and just replacing the construction of the Laplace stiffness
>     matrix by a read from a file using
>     EpetraExt::MatrixMarketFileToCrsMatrix.
>     Are there any constraints on the input matrices to AztecOO? I
>     could send
>     the test example (its ~ 3 MB so I think it's not a good idea to attach
>     it to this mail) if someone is interested ...
>
>     I'm sorry if this problem has already been discussed elsewhere: I
>     tried
>     to search the trilinos-user archive but couldn't get beyound the
>     password pop-up.
>
>     Thanks,
>     Dorian
>
>     Dorian Krause wrote:
>     > Hi,
>     >
>     > using AztecOO to solve a distributed system, gives me the error
>     >
>     > AZ_extract_comm_info: Received elements must be stored after
>     >                    all 1107 local elements
>     >
>     > My code looks as follows:
>     >
>     > ========================================
>     >
>     > map  = new Epetra_Map(-1, nrows, 0, *communicator);
>     > vecX = new Epetra_Vector(*map);
>     > vecB = new Epetra_Vector(*map);
>     > mat  = new Epetra_CrsMatrix(Copy,*map, /* maximal bandwith */);
>     > /* Insert into matrix */
>     >
>     > problem = new Epetra_LinearProblem(mat, vecX, vecB);
>     > solver  = new AztecOO(*problem);
>     >
>     > solver->SetAztecOption( AZ_solver, AZ_cg);
>     > solver->SetAztecOption( AZ_precond, AZ_Jacobi);
>     >
>     > /* initialize vecX and vecB */
>     >
>     > solver->Iterate(maxiter,TOL);
>     >
>     > ========================================
>     >
>     > If I understand it correctly the matrix mat created with this code is
>     > distributed rowwise on the vectors, right?
>     > Where might this error come from? I checked that all indices are in
>     > range but I don't know what else I should look at ...
>     >
>     > Many thanks in advance,
>     > Dorian
>     >
>     >
>     >
>     > _______________________________________________
>     > Trilinos-Users mailing list
>     > Trilinos-Users at software.sandia.gov
>     > http://software.sandia.gov/mailman/listinfo/trilinos-users
>     >
>     >  
>
>
>     _______________________________________________
>     Trilinos-Users mailing list
>     Trilinos-Users at software.sandia.gov
>     http://software.sandia.gov/mailman/listinfo/trilinos-users
>


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: ex1.cpp
Url: https://software.sandia.gov/pipermail/trilinos-users/attachments/20090918/523e2208/attachment-0001.pl 


More information about the Trilinos-Users mailing list