[petsc-users] MatPermute problem because of wrong sized index set

[Eda Oktay]

Hello everyone,

I am trying to permute a matrix by using a sorted eigenvector in order to
partition the matrix. I sorted the vector and got the index set idx. I am
using 2 processors and I have to divide the vector into 2 according to the
sign of elements (first negative signed ones, then positive signed ones)
first, so I wrote the code below (if-else part). After sorting and dividing
them, I need to permute the matrix with these index sets.

However, in last line (MatPermute), I got the following error:

[0]PETSC ERROR: --------------------- Error Message
--------------------------------------------------------------
[0]PETSC ERROR: Argument out of range
[0]PETSC ERROR: Index 13 is out of range
[0]PETSC ERROR: See http://www.mcs.anl.gov/petsc/documentation/faq.html for
trouble shooting.
[0]PETSC ERROR: Petsc Release Version 3.11.1, Apr, 12, 2019
[0]PETSC ERROR: ./son_without_parmetis on a arch-linux2-c-debug named
5470.wls.metu.edu.tr by edaoktay Mon Jun 17 10:11:54 2019
[0]PETSC ERROR: Configure options --download-mpich --download-openblas
--download-slepc --download-metis --download-parmetis --download-chaco
--with-X=1
[0]PETSC ERROR: #1 PetscLayoutFindOwner() line 249 in
/home/edaoktay/petsc-3.11.1/include/petscis.h
[0]PETSC ERROR: #2 PetscSFSetGraphLayout() line 545 in
/home/edaoktay/petsc-3.11.1/src/vec/is/utils/pmap.c
[0]PETSC ERROR: #3 MatPermute_MPIAIJ() line 1622 in
/home/edaoktay/petsc-3.11.1/src/mat/impls/aij/mpi/mpiaij.c
[0]PETSC ERROR: #4 MatPermute() line 5095 in
/home/edaoktay/petsc-3.11.1/src/mat/interface/matrix.c
[0]PETSC ERROR: #5 main() line 354 in
/home/edaoktay/petsc-3.11.1/arch-linux2-c-debug/share/slepc/examples/src/eda/son_without_parmetis.c
[0]PETSC ERROR: PETSc Option Table entries:
[0]PETSC ERROR: -f
/home/edaoktay/petsc-3.11.1/share/petsc/datafiles/matrices/binary_files/LFAT5_binary
[0]PETSC ERROR: -weighted
[0]PETSC ERROR: ----------------End of Error Message -------send entire
error message to petsc-maint at mcs.anl.gov----------

I also found out that when I look at the submatrix Ais, I got different
sized matrix. For example, I used 14*14 matrix, I get the vector of length
14 but then idx is divided into 5 and 7 elements so Ais is of size 12. 2 of
the positive signed elements are not counted and I didn't understand why.

I also checked the if-else part and I found out that this portion of the
code is ran by processor 0.  Can be this the problem for different sized
Ais? And how can I fix that problem?

Thanks,

Eda


PetscMPIInt rank,size;
  MPI_Comm_rank(PETSC_COMM_WORLD, &rank);
  MPI_Comm_size(PETSC_COMM_WORLD, &size);

  PetscInt kk;

  for (i=0;i<siz;i++){
      if (avr[i] < 0){
          ++kk;
      }
  }

  PetscInt *idxx;
  IS iskk;
  PetscInt sizeofidxx;
  if (rank == 0){

      PetscMalloc1(kk,&idxx);
      j = 0;
      for (i=0; i<kk; i++){
          idxx[j] = idx[i];
          j++;
      }
  } else{
      PetscMalloc1(siz-kk,&idxx);
      j = 0;
      for (i=siz-kk-2 ;i<siz; i++){
          idxx[j] = idx[i];
          j++;
      }
  }

  sizeofidxx = j;
  ISCreateGeneral(PETSC_COMM_WORLD,sizeofidxx,idxx,PETSC_COPY_VALUES,&is);
  ISView(is,PETSC_VIEWER_STDOUT_WORLD);

  /*Permute matrix L (spy(A(p1,p1)))*/

  Mat Ais;
  MatCreateSubMatrix(A,is,is,MAT_INITIAL_MATRIX,&Ais);
  ierr =ISSetPermutation(is);CHKERRQ(ierr);

    /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
                    Create Partitioning
     - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
    PetscInt mA,nA,mL,nL;
    MatGetSize(Ais,&mA,&nA);
    MatGetLocalSize(Ais,&mL,&nL);
    PetscPrintf(PETSC_COMM_WORLD," Size of Ais: %D,%D\n",mA,nA);
    PetscPrintf(PETSC_COMM_WORLD," Size of local Ais: %D,%D\n",mL,nL);

  ierr = MatPermute(Ais,is,is,&PL);CHKERRQ(ierr);
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20190617/7f79e8eb/attachment.html>