[petsc-users] MatSetValue in Fortran

Wed Feb 9 09:36:08 CST 2022

Sorry for the bad subject , i made a mistake.
The correct email subject and message was :

I'm struggling for a very simple error that i can not see. I'm running 
in a sequential program, for the test.

MatView is giving me for the row 125 :
row 125: (125, 0.)  (107, 0.)

I'm getting those values with MatGetRow :
       row=125
       CALL MatGetRow(MATGLOB,row,nb,testcols,testvalues,IER)
       write(*,*)row,testcols(1),testvalues(1),
      &          testcols(2),testvalues(2)
       CALL MatRestoreRow(MATGLOB,row,nb,testcols,testvalues,IER)

The output is :
       125    125  0.000000000000000E+000     107    0.000000000000000E+000
Which is what i want. It's ok.

Then i'm doing the MatSetValue :
       val = -1
       row = 125
       col = 107
       CALL MatSetValue(MATGLOB,row,col,val,
      &                  INSERT_VALUES, IER)

And i've got the error :
[0]PETSC ERROR: --------------------- Error Message 
--------------------------------------------------------------
[0]PETSC ERROR: Argument out of range
[0]PETSC ERROR: New nonzero at (125,107) caused a malloc
Use MatSetOption(A, MAT_NEW_NONZERO_ALLOCATION_ERR, PETSC_FALSE) to turn 
off this check
[0]PETSC ERROR: See https://petsc.org/release/faq/ for trouble shooting.
[0]PETSC ERROR: Petsc Release Version 3.16.4, unknown
[0]PETSC ERROR: 
/home/jobic/projet/fe-utils/marcus/3.16/test_MatSetValue_loem_3.16p4_openmpi_intel 
on a  named leto4.iusti-calcul.recherche by jobic Wed Feb  9 16:01:51 2022
[0]PETSC ERROR: Configure options 
--prefix=/local/lib/petsc/3.16/p4/17/openmpi_intel-mkl-works 
--with-single-library=0 --with-large-file-io=1 --with-debugging=0 
--with-blacs=1 
--with-blacs-dir=/opt/intel/compilers_and_libraries_2017.1.132/linux/compiler/lib/intel64/ 
--download-scalapack=1 --download-parmetis=1 --download-make=1 
--download-mumps=1 --LIBS=" 
-Wl,-rpath,/opt/intel/compilers_and_libraries_2017.1.132/linux/compiler/lib/intel64/" 
--with-blaslapack-dir=/opt/intel/compilers_and_libraries_2017.1.132/linux/compiler/lib/intel64/ 
--download-metis=1 --download-parmetis=1 --download-ptscotch=1 
--download-cmake=1 --download-slepc=1 --download-hdf5=1 --with-zlib=1 
--download-szlib=1 --download-suitesparse=1 --download-p4est=1 
--download-netcdf=1 --download-triangle=1 --with-shared-libraries=0 
--with-cxx-dialect=C++11 -CFLAGS=" -O3 -mtune=core-avx2 -mkl" 
--COPTFLAGS="-D_POSIX_C_SOURCE=199309L" -CXXFLAGS=" -O3 -mtune=core-avx2 
-mkl" -FFLAGS=" -O3 -mtune=core-avx2 -mkl" 
PETSC_ARCH=openmpi_intel-mkl-17-works
[0]PETSC ERROR: #1 MatSetValues_SeqAIJ() at 
/home/devel/src_linux/petsc-3.16.4/src/mat/impls/aij/seq/aij.c:520
[0]PETSC ERROR: #2 MatSetValues() at 
/home/devel/src_linux/petsc-3.16.4/src/mat/interface/matrix.c:1398
[0]PETSC ERROR: #3 MatGetRow() at 
/home/devel/src_linux/petsc-3.16.4/src/mat/interface/matrix.c:558
[0]PETSC ERROR: #4 MatRestoreRow_Fortran() at 
/home/devel/src_linux/petsc-3.16.4/src/mat/interface/ftn-custom/zmatrixf.c:582

I'm obviously doing something wrong, but where ?

Thanks,

Yann

Le 2/9/2022 à 12:46 PM, Matthew Knepley a écrit :
> On Wed, Feb 9, 2022 at 5:56 AM Yann Jobic <yann.jobic at univ-amu.fr 
> <mailto:yann.jobic at univ-amu.fr>> wrote:
> 
>     Dear All,
> 
>     I'm facing a strange problem. I did not succeed in putting some values
>     in an MPI matrix. I'm using Petsc 3.16.4. The matrix is pre-allocated,
>     with some zeroes at the right position.
> 
>     To explain the context, it's a finit elements code, thus in the tangent
>     matrix creation, i've got a first loop over the elements, and feed the
>     matrix accordingly. This part is working. I'm using ADD_VALUES.
> 
>     I then have to put the periodic boundary conditions on some nodes, that
>     is to say that i've got 1 at the diagonal, and -1 on the mirror element
>     of the designated ddl  (i've got a Newton-Raphson minimisation
>     procedure). I'm using INSERT_VALUES here.
> 
>     According to the documentation, i should do :
> 
>     Loop over Elements
>     call MatSetValues
>     end loop
>     call MatAssemblyBegin(A,MAT_FLUSH_ASSEMBLY,ierr);CHKERRA(ierr)
>     call MatAssemblyEnd(A,MAT_FLUSH_ASSEMBLY,ierr);CHKERRA(ierr)
> 
>     Loop over periodic nodes
>     call MatSetValue
>     end loop
>     call MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY,ierr);CHKERRA(ierr)
>     call MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY,ierr);CHKERRA(ierr)
> 
>     But it's not working. I don't have those values from the periodic
>     boundary condition in the global matrix. The return value of
>     MatSetValue
>     is 0.
> 
>     I tried valgrind, but memory access looks ok. I really don't know
>     how to
>     debug this. Do you have any idea of what could happen here ?
>     Debug ideas ?
> 
> 
> This is strange since we have tests for this kind of insertion. I would 
> make a tiny
> code that adds two neighboring cells and a periodic boundary between the 
> other side.
> If that fails it will be easy for us to look at the figure out what is 
> happening.
> 
>    Thanks,
> 
>       Matt
> 
>     Many thanks,
> 
>     Best Regards,
> 
>     Yann
> 
> 
> 
> -- 
> 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
> 
> https://www.cse.buffalo.edu/~knepley/ <http://www.cse.buffalo.edu/~knepley/>