I reckon that what corresponds best to my situation is to create my own
MPIAIJ matrix, recover the DMDA local to global mapping, preallocate
myself and fill my values with MatSetValues.
However, what is the correct way to call
ISLocalToGlobalMappingGetIndices in Fortran? It seems there's something
missing in the documentation as my code just won't work.
Some examples include a PetscOffset argument which is not present in the
documentation and there is no manual page for
ISLocalToGlobalMappingGetIndicesF90: the only information I could find
is on /src/vec/is/utils/f90-custom/zisltogf90.c. Are you supposed to
allocate the index array yourself? I tried declaring it as a pointer
too. I keep getting an error
[0]PETSC ERROR: #1 User provided function() line 0 in User file
application called MPI_Abort(MPI_COMM_SELF, -42924) - process 0
Minimal example:
PROGRAM test_dmda
#include <petsc/finclude/petscdm.h>
#include <petsc/finclude/petscdmda.h>
USE PetscDMDA
IMPLICIT NONE
PetscErrorCode :: ierr
PetscInt :: irank, icx, icy, lx, ly, n_ltog
DM :: da
ISLocalToGlobalMapping :: ltog
PetscInt, DIMENSION(:), ALLOCATABLE :: id_ltog
!PetscInt, DIMENSION(:), POINTER :: id_ltog
PetscInt, PARAMETER :: nx=24, ny=9, neq=4
CHARACTER(LEN=100) :: err_msg
CALL PetscInitialize(PETSC_NULL_CHARACTER,ierr)
CALL MPI_COMM_RANK(PETSC_COMM_WORLD,irank,ierr)
CALL
DMDACreate2D(PETSC_COMM_WORLD,DM_BOUNDARY_NONE,DM_BOUNDARY_NONE,DMDA_STENCIL_STAR,(nx+1),(ny+1),
&
PETSC_DECIDE,PETSC_DECIDE,neq,0,PETSC_NULL_INTEGER,PETSC_NULL_INTEGER,da,ierr)
CHKERRA(ierr)
CALL DMSetUp(da,ierr)
CHKERRA(ierr)
CALL
DMDAGetCorners(da,icx,icy,PETSC_NULL_INTEGER,lx,ly,PETSC_NULL_INTEGER,ierr)
CHKERRA(ierr)
WRITE(*,'(*(1X,A,I3))') 'I am process #', irank, ' - my lower left
corner is icx=', icx, &
'and icy=', icy, ' - the width is lx=', lx, '
and ly=', ly
CALL DMGetLocalToGlobalMapping(da,ltog,ierr)
CHKERRA(ierr)
CALL ISLocalToGlobalMappingGetSize(ltog,n_ltog,ierr)
CHKERRA(ierr)
ALLOCATE(id_ltog(1:n_ltog),STAT=ierr,ERRMSG=err_msg)
CALL ISLocalToGlobalMappingGetIndices(ltog,id_ltog,ierr)
CHKERRA(ierr)
CALL ISLocalToGlobalMappingRestoreIndices(ltog,id_ltog,ierr)
CHKERRA(ierr)
DEALLOCATE(id_ltog,STAT=ierr,ERRMSG=err_msg)
CALL DMDestroy(da,ierr)
CHKERRA(ierr)
CALL PetscFinalize(ierr)
END PROGRAM test_dmda
On 21/02/2019 17:49, Matthew Knepley wrote:
On Thu, Feb 21, 2019 at 11:16 AM Thibaut Appel via petsc-users
<[email protected] <mailto:[email protected]>> wrote:
Dear PETSc developers/users,
I’m solving linear PDEs on a regular grid with high-order finite
differences, assembling an MPIAIJ matrix to solve linear systems
or eigenvalue problems. I’ve been using vertex major, natural
ordering for the parallelism with PetscSplitOwnership (yielding
rectangular slices of the physical domain) and wanted to move to
DMDA to have a more square-ish domain decomposition and minimize
communication between processes.
However, my application is memory critical, and I have
finely-tuned matrix preallocation routines for allocating memory
“optimally”. It seems the memory of a DMDA matrix is allocated
along the value of the stencil width of DMDACreate and the manual
says about it
“These DMDA stencils have nothing directly to do with any finite
difference stencils one might chose to use for a discretization”
And despite reading the manual pages there must be something I do
not understand in the DM topology, what is that "stencil width"
for then? I will not use ghost values for my FD-method, right?
What this is saying is, "You might be using some stencil that is not
STAR or BOX, but we are preallocating according to one of those".
If you really care about how much memory is preallocated, which it
seems you do, then you might be able to use
https://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/DMDA/DMDASetBlockFills.html
to tell use exactly how to preallocate.
I was then wondering if I could just create a MPIAIJ matrix, and
with a PETSc routine get the global indices of the domain for each
process: in other words, an equivalent of PetscSplitOwnership that
gives me the DMDA unknown ordering. So I can feed and loop on that
in my preallocation and assembly routines.
You can make an MPIAIJ matrix yourself of course. It should have the
same division of rows as the DMDA division of dofs. Also,
MatSetValuesStencil() will not work for a custom matrix.
Thanks,
Matt
Thanks very much,
Thibaut
--
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/>
