DMCompositeSetCoupling#

Sets user provided routines that compute the coupling between the separate components DM in a DMCOMPOSITE to build the correct matrix nonzero structure.

Synopsis#

#include "petscdmcomposite.h"  
PetscErrorCode DMCompositeSetCoupling(DM dm, PetscErrorCode (*FormCoupleLocations)(DM, Mat, PetscInt *, PetscInt *, PetscInt, PetscInt, PetscInt, PetscInt))

Logically Collective; No Fortran Support

Input Parameters#

dm - the composite object
FormCoupleLocations - routine to set the nonzero locations in the matrix

Notes#

See DMSetApplicationContext() and DMGetApplicationContext() for how to get user information into this routine

The provided function should have a signature matching

   PetscErrorCode your_form_couple_method(DM dm, Mat J, PetscInt *dnz, PetscInt *onz, PetscInt rstart, PetscInt nrows, PetscInt start, PetscInt end);

where

dm - the composite object
J - the constructed matrix, or NULL. If provided, the function should fill the existing nonzero pattern with zeros (only dm and rstart are valid in this case).
dnz - array counting the number of on-diagonal non-zero entries per row, where on-diagonal means that this process owns both the row and column
onz - array counting the number of off-diagonal non-zero entries per row, where off-diagonal means that this process owns the row
rstart - offset into *nz arrays, for local row index r, update onz[r - rstart] or dnz[r - rstart]
nrows - number of owned global rows
start - the first owned global index
end - the last owned global index + 1

If J is not NULL, then the only other valid parameter is rstart

The user coupling function has a weird and poorly documented interface and is not tested, it should be removed

Level#

advanced

Location#

src/dm/impls/composite/pack.c

Index of all DMComposite routines
Table of Contents for all manual pages
Index of all manual pages