#include "petscdm.h" #include "petscdmlabel.h" #include "petscds.h" PetscErrorCode DMGetCompatibility(DM dm1,DM dm2,PetscBool *compatible,PetscBool *set)Collective
dm1 | - the first DM | |
dm2 | - the second DM |
compatible | - whether or not the two DMs are compatible | |
set | - whether or not the compatible value was set |
Typically, one would confirm compatibility if intending to simultaneously iterate over a pair of vectors obtained from different DMs.
For example, two DMDA objects are compatible if they have the same local and global sizes and the same stencil width. They can have different numbers of degrees of freedom per node. Thus, one could use the node numbering from either DM in bounds for a loop over vectors derived from either DM.
Consider the operation of summing data living on a 2-dof DMDA to data living on a 1-dof DMDA, which should be compatible, as in the following snippet.
... ierr = DMGetCompatibility(da1,da2,&compatible,&set);CHKERRQ(ierr); if (set && compatible) { ierr = DMDAVecGetArrayDOF(da1,vec1,&arr1);CHKERRQ(ierr); ierr = DMDAVecGetArrayDOF(da2,vec2,&arr2);CHKERRQ(ierr); ierr = DMDAGetCorners(da1,&x,&y,NULL,&m,&n,NULL);CHKERRQ(ierr); for (j=y; j<y+n; ++j) { for (i=x; i<x+m, ++i) { arr1[j][i][0] = arr2[j][i][0] + arr2[j][i][1]; } } ierr = DMDAVecRestoreArrayDOF(da1,vec1,&arr1);CHKERRQ(ierr); ierr = DMDAVecRestoreArrayDOF(da2,vec2,&arr2);CHKERRQ(ierr); } else { SETERRQ(PetscObjectComm((PetscObject)da1,PETSC_ERR_ARG_INCOMP,"DMDA objects incompatible"); } ...
Checking compatibility might be expensive for a given implementation of DM, or might be impossible to unambiguously confirm or deny. For this reason, this function may decline to determine compatibility, and hence users should always check the "set" output parameter.
A DM is always compatible with itself.
In the current implementation, DMs which live on "unequal" communicators (MPI_UNEQUAL in the terminology of MPI_Comm_compare()) are always deemed incompatible.
This function is labeled "Collective," as information about all subdomains is required on each rank. However, in DM implementations which store all this information locally, this function may be merely "Logically Collective".