PetscViewerGetSubViewer#

Creates a new PetscViewer (same type as the old) that lives on a subcommunicator of the original viewer’s communicator

Synopsis#

#include "petscviewer.h" 
PetscErrorCode PetscViewerGetSubViewer(PetscViewer viewer, MPI_Comm comm, PetscViewer *outviewer)

Collective

Input Parameters#

viewer - the PetscViewer to be reproduced
comm - the sub communicator to use

Output Parameter#

outviewer - new PetscViewer

Notes#

The output of the subviewers is synchronized against the original viewer. For example, if a viewer on two MPI processes is decomposed into two subviewers, the output from the first viewer is all printed before the output from the second viewer. You must call PetscViewerFlush() after the call to PetscViewerRestoreSubViewer().

Call PetscViewerRestoreSubViewer() to destroy this PetscViewer, NOT PetscViewerDestroy()

This is most commonly used to view a sequential object that is part of a parallel object. For example PCView() on a PCBJACOBI could use this to obtain a PetscViewer that is used with the sequential KSP on one block of the preconditioner.

Between the calls to PetscViewerGetSubViewer() and PetscViewerRestoreSubViewer() the original viewer should not be used

PETSCVIEWERDRAW and PETSCVIEWERBINARY only support returning a singleton viewer on MPI rank 0, all other ranks will return a NULL viewer

Developer Notes#

There is currently incomplete error checking to ensure the user does not use the original viewer between the the calls to PetscViewerGetSubViewer() and PetscViewerRestoreSubViewer(). If the user does there could be errors in the viewing that go undetected or crash the code.

It would be nice if the call to PetscViewerFlush() was not required and was handled by PetscViewerRestoreSubViewer()