MatCreateShell#

Creates a new matrix of MatType MATSHELL for use with a user-defined private data storage format.

Synopsis#

#include "petscmat.h" 
PetscErrorCode MatCreateShell(MPI_Comm comm, PetscInt m, PetscInt n, PetscInt M, PetscInt N, void *ctx, Mat *A)

Collective

Input Parameters#

  • comm - MPI communicator

  • m - number of local rows (must be given)

  • n - number of local columns (must be given)

  • M - number of global rows (may be PETSC_DETERMINE)

  • N - number of global columns (may be PETSC_DETERMINE)

  • ctx - pointer to data needed by the shell matrix routines

Output Parameter#

  • A - the matrix

Usage#

    extern PetscErrorCode mult(Mat,Vec,Vec);
    MatCreateShell(comm,m,n,M,N,ctx,&mat);
    MatShellSetOperation(mat,MATOP_MULT,(void(*)(void))mult);
    [ Use matrix for operations that have been set ]
    MatDestroy(mat);

Notes#

The shell matrix type is intended to provide a simple class to use with KSP (such as, for use with matrix-free methods). You should not use the shell type if you plan to define a complete matrix class.

PETSc requires that matrices and vectors being used for certain operations are partitioned accordingly. For example, when creating a shell matrix, A, that supports parallel matrix-vector products using MatMult(A,x,y) the user should set the number of local matrix rows to be the number of local elements of the corresponding result vector, y. Note that this is information is required for use of the matrix interface routines, even though the shell matrix may not actually be physically partitioned. For example,

     Vec x, y
     extern PetscErrorCode mult(Mat,Vec,Vec);
     Mat A

     VecCreateMPI(comm,PETSC_DECIDE,M,&y);
     VecCreateMPI(comm,PETSC_DECIDE,N,&x);
     VecGetLocalSize(y,&m);
     VecGetLocalSize(x,&n);
     MatCreateShell(comm,m,n,M,N,ctx,&A);
     MatShellSetOperation(mat,MATOP_MULT,(void(*)(void))mult);
     MatMult(A,x,y);
     MatDestroy(&A);
     VecDestroy(&y);
     VecDestroy(&x);

MATSHELL handles MatShift(), MatDiagonalSet(), MatDiagonalScale(), MatAXPY(), MatScale(), MatZeroRows() and MatZeroRowsColumns() internally, so these operations cannot be overwritten unless MatShellSetManageScalingShifts() is called.

Developer Notes#

For rectangular matrices do all the scalings and shifts make sense?

Regarding shifting and scaling. The general form is

diag(left)(vscale*A + diag(dshift) + vshift I)diag(right)

The order you apply the operations is important. For example if you have a dshift then apply a MatScale(s) you get svscaleA + sdiag(shift). But if you first scale and then shift you get svscale*A + diag(shift)

A is the user provided function.

KSP/PC uses changes in the Mat’s “state” to decide if preconditioners need to be rebuilt PCSetUp() only calls the setup() for for the PC implementation if the Mat state has increased from the previous call. Thus to get changes in a MATSHELL to trigger an update in the preconditioner you must call MatAssemblyBegin() and MatAssemblyEnd() or PetscObjectStateIncrease((PetscObject)mat); each time the MATSHELL matrix has changed.

Matrix product operations (i.e. MatMat(), MatTransposeMat() etc) can be specified using MatShellSetMatProductOperation()

Calling MatAssemblyBegin()/MatAssemblyEnd() on a MATSHELL removes any previously supplied shift and scales that were provided with MatDiagonalSet(), MatShift(), MatScale(), or MatDiagonalScale().

Fortran Note#

To use this from Fortran with a ctx you must write an interface definition for this function and for MatShellGetContext() that tells Fortran the Fortran derived data type you are passing in as the ctx argument.

See Also#

Matrices, Mat, MATSHELL, MatShellSetOperation(), MatHasOperation(), MatShellGetContext(), MatShellSetContext(), MATSHELL, MatShellSetManageScalingShifts(), MatShellSetMatProductOperation()

Level#

advanced

Location#

src/mat/impls/shell/shell.c

Examples#

src/ksp/ksp/tutorials/ex14f.F90
src/mat/tutorials/ex6f.F90
src/snes/tutorials/ex36.c
src/tao/bound/tutorials/plate2.c
src/tao/pde_constrained/tutorials/elliptic.c
src/tao/pde_constrained/tutorials/hyperbolic.c
src/tao/pde_constrained/tutorials/parabolic.c
src/tao/unconstrained/tutorials/eptorsion1.c
src/tao/unconstrained/tutorials/eptorsion3.c
src/ts/tutorials/ex20opt_ic.c
src/ts/tutorials/ex22f_mf.F90


Edit on GitLab

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