xref: /petsc/src/ksp/pc/impls/mg/fmg.c (revision 4b9ad92859ccb93b5e851e53cb8c4c04ac10e155)

/*$Id: fmg.c,v 1.26 2001/08/21 21:03:20 bsmith Exp $*/
/*
     Full multigrid using either additive or multiplicative V or W cycle
*/
#include "src/ksp/pc/impls/mg/mgimpl.h"

EXTERN int MGMCycle_Private(MG *,PetscTruth*);

/*
       MGFCycle_Private - Given an MG structure created with MGCreate() runs
               full multigrid.

    Iput Parameters:
.   mg - structure created with MGCreate().

    Note: This may not be what others call full multigrid. What we
          do is restrict the rhs to all levels, then starting
          on the coarsest level work our way up generating
          initial guess for the next level. This provides an
          improved preconditioner but not a great improvement.
*/
#undef __FUNCT__
#define __FUNCT__ "MGFCycle_Private"
int MGFCycle_Private(MG *mg)
{
  int    i,l = mg[0]->levels,ierr;
  PetscScalar zero = 0.0;

  PetscFunctionBegin;
  /* restrict the RHS through all levels to coarsest. */
  for (i=l-1; i>0; i--){
    ierr = MatRestrict(mg[i]->restrct,mg[i]->b,mg[i-1]->b);CHKERRQ(ierr);
  }

  /* work our way up through the levels */
  ierr = VecSet(&zero,mg[0]->x);CHKERRQ(ierr);
  for (i=0; i<l-1; i++) {
    ierr = MGMCycle_Private(&mg[i],PETSC_NULL);CHKERRQ(ierr);
    ierr = MatInterpolate(mg[i+1]->interpolate,mg[i]->x,mg[i+1]->x);CHKERRQ(ierr);
  }
  ierr = MGMCycle_Private(&mg[l-1],PETSC_NULL);CHKERRQ(ierr);
  PetscFunctionReturn(0);
}

/*
       MGKCycle_Private - Given an MG structure created with MGCreate() runs
               full Kascade MG solve.

    Iput Parameters:
.   mg - structure created with MGCreate().

    Note: This may not be what others call Kascadic MG.
*/
#undef __FUNCT__
#define __FUNCT__ "MGKCycle_Private"
int MGKCycle_Private(MG *mg)
{
  int    i,l = mg[0]->levels,ierr;
  PetscScalar zero = 0.0;

  PetscFunctionBegin;
  /* restrict the RHS through all levels to coarsest. */
  for (i=l-1; i>0; i--){
    ierr = MatRestrict(mg[i]->restrct,mg[i]->b,mg[i-1]->b);CHKERRQ(ierr);
  }

  /* work our way up through the levels */
  ierr = VecSet(&zero,mg[0]->x);CHKERRQ(ierr);
  for (i=0; i<l-1; i++) {
    if (mg[i]->eventsolve) {ierr = PetscLogEventBegin(mg[i]->eventsolve,0,0,0,0);CHKERRQ(ierr);}
    ierr = KSPSetRhs(mg[i]->smoothd,mg[i]->b);CHKERRQ(ierr);
    ierr = KSPSetSolution(mg[i]->smoothd,mg[i]->x);CHKERRQ(ierr);
    ierr = KSPSolve(mg[i]->smoothd);CHKERRQ(ierr);
    if (mg[i]->eventsolve) {ierr = PetscLogEventEnd(mg[i]->eventsolve,0,0,0,0);CHKERRQ(ierr);}
    ierr = MatInterpolate(mg[i+1]->interpolate,mg[i]->x,mg[i+1]->x);CHKERRQ(ierr);
  }
  if (mg[l-1]->eventsolve) {ierr = PetscLogEventBegin(mg[l-1]->eventsolve,0,0,0,0);CHKERRQ(ierr);}
  ierr = KSPSetRhs(mg[l-1]->smoothd,mg[l-1]->b);CHKERRQ(ierr);
  ierr = KSPSetSolution(mg[l-1]->smoothd,mg[l-1]->x);CHKERRQ(ierr);
  ierr = KSPSolve(mg[l-1]->smoothd);CHKERRQ(ierr);
  if (mg[l-1]->eventsolve) {ierr = PetscLogEventEnd(mg[l-1]->eventsolve,0,0,0,0);CHKERRQ(ierr);}

  PetscFunctionReturn(0);
}