[petsc-dev] PetscOptionsTail

[Mark F. Adams]

On Apr 2, 2012, at 3:43 PM, Peter Brune wrote:

> Does PCGAMGSetType() call PetscOptionsHead() from KSP/PCSetFromOptions on some subsolver?  I had a similar issue recently with FAS.  The fix is to move other function calls that might do this out of your PetscOptionsHead/Tail() pair.

I don't think so ... I think this is the code stack that is invoked with PCGAMGSetType:

PetscErrorCode PCGAMGSetType( PC pc, const PCGAMGType type )
{
  PetscErrorCode ierr;
  
  PetscFunctionBegin;
  PetscValidHeaderSpecific(pc,PC_CLASSID,1);
  ierr = PetscTryMethod(pc,"PCGAMGSetType_C",(PC,const PCGAMGType),(pc,type));
  CHKERRQ(ierr);
  PetscFunctionReturn(0);
}
  
EXTERN_C_BEGIN
#undef __FUNCT__  
#define __FUNCT__ "PCGAMGSetType_GAMG"
PetscErrorCode PCGAMGSetType_GAMG( PC pc, const PCGAMGType type )
{
  PetscErrorCode ierr,(*r)(PC);
  
  PetscFunctionBegin;
  ierr = PetscFListFind(GAMGList,((PetscObject)pc)->comm,type,PETSC_FALSE,(PetscVoidStarFunction)&r); 
  CHKERRQ(ierr);

  if (!r) SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_UNKNOWN_TYPE,"Unknown GAMG type %s given",type);

  /* call sub create method */
  ierr = (*r)(pc); CHKERRQ(ierr);

  PetscFunctionReturn(0);
}
EXTERN_C_END

PetscErrorCode  PCCreateGAMG_AGG( PC pc )
{
  PetscErrorCode  ierr;
  PC_MG           *mg = (PC_MG*)pc->data;
  PC_GAMG         *pc_gamg = (PC_GAMG*)mg->innerctx;
  PC_GAMG_AGG      *pc_gamg_agg;

  PetscFunctionBegin;
  /* create sub context for SA */
  ierr = PetscNewLog( pc, PC_GAMG_AGG, &pc_gamg_agg ); CHKERRQ(ierr);
  assert(!pc_gamg->subctx);
  pc_gamg->subctx = pc_gamg_agg;
  
  pc->ops->setfromoptions = PCSetFromOptions_GAMG_AGG;
  pc->ops->destroy        = PCDestroy_AGG;
  /* reset does not do anything; setup not virtual */

  /* set internal function pointers */
  pc_gamg->graph = PCGAMGgraph_AGG;
  pc_gamg->coarsen = PCGAMGCoarsen_AGG;
  pc_gamg->prolongator = PCGAMGProlongator_AGG;
  pc_gamg->optprol = PCGAMGOptprol_AGG;
 
  pc_gamg->createdefaultdata = PCSetData_AGG;

  ierr = PetscObjectComposeFunctionDynamic( (PetscObject)pc,
                                            "PCSetCoordinates_C",
                                            "PCSetCoordinates_AGG",
                                            PCSetCoordinates_AGG);
  PetscFunctionReturn(0);
}

Mark

> 
> - Peter
> 
> On Mon, Apr 2, 2012 at 2:40 PM, Mark F. Adams <mark.adams at columbia.edu> wrote:
> I am getting strange behavior with PetscOptionsTail.  I have this code:
> 
>  /* instantiate derived type */
>  ierr = PetscOptionsHead("GAMG options"); CHKERRQ(ierr);
>  {
>    char tname[256] = GAMGAGG;
>    ierr = PetscOptionsList("-pc_gamg_type","Type of GAMG method","PCGAMGSetType",
>                            GAMGList, tname, tname, sizeof(tname), PETSC_NULL );
>    CHKERRQ(ierr);
>    ierr = PCGAMGSetType( pc, tname ); CHKERRQ(ierr);
>  }
> PetscPrintf(wcomm,"******* [%d]%s 5555\n",0,__FUNCT__);
>  ierr = PetscOptionsTail();   CHKERRQ(ierr);
> PetscPrintf(wcomm,"******* [%d]%s 6666\n",0,__FUNCT__);
> 
> and this last line does not get executed.  It is as if CHKERRQ is silently returning.  I see the usage in KSPSetFromOptions_GMRES has this code (PetscOptionsTail) at the very end and so this behavior  would not be noticed.
> 
> Mark
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120402/0a87ba16/attachment.html>