// ============================================================================
int ML_Epetra::MultiLevelPreconditioner::
VisualizeSmoothers(int NumPreCycles, int NumPostCycles)
{
bool viz = List_.get("viz: enable",false);
if (viz) {
if (IsPreconditionerComputed() == false)
ML_CHK_ERR(-1); // need an already computed preconditioner
bool VizPreSmoother = false;
bool VizPostSmoother = false;
if (NumPreCycles != 0)
VizPreSmoother = true;
if (NumPostCycles != 0)
VizPostSmoother = true;
int ierr = Visualize(false, VizPreSmoother, VizPostSmoother,
false, NumPreCycles, NumPostCycles, -1);
ML_CHK_ERR(ierr);
}
else
{
std::cout << PrintMsg_ << "You need to specify `viz: enable' = true" << std::endl;
std::cout << PrintMsg_ << "in the parameter list before building the ML" << std::endl;
std::cout << PrintMsg_ << "preconditioner in order to visualize" << std::endl;
ML_CHK_ERR(-1);
}
return(0);
}
// ================================================ ====== ==== ==== == =
int ML_Epetra::FaceMatrixFreePreconditioner::NodeAggregate(ML_Aggregate_Struct *&MLAggr,ML_Operator *&P,ML_Operator* TMT_ML,int &NumAggregates){
/* Pull Teuchos Options */
string CoarsenType = List_.get("aggregation: type", "Uncoupled");
double Threshold = List_.get("aggregation: threshold", 0.0);
int NodesPerAggr = List_.get("aggregation: nodes per aggregate",
ML_Aggregate_Get_OptimalNumberOfNodesPerAggregate());
string PrintMsg_ = "FMFP (Level 0): ";
ML_Aggregate_Create(&MLAggr);
ML_Aggregate_Set_MaxLevels(MLAggr, 2);
ML_Aggregate_Set_StartLevel(MLAggr, 0);
ML_Aggregate_Set_Threshold(MLAggr, Threshold);
ML_Aggregate_Set_MaxCoarseSize(MLAggr,1);
MLAggr->cur_level = 0;
ML_Aggregate_Set_Reuse(MLAggr);
MLAggr->keep_agg_information = 1;
P = ML_Operator_Create(ml_comm_);
/* Process Teuchos Options */
if (CoarsenType == "Uncoupled")
ML_Aggregate_Set_CoarsenScheme_Uncoupled(MLAggr);
else if (CoarsenType == "Uncoupled-MIS"){
ML_Aggregate_Set_CoarsenScheme_UncoupledMIS(MLAggr);
}
else if (CoarsenType == "METIS"){
ML_Aggregate_Set_CoarsenScheme_METIS(MLAggr);
ML_Aggregate_Set_NodesPerAggr(0, MLAggr, 0, NodesPerAggr);
}/*end if*/
else {
if(!Comm_->MyPID()) printf("FMFP: Unsupported (1,1) block aggregation type(%s), resetting to uncoupled-mis\n",CoarsenType.c_str());
ML_Aggregate_Set_CoarsenScheme_UncoupledMIS(MLAggr);
}
/* Aggregate Nodes */
int printlevel=ML_Get_PrintLevel();
ML_Set_PrintLevel(10);
NumAggregates = ML_Aggregate_Coarsen(MLAggr, TMT_ML, &P, ml_comm_);
ML_Set_PrintLevel(printlevel);
if (NumAggregates == 0){
cerr << "Found 0 aggregates, perhaps the problem is too small." << endl;
ML_CHK_ERR(-2);
}/*end if*/
else if(very_verbose_) printf("[%d] FMFP: %d aggregates created invec_leng=%d\n",Comm_->MyPID(),NumAggregates,P->invec_leng);
int globalAggs;
Comm_->SumAll(&NumAggregates,&globalAggs,1);
if( verbose_ && !Comm_->MyPID()) {
std::cout << PrintMsg_ << "Aggregation threshold = " << Threshold << std::endl;
std::cout << PrintMsg_ << "Global aggregates = " << globalAggs << std::endl;
//ML_Aggregate_Print_Complexity(MLAggr);
}
if(P==0) {fprintf(stderr,"%s","ERROR: No tentative prolongator found\n");ML_CHK_ERR(-5);}
return 0;
}
// ================================================ ====== ==== ==== == =
// Forms the coarse matrix, given the prolongator
int ML_Epetra::FaceMatrixFreePreconditioner::FormCoarseMatrix()
{
CoarseMat_ML = ML_Operator_Create(ml_comm_);
CoarseMat_ML->data_destroy=free;
ML_Operator *Temp_ML=0;
ML_Operator *R= ML_Operator_Create(ml_comm_);
ML_Operator *P= ML_Operator_Create(ml_comm_);
/* Build ML_Operator version of Prolongator_, Restriction Operator */
ML_CHK_ERR(ML_Operator_WrapEpetraCrsMatrix(Prolongator_,P,verbose_));
P->num_rigid=P->num_PDEs=dim;
//NTS: ML_CHK_ERR won't work on this: it returns 1
ML_Operator_Transpose_byrow(P, R);
/* OPTION: Disable the addon */
const Epetra_CrsMatrix *Op11crs = dynamic_cast<const Epetra_CrsMatrix*>(&*Operator_);
const Epetra_Operator_With_MatMat *Op11mm = dynamic_cast<const Epetra_Operator_With_MatMat*>(&*Operator_);
/* Do the A*P with or without addon*/
if(Op11crs){
if(verbose_ && !Comm_->MyPID()) printf("FMFP: Running *without* addon\n");
ML_Operator *SM_ML = ML_Operator_Create(ml_comm_);
Temp_ML = ML_Operator_Create(ml_comm_);
ML_Operator_WrapEpetraCrsMatrix((Epetra_CrsMatrix*)Op11crs,SM_ML,verbose_);
ML_2matmult(SM_ML,P,Temp_ML,ML_CSR_MATRIX);
ML_Operator_Destroy(&SM_ML);
}
else if(Op11mm){
if(verbose_ && !Comm_->MyPID()) printf("FMFP: Running with addon\n");
ML_CHK_ERR(Op11mm->MatrixMatrix_Multiply(*Prolongator_,ml_comm_,&Temp_ML));
}
else{
if(!Comm_->MyPID()) printf("ERROR: FMFP Illegal Operator\n");
delete R;
ML_CHK_ERR(-1);
}
/* Do R * AP */
R->num_rigid=R->num_PDEs=dim;
ML_2matmult_block(R, Temp_ML,CoarseMat_ML,ML_CSR_MATRIX);
/* Wrap to Epetra-land */
int nnz=100;
double time;
ML_Operator2EpetraCrsMatrix(CoarseMat_ML,CoarseMatrix,nnz,true,time,0,verbose_);
// NTS: This is a hack to get around the sticking ones on the diagonal issue;
/* Cleanup */
ML_Operator_Destroy(&P);
ML_Operator_Destroy(&R);
ML_Operator_Destroy(&Temp_ML);
ML_Operator_Destroy(&CoarseMat_ML);CoarseMat_ML=0;//HAX
return 0;
}/*end FormCoarseMatrix*/
// ============================================================================
int ML_Epetra::MultiLevelPreconditioner::
VisualizeAggregates()
{
bool viz = List_.get("viz: enable",false);
if (viz) {
if (IsPreconditionerComputed() == false)
ML_CHK_ERR(-1); // need an already computed preconditioner
ML_CHK_ERR(Visualize(true, false, false, false, -1, -1, -1));
}
return(0);
}
int ML_Amesos_Solve( void *data, double x[], double rhs[] )
{
Amesos_Handle_Type *Amesos_Handle = (Amesos_Handle_Type *) data;
if (Amesos_Handle->A_Base == 0) return 0;
Amesos_BaseSolver *A_Base = (Amesos_BaseSolver *) Amesos_Handle->A_Base ;
Epetra_Time Time(A_Base->Comm());
Epetra_LinearProblem *Amesos_LinearProblem = (Epetra_LinearProblem *)A_Base->GetProblem() ;
const Epetra_BlockMap & map = Amesos_LinearProblem->GetOperator()->OperatorDomainMap() ;
Epetra_Vector EV_rhs( View, map, rhs ) ;
Epetra_Vector EV_lhs( View, map, x ) ;
Amesos_LinearProblem->SetRHS( &EV_rhs ) ;
Amesos_LinearProblem->SetLHS( &EV_lhs ) ;
A_Base->Solve() ;
TimeForSolve__ += Time.ElapsedTime();
NumSolves__++;
#ifdef ML_AMESOS_DEBUG
// verify that the residual is actually small (and print the max
// in the destruction phase)
Epetra_Vector Ax(map);
(Amesos_LinearProblem->GetMatrix())->Multiply(false,EV_lhs,Ax);
ML_CHK_ERR(Ax.Update(1.0, EV_rhs, -1.0));
double residual;
ML_CHK_ERR(Ax.Norm2(&residual));
if( residual > MaxError__ ) MaxError__ = residual;
#endif
return 0;
} //ML_Amesos_Solve()
// ============================================================================
int ML_Epetra::MultiLevelPreconditioner::
VisualizeCycle(int NumCycles)
{
bool viz = List_.get("viz: enable",false);
if (viz) {
if (IsPreconditionerComputed() == false)
ML_CHK_ERR(-1); // need an already computed preconditioner
int ierr = Visualize(false, false, false, true,
-1, -1, NumCycles);
ML_CHK_ERR(ierr);
}
else
{
cout << PrintMsg_ << "You need to specify `viz: enable' = true" << endl;
cout << PrintMsg_ << "in the parameter list before building the ML" << endl;
cout << PrintMsg_ << "preconditioner in order to visualize" << endl;
ML_CHK_ERR(-1);
}
return(0);
}
// ================================================ ====== ==== ==== == =
// Computes C= <me> * A
int ML_Epetra::Epetra_Multi_CrsMatrix::MatrixMatrix_Multiply(const Epetra_CrsMatrix & A, ML_Comm *comm, ML_Operator **C) const
{
int rv=0;
ML_Comm* temp = global_comm;
/* Setup for 1st Matmat */
ML_Operator * MV[2]={0,0},*CV;
MV[(NumMatrices_-1)%2]= ML_Operator_Create(comm);
rv=ML_Operator_WrapEpetraCrsMatrix((Epetra_CrsMatrix*)&A,MV[(NumMatrices_-1)%2]);
ML_CHK_ERR(rv);
/* Do the matmats */
for(int i=NumMatrices_-1;rv==0 && i>=0;i--){
/* Do pre-wraps */
if(MV[(i+1)%2] && i!=NumMatrices_-1) ML_Operator_Destroy(&MV[(i+1)%2]);
MV[(i+1)%2]=ML_Operator_Create(comm);
CV=ML_Operator_Create(comm);
rv=ML_Operator_WrapEpetraCrsMatrix(CrsMatrices_[i],CV);
ML_CHK_ERR(rv);
/* Do matmat */
ML_2matmult(CV,MV[i%2],MV[(i+1)%2],ML_CSR_MATRIX);
ML_Operator_Destroy(&CV);
}/*end for*/
global_comm = temp;
/* Final Postwrap */
*C=MV[1];
/* Cleanup */
if(MV[0]) ML_Operator_Destroy(&MV[0]);
return rv;
}/*end MatrixMatrix_Multiply*/
// ================================================ ====== ==== ==== == =
//! Build the face-to-node prolongator described by Bochev, Siefert, Tuminaro, Xu and Zhu (2007).
int ML_Epetra::FaceMatrixFreePreconditioner::PBuildSparsity(ML_Operator *P, Epetra_CrsMatrix *&Psparse){
/* Create wrapper to do abs(T) */
// NTS: Assume D0 has already been reindexed by now.
ML_Operator* AbsFN_ML = ML_Operator_Create(ml_comm_);
ML_CHK_ERR(ML_Operator_WrapEpetraCrsMatrix(const_cast<Epetra_CrsMatrix*>(&*FaceNode_Matrix_),AbsFN_ML,verbose_));
ML_Operator_Set_Getrow(AbsFN_ML,AbsFN_ML->outvec_leng,CSR_getrow_ones);
/* Form abs(T) * P_n */
ML_Operator* AbsFNP = ML_Operator_Create(ml_comm_);
ML_2matmult(AbsFN_ML,P,AbsFNP, ML_CSR_MATRIX);
/* Wrap P_n into Epetra-land */
Epetra_CrsMatrix_Wrap_ML_Operator(AbsFNP,*Comm_,*FaceRangeMap_,&Psparse,Copy,0);
/* Nuke the rows in Psparse */
if(BCfaces_.size()>0) Apply_BCsToMatrixRows(BCfaces_.get(),BCfaces_.size(),*Psparse);
// Cleanup
ML_Operator_Destroy(&AbsFN_ML);
ML_Operator_Destroy(&AbsFNP);
return 0;
}
// ================================================ ====== ==== ==== == =
//! Apply the preconditioner to an Epetra_MultiVector X, puts the result in Y
int ML_Epetra::FaceMatrixFreePreconditioner::ApplyInverse(const Epetra_MultiVector& B_, Epetra_MultiVector& X) const{
const Epetra_MultiVector *B;
Epetra_MultiVector *Bcopy=0;
/* Sanity Checks */
int NumVectors=B_.NumVectors();
if (!B_.Map().SameAs(*FaceDomainMap_)) ML_CHK_ERR(-1);
if (NumVectors != X.NumVectors()) ML_CHK_ERR(-1);
Epetra_MultiVector r_edge(*FaceDomainMap_,NumVectors,false);
Epetra_MultiVector e_edge(*FaceDomainMap_,NumVectors,false);
Epetra_MultiVector e_node(*CoarseMap_,NumVectors,false);
Epetra_MultiVector r_node(*CoarseMap_,NumVectors,false);
/* Deal with the B==X case */
if (B_.Pointers()[0] == X.Pointers()[0]){
Bcopy=new Epetra_MultiVector(B_);
B=Bcopy;
X.PutScalar(0.0);
}
else B=&B_;
for(int i=0;i<num_cycles;i++){
/* Pre-smoothing */
#ifdef HAVE_ML_IFPACK
if(Smoother_) ML_CHK_ERR(Smoother_->ApplyInverse(*B,X));
#endif
if(MaxLevels > 0){
if(i != 0
#ifdef HAVE_ML_IFPACK
|| Smoother_
#endif
){
/* Calculate Residual (r_e = b - (S+M+Addon) * x) */
ML_CHK_ERR(Operator_->Apply(X,r_edge));
ML_CHK_ERR(r_edge.Update(1.0,*B,-1.0));
/* Xfer to coarse grid (r_n = P' * r_e) */
ML_CHK_ERR(Prolongator_->Multiply(true,r_edge,r_node));
}
else{
/* Xfer to coarse grid (r_n = P' * r_e) */
ML_CHK_ERR(Prolongator_->Multiply(true,*B,r_node));
}
/* AMG on coarse grid (e_n = (CoarseMatrix)^{-1} r_n) */
ML_CHK_ERR(CoarsePC->ApplyInverse(r_node,e_node));
/* Xfer back to fine grid (e_e = P * e_n) */
ML_CHK_ERR(Prolongator_->Multiply(false,e_node,e_edge));
/* Add in correction (x = x + e_e) */
ML_CHK_ERR(X.Update(1.0,e_edge,1.0));
}/*end if*/
/* Post-Smoothing */
#ifdef HAVE_ML_IFPACK
if(Smoother_) ML_CHK_ERR(Smoother_->ApplyInverse(*B,X));
#endif
}/*end for*/
/* Cleanup */
if(Bcopy) delete Bcopy;
return 0;
}/*end ApplyInverse*/
// ================================================ ====== ==== ==== == =
//! Build the face-to-node prolongator described by Bochev, Siefert, Tuminaro, Xu and Zhu (2007).
int ML_Epetra::FaceMatrixFreePreconditioner::BuildProlongator()
{
/* Wrap TMT_Matrix in a ML_Operator */
ML_Operator* TMT_ML = ML_Operator_Create(ml_comm_);
ML_Operator_WrapEpetraCrsMatrix(const_cast<Epetra_CrsMatrix*>(&*TMT_Matrix_),TMT_ML);
/* Nodal Aggregation */
ML_Aggregate_Struct *MLAggr=0;
ML_Operator *P=0;
int NumAggregates;
int rv=ML_Epetra::RefMaxwell_Aggregate_Nodes(*TMT_Matrix_,List_,ml_comm_,std::string("FMFP (level 0) :"),MLAggr,P,NumAggregates);
if(rv || !P) {if(!Comm_->MyPID()) printf("ERROR: Building nodal P\n");ML_CHK_ERR(-1);}
/* Build 1-unknown sparsity of prolongator */
Epetra_CrsMatrix *Psparse=0;
PBuildSparsity(P,Psparse);
if(!Psparse) {if(!Comm_->MyPID()) printf("ERROR: Building Psparse\n");ML_CHK_ERR(-2);}
/* Build the "nullspace" */
Epetra_MultiVector *nullspace;
BuildNullspace(nullspace);
if(!nullspace) {if(!Comm_->MyPID()) printf("ERROR: Building Nullspace\n");ML_CHK_ERR(-3);}
/* Build the DomainMap of the new operator*/
const Epetra_Map & FineColMap = Psparse->ColMap();
CoarseMap_=new Epetra_Map(-1,NumAggregates*dim,0,*Comm_);
/* Allocate the Prolongator_ */
int max_nz_per_row=Psparse->MaxNumEntries();
Prolongator_=new Epetra_CrsMatrix(Copy,*FaceRangeMap_,0);
int ne1, *idx1, *idx2;
idx2=new int [dim*max_nz_per_row];
double *vals1, *vals2;
vals2=new double[dim*max_nz_per_row];
int nonzeros;
for(int i=0;i<Prolongator_->NumMyRows();i++){
Psparse->ExtractMyRowView(i,ne1,vals1,idx1);
nonzeros=0;
for(int j=0;j<ne1;j++) nonzeros+=ABS(vals1[j])>0;
for(int j=0;j<ne1;j++){
for(int k=0;k<dim;k++) {
idx2[j*dim+k]=FineColMap.GID(idx1[j])*dim+k;
//FIX: This works only because there's an implicit linear mapping which
//we're exploiting.
if(idx2[j*dim+k]==-1) printf("[%d] ERROR: idx1[j]=%d / idx1[j]*dim+k=%d does not have a GID!\n",Comm_->MyPID(),idx1[j],idx1[j]*dim+k);
if(vals1[j]==0 ) vals2[j*dim+k]=0;
else vals2[j*dim+k]=(*nullspace)[k][i] / nonzeros;
}/*end for*/
}/*end for*/
Prolongator_->InsertGlobalValues(FaceRangeMap_->GID(i),dim*ne1,vals2,idx2);
}/*end for*/
/* FillComplete / OptimizeStorage for Prolongator*/
Prolongator_->FillComplete(*CoarseMap_,*FaceRangeMap_);
Prolongator_->OptimizeStorage();
#ifndef NO_OUTPUT
/* DEBUG: Dump aggregates */
Epetra_IntVector AGG(View,*NodeDomainMap_,MLAggr->aggr_info[0]);
IVOUT(AGG,"agg.dat");
EpetraExt::RowMatrixToMatlabFile("fmfp_psparse.dat",*Psparse);
EpetraExt::RowMatrixToMatlabFile("fmfp_prolongator.dat",*Prolongator_);
EpetraExt::VectorToMatrixMarketFile("fmfp_null0.dat",*(*nullspace)(0));
EpetraExt::VectorToMatrixMarketFile("fmfp_null1.dat",*(*nullspace)(1));
EpetraExt::VectorToMatrixMarketFile("fmfp_null2.dat",*(*nullspace)(2));
#endif
/* EXPERIMENTAL: Normalize Prolongator Columns */
bool normalize_prolongator=List_.get("face matrix free: normalize prolongator",false);
if(normalize_prolongator){
Epetra_Vector n_vector(*CoarseMap_,false);
Prolongator_->InvColSums(n_vector);
Prolongator_->RightScale(n_vector);
}/*end if*/
/* Post-wrapping to convert to ML indexing */
#ifdef HAVE_ML_EPETRAEXT
Prolongator_ = dynamic_cast<Epetra_CrsMatrix*>(ModifyEpetraMatrixColMap(*Prolongator_,ProlongatorColMapTrans_,"Prolongator",(verbose_&&!Comm_->MyPID())));
#endif
/* Cleanup */
ML_qr_fix_Destroy();
ML_Aggregate_Destroy(&MLAggr);
ML_Operator_Destroy(&TMT_ML);
ML_Operator_Destroy(&P);
delete nullspace;
delete Psparse;
delete [] idx2;
delete [] vals2;
return 0;
}/*end BuildProlongator_*/
// ================================================ ====== ==== ==== == =
//! Build pi operator described by Bochev, Siefert, Tuminaro, Xu and Zhu (2007).
int ML_Epetra::FaceMatrixFreePreconditioner::BuildNullspace(Epetra_MultiVector *& nullspace){
int Nf=FaceRangeMap_->NumMyElements();
/* Pull the coordinates from Teuchos */
double * xcoord=List_.get("x-coordinates",(double*)0);
double * ycoord=List_.get("y-coordinates",(double*)0);
double * zcoord=List_.get("z-coordinates",(double*)0);
dim=(xcoord!=0) + (ycoord!=0) + (zcoord!=0);
// Sanity check
if(dim!=3){if(!Comm_->MyPID()) printf("ERROR: FaceMatrixFreePreconditioner only works in 3D"); ML_CHK_ERR(-1);}
// Build the (unimported) coordinate multivector
double **d_coords=new double* [dim];
d_coords[0]=xcoord; d_coords[1]=ycoord;
if(dim==3) d_coords[2]=zcoord;
Epetra_MultiVector n_coords_domain(View,*NodeDomainMap_,d_coords,dim);
Epetra_MultiVector *n_coords;
// Import coordinate info
if(FaceNode_Matrix_->Importer()){
n_coords=new Epetra_MultiVector(FaceNode_Matrix_->ColMap(),dim);
n_coords->PutScalar(0.0);
n_coords->Import(n_coords_domain,*FaceNode_Matrix_->Importer(),Add);
}
else n_coords=&n_coords_domain;
// Sanity HAQ - Only works on Hexes
if(FaceNode_Matrix_->GlobalMaxNumEntries()!=4)
{if(!Comm_->MyPID()) printf("ERROR: FaceMatrixFreePreconditioner only works on Hexes"); ML_CHK_ERR(-2);}
// Allocate vector
nullspace=new Epetra_MultiVector(*FaceDomainMap_,3);
// Fill matrix - NTS this will *NOT* do periodic BCs correctly.
double *a=new double[dim];
double *b=new double[dim];
double *c=new double[dim];
for(int i=0;i<Nf;i++){
int Ni,*indices;
double *values;
FaceNode_Matrix_->ExtractMyRowView(i,Ni,values,indices);
if(Ni != 4){
printf("ERROR: Face %d has only %d nodes\n",i,Ni);
ML_free(a);
ML_free(b);
ML_free(c);
ML_CHK_ERR(-1);
}
a[0] = (*n_coords)[0][indices[1]] - (*n_coords)[0][indices[0]];
a[1] = (*n_coords)[1][indices[1]] - (*n_coords)[1][indices[0]];
a[2] = (*n_coords)[2][indices[1]] - (*n_coords)[2][indices[0]];
b[0] = (*n_coords)[0][indices[2]] - (*n_coords)[0][indices[0]];
b[1] = (*n_coords)[1][indices[2]] - (*n_coords)[1][indices[0]];
b[2] = (*n_coords)[2][indices[2]] - (*n_coords)[2][indices[0]];
cross_product(a,b,c);
// HAQ - Hardwiring for hexes
// HAQ - Absolute value, presuming all hexes are actually pointed the same way. This is a HAQ!!!
(*nullspace)[0][i]=ABS(c[0])/6.0;
(*nullspace)[1][i]=ABS(c[1])/6.0;
(*nullspace)[2][i]=ABS(c[2])/6.0;
}
/* Cleanup */
if(FaceNode_Matrix_->Importer()) delete n_coords;
delete [] a; delete [] b; delete [] c;
delete [] d_coords;
return 0;
}
// ================================================ ====== ==== ==== == =
int ML_Epetra::RefMaxwell_Aggregate_Nodes(const Epetra_CrsMatrix & A, Teuchos::ParameterList & List, ML_Comm * ml_comm, std::string PrintMsg,
ML_Aggregate_Struct *& MLAggr,ML_Operator *&P, int &NumAggregates){
/* Output level */
bool verbose, very_verbose;
int OutputLevel = List.get("ML output", -47);
if(OutputLevel == -47) OutputLevel = List.get("output", 1);
if(OutputLevel>=15) very_verbose=verbose=true;
if(OutputLevel > 5) {very_verbose=false;verbose=true;}
else very_verbose=verbose=false;
/* Wrap A in a ML_Operator */
ML_Operator* A_ML = ML_Operator_Create(ml_comm);
ML_Operator_WrapEpetraCrsMatrix(const_cast<Epetra_CrsMatrix*>(&A),A_ML);
/* Pull Teuchos Options */
std::string CoarsenType = List.get("aggregation: type", "Uncoupled");
double Threshold = List.get("aggregation: threshold", 0.0);
int NodesPerAggr = List.get("aggregation: nodes per aggregate",
ML_Aggregate_Get_OptimalNumberOfNodesPerAggregate());
bool UseAux = List.get("aggregation: aux: enable",false);
double AuxThreshold = List.get("aggregation: aux: threshold",0.0);
int MaxAuxLevels = List.get("aggregation: aux: max levels",10);
ML_Aggregate_Create(&MLAggr);
ML_Aggregate_Set_MaxLevels(MLAggr, 2);
ML_Aggregate_Set_StartLevel(MLAggr, 0);
ML_Aggregate_Set_Threshold(MLAggr, Threshold);
ML_Aggregate_Set_MaxCoarseSize(MLAggr,1);
MLAggr->cur_level = 0;
ML_Aggregate_Set_Reuse(MLAggr);
MLAggr->keep_agg_information = 1;
P = ML_Operator_Create(ml_comm);
/* Process Teuchos Options */
if (CoarsenType == "Uncoupled")
ML_Aggregate_Set_CoarsenScheme_Uncoupled(MLAggr);
else if (CoarsenType == "Uncoupled-MIS"){
ML_Aggregate_Set_CoarsenScheme_UncoupledMIS(MLAggr);
}
else if (CoarsenType == "METIS"){
ML_Aggregate_Set_CoarsenScheme_METIS(MLAggr);
ML_Aggregate_Set_NodesPerAggr(0, MLAggr, 0, NodesPerAggr);
}/*end if*/
else {
if(!A.Comm().MyPID()) printf("%s Unsupported (1,1) block aggregation type(%s), resetting to uncoupled-mis\n",PrintMsg.c_str(),CoarsenType.c_str());
ML_Aggregate_Set_CoarsenScheme_UncoupledMIS(MLAggr);
}
/* Setup Aux Data */
if(UseAux) {
A_ML->aux_data->enable=1;
A_ML->aux_data->threshold=AuxThreshold;
A_ML->aux_data->max_level=MaxAuxLevels;
ML_Init_Aux(A_ML,List);
if(verbose && !A.Comm().MyPID()) {
printf("%s Using auxiliary matrix\n",PrintMsg.c_str());
printf("%s aux threshold = %e\n",PrintMsg.c_str(),A_ML->aux_data->threshold);
}
}
/* Aggregate Nodes */
int printlevel=ML_Get_PrintLevel();
if(verbose) ML_Set_PrintLevel(10);
NumAggregates = ML_Aggregate_Coarsen(MLAggr,A_ML, &P, ml_comm);
if(verbose) ML_Set_PrintLevel(printlevel);
if (NumAggregates == 0){
std::cerr << "Found 0 aggregates, perhaps the problem is too small." << std::endl;
ML_CHK_ERR(-2);
}/*end if*/
else if(very_verbose) printf("[%d] %s %d aggregates created invec_leng=%d\n",A.Comm().MyPID(),PrintMsg.c_str(),NumAggregates,P->invec_leng);
if(verbose){
int globalAggs=0;
A.Comm().SumAll(&NumAggregates,&globalAggs,1);
if(!A.Comm().MyPID()) {
printf("%s Aggregation threshold = %e\n",PrintMsg.c_str(),Threshold);
printf("%s Global aggregates = %d\n",PrintMsg.c_str(),globalAggs);
}
}
/* Cleanup */
ML_qr_fix_Destroy();
if(UseAux) ML_Finalize_Aux(A_ML);
ML_Operator_Destroy(&A_ML);
return 0;
}
// ============================================================================
int ML_Epetra::MultiLevelPreconditioner::
CreateAuxiliaryMatrixVbr(Epetra_VbrMatrix* &FakeMatrix)
{
// FakeMatrix has already been created before
if (FakeMatrix == 0)
ML_CHK_ERR(-1); // something went wrong
int NumDimensions = 0;
double* x_coord = List_.get("x-coordinates", (double *)0);
if (x_coord != 0) ++NumDimensions;
// at least x-coordinates must be not null
if( NumDimensions == 0 ) {
std::cerr << ErrorMsg_ << "Option `aggregation: use auxiliary matrix' == true" << std::endl
<< ErrorMsg_ << "requires x-, y-, or z-coordinates." << std::endl
<< ErrorMsg_ << "You must specify them using options" << std::endl
<< ErrorMsg_ << "`x-coordinates' (and equivalently for" << std::endl
<< ErrorMsg_ << "y- and z-)." << std::endl;
ML_CHK_ERR(-2); // wrong parameters
}
double* y_coord = List_.get("y-coordinates", (double *)0);
if (y_coord != 0) ++NumDimensions;
double* z_coord = List_.get("z-coordinates", (double *)0);
if (z_coord != 0) ++NumDimensions;
// small check to avoid strange behavior
if( z_coord != 0 && y_coord == 0 ) {
std::cerr << ErrorMsg_ << "Something wrong: `y-coordinates'" << std::endl
<< ErrorMsg_ << "is null, while `z-coordinates' is not null" << std::endl;
ML_CHK_ERR(-3); // something went wrong
}
// usual crap to clutter the output
if( verbose_ ) {
std::cout << std::endl;
std::cout << PrintMsg_ << "*** Using auxiliary matrix to create the aggregates" << std::endl
<< PrintMsg_ << "*** Number of dimensions = " << NumDimensions << std::endl
<< PrintMsg_ << "*** (the version for Epetra_VbrMatrix is currently used)" << std::endl;
std::cout << std::endl;
}
const Epetra_BlockMap& RowMap = FakeMatrix->RowMap();
const Epetra_BlockMap& ColMap = FakeMatrix->ColMap();
int NumMyRowElements = RowMap.NumMyElements();
int NumMyColElements = ColMap.NumMyElements();
int* MyGlobalRowElements = RowMap.MyGlobalElements();
int* MyGlobalColElements = ColMap.MyGlobalElements();
// use point map to exchange coordinates
Epetra_Map PointRowMap(-1,NumMyRowElements,MyGlobalRowElements,0,Comm());
Epetra_Map PointColMap(-1,NumMyColElements,MyGlobalColElements,0,Comm());
Epetra_Vector RowX(PointRowMap);
Epetra_Vector RowY(PointRowMap);
Epetra_Vector RowZ(PointRowMap);
for (int i = 0 ; i < NumMyRowElements ; ++i) {
RowX[i] = x_coord[i];
if (NumDimensions > 1) RowY[i] = y_coord[i];
if (NumDimensions > 2) RowZ[i] = z_coord[i];
}
// create vectors containing coordinates for columns
// (this is useful only if MIS/ParMETIS are used)
Epetra_Vector ColX(PointColMap);
Epetra_Vector ColY(PointColMap);
Epetra_Vector ColZ(PointColMap);
// get coordinates for non-local nodes (in column map)
Epetra_Import Importer(PointColMap,PointRowMap);
ColX.Import(RowX,Importer,Insert);
if (NumDimensions > 1) ColY.Import(RowY,Importer,Insert);
if (NumDimensions > 2) ColZ.Import(RowZ,Importer,Insert);
// room for getrow()
int MaxNnz = FakeMatrix->MaxNumEntries();
std::vector<int> colInd(MaxNnz);
std::vector<double> colVal(MaxNnz);
std::vector<double> coord_i(3);
std::vector<double> coord_j(3);
// change the entries of FakeMatrix so that it corresponds to a discrete
// Laplacian. Note: This is not exactly the same as in the Crs case.
FakeMatrix->PutScalar(0.0);
for (int LocalRow = 0; LocalRow < NumMyRowElements ; ++LocalRow) {
int RowDim, NumBlockEntries;
int* BlockIndices;
Epetra_SerialDenseMatrix** RowValues;
switch (NumDimensions) {
case 3:
coord_i[2] = RowZ[LocalRow];
case 2:
coord_i[1] = RowY[LocalRow];
case 1:
coord_i[0] = RowX[LocalRow];
}
FakeMatrix->ExtractMyBlockRowView(LocalRow,RowDim,NumBlockEntries,
BlockIndices,RowValues);
// accumulator for zero row-sum
double total = 0.0;
for (int j = 0 ; j < NumBlockEntries ; ++j) {
int LocalCol = BlockIndices[j];
// insert diagonal later
if (LocalCol != LocalRow) {
// get coordinates of this node
switch (NumDimensions) {
case 3:
coord_j[2] = ColZ[LocalCol];
case 2:
coord_j[1] = ColY[LocalCol];
case 1:
coord_j[0] = ColX[LocalCol];
}
// d2 is the square of the distance between node `i' and
// node `j'
double d2 = (coord_i[0] - coord_j[0]) * (coord_i[0] - coord_j[0]) +
(coord_i[1] - coord_j[1]) * (coord_i[1] - coord_j[1]) +
(coord_i[2] - coord_j[2]) * (coord_i[2] - coord_j[2]);
if (d2 == 0.0) {
std::cerr << std::endl;
std::cerr << ErrorMsg_ << "distance between node " << LocalRow << " and node "
<< LocalCol << std::endl
<< ErrorMsg_ << "is zero. Coordinates of these nodes are" << std::endl
<< ErrorMsg_ << "x_i = " << coord_i[0] << ", x_j = " << coord_j[0] << std::endl
<< ErrorMsg_ << "y_i = " << coord_i[1] << ", y_j = " << coord_j[1] << std::endl
<< ErrorMsg_ << "z_i = " << coord_i[2] << ", z_j = " << coord_j[2] << std::endl
<< ErrorMsg_ << "Now proceeding with distance = 1.0" << std::endl;
std::cerr << std::endl;
d2 = 1.0;
}
for (int k = 0 ; k < RowValues[j]->M() ; ++k) {
for (int h = 0 ; h < RowValues[j]->N() ; ++h) {
if (k == h)
(*RowValues[j])(k,h) = - 1.0 / d2;
}
}
total += 1.0 /d2;
}
}
// check that the diagonal block exists
bool ok = false;
int DiagonalBlock = 0;
for (int j = 0 ; j < NumBlockEntries ; ++j) {
if (BlockIndices[j] == LocalRow) {
DiagonalBlock = j;
ok = true;
break;
}
}
assert (ok == true);
for (int k = 0 ; k < RowValues[DiagonalBlock]->N() ; ++k) {
(*RowValues[DiagonalBlock])(k,k) = total;
}
}
// tell ML to keep the tentative prolongator
ML_Aggregate_Set_Reuse(agg_);
// pray that no bugs will tease us
return(0);
}
// ================================================ ====== ==== ==== == =
// the tentative null space is in input because the user
// has to remember to allocate and fill it, and then to delete
// it after calling this method.
int ML_Epetra::MultiLevelPreconditioner::
ComputeAdaptivePreconditioner(int TentativeNullSpaceSize,
double* TentativeNullSpace)
{
if ((TentativeNullSpaceSize == 0) || (TentativeNullSpace == 0))
ML_CHK_ERR(-1);
// ================================== //
// get parameters from the input list //
// ================================== //
// maximum number of relaxation sweeps
int MaxSweeps = List_.get("adaptive: max sweeps", 10);
// number of std::vector to be added to the tentative null space
int NumAdaptiveVectors = List_.get("adaptive: num vectors", 1);
if (verbose_) {
std::cout << PrintMsg_ << "*** Adaptive Smoother Aggregation setup ***" << std::endl;
std::cout << PrintMsg_ << " Maximum relaxation sweeps = " << MaxSweeps << std::endl;
std::cout << PrintMsg_ << " Additional vectors to compute = " << NumAdaptiveVectors << std::endl;
}
// ==================================================== //
// compute the preconditioner, set null space from user //
// (who will have to delete std::vector TentativeNullSpace) //
// ==================================================== //
double* NewNullSpace = 0;
double* OldNullSpace = TentativeNullSpace;
int OldNullSpaceSize = TentativeNullSpaceSize;
// need some work otherwise matvec() with Epetra_Vbr fails.
// Also, don't differentiate between range and domain here
// as ML will not work if range != domain
const Epetra_VbrMatrix* VbrA = NULL;
VbrA = dynamic_cast<const Epetra_VbrMatrix*>(RowMatrix_);
Epetra_Vector* LHS = 0;
Epetra_Vector* RHS = 0;
if (VbrA != 0) {
LHS = new Epetra_Vector(VbrA->DomainMap());
RHS = new Epetra_Vector(VbrA->DomainMap());
} else {
LHS = new Epetra_Vector(RowMatrix_->OperatorDomainMap());
RHS = new Epetra_Vector(RowMatrix_->OperatorDomainMap());
}
// destroy what we may already have
if (IsComputePreconditionerOK_ == true) {
DestroyPreconditioner();
}
// build the preconditioner for the first time
List_.set("null space: type", "pre-computed");
List_.set("null space: dimension", OldNullSpaceSize);
List_.set("null space: vectors", OldNullSpace);
ComputePreconditioner();
// ====================== //
// add one std::vector at time //
// ====================== //
for (int istep = 0 ; istep < NumAdaptiveVectors ; ++istep) {
if (verbose_) {
std::cout << PrintMsg_ << "\tAdaptation step " << istep << std::endl;
std::cout << PrintMsg_ << "\t---------------" << std::endl;
}
// ==================== //
// look for "bad" modes //
// ==================== //
// note: should an error occur, ML_CHK_ERR will return,
// and LHS and RHS will *not* be delete'd (--> memory leak).
// Anyway, this means that something wrong happened in the code
// and should be fixed by the user.
LHS->Random();
double Norm2;
for (int i = 0 ; i < MaxSweeps ; ++i) {
// RHS = (I - ML^{-1} A) LHS
ML_CHK_ERR(RowMatrix_->Multiply(false,*LHS,*RHS));
// FIXME: can do something slightly better here
ML_CHK_ERR(ApplyInverse(*RHS,*RHS));
ML_CHK_ERR(LHS->Update(-1.0,*RHS,1.0));
LHS->Norm2(&Norm2);
if (verbose_) {
std::cout << PrintMsg_ << "\titer " << i << ", ||x||_2 = ";
std::cout << Norm2 << std::endl;
}
}
// scaling vectors
double NormInf;
LHS->NormInf(&NormInf);
LHS->Scale(1.0 / NormInf);
// ========================================================= //
// copy tentative and computed null space into NewNullSpace, //
// which now becomes the standard null space //
// ========================================================= //
int NewNullSpaceSize = OldNullSpaceSize + 1;
NewNullSpace = new double[NumMyRows() * NewNullSpaceSize];
assert (NewNullSpace != 0);
int itmp = OldNullSpaceSize * NumMyRows();
for (int i = 0 ; i < itmp ; ++i) {
NewNullSpace[i] = OldNullSpace[i];
}
for (int j = 0 ; j < NumMyRows() ; ++j) {
NewNullSpace[itmp + j] = (*LHS)[j];
}
// =============== //
// visualize modes //
// =============== //
if (List_.get("adaptive: visualize", false)) {
double* x_coord = List_.get("viz: x-coordinates", (double*)0);
double* y_coord = List_.get("viz: y-coordinates", (double*)0);
double* z_coord = List_.get("viz: z-coordinates", (double*)0);
assert (x_coord != 0);
std::vector<double> plot_me(NumMyRows()/NumPDEEqns_);
ML_Aggregate_Viz_Stats info;
info.Amatrix = &(ml_->Amat[LevelID_[0]]);
info.x = x_coord;
info.y = y_coord;
info.z = z_coord;
info.Nlocal = NumMyRows() / NumPDEEqns_;
info.Naggregates = 1;
ML_Operator_AmalgamateAndDropWeak(&(ml_->Amat[LevelID_[0]]),
NumPDEEqns_, 0.0);
for (int ieqn = 0 ; ieqn < NumPDEEqns_ ; ++ieqn) {
for (int j = 0 ; j < NumMyRows() ; j+=NumPDEEqns_) {
plot_me[j / NumPDEEqns_] = (*LHS)[j + ieqn];
}
char FileName[80];
sprintf(FileName,"nullspace-mode%d-eq%d.xyz", istep, ieqn);
if (verbose_)
std::cout << PrintMsg_ << "writing file " << FileName << "..." << std::endl;
ML_Aggregate_VisualizeXYZ(info,FileName,
ml_->comm,&plot_me[0]);
}
ML_Operator_UnAmalgamateAndDropWeak(&(ml_->Amat[LevelID_[0]]),
NumPDEEqns_, 0.0);
}
// Destroy the old preconditioner
DestroyPreconditioner();
// ==================================================== //
// build the new preconditioner with the new null space //
// ==================================================== //
List_.set("null space: type", "pre-computed");
List_.set("null space: dimension", NewNullSpaceSize);
List_.set("null space: vectors", NewNullSpace);
ML_CHK_ERR(ComputePreconditioner());
if (istep && (istep != NumAdaptiveVectors))
delete OldNullSpace;
OldNullSpace = NewNullSpace;
OldNullSpaceSize = NewNullSpaceSize;
}
// keep trace of this pointer, it will be delete'd later
NullSpaceToFree_ = NewNullSpace;
delete LHS;
delete RHS;
return(0);
}
// ================================================ ====== ==== ==== == =
// Computes the preconditioner
int ML_Epetra::FaceMatrixFreePreconditioner::ComputePreconditioner(const bool CheckFiltering)
{
Teuchos::ParameterList & ListCoarse=List_.sublist("face matrix free: coarse");
/* ML Communicator */
ML_Comm_Create(&ml_comm_);
/* Parameter List Options */
int SmootherSweeps = List_.get("smoother: sweeps", 0);
MaxLevels = List_.get("max levels",10);
print_hierarchy= List_.get("print hierarchy",false);
num_cycles = List_.get("cycle applications",1);
/* Sanity Checking*/
int OperatorDomainPoints = OperatorDomainMap().NumGlobalPoints();
int OperatorRangePoints = OperatorRangeMap().NumGlobalPoints();
if (OperatorDomainPoints != OperatorRangePoints)
ML_CHK_ERR(-1); // only square matrices
/* Output Header */
if(verbose_ && !Comm_->MyPID()) {
printf("------------------------------------------------------------------------------\n");
printf("***\n");
printf("*** ML_Epetra::FaceMatrixFreePreconditioner [%s]\n",Label());
printf("***\n");
}
/* Invert non-zeros on the diagonal */
if(SmootherSweeps){
for (int i = 0; i < InvDiagonal_->MyLength(); ++i)
if ((*InvDiagonal_)[i] != 0.0)
(*InvDiagonal_)[i] = 1.0 / (*InvDiagonal_)[i];
double nrm;
InvDiagonal_->Norm2(&nrm);
if(verbose_ && !Comm_->MyPID()) printf("Inverse Diagonal Norm = %6.4e\n",nrm);
}/*end if*/
/* Do the eigenvalue estimation for Chebyshev */
if(SmootherSweeps)
ML_CHK_ERR(SetupSmoother());
if(MaxLevels > 0) {
/* Build the prolongator */
ML_CHK_ERR(BuildProlongator());
/* DEBUG: Output matrices */
if(print_hierarchy) EpetraExt::RowMatrixToMatlabFile("prolongator.dat",*Prolongator_);
/* Form the coarse matrix */
ML_CHK_ERR(FormCoarseMatrix());
/* DEBUG: Output matrices */
if(print_hierarchy) EpetraExt::RowMatrixToMatlabFile("coarsemat.dat",*CoarseMatrix);
/* Setup Preconditioner on Coarse Matrix */
CoarsePC = new MultiLevelPreconditioner(*CoarseMatrix,ListCoarse);
if(!CoarsePC) ML_CHK_ERR(-2);
/* Clean Up */
//delete nullspace;
}/*end if*/
return 0;
}/*end ComputePreconditioner*/
Epetra_Operator* ML_Gen_Smoother_Ifpack_Epetra(const Epetra_Operator *A,const Epetra_Vector *InvDiagonal, Teuchos::ParameterList & List,string printMsg,bool verbose){
/* Variables */
double lambda_min = 0.0;
double lambda_max = 0.0;
Teuchos::ParameterList IFPACKList=List.sublist("smoother: ifpack list");
/* Parameter-list Options */
string PreOrPostSmoother = List.get("smoother: pre or post","both");
string SmooType = List.get("smoother: type", "Chebyshev");
if(SmooType=="IFPACK") SmooType=List.get("smoother: ifpack type","Chebyshev");
int Sweeps = List.get("smoother: sweeps", 3);
int IfpackOverlap = List.get("smoother: ifpack overlap",0);
double omega = List.get("smoother: damping factor",1.0);
Ifpack_Chebyshev* SmootherC_=0;
Ifpack_Preconditioner* SmootherP_=0;
/* Sanity Check*/
if(Sweeps==0) return 0;
/* Early Output*/
int Nrows=A->OperatorDomainMap().NumGlobalElements();
int Nnz=-1;
const Epetra_RowMatrix *Arow=dynamic_cast<const Epetra_RowMatrix*>(A);
if(Arow) Nnz=Arow->NumGlobalNonzeros();
if(verbose && !A->Comm().MyPID())
cout <<printMsg<<"# global rows = "<<Nrows<<" # estim. global nnz = "<<Nnz<<endl;
/**********************************************/
/*** Chebyshev (Including Block) ***/
/**********************************************/
if(SmooType=="Chebyshev" || SmooType=="MLS" || SmooType=="IFPACK-Chebyshev" || SmooType=="IFPACK-Block Chebyshev"){
bool allocated_inv_diagonal=false;
int MaximumIterations = List.get("eigen-analysis: max iters", 10);
string EigenType_ = List.get("eigen-analysis: type", "cg");
double boost = List.get("eigen-analysis: boost for lambda max", 1.0);
double alpha = List.get("chebyshev: alpha",30.0001);
Epetra_Vector *InvDiagonal_=0;
/* Block Chebyshev stuff if needed */
int MyCheby_nBlocks= List.get("smoother: Block Chebyshev number of blocks",0);
int* MyCheby_blockIndices=List.get("smoother: Block Chebyshev block list",(int*)0);
int* MyCheby_blockStarts= List.get("smoother: Block Chebyshev block starts",(int*)0);
bool MyCheby_NE= List.get("smoother: chebyshev solve normal equations",false);
if(SmooType == "IFPACK-Block Chebyshev" && MyCheby_blockIndices && MyCheby_blockStarts){
// If we're using Block Chebyshev, it can compute it's own eigenvalue estimate..
Teuchos::ParameterList PermuteList,BlockList;
BlockList.set("apply mode","invert");
PermuteList.set("number of local blocks",MyCheby_nBlocks);
PermuteList.set("block start index",MyCheby_blockStarts);
// if(is_lid) PermuteList.set("block entry lids",Blockids_);
//NTS: Add LID support
PermuteList.set("block entry gids",MyCheby_blockIndices);
PermuteList.set("blockdiagmatrix: list",BlockList);
IFPACKList.set("chebyshev: use block mode",true);
IFPACKList.set("chebyshev: block list",PermuteList);
IFPACKList.set("chebyshev: eigenvalue max iterations",10);
// EXPERIMENTAL: Cheby-NE
IFPACKList.set("chebyshev: solve normal equations",MyCheby_NE);
}
else {
/* Non-Blocked Chebyshev */
/* Grab Diagonal & invert if not provided */
if(InvDiagonal) InvDiagonal_=const_cast<Epetra_Vector *>(InvDiagonal);
else{
const Epetra_CrsMatrix* Acrs=dynamic_cast<const Epetra_CrsMatrix*>(A);
if(!Acrs) return 0;
allocated_inv_diagonal=true;
InvDiagonal_ = new Epetra_Vector(Acrs->RowMap());
Acrs->ExtractDiagonalCopy(*InvDiagonal_);
for (int i = 0; i < InvDiagonal_->MyLength(); ++i)
if ((*InvDiagonal_)[i] != 0.0)
(*InvDiagonal_)[i] = 1.0 / (*InvDiagonal_)[i];
}
/* Do the eigenvalue estimation*/
if (EigenType_ == "power-method") Ifpack_Chebyshev::PowerMethod(*A,*InvDiagonal_,MaximumIterations,lambda_max);
else if(EigenType_ == "cg") Ifpack_Chebyshev::CG(*A,*InvDiagonal_,MaximumIterations,lambda_min,lambda_max);
else ML_CHK_ERR(0); // not recognized
lambda_min=lambda_max / alpha;
/* Setup the Smoother's List*/
IFPACKList.set("chebyshev: min eigenvalue", lambda_min);
IFPACKList.set("chebyshev: max eigenvalue", boost * lambda_max);
IFPACKList.set("chebyshev: ratio eigenvalue",alpha);
IFPACKList.set("chebyshev: operator inv diagonal", InvDiagonal_);
}
/* Setup the Smoother's List*/
IFPACKList.set("chebyshev: ratio eigenvalue", alpha);
IFPACKList.set("chebyshev: degree", Sweeps);
IFPACKList.set("chebyshev: zero starting solution",false);
// Setup
SmootherC_= new Ifpack_Chebyshev(A);
if (SmootherC_ == 0) return 0;
SmootherC_->SetParameters(IFPACKList);
SmootherC_->Initialize();
SmootherC_->Compute();
// Grab the lambda's if needed
if(SmooType=="IFPACK-Block Chebyshev"){
lambda_min=SmootherC_->GetLambdaMin();
lambda_max=SmootherC_->GetLambdaMax();
}
// Smoother Info Output
if(verbose && !A->Comm().MyPID()){
if(SmooType=="IFPACK-Block Chebyshev") {
cout << printMsg << "MLS/Block-Chebyshev, polynomial order = "
<< Sweeps
<< ", alpha = " << alpha << endl;
cout << printMsg << "lambda_min = " << lambda_min
<< ", lambda_max = " << lambda_max << endl;
}
else {
cout << printMsg << "MLS/Chebyshev, polynomial order = "
<< Sweeps
<< ", alpha = " << alpha << endl;
cout << printMsg << "lambda_min = " << lambda_min
<< ", lambda_max = " << boost*lambda_max << endl;
}
}
// Cleanup: Since Chebyshev will keep it's own copy of the Inverse Diagonal...
if (allocated_inv_diagonal) delete InvDiagonal_;
return SmootherC_;
}
/**********************************************/
/*** Point Relaxation ***/
/**********************************************/
else if(SmooType=="Gauss-Seidel" || SmooType=="symmetric Gauss-Seidel" || SmooType=="Jacobi"
|| SmooType=="point relaxation stand-alone" || SmooType=="point relaxation" ){
const Epetra_CrsMatrix* Acrs=dynamic_cast<const Epetra_CrsMatrix*>(A);
if(!Acrs) return 0;
string MyIfpackType="point relaxation stand-alone";
if(IfpackOverlap > 0) MyIfpackType="point relaxation";
string MyRelaxType="symmetric Gauss-Seidel";
if(SmooType=="symmetric Gauss-Seidel") MyRelaxType=SmooType;
else if(SmooType=="Jacobi") MyRelaxType=SmooType;
IFPACKList.set("relaxation: type", IFPACKList.get("relaxation: type",MyRelaxType));
IFPACKList.set("relaxation: sweeps", Sweeps);
IFPACKList.set("relaxation: damping factor", omega);
IFPACKList.set("relaxation: zero starting solution",false);
if(verbose && !A->Comm().MyPID()){
cout << printMsg << IFPACKList.get("relaxation: type",MyRelaxType).c_str()<<" (sweeps="
<< Sweeps << ",omega=" << omega << ")" <<endl;
}
Ifpack Factory;
SmootherP_ = Factory.Create(MyIfpackType,const_cast<Epetra_CrsMatrix*>(Acrs),IfpackOverlap);
if (SmootherP_ == 0) return 0;
SmootherP_->SetParameters(IFPACKList);
SmootherP_->Initialize();
SmootherP_->Compute();
return SmootherP_;
}
/**********************************************/
/*** Block Relaxation ***/
/**********************************************/
else if(SmooType=="block Gauss-Seidel" || SmooType=="symmetric block Gauss-Seidel" || SmooType=="block Jacobi"
|| SmooType=="block relaxation stand-alone" || SmooType=="block relaxation" ){
const Epetra_CrsMatrix* Acrs=dynamic_cast<const Epetra_CrsMatrix*>(A);
if(!Acrs) return 0;
string MyIfpackType="block relaxation stand-alone";
if(IfpackOverlap > 0) MyIfpackType="block relaxation";
string MyRelaxType="symmetric Gauss-Seidel";
if(SmooType=="block Gauss-Seidel") MyRelaxType="Gauss-Seidel";
else if(SmooType=="block Jacobi") MyRelaxType="Jacobi";
IFPACKList.set("relaxation: type", IFPACKList.get("relaxation: type",MyRelaxType));
IFPACKList.set("relaxation: sweeps", Sweeps);
IFPACKList.set("relaxation: damping factor", omega);
IFPACKList.set("relaxation: zero starting solution",false);
if(verbose && !A->Comm().MyPID()){
cout << printMsg << "block " << IFPACKList.get("relaxation: type",MyRelaxType).c_str()<<" (sweeps="
<< Sweeps << ",omega=" << omega << ")" <<endl;
}
Ifpack Factory;
SmootherP_ = Factory.Create(MyIfpackType,const_cast<Epetra_CrsMatrix*>(Acrs),IfpackOverlap);
if (SmootherP_ == 0) return 0;
SmootherP_->SetParameters(IFPACKList);
SmootherP_->Initialize();
SmootherP_->Compute();
return SmootherP_;
}
/**********************************************/
/*** Incomplete Factorization ***/
/**********************************************/
else if(SmooType == "ILU" || SmooType == "IC" || SmooType == "ILUT" ||
SmooType == "ICT" || SmooType == "SILU") {
const Epetra_RowMatrix* Arow=dynamic_cast<const Epetra_RowMatrix*>(A);
double MyLOF=0.0;
if(SmooType=="ILUT" || SmooType=="ICT") MyLOF=List.get("smoother: ifpack level-of-fill",1.0);
else MyLOF=List.get("smoother: ifpack level-of-fill",0.0);
int MyIfpackOverlap = List.get("smoother: ifpack overlap", 0);
double MyIfpackRT = List.get("smoother: ifpack relative threshold", 1.0);
double MyIfpackAT = List.get("smoother: ifpack absolute threshold", 0.0);
IFPACKList.set("ILU: sweeps",Sweeps);
// Set the fact: LOF options, but only if they're not set already... All this sorcery is because level-of-fill
// is an int for ILU and a double for ILUT. Lovely.
if(SmooType=="ILUT" || SmooType=="ICT"){
IFPACKList.set("fact: level-of-fill", IFPACKList.get("fact: level-of-fill",MyLOF));
IFPACKList.set("fact: ilut level-of-fill", IFPACKList.get("fact: ilut level-of-fill",MyLOF));
IFPACKList.set("fact: ict level-of-fill", IFPACKList.get("fact: ict level-of-fill",MyLOF));
MyLOF=IFPACKList.get("fact: level-of-fill",MyLOF);
}
else{
IFPACKList.set("fact: level-of-fill", (int) IFPACKList.get("fact: level-of-fill",(int)MyLOF));
MyLOF=IFPACKList.get("fact: level-of-fill",(int)MyLOF);
}
IFPACKList.set("fact: relative threshold", MyIfpackRT);
IFPACKList.set("fact: absolute threshold", MyIfpackAT);
if(verbose && !A->Comm().MyPID()){
cout << printMsg << "IFPACK, type=`" << SmooType << "'," << endl
<< printMsg << PreOrPostSmoother << ",overlap=" << MyIfpackOverlap << endl;
cout << printMsg << "level-of-fill=" << MyLOF;
cout << ",rel. threshold=" << MyIfpackRT
<< ",abs. threshold=" << MyIfpackAT << endl;
}
Ifpack Factory;
SmootherP_ = Factory.Create(SmooType,const_cast<Epetra_RowMatrix*>(Arow),IfpackOverlap);
if (SmootherP_ == 0) return 0;
SmootherP_->SetParameters(IFPACKList);
SmootherP_->Initialize();
SmootherP_->Compute();
return SmootherP_;
}
/**********************************************/
/*** SORa ***/
/**********************************************/
else if(SmooType=="SORa"){
const Epetra_RowMatrix* Arow=dynamic_cast<const Epetra_RowMatrix*>(A);
if(verbose && !A->Comm().MyPID()){
cout << printMsg << "IFPACK/SORa("<<IFPACKList.get("sora: alpha",1.5)<<","<<IFPACKList.get("sora: gamma",1.0)<<")"
<< ", sweeps = " <<IFPACKList.get("sora: sweeps",1)<<endl;
if(IFPACKList.get("sora: oaz boundaries",false))
cout << printMsg << "oaz boundary handling enabled"<<endl;
if(IFPACKList.get("sora: use interproc damping",false))
cout << printMsg << "interproc damping enabled"<<endl;
if(IFPACKList.get("sora: use global damping",false))
cout << printMsg << "global damping enabled"<<endl;
}
Ifpack Factory;
SmootherP_ = Factory.Create(SmooType,const_cast<Epetra_RowMatrix*>(Arow),IfpackOverlap);
if (SmootherP_ == 0) return 0;
SmootherP_->SetParameters(IFPACKList);
SmootherP_->Initialize();
SmootherP_->Compute();
return SmootherP_;
}
else{
printf("ML_Gen_Smoother_Ifpack_New: Unknown preconditioner\n");
ML_CHK_ERR(0);
}
return 0;
}/*ML_Gen_Smoother_Ifpack_New*/
// ============================================================================
int ML_Epetra::MultiLevelPreconditioner::
CreateAuxiliaryMatrixCrs(Epetra_FECrsMatrix* &FakeMatrix)
{
int NumMyRows = RowMatrix_->NumMyRows();
const Epetra_Map& RowMap = RowMatrix_->RowMatrixRowMap();
const Epetra_Map& ColMap = RowMatrix_->RowMatrixColMap();
FakeMatrix = new Epetra_FECrsMatrix(Copy,RowMap,
2*RowMatrix_->MaxNumEntries());
if (FakeMatrix == 0)
ML_CHK_ERR(-1); // something went wrong
int NumDimensions = 0;
double* x_coord = List_.get("x-coordinates", (double *)0);
if (x_coord != 0) ++NumDimensions;
// at least x-coordinates must be not null
if( NumDimensions == 0 ) {
std::cerr << ErrorMsg_ << "Option `aggregation: use auxiliary matrix' == true" << std::endl
<< ErrorMsg_ << "requires x-, y-, or z-coordinates." << std::endl
<< ErrorMsg_ << "You must specify them using options" << std::endl
<< ErrorMsg_ << "`x-coordinates' (and equivalently for" << std::endl
<< ErrorMsg_ << "y- and z-." << std::endl;
ML_CHK_ERR(-2); // wrong parameters
}
double* y_coord = List_.get("y-coordinates", (double *)0);
if (y_coord != 0) ++NumDimensions;
double* z_coord = List_.get("z-coordinates", (double *)0);
if (z_coord != 0) ++NumDimensions;
// small check to avoid strange behavior
if( z_coord != 0 && y_coord == 0 ) {
std::cerr << ErrorMsg_ << "Something wrong: `y-coordinates'" << std::endl
<< ErrorMsg_ << "is null, while `z-coordinates' is null" << std::endl;
ML_CHK_ERR(-3); // something went wrong
}
double theta = List_.get("aggregation: theta",0.0);
bool SymmetricPattern = List_.get("aggregation: use symmetric pattern",false);
// usual crap to clutter the output
if( verbose_ ) {
std::cout << std::endl;
std::cout << PrintMsg_ << "*** Using auxiliary matrix to create the aggregates" << std::endl
<< PrintMsg_ << "*** Number of dimensions = " << NumDimensions << std::endl
<< PrintMsg_ << "*** theta = " << theta;
if( SymmetricPattern ) std::cout << ", using symmetric pattern" << std::endl;
else std::cout << ", using original pattern" << std::endl;
std::cout << std::endl;
}
// create vectors containing coordinates, replicated for all unknonws
// FIXME: I don't really need Z in all cases
// for west
// I am over-allocating, for large number of equations per node
// this is not optimal. However, it is a only-once importing
// of some more data. It should harm too much...
//
// The following will work for constant number of equations per
// node only. The fix should be easy, though.
Epetra_Vector RowX(RowMap); RowX.PutScalar(0.0);
Epetra_Vector RowY(RowMap); RowY.PutScalar(0.0);
Epetra_Vector RowZ(RowMap); RowZ.PutScalar(0.0);
for (int i = 0 ; i < NumMyRows ; i += NumPDEEqns_) {
RowX[i] = x_coord[i / NumPDEEqns_];
if (NumDimensions > 1) RowY[i] = y_coord[i / NumPDEEqns_];
if (NumDimensions > 2) RowZ[i] = z_coord[i / NumPDEEqns_];
}
// create vectors containing coordinates for columns
// (this is useful only if MIS/ParMETIS are used)
Epetra_Vector ColX(ColMap); ColX.PutScalar(0.0);
Epetra_Vector ColY(ColMap); ColY.PutScalar(0.0);
Epetra_Vector ColZ(ColMap); ColZ.PutScalar(0.0);
// get coordinates for non-local nodes (in column map)
Epetra_Import Importer(ColMap,RowMap);
ColX.Import(RowX,Importer,Insert);
if (NumDimensions > 1) ColY.Import(RowY,Importer,Insert);
if (NumDimensions > 2) ColZ.Import(RowZ,Importer,Insert);
// global row and column numbering
int* MyGlobalRowElements = RowMap.MyGlobalElements();
int* MyGlobalColElements = ColMap.MyGlobalElements();
// room for getrow()
int MaxNnz = RowMatrix_->MaxNumEntries();
std::vector<int> colInd(MaxNnz);
std::vector<double> colVal(MaxNnz);
std::vector<double> coord_i(3);
std::vector<double> coord_j(3);
// =================== //
// cycle over all rows //
// =================== //
for (int i = 0; i < NumMyRows ; i += NumPDEEqns_) {
int GlobalRow = MyGlobalRowElements[i];
if( i%NumPDEEqns_ == 0 ) { // do it just once for each block row
switch( NumDimensions ) {
case 3:
coord_i[2] = RowZ[i];
case 2:
coord_i[1] = RowY[i];
case 1:
coord_i[0] = RowX[i];
}
int NumEntries;
ML_CHK_ERR(RowMatrix_->ExtractMyRowCopy(i,MaxNnz,NumEntries,
&colVal[0],&colInd[0]));
// NOTE: for VBR matrices, the "real" value that will be used in
// the subsequent part of the code is only the one for the first
// equations. For each block, I replace values with the sum of
// the std::abs of each block entry.
for (int j = 0 ; j < NumEntries ; j += NumPDEEqns_) {
colVal[j] = std::fabs(colVal[j]);
for (int k = 1 ; k < NumPDEEqns_ ; ++k) {
colVal[j] += std::fabs(colVal[j+k]);
}
}
// work only on the first equations. Theta will blend the
// coordinate part with the sub of std::abs of row elements.
int GlobalCol;
double total = 0.0;
for (int j = 0 ; j < NumEntries ; j += NumPDEEqns_) {
if (colInd[j] >= NumMyRows)
continue;
if (colInd[j]%NumPDEEqns_ == 0) {
// insert diagonal later
if (colInd[j] != i) {
// get coordinates of this node
switch (NumDimensions) {
case 3:
coord_j[2] = ColZ[colInd[j]];
case 2:
coord_j[1] = ColY[colInd[j]];
case 1:
coord_j[0] = ColX[colInd[j]];
}
// d2 is the square of the distance between node `i' and
// node `j'
double d2 = (coord_i[0] - coord_j[0]) * (coord_i[0] - coord_j[0]) +
(coord_i[1] - coord_j[1]) * (coord_i[1] - coord_j[1]) +
(coord_i[2] - coord_j[2]) * (coord_i[2] - coord_j[2]);
if (d2 == 0.0) {
std::cerr << std::endl;
std::cerr << ErrorMsg_ << "distance between node " << i/NumPDEEqns_ << " and node "
<< colInd[j]/NumPDEEqns_ << std::endl
<< ErrorMsg_ << "is zero. Coordinates of these nodes are" << std::endl
<< ErrorMsg_ << "x_i = " << coord_i[0] << ", x_j = " << coord_j[0] << std::endl
<< ErrorMsg_ << "y_i = " << coord_i[1] << ", y_j = " << coord_j[1] << std::endl
<< ErrorMsg_ << "z_i = " << coord_i[2] << ", z_j = " << coord_j[2] << std::endl
<< ErrorMsg_ << "Now proceeding with distance = 1.0" << std::endl;
std::cerr << std::endl;
d2 = 1.0;
}
// blend d2 with the actual values of the matrix
// FIXME: am I useful?
double val = -(1.0 - theta) * (1.0 / d2) + theta * (colVal[j]);
GlobalCol = MyGlobalColElements[colInd[j]];
// insert this value on all rows
for (int k = 0 ; k < NumPDEEqns_ ; ++k) {
int row = GlobalRow + k;
int col = GlobalCol + k;
if (FakeMatrix->SumIntoGlobalValues(1,&row,1,&col,&val) != 0) {
ML_CHK_ERR(FakeMatrix->InsertGlobalValues(1,&row,1,&col,&val));
}
}
total -= val;
// put (j,i) element as well, only for in-process stuff.
// I have some problems with off-processor elements.
// It is here that I need the FE matrix.
if (SymmetricPattern == true && colInd[j] < NumMyRows ) {
for( int k=0 ; k<NumPDEEqns_ ; ++k ) {
int row = GlobalCol+k;
int col = GlobalRow+k;
if( FakeMatrix->SumIntoGlobalValues(1,&row,1,&col,&val) != 0 ) {
ML_CHK_ERR(FakeMatrix->InsertGlobalValues(1,&row,1,&col,&val));
}
}
total -= val;
}
}
}
}
// create lines with zero-row sum
for (int k = 0 ; k < NumPDEEqns_ ; ++k) {
int row = GlobalRow + k;
if (FakeMatrix->SumIntoGlobalValues(1,&row,1,&row,&total) != 0) {
if (FakeMatrix->InsertGlobalValues(1,&row,1,&row,&total) != 0)
ML_CHK_ERR(-9); // something went wrong
}
}
}
}
if (FakeMatrix->FillComplete())
ML_CHK_ERR(-5); // something went wrong
// stick pointer in Amat for level 0 (finest level)
ml_->Amat[LevelID_[0]].data = (void *)FakeMatrix;
// tell ML to keep the tentative prolongator
ML_Aggregate_Set_Reuse(agg_);
// pray that no bugs will tease us
return(0);
}
