Teuchos::RCP<Map>
Cartesian2D(const Teuchos::RCP<const Teuchos::Comm<int> > & comm, const GlobalOrdinal nx, const GlobalOrdinal ny,
const GlobalOrdinal mx, const GlobalOrdinal my)
{
if (nx <= 0 || ny <= 0 || mx <= 0 || my <= 0 || (mx > nx) || (my > ny))
throw(Exception(__FILE__, __LINE__,
"Incorrect input parameter to Maps::Cartesian2D()",
"nx = " + toString(nx) +
", ny = " + toString(ny) +
", mx = " + toString(mx) +
", my = " + toString(my)));
int MyPID = comm->getRank();
GlobalOrdinal startx, starty, endx, endy;
GlobalOrdinal xpid = MyPID % mx;
GlobalOrdinal ypid = MyPID / mx;
GlobalOrdinal PerProcSmallXDir = Teuchos::as<GlobalOrdinal>(Teuchos::as<double>(nx)/Teuchos::as<double>(mx));
GlobalOrdinal PerProcSmallYDir = Teuchos::as<GlobalOrdinal>(Teuchos::as<double>(ny)/Teuchos::as<double>(my));
GlobalOrdinal NBigXDir = nx - PerProcSmallXDir*mx;
GlobalOrdinal NBigYDir = ny - PerProcSmallYDir*my;
if (xpid < NBigXDir) startx = xpid*(PerProcSmallXDir+1);
else startx = (xpid-NBigXDir)*PerProcSmallXDir +
NBigXDir*(PerProcSmallXDir+1);
endx = startx + PerProcSmallXDir;
if (xpid < NBigXDir) endx++;
if (ypid < NBigYDir) starty = ypid*(PerProcSmallYDir+1);
else starty = (ypid-NBigYDir)*PerProcSmallYDir +
NBigYDir*(PerProcSmallYDir+1);
endy = starty + PerProcSmallYDir;
if ( ypid < NBigYDir) endy++;
size_t NumMyElements = (endx - startx) * (endy - starty);
vector<GlobalOrdinal> MyGlobalElements(NumMyElements);
size_t count = 0;
for (GlobalOrdinal i = startx ; i < endx ; ++i)
for (GlobalOrdinal j = starty ; j < endy ; ++j)
MyGlobalElements[count++] = i + j * nx;
global_size_t numGlobalElements = nx * ny;
const Teuchos::ArrayView<const GlobalOrdinal> elementList(MyGlobalElements);
return MapTraits<GlobalOrdinal,Map>::Build(numGlobalElements, elementList, 0/*indexBase*/, comm /*TODO:node*/);
} // Cartesian2D()
// ======================================================================
Operator GetPtent1D(const MultiVector& D, const int offset = 0)
{
if (D.GetNumVectors() != 1)
ML_THROW("D.GetNumVectors() != 1", -1);
int size = D.GetMyLength();
if (size == 0)
ML_THROW("empty diagonal vector in input", -1);
double* diag = new double[size];
for (int i = 0 ; i < size ; ++i)
diag[i] = D(i);
// creates the ML operator and store the diag pointer,
// as well as the function pointers
ML_Operator* MLDiag = ML_Operator_Create(GetML_Comm());
int invec_leng = size / 3 + size % 3;
int outvec_leng = size;
MLDiag->invec_leng = invec_leng;
MLDiag->outvec_leng = outvec_leng;
MLDiag->data = (void*)diag;
MLDiag->data_destroy = Ptent1D_destroy;
MLDiag->matvec->func_ptr = Ptent1D_matvec;
MLDiag->matvec->ML_id = ML_NONEMPTY;
MLDiag->matvec->Nrows = outvec_leng;
MLDiag->from_an_ml_operator = 0;
MLDiag->getrow->func_ptr = Ptent1D_getrows;
MLDiag->getrow->ML_id = ML_NONEMPTY;
MLDiag->getrow->Nrows = outvec_leng;
// creates the domain space
vector<int> MyGlobalElements(invec_leng);
for (int i = 0 ; i < invec_leng ; ++i)
MyGlobalElements[i] = D.GetVectorSpace()(i * 3) / 3;
Space DomainSpace(invec_leng, -1, &MyGlobalElements[0]);
Space RangeSpace = D.GetVectorSpace();
// creates the MLAPI wrapper
Operator Diag(DomainSpace,RangeSpace,MLDiag,true);
return(Diag);
}
//=============================================================================
Epetra_Map* Epetra_Map::ReplaceCommWithSubset(const Epetra_Comm * theComm) const
{
// mfh 26 Mar 2013: The lazy way to do this is simply to recreate
// the Map by calling its ordinary public constructor, using the
// original Map's data. This only involves O(1) all-reduces over
// the new communicator, which in the common case only includes a
// small number of processes.
Epetra_Map * NewMap=0;
// Create the Map to return (unless theComm is NULL, in which case
// we return zero).
if(theComm) {
// Map requires that the index base equal the global min GID.
// Figuring out the global min GID requires a reduction over all
// processes in the new communicator. It could be that some (or
// even all) of these processes contain zero entries. (Recall
// that this method, unlike removeEmptyProcesses(), may remove
// an arbitrary subset of processes.) We deal with this by
// doing a min over the min GID on each process if the process
// has more than zero entries, or the global max GID, if that
// process has zero entries. If no processes have any entries,
// then the index base doesn't matter anyway.
#ifndef EPETRA_NO_32BIT_GLOBAL_INDICES
if(GlobalIndicesInt()) {
int MyMin, theIndexBase;
MyMin = NumMyElements() > 0 ? MinMyGID() : MaxAllGID();
theComm->MinAll(&MyMin,&theIndexBase,1);
NewMap = new Epetra_Map(-1,NumMyElements(),MyGlobalElements(),theIndexBase,*theComm);
}
else
#endif
#ifndef EPETRA_NO_64BIT_GLOBAL_INDICES
if(GlobalIndicesLongLong()) {
long long MyMin, theIndexBase;
MyMin = NumMyElements() > 0 ? MinMyGID64() : MaxAllGID64();
theComm->MinAll(&MyMin,&theIndexBase,1);
NewMap = new Epetra_Map(-1,NumMyElements(),MyGlobalElements64(),theIndexBase,*theComm);
}
else
#endif
throw ReportError("Epetra_Map::ReplaceCommWithSubset ERROR, GlobalIndices type unknown.",-1);
}
return NewMap;
}
int main(int argc, char *argv[])
{
int i;
bool ierr, gerr;
gerr = true;
#ifdef HAVE_MPI
// Initialize MPI and setup an Epetra communicator
MPI_Init(&argc,&argv);
Teuchos::RCP<Epetra_MpiComm> Comm = Teuchos::rcp( new Epetra_MpiComm(MPI_COMM_WORLD) );
#else
// If we aren't using MPI, then setup a serial communicator.
Teuchos::RCP<Epetra_SerialComm> Comm = Teuchos::rcp( new Epetra_SerialComm() );
#endif
// number of global elements
int dim = 100;
int blockSize = 5;
bool verbose = false;
if (argc>1) {
if (argv[1][0]=='-' && argv[1][1]=='v') {
verbose = true;
}
}
// Construct a Map that puts approximately the same number of
// equations on each processor.
Teuchos::RCP<Epetra_Map> Map = Teuchos::rcp( new Epetra_Map(dim, 0, *Comm) );
// Get update list and number of local equations from newly created Map.
int NumMyElements = Map->NumMyElements();
std::vector<int> MyGlobalElements(NumMyElements);
Map->MyGlobalElements(&MyGlobalElements[0]);
// Create an integer std::vector NumNz that is used to build the Petra Matrix.
// NumNz[i] is the Number of OFF-DIAGONAL term for the ith global equation
// on this processor
std::vector<int> NumNz(NumMyElements);
// We are building a tridiagonal matrix where each row has (-1 2 -1)
// So we need 2 off-diagonal terms (except for the first and last equation)
for (i=0; i<NumMyElements; i++) {
if (MyGlobalElements[i]==0 || MyGlobalElements[i] == dim-1) {
NumNz[i] = 2;
}
else {
NumNz[i] = 3;
}
}
// Create an Epetra_Matrix
Teuchos::RCP<Epetra_CrsMatrix> A = Teuchos::rcp( new Epetra_CrsMatrix(Copy, *Map, &NumNz[0]) );
// Add rows one-at-a-time
// Need some vectors to help
// Off diagonal Values will always be -1
std::vector<double> Values(2);
Values[0] = -1.0; Values[1] = -1.0;
std::vector<int> Indices(2);
double two = 2.0;
int NumEntries;
for (i=0; i<NumMyElements; i++) {
if (MyGlobalElements[i]==0) {
Indices[0] = 1;
NumEntries = 1;
}
else if (MyGlobalElements[i] == dim-1) {
Indices[0] = dim-2;
NumEntries = 1;
}
else {
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
NumEntries = 2;
}
ierr = A->InsertGlobalValues(MyGlobalElements[i],NumEntries,&Values[0],&Indices[0]);
assert(ierr==0);
// Put in the diagonal entry
ierr = A->InsertGlobalValues(MyGlobalElements[i],1,&two,&MyGlobalElements[i]);
assert(ierr==0);
}
// Finish building the epetra matrix A
ierr = A->FillComplete();
assert(ierr==0);
// Issue several useful typedefs;
typedef Belos::MultiVec<double> EMV;
typedef Belos::Operator<double> EOP;
// Create an Epetra_MultiVector for an initial std::vector to start the solver.
// Note that this needs to have the same number of columns as the blocksize.
Teuchos::RCP<Belos::EpetraMultiVec> ivec = Teuchos::rcp( new Belos::EpetraMultiVec(*Map, blockSize) );
ivec->Random();
// Create an output manager to handle the I/O from the solver
Teuchos::RCP<Belos::OutputManager<double> > MyOM = Teuchos::rcp( new Belos::OutputManager<double>() );
if (verbose) {
MyOM->setVerbosity( Belos::Warnings );
}
// test the Epetra adapter multivector
ierr = Belos::TestMultiVecTraits<double,EMV>(MyOM,ivec);
gerr &= ierr;
if (ierr) {
MyOM->print(Belos::Warnings,"*** EpetraAdapter PASSED TestMultiVecTraits()\n");
}
else {
MyOM->print(Belos::Warnings,"*** EpetraAdapter FAILED TestMultiVecTraits() ***\n\n");
}
#ifdef HAVE_MPI
MPI_Finalize();
#endif
if (gerr == false) {
MyOM->print(Belos::Warnings,"End Result: TEST FAILED\n");
return -1;
}
//
// Default return value
//
MyOM->print(Belos::Warnings,"End Result: TEST PASSED\n");
return 0;
}
int main(int argc, char *argv[])
{
int ierr = 0, i, forierr = 0;
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
int rank; // My process ID
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
int rank = 0;
Epetra_SerialComm Comm;
#endif
bool verbose = false;
// Check if we should print results to standard out
if (argc>1) if (argv[1][0]=='-' && argv[1][1]=='v') verbose = true;
int verbose_int = verbose ? 1 : 0;
Comm.Broadcast(&verbose_int, 1, 0);
verbose = verbose_int==1 ? true : false;
// char tmp;
// if (rank==0) cout << "Press any key to continue..."<< endl;
// if (rank==0) cin >> tmp;
// Comm.Barrier();
Comm.SetTracebackMode(0); // This should shut down any error traceback reporting
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
if(verbose && MyPID==0)
cout << Epetra_Version() << endl << endl;
if (verbose) cout << "Processor "<<MyPID<<" of "<< NumProc
<< " is alive."<<endl;
// Redefine verbose to only print on PE 0
if(verbose && rank!=0)
verbose = false;
int NumMyEquations = 10000;
long long NumGlobalEquations = (NumMyEquations * NumProc) + EPETRA_MIN(NumProc,3);
if(MyPID < 3)
NumMyEquations++;
// Construct a Map that puts approximately the same Number of equations on each processor
Epetra_Map Map(NumGlobalEquations, NumMyEquations, 0LL, Comm);
// Get update list and number of local equations from newly created Map
vector<long long> MyGlobalElements(Map.NumMyElements());
Map.MyGlobalElements(&MyGlobalElements[0]);
// Create an integer vector NumNz that is used to build the Petra Matrix.
// NumNz[i] is the Number of OFF-DIAGONAL term for the ith global equation on this processor
vector<int> NumNz(NumMyEquations);
// We are building a tridiagonal matrix where each row has (-1 2 -1)
// So we need 2 off-diagonal terms (except for the first and last equation)
for(i = 0; i < NumMyEquations; i++)
if((MyGlobalElements[i] == 0) || (MyGlobalElements[i] == NumGlobalEquations - 1))
NumNz[i] = 1;
else
NumNz[i] = 2;
// Create a Epetra_Matrix
Epetra_CrsMatrix A(Copy, Map, &NumNz[0]);
EPETRA_TEST_ERR(A.IndicesAreGlobal(),ierr);
EPETRA_TEST_ERR(A.IndicesAreLocal(),ierr);
// Add rows one-at-a-time
// Need some vectors to help
// Off diagonal Values will always be -1
vector<double> Values(2);
Values[0] = -1.0;
Values[1] = -1.0;
vector<long long> Indices(2);
double two = 2.0;
int NumEntries;
forierr = 0;
for(i = 0; i < NumMyEquations; i++) {
if(MyGlobalElements[i] == 0) {
Indices[0] = 1;
NumEntries = 1;
}
else if (MyGlobalElements[i] == NumGlobalEquations-1) {
Indices[0] = NumGlobalEquations-2;
NumEntries = 1;
}
else {
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
NumEntries = 2;
}
forierr += !(A.InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0])==0);
forierr += !(A.InsertGlobalValues(MyGlobalElements[i], 1, &two, &MyGlobalElements[i])>0); // Put in the diagonal entry
}
EPETRA_TEST_ERR(forierr,ierr);
// Finish up
A.FillComplete();
A.OptimizeStorage();
Epetra_JadMatrix JadA(A);
Epetra_JadMatrix JadA1(A);
Epetra_JadMatrix JadA2(A);
// Create vectors for Power method
Epetra_Vector q(Map);
Epetra_Vector z(Map); z.Random();
Epetra_Vector resid(Map);
Epetra_Flops flopcounter;
A.SetFlopCounter(flopcounter);
q.SetFlopCounter(A);
z.SetFlopCounter(A);
resid.SetFlopCounter(A);
JadA.SetFlopCounter(A);
JadA1.SetFlopCounter(A);
JadA2.SetFlopCounter(A);
if (verbose) cout << "=======================================" << endl
<< "Testing Jad using CrsMatrix as input..." << endl
<< "=======================================" << endl;
A.ResetFlops();
powerMethodTests(A, JadA, Map, q, z, resid, verbose);
// Increase diagonal dominance
if (verbose) cout << "\n\nIncreasing the magnitude of first diagonal term and solving again\n\n"
<< endl;
if (A.MyGlobalRow(0)) {
int numvals = A.NumGlobalEntries(0);
vector<double> Rowvals(numvals);
vector<long long> Rowinds(numvals);
A.ExtractGlobalRowCopy(0, numvals, numvals, &Rowvals[0], &Rowinds[0]); // Get A[0,0]
for (i=0; i<numvals; i++) if (Rowinds[i] == 0) Rowvals[i] *= 10.0;
A.ReplaceGlobalValues(0, numvals, &Rowvals[0], &Rowinds[0]);
}
JadA.UpdateValues(A);
A.ResetFlops();
powerMethodTests(A, JadA, Map, q, z, resid, verbose);
if (verbose) cout << "================================================================" << endl
<< "Testing Jad using Jad matrix as input matrix for construction..." << endl
<< "================================================================" << endl;
JadA1.ResetFlops();
powerMethodTests(JadA1, JadA2, Map, q, z, resid, verbose);
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
return ierr ;
}
int main(int argc, char *argv[])
{
int ierr = 0, i;
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
Epetra_SerialComm Comm;
#endif
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
bool verbose = (MyPID==0);
if (verbose)
cout << Epetra_Version() << endl << endl;
cout << Comm << endl;
// Get the number of local equations from the command line
if (argc!=2)
{
if (verbose)
cout << "Usage: " << argv[0] << " number_of_equations" << endl;
std::exit(1);
}
long long NumGlobalElements = std::atoi(argv[1]);
if (NumGlobalElements < NumProc)
{
if (verbose)
cout << "numGlobalBlocks = " << NumGlobalElements
<< " cannot be < number of processors = " << NumProc << endl;
std::exit(1);
}
// Construct a Map that puts approximately the same number of
// equations on each processor.
Epetra_Map Map(NumGlobalElements, 0LL, Comm);
// Get update list and number of local equations from newly created Map.
int NumMyElements = Map.NumMyElements();
std::vector<long long> MyGlobalElements(NumMyElements);
Map.MyGlobalElements(&MyGlobalElements[0]);
// Create an integer vector NumNz that is used to build the Petra Matrix.
// NumNz[i] is the Number of OFF-DIAGONAL term for the ith global equation
// on this processor
std::vector<int> NumNz(NumMyElements);
// We are building a tridiagonal matrix where each row has (-1 2 -1)
// So we need 2 off-diagonal terms (except for the first and last equation)
for (i=0; i<NumMyElements; i++)
if (MyGlobalElements[i]==0 || MyGlobalElements[i] == NumGlobalElements-1)
NumNz[i] = 2;
else
NumNz[i] = 3;
// Create a Epetra_Matrix
Epetra_CrsMatrix A(Copy, Map, &NumNz[0]);
// Add rows one-at-a-time
// Need some vectors to help
// Off diagonal Values will always be -1
std::vector<double> Values(2);
Values[0] = -1.0; Values[1] = -1.0;
std::vector<long long> Indices(2);
double two = 2.0;
int NumEntries;
for (i=0; i<NumMyElements; i++)
{
if (MyGlobalElements[i]==0)
{
Indices[0] = 1;
NumEntries = 1;
}
else if (MyGlobalElements[i] == NumGlobalElements-1)
{
Indices[0] = NumGlobalElements-2;
NumEntries = 1;
}
else
{
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
NumEntries = 2;
}
ierr = A.InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert(ierr==0);
// Put in the diagonal entry
ierr = A.InsertGlobalValues(MyGlobalElements[i], 1, &two, &MyGlobalElements[i]);
assert(ierr==0);
}
// Finish up
ierr = A.FillComplete();
assert(ierr==0);
// Create vectors for Power method
// variable needed for iteration
double lambda = 0.0;
int niters = (int) NumGlobalElements*10;
double tolerance = 1.0e-2;
// Iterate
Epetra_Flops counter;
A.SetFlopCounter(counter);
Epetra_Time timer(Comm);
ierr += power_method(A, lambda, niters, tolerance, verbose);
double elapsed_time = timer.ElapsedTime();
double total_flops =counter.Flops();
double MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose)
cout << "\n\nTotal MFLOPs for first solve = " << MFLOPs << endl<< endl;
// Increase diagonal dominance
if (verbose)
cout << "\nIncreasing magnitude of first diagonal term, solving again\n\n"
<< endl;
if (A.MyGlobalRow(0)) {
int numvals = A.NumGlobalEntries(0);
std::vector<double> Rowvals(numvals);
std::vector<long long> Rowinds(numvals);
A.ExtractGlobalRowCopy(0, numvals, numvals, &Rowvals[0], &Rowinds[0]); // Get A[0,0]
for (i=0; i<numvals; i++) if (Rowinds[i] == 0) Rowvals[i] *= 10.0;
A.ReplaceGlobalValues(0, numvals, &Rowvals[0], &Rowinds[0]);
}
// Iterate (again)
lambda = 0.0;
timer.ResetStartTime();
counter.ResetFlops();
ierr += power_method(A, lambda, niters, tolerance, verbose);
elapsed_time = timer.ElapsedTime();
total_flops = counter.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose)
cout << "\n\nTotal MFLOPs for second solve = " << MFLOPs << endl<< endl;
// Release all objects
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
/* end main
*/
return ierr ;
}
int main(int argc, char *argv[])
{
int ierr = 0, i, forierr = 0;
#ifdef HAVE_MPI
CT_Epetra_MpiComm_ID_Flex_t Comm;
// Initialize MPI
MPI_Init(&argc,&argv);
int rank; // My process ID
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
Comm.Epetra_MpiComm = Epetra_MpiComm_Create( MPI_COMM_WORLD );
#else
int rank = 0;
CT_Epetra_SerialComm_ID_Flex_t Comm;
Comm.Epetra_SerialComm = Epetra_SerialComm_Create();
#endif
bool verbose = false;
// Check if we should print results to standard out
if (argc>1) if (argv[1][0]=='-' && argv[1][1]=='v') verbose = true;
int verbose_int = verbose ? 1 : 0;
Epetra_Comm_Broadcast_Int(Comm.Epetra_Comm, &verbose_int, 1, 0);
verbose = verbose_int==1 ? true : false;
// char tmp;
// if (rank==0) cout << "Press any key to continue..."<< endl;
// if (rank==0) cin >> tmp;
// Epetra_Comm_Barrier(Comm.Epetra_Comm);
// Epetra_Comm_SetTracebackMode(Comm.Epetra_Comm, 0); // This should shut down any error traceback reporting
int MyPID = Epetra_Comm_MyPID(Comm.Epetra_Comm);
int NumProc = Epetra_Comm_NumProc(Comm.Epetra_Comm);
if(verbose && MyPID==0)
cout << Epetra_Version() << endl << endl;
if (verbose) cout << "Processor "<<MyPID<<" of "<< NumProc
<< " is alive."<<endl;
// Redefine verbose to only print on PE 0
if(verbose && rank!=0)
verbose = false;
int NumMyEquations = 10000;
int NumGlobalEquations = (NumMyEquations * NumProc) + EPETRA_MIN(NumProc,3);
if(MyPID < 3)
NumMyEquations++;
// Construct a Map that puts approximately the same Number of equations on each processor
CT_Epetra_Map_ID_Flex_t Map;
Map.Epetra_Map = Epetra_Map_Create_Linear(NumGlobalEquations, NumMyEquations, 0, Comm.Epetra_Comm);
// Get update list and number of local equations from newly created Map
vector<int> MyGlobalElements(Epetra_BlockMap_NumMyElements(Map.Epetra_BlockMap));
Epetra_BlockMap_MyGlobalElements_Fill(Map.Epetra_BlockMap, &MyGlobalElements[0]);
// Create an integer vector NumNz that is used to build the Petra Matrix.
// NumNz[i] is the Number of OFF-DIAGONAL term for the ith global equation on this processor
vector<int> NumNz(NumMyEquations);
// We are building a tridiagonal matrix where each row has (-1 2 -1)
// So we need 2 off-diagonal terms (except for the first and last equation)
for(i = 0; i < NumMyEquations; i++)
if((MyGlobalElements[i] == 0) || (MyGlobalElements[i] == NumGlobalEquations - 1))
NumNz[i] = 1;
else
NumNz[i] = 2;
// Create a Epetra_Matrix
CT_Epetra_CrsMatrix_ID_Flex_t A;
A.Epetra_CrsMatrix = Epetra_CrsMatrix_Create_VarPerRow(
CT_Epetra_DataAccess_E_Copy, Map.Epetra_Map, &NumNz[0], FALSE);
EPETRA_TEST_ERR(Epetra_CrsMatrix_IndicesAreGlobal(A.Epetra_CrsMatrix),ierr);
EPETRA_TEST_ERR(Epetra_CrsMatrix_IndicesAreLocal(A.Epetra_CrsMatrix),ierr);
// Add rows one-at-a-time
// Need some vectors to help
// Off diagonal Values will always be -1
vector<double> Values(2);
Values[0] = -1.0;
Values[1] = -1.0;
vector<int> Indices(2);
double two = 2.0;
int NumEntries;
forierr = 0;
for(i = 0; i < NumMyEquations; i++) {
if(MyGlobalElements[i] == 0) {
Indices[0] = 1;
NumEntries = 1;
}
else if (MyGlobalElements[i] == NumGlobalEquations-1) {
Indices[0] = NumGlobalEquations-2;
NumEntries = 1;
}
else {
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
NumEntries = 2;
}
forierr += !(Epetra_CrsMatrix_InsertGlobalValues(A.Epetra_CrsMatrix,
MyGlobalElements[i], NumEntries, &Values[0], &Indices[0])==0);
forierr += !(Epetra_CrsMatrix_InsertGlobalValues(A.Epetra_CrsMatrix,
MyGlobalElements[i], 1, &two, &MyGlobalElements[i])>0); // Put in the diagonal entry
}
EPETRA_TEST_ERR(forierr,ierr);
// Finish up
Epetra_CrsMatrix_FillComplete(A.Epetra_CrsMatrix, TRUE);
Epetra_CrsMatrix_OptimizeStorage(A.Epetra_CrsMatrix);
CT_Epetra_JadMatrix_ID_Flex_t JadA, JadA1, JadA2;
JadA.Epetra_JadMatrix = Epetra_JadMatrix_Create(A.Epetra_RowMatrix);
JadA1.Epetra_JadMatrix = Epetra_JadMatrix_Create(A.Epetra_RowMatrix);
JadA2.Epetra_JadMatrix = Epetra_JadMatrix_Create(A.Epetra_RowMatrix);
// Create vectors for Power method
CT_Epetra_Vector_ID_Flex_t q, z, resid;
q.Epetra_Vector = Epetra_Vector_Create(Map.Epetra_BlockMap, TRUE);
z.Epetra_Vector = Epetra_Vector_Create(Map.Epetra_BlockMap, TRUE);
Epetra_MultiVector_Random(z.Epetra_MultiVector);
resid.Epetra_Vector = Epetra_Vector_Create(Map.Epetra_BlockMap, TRUE);
CT_Epetra_Flops_ID_t flopcounter = Epetra_Flops_Create();
Epetra_CompObject_SetFlopCounter(A.Epetra_CompObject, flopcounter);
Epetra_CompObject_SetFlopCounter_Matching(q.Epetra_CompObject, A.Epetra_CompObject);
Epetra_CompObject_SetFlopCounter_Matching(z.Epetra_CompObject, A.Epetra_CompObject);
Epetra_CompObject_SetFlopCounter_Matching(resid.Epetra_CompObject, A.Epetra_CompObject);
Epetra_CompObject_SetFlopCounter_Matching(JadA.Epetra_CompObject, A.Epetra_CompObject);
Epetra_CompObject_SetFlopCounter_Matching(JadA1.Epetra_CompObject, A.Epetra_CompObject);
Epetra_CompObject_SetFlopCounter_Matching(JadA2.Epetra_CompObject, A.Epetra_CompObject);
if (verbose) cout << "=======================================" << endl
<< "Testing Jad using CrsMatrix as input..." << endl
<< "=======================================" << endl;
Epetra_CompObject_ResetFlops(A.Epetra_CompObject);
powerMethodTests(A.Epetra_RowMatrix, JadA.Epetra_RowMatrix, Map, q, z, resid, verbose);
// Increase diagonal dominance
if (verbose) cout << "\n\nIncreasing the magnitude of first diagonal term and solving again\n\n"
<< endl;
if (Epetra_CrsMatrix_MyGlobalRow(A.Epetra_CrsMatrix, 0)) {
int numvals = Epetra_CrsMatrix_NumGlobalEntries(A.Epetra_CrsMatrix, 0);
vector<double> Rowvals(numvals);
vector<int> Rowinds(numvals);
Epetra_CrsMatrix_ExtractGlobalRowCopy_WithIndices(A.Epetra_CrsMatrix, 0,
numvals, &numvals, &Rowvals[0], &Rowinds[0]); // Get A[0,0]
for (i=0; i<numvals; i++) if (Rowinds[i] == 0) Rowvals[i] *= 10.0;
Epetra_CrsMatrix_ReplaceGlobalValues(A.Epetra_CrsMatrix, 0, numvals, &Rowvals[0], &Rowinds[0]);
}
Epetra_JadMatrix_UpdateValues(JadA.Epetra_JadMatrix, A.Epetra_RowMatrix, FALSE);
Epetra_CompObject_ResetFlops(A.Epetra_CompObject);
powerMethodTests(A.Epetra_RowMatrix, JadA.Epetra_RowMatrix, Map, q, z, resid, verbose);
if (verbose) cout << "================================================================" << endl
<< "Testing Jad using Jad matrix as input matrix for construction..." << endl
<< "================================================================" << endl;
Epetra_CompObject_ResetFlops(JadA1.Epetra_CompObject);
powerMethodTests(JadA1.Epetra_RowMatrix, JadA2.Epetra_RowMatrix, Map, q, z, resid, verbose);
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return ierr ;
}
int Amesos_Scalapack::RedistributeA( ) {
if( debug_ == 1 ) std::cout << "Entering `RedistributeA()'" << std::endl;
Time_->ResetStartTime();
Epetra_RowMatrix *RowMatrixA = dynamic_cast<Epetra_RowMatrix *>(Problem_->GetOperator());
EPETRA_CHK_ERR( RowMatrixA == 0 ) ;
const Epetra_Map &OriginalMap = RowMatrixA->RowMatrixRowMap() ;
int NumberOfProcesses = Comm().NumProc() ;
//
// Compute a uniform distribution as ScaLAPACK would want it
// MyFirstElement - The first element which this processor would have
// NumExpectedElemetns - The number of elements which this processor would have
//
int NumRows_ = RowMatrixA->NumGlobalRows() ;
int NumColumns_ = RowMatrixA->NumGlobalCols() ;
if ( MaxProcesses_ > 0 ) {
NumberOfProcesses = EPETRA_MIN( NumberOfProcesses, MaxProcesses_ ) ;
}
else {
int ProcessNumHeuristic = (1+NumRows_/200)*(1+NumRows_/200);
NumberOfProcesses = EPETRA_MIN( NumberOfProcesses, ProcessNumHeuristic );
}
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:171" << std::endl;
//
// Create the ScaLAPACK data distribution.
// The TwoD data distribution is created in a completely different
// manner and is not transposed (whereas the SaLAPACK 1D data
// distribution was transposed)
//
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:163" << std::endl;
Comm().Barrier();
if ( TwoD_distribution_ ) {
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:166" << std::endl;
Comm().Barrier();
npcol_ = EPETRA_MIN( NumberOfProcesses,
EPETRA_MAX ( 2, (int) sqrt( NumberOfProcesses * 0.5 ) ) ) ;
nprow_ = NumberOfProcesses / npcol_ ;
//
// Create the map for FatA - our first intermediate matrix
//
int NumMyElements = RowMatrixA->RowMatrixRowMap().NumMyElements() ;
std::vector<int> MyGlobalElements( NumMyElements );
RowMatrixA->RowMatrixRowMap().MyGlobalElements( &MyGlobalElements[0] ) ;
int NumMyColumns = RowMatrixA->RowMatrixColMap().NumMyElements() ;
std::vector<int> MyGlobalColumns( NumMyColumns );
RowMatrixA->RowMatrixColMap().MyGlobalElements( &MyGlobalColumns[0] ) ;
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:194" << std::endl;
std::vector<int> MyFatElements( NumMyElements * npcol_ );
for( int LocalRow=0; LocalRow<NumMyElements; LocalRow++ ) {
for (int i = 0 ; i < npcol_; i++ ){
MyFatElements[LocalRow*npcol_+i] = MyGlobalElements[LocalRow]*npcol_+i;
}
}
Epetra_Map FatInMap( npcol_*NumRows_, NumMyElements*npcol_,
&MyFatElements[0], 0, Comm() );
//
// Create FatIn, our first intermediate matrix
//
Epetra_CrsMatrix FatIn( Copy, FatInMap, 0 );
std::vector<std::vector<int> > FatColumnIndices(npcol_,std::vector<int>(1));
std::vector<std::vector<double> > FatMatrixValues(npcol_,std::vector<double>(1));
std::vector<int> FatRowPtrs(npcol_); // A FatRowPtrs[i] = the number
// of entries in local row LocalRow*npcol_ + i
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:219" << std::endl;
//
mypcol_ = iam_%npcol_;
myprow_ = (iam_/npcol_)%nprow_;
if ( iam_ >= nprow_ * npcol_ ) {
myprow_ = nprow_;
mypcol_ = npcol_;
}
// Each row is split into npcol_ rows, with each of the
// new rows containing only those elements belonging to
// its process column (in the ScaLAPACK 2D process grid)
//
int MaxNumIndices = RowMatrixA->MaxNumEntries();
int NumIndices;
std::vector<int> ColumnIndices(MaxNumIndices);
std::vector<double> MatrixValues(MaxNumIndices);
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:232 NumMyElements = "
<< NumMyElements
<< std::endl;
nb_ = grid_nb_;
for( int LocalRow=0; LocalRow<NumMyElements; ++LocalRow ) {
RowMatrixA->ExtractMyRowCopy( LocalRow,
MaxNumIndices,
NumIndices,
&MatrixValues[0],
&ColumnIndices[0] );
for (int i=0; i<npcol_; i++ ) FatRowPtrs[i] = 0 ;
//
// Deal the individual matrix entries out to the row owned by
// the process to which this matrix entry will belong.
//
for( int i=0 ; i<NumIndices ; ++i ) {
int GlobalCol = MyGlobalColumns[ ColumnIndices[i] ];
int pcol_i = pcolnum( GlobalCol, nb_, npcol_ ) ;
if ( FatRowPtrs[ pcol_i ]+1 >= FatColumnIndices[ pcol_i ].size() ) {
FatColumnIndices[ pcol_i ]. resize( 2 * FatRowPtrs[ pcol_i ]+1 );
FatMatrixValues[ pcol_i ]. resize( 2 * FatRowPtrs[ pcol_i ]+1 );
}
FatColumnIndices[pcol_i][FatRowPtrs[pcol_i]] = GlobalCol ;
FatMatrixValues[pcol_i][FatRowPtrs[pcol_i]] = MatrixValues[i];
FatRowPtrs[ pcol_i ]++;
}
//
// Insert each of the npcol_ rows individually
//
for ( int pcol_i = 0 ; pcol_i < npcol_ ; pcol_i++ ) {
FatIn.InsertGlobalValues( MyGlobalElements[LocalRow]*npcol_ + pcol_i,
FatRowPtrs[ pcol_i ],
&FatMatrixValues[ pcol_i ][0],
&FatColumnIndices[ pcol_i ][0] );
}
}
FatIn.FillComplete( false );
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:260" << std::endl;
if ( debug_ == 1) std::cout << "Amesos_Scalapack.cpp:265B"
<< " iam_ = " << iam_
<< " nb_ = " << nb_
<< " nprow_ = " << nprow_
<< " npcol_ = " << npcol_
<< std::endl;
//
// Compute the map for our second intermediate matrix, FatOut
//
// Compute directly
int UniformRows = ( NumRows_ / ( nprow_ * nb_ ) ) * nb_ ;
int AllExcessRows = NumRows_ - UniformRows * nprow_ ;
int OurExcessRows = EPETRA_MIN( nb_, AllExcessRows - ( myprow_ * nb_ ) ) ;
OurExcessRows = EPETRA_MAX( 0, OurExcessRows );
NumOurRows_ = UniformRows + OurExcessRows ;
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:277" << std::endl;
int UniformColumns = ( NumColumns_ / ( npcol_ * nb_ ) ) * nb_ ;
int AllExcessColumns = NumColumns_ - UniformColumns * npcol_ ;
int OurExcessColumns = EPETRA_MIN( nb_, AllExcessColumns - ( mypcol_ * nb_ ) ) ;
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:281" << std::endl;
OurExcessColumns = EPETRA_MAX( 0, OurExcessColumns );
NumOurColumns_ = UniformColumns + OurExcessColumns ;
if ( iam_ >= nprow_ * npcol_ ) {
UniformRows = 0;
NumOurRows_ = 0;
NumOurColumns_ = 0;
}
Comm().Barrier();
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:295" << std::endl;
#if 0
// Compute using ScaLAPACK's numroc routine, assert agreement
int izero = 0; // All matrices start at process 0
int NumRocSays = numroc_( &NumRows_, &nb_, &myprow_, &izero, &nprow_ );
assert( NumOurRows_ == NumRocSays );
#endif
//
// Compute the rows which this process row owns in the ScaLAPACK 2D
// process grid.
//
std::vector<int> AllOurRows(NumOurRows_);
int RowIndex = 0 ;
int BlockRow = 0 ;
for ( ; BlockRow < UniformRows / nb_ ; BlockRow++ ) {
for ( int RowOffset = 0; RowOffset < nb_ ; RowOffset++ ) {
AllOurRows[RowIndex++] = BlockRow*nb_*nprow_ + myprow_*nb_ + RowOffset ;
}
}
Comm().Barrier();
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:315" << std::endl;
assert ( BlockRow == UniformRows / nb_ ) ;
for ( int RowOffset = 0; RowOffset < OurExcessRows ; RowOffset++ ) {
AllOurRows[RowIndex++] = BlockRow*nb_*nprow_ + myprow_*nb_ + RowOffset ;
}
assert( RowIndex == NumOurRows_ );
//
// Distribute those rows amongst all the processes in that process row
// This is an artificial distribution with the following properties:
// 1) It is a 1D data distribution (each row belogs entirely to
// a single process
// 2) All data which will eventually belong to a given process row,
// is entirely contained within the processes in that row.
//
Comm().Barrier();
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:312" << std::endl;
//
// Compute MyRows directly
//
std::vector<int>MyRows(NumOurRows_);
RowIndex = 0 ;
BlockRow = 0 ;
for ( ; BlockRow < UniformRows / nb_ ; BlockRow++ ) {
for ( int RowOffset = 0; RowOffset < nb_ ; RowOffset++ ) {
MyRows[RowIndex++] = BlockRow*nb_*nprow_*npcol_ +
myprow_*nb_*npcol_ + RowOffset*npcol_ + mypcol_ ;
}
}
Comm().Barrier();
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:326" << std::endl;
assert ( BlockRow == UniformRows / nb_ ) ;
for ( int RowOffset = 0; RowOffset < OurExcessRows ; RowOffset++ ) {
MyRows[RowIndex++] = BlockRow*nb_*nprow_*npcol_ +
myprow_*nb_*npcol_ + RowOffset*npcol_ + mypcol_ ;
}
Comm().Barrier();
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:334" << std::endl;
Comm().Barrier();
for (int i=0; i < NumOurRows_; i++ ) {
assert( MyRows[i] == AllOurRows[i]*npcol_+mypcol_ );
}
Comm().Barrier();
if ( debug_ == 1) std::cout << "Amesos_Scalapack.cpp:340"
<< " iam_ = " << iam_
<< " myprow_ = " << myprow_
<< " mypcol_ = " << mypcol_
<< " NumRows_ = " << NumRows_
<< " NumOurRows_ = " << NumOurRows_
<< std::endl;
Comm().Barrier();
Epetra_Map FatOutMap( npcol_*NumRows_, NumOurRows_, &MyRows[0], 0, Comm() );
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:344" << std::endl;
Comm().Barrier();
if ( FatOut_ ) delete FatOut_ ;
FatOut_ = new Epetra_CrsMatrix( Copy, FatOutMap, 0 ) ;
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:348" << std::endl;
Epetra_Export ExportToFatOut( FatInMap, FatOutMap ) ;
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:360" << std::endl;
FatOut_->Export( FatIn, ExportToFatOut, Add );
FatOut_->FillComplete( false );
//
// Create a map to allow us to redistribute the vectors X and B
//
Epetra_RowMatrix *RowMatrixA = dynamic_cast<Epetra_RowMatrix *>(Problem_->GetOperator());
const Epetra_Map &OriginalMap = RowMatrixA->RowMatrixRowMap() ;
assert( NumGlobalElements_ == OriginalMap.NumGlobalElements() ) ;
int NumMyVecElements = 0 ;
if ( mypcol_ == 0 ) {
NumMyVecElements = NumOurRows_;
}
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:385" << std::endl;
if (VectorMap_) { delete VectorMap_ ; VectorMap_ = 0 ; }
VectorMap_ = new Epetra_Map( NumGlobalElements_,
NumMyVecElements,
&AllOurRows[0],
0,
Comm() );
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << " Amesos_Scalapack.cpp:393 debug_ = "
<< debug_ << std::endl;
} else {
nprow_ = 1 ;
npcol_ = NumberOfProcesses / nprow_ ;
assert ( nprow_ * npcol_ == NumberOfProcesses ) ;
m_per_p_ = ( NumRows_ + NumberOfProcesses - 1 ) / NumberOfProcesses ;
int MyFirstElement = EPETRA_MIN( iam_ * m_per_p_, NumRows_ ) ;
int MyFirstNonElement = EPETRA_MIN( (iam_+1) * m_per_p_, NumRows_ ) ;
int NumExpectedElements = MyFirstNonElement - MyFirstElement ;
assert( NumRows_ == RowMatrixA->NumGlobalRows() ) ;
if ( ScaLAPACK1DMap_ ) delete( ScaLAPACK1DMap_ ) ;
ScaLAPACK1DMap_ = new Epetra_Map( NumRows_, NumExpectedElements, 0, Comm() );
if ( ScaLAPACK1DMatrix_ ) delete( ScaLAPACK1DMatrix_ ) ;
ScaLAPACK1DMatrix_ = new Epetra_CrsMatrix(Copy, *ScaLAPACK1DMap_, 0);
Epetra_Export ExportToScaLAPACK1D_( OriginalMap, *ScaLAPACK1DMap_);
ScaLAPACK1DMatrix_->Export( *RowMatrixA, ExportToScaLAPACK1D_, Add );
ScaLAPACK1DMatrix_->FillComplete( false ) ;
}
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << " Amesos_Scalapack.cpp:417 debug_ = "
<< debug_ << std::endl;
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:402"
<< " nprow_ = " << nprow_
<< " npcol_ = " << npcol_ << std::endl ;
int info;
const int zero = 0 ;
if ( ictxt_ == -1313 ) {
ictxt_ = 0 ;
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:408" << std::endl;
SL_INIT_F77(&ictxt_, &nprow_, &npcol_) ;
}
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:410A" << std::endl;
int nprow;
int npcol;
int myrow;
int mycol;
BLACS_GRIDINFO_F77(&ictxt_, &nprow, &npcol, &myrow, &mycol) ;
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "iam_ = " << iam_ << " Amesos_Scalapack.cpp:410" << std::endl;
if ( iam_ < nprow_ * npcol_ ) {
assert( nprow == nprow_ ) ;
if ( npcol != npcol_ ) std::cout << "Amesos_Scalapack.cpp:430 npcol = " <<
npcol << " npcol_ = " << npcol_ << std::endl ;
assert( npcol == npcol_ ) ;
if ( TwoD_distribution_ ) {
assert( myrow == myprow_ ) ;
assert( mycol == mypcol_ ) ;
lda_ = EPETRA_MAX(1,NumOurRows_) ;
} else {
assert( myrow == 0 ) ;
assert( mycol == iam_ ) ;
nb_ = m_per_p_;
lda_ = EPETRA_MAX(1,NumGlobalElements_);
}
if ( debug_ == 1) std::cout << "iam_ = " << iam_
<< "Amesos_Scalapack.cpp: " << __LINE__
<< " TwoD_distribution_ = " << TwoD_distribution_
<< " NumGlobalElements_ = " << NumGlobalElements_
<< " debug_ = " << debug_
<< " nb_ = " << nb_
<< " lda_ = " << lda_
<< " nprow_ = " << nprow_
<< " npcol_ = " << npcol_
<< " myprow_ = " << myprow_
<< " mypcol_ = " << mypcol_
<< " iam_ = " << iam_ << std::endl ;
AMESOS_PRINT( myprow_ );
DESCINIT_F77(DescA_,
&NumGlobalElements_,
&NumGlobalElements_,
&nb_,
&nb_,
&zero,
&zero,
&ictxt_,
&lda_,
&info) ;
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:441" << std::endl;
assert( info == 0 ) ;
} else {
DescA_[0] = -13;
if ( debug_ == 1) std::cout << "iam_ = " << iam_ << "Amesos_Scalapack.cpp:458 nprow = " << nprow << std::endl;
assert( nprow == -1 ) ;
}
if ( debug_ == 1) std::cout << "Amesos_Scalapack.cpp:446" << std::endl;
MatTime_ += Time_->ElapsedTime();
return 0;
}
int main(int argc, char *argv[]) {
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
bool testFailed;
bool boolret;
int MyPID = Comm.MyPID();
bool verbose = true;
bool debug = false;
std::string which("SM");
Teuchos::CommandLineProcessor cmdp(false,true);
cmdp.setOption("verbose","quiet",&verbose,"Print messages and results.");
cmdp.setOption("debug","nodebug",&debug,"Print debugging information.");
cmdp.setOption("sort",&which,"Targetted eigenvalues (SM,LM,SR,LR,SI,or LI).");
if (cmdp.parse(argc,argv) != Teuchos::CommandLineProcessor::PARSE_SUCCESSFUL) {
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return -1;
}
typedef double ScalarType;
typedef Teuchos::ScalarTraits<ScalarType> ScalarTypeTraits;
typedef ScalarTypeTraits::magnitudeType MagnitudeType;
typedef Epetra_MultiVector MV;
typedef Epetra_Operator OP;
typedef Anasazi::MultiVecTraits<ScalarType,MV> MVTraits;
typedef Anasazi::OperatorTraits<ScalarType,MV,OP> OpTraits;
// Dimension of the matrix
int nx = 10; // Discretization points in any one direction.
int NumGlobalElements = nx*nx; // Size of matrix nx*nx
// Construct a Map that puts approximately the same number of
// equations on each processor.
Epetra_Map Map(NumGlobalElements, 0, Comm);
// Get update list and number of local equations from newly created Map.
int NumMyElements = Map.NumMyElements();
std::vector<int> MyGlobalElements(NumMyElements);
Map.MyGlobalElements(&MyGlobalElements[0]);
// Create an integer vector NumNz that is used to build the Petra Matrix.
// NumNz[i] is the Number of OFF-DIAGONAL term for the ith global equation
// on this processor
std::vector<int> NumNz(NumMyElements);
/* We are building a matrix of block structure:
| T -I |
|-I T -I |
| -I T |
| ... -I|
| -I T|
where each block is dimension nx by nx and the matrix is on the order of
nx*nx. The block T is a tridiagonal matrix.
*/
for (int i=0; i<NumMyElements; i++) {
if (MyGlobalElements[i] == 0 || MyGlobalElements[i] == NumGlobalElements-1 ||
MyGlobalElements[i] == nx-1 || MyGlobalElements[i] == nx*(nx-1) ) {
NumNz[i] = 3;
}
else if (MyGlobalElements[i] < nx || MyGlobalElements[i] > nx*(nx-1) ||
MyGlobalElements[i]%nx == 0 || (MyGlobalElements[i]+1)%nx == 0) {
NumNz[i] = 4;
}
else {
NumNz[i] = 5;
}
}
// Create an Epetra_Matrix
Teuchos::RCP<Epetra_CrsMatrix> A = Teuchos::rcp( new Epetra_CrsMatrix(Copy, Map, &NumNz[0]) );
// Diffusion coefficient, can be set by user.
// When rho*h/2 <= 1, the discrete convection-diffusion operator has real eigenvalues.
// When rho*h/2 > 1, the operator has complex eigenvalues.
double rho = 2*(nx+1);
// Compute coefficients for discrete convection-diffution operator
const double one = 1.0;
std::vector<double> Values(4);
std::vector<int> Indices(4);
double h = one /(nx+1);
double h2 = h*h;
double c = 5.0e-01*rho/ h;
Values[0] = -one/h2 - c; Values[1] = -one/h2 + c; Values[2] = -one/h2; Values[3]= -one/h2;
double diag = 4.0 / h2;
int NumEntries, info;
for (int i=0; i<NumMyElements; i++)
{
if (MyGlobalElements[i]==0)
{
Indices[0] = 1;
Indices[1] = nx;
NumEntries = 2;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[1], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i] == nx*(nx-1))
{
Indices[0] = nx*(nx-1)+1;
Indices[1] = nx*(nx-2);
NumEntries = 2;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[1], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i] == nx-1)
{
Indices[0] = nx-2;
NumEntries = 1;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
Indices[0] = 2*nx-1;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[2], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i] == NumGlobalElements-1)
{
Indices[0] = NumGlobalElements-2;
NumEntries = 1;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
Indices[0] = nx*(nx-1)-1;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[2], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i] < nx)
{
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
Indices[2] = MyGlobalElements[i]+nx;
NumEntries = 3;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i] > nx*(nx-1))
{
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
Indices[2] = MyGlobalElements[i]-nx;
NumEntries = 3;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
}
else if (MyGlobalElements[i]%nx == 0)
{
Indices[0] = MyGlobalElements[i]+1;
Indices[1] = MyGlobalElements[i]-nx;
Indices[2] = MyGlobalElements[i]+nx;
NumEntries = 3;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[1], &Indices[0]);
assert( info==0 );
}
else if ((MyGlobalElements[i]+1)%nx == 0)
{
Indices[0] = MyGlobalElements[i]-nx;
Indices[1] = MyGlobalElements[i]+nx;
NumEntries = 2;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[2], &Indices[0]);
assert( info==0 );
Indices[0] = MyGlobalElements[i]-1;
NumEntries = 1;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
}
else
{
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
Indices[2] = MyGlobalElements[i]-nx;
Indices[3] = MyGlobalElements[i]+nx;
NumEntries = 4;
info = A->InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert( info==0 );
}
// Put in the diagonal entry
info = A->InsertGlobalValues(MyGlobalElements[i], 1, &diag, &MyGlobalElements[i]);
assert( info==0 );
}
// Finish up
info = A->FillComplete();
assert( info==0 );
A->SetTracebackMode(1); // Shutdown Epetra Warning tracebacks
//************************************
// Start the block Davidson iteration
//***********************************
//
// Variables used for the Generalized Davidson Method
//
int nev = 4;
int blockSize = 1;
int maxDim = 50;
int restartDim = 10;
int maxRestarts = 500;
double tol = 1e-10;
// Set verbosity level
int verbosity = Anasazi::Errors + Anasazi::Warnings;
if (verbose) {
verbosity += Anasazi::FinalSummary + Anasazi::TimingDetails;
}
if (debug) {
verbosity += Anasazi::Debug;
}
//
// Create parameter list to pass into solver manager
//
Teuchos::ParameterList MyPL;
MyPL.set( "Verbosity", verbosity );
MyPL.set( "Which", which );
MyPL.set( "Block Size", blockSize );
MyPL.set( "Maximum Subspace Dimension", maxDim);
MyPL.set( "Restart Dimension", restartDim);
MyPL.set( "Maximum Restarts", maxRestarts );
MyPL.set( "Convergence Tolerance", tol );
MyPL.set( "Relative Convergence Tolerance", true );
MyPL.set( "Initial Guess", "User" );
// Create an Epetra_MultiVector for an initial vector to start the solver.
// Note: This needs to have the same number of columns as the blocksize.
Teuchos::RCP<Epetra_MultiVector> ivec = Teuchos::rcp( new Epetra_MultiVector(Map, blockSize) );
ivec->Random();
// Create the eigenproblem.
Teuchos::RCP<Anasazi::BasicEigenproblem<double, MV, OP> > MyProblem = Teuchos::rcp(
new Anasazi::BasicEigenproblem<double,MV,OP>() );
MyProblem->setA(A);
MyProblem->setInitVec(ivec);
// Inform the eigenproblem that the operator A is non-Hermitian
MyProblem->setHermitian(false);
// Set the number of eigenvalues requested
MyProblem->setNEV( nev );
// Inform the eigenproblem that you are finishing passing it information
boolret = MyProblem->setProblem();
if (boolret != true) {
if (verbose && MyPID == 0) {
std::cout << "Anasazi::BasicEigenproblem::setProblem() returned with error." << std::endl;
}
#ifdef HAVE_MPI
MPI_Finalize() ;
#endif
return -1;
}
// Initialize the Block Arnoldi solver
Anasazi::GeneralizedDavidsonSolMgr<double, MV, OP> MySolverMgr(MyProblem, MyPL);
// Solve the problem to the specified tolerances or length
Anasazi::ReturnType returnCode = MySolverMgr.solve();
testFailed = false;
if (returnCode != Anasazi::Converged && MyPID==0 && verbose) {
testFailed = true;
}
// Get the eigenvalues and eigenvectors from the eigenproblem
Anasazi::Eigensolution<ScalarType,MV> sol = MyProblem->getSolution();
std::vector<Anasazi::Value<ScalarType> > evals = sol.Evals;
Teuchos::RCP<MV> evecs = sol.Evecs;
std::vector<int> index = sol.index;
int numev = sol.numVecs;
// Output computed eigenvalues and their direct residuals
if (verbose && MyPID==0) {
int numritz = (int)evals.size();
std::cout.setf(std::ios_base::right, std::ios_base::adjustfield);
std::cout<<std::endl<< "Computed Ritz Values"<< std::endl;
std::cout<< std::setw(16) << "Real Part"
<< std::setw(16) << "Imag Part"
<< std::endl;
std::cout<<"-----------------------------------------------------------"<<std::endl;
for (int i=0; i<numritz; i++) {
std::cout<< std::setw(16) << evals[i].realpart
<< std::setw(16) << evals[i].imagpart
<< std::endl;
}
std::cout<<"-----------------------------------------------------------"<<std::endl;
}
if (numev > 0) {
// Compute residuals.
Teuchos::LAPACK<int,double> lapack;
std::vector<double> normA(numev);
// The problem is non-Hermitian.
int i=0;
std::vector<int> curind(1);
std::vector<double> resnorm(1), tempnrm(1);
Teuchos::RCP<MV> tempAevec;
Teuchos::RCP<const MV> evecr, eveci;
Epetra_MultiVector Aevec(Map,numev);
// Compute A*evecs
OpTraits::Apply( *A, *evecs, Aevec );
Teuchos::SerialDenseMatrix<int,double> Breal(1,1), Bimag(1,1);
while (i<numev) {
if (index[i]==0) {
// Get a view of the current eigenvector (evecr)
curind[0] = i;
evecr = MVTraits::CloneView( *evecs, curind );
// Get a copy of A*evecr
tempAevec = MVTraits::CloneCopy( Aevec, curind );
// Compute A*evecr - lambda*evecr
Breal(0,0) = evals[i].realpart;
MVTraits::MvTimesMatAddMv( -1.0, *evecr, Breal, 1.0, *tempAevec );
// Compute the norm of the residual and increment counter
MVTraits::MvNorm( *tempAevec, resnorm );
normA[i] = resnorm[0] / Teuchos::ScalarTraits<MagnitudeType>::magnitude( evals[i].realpart );
i++;
} else {
// Get a view of the real part of the eigenvector (evecr)
curind[0] = i;
evecr = MVTraits::CloneView( *evecs, curind );
// Get a copy of A*evecr
tempAevec = MVTraits::CloneCopy( Aevec, curind );
// Get a view of the imaginary part of the eigenvector (eveci)
curind[0] = i+1;
eveci = MVTraits::CloneView( *evecs, curind );
// Set the eigenvalue into Breal and Bimag
Breal(0,0) = evals[i].realpart;
Bimag(0,0) = evals[i].imagpart;
// Compute A*evecr - evecr*lambdar + eveci*lambdai
MVTraits::MvTimesMatAddMv( -1.0, *evecr, Breal, 1.0, *tempAevec );
MVTraits::MvTimesMatAddMv( 1.0, *eveci, Bimag, 1.0, *tempAevec );
MVTraits::MvNorm( *tempAevec, tempnrm );
// Get a copy of A*eveci
tempAevec = MVTraits::CloneCopy( Aevec, curind );
// Compute A*eveci - eveci*lambdar - evecr*lambdai
MVTraits::MvTimesMatAddMv( -1.0, *evecr, Bimag, 1.0, *tempAevec );
MVTraits::MvTimesMatAddMv( -1.0, *eveci, Breal, 1.0, *tempAevec );
MVTraits::MvNorm( *tempAevec, resnorm );
// Compute the norms and scale by magnitude of eigenvalue
normA[i] = lapack.LAPY2( tempnrm[0], resnorm[0] ) /
lapack.LAPY2( evals[i].realpart, evals[i].imagpart );
normA[i+1] = normA[i];
i=i+2;
}
}
// Output computed eigenvalues and their direct residuals
if (verbose && MyPID==0) {
std::cout.setf(std::ios_base::right, std::ios_base::adjustfield);
std::cout<<std::endl<< "Actual Residuals"<<std::endl;
std::cout<< std::setw(16) << "Real Part"
<< std::setw(16) << "Imag Part"
<< std::setw(20) << "Direct Residual"<< std::endl;
std::cout<<"-----------------------------------------------------------"<<std::endl;
for (int j=0; j<numev; j++) {
std::cout<< std::setw(16) << evals[j].realpart
<< std::setw(16) << evals[j].imagpart
<< std::setw(20) << normA[j] << std::endl;
if ( normA[j] > tol ) {
testFailed = true;
}
}
std::cout<<"-----------------------------------------------------------"<<std::endl;
}
}
#ifdef EPETRA_MPI
MPI_Finalize();
#endif
if (testFailed) {
if (verbose && MyPID==0) {
std::cout << "End Result: TEST FAILED" << std::endl;
}
return -1;
}
//
// Default return value
//
if (verbose && MyPID==0) {
std::cout << "End Result: TEST PASSED" << std::endl;
}
return 0;
}
int main(int argc, char *argv[])
{
int ierr = 0, i;
#ifdef HAVE_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
Epetra_SerialComm Comm;
#endif
bool success = false;
bool verbose = true;
try {
//int myRank = Comm.MyPID();
//int numGlobalElements = 10000000;
int numGlobalElements = 100;
Teuchos::CommandLineProcessor cmdp(false,true);
cmdp.setOption("numGlobalElements",&numGlobalElements,"Global problem size.");
if (cmdp.parse(argc,argv) != Teuchos::CommandLineProcessor::PARSE_SUCCESSFUL) {
throw -1;
}
Epetra_Map Map(numGlobalElements, 0, Comm);
int NumMyElements = Map.NumMyElements();
std::vector<int> MyGlobalElements(NumMyElements);
Map.MyGlobalElements(&MyGlobalElements[0]);
int NumNz = 3;
// std::vector<int> NumNz(NumMyElements);
// for (i=0; i<NumMyElements; i++)
// if (MyGlobalElements[i]==0 || MyGlobalElements[i] == numGlobalElements-1)
// NumNz[i] = 2;
// else
// NumNz[i] = 3;
// Epetra_CrsMatrix A(Copy, Map, &NumNz[0]);
MemoryUsageStart("Epetra");
PrintMemoryUsage("Initial memory usage", "epetra-init.heap");
Epetra_CrsMatrix A(Copy, Map, NumNz);
PrintMemoryUsage("Memory after CrsMatrix constructor", "epetra-after-ctor.heap");
std::vector<double> Values(2);
Values[0] = -1.0; Values[1] = -1.0;
std::vector<int> Indices(2);
double two = 2.0;
int NumEntries;
for (i=0; i<NumMyElements; i++) {
if (MyGlobalElements[i]==0) {
Indices[0] = 1;
NumEntries = 1;
} else if (MyGlobalElements[i] == numGlobalElements-1) {
Indices[0] = numGlobalElements-2;
NumEntries = 1;
} else {
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
NumEntries = 2;
}
ierr = A.InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0]);
assert(ierr==0);
// Put in the diagonal entry
ierr = A.InsertGlobalValues(MyGlobalElements[i], 1, &two, &MyGlobalElements[i]);
assert(ierr==0);
}
PrintMemoryUsage("Memory after InsertGlobalValues()", "epetra-after-insert.heap");
ierr = A.FillComplete();
assert(ierr == 0);
PrintMemoryUsage("Memory after FillComplete()", "epetra-after-fillcomplete.heap");
MemoryUsageStop();
success = true;
}
TEUCHOS_STANDARD_CATCH_STATEMENTS(verbose, std::cerr, success);
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return ( success ? EXIT_SUCCESS : EXIT_FAILURE );
}
int main(int argc, char *argv[])
{
int i;
bool ierr, gerr;
gerr = true;
#ifdef HAVE_MPI
// Initialize MPI and setup an Epetra communicator
MPI_Init(&argc,&argv);
Teuchos::RCP<Epetra_MpiComm> Comm = Teuchos::rcp( new Epetra_MpiComm(MPI_COMM_WORLD) );
#else
// If we aren't using MPI, then setup a serial communicator.
Teuchos::RCP<Epetra_SerialComm> Comm = Teuchos::rcp( new Epetra_SerialComm() );
#endif
// number of global elements
const int dim = 100;
const int blockSize = 5;
bool verbose = false;
if (argc>1) {
if (argv[1][0]=='-' && argv[1][1]=='v') {
verbose = true;
}
}
// Create an output manager to handle the I/O from the solver
Teuchos::RCP<Anasazi::OutputManager<double> > MyOM = Teuchos::rcp( new Anasazi::BasicOutputManager<double>() );
if (verbose) {
MyOM->setVerbosity( Anasazi::Warnings );
}
#ifndef HAVE_EPETRA_THYRA
MyOM->stream(Anasazi::Warnings)
<< "Please configure Anasazi with:" << std::endl
<< "--enable-epetra-thyra" << std::endl
<< "--enable-anasazi-thyra" << std::endl;
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return -1;
#endif
// Construct a Map that puts approximately the same number of
// equations on each processor.
Teuchos::RCP<Epetra_Map> Map = Teuchos::rcp( new Epetra_Map(dim, 0, *Comm) );
// Get update list and number of local equations from newly created Map.
int NumMyElements = Map->NumMyElements();
std::vector<int> MyGlobalElements(NumMyElements);
Map->MyGlobalElements(&MyGlobalElements[0]);
// Create an integer vector NumNz that is used to build the Petra Matrix.
// NumNz[i] is the Number of OFF-DIAGONAL term for the ith global equation
// on this processor
std::vector<int> NumNz(NumMyElements);
// We are building a tridiagonal matrix where each row has (-1 2 -1)
// So we need 2 off-diagonal terms (except for the first and last equation)
for (i=0; i<NumMyElements; i++) {
if (MyGlobalElements[i]==0 || MyGlobalElements[i] == dim-1) {
NumNz[i] = 2;
}
else {
NumNz[i] = 3;
}
}
// Create an Epetra_Matrix
Teuchos::RCP<Epetra_CrsMatrix> A = Teuchos::rcp( new Epetra_CrsMatrix(Copy, *Map, &NumNz[0]) );
// Add rows one-at-a-time
// Need some vectors to help
// Off diagonal Values will always be -1
std::vector<double> Values(2);
Values[0] = -1.0; Values[1] = -1.0;
std::vector<int> Indices(2);
double two = 2.0;
int NumEntries;
for (i=0; i<NumMyElements; i++) {
if (MyGlobalElements[i]==0) {
Indices[0] = 1;
NumEntries = 1;
}
else if (MyGlobalElements[i] == dim-1) {
Indices[0] = dim-2;
NumEntries = 1;
}
else {
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
NumEntries = 2;
}
ierr = A->InsertGlobalValues(MyGlobalElements[i],NumEntries,&Values[0],&Indices[0]);
assert(ierr==0);
// Put in the diagonal entry
ierr = A->InsertGlobalValues(MyGlobalElements[i],1,&two,&MyGlobalElements[i]);
assert(ierr==0);
}
// Finish building the epetra matrix A
ierr = A->FillComplete();
assert(ierr==0);
#ifdef HAVE_EPETRA_THYRA
typedef Thyra::MultiVectorBase<double> TMVB;
typedef Thyra::LinearOpBase<double> TLOB;
// first, create a Thyra::VectorSpaceBase from an Epetra_Map using the Epetra-Thyra wrappers
Teuchos::RCP<const Thyra::VectorSpaceBase<double> > space = Thyra::create_VectorSpace(Map);
// then, create a Thyra::MultiVectorBase from the Thyra::VectorSpaceBase using Thyra creational functions
Teuchos::RCP<Thyra::MultiVectorBase<double> > thyra_ivec = Thyra::createMembers(space,blockSize);
// then, create a Thyra::LinearOpBase from the Epetra_CrsMatrix using the Epetra-Thyra wrappers
Teuchos::RCP<const Thyra::LinearOpBase<double> > thyra_op = Thyra::epetraLinearOp(A);
// test the Thyra multivector adapter
ierr = Anasazi::TestMultiVecTraits<double,TMVB>(MyOM,thyra_ivec);
gerr |= ierr;
if (ierr) {
MyOM->stream(Anasazi::Warnings) << "*** ThyraAdapter PASSED TestMultiVecTraits()" << std::endl;
}
else {
MyOM->stream(Anasazi::Warnings) << "*** ThyraAdapter FAILED TestMultiVecTraits() ***" << std::endl << std::endl;
}
// test the Thyra operator adapter
ierr = Anasazi::TestOperatorTraits<double,TMVB,TLOB>(MyOM,thyra_ivec,thyra_op);
gerr |= ierr;
if (ierr) {
MyOM->stream(Anasazi::Warnings) << "*** ThyraAdapter PASSED TestOperatorTraits()" << std::endl;
}
else {
MyOM->stream(Anasazi::Warnings) << "*** ThyraAdapter FAILED TestOperatorTraits() ***" << std::endl << std::endl;
}
#endif
#ifdef HAVE_MPI
MPI_Finalize();
#endif
if (gerr == false) {
MyOM->print(Anasazi::Warnings,"End Result: TEST FAILED\n");
return -1;
}
//
// Default return value
//
MyOM->print(Anasazi::Warnings,"End Result: TEST PASSED\n");
return 0;
}
int main(int argc, char *argv[])
{
int ierr = 0, i, forierr = 0;
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
int rank; // My process ID
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
int rank = 0;
Epetra_SerialComm Comm;
#endif
bool verbose = false;
// Check if we should print results to standard out
if (argc>1) if (argv[1][0]=='-' && argv[1][1]=='v') verbose = true;
int verbose_int = verbose ? 1 : 0;
Comm.Broadcast(&verbose_int, 1, 0);
verbose = verbose_int==1 ? true : false;
// char tmp;
// if (rank==0) cout << "Press any key to continue..."<< endl;
// if (rank==0) cin >> tmp;
// Comm.Barrier();
Comm.SetTracebackMode(0); // This should shut down any error traceback reporting
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
if(verbose && MyPID==0)
cout << Epetra_Version() << endl << endl;
if (verbose) cout << "Processor "<<MyPID<<" of "<< NumProc
<< " is alive."<<endl;
// Redefine verbose to only print on PE 0
if(verbose && rank!=0)
verbose = false;
int NumMyEquations = 10000;
int NumGlobalEquations = (NumMyEquations * NumProc) + EPETRA_MIN(NumProc,3);
if(MyPID < 3)
NumMyEquations++;
// Construct a Map that puts approximately the same Number of equations on each processor
Epetra_Map Map(NumGlobalEquations, NumMyEquations, 0, Comm);
// Get update list and number of local equations from newly created Map
vector<int> MyGlobalElements(Map.NumMyElements());
Map.MyGlobalElements(&MyGlobalElements[0]);
// Create an integer vector NumNz that is used to build the Petra Matrix.
// NumNz[i] is the Number of OFF-DIAGONAL term for the ith global equation on this processor
vector<int> NumNz(NumMyEquations);
// We are building a tridiagonal matrix where each row has (-1 2 -1)
// So we need 2 off-diagonal terms (except for the first and last equation)
for(i = 0; i < NumMyEquations; i++)
if((MyGlobalElements[i] == 0) || (MyGlobalElements[i] == NumGlobalEquations - 1))
NumNz[i] = 1;
else
NumNz[i] = 2;
// Create a Epetra_Matrix
Epetra_CrsMatrix A(Copy, Map, &NumNz[0]);
EPETRA_TEST_ERR(A.IndicesAreGlobal(),ierr);
EPETRA_TEST_ERR(A.IndicesAreLocal(),ierr);
// Add rows one-at-a-time
// Need some vectors to help
// Off diagonal Values will always be -1
vector<double> Values(2);
Values[0] = -1.0;
Values[1] = -1.0;
vector<int> Indices(2);
double two = 2.0;
int NumEntries;
forierr = 0;
for(i = 0; i < NumMyEquations; i++) {
if(MyGlobalElements[i] == 0) {
Indices[0] = 1;
NumEntries = 1;
}
else if (MyGlobalElements[i] == NumGlobalEquations-1) {
Indices[0] = NumGlobalEquations-2;
NumEntries = 1;
}
else {
Indices[0] = MyGlobalElements[i]-1;
Indices[1] = MyGlobalElements[i]+1;
NumEntries = 2;
}
forierr += !(A.InsertGlobalValues(MyGlobalElements[i], NumEntries, &Values[0], &Indices[0])==0);
forierr += !(A.InsertGlobalValues(MyGlobalElements[i], 1, &two, &MyGlobalElements[i])>0); // Put in the diagonal entry
}
EPETRA_TEST_ERR(forierr,ierr);
// Finish up
A.FillComplete();
A.OptimizeStorage();
HYPRE_IJMatrix Matrix;
int ilower = Map.MinMyGID();
int iupper = Map.MaxMyGID();
//printf("Proc[%d], ilower = %d, iupper = %d.\n", MyPID, ilower, iupper);
HYPRE_IJMatrixCreate(MPI_COMM_WORLD, ilower, iupper, ilower, iupper, &Matrix);
HYPRE_IJMatrixSetObjectType(Matrix, HYPRE_PARCSR);
HYPRE_IJMatrixInitialize(Matrix);
for(i = 0; i < A.NumMyRows(); i++){
int numElements;
A.NumMyRowEntries(i, numElements);
vector<int> my_indices; my_indices.resize(numElements);
vector<double> my_values; my_values.resize(numElements);
int numEntries;
A.ExtractMyRowCopy(i, numElements, numEntries, &my_values[0], &my_indices[0]);
for(int j = 0; j < numEntries; j++) {
my_indices[j] = A.GCID(my_indices[j]);
}
int GlobalRow[1];
GlobalRow[0] = A.GRID(i);
HYPRE_IJMatrixSetValues(Matrix, 1, &numEntries, GlobalRow, &my_indices[0], &my_values[0]);
}
HYPRE_IJMatrixAssemble(Matrix);
EpetraExt_HypreIJMatrix JadA(Matrix);
JadA.SetMaps(JadA.RowMatrixRowMap(), A.RowMatrixColMap());
// Create vectors for Power method
Epetra_Vector q(Map);
Epetra_Vector z(Map); z.Random();
Epetra_Vector resid(Map);
Epetra_Flops flopcounter;
A.SetFlopCounter(flopcounter);
q.SetFlopCounter(A);
z.SetFlopCounter(A);
resid.SetFlopCounter(A);
JadA.SetFlopCounter(A);
if (verbose) cout << "=======================================" << endl
<< "Testing Jad using CrsMatrix as input..." << endl
<< "=======================================" << endl;
A.ResetFlops();
powerMethodTests(A, JadA, Map, q, z, resid, verbose);
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
return ierr ;
}
int Epetra_BlockMap::MyGlobalElementsPtr(int *& MyGlobalElementList) const
{
MyGlobalElementList = MyGlobalElements();
return(0);
}
//==============================================================================
void Epetra_BlockMap::Print(ostream & os) const
{
int * FirstPointInElementList1 = 0;
int * ElementSizeList1 = 0;
if (!ConstantElementSize()) {
FirstPointInElementList1 = FirstPointInElementList();
ElementSizeList1 = ElementSizeList();
}
int MyPID = Comm().MyPID();
int NumProc = Comm().NumProc();
for (int iproc = 0; iproc < NumProc; iproc++) {
if (MyPID == iproc) {
if (MyPID == 0) {
os << "\nNumber of Global Elements = "; os << NumGlobalElements64(); os << endl;
os << "Number of Global Points = "; os << NumGlobalPoints64(); os << endl;
os << "Maximum of all GIDs = "; os << MaxAllGID64(); os << endl;
os << "Minimum of all GIDs = "; os << MinAllGID64(); os << endl;
os << "Index Base = "; os << IndexBase(); os << endl;
if (ConstantElementSize())
os << "Constant Element Size = "; os << ElementSize(); os << endl;
}
os << endl;
os << "Number of Local Elements = "; os << NumMyElements(); os << endl;
os << "Number of Local Points = "; os << NumMyPoints(); os << endl;
os << "Maximum of my GIDs = "; os << MaxMyGID64(); os << endl;
os << "Minimum of my GIDs = "; os << MinMyGID64(); os << endl;
os << endl;
os.width(14);
os << " MyPID"; os << " ";
os.width(14);
os << " Local Index "; os << " ";
os.width(14);
os << " Global Index "; os << " ";
if (!ConstantElementSize()) {
os.width(14);
os <<" FirstPointInElement "; os << " ";
os.width(14);
os <<" ElementSize "; os << " ";
}
os << endl;
for (int i = 0; i < NumMyElements(); i++) {
os.width(14);
os << MyPID; os << " ";
os.width(14);
os << i; os << " ";
os.width(14);
if(BlockMapData_->GlobalIndicesLongLong_)
{
#ifndef EPETRA_NO_64BIT_GLOBAL_INDICES
long long * MyGlobalElements1 = MyGlobalElements64();
os << MyGlobalElements1[i]; os << " ";
#else
throw ReportError("Epetra_BlockMap::Print: ERROR, GlobalIndicesLongLong but no API for it.",-1);
#endif
}
else if(BlockMapData_->GlobalIndicesInt_)
{
#ifndef EPETRA_NO_32BIT_GLOBAL_INDICES
int * MyGlobalElements1 = MyGlobalElements();
os << MyGlobalElements1[i]; os << " ";
#else
throw ReportError("Epetra_BlockMap::Print: ERROR, no GlobalIndicesLongLong but no API for it.",-1);
#endif
}
if (!ConstantElementSize()) {
os.width(14);
os << FirstPointInElementList1[i]; os << " ";
os.width(14);
os << ElementSizeList1[i]; os << " ";
}
os << endl;
}
os << flush;
}
// Do a few global ops to give I/O a chance to complete
Comm().Barrier();
Comm().Barrier();
Comm().Barrier();
}
return;
}
int main(int argc, char *argv[]) {
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init( &argc, &argv );
//int size, rank; // Number of MPI processes, My process ID
//MPI_Comm_size(MPI_COMM_WORLD, &size);
//MPI_Comm_rank(MPI_COMM_WORLD, &rank);
#else
//int size = 1; // Serial case (not using MPI)
//int rank = 0;
#endif
bool verbose = false;
int nx = 5;
int ny = 5;
if( argc > 1 )
{
if( argc > 4 )
{
cout << "Usage: " << argv[0] << " [-v [nx [ny]]]" << endl;
exit(1);
}
int loc = 1;
// Check if we should print results to standard out
if(argv[loc][0]=='-' && argv[loc][1]=='v')
{ verbose = true; ++loc; }
if (loc < argc) nx = atoi( argv[loc++] );
if( loc < argc) ny = atoi( argv[loc] );
}
#ifdef EPETRA_MPI
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
bool verbose1 = false;
if(verbose) verbose1 = (MyPID==0);
if(verbose1)
cout << EpetraExt::EpetraExt_Version() << endl << endl;
Comm.Barrier();
if(verbose) cout << Comm << endl << flush;
Comm.Barrier();
int NumGlobalElements = nx * ny;
if( NumGlobalElements < NumProc )
{
cout << "NumGlobalElements = " << NumGlobalElements <<
" cannot be < number of processors = " << NumProc;
exit(1);
}
int IndexBase = 0;
Epetra_Map Map( NumGlobalElements, IndexBase, Comm );
// Extract the global indices of the elements local to this processor
int NumMyElements = Map.NumMyElements();
std::vector<int> MyGlobalElements( NumMyElements );
Map.MyGlobalElements( &MyGlobalElements[0] );
if( verbose ) cout << Map;
// Create the number of non-zeros for a tridiagonal (1D problem) or banded
// (2D problem) matrix
std::vector<int> NumNz( NumMyElements, 5 );
int global_i;
int global_j;
for (int i = 0; i < NumMyElements; ++i)
{
global_j = MyGlobalElements[i] / nx;
global_i = MyGlobalElements[i] - global_j * nx;
if (global_i == 0) NumNz[i] -= 1; // By having separate statements,
if (global_i == nx-1) NumNz[i] -= 1; // this works for 2D as well as 1D
if (global_j == 0) NumNz[i] -= 1; // systems (i.e. nx x 1 or 1 x ny)
if (global_j == ny-1) NumNz[i] -= 1; // or even a 1 x 1 system
}
if(verbose)
{
cout << endl << "NumNz: ";
for (int i = 0; i < NumMyElements; i++) cout << NumNz[i] << " ";
cout << endl;
} // end if
// Create the Epetra Compressed Row Sparse Graph
Epetra_CrsGraph A( Copy, Map, &NumNz[0] );
std::vector<int> Indices(5);
int NumEntries;
for (int i = 0; i < NumMyElements; ++i )
{
global_j = MyGlobalElements[i] / nx;
global_i = MyGlobalElements[i] - global_j * nx;
NumEntries = 0;
// (i,j-1) entry
if (global_j > 0 && ny > 1)
Indices[NumEntries++] = global_i + (global_j-1)*nx;
// (i-1,j) entry
if (global_i > 0)
Indices[NumEntries++] = global_i-1 + global_j *nx;
// (i,j) entry
Indices[NumEntries++] = MyGlobalElements[i];
// (i+1,j) entry
if (global_i < nx-1)
Indices[NumEntries++] = global_i+1 + global_j *nx;
// (i,j+1) entry
if (global_j < ny-1 && ny > 1)
Indices[NumEntries++] = global_i + (global_j+1)*nx;
// Insert the global indices
A.InsertGlobalIndices( MyGlobalElements[i], NumEntries, &Indices[0] );
} // end i loop
// Finish up graph construction
A.FillComplete();
EpetraExt::CrsGraph_MapColoring
Greedy0MapColoringTransform( EpetraExt::CrsGraph_MapColoring::GREEDY,
0, false, verbose );
Epetra_MapColoring & Greedy0ColorMap = Greedy0MapColoringTransform( A );
printColoring(Greedy0ColorMap, &A,verbose);
EpetraExt::CrsGraph_MapColoring
Greedy1MapColoringTransform( EpetraExt::CrsGraph_MapColoring::GREEDY,
1, false, verbose );
Epetra_MapColoring & Greedy1ColorMap = Greedy1MapColoringTransform( A );
printColoring(Greedy1ColorMap, &A,verbose);
EpetraExt::CrsGraph_MapColoring
Greedy2MapColoringTransform( EpetraExt::CrsGraph_MapColoring::GREEDY,
2, false, verbose );
Epetra_MapColoring & Greedy2ColorMap = Greedy2MapColoringTransform( A );
printColoring(Greedy2ColorMap, &A,verbose);
EpetraExt::CrsGraph_MapColoring
Lubi0MapColoringTransform( EpetraExt::CrsGraph_MapColoring::LUBY,
0, false, verbose );
Epetra_MapColoring & Lubi0ColorMap = Lubi0MapColoringTransform( A );
printColoring(Lubi0ColorMap, &A,verbose);
EpetraExt::CrsGraph_MapColoring
Lubi1MapColoringTransform( EpetraExt::CrsGraph_MapColoring::LUBY,
1, false, verbose );
Epetra_MapColoring & Lubi1ColorMap = Lubi1MapColoringTransform( A );
printColoring(Lubi1ColorMap, &A,verbose);
EpetraExt::CrsGraph_MapColoring
Lubi2MapColoringTransform( EpetraExt::CrsGraph_MapColoring::LUBY,
2, false, verbose );
Epetra_MapColoring & Lubi2ColorMap = Lubi2MapColoringTransform( A );
printColoring(Lubi2ColorMap, &A,verbose);
#ifdef EPETRA_MPI
if( verbose ) cout << "Parallel Map Coloring 1!\n";
EpetraExt::CrsGraph_MapColoring
Parallel1MapColoringTransform( EpetraExt::CrsGraph_MapColoring::PSEUDO_PARALLEL,
0, false, verbose );
Epetra_MapColoring & Parallel1ColorMap = Parallel1MapColoringTransform( A );
printColoring(Parallel1ColorMap, &A,verbose);
if( verbose ) cout << "Parallel Map Coloring 2!\n";
EpetraExt::CrsGraph_MapColoring
Parallel2MapColoringTransform( EpetraExt::CrsGraph_MapColoring::JONES_PLASSMAN,
0, false, verbose );
Epetra_MapColoring & Parallel2ColorMap = Parallel2MapColoringTransform( A );
printColoring(Parallel2ColorMap, &A,verbose);
#endif
#ifdef EPETRA_MPI
MPI_Finalize();
#endif
return 0;
}