void runJadMatrixTests(Epetra_JadMatrix * A, Epetra_MultiVector * b, Epetra_MultiVector * bt,
Epetra_MultiVector * xexact, bool StaticProfile, bool verbose, bool summary) {
Epetra_MultiVector z(*b);
Epetra_MultiVector r(*b);
Epetra_SerialDenseVector resvec(b->NumVectors());
//Timings
Epetra_Flops flopcounter;
A->SetFlopCounter(flopcounter);
Epetra_Time timer(A->Comm());
for (int j=0; j<2; j++) { // j = 0 is notrans, j = 1 is trans
bool TransA = (j==1);
A->SetUseTranspose(TransA);
flopcounter.ResetFlops();
timer.ResetStartTime();
//10 matvecs
for( int i = 0; i < 10; ++i )
A->Apply(*xexact, z); // Compute z = A*xexact or z = A'*xexact
double elapsed_time = timer.ElapsedTime();
double total_flops = A->Flops();
// Compute residual
if (TransA)
r.Update(-1.0, z, 1.0, *bt, 0.0); // r = bt - z
else
r.Update(-1.0, z, 1.0, *b, 0.0); // r = b - z
r.Norm2(resvec.Values());
if (verbose) cout << "ResNorm = " << resvec.NormInf() << ": ";
double MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Total MFLOPs for 10 " << " Jagged Diagonal MatVec's with (Trans = " << TransA
<< ") " << MFLOPs << " (" << elapsed_time << " s)" <<endl;
if (summary) {
if (A->Comm().NumProc()==1) {
if (TransA) cout << "TransMv" << '\t';
else cout << "NoTransMv" << '\t';
}
cout << MFLOPs << endl;
}
}
return;
}
int main(int argc, char *argv[])
{
int ierr = 0, i, j, k;
bool debug = false;
#ifdef EPETRA_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
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;
if (verbose && Comm.MyPID()==0)
cout << Epetra_Version() << endl << endl;
int rank = Comm.MyPID();
// 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
if (verbose) cout << Comm <<endl;
// bool verbose1 = verbose;
// Redefine verbose to only print on PE 0
if (verbose && rank!=0) verbose = false;
int N = 20;
int NRHS = 4;
double * A = new double[N*N];
double * A1 = new double[N*N];
double * X = new double[(N+1)*NRHS];
double * X1 = new double[(N+1)*NRHS];
int LDX = N+1;
int LDX1 = N+1;
double * B = new double[N*NRHS];
double * B1 = new double[N*NRHS];
int LDB = N;
int LDB1 = N;
int LDA = N;
int LDA1 = LDA;
double OneNorm1;
bool Transpose = false;
Epetra_SerialDenseSolver solver;
Epetra_SerialDenseMatrix * Matrix;
for (int kk=0; kk<2; kk++) {
for (i=1; i<=N; i++) {
GenerateHilbert(A, LDA, i);
OneNorm1 = 0.0;
for (j=1; j<=i; j++) OneNorm1 += 1.0/((double) j); // 1-Norm = 1 + 1/2 + ...+1/n
if (kk==0) {
Matrix = new Epetra_SerialDenseMatrix(View, A, LDA, i, i);
LDA1 = LDA;
}
else {
Matrix = new Epetra_SerialDenseMatrix(Copy, A, LDA, i, i);
LDA1 = i;
}
GenerateHilbert(A1, LDA1, i);
if (kk==1) {
solver.FactorWithEquilibration(true);
solver.SolveWithTranspose(true);
Transpose = true;
solver.SolveToRefinedSolution(true);
}
for (k=0; k<NRHS; k++)
for (j=0; j<i; j++) {
B[j+k*LDB] = 1.0/((double) (k+3)*(j+3));
B1[j+k*LDB1] = B[j+k*LDB1];
}
Epetra_SerialDenseMatrix Epetra_B(View, B, LDB, i, NRHS);
Epetra_SerialDenseMatrix Epetra_X(View, X, LDX, i, NRHS);
solver.SetMatrix(*Matrix);
solver.SetVectors(Epetra_X, Epetra_B);
ierr = check(solver, A1, LDA1, i, NRHS, OneNorm1, B1, LDB1, X1, LDX1, Transpose, verbose);
assert (ierr>-1);
delete Matrix;
if (ierr!=0) {
if (verbose) cout << "Factorization failed due to bad conditioning. This is normal if RCOND is small."
<< endl;
break;
}
}
}
delete [] A;
delete [] A1;
delete [] X;
delete [] X1;
delete [] B;
delete [] B1;
/////////////////////////////////////////////////////////////////////
// Now test norms and scaling functions
/////////////////////////////////////////////////////////////////////
Epetra_SerialDenseMatrix D;
double ScalarA = 2.0;
int DM = 10;
int DN = 8;
D.Shape(DM, DN);
for (j=0; j<DN; j++)
for (i=0; i<DM; i++) D[j][i] = (double) (1+i+j*DM) ;
//cout << D << endl;
double NormInfD_ref = (double)(DM*(DN*(DN+1))/2);
double NormOneD_ref = (double)((DM*DN*(DM*DN+1))/2 - (DM*(DN-1)*(DM*(DN-1)+1))/2 );
double NormInfD = D.NormInf();
double NormOneD = D.NormOne();
if (verbose) {
cout << " *** Before scaling *** " << endl
<< " Computed one-norm of test matrix = " << NormOneD << endl
<< " Expected one-norm = " << NormOneD_ref << endl
<< " Computed inf-norm of test matrix = " << NormInfD << endl
<< " Expected inf-norm = " << NormInfD_ref << endl;
}
D.Scale(ScalarA); // Scale entire D matrix by this value
NormInfD = D.NormInf();
NormOneD = D.NormOne();
if (verbose) {
cout << " *** After scaling *** " << endl
<< " Computed one-norm of test matrix = " << NormOneD << endl
<< " Expected one-norm = " << NormOneD_ref*ScalarA << endl
<< " Computed inf-norm of test matrix = " << NormInfD << endl
<< " Expected inf-norm = " << NormInfD_ref*ScalarA << endl;
}
/////////////////////////////////////////////////////////////////////
// Now test that A.Multiply(false, x, y) produces the same result
// as y.Multiply('N','N', 1.0, A, x, 0.0).
/////////////////////////////////////////////////////////////////////
N = 10;
int M = 10;
LDA = N;
Epetra_SerialDenseMatrix smallA(N, M, false);
Epetra_SerialDenseMatrix x(N, 1, false);
Epetra_SerialDenseMatrix y1(N, 1, false);
Epetra_SerialDenseMatrix y2(N, 1, false);
for(i=0; i<N; ++i) {
for(j=0; j<M; ++j) {
smallA(i,j) = 1.0*i+2.0*j+1.0;
}
x(i,0) = 1.0;
y1(i,0) = 0.0;
y2(i,0) = 0.0;
}
//quick check of operator==
if (x == y1) {
if (verbose) cout << "err in Epetra_SerialDenseMatrix::operator==, "
<< "erroneously returned true." << std::endl;
return(-1);
}
//quick check of operator!=
if (x != x) {
if (verbose) cout << "err in Epetra_SerialDenseMatrix::operator==, "
<< "erroneously returned true." << std::endl;
return(-1);
}
int err1 = smallA.Multiply(false, x, y1);
int err2 = y2.Multiply('N','N', 1.0, smallA, x, 0.0);
if (err1 != 0 || err2 != 0) {
if (verbose) cout << "err in Epetra_SerialDenseMatrix::Multiply"<<endl;
return(err1+err2);
}
for(i=0; i<N; ++i) {
if (y1(i,0) != y2(i,0)) {
if (verbose) cout << "different versions of Multiply don't match."<<endl;
return(-99);
}
}
/////////////////////////////////////////////////////////////////////
// Now test for larger system, both correctness and performance.
/////////////////////////////////////////////////////////////////////
N = 2000;
NRHS = 5;
LDA = N;
LDB = N;
LDX = N;
if (verbose) cout << "\n\nComputing factor of an " << N << " x " << N << " general matrix...Please wait.\n\n" << endl;
// Define A and X
A = new double[LDA*N];
X = new double[LDB*NRHS];
for (j=0; j<N; j++) {
for (k=0; k<NRHS; k++) X[j+k*LDX] = 1.0/((double) (j+5+k));
for (i=0; i<N; i++) {
if (i==((j+2)%N)) A[i+j*LDA] = 100.0 + i;
else A[i+j*LDA] = -11.0/((double) (i+5)*(j+2));
}
}
// Define Epetra_SerialDenseMatrix object
Epetra_SerialDenseMatrix BigMatrix(Copy, A, LDA, N, N);
Epetra_SerialDenseMatrix OrigBigMatrix(View, A, LDA, N, N);
Epetra_SerialDenseSolver BigSolver;
BigSolver.FactorWithEquilibration(true);
BigSolver.SetMatrix(BigMatrix);
// Time factorization
Epetra_Flops counter;
BigSolver.SetFlopCounter(counter);
Epetra_Time Timer(Comm);
double tstart = Timer.ElapsedTime();
ierr = BigSolver.Factor();
if (ierr!=0 && verbose) cout << "Error in factorization = "<<ierr<< endl;
assert(ierr==0);
double time = Timer.ElapsedTime() - tstart;
double FLOPS = counter.Flops();
double MFLOPS = FLOPS/time/1000000.0;
if (verbose) cout << "MFLOPS for Factorization = " << MFLOPS << endl;
// Define Left hand side and right hand side
Epetra_SerialDenseMatrix LHS(View, X, LDX, N, NRHS);
Epetra_SerialDenseMatrix RHS;
RHS.Shape(N,NRHS); // Allocate RHS
// Compute RHS from A and X
Epetra_Flops RHS_counter;
RHS.SetFlopCounter(RHS_counter);
tstart = Timer.ElapsedTime();
RHS.Multiply('N', 'N', 1.0, OrigBigMatrix, LHS, 0.0);
time = Timer.ElapsedTime() - tstart;
Epetra_SerialDenseMatrix OrigRHS = RHS;
FLOPS = RHS_counter.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) cout << "MFLOPS to build RHS (NRHS = " << NRHS <<") = " << MFLOPS << endl;
// Set LHS and RHS and solve
BigSolver.SetVectors(LHS, RHS);
tstart = Timer.ElapsedTime();
ierr = BigSolver.Solve();
if (ierr==1 && verbose) cout << "LAPACK guidelines suggest this matrix might benefit from equilibration." << endl;
else if (ierr!=0 && verbose) cout << "Error in solve = "<<ierr<< endl;
assert(ierr>=0);
time = Timer.ElapsedTime() - tstart;
FLOPS = BigSolver.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) cout << "MFLOPS for Solve (NRHS = " << NRHS <<") = " << MFLOPS << endl;
double * resid = new double[NRHS];
bool OK = Residual(N, NRHS, A, LDA, BigSolver.Transpose(), BigSolver.X(), BigSolver.LDX(),
OrigRHS.A(), OrigRHS.LDA(), resid);
if (verbose) {
if (!OK) cout << "************* Residual do not meet tolerance *************" << endl;
for (i=0; i<NRHS; i++)
cout << "Residual[" << i <<"] = "<< resid[i] << endl;
cout << endl;
}
// Solve again using the Epetra_SerialDenseVector class for LHS and RHS
Epetra_SerialDenseVector X2;
Epetra_SerialDenseVector B2;
X2.Size(BigMatrix.N());
B2.Size(BigMatrix.M());
int length = BigMatrix.N();
{for (int kk=0; kk<length; kk++) X2[kk] = ((double ) kk)/ ((double) length);} // Define entries of X2
RHS_counter.ResetFlops();
B2.SetFlopCounter(RHS_counter);
tstart = Timer.ElapsedTime();
B2.Multiply('N', 'N', 1.0, OrigBigMatrix, X2, 0.0); // Define B2 = A*X2
time = Timer.ElapsedTime() - tstart;
Epetra_SerialDenseVector OrigB2 = B2;
FLOPS = RHS_counter.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) cout << "MFLOPS to build single RHS = " << MFLOPS << endl;
// Set LHS and RHS and solve
BigSolver.SetVectors(X2, B2);
tstart = Timer.ElapsedTime();
ierr = BigSolver.Solve();
time = Timer.ElapsedTime() - tstart;
if (ierr==1 && verbose) cout << "LAPACK guidelines suggest this matrix might benefit from equilibration." << endl;
else if (ierr!=0 && verbose) cout << "Error in solve = "<<ierr<< endl;
assert(ierr>=0);
FLOPS = counter.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) cout << "MFLOPS to solve single RHS = " << MFLOPS << endl;
OK = Residual(N, 1, A, LDA, BigSolver.Transpose(), BigSolver.X(), BigSolver.LDX(), OrigB2.A(),
OrigB2.LDA(), resid);
if (verbose) {
if (!OK) cout << "************* Residual do not meet tolerance *************" << endl;
cout << "Residual = "<< resid[0] << endl;
}
delete [] resid;
delete [] A;
delete [] X;
///////////////////////////////////////////////////
// Now test default constructor and index operators
///////////////////////////////////////////////////
N = 5;
Epetra_SerialDenseMatrix C; // Implicit call to default constructor, should not need to call destructor
C.Shape(5,5); // Make it 5 by 5
double * C1 = new double[N*N];
GenerateHilbert(C1, N, N); // Generate Hilber matrix
C1[1+2*N] = 1000.0; // Make matrix nonsymmetric
// Fill values of C with Hilbert values
for (i=0; i<N; i++)
for (j=0; j<N; j++)
C(i,j) = C1[i+j*N];
// Test if values are correctly written and read
for (i=0; i<N; i++)
for (j=0; j<N; j++) {
assert(C(i,j) == C1[i+j*N]);
assert(C(i,j) == C[j][i]);
}
if (verbose)
cout << "Default constructor and index operator check OK. Values of Hilbert matrix = "
<< endl << C << endl
<< "Values should be 1/(i+j+1), except value (1,2) should be 1000" << endl;
delete [] C1;
// now test sized/shaped constructor
Epetra_SerialDenseMatrix shapedMatrix(10, 12);
assert(shapedMatrix.M() == 10);
assert(shapedMatrix.N() == 12);
for(i = 0; i < 10; i++)
for(j = 0; j < 12; j++)
assert(shapedMatrix(i, j) == 0.0);
Epetra_SerialDenseVector sizedVector(20);
assert(sizedVector.Length() == 20);
for(i = 0; i < 20; i++)
assert(sizedVector(i) == 0.0);
if (verbose)
cout << "Shaped/sized constructors check OK." << endl;
// test Copy/View mode in op= and cpy ctr
int temperr = 0;
temperr = matrixAssignment(verbose, debug);
if(verbose && temperr == 0)
cout << "Operator = checked OK." << endl;
EPETRA_TEST_ERR(temperr, ierr);
temperr = matrixCpyCtr(verbose, debug);
if(verbose && temperr == 0)
cout << "Copy ctr checked OK." << endl;
EPETRA_TEST_ERR(temperr, ierr);
// Test some vector methods
Epetra_SerialDenseVector v1(3);
v1[0] = 1.0;
v1[1] = 3.0;
v1[2] = 2.0;
Epetra_SerialDenseVector v2(3);
v2[0] = 2.0;
v2[1] = 1.0;
v2[2] = -2.0;
temperr = 0;
if (v1.Norm1()!=6.0) temperr++;
if (fabs(sqrt(14.0)-v1.Norm2())>1.0e-6) temperr++;
if (v1.NormInf()!=3.0) temperr++;
if(verbose && temperr == 0)
cout << "Vector Norms checked OK." << endl;
temperr = 0;
if (v1.Dot(v2)!=1.0) temperr++;
if(verbose && temperr == 0)
cout << "Vector Dot product checked OK." << endl;
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
/* end main
*/
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[]) {
#ifdef EPETRA_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm (MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
cout << Comm << endl;
int MyPID = Comm.MyPID();
bool verbose = false;
if (MyPID==0) verbose = true;
if(argc < 2 && verbose) {
cerr << "Usage: " << argv[0]
<< " HB_filename [level_fill [level_overlap [absolute_threshold [ relative_threshold]]]]" << endl
<< "where:" << endl
<< "HB_filename - filename and path of a Harwell-Boeing data set" << endl
<< "level_fill - The amount of fill to use for ILU(k) preconditioner (default 0)" << endl
<< "level_overlap - The amount of overlap used for overlapping Schwarz subdomains (default 0)" << endl
<< "absolute_threshold - The minimum value to place on the diagonal prior to factorization (default 0.0)" << endl
<< "relative_threshold - The relative amount to perturb the diagonal prior to factorization (default 1.0)" << endl << endl
<< "To specify a non-default value for one of these parameters, you must specify all" << endl
<< " preceding values but not any subsequent parameters. Example:" << endl
<< "ifpackHbSerialMsr.exe mymatrix.hb 1 - loads mymatrix.hb, uses level fill of one, all other values are defaults" << endl
<< endl;
return(1);
}
// Uncomment the next three lines to debug in mpi mode
//int tmp;
//if (MyPID==0) cin >> tmp;
//Comm.Barrier();
Epetra_Map * readMap;
Epetra_CrsMatrix * readA;
Epetra_Vector * readx;
Epetra_Vector * readb;
Epetra_Vector * readxexact;
// Call routine to read in HB problem
Trilinos_Util_ReadHb2Epetra(argv[1], Comm, readMap, readA, readx, readb, readxexact);
// Create uniform distributed map
Epetra_Map map(readMap->NumGlobalElements(), 0, Comm);
// Create Exporter to distribute read-in matrix and vectors
Epetra_Export exporter(*readMap, map);
Epetra_CrsMatrix A(Copy, map, 0);
Epetra_Vector x(map);
Epetra_Vector b(map);
Epetra_Vector xexact(map);
Epetra_Time FillTimer(Comm);
x.Export(*readx, exporter, Add);
b.Export(*readb, exporter, Add);
xexact.Export(*readxexact, exporter, Add);
Comm.Barrier();
double vectorRedistributeTime = FillTimer.ElapsedTime();
A.Export(*readA, exporter, Add);
Comm.Barrier();
double matrixRedistributeTime = FillTimer.ElapsedTime() - vectorRedistributeTime;
assert(A.FillComplete()==0);
Comm.Barrier();
double fillCompleteTime = FillTimer.ElapsedTime() - matrixRedistributeTime;
if (Comm.MyPID()==0) {
cout << "\n\n****************************************************" << endl;
cout << "\n Vector redistribute time (sec) = " << vectorRedistributeTime<< endl;
cout << " Matrix redistribute time (sec) = " << matrixRedistributeTime << endl;
cout << " Transform to Local time (sec) = " << fillCompleteTime << endl<< endl;
}
Epetra_Vector tmp1(*readMap);
Epetra_Vector tmp2(map);
readA->Multiply(false, *readxexact, tmp1);
A.Multiply(false, xexact, tmp2);
double residual;
tmp1.Norm2(&residual);
if (verbose) cout << "Norm of Ax from file = " << residual << endl;
tmp2.Norm2(&residual);
if (verbose) cout << "Norm of Ax after redistribution = " << residual << endl << endl << endl;
//cout << "A from file = " << *readA << endl << endl << endl;
//cout << "A after dist = " << A << endl << endl << endl;
delete readA;
delete readx;
delete readb;
delete readxexact;
delete readMap;
Comm.Barrier();
// Construct ILU preconditioner
double elapsed_time, total_flops, MFLOPs;
Epetra_Time timer(Comm);
int LevelFill = 0;
if (argc > 2) LevelFill = atoi(argv[2]);
if (verbose) cout << "Using Level Fill = " << LevelFill << endl;
int Overlap = 0;
if (argc > 3) Overlap = atoi(argv[3]);
if (verbose) cout << "Using Level Overlap = " << Overlap << endl;
double Athresh = 0.0;
if (argc > 4) Athresh = atof(argv[4]);
if (verbose) cout << "Using Absolute Threshold Value of = " << Athresh << endl;
double Rthresh = 1.0;
if (argc > 5) Rthresh = atof(argv[5]);
if (verbose) cout << "Using Relative Threshold Value of = " << Rthresh << endl;
Ifpack_IlukGraph * IlukGraph = 0;
Ifpack_CrsRiluk * ILUK = 0;
if (LevelFill>-1) {
elapsed_time = timer.ElapsedTime();
IlukGraph = new Ifpack_IlukGraph(A.Graph(), LevelFill, Overlap);
assert(IlukGraph->ConstructFilledGraph()==0);
elapsed_time = timer.ElapsedTime() - elapsed_time;
if (verbose) cout << "Time to construct ILUK graph = " << elapsed_time << endl;
Epetra_Flops fact_counter;
elapsed_time = timer.ElapsedTime();
ILUK = new Ifpack_CrsRiluk(*IlukGraph);
ILUK->SetFlopCounter(fact_counter);
ILUK->SetAbsoluteThreshold(Athresh);
ILUK->SetRelativeThreshold(Rthresh);
//assert(ILUK->InitValues()==0);
int initerr = ILUK->InitValues(A);
if (initerr!=0) cout << Comm << "InitValues error = " << initerr;
assert(ILUK->Factor()==0);
elapsed_time = timer.ElapsedTime() - elapsed_time;
total_flops = ILUK->Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Time to compute preconditioner values = "
<< elapsed_time << endl
<< "MFLOPS for Factorization = " << MFLOPs << endl;
//cout << *ILUK << endl;
}
double Condest;
ILUK->Condest(false, Condest);
if (verbose) cout << "Condition number estimate for this preconditioner = " << Condest << endl;
int Maxiter = 500;
double Tolerance = 1.0E-14;
Epetra_Vector xcomp(map);
Epetra_Vector resid(map);
Epetra_Flops counter;
A.SetFlopCounter(counter);
xcomp.SetFlopCounter(A);
b.SetFlopCounter(A);
resid.SetFlopCounter(A);
ILUK->SetFlopCounter(A);
elapsed_time = timer.ElapsedTime();
BiCGSTAB(A, xcomp, b, ILUK, Maxiter, Tolerance, &residual, verbose);
elapsed_time = timer.ElapsedTime() - elapsed_time;
total_flops = counter.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Time to compute solution = "
<< elapsed_time << endl
<< "Number of operations in solve = " << total_flops << endl
<< "MFLOPS for Solve = " << MFLOPs<< endl << endl;
resid.Update(1.0, xcomp, -1.0, xexact, 0.0); // resid = xcomp - xexact
resid.Norm2(&residual);
if (verbose) cout << "Norm of the difference between exact and computed solutions = " << residual << endl;
if (ILUK!=0) delete ILUK;
if (IlukGraph!=0) delete IlukGraph;
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
return 0 ;
}
int
main (int argc, char *argv[])
{
// These "using" statements make the code a bit more concise.
using std::cout;
using std::endl;
int ierr = 0, i;
// If Trilinos was built with MPI, initialize MPI, otherwise
// initialize the serial "communicator" that stands in for MPI.
#ifdef EPETRA_MPI
MPI_Init (&argc,&argv);
Epetra_MpiComm Comm (MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
const int MyPID = Comm.MyPID();
const int NumProc = Comm.NumProc();
// We only allow (MPI) Process 0 to write to stdout.
const bool verbose = (MyPID == 0);
const int NumGlobalElements = 100;
if (verbose)
cout << Epetra_Version() << endl << endl;
// Asking the Epetra_Comm to print itself is a good test for whether
// you are running in an MPI environment. However, it will print
// something on all MPI processes, so you should remove it for a
// large-scale parallel run.
cout << Comm << endl;
if (NumGlobalElements < NumProc)
{
if (verbose)
cout << "numGlobalBlocks = " << NumGlobalElements
<< " cannot be < number of processors = " << NumProc << endl;
std::exit (EXIT_FAILURE);
}
// Construct a Map that puts approximately the same number of rows
// of the matrix A 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]);
// NumNz[i] is the number of nonzero elements in row i of the sparse
// matrix on this MPI process. Epetra_CrsMatrix uses this to figure
// out how much space to allocate.
std::vector<int> NumNz (NumMyElements);
// We are building a tridiagonal matrix where each row contains the
// nonzero elements (-1 2 -1). Thus, we need 2 off-diagonal terms,
// except for the first and last row of the matrix.
for (int i = 0; i < NumMyElements; ++i)
if (MyGlobalElements[i] == 0 || MyGlobalElements[i] == NumGlobalElements-1)
NumNz[i] = 2; // First or last row
else
NumNz[i] = 3; // Not the (first or last row)
// Create the Epetra_CrsMatrix.
Epetra_CrsMatrix A (Copy, Map, &NumNz[0]);
//
// Add rows to the sparse matrix one at a time.
//
std::vector<double> Values(2);
Values[0] = -1.0; Values[1] = -1.0;
std::vector<int> Indices(2);
const double two = 2.0;
int NumEntries;
for (int i = 0; i < NumMyElements; ++i)
{
if (MyGlobalElements[i] == 0)
{ // The first row of the matrix.
Indices[0] = 1;
NumEntries = 1;
}
else if (MyGlobalElements[i] == NumGlobalElements - 1)
{ // The last row of the matrix.
Indices[0] = NumGlobalElements-2;
NumEntries = 1;
}
else
{ // Any row of the matrix other than the first or last.
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);
// Insert the diagonal entry.
ierr = A.InsertGlobalValues(MyGlobalElements[i], 1, &two, &MyGlobalElements[i]);
assert(ierr==0);
}
// Finish up. We can call FillComplete() with no arguments, because
// the matrix is square.
ierr = A.FillComplete ();
assert (ierr==0);
// Parameters for the power method.
const int niters = NumGlobalElements*10;
const double tolerance = 1.0e-2;
//
// Run the power method. Keep track of the flop count and the total
// elapsed time.
//
Epetra_Flops counter;
A.SetFlopCounter(counter);
Epetra_Time timer(Comm);
double lambda = 0.0;
ierr += powerMethod (lambda, A, niters, tolerance, verbose);
double elapsedTime = timer.ElapsedTime ();
double totalFlops =counter.Flops ();
// Mflop/s: Million floating-point arithmetic operations per second.
double Mflop_per_s = totalFlops / elapsedTime / 1000000.0;
if (verbose)
cout << endl << endl << "Total Mflop/s for first solve = "
<< Mflop_per_s << endl<< endl;
// Increase the first (0,0) diagonal entry of the matrix.
if (verbose)
cout << endl << "Increasing magnitude of first diagonal term, solving again"
<< endl << endl << endl;
if (A.MyGlobalRow (0)) {
int numvals = A.NumGlobalEntries (0);
std::vector<double> Rowvals (numvals);
std::vector<int> Rowinds (numvals);
A.ExtractGlobalRowCopy (0, numvals, numvals, &Rowvals[0], &Rowinds[0]); // Get A(0,0)
for (int i = 0; i < numvals; ++i)
if (Rowinds[i] == 0)
Rowvals[i] *= 10.0;
A.ReplaceGlobalValues (0, numvals, &Rowvals[0], &Rowinds[0]);
}
//
// Run the power method again. Keep track of the flop count and the
// total elapsed time.
//
lambda = 0.0;
timer.ResetStartTime();
counter.ResetFlops();
ierr += powerMethod (lambda, A, niters, tolerance, verbose);
elapsedTime = timer.ElapsedTime();
totalFlops = counter.Flops();
Mflop_per_s = totalFlops / elapsedTime / 1000000.0;
if (verbose)
cout << endl << endl << "Total Mflop/s for second solve = "
<< Mflop_per_s << endl << endl;
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
return ierr;
}
// *************************************************************
// main program - This benchmark code reads a Harwell-Boeing data
// set and finds the minimal eigenvalue of the matrix
// using inverse iteration.
// *************************************************************
int main(int argc, char *argv[]) {
#ifdef EPETRA_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm (MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
cout << Comm << endl;
int MyPID = Comm.MyPID();
bool verbose = false;
if (MyPID==0) verbose = true; // Print out detailed results (turn off for best performance)
if(argc != 2) {
if (verbose) cerr << "Usage: " << argv[0] << " HB_data_file" << endl;
exit(1); // Error
}
// Define pointers that will be set by HB read function
Epetra_Map * readMap;
Epetra_CrsMatrix * readA;
Epetra_Vector * readx;
Epetra_Vector * readb;
Epetra_Vector * readxexact;
// Call function to read in HB problem
Trilinos_Util_ReadHb2Epetra(argv[1], Comm, readMap, readA, readx, readb, readxexact);
// Not interested in x, b or xexact for an eigenvalue problem
delete readx;
delete readb;
delete readxexact;
#ifdef EPETRA_MPI // If running in parallel, we need to distribute matrix across all PEs.
// Create uniform distributed map
Epetra_Map map(readMap->NumGlobalElements(), 0, Comm);
// Create Exporter to distribute read-in matrix and vectors
Epetra_Export exporter(*readMap, map);
Epetra_CrsMatrix A(Copy, map, 0);
A.Export(*readA, exporter, Add);
assert(A.FillComplete()==0);
delete readA;
delete readMap;
#else // If not running in parallel, we do not need to distribute the matrix
Epetra_CrsMatrix & A = *readA;
#endif
// Create flop counter to collect all FLOPS
Epetra_Flops counter;
A.SetFlopCounter(counter);
double lambda = 0; // Minimal eigenvalue returned here
// Call inverse iteration solver
Epetra_Time timer(Comm);
invIteration(A, lambda, verbose);
double elapsedTime = timer.ElapsedTime();
double totalFlops = counter.Flops();
double MFLOPS = totalFlops/elapsedTime/1000000.0;
cout << endl
<< "*************************************************" << endl
<< " Approximate smallest eigenvalue = " << lambda << endl
<< " Total Time = " << elapsedTime << endl
<< " Total FLOPS = " << totalFlops << endl
<< " Total MFLOPS = " << MFLOPS << endl
<< "*************************************************" << endl;
// All done
#ifdef EPETRA_MPI
MPI_Finalize();
#else
delete readA;
delete readMap;
#endif
return (0);
}
//=========================================================================================
void runLUMatrixTests(Epetra_CrsMatrix * L, Epetra_MultiVector * bL, Epetra_MultiVector * btL, Epetra_MultiVector * xexactL,
Epetra_CrsMatrix * U, Epetra_MultiVector * bU, Epetra_MultiVector * btU, Epetra_MultiVector * xexactU,
bool StaticProfile, bool verbose, bool summary) {
if (L->NoDiagonal()) {
bL->Update(1.0, *xexactL, 1.0); // Add contribution of a unit diagonal to bL
btL->Update(1.0, *xexactL, 1.0); // Add contribution of a unit diagonal to btL
}
if (U->NoDiagonal()) {
bU->Update(1.0, *xexactU, 1.0); // Add contribution of a unit diagonal to bU
btU->Update(1.0, *xexactU, 1.0); // Add contribution of a unit diagonal to btU
}
Epetra_MultiVector z(*bL);
Epetra_MultiVector r(*bL);
Epetra_SerialDenseVector resvec(bL->NumVectors());
//Timings
Epetra_Flops flopcounter;
L->SetFlopCounter(flopcounter);
U->SetFlopCounter(flopcounter);
Epetra_Time timer(L->Comm());
std::string statdyn = "dynamic";
if (StaticProfile) statdyn = "static ";
for (int j=0; j<4; j++) { // j = 0/2 is notrans, j = 1/3 is trans
bool TransA = (j==1 || j==3);
std::string contig = "without";
if (j>1) contig = "with ";
if (j==2) {
L->OptimizeStorage();
U->OptimizeStorage();
}
flopcounter.ResetFlops();
timer.ResetStartTime();
//10 lower solves
bool Upper = false;
bool UnitDiagonal = L->NoDiagonal(); // If no diagonal, then unit must be used
Epetra_MultiVector * b = TransA ? btL : bL; // solve with the appropriate b vector
for( int i = 0; i < 10; ++i )
L->Solve(Upper, TransA, UnitDiagonal, *b, z); // Solve Lz = bL or L'z = bLt
double elapsed_time = timer.ElapsedTime();
double total_flops = L->Flops();
// Compute residual
r.Update(-1.0, z, 1.0, *xexactL, 0.0); // r = bt - z
r.Norm2(resvec.Values());
if (resvec.NormInf()>0.000001) {
cout << "resvec = " << resvec << endl;
cout << "z = " << z << endl;
cout << "xexactL = " << *xexactL << endl;
cout << "r = " << r << endl;
}
if (verbose) cout << "ResNorm = " << resvec.NormInf() << ": ";
double MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Total MFLOPs for 10 " << " Lower solves " << statdyn << " Profile (Trans = " << TransA
<< ") and " << contig << " opt storage = " << MFLOPs << " (" << elapsed_time << " s)" <<endl;
if (summary) {
if (L->Comm().NumProc()==1) {
if (TransA) cout << "TransLSv" << statdyn<< "Prof" << contig << "OptStor" << '\t';
else cout << "NoTransLSv" << statdyn << "Prof" << contig << "OptStor" << '\t';
}
cout << MFLOPs << endl;
}
flopcounter.ResetFlops();
timer.ResetStartTime();
//10 upper solves
Upper = true;
UnitDiagonal = U->NoDiagonal(); // If no diagonal, then unit must be used
b = TransA ? btU : bU; // solve with the appropriate b vector
for( int i = 0; i < 10; ++i )
U->Solve(Upper, TransA, UnitDiagonal, *b, z); // Solve Lz = bL or L'z = bLt
elapsed_time = timer.ElapsedTime();
total_flops = U->Flops();
// Compute residual
r.Update(-1.0, z, 1.0, *xexactU, 0.0); // r = bt - z
r.Norm2(resvec.Values());
if (resvec.NormInf()>0.001) {
cout << "U = " << *U << endl;
//cout << "resvec = " << resvec << endl;
cout << "z = " << z << endl;
cout << "xexactU = " << *xexactU << endl;
//cout << "r = " << r << endl;
cout << "b = " << *b << endl;
}
if (verbose) cout << "ResNorm = " << resvec.NormInf() << ": ";
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Total MFLOPs for 10 " << " Upper solves " << statdyn << " Profile (Trans = " << TransA
<< ") and " << contig << " opt storage = " << MFLOPs <<endl;
if (summary) {
if (L->Comm().NumProc()==1) {
if (TransA) cout << "TransUSv" << statdyn<< "Prof" << contig << "OptStor" << '\t';
else cout << "NoTransUSv" << statdyn << "Prof" << contig << "OptStor" << '\t';
}
cout << MFLOPs << endl;
}
}
return;
}
void runMatrixTests(Epetra_CrsMatrix * A, Epetra_MultiVector * b, Epetra_MultiVector * bt,
Epetra_MultiVector * xexact, bool StaticProfile, bool verbose, bool summary) {
Epetra_MultiVector z(*b);
Epetra_MultiVector r(*b);
Epetra_SerialDenseVector resvec(b->NumVectors());
//Timings
Epetra_Flops flopcounter;
A->SetFlopCounter(flopcounter);
Epetra_Time timer(A->Comm());
std::string statdyn = "dynamic";
if (StaticProfile) statdyn = "static ";
for (int j=0; j<4; j++) { // j = 0/2 is notrans, j = 1/3 is trans
bool TransA = (j==1 || j==3);
std::string contig = "without";
if (j>1) contig = "with ";
#ifdef EPETRA_SHORT_PERFTEST
int kstart = 1;
#else
int kstart = 0;
#endif
for (int k=kstart; k<2; k++) { // Loop over old multiply vs. new multiply
std::string oldnew = "old";
if (k>0) oldnew = "new";
if (j==2) A->OptimizeStorage();
flopcounter.ResetFlops();
timer.ResetStartTime();
if (k==0) {
//10 matvecs
#ifndef EPETRA_SHORT_PERFTEST
for( int i = 0; i < 10; ++i )
A->Multiply1(TransA, *xexact, z); // Compute z = A*xexact or z = A'*xexact using old Multiply method
#endif
}
else {
//10 matvecs
for( int i = 0; i < 10; ++i )
A->Multiply(TransA, *xexact, z); // Compute z = A*xexact or z = A'*xexact
}
double elapsed_time = timer.ElapsedTime();
double total_flops = A->Flops();
// Compute residual
if (TransA)
r.Update(-1.0, z, 1.0, *bt, 0.0); // r = bt - z
else
r.Update(-1.0, z, 1.0, *b, 0.0); // r = b - z
r.Norm2(resvec.Values());
if (verbose) cout << "ResNorm = " << resvec.NormInf() << ": ";
double MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Total MFLOPs for 10 " << oldnew << " MatVec's with " << statdyn << " Profile (Trans = " << TransA
<< ") and " << contig << " optimized storage = " << MFLOPs << " (" << elapsed_time << " s)" <<endl;
if (summary) {
if (A->Comm().NumProc()==1) {
if (TransA) cout << "TransMv" << statdyn<< "Prof" << contig << "OptStor" << '\t';
else cout << "NoTransMv" << statdyn << "Prof" << contig << "OptStor" << '\t';
}
cout << MFLOPs << endl;
}
}
}
return;
}
int main(int argc, char *argv[])
{
int ierr = 0;
double elapsed_time;
double total_flops;
double MFLOPs;
#ifdef EPETRA_MPI
// Initialize MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm comm( MPI_COMM_WORLD );
#else
Epetra_SerialComm comm;
#endif
bool verbose = false;
bool summary = false;
// Check if we should print verbose results to standard out
if (argc>6) if (argv[6][0]=='-' && argv[6][1]=='v') verbose = true;
// Check if we should print verbose results to standard out
if (argc>6) if (argv[6][0]=='-' && argv[6][1]=='s') summary = true;
if(argc < 6) {
cerr << "Usage: " << argv[0]
<< " NumNodesX NumNodesY NumProcX NumProcY NumPoints [-v|-s]" << endl
<< "where:" << endl
<< "NumNodesX - Number of mesh nodes in X direction per processor" << endl
<< "NumNodesY - Number of mesh nodes in Y direction per processor" << endl
<< "NumProcX - Number of processors to use in X direction" << endl
<< "NumProcY - Number of processors to use in Y direction" << endl
<< "NumPoints - Number of points to use in stencil (5, 9 or 25 only)" << endl
<< "-v|-s - (Optional) Run in verbose mode if -v present or summary mode if -s present" << endl
<< " NOTES: NumProcX*NumProcY must equal the number of processors used to run the problem." << endl << endl
<< " Serial example:" << endl
<< argv[0] << " 16 12 1 1 25 -v" << endl
<< " Run this program in verbose mode on 1 processor using a 16 X 12 grid with a 25 point stencil."<< endl <<endl
<< " MPI example:" << endl
<< "mpirun -np 32 " << argv[0] << " 10 12 4 8 9 -v" << endl
<< " Run this program in verbose mode on 32 processors putting a 10 X 12 subgrid on each processor using 4 processors "<< endl
<< " in the X direction and 8 in the Y direction. Total grid size is 40 points in X and 96 in Y with a 9 point stencil."<< endl
<< endl;
return(1);
}
//char tmp;
//if (comm.MyPID()==0) cout << "Press any key to continue..."<< endl;
//if (comm.MyPID()==0) cin >> tmp;
//comm.Barrier();
comm.SetTracebackMode(0); // This should shut down any error traceback reporting
if (verbose && comm.MyPID()==0)
cout << Epetra_Version() << endl << endl;
if (summary && comm.MyPID()==0) {
if (comm.NumProc()==1)
cout << Epetra_Version() << endl << endl;
else
cout << endl << endl; // Print two blank line to keep output columns lined up
}
if (verbose) cout << comm <<endl;
// Redefine verbose to only print on PE 0
if (verbose && comm.MyPID()!=0) verbose = false;
if (summary && comm.MyPID()!=0) summary = false;
int numNodesX = atoi(argv[1]);
int numNodesY = atoi(argv[2]);
int numProcsX = atoi(argv[3]);
int numProcsY = atoi(argv[4]);
int numPoints = atoi(argv[5]);
if (verbose || (summary && comm.NumProc()==1)) {
cout << " Number of local nodes in X direction = " << numNodesX << endl
<< " Number of local nodes in Y direction = " << numNodesY << endl
<< " Number of global nodes in X direction = " << numNodesX*numProcsX << endl
<< " Number of global nodes in Y direction = " << numNodesY*numProcsY << endl
<< " Number of local nonzero entries = " << numNodesX*numNodesY*numPoints << endl
<< " Number of global nonzero entries = " << numNodesX*numNodesY*numPoints*numProcsX*numProcsY << endl
<< " Number of Processors in X direction = " << numProcsX << endl
<< " Number of Processors in Y direction = " << numProcsY << endl
<< " Number of Points in stencil = " << numPoints << endl << endl;
}
// Print blank line to keep output columns lined up
if (summary && comm.NumProc()>1)
cout << endl << endl << endl << endl << endl << endl << endl << endl<< endl << endl;
if (numProcsX*numProcsY!=comm.NumProc()) {
cerr << "Number of processors = " << comm.NumProc() << endl
<< " is not the product of " << numProcsX << " and " << numProcsY << endl << endl;
return(1);
}
if (numPoints!=5 && numPoints!=9 && numPoints!=25) {
cerr << "Number of points specified = " << numPoints << endl
<< " is not 5, 9, 25" << endl << endl;
return(1);
}
if (numNodesX*numNodesY<=0) {
cerr << "Product of number of nodes is <= zero" << endl << endl;
return(1);
}
Epetra_IntSerialDenseVector Xoff, XLoff, XUoff;
Epetra_IntSerialDenseVector Yoff, YLoff, YUoff;
if (numPoints==5) {
// Generate a 5-point 2D Finite Difference matrix
Xoff.Size(5);
Yoff.Size(5);
Xoff[0] = -1; Xoff[1] = 1; Xoff[2] = 0; Xoff[3] = 0; Xoff[4] = 0;
Yoff[0] = 0; Yoff[1] = 0; Yoff[2] = 0; Yoff[3] = -1; Yoff[4] = 1;
// Generate a 2-point 2D Lower triangular Finite Difference matrix
XLoff.Size(2);
YLoff.Size(2);
XLoff[0] = -1; XLoff[1] = 0;
YLoff[0] = 0; YLoff[1] = -1;
// Generate a 3-point 2D upper triangular Finite Difference matrix
XUoff.Size(3);
YUoff.Size(3);
XUoff[0] = 0; XUoff[1] = 1; XUoff[2] = 0;
YUoff[0] = 0; YUoff[1] = 0; YUoff[2] = 1;
}
else if (numPoints==9) {
// Generate a 9-point 2D Finite Difference matrix
Xoff.Size(9);
Yoff.Size(9);
Xoff[0] = -1; Xoff[1] = 0; Xoff[2] = 1;
Yoff[0] = -1; Yoff[1] = -1; Yoff[2] = -1;
Xoff[3] = -1; Xoff[4] = 0; Xoff[5] = 1;
Yoff[3] = 0; Yoff[4] = 0; Yoff[5] = 0;
Xoff[6] = -1; Xoff[7] = 0; Xoff[8] = 1;
Yoff[6] = 1; Yoff[7] = 1; Yoff[8] = 1;
// Generate a 5-point lower triangular 2D Finite Difference matrix
XLoff.Size(5);
YLoff.Size(5);
XLoff[0] = -1; XLoff[1] = 0; Xoff[2] = 1;
YLoff[0] = -1; YLoff[1] = -1; Yoff[2] = -1;
XLoff[3] = -1; XLoff[4] = 0;
YLoff[3] = 0; YLoff[4] = 0;
// Generate a 4-point upper triangular 2D Finite Difference matrix
XUoff.Size(4);
YUoff.Size(4);
XUoff[0] = 1;
YUoff[0] = 0;
XUoff[1] = -1; XUoff[2] = 0; XUoff[3] = 1;
YUoff[1] = 1; YUoff[2] = 1; YUoff[3] = 1;
}
else {
// Generate a 25-point 2D Finite Difference matrix
Xoff.Size(25);
Yoff.Size(25);
int xi = 0, yi = 0;
int xo = -2, yo = -2;
Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++;
Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ;
xo = -2, yo++;
Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++;
Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ;
xo = -2, yo++;
Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++;
Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ;
xo = -2, yo++;
Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++;
Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ;
xo = -2, yo++;
Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++; Xoff[xi++] = xo++;
Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ; Yoff[yi++] = yo ;
// Generate a 13-point lower triangular 2D Finite Difference matrix
XLoff.Size(13);
YLoff.Size(13);
xi = 0, yi = 0;
xo = -2, yo = -2;
XLoff[xi++] = xo++; XLoff[xi++] = xo++; XLoff[xi++] = xo++; XLoff[xi++] = xo++; XLoff[xi++] = xo++;
YLoff[yi++] = yo ; YLoff[yi++] = yo ; YLoff[yi++] = yo ; YLoff[yi++] = yo ; YLoff[yi++] = yo ;
xo = -2, yo++;
XLoff[xi++] = xo++; XLoff[xi++] = xo++; XLoff[xi++] = xo++; XLoff[xi++] = xo++; XLoff[xi++] = xo++;
YLoff[yi++] = yo ; YLoff[yi++] = yo ; YLoff[yi++] = yo ; YLoff[yi++] = yo ; YLoff[yi++] = yo ;
xo = -2, yo++;
XLoff[xi++] = xo++; XLoff[xi++] = xo++; XLoff[xi++] = xo++;
YLoff[yi++] = yo ; YLoff[yi++] = yo ; YLoff[yi++] = yo ;
// Generate a 13-point upper triangular 2D Finite Difference matrix
XUoff.Size(13);
YUoff.Size(13);
xi = 0, yi = 0;
xo = 0, yo = 0;
XUoff[xi++] = xo++; XUoff[xi++] = xo++; XUoff[xi++] = xo++;
YUoff[yi++] = yo ; YUoff[yi++] = yo ; YUoff[yi++] = yo ;
xo = -2, yo++;
XUoff[xi++] = xo++; XUoff[xi++] = xo++; XUoff[xi++] = xo++; XUoff[xi++] = xo++; XUoff[xi++] = xo++;
YUoff[yi++] = yo ; YUoff[yi++] = yo ; YUoff[yi++] = yo ; YUoff[yi++] = yo ; YUoff[yi++] = yo ;
xo = -2, yo++;
XUoff[xi++] = xo++; XUoff[xi++] = xo++; XUoff[xi++] = xo++; XUoff[xi++] = xo++; XUoff[xi++] = xo++;
YUoff[yi++] = yo ; YUoff[yi++] = yo ; YUoff[yi++] = yo ; YUoff[yi++] = yo ; YUoff[yi++] = yo ;
}
Epetra_Map * map;
Epetra_Map * mapL;
Epetra_Map * mapU;
Epetra_CrsMatrix * A;
Epetra_CrsMatrix * L;
Epetra_CrsMatrix * U;
Epetra_MultiVector * b;
Epetra_MultiVector * bt;
Epetra_MultiVector * xexact;
Epetra_MultiVector * bL;
Epetra_MultiVector * btL;
Epetra_MultiVector * xexactL;
Epetra_MultiVector * bU;
Epetra_MultiVector * btU;
Epetra_MultiVector * xexactU;
Epetra_SerialDenseVector resvec(0);
//Timings
Epetra_Flops flopcounter;
Epetra_Time timer(comm);
#ifdef EPETRA_VERY_SHORT_PERFTEST
int jstop = 1;
#elif EPETRA_SHORT_PERFTEST
int jstop = 1;
#else
int jstop = 2;
#endif
for (int j=0; j<jstop; j++) {
for (int k=1; k<17; k++) {
#ifdef EPETRA_VERY_SHORT_PERFTEST
if (k<3 || (k%4==0 && k<9)) {
#elif EPETRA_SHORT_PERFTEST
if (k<6 || k%4==0) {
#else
if (k<7 || k%2==0) {
#endif
int nrhs=k;
if (verbose) cout << "\n*************** Results for " << nrhs << " RHS with ";
bool StaticProfile = (j!=0);
if (verbose) {
if (StaticProfile) cout << " static profile\n";
else cout << " dynamic profile\n";
}
GenerateCrsProblem(numNodesX, numNodesY, numProcsX, numProcsY, numPoints,
Xoff.Values(), Yoff.Values(), nrhs, comm, verbose, summary,
map, A, b, bt, xexact, StaticProfile, false);
#ifdef EPETRA_HAVE_JADMATRIX
timer.ResetStartTime();
Epetra_JadMatrix JA(*A);
elapsed_time = timer.ElapsedTime();
if (verbose) cout << "Time to create Jagged diagonal matrix = " << elapsed_time << endl;
//cout << "A = " << *A << endl;
//cout << "JA = " << JA << endl;
runJadMatrixTests(&JA, b, bt, xexact, StaticProfile, verbose, summary);
#endif
runMatrixTests(A, b, bt, xexact, StaticProfile, verbose, summary);
delete A;
delete b;
delete bt;
delete xexact;
GenerateCrsProblem(numNodesX, numNodesY, numProcsX, numProcsY, XLoff.Length(),
XLoff.Values(), YLoff.Values(), nrhs, comm, verbose, summary,
mapL, L, bL, btL, xexactL, StaticProfile, true);
GenerateCrsProblem(numNodesX, numNodesY, numProcsX, numProcsY, XUoff.Length(),
XUoff.Values(), YUoff.Values(), nrhs, comm, verbose, summary,
mapU, U, bU, btU, xexactU, StaticProfile, true);
runLUMatrixTests(L, bL, btL, xexactL, U, bU, btU, xexactU, StaticProfile, verbose, summary);
delete L;
delete bL;
delete btL;
delete xexactL;
delete mapL;
delete U;
delete bU;
delete btU;
delete xexactU;
delete mapU;
Epetra_MultiVector q(*map, nrhs);
Epetra_MultiVector z(q);
Epetra_MultiVector r(q);
delete map;
q.SetFlopCounter(flopcounter);
z.SetFlopCounter(q);
r.SetFlopCounter(q);
resvec.Resize(nrhs);
flopcounter.ResetFlops();
timer.ResetStartTime();
//10 norms
for( int i = 0; i < 10; ++i )
q.Norm2( resvec.Values() );
elapsed_time = timer.ElapsedTime();
total_flops = q.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "\nTotal MFLOPs for 10 Norm2's= " << MFLOPs << endl;
if (summary) {
if (comm.NumProc()==1) cout << "Norm2" << '\t';
cout << MFLOPs << endl;
}
flopcounter.ResetFlops();
timer.ResetStartTime();
//10 dot's
for( int i = 0; i < 10; ++i )
q.Dot(z, resvec.Values());
elapsed_time = timer.ElapsedTime();
total_flops = q.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Total MFLOPs for 10 Dot's = " << MFLOPs << endl;
if (summary) {
if (comm.NumProc()==1) cout << "DotProd" << '\t';
cout << MFLOPs << endl;
}
flopcounter.ResetFlops();
timer.ResetStartTime();
//10 dot's
for( int i = 0; i < 10; ++i )
q.Update(1.0, z, 1.0, r, 0.0);
elapsed_time = timer.ElapsedTime();
total_flops = q.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Total MFLOPs for 10 Updates= " << MFLOPs << endl;
if (summary) {
if (comm.NumProc()==1) cout << "Update" << '\t';
cout << MFLOPs << endl;
}
}
}
}
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
return ierr ;
}
// Constructs a 2D PDE finite difference matrix using the list of x and y offsets.
//
// nx (In) - number of grid points in x direction
// ny (In) - number of grid points in y direction
// The total number of equations will be nx*ny ordered such that the x direction changes
// most rapidly:
// First equation is at point (0,0)
// Second at (1,0)
// ...
// nx equation at (nx-1,0)
// nx+1st equation at (0,1)
// numPoints (In) - number of points in finite difference stencil
// xoff (In) - stencil offsets in x direction (of length numPoints)
// yoff (In) - stencil offsets in y direction (of length numPoints)
// A standard 5-point finite difference stencil would be described as:
// numPoints = 5
// xoff = [-1, 1, 0, 0, 0]
// yoff = [ 0, 0, 0, -1, 1]
// nrhs - Number of rhs to generate. (First interface produces vectors, so nrhs is not needed
// comm (In) - an Epetra_Comm object describing the parallel machine (numProcs and my proc ID)
// map (Out) - Epetra_Map describing distribution of matrix and vectors/multivectors
// A (Out) - Epetra_CrsMatrix constructed for nx by ny grid using prescribed stencil
// Off-diagonal values are random between 0 and 1. If diagonal is part of stencil,
// diagonal will be slightly diag dominant.
// b (Out) - Generated RHS. Values satisfy b = A*xexact
// bt (Out) - Generated RHS. Values satisfy b = A'*xexact
// xexact (Out) - Generated exact solution to Ax = b and b' = A'xexact
// Note: Caller of this function is responsible for deleting all output objects.
void GenerateCrsProblem(int numNodesX, int numNodesY, int numProcsX, int numProcsY, int numPoints,
int * xoff, int * yoff,
const Epetra_Comm &comm, bool verbose, bool summary,
Epetra_Map *& map,
Epetra_CrsMatrix *& A,
Epetra_Vector *& b,
Epetra_Vector *& bt,
Epetra_Vector *&xexact, bool StaticProfile, bool MakeLocalOnly) {
Epetra_MultiVector * b1, * bt1, * xexact1;
GenerateCrsProblem(numNodesX, numNodesY, numProcsX, numProcsY, numPoints,
xoff, yoff, 1, comm, verbose, summary,
map, A, b1, bt1, xexact1, StaticProfile, MakeLocalOnly);
b = dynamic_cast<Epetra_Vector *>(b1);
bt = dynamic_cast<Epetra_Vector *>(bt1);
xexact = dynamic_cast<Epetra_Vector *>(xexact1);
return;
}
void GenerateCrsProblem(int numNodesX, int numNodesY, int numProcsX, int numProcsY, int numPoints,
int * xoff, int * yoff, int nrhs,
const Epetra_Comm &comm, bool verbose, bool summary,
Epetra_Map *& map,
Epetra_CrsMatrix *& A,
Epetra_MultiVector *& b,
Epetra_MultiVector *& bt,
Epetra_MultiVector *&xexact, bool StaticProfile, bool MakeLocalOnly) {
Epetra_Time timer(comm);
// Determine my global IDs
long long * myGlobalElements;
GenerateMyGlobalElements(numNodesX, numNodesY, numProcsX, numProcsY, comm.MyPID(), myGlobalElements);
int numMyEquations = numNodesX*numNodesY;
map = new Epetra_Map((long long)-1, numMyEquations, myGlobalElements, 0, comm); // Create map with 2D block partitioning.
delete [] myGlobalElements;
long long numGlobalEquations = map->NumGlobalElements64();
int profile = 0; if (StaticProfile) profile = numPoints;
#ifdef EPETRA_HAVE_STATICPROFILE
if (MakeLocalOnly)
A = new Epetra_CrsMatrix(Copy, *map, *map, profile, StaticProfile); // Construct matrix with rowmap=colmap
else
A = new Epetra_CrsMatrix(Copy, *map, profile, StaticProfile); // Construct matrix
#else
if (MakeLocalOnly)
A = new Epetra_CrsMatrix(Copy, *map, *map, profile); // Construct matrix with rowmap=colmap
else
A = new Epetra_CrsMatrix(Copy, *map, profile); // Construct matrix
#endif
long long * indices = new long long[numPoints];
double * values = new double[numPoints];
double dnumPoints = (double) numPoints;
int nx = numNodesX*numProcsX;
for (int i=0; i<numMyEquations; i++) {
long long rowID = map->GID64(i);
int numIndices = 0;
for (int j=0; j<numPoints; j++) {
long long colID = rowID + xoff[j] + nx*yoff[j]; // Compute column ID based on stencil offsets
if (colID>-1 && colID<numGlobalEquations) {
indices[numIndices] = colID;
double value = - ((double) rand())/ ((double) RAND_MAX);
if (colID==rowID)
values[numIndices++] = dnumPoints - value; // Make diagonal dominant
else
values[numIndices++] = value;
}
}
//cout << "Building row " << rowID << endl;
A->InsertGlobalValues(rowID, numIndices, values, indices);
}
delete [] indices;
delete [] values;
double insertTime = timer.ElapsedTime();
timer.ResetStartTime();
A->FillComplete(false);
double fillCompleteTime = timer.ElapsedTime();
if (verbose)
cout << "Time to insert matrix values = " << insertTime << endl
<< "Time to complete fill = " << fillCompleteTime << endl;
if (summary) {
if (comm.NumProc()==1) cout << "InsertTime" << '\t';
cout << insertTime << endl;
if (comm.NumProc()==1) cout << "FillCompleteTime" << '\t';
cout << fillCompleteTime << endl;
}
if (nrhs<=1) {
b = new Epetra_Vector(*map);
bt = new Epetra_Vector(*map);
xexact = new Epetra_Vector(*map);
}
else {
b = new Epetra_MultiVector(*map, nrhs);
bt = new Epetra_MultiVector(*map, nrhs);
xexact = new Epetra_MultiVector(*map, nrhs);
}
xexact->Random(); // Fill xexact with random values
A->Multiply(false, *xexact, *b);
A->Multiply(true, *xexact, *bt);
return;
}
int main(int argc, char *argv[])
{
int ierr = 0, i, j, k;
#ifdef EPETRA_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm( MPI_COMM_WORLD );
#else
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;
if(verbose && Comm.MyPID()==0)
std::cout << Epetra_Version() << std::endl << std::endl;
int rank = Comm.MyPID();
// char tmp;
// if (rank==0) std::cout << "Press any key to continue..."<< std::endl;
// if (rank==0) cin >> tmp;
// Comm.Barrier();
Comm.SetTracebackMode(0); // This should shut down any error traceback reporting
if (verbose) std::cout << Comm << std::endl;
// bool verbose1 = verbose;
// Redefine verbose to only print on PE 0
if (verbose && rank!=0) verbose = false;
int N = 20;
int NRHS = 4;
double * A = new double[N*N];
double * A1 = new double[N*N];
double * X = new double[(N+1)*NRHS];
double * X1 = new double[(N+1)*NRHS];
int LDX = N+1;
int LDX1 = N+1;
double * B = new double[N*NRHS];
double * B1 = new double[N*NRHS];
int LDB = N;
int LDB1 = N;
int LDA = N;
int LDA1 = LDA;
double OneNorm1;
bool Upper = false;
Epetra_SerialSpdDenseSolver solver;
Epetra_SerialSymDenseMatrix * Matrix;
for (int kk=0; kk<2; kk++) {
for (i=1; i<=N; i++) {
GenerateHilbert(A, LDA, i);
OneNorm1 = 0.0;
for (j=1; j<=i; j++) OneNorm1 += 1.0/((double) j); // 1-Norm = 1 + 1/2 + ...+1/n
if (kk==0) {
Matrix = new Epetra_SerialSymDenseMatrix(View, A, LDA, i);
LDA1 = LDA;
}
else {
Matrix = new Epetra_SerialSymDenseMatrix(Copy, A, LDA, i);
LDA1 = i;
}
GenerateHilbert(A1, LDA1, i);
if (kk==1) {
solver.FactorWithEquilibration(true);
Matrix->SetUpper();
Upper = true;
solver.SolveToRefinedSolution(false);
}
for (k=0; k<NRHS; k++)
for (j=0; j<i; j++) {
B[j+k*LDB] = 1.0/((double) (k+3)*(j+3));
B1[j+k*LDB1] = B[j+k*LDB1];
}
Epetra_SerialDenseMatrix Epetra_B(View, B, LDB, i, NRHS);
Epetra_SerialDenseMatrix Epetra_X(View, X, LDX, i, NRHS);
solver.SetMatrix(*Matrix);
solver.SetVectors(Epetra_X, Epetra_B);
ierr = check(solver, A1, LDA1, i, NRHS, OneNorm1, B1, LDB1, X1, LDX1, Upper, verbose);
assert (ierr>-1);
delete Matrix;
if (ierr!=0) {
if (verbose) std::cout << "Factorization failed due to bad conditioning. This is normal if SCOND is small."
<< std::endl;
break;
}
}
}
delete [] A;
delete [] A1;
delete [] X;
delete [] X1;
delete [] B;
delete [] B1;
/////////////////////////////////////////////////////////////////////
// Now test norms and scaling functions
/////////////////////////////////////////////////////////////////////
Epetra_SerialSymDenseMatrix D;
double ScalarA = 2.0;
int DM = 10;
int DN = 10;
D.Shape(DM);
for (j=0; j<DN; j++)
for (i=0; i<DM; i++) D[j][i] = (double) (1+i+j*DM) ;
//std::cout << D << std::endl;
double NormInfD_ref = (double)(DM*(DN*(DN+1))/2);
double NormOneD_ref = NormInfD_ref;
double NormInfD = D.NormInf();
double NormOneD = D.NormOne();
if (verbose) {
std::cout << " *** Before scaling *** " << std::endl
<< " Computed one-norm of test matrix = " << NormOneD << std::endl
<< " Expected one-norm = " << NormOneD_ref << std::endl
<< " Computed inf-norm of test matrix = " << NormInfD << std::endl
<< " Expected inf-norm = " << NormInfD_ref << std::endl;
}
D.Scale(ScalarA); // Scale entire D matrix by this value
//std::cout << D << std::endl;
NormInfD = D.NormInf();
NormOneD = D.NormOne();
if (verbose) {
std::cout << " *** After scaling *** " << std::endl
<< " Computed one-norm of test matrix = " << NormOneD << std::endl
<< " Expected one-norm = " << NormOneD_ref*ScalarA << std::endl
<< " Computed inf-norm of test matrix = " << NormInfD << std::endl
<< " Expected inf-norm = " << NormInfD_ref*ScalarA << std::endl;
}
/////////////////////////////////////////////////////////////////////
// Now test for larger system, both correctness and performance.
/////////////////////////////////////////////////////////////////////
N = 2000;
NRHS = 5;
LDA = N;
LDB = N;
LDX = N;
if (verbose) std::cout << "\n\nComputing factor of an " << N << " x " << N << " SPD matrix...Please wait.\n\n" << std::endl;
// Define A and X
A = new double[LDA*N];
X = new double[LDB*NRHS];
for (j=0; j<N; j++) {
for (k=0; k<NRHS; k++) X[j+k*LDX] = 1.0/((double) (j+5+k));
for (i=0; i<N; i++) {
if (i==j) A[i+j*LDA] = 100.0 + i;
else A[i+j*LDA] = -1.0/((double) (i+10)*(j+10));
}
}
// Define Epetra_SerialDenseMatrix object
Epetra_SerialSymDenseMatrix BigMatrix(Copy, A, LDA, N);
Epetra_SerialSymDenseMatrix OrigBigMatrix(View, A, LDA, N);
Epetra_SerialSpdDenseSolver BigSolver;
BigSolver.FactorWithEquilibration(true);
BigSolver.SetMatrix(BigMatrix);
// Time factorization
Epetra_Flops counter;
BigSolver.SetFlopCounter(counter);
Epetra_Time Timer(Comm);
double tstart = Timer.ElapsedTime();
ierr = BigSolver.Factor();
if (ierr!=0 && verbose) std::cout << "Error in factorization = "<<ierr<< std::endl;
assert(ierr==0);
double time = Timer.ElapsedTime() - tstart;
double FLOPS = counter.Flops();
double MFLOPS = FLOPS/time/1000000.0;
if (verbose) std::cout << "MFLOPS for Factorization = " << MFLOPS << std::endl;
// Define Left hand side and right hand side
Epetra_SerialDenseMatrix LHS(View, X, LDX, N, NRHS);
Epetra_SerialDenseMatrix RHS;
RHS.Shape(N,NRHS); // Allocate RHS
// Compute RHS from A and X
Epetra_Flops RHS_counter;
RHS.SetFlopCounter(RHS_counter);
tstart = Timer.ElapsedTime();
RHS.Multiply('L', 1.0, OrigBigMatrix, LHS, 0.0); // Symmetric Matrix-multiply
time = Timer.ElapsedTime() - tstart;
Epetra_SerialDenseMatrix OrigRHS = RHS;
FLOPS = RHS_counter.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) std::cout << "MFLOPS to build RHS (NRHS = " << NRHS <<") = " << MFLOPS << std::endl;
// Set LHS and RHS and solve
BigSolver.SetVectors(LHS, RHS);
tstart = Timer.ElapsedTime();
ierr = BigSolver.Solve();
if (ierr==1 && verbose) std::cout << "LAPACK guidelines suggest this matrix might benefit from equilibration." << std::endl;
else if (ierr!=0 && verbose) std::cout << "Error in solve = "<<ierr<< std::endl;
assert(ierr>=0);
time = Timer.ElapsedTime() - tstart;
FLOPS = BigSolver.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) std::cout << "MFLOPS for Solve (NRHS = " << NRHS <<") = " << MFLOPS << std::endl;
double * resid = new double[NRHS];
bool OK = Residual(N, NRHS, A, LDA, BigSolver.X(), BigSolver.LDX(),
OrigRHS.A(), OrigRHS.LDA(), resid);
if (verbose) {
if (!OK) std::cout << "************* Residual do not meet tolerance *************" << std::endl;
for (i=0; i<NRHS; i++)
std::cout << "Residual[" << i <<"] = "<< resid[i] << std::endl;
std::cout << std::endl;
}
// Solve again using the Epetra_SerialDenseVector class for LHS and RHS
Epetra_SerialDenseVector X2;
Epetra_SerialDenseVector B2;
X2.Size(BigMatrix.N());
B2.Size(BigMatrix.M());
int length = BigMatrix.N();
{for (int kk=0; kk<length; kk++) X2[kk] = ((double ) kk)/ ((double) length);} // Define entries of X2
RHS_counter.ResetFlops();
B2.SetFlopCounter(RHS_counter);
tstart = Timer.ElapsedTime();
B2.Multiply('N', 'N', 1.0, OrigBigMatrix, X2, 0.0); // Define B2 = A*X2
time = Timer.ElapsedTime() - tstart;
Epetra_SerialDenseVector OrigB2 = B2;
FLOPS = RHS_counter.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) std::cout << "MFLOPS to build single RHS = " << MFLOPS << std::endl;
// Set LHS and RHS and solve
BigSolver.SetVectors(X2, B2);
tstart = Timer.ElapsedTime();
ierr = BigSolver.Solve();
time = Timer.ElapsedTime() - tstart;
if (ierr==1 && verbose) std::cout << "LAPACK guidelines suggest this matrix might benefit from equilibration." << std::endl;
else if (ierr!=0 && verbose) std::cout << "Error in solve = "<<ierr<< std::endl;
assert(ierr>=0);
FLOPS = counter.Flops();
MFLOPS = FLOPS/time/1000000.0;
if (verbose) std::cout << "MFLOPS to solve single RHS = " << MFLOPS << std::endl;
OK = Residual(N, 1, A, LDA, BigSolver.X(), BigSolver.LDX(), OrigB2.A(),
OrigB2.LDA(), resid);
if (verbose) {
if (!OK) std::cout << "************* Residual do not meet tolerance *************" << std::endl;
std::cout << "Residual = "<< resid[0] << std::endl;
}
delete [] resid;
delete [] A;
delete [] X;
///////////////////////////////////////////////////
// Now test default constructor and index operators
///////////////////////////////////////////////////
N = 5;
Epetra_SerialSymDenseMatrix C; // Implicit call to default constructor, should not need to call destructor
C.Shape(5); // Make it 5 by 5
double * C1 = new double[N*N];
GenerateHilbert(C1, N, N); // Generate Hilber matrix
C1[1+2*N] = 1000.0; // Make matrix nonsymmetric
// Fill values of C with Hilbert values
for (i=0; i<N; i++)
for (j=0; j<N; j++)
C(i,j) = C1[i+j*N];
// Test if values are correctly written and read
for (i=0; i<N; i++)
for (j=0; j<N; j++) {
assert(C(i,j) == C1[i+j*N]);
assert(C(i,j) == C[j][i]);
}
if (verbose)
std::cout << "Default constructor and index operator check OK. Values of Hilbert matrix = "
<< std::endl << C << std::endl
<< "Values should be 1/(i+j+1), except value (1,2) should be 1000" << std::endl;
delete [] C1;
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
/* end main
*/
return ierr ;
}
int main(int argc, char *argv[])
{
int ierr = 0, forierr = 0;
bool debug = false;
#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..."<< std::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() << std::endl << std::endl;
if (verbose) cout << "Processor "<<MyPID<<" of "<< NumProc
<< " is alive."<<endl;
bool verbose1 = verbose;
// 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
int* MyGlobalElements = new int[Map.NumMyElements()];
Map.MyGlobalElements(MyGlobalElements);
// 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
int* NumNz = new int[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 (int 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);
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
double* Values = new double[2];
Values[0] = -1.0;
Values[1] = -1.0;
int* Indices = new int[2];
double two = 2.0;
int NumEntries;
forierr = 0;
for (int 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, Indices)==0);
forierr += !(A.InsertGlobalValues(MyGlobalElements[i], 1, &two, MyGlobalElements+i)>0); // Put in the diagonal entry
}
EPETRA_TEST_ERR(forierr,ierr);
int * indexOffsetTmp;
int * indicesTmp;
double * valuesTmp;
// Finish up
EPETRA_TEST_ERR(!(A.IndicesAreGlobal()),ierr);
EPETRA_TEST_ERR(!(A.ExtractCrsDataPointers(indexOffsetTmp, indicesTmp, valuesTmp)==-1),ierr); // Should fail
EPETRA_TEST_ERR(!(A.FillComplete(false)==0),ierr);
EPETRA_TEST_ERR(!(A.ExtractCrsDataPointers(indexOffsetTmp, indicesTmp, valuesTmp)==-1),ierr); // Should fail
EPETRA_TEST_ERR(!(A.IndicesAreLocal()),ierr);
EPETRA_TEST_ERR(A.StorageOptimized(),ierr);
A.OptimizeStorage();
EPETRA_TEST_ERR(!(A.StorageOptimized()),ierr);
EPETRA_TEST_ERR(!(A.ExtractCrsDataPointers(indexOffsetTmp, indicesTmp, valuesTmp)==0),ierr); // Should succeed
const Epetra_CrsGraph & GofA = A.Graph();
EPETRA_TEST_ERR((indicesTmp!=GofA[0] || valuesTmp!=A[0]),ierr); // Extra check to see if operator[] is consistent
EPETRA_TEST_ERR(A.UpperTriangular(),ierr);
EPETRA_TEST_ERR(A.LowerTriangular(),ierr);
int NumMyNonzeros = 3 * NumMyEquations;
if(A.LRID(0) >= 0)
NumMyNonzeros--; // If I own first global row, then there is one less nonzero
if(A.LRID(NumGlobalEquations-1) >= 0)
NumMyNonzeros--; // If I own last global row, then there is one less nonzero
EPETRA_TEST_ERR(check(A, NumMyEquations, NumGlobalEquations, NumMyNonzeros, 3*NumGlobalEquations-2,
MyGlobalElements, verbose),ierr);
forierr = 0;
for (int i = 0; i < NumMyEquations; i++)
forierr += !(A.NumGlobalEntries(MyGlobalElements[i])==NumNz[i]+1);
EPETRA_TEST_ERR(forierr,ierr);
forierr = 0;
for (int i = 0; i < NumMyEquations; i++)
forierr += !(A.NumMyEntries(i)==NumNz[i]+1);
EPETRA_TEST_ERR(forierr,ierr);
if (verbose) cout << "\n\nNumEntries function check OK" << std::endl<< std::endl;
EPETRA_TEST_ERR(check_graph_sharing(Comm),ierr);
// Create vectors for Power method
Epetra_Vector q(Map);
Epetra_Vector z(Map);
Epetra_Vector resid(Map);
// variable needed for iteration
double lambda = 0.0;
// int niters = 10000;
int niters = 200;
double tolerance = 1.0e-1;
/////////////////////////////////////////////////////////////////////////////////////////////////
// Iterate
Epetra_Flops flopcounter;
A.SetFlopCounter(flopcounter);
q.SetFlopCounter(A);
z.SetFlopCounter(A);
resid.SetFlopCounter(A);
Epetra_Time timer(Comm);
EPETRA_TEST_ERR(power_method(false, A, q, z, resid, &lambda, niters, tolerance, verbose),ierr);
double elapsed_time = timer.ElapsedTime();
double total_flops = A.Flops() + q.Flops() + z.Flops() + resid.Flops();
double MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "\n\nTotal MFLOPs for first solve = " << MFLOPs << std::endl<< std::endl;
/////////////////////////////////////////////////////////////////////////////////////////////////
// Solve transpose problem
if (verbose) cout << "\n\nUsing transpose of matrix and solving again (should give same result).\n\n"
<< std::endl;
// Iterate
lambda = 0.0;
flopcounter.ResetFlops();
timer.ResetStartTime();
EPETRA_TEST_ERR(power_method(true, A, q, z, resid, &lambda, niters, tolerance, verbose),ierr);
elapsed_time = timer.ElapsedTime();
total_flops = A.Flops() + q.Flops() + z.Flops() + resid.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "\n\nTotal MFLOPs for transpose solve = " << MFLOPs << std::endl<< endl;
/////////////////////////////////////////////////////////////////////////////////////////////////
// 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);
double * Rowvals = new double [numvals];
int * Rowinds = new int [numvals];
A.ExtractGlobalRowCopy(0, numvals, numvals, Rowvals, Rowinds); // Get A[0,0]
for (int i=0; i<numvals; i++) if (Rowinds[i] == 0) Rowvals[i] *= 10.0;
A.ReplaceGlobalValues(0, numvals, Rowvals, Rowinds);
delete [] Rowvals;
delete [] Rowinds;
}
// Iterate (again)
lambda = 0.0;
flopcounter.ResetFlops();
timer.ResetStartTime();
EPETRA_TEST_ERR(power_method(false, A, q, z, resid, &lambda, niters, tolerance, verbose),ierr);
elapsed_time = timer.ElapsedTime();
total_flops = A.Flops() + q.Flops() + z.Flops() + resid.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "\n\nTotal MFLOPs for second solve = " << MFLOPs << endl<< endl;
/////////////////////////////////////////////////////////////////////////////////////////////////
// Solve transpose problem
if (verbose) cout << "\n\nUsing transpose of matrix and solving again (should give same result).\n\n"
<< endl;
// Iterate (again)
lambda = 0.0;
flopcounter.ResetFlops();
timer.ResetStartTime();
EPETRA_TEST_ERR(power_method(true, A, q, z, resid, &lambda, niters, tolerance, verbose),ierr);
elapsed_time = timer.ElapsedTime();
total_flops = A.Flops() + q.Flops() + z.Flops() + resid.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "\n\nTotal MFLOPs for tranpose of second solve = " << MFLOPs << endl<< endl;
if (verbose) cout << "\n\n*****Testing constant entry constructor" << endl<< endl;
Epetra_CrsMatrix AA(Copy, Map, 5);
if (debug) Comm.Barrier();
double dble_one = 1.0;
for (int i=0; i< NumMyEquations; i++) AA.InsertGlobalValues(MyGlobalElements[i], 1, &dble_one, MyGlobalElements+i);
// Note: All processors will call the following Insert routines, but only the processor
// that owns it will actually do anything
int One = 1;
if (AA.MyGlobalRow(0)) {
EPETRA_TEST_ERR(!(AA.InsertGlobalValues(0, 0, &dble_one, &One)==0),ierr);
}
else EPETRA_TEST_ERR(!(AA.InsertGlobalValues(0, 1, &dble_one, &One)==-1),ierr);
EPETRA_TEST_ERR(!(AA.FillComplete(false)==0),ierr);
EPETRA_TEST_ERR(AA.StorageOptimized(),ierr);
EPETRA_TEST_ERR(!(AA.UpperTriangular()),ierr);
EPETRA_TEST_ERR(!(AA.LowerTriangular()),ierr);
if (debug) Comm.Barrier();
EPETRA_TEST_ERR(check(AA, NumMyEquations, NumGlobalEquations, NumMyEquations, NumGlobalEquations,
MyGlobalElements, verbose),ierr);
if (debug) Comm.Barrier();
forierr = 0;
for (int i=0; i<NumMyEquations; i++) forierr += !(AA.NumGlobalEntries(MyGlobalElements[i])==1);
EPETRA_TEST_ERR(forierr,ierr);
if (verbose) cout << "\n\nNumEntries function check OK" << endl<< endl;
if (debug) Comm.Barrier();
if (verbose) cout << "\n\n*****Testing copy constructor" << endl<< endl;
Epetra_CrsMatrix B(AA);
EPETRA_TEST_ERR(check(B, NumMyEquations, NumGlobalEquations, NumMyEquations, NumGlobalEquations,
MyGlobalElements, verbose),ierr);
forierr = 0;
for (int i=0; i<NumMyEquations; i++) forierr += !(B.NumGlobalEntries(MyGlobalElements[i])==1);
EPETRA_TEST_ERR(forierr,ierr);
if (verbose) cout << "\n\nNumEntries function check OK" << endl<< endl;
if (debug) Comm.Barrier();
if (verbose) cout << "\n\n*****Testing local view constructor" << endl<< endl;
Epetra_CrsMatrix BV(View, AA.RowMap(), AA.ColMap(), 0);
forierr = 0;
int* Inds;
double* Vals;
for (int i = 0; i < NumMyEquations; i++) {
forierr += !(AA.ExtractMyRowView(i, NumEntries, Vals, Inds)==0);
forierr += !(BV.InsertMyValues(i, NumEntries, Vals, Inds)==0);
}
BV.FillComplete(false);
EPETRA_TEST_ERR(check(BV, NumMyEquations, NumGlobalEquations, NumMyEquations, NumGlobalEquations,
MyGlobalElements, verbose),ierr);
forierr = 0;
for (int i=0; i<NumMyEquations; i++) forierr += !(BV.NumGlobalEntries(MyGlobalElements[i])==1);
EPETRA_TEST_ERR(forierr,ierr);
if (verbose) cout << "\n\nNumEntries function check OK" << endl<< endl;
if (debug) Comm.Barrier();
if (verbose) cout << "\n\n*****Testing post construction modifications" << endl<< endl;
EPETRA_TEST_ERR(!(B.InsertGlobalValues(0, 1, &dble_one, &One)==-2),ierr);
// Release all objects
delete [] NumNz;
delete [] Values;
delete [] Indices;
delete [] MyGlobalElements;
if (verbose1) {
// Test ostream << operator (if verbose1)
// Construct a Map that puts 2 equations on each PE
int NumMyElements1 = 2;
int NumMyEquations1 = NumMyElements1;
int NumGlobalEquations1 = NumMyEquations1*NumProc;
Epetra_Map Map1(-1, NumMyElements1, 0, Comm);
// Get update list and number of local equations from newly created Map
int * MyGlobalElements1 = new int[Map1.NumMyElements()];
Map1.MyGlobalElements(MyGlobalElements1);
// 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
int * NumNz1 = new int[NumMyEquations1];
// 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 (int i=0; i<NumMyEquations1; i++)
if (MyGlobalElements1[i]==0 || MyGlobalElements1[i] == NumGlobalEquations1-1)
NumNz1[i] = 1;
else
NumNz1[i] = 2;
// Create a Epetra_Matrix
Epetra_CrsMatrix A1(Copy, Map1, NumNz1);
// Add rows one-at-a-time
// Need some vectors to help
// Off diagonal Values will always be -1
double *Values1 = new double[2];
Values1[0] = -1.0; Values1[1] = -1.0;
int *Indices1 = new int[2];
double two1 = 2.0;
int NumEntries1;
forierr = 0;
for (int i=0; i<NumMyEquations1; i++)
{
if (MyGlobalElements1[i]==0)
{
Indices1[0] = 1;
NumEntries1 = 1;
}
else if (MyGlobalElements1[i] == NumGlobalEquations1-1)
{
Indices1[0] = NumGlobalEquations1-2;
NumEntries1 = 1;
}
else
{
Indices1[0] = MyGlobalElements1[i]-1;
Indices1[1] = MyGlobalElements1[i]+1;
NumEntries1 = 2;
}
forierr += !(A1.InsertGlobalValues(MyGlobalElements1[i], NumEntries1, Values1, Indices1)==0);
forierr += !(A1.InsertGlobalValues(MyGlobalElements1[i], 1, &two1, MyGlobalElements1+i)>0); // Put in the diagonal entry
}
EPETRA_TEST_ERR(forierr,ierr);
delete [] Indices1;
delete [] Values1;
// Finish up
EPETRA_TEST_ERR(!(A1.FillComplete(false)==0),ierr);
// Test diagonal extraction function
Epetra_Vector checkDiag(Map1);
EPETRA_TEST_ERR(!(A1.ExtractDiagonalCopy(checkDiag)==0),ierr);
forierr = 0;
for (int i=0; i<NumMyEquations1; i++) forierr += !(checkDiag[i]==two1);
EPETRA_TEST_ERR(forierr,ierr);
// Test diagonal replacement method
forierr = 0;
for (int i=0; i<NumMyEquations1; i++) checkDiag[i]=two1*two1;
EPETRA_TEST_ERR(forierr,ierr);
EPETRA_TEST_ERR(!(A1.ReplaceDiagonalValues(checkDiag)==0),ierr);
Epetra_Vector checkDiag1(Map1);
EPETRA_TEST_ERR(!(A1.ExtractDiagonalCopy(checkDiag1)==0),ierr);
forierr = 0;
for (int i=0; i<NumMyEquations1; i++) forierr += !(checkDiag[i]==checkDiag1[i]);
EPETRA_TEST_ERR(forierr,ierr);
if (verbose) cout << "\n\nDiagonal extraction and replacement OK.\n\n" << endl;
double orignorm = A1.NormOne();
EPETRA_TEST_ERR(!(A1.Scale(4.0)==0),ierr);
EPETRA_TEST_ERR(!(A1.NormOne()!=orignorm),ierr);
if (verbose) cout << "\n\nMatrix scale OK.\n\n" << endl;
if (verbose) cout << "\n\nPrint out tridiagonal matrix, each part on each processor.\n\n" << endl;
cout << A1 << endl;
// Release all objects
delete [] NumNz1;
delete [] MyGlobalElements1;
}
if (verbose) cout << "\n\n*****Testing LeftScale and RightScale" << endl << endl;
int NumMyElements2 = 7;
int NumMyRows2 = 1;//This value should not be changed without editing the
// code below.
Epetra_Map RowMap(-1,NumMyRows2,0,Comm);
Epetra_Map ColMap(NumMyElements2,NumMyElements2,0,Comm);
// The DomainMap needs to be different from the ColMap for the test to
// be meaningful.
Epetra_Map DomainMap(NumMyElements2,0,Comm);
int NumMyRangeElements2 = 0;
// We need to distribute the elements differently for the range map also.
if (MyPID % 2 == 0)
NumMyRangeElements2 = NumMyRows2*2; //put elements on even number procs
if (NumProc % 2 == 1 && MyPID == NumProc-1)
NumMyRangeElements2 = NumMyRows2; //If number of procs is odd, put
// the last NumMyElements2 elements on the last proc
Epetra_Map RangeMap(-1,NumMyRangeElements2,0,Comm);
Epetra_CrsMatrix A2(Copy,RowMap,ColMap,NumMyElements2);
double * Values2 = new double[NumMyElements2];
int * Indices2 = new int[NumMyElements2];
for (int i=0; i<NumMyElements2; i++) {
Values2[i] = i+MyPID;
Indices2[i]=i;
}
A2.InsertMyValues(0,NumMyElements2,Values2,Indices2);
A2.FillComplete(DomainMap,RangeMap,false);
Epetra_CrsMatrix A2copy(A2);
double * RowLeftScaleValues = new double[NumMyRows2];
double * ColRightScaleValues = new double[NumMyElements2];
int RowLoopLength = RowMap.MaxMyGID()-RowMap.MinMyGID()+1;
for (int i=0; i<RowLoopLength; i++)
RowLeftScaleValues[i] = (i + RowMap.MinMyGID() ) % 2 + 1;
// For the column map, all procs own all elements
for (int i=0; i<NumMyElements2;i++)
ColRightScaleValues[i] = i % 2 + 1;
int RangeLoopLength = RangeMap.MaxMyGID()-RangeMap.MinMyGID()+1;
double * RangeLeftScaleValues = new double[RangeLoopLength];
int DomainLoopLength = DomainMap.MaxMyGID()-DomainMap.MinMyGID()+1;
double * DomainRightScaleValues = new double[DomainLoopLength];
for (int i=0; i<RangeLoopLength; i++)
RangeLeftScaleValues[i] = 1.0/((i + RangeMap.MinMyGID() ) % 2 + 1);
for (int i=0; i<DomainLoopLength;i++)
DomainRightScaleValues[i] = 1.0/((i + DomainMap.MinMyGID() ) % 2 + 1);
Epetra_Vector xRow(View,RowMap,RowLeftScaleValues);
Epetra_Vector xCol(View,ColMap,ColRightScaleValues);
Epetra_Vector xRange(View,RangeMap,RangeLeftScaleValues);
Epetra_Vector xDomain(View,DomainMap,DomainRightScaleValues);
double A2infNorm = A2.NormInf();
double A2oneNorm = A2.NormOne();
if (verbose1) cout << A2;
EPETRA_TEST_ERR(A2.LeftScale(xRow),ierr);
double A2infNorm1 = A2.NormInf();
double A2oneNorm1 = A2.NormOne();
bool ScalingBroke = false;
if (A2infNorm1>2*A2infNorm||A2infNorm1<A2infNorm) {
EPETRA_TEST_ERR(-31,ierr);
ScalingBroke = true;
}
if (A2oneNorm1>2*A2oneNorm||A2oneNorm1<A2oneNorm) {
EPETRA_TEST_ERR(-32,ierr);
ScalingBroke = true;
}
if (verbose1) cout << A2;
EPETRA_TEST_ERR(A2.RightScale(xCol),ierr);
double A2infNorm2 = A2.NormInf();
double A2oneNorm2 = A2.NormOne();
if (A2infNorm2>=2*A2infNorm1||A2infNorm2<=A2infNorm1) {
EPETRA_TEST_ERR(-33,ierr);
ScalingBroke = true;
}
if (A2oneNorm2>2*A2oneNorm1||A2oneNorm2<=A2oneNorm1) {
EPETRA_TEST_ERR(-34,ierr);
ScalingBroke = true;
}
if (verbose1) cout << A2;
EPETRA_TEST_ERR(A2.RightScale(xDomain),ierr);
double A2infNorm3 = A2.NormInf();
double A2oneNorm3 = A2.NormOne();
// The last two scaling ops cancel each other out
if (A2infNorm3!=A2infNorm1) {
EPETRA_TEST_ERR(-35,ierr)
ScalingBroke = true;
}
if (A2oneNorm3!=A2oneNorm1) {
EPETRA_TEST_ERR(-36,ierr)
ScalingBroke = true;
}
if (verbose1) cout << A2;
EPETRA_TEST_ERR(A2.LeftScale(xRange),ierr);
double A2infNorm4 = A2.NormInf();
double A2oneNorm4 = A2.NormOne();
// The 4 scaling ops all cancel out
if (A2infNorm4!=A2infNorm) {
EPETRA_TEST_ERR(-37,ierr)
ScalingBroke = true;
}
if (A2oneNorm4!=A2oneNorm) {
EPETRA_TEST_ERR(-38,ierr)
ScalingBroke = true;
}
//
// Now try changing the values underneath and make sure that
// telling one process about the change causes NormInf() and
// NormOne() to recompute the norm on all processes.
//
double *values;
int num_my_rows = A2.NumMyRows() ;
int num_entries;
for ( int i=0 ; i< num_my_rows; i++ ) {
EPETRA_TEST_ERR( A2.ExtractMyRowView( i, num_entries, values ), ierr );
for ( int j = 0 ; j <num_entries; j++ ) {
values[j] *= 2.0;
}
}
if ( MyPID == 0 )
A2.SumIntoGlobalValues( 0, 0, 0, 0 ) ;
double A2infNorm5 = A2.NormInf();
double A2oneNorm5 = A2.NormOne();
if (A2infNorm5!=2.0 * A2infNorm4) {
EPETRA_TEST_ERR(-39,ierr)
ScalingBroke = true;
}
if (A2oneNorm5!= 2.0 * A2oneNorm4) {
EPETRA_TEST_ERR(-40,ierr)
ScalingBroke = true;
}
//
// Restore the values underneath
//
for ( int i=0 ; i< num_my_rows; i++ ) {
EPETRA_TEST_ERR( A2.ExtractMyRowView( i, num_entries, values ), ierr );
for ( int j = 0 ; j <num_entries; j++ ) {
values[j] /= 2.0;
}
}
if (verbose1) cout << A2;
if (ScalingBroke) {
if (verbose) cout << endl << "LeftScale and RightScale tests FAILED" << endl << endl;
}
else {
if (verbose) cout << endl << "LeftScale and RightScale tests PASSED" << endl << endl;
}
Comm.Barrier();
if (verbose) cout << "\n\n*****Testing InvRowMaxs and InvColMaxs" << endl << endl;
if (verbose1) cout << A2 << endl;
EPETRA_TEST_ERR(A2.InvRowMaxs(xRow),ierr);
EPETRA_TEST_ERR(A2.InvRowMaxs(xRange),ierr);
if (verbose1) cout << xRow << endl << xRange << endl;
if (verbose) cout << "\n\n*****Testing InvRowSums and InvColSums" << endl << endl;
bool InvSumsBroke = false;
// Works!
EPETRA_TEST_ERR(A2.InvRowSums(xRow),ierr);
if (verbose1) cout << xRow;
EPETRA_TEST_ERR(A2.LeftScale(xRow),ierr);
float A2infNormFloat = A2.NormInf();
if (verbose1) cout << A2 << endl;
if (fabs(1.0-A2infNormFloat) > 1.e-5) {
EPETRA_TEST_ERR(-41,ierr);
InvSumsBroke = true;
}
// Works
int expectedcode = 1;
if (Comm.NumProc()>1) expectedcode = 0;
EPETRA_TEST_ERR(!(A2.InvColSums(xDomain)==expectedcode),ierr); // This matrix has a single row, the first column has a zero, so a warning is issued.
if (verbose1) cout << xDomain << endl;
EPETRA_TEST_ERR(A2.RightScale(xDomain),ierr);
float A2oneNormFloat2 = A2.NormOne();
if (verbose1) cout << A2;
if (fabs(1.0-A2oneNormFloat2)>1.e-5) {
EPETRA_TEST_ERR(-42,ierr)
InvSumsBroke = true;
}
// Works!
EPETRA_TEST_ERR(A2.InvRowSums(xRange),ierr);
if (verbose1) cout << xRange;
EPETRA_TEST_ERR(A2.LeftScale(xRange),ierr);
float A2infNormFloat2 = A2.NormInf(); // We use a float so that rounding error
// will not prevent the sum from being 1.0.
if (verbose1) cout << A2;
if (fabs(1.0-A2infNormFloat2)>1.e-5) {
cout << "InfNorm should be = 1, but InfNorm = " << A2infNormFloat2 << endl;
EPETRA_TEST_ERR(-43,ierr);
InvSumsBroke = true;
}
// Doesn't work - may not need this test because column ownership is not unique
/* EPETRA_TEST_ERR(A2.InvColSums(xCol),ierr);
cout << xCol;
EPETRA_TEST_ERR(A2.RightScale(xCol),ierr);
float A2oneNormFloat = A2.NormOne();
cout << A2;
if (fabs(1.0-A2oneNormFloat)>1.e-5) {
EPETRA_TEST_ERR(-44,ierr);
InvSumsBroke = true;
}
*/
delete [] ColRightScaleValues;
delete [] DomainRightScaleValues;
if (verbose) cout << "Begin partial sum testing." << endl;
// Test with a matrix that has partial sums for a subset of the rows
// on multiple processors. (Except for the serial case, of course.)
int NumMyRows3 = 2; // Changing this requires further changes below
int * myGlobalElements = new int[NumMyRows3];
for (int i=0; i<NumMyRows3; i++) myGlobalElements[i] = MyPID+i;
Epetra_Map RowMap3(NumProc*2, NumMyRows3, myGlobalElements, 0, Comm);
int NumMyElements3 = 5;
Epetra_CrsMatrix A3(Copy, RowMap3, NumMyElements3);
double * Values3 = new double[NumMyElements3];
int * Indices3 = new int[NumMyElements3];
for (int i=0; i < NumMyElements3; i++) {
Values3[i] = (int) (MyPID + (i+1));
Indices3[i]=i;
}
for (int i=0; i<NumMyRows3; i++) {
A3.InsertGlobalValues(myGlobalElements[i],NumMyElements3,Values3,Indices3);
}
Epetra_Map RangeMap3(NumProc+1, 0, Comm);
Epetra_Map DomainMap3(NumMyElements3, 0, Comm);
EPETRA_TEST_ERR(A3.FillComplete(DomainMap3, RangeMap3,false),ierr);
if (verbose1) cout << A3;
Epetra_Vector xRange3(RangeMap3,false);
Epetra_Vector xDomain3(DomainMap3,false);
EPETRA_TEST_ERR(A3.InvRowSums(xRange3),ierr);
if (verbose1) cout << xRange3;
EPETRA_TEST_ERR(A3.LeftScale(xRange3),ierr);
float A3infNormFloat = A3.NormInf();
if (verbose1) cout << A3;
if (1.0!=A3infNormFloat) {
cout << "InfNorm should be = 1, but InfNorm = " << A3infNormFloat <<endl;
EPETRA_TEST_ERR(-61,ierr);
InvSumsBroke = true;
}
// we want to take the transpose of our matrix and fill in different values.
int NumMyColumns3 = NumMyRows3;
Epetra_Map ColMap3cm(RowMap3);
Epetra_Map RowMap3cm(A3.ColMap());
Epetra_CrsMatrix A3cm(Copy,RowMap3cm,ColMap3cm,NumProc+1);
double *Values3cm = new double[NumMyColumns3];
int * Indices3cm = new int[NumMyColumns3];
for (int i=0; i<NumMyColumns3; i++) {
Values3cm[i] = MyPID + i + 1;
Indices3cm[i]= i + MyPID;
}
for (int ii=0; ii<NumMyElements3; ii++) {
A3cm.InsertGlobalValues(ii, NumMyColumns3, Values3cm, Indices3cm);
}
// The DomainMap and the RangeMap from the last test will work fine for
// the RangeMap and DomainMap, respectively, but I will make copies to
// avaoid confusion when passing what looks like a DomainMap where we
// need a RangeMap and vice vera.
Epetra_Map RangeMap3cm(DomainMap3);
Epetra_Map DomainMap3cm(RangeMap3);
EPETRA_TEST_ERR(A3cm.FillComplete(DomainMap3cm,RangeMap3cm),ierr);
if (verbose1) cout << A3cm << endl;
// Again, we can copy objects from the last example.
//Epetra_Vector xRange3cm(xDomain3); //Don't use at this time
Epetra_Vector xDomain3cm(DomainMap3cm,false);
EPETRA_TEST_ERR(A3cm.InvColSums(xDomain3cm),ierr);
if (verbose1) cout << xDomain3cm << endl;
EPETRA_TEST_ERR(A3cm.RightScale(xDomain3cm),ierr);
float A3cmOneNormFloat = A3cm.NormOne();
if (verbose1) cout << A3cm << endl;
if (1.0!=A3cmOneNormFloat) {
cout << "OneNorm should be = 1, but OneNorm = " << A3cmOneNormFloat << endl;
EPETRA_TEST_ERR(-62,ierr);
InvSumsBroke = true;
}
if (verbose) cout << "End partial sum testing" << endl;
if (verbose) cout << "Begin replicated testing" << endl;
// We will now view the shared row as a repliated row, rather than one
// that has partial sums of its entries on mulitple processors.
// We will reuse much of the data used for the partial sum tesitng.
Epetra_Vector xRow3(RowMap3,false);
Epetra_CrsMatrix A4(Copy, RowMap3, NumMyElements3);
for (int ii=0; ii < NumMyElements3; ii++) {
Values3[ii] = (int)((ii*.6)+1.0);
}
for (int ii=0; ii<NumMyRows3; ii++) {
A4.InsertGlobalValues(myGlobalElements[ii],NumMyElements3,Values3,Indices3);
}
EPETRA_TEST_ERR(A4.FillComplete(DomainMap3, RangeMap3,false),ierr);
if (verbose1) cout << A4 << endl;
// The next two lines should be expanded into a verifiable test.
EPETRA_TEST_ERR(A4.InvRowMaxs(xRow3),ierr);
EPETRA_TEST_ERR(A4.InvRowMaxs(xRange3),ierr);
if (verbose1) cout << xRow3 << xRange3;
EPETRA_TEST_ERR(A4.InvRowSums(xRow3),ierr);
if (verbose1) cout << xRow3;
EPETRA_TEST_ERR(A4.LeftScale(xRow3),ierr);
float A4infNormFloat = A4.NormInf();
if (verbose1) cout << A4;
if (2.0!=A4infNormFloat && NumProc != 1) {
if (verbose1) cout << "InfNorm should be = 2 (because one column is replicated on two processors and NormOne() does not handle replication), but InfNorm = " << A4infNormFloat <<endl;
EPETRA_TEST_ERR(-63,ierr);
InvSumsBroke = true;
}
else if (1.0!=A4infNormFloat && NumProc == 1) {
if (verbose1) cout << "InfNorm should be = 1, but InfNorm = " << A4infNormFloat <<endl;
EPETRA_TEST_ERR(-63,ierr);
InvSumsBroke = true;
}
Epetra_Vector xCol3cm(ColMap3cm,false);
Epetra_CrsMatrix A4cm(Copy, RowMap3cm, ColMap3cm, NumProc+1);
//Use values from A3cm
for (int ii=0; ii<NumMyElements3; ii++) {
A4cm.InsertGlobalValues(ii,NumMyColumns3,Values3cm,Indices3cm);
}
EPETRA_TEST_ERR(A4cm.FillComplete(DomainMap3cm, RangeMap3cm,false),ierr);
if (verbose1) cout << A4cm << endl;
// The next two lines should be expanded into a verifiable test.
EPETRA_TEST_ERR(A4cm.InvColMaxs(xCol3cm),ierr);
EPETRA_TEST_ERR(A4cm.InvColMaxs(xDomain3cm),ierr);
if (verbose1) cout << xCol3cm << xDomain3cm;
EPETRA_TEST_ERR(A4cm.InvColSums(xCol3cm),ierr);
if (verbose1) cout << xCol3cm << endl;
EPETRA_TEST_ERR(A4cm.RightScale(xCol3cm),ierr);
float A4cmOneNormFloat = A4cm.NormOne();
if (verbose1) cout << A4cm << endl;
if (2.0!=A4cmOneNormFloat && NumProc != 1) {
if (verbose1) cout << "OneNorm should be = 2 (because one column is replicated on two processors and NormOne() does not handle replication), but OneNorm = " << A4cmOneNormFloat << endl;
EPETRA_TEST_ERR(-64,ierr);
InvSumsBroke = true;
}
else if (1.0!=A4cmOneNormFloat && NumProc == 1) {
if (verbose1) cout << "OneNorm should be = 1, but OneNorm = " << A4infNormFloat <<endl;
EPETRA_TEST_ERR(-64,ierr);
InvSumsBroke = true;
}
if (verbose) cout << "End replicated testing" << endl;
if (InvSumsBroke) {
if (verbose) cout << endl << "InvRowSums tests FAILED" << endl << endl;
}
else
if (verbose) cout << endl << "InvRowSums tests PASSED" << endl << endl;
A3cm.PutScalar(2.0);
int nnz_A3cm = A3cm.Graph().NumGlobalNonzeros();
double check_frobnorm = sqrt(nnz_A3cm*4.0);
double frobnorm = A3cm.NormFrobenius();
bool frobnorm_test_failed = false;
if (fabs(check_frobnorm-frobnorm) > 5.e-5) {
frobnorm_test_failed = true;
}
if (frobnorm_test_failed) {
if (verbose) std::cout << "Frobenius-norm test FAILED."<<std::endl;
EPETRA_TEST_ERR(-65, ierr);
}
delete [] Values2;
delete [] Indices2;
delete [] myGlobalElements;
delete [] Values3;
delete [] Indices3;
delete [] Values3cm;
delete [] Indices3cm;
delete [] RangeLeftScaleValues;
delete [] RowLeftScaleValues;
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
/* end main
*/
return ierr ;
}
int main(int argc, char *argv[])
{
#ifdef HAVE_MPI
MPI_Init(&argc, &argv);
Epetra_MpiComm Comm(MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
bool verbose = (Comm.MyPID() == 0);
// set global dimension to 5, could be any number
int NumGlobalElements = 5;
// create a map
Epetra_Map Map(NumGlobalElements,0,Comm);
// local number of rows
int NumMyElements = Map.NumMyElements();
// get update list
int * MyGlobalElements = Map.MyGlobalElements( );
// ============= CONSTRUCTION OF THE MATRIX ===========================
// Create a Epetra_Matrix
Epetra_CrsMatrix A(Copy,Map,3);
// Add rows one-at-a-time
double *Values = new double[2];
Values[0] = -1.0; Values[1] = -1.0;
int *Indices = new int[2];
double two = 2.0;
int NumEntries;
for( int 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;
}
A.InsertGlobalValues(MyGlobalElements[i], NumEntries, Values, Indices);
// Put in the diagonal entry
A.InsertGlobalValues(MyGlobalElements[i], 1, &two, MyGlobalElements+i);
}
// Finish up
A.FillComplete();
// ================ CONSTRUCTION OF VECTORS =======================
// build up two distributed vectors q and z, and compute
// q = A * z
Epetra_Vector q(A.RowMap());
Epetra_Vector z(A.RowMap());
// Fill z with 1's
z.PutScalar( 1.0 );
// ================ USE OF TIME AND FLOPS =========================
Epetra_Flops counter;
A.SetFlopCounter(counter);
Epetra_Time timer(Comm);
A.Multiply(false, z, q); // Compute q = A*z
double elapsed_time = timer.ElapsedTime();
double total_flops =counter.Flops();
if (verbose)
cout << "Total ops: " << total_flops << endl;
double MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose)
cout << "Total MFLOPs for mat-vec = " << MFLOPs << endl<< endl;
double dotProduct;
z.SetFlopCounter(counter);
timer.ResetStartTime();
z.Dot(q, &dotProduct);
total_flops =counter.Flops();
if (verbose)
cout << "Total ops: " << total_flops << endl;
elapsed_time = timer.ElapsedTime();
if (elapsed_time != 0.0)
MFLOPs = (total_flops / elapsed_time) / 1000000.0;
else
MFLOPs = 0;
if (verbose)
{
cout << "Total MFLOPs for vec-vec = " << MFLOPs << endl<< endl;
cout << "q dot z = " << dotProduct << endl;
}
#ifdef HAVE_MPI
MPI_Finalize();
#endif
return( 0 );
} /* main */
int main(int argc, char *argv[]) {
#ifdef EPETRA_MPI
MPI_Init(&argc,&argv);
Epetra_MpiComm Comm (MPI_COMM_WORLD);
#else
Epetra_SerialComm Comm;
#endif
cout << Comm << endl;
int MyPID = Comm.MyPID();
bool verbose = false;
bool verbose1 = true;
if (MyPID==0) verbose = true;
if(argc < 2 && verbose) {
cerr << "Usage: " << argv[0]
<< " HB_filename [level_fill [level_overlap [absolute_threshold [ relative_threshold]]]]" << endl
<< "where:" << endl
<< "HB_filename - filename and path of a Harwell-Boeing data set" << endl
<< "level_fill - The amount of fill to use for ILU(k) preconditioner (default 0)" << endl
<< "level_overlap - The amount of overlap used for overlapping Schwarz subdomains (default 0)" << endl
<< "absolute_threshold - The minimum value to place on the diagonal prior to factorization (default 0.0)" << endl
<< "relative_threshold - The relative amount to perturb the diagonal prior to factorization (default 1.0)" << endl << endl
<< "To specify a non-default value for one of these parameters, you must specify all" << endl
<< " preceding values but not any subsequent parameters. Example:" << endl
<< "ifpackHpcSerialMsr.exe mymatrix.hpc 1 - loads mymatrix.hpc, uses level fill of one, all other values are defaults" << endl
<< endl;
return(1);
}
// Uncomment the next three lines to debug in mpi mode
//int tmp;
//if (MyPID==0) cin >> tmp;
//Comm.Barrier();
Epetra_Map * readMap;
Epetra_CrsMatrix * readA;
Epetra_Vector * readx;
Epetra_Vector * readb;
Epetra_Vector * readxexact;
// Call routine to read in HB problem
Trilinos_Util_ReadHb2Epetra(argv[1], Comm, readMap, readA, readx, readb, readxexact);
// Create uniform distributed map
Epetra_Map map(readMap->NumGlobalElements(), 0, Comm);
// Create Exporter to distribute read-in matrix and vectors
Epetra_Export exporter(*readMap, map);
Epetra_CrsMatrix A(Copy, map, 0);
Epetra_Vector x(map);
Epetra_Vector b(map);
Epetra_Vector xexact(map);
Epetra_Time FillTimer(Comm);
x.Export(*readx, exporter, Add);
b.Export(*readb, exporter, Add);
xexact.Export(*readxexact, exporter, Add);
Comm.Barrier();
double vectorRedistributeTime = FillTimer.ElapsedTime();
A.Export(*readA, exporter, Add);
Comm.Barrier();
double matrixRedistributeTime = FillTimer.ElapsedTime() - vectorRedistributeTime;
assert(A.FillComplete()==0);
Comm.Barrier();
double fillCompleteTime = FillTimer.ElapsedTime() - matrixRedistributeTime;
if (Comm.MyPID()==0) {
cout << "\n\n****************************************************" << endl;
cout << "\n Vector redistribute time (sec) = " << vectorRedistributeTime<< endl;
cout << " Matrix redistribute time (sec) = " << matrixRedistributeTime << endl;
cout << " Transform to Local time (sec) = " << fillCompleteTime << endl<< endl;
}
Epetra_Vector tmp1(*readMap);
Epetra_Vector tmp2(map);
readA->Multiply(false, *readxexact, tmp1);
A.Multiply(false, xexact, tmp2);
double residual;
tmp1.Norm2(&residual);
if (verbose) cout << "Norm of Ax from file = " << residual << endl;
tmp2.Norm2(&residual);
if (verbose) cout << "Norm of Ax after redistribution = " << residual << endl << endl << endl;
//cout << "A from file = " << *readA << endl << endl << endl;
//cout << "A after dist = " << A << endl << endl << endl;
delete readA;
delete readx;
delete readb;
delete readxexact;
delete readMap;
Comm.Barrier();
bool smallProblem = false;
if (A.RowMap().NumGlobalElements()<100) smallProblem = true;
if (smallProblem)
cout << "Original Matrix = " << endl << A << endl;
x.PutScalar(0.0);
Epetra_LinearProblem FullProblem(&A, &x, &b);
double normb, norma;
b.NormInf(&normb);
norma = A.NormInf();
if (verbose)
cout << "Inf norm of Original Matrix = " << norma << endl
<< "Inf norm of Original RHS = " << normb << endl;
Epetra_Time ReductionTimer(Comm);
Epetra_CrsSingletonFilter SingletonFilter;
Comm.Barrier();
double reduceInitTime = ReductionTimer.ElapsedTime();
SingletonFilter.Analyze(&A);
Comm.Barrier();
double reduceAnalyzeTime = ReductionTimer.ElapsedTime() - reduceInitTime;
if (SingletonFilter.SingletonsDetected())
cout << "Singletons found" << endl;
else {
cout << "Singletons not found" << endl;
exit(1);
}
SingletonFilter.ConstructReducedProblem(&FullProblem);
Comm.Barrier();
double reduceConstructTime = ReductionTimer.ElapsedTime() - reduceInitTime;
double totalReduceTime = ReductionTimer.ElapsedTime();
if (verbose)
cout << "\n\n****************************************************" << endl
<< " Reduction init time (sec) = " << reduceInitTime<< endl
<< " Reduction Analyze time (sec) = " << reduceAnalyzeTime << endl
<< " Construct Reduced Problem time (sec) = " << reduceConstructTime << endl
<< " Reduction Total time (sec) = " << totalReduceTime << endl<< endl;
Statistics(SingletonFilter);
Epetra_LinearProblem * ReducedProblem = SingletonFilter.ReducedProblem();
Epetra_CrsMatrix * Ap = dynamic_cast<Epetra_CrsMatrix *>(ReducedProblem->GetMatrix());
Epetra_Vector * bp = (*ReducedProblem->GetRHS())(0);
Epetra_Vector * xp = (*ReducedProblem->GetLHS())(0);
if (smallProblem)
cout << " Reduced Matrix = " << endl << *Ap << endl
<< " LHS before sol = " << endl << *xp << endl
<< " RHS = " << endl << *bp << endl;
// Construct ILU preconditioner
double elapsed_time, total_flops, MFLOPs;
Epetra_Time timer(Comm);
int LevelFill = 0;
if (argc > 2) LevelFill = atoi(argv[2]);
if (verbose) cout << "Using Level Fill = " << LevelFill << endl;
int Overlap = 0;
if (argc > 3) Overlap = atoi(argv[3]);
if (verbose) cout << "Using Level Overlap = " << Overlap << endl;
double Athresh = 0.0;
if (argc > 4) Athresh = atof(argv[4]);
if (verbose) cout << "Using Absolute Threshold Value of = " << Athresh << endl;
double Rthresh = 1.0;
if (argc > 5) Rthresh = atof(argv[5]);
if (verbose) cout << "Using Relative Threshold Value of = " << Rthresh << endl;
Ifpack_IlukGraph * IlukGraph = 0;
Ifpack_CrsRiluk * ILUK = 0;
if (LevelFill>-1) {
elapsed_time = timer.ElapsedTime();
IlukGraph = new Ifpack_IlukGraph(Ap->Graph(), LevelFill, Overlap);
assert(IlukGraph->ConstructFilledGraph()==0);
elapsed_time = timer.ElapsedTime() - elapsed_time;
if (verbose) cout << "Time to construct ILUK graph = " << elapsed_time << endl;
Epetra_Flops fact_counter;
elapsed_time = timer.ElapsedTime();
ILUK = new Ifpack_CrsRiluk(*IlukGraph);
ILUK->SetFlopCounter(fact_counter);
ILUK->SetAbsoluteThreshold(Athresh);
ILUK->SetRelativeThreshold(Rthresh);
//assert(ILUK->InitValues()==0);
int initerr = ILUK->InitValues(*Ap);
if (initerr!=0) {
cout << endl << Comm << endl << " InitValues error = " << initerr;
if (initerr==1) cout << " Zero diagonal found, warning error only";
cout << endl << endl;
}
assert(ILUK->Factor()==0);
elapsed_time = timer.ElapsedTime() - elapsed_time;
total_flops = ILUK->Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Time to compute preconditioner values = "
<< elapsed_time << endl
<< "MFLOPS for Factorization = " << MFLOPs << endl;
//cout << *ILUK << endl;
double Condest;
ILUK->Condest(false, Condest);
if (verbose) cout << "Condition number estimate for this preconditioner = " << Condest << endl;
}
int Maxiter = 100;
double Tolerance = 1.0E-8;
Epetra_Flops counter;
Ap->SetFlopCounter(counter);
xp->SetFlopCounter(*Ap);
bp->SetFlopCounter(*Ap);
if (ILUK!=0) ILUK->SetFlopCounter(*Ap);
elapsed_time = timer.ElapsedTime();
double normreducedb, normreduceda;
bp->NormInf(&normreducedb);
normreduceda = Ap->NormInf();
if (verbose)
cout << "Inf norm of Reduced Matrix = " << normreduceda << endl
<< "Inf norm of Reduced RHS = " << normreducedb << endl;
BiCGSTAB(*Ap, *xp, *bp, ILUK, Maxiter, Tolerance, &residual, verbose);
elapsed_time = timer.ElapsedTime() - elapsed_time;
total_flops = counter.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Time to compute solution = "
<< elapsed_time << endl
<< "Number of operations in solve = " << total_flops << endl
<< "MFLOPS for Solve = " << MFLOPs<< endl << endl;
SingletonFilter.ComputeFullSolution();
if (smallProblem)
cout << " Reduced LHS after sol = " << endl << *xp << endl
<< " Full LHS after sol = " << endl << x << endl
<< " Full Exact LHS = " << endl << xexact << endl;
Epetra_Vector resid(x);
resid.Update(1.0, x, -1.0, xexact, 0.0); // resid = xcomp - xexact
resid.Norm2(&residual);
double normx, normxexact;
x.Norm2(&normx);
xexact.Norm2(&normxexact);
if (verbose)
cout << "2-norm of computed solution = " << normx << endl
<< "2-norm of exact solution = " << normxexact << endl
<< "2-norm of difference between computed and exact solution = " << residual << endl;
if (verbose1 && residual>1.0e-5) {
if (verbose)
cout << "Difference between computed and exact solution appears large..." << endl
<< "Computing norm of A times this difference. If this norm is small, then matrix is singular"
<< endl;
Epetra_Vector bdiff(b);
assert(A.Multiply(false, resid, bdiff)==0);
assert(bdiff.Norm2(&residual)==0);
if (verbose)
cout << "2-norm of A times difference between computed and exact solution = " << residual << endl;
}
if (verbose)
cout << "********************************************************" << endl
<< " Solving again with 2*Ax=2*b" << endl
<< "********************************************************" << endl;
A.Scale(1.0); // A = 2*A
b.Scale(1.0); // b = 2*b
x.PutScalar(0.0);
b.NormInf(&normb);
norma = A.NormInf();
if (verbose)
cout << "Inf norm of Original Matrix = " << norma << endl
<< "Inf norm of Original RHS = " << normb << endl;
double updateReducedProblemTime = ReductionTimer.ElapsedTime();
SingletonFilter.UpdateReducedProblem(&FullProblem);
Comm.Barrier();
updateReducedProblemTime = ReductionTimer.ElapsedTime() - updateReducedProblemTime;
if (verbose)
cout << "\n\n****************************************************" << endl
<< " Update Reduced Problem time (sec) = " << updateReducedProblemTime<< endl
<< "****************************************************" << endl;
Statistics(SingletonFilter);
if (LevelFill>-1) {
Epetra_Flops fact_counter;
elapsed_time = timer.ElapsedTime();
int initerr = ILUK->InitValues(*Ap);
if (initerr!=0) {
cout << endl << Comm << endl << " InitValues error = " << initerr;
if (initerr==1) cout << " Zero diagonal found, warning error only";
cout << endl << endl;
}
assert(ILUK->Factor()==0);
elapsed_time = timer.ElapsedTime() - elapsed_time;
total_flops = ILUK->Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Time to compute preconditioner values = "
<< elapsed_time << endl
<< "MFLOPS for Factorization = " << MFLOPs << endl;
double Condest;
ILUK->Condest(false, Condest);
if (verbose) cout << "Condition number estimate for this preconditioner = " << Condest << endl;
}
bp->NormInf(&normreducedb);
normreduceda = Ap->NormInf();
if (verbose)
cout << "Inf norm of Reduced Matrix = " << normreduceda << endl
<< "Inf norm of Reduced RHS = " << normreducedb << endl;
BiCGSTAB(*Ap, *xp, *bp, ILUK, Maxiter, Tolerance, &residual, verbose);
elapsed_time = timer.ElapsedTime() - elapsed_time;
total_flops = counter.Flops();
MFLOPs = total_flops/elapsed_time/1000000.0;
if (verbose) cout << "Time to compute solution = "
<< elapsed_time << endl
<< "Number of operations in solve = " << total_flops << endl
<< "MFLOPS for Solve = " << MFLOPs<< endl << endl;
SingletonFilter.ComputeFullSolution();
if (smallProblem)
cout << " Reduced LHS after sol = " << endl << *xp << endl
<< " Full LHS after sol = " << endl << x << endl
<< " Full Exact LHS = " << endl << xexact << endl;
resid.Update(1.0, x, -1.0, xexact, 0.0); // resid = xcomp - xexact
resid.Norm2(&residual);
x.Norm2(&normx);
xexact.Norm2(&normxexact);
if (verbose)
cout << "2-norm of computed solution = " << normx << endl
<< "2-norm of exact solution = " << normxexact << endl
<< "2-norm of difference between computed and exact solution = " << residual << endl;
if (verbose1 && residual>1.0e-5) {
if (verbose)
cout << "Difference between computed and exact solution appears large..." << endl
<< "Computing norm of A times this difference. If this norm is small, then matrix is singular"
<< endl;
Epetra_Vector bdiff(b);
assert(A.Multiply(false, resid, bdiff)==0);
assert(bdiff.Norm2(&residual)==0);
if (verbose)
cout << "2-norm of A times difference between computed and exact solution = " << residual << endl;
}
if (ILUK!=0) delete ILUK;
if (IlukGraph!=0) delete IlukGraph;
#ifdef EPETRA_MPI
MPI_Finalize() ;
#endif
return 0 ;
}
