void ewol::object::Manager::timeCall(int64_t _localTime) {
int64_t previousTime = m_lastPeriodicCallTime;
m_lastPeriodicCallTime = _localTime;
if (periodicCall.getNumberConnected() <= 0) {
return;
}
float deltaTime = (float)(_localTime - previousTime)/1000000.0;
ewol::event::Time myTime(_localTime, m_applWakeUpTime, deltaTime, deltaTime);
periodicCall.emit(myTime);
}
//=========================================================================
RowMatrix_Transpose::NewTypeRef
RowMatrix_Transpose::
operator()( OriginalTypeRef orig )
{
origObj_ = &orig;
if( !TransposeRowMap_ )
{
if( IgnoreNonLocalCols_ )
TransposeRowMap_ = (Epetra_Map *) &(orig.OperatorRangeMap()); // Should be replaced with refcount =
else
TransposeRowMap_ = (Epetra_Map *) &(orig.OperatorDomainMap()); // Should be replaced with refcount =
}
NumMyRows_ = orig.NumMyRows();
NumMyCols_ = orig.NumMyCols();
// This routine will work for any RowMatrix object, but will attempt cast the matrix to a CrsMatrix if
// possible (because we can then use a View of the matrix and graph, which is much cheaper).
// First get the local indices to count how many nonzeros will be in the
// transpose graph on each processor
Epetra_CrsMatrix * TempTransA1 = CreateTransposeLocal(orig);
if(!TempTransA1->Exporter()) {
// Short Circuit: There is no need to make another matrix since TransposeRowMap_== TransMap
newObj_ = TransposeMatrix_ = TempTransA1;
return *newObj_;
}
#ifdef ENABLE_TRANSPOSE_TIMINGS
Teuchos::Time myTime("global");
Teuchos::TimeMonitor MM(myTime);
Teuchos::RCP<Teuchos::Time> mtime;
mtime=MM.getNewTimer("Transpose: Final FusedExport");
mtime->start();
#endif
// Now that transpose matrix with shared rows is entered, create a new matrix that will
// get the transpose with uniquely owned rows (using the same row distribution as A).
TransposeMatrix_ = new Epetra_CrsMatrix(*TempTransA1,*TempTransA1->Exporter(),NULL,0,TransposeRowMap_, false);
#ifdef ENABLE_TRANSPOSE_TIMINGS
mtime->stop();
#endif
newObj_ = TransposeMatrix_;
delete TempTransA1;
return *newObj_;
}
//=========================================================================
Epetra_CrsMatrix* EpetraExt::RowMatrix_Transpose::CreateTransposeLocal(OriginalTypeRef orig)
{
#ifdef ENABLE_TRANSPOSE_TIMINGS
Teuchos::Time myTime("global");
Teuchos::TimeMonitor MM(myTime);
Teuchos::RCP<Teuchos::Time> mtime;
mtime=MM.getNewTimer("Transpose: CreateTransposeLocal 1");
mtime->start();
#endif
int i,j,err;
const Epetra_CrsMatrix * OrigCrsMatrix = dynamic_cast<const Epetra_CrsMatrix*>(&orig);
if(OrigCrsMatrix) OrigMatrixIsCrsMatrix_ = true;
else OrigMatrixIsCrsMatrix_ = false;
const Epetra_Map & TransMap = orig.RowMatrixColMap();
int TransNnz = orig.NumMyNonzeros();
int NumIndices;
Epetra_CrsMatrix *TempTransA1 = new Epetra_CrsMatrix(Copy, TransMap,orig.RowMatrixRowMap(),0);
Epetra_IntSerialDenseVector & TransRowptr = TempTransA1->ExpertExtractIndexOffset();
Epetra_IntSerialDenseVector & TransColind = TempTransA1->ExpertExtractIndices();
double *& TransVals = TempTransA1->ExpertExtractValues();
NumMyRows_ = orig.NumMyRows();
NumMyCols_ = orig.NumMyCols();
TransRowptr.Resize(NumMyCols_+1);
TransColind.Resize(TransNnz);
resize_doubles(0,TransNnz,TransVals);
std::vector<int> CurrentStart(NumMyCols_,0);
// Count up nnz using the Rowptr to count the number of non-nonzeros.
if (OrigMatrixIsCrsMatrix_)
{
const Epetra_CrsGraph & OrigGraph = OrigCrsMatrix->Graph(); // Get matrix graph
for (i=0; i<NumMyRows_; i++)
{
err = OrigGraph.ExtractMyRowView(i, NumIndices, Indices_); // Get view of ith row
if (err != 0) throw OrigGraph.ReportError("ExtractMyRowView failed",err);
for (j=0; j<NumIndices; j++) ++CurrentStart[Indices_[j]];
}
}
else // Original is not a CrsMatrix
{
MaxNumEntries_ = 0;
MaxNumEntries_ = orig.MaxNumEntries();
delete [] Indices_; delete [] Values_;
Indices_ = new int[MaxNumEntries_];
Values_ = new double[MaxNumEntries_];
for (i=0; i<NumMyRows_; i++)
{
err = orig.ExtractMyRowCopy(i, MaxNumEntries_, NumIndices, Values_, Indices_);
if (err != 0) {
std::cerr << "ExtractMyRowCopy failed."<<std::endl;
throw err;
}
for (j=0; j<NumIndices; j++) ++CurrentStart[Indices_[j]];
}
}
// Scansum the rowptr; reset currentstart
TransRowptr[0] = 0;
for (i=1;i<NumMyCols_+1; i++) TransRowptr[i] = CurrentStart[i-1] + TransRowptr[i-1];
for (i=0;i<NumMyCols_; i++) CurrentStart[i] = TransRowptr[i];
// Now copy values and global indices into newly create transpose storage
for (i=0; i<NumMyRows_; i++)
{
if (OrigMatrixIsCrsMatrix_)
err = OrigCrsMatrix->ExtractMyRowView(i, NumIndices, Values_, Indices_);
else
err = orig.ExtractMyRowCopy(i, MaxNumEntries_, NumIndices, Values_, Indices_);
if (err != 0) {
std::cerr << "ExtractMyRowCopy failed."<<std::endl;
throw err;
}
for (j=0; j<NumIndices; j++)
{
int idx = CurrentStart[Indices_[j]];
TransColind[idx] = i;
TransVals[idx] = Values_[j];
++CurrentStart[Indices_[j]];// increment the counter into the local row
}
}
#ifdef ENABLE_TRANSPOSE_TIMINGS
mtime->stop();
mtime=MM.getNewTimer("Transpose: CreateTransposeLocal 2");
mtime->start();
#endif
// Prebuild the importers and exporters the no-communication way, flipping the importers
// and exporters around.
Epetra_Import * myimport = 0;
Epetra_Export * myexport = 0;
if(OrigMatrixIsCrsMatrix_ && OrigCrsMatrix->Importer())
myexport = new Epetra_Export(*OrigCrsMatrix->Importer());
if(OrigMatrixIsCrsMatrix_ && OrigCrsMatrix->Exporter())
myimport = new Epetra_Import(*OrigCrsMatrix->Exporter());
#ifdef ENABLE_TRANSPOSE_TIMINGS
mtime->stop();
mtime=MM.getNewTimer("Transpose: CreateTransposeLocal 3");
mtime->start();
#endif
// Call ExpertStaticFillComplete
err = TempTransA1->ExpertStaticFillComplete(orig.OperatorRangeMap(),orig.OperatorDomainMap(),myimport,myexport);
if (err != 0) {
throw TempTransA1->ReportError("ExpertStaticFillComplete failed.",err);
}
#ifdef ENABLE_TRANSPOSE_TIMINGS
mtime->stop();
#endif
return TempTransA1;
}
int main(int argc, char* argv[])
{
typedef long Int;
typedef double Entry;
typedef Kokkos::OpenMP Exe_Space;
if(argc < 2)
{
std::cout << "Inputs nthreads matrix.mtx (optional rhs.mtx)" << std::endl;
return -1;
}
Int nthreads = atoi(argv[1]);
std::string mname = std::string(argv[2]);
std::string vname;
//Load inital information
//Matrix
Int m, n, nnz;
m = n = nnz = 0;
Int *col_ptr, *row_idx;
Entry *val;
col_ptr = NULL;
std::cout << "Matrix read" << std::endl;
double rmatrix = myTime();
readMatrix<Int,Entry>(mname, m, n, nnz,
&col_ptr, &row_idx, &val);
std::cout << "Read Matrix, Time: "
<< totalTime(rmatrix,myTime()) << std::endl;
//RHS
Int vn, vm;
vn = vm = 0;
Entry *x, *xhat, *y;
x = xhat = y = NULL;
vn = n;
x = new Entry[vn]();
xhat = new Entry[vn]();
if(argc == 4)
{
vname = std::string(argv[3]);
//std::cout << "reading vector " << vname << std::endl;
readVector<Int,Entry>(vname, vm, &y);
}
else
{
vm = m;
y = new Entry[m]();
for(Int i = 0; i < vm; i++)
{
xhat[i] = (Entry) i;
}
multiply<Int,Entry>(m,n,col_ptr,row_idx,val, xhat, y);
for(Int i = 0; i < vm; i++)
{
//std::cout << "y " << y[i] << std::endl;
xhat[i] = (Entry) 0.0;
}
}
//Starting up Kokkos
Exe_Space::initialize(nthreads);
std::cout << "Kokkos Settings" << std::endl;
std::cout << "hwloc aval: "
<< Kokkos::hwloc::available()<< std::endl;
std::cout << "numa count: "
<< Kokkos::hwloc::get_available_numa_count()
<< std::endl;
std::cout << "thrd numa: "
<< Kokkos::hwloc::get_available_cores_per_numa()
<< std::endl;
//Start Basker
{
// int result = 0; // NDE warning unused
BaskerNS::Basker<Int, Entry, Exe_Space> mybasker;
//---Options
mybasker.Options.same_pattern = BASKER_FALSE;
mybasker.Options.verbose = BASKER_FALSE;
mybasker.Options.verbose_matrix_out = BASKER_FALSE;
mybasker.Options.realloc = BASKER_TRUE;
mybasker.Options.transpose = BASKER_FALSE;
mybasker.Options.symmetric = BASKER_FALSE;
mybasker.Options.AtA = BASKER_TRUE;
mybasker.Options.A_plus_At = BASKER_TRUE;
mybasker.Options.matching = BASKER_TRUE;
mybasker.Options.matching_type = BASKER_MATCHING_BN;
mybasker.Options.btf = BASKER_TRUE;
mybasker.Options.btf_max_percent = BASKER_BTF_MAX_PERCENT;
mybasker.Options.btf_large = BASKER_BTF_LARGE;
mybasker.Options.no_pivot = BASKER_FALSE;
//mybasker.Options.pivot_tol = .001;
//mybasker.Options.pivot_bias = .001;
//mybasker.Options.btf_prune_size = 2;
mybasker.SetThreads(nthreads);
std::cout << "Setting Threads:" << nthreads << std::endl;
double stime = myTime();
mybasker.Symbolic(m,n,nnz,col_ptr,row_idx,val);
std::cout << "Done with Symbolic, Time: "
<< totalTime(stime, myTime()) << std::endl;
double ftime = myTime();
mybasker.Factor(m,n,nnz,col_ptr,row_idx,val);
std::cout << "Done with Factor, Time: "
<< totalTime(ftime, myTime()) << std::endl;
//mybasker.DEBUG_PRINT();
double ttime = myTime();
mybasker.SolveTest();
Int *lperm;
Int *rperm;
mybasker.GetPerm(&lperm,&rperm);
mybasker.Solve(y,x);
std::cout << "Done with Solve, Time: "
<< totalTime(ttime, myTime()) << std::endl;
multiply<Int,Entry>(m,n,col_ptr,row_idx,val, x, xhat);
for(Int i = 0; i < m; i++)
{
xhat[i] = y[i] - xhat[i];
}
/*
for(Int i = 0; i < n; i++)
{
std::cout << "idx: " << i
<< "x: " << x[i]
<< " xhat: " << xhat[i]
<< " y: " << y[i]
<< std::endl;
}
*/
std::cout << "||X||: " << norm2<Int,Entry>(n,x)
<< " ||Y-AX||: " << norm2<Int,Entry>(m,xhat)
<< std::endl;
/*
//Refactor
double rftime = myTime();
mybasker.Factor(m,n,nnz,col_ptr,row_idx,val);
std::cout << "Done with Refactor Factor, Time: "
<< totalTime(rftime, myTime()) << std::endl;
//ReSolve
double rttime = myTime();
mybasker.Solve(y,x);
std::cout << "Done with Refactor Solve, Time: "
<< totalTime(rttime, myTime()) << std::endl;
multiply<Int,Entry>(m,n,col_ptr,row_idx,val, x, xhat);
for(Int i = 0; i < m; i++)
{
xhat[i] = y[i] - xhat[i];
}
std::cout << "||X||: " << norm2<Int,Entry>(n,x)
<< " ||Y-AX||: " << norm2<Int,Entry>(m,xhat)
<< std::endl;
*/
//mybasker.GetPerm()
mybasker.Finalize();
}
Kokkos::finalize();
}//end main
