void MUMPS_solver::get_factorization( Mat<double, Gen<>, SparseLine<> > &mat, bool want_free ) {
if ( already_factorized ) return;
else already_factorized = true;
init_MPI();
/// initialisation de MUMPS pour un seul processeur
id.par = 1; /// le maître participe à la factorisation
id.sym = 0; /// matrice générale
id.comm_fortran = use_comm_word; /// communicator de MPI
id.job = job_init;
dmumps_c( &id );
id.ICNTL( 1 ) = -1; /*! No outputs */
id.ICNTL( 2 ) = -1; /*! No outputs */
id.ICNTL( 3 ) = -1; /*! No outputs */
id.ICNTL( 4 ) = -1; /*! No outputs */
/*! Define the problem on the host */
if ( myid == 0 ) {
free();
/// construction des données nécessaires à la factorisation (excepté a qui peut ne pas être nécessaire )
load_matrix( mat );
id.job = job_analyse;
dmumps_c( &id );
id.job = job_factorize;
dmumps_c( &id );
if ( want_free )
mat.free();
}
}
int MAIN__()
{
int argc = 1;
char * name = "c_example";
char ** argv;
#else
int main(int argc, char ** argv)
{
#endif
DMUMPS_STRUC_C id;
MUMPS_INT n = 2;
MUMPS_INT nz = 2;
MUMPS_INT irn[] = { 1, 2 };
MUMPS_INT jcn[] = { 1, 2 };
double a[2];
double rhs[2];
MUMPS_INT myid, ierr;
#if defined(MAIN_COMP)
argv = &name;
#endif
ierr = MPI_Init(&argc, &argv);
ierr = MPI_Comm_rank(MPI_COMM_WORLD, &myid);
/* Define A and rhs */
rhs[0] = 1.0; rhs[1] = 4.0;
a[0] = 1.0; a[1] = 2.0;
/* Initialize a MUMPS instance. Use MPI_COMM_WORLD */
id.job = JOB_INIT; id.par = 1; id.sym = 0; id.comm_fortran = USE_COMM_WORLD;
dmumps_c(&id);
/* Define the problem on the host */
if (myid == 0) {
id.n = n; id.nz = nz; id.irn = irn; id.jcn = jcn;
id.a = a; id.rhs = rhs;
}
#define ICNTL(I) icntl[(I)-1] /* macro s.t. indices match documentation */
/* No outputs */
id.ICNTL(1) = -1; id.ICNTL(2) = -1; id.ICNTL(3) = -1; id.ICNTL(4) = 0;
/* Call the MUMPS package. */
id.job = 6;
dmumps_c(&id);
id.job = JOB_END; dmumps_c(&id); /* Terminate instance */
if (myid == 0) {
printf("Solution is : (%8.2f %8.2f)\n", rhs[0], rhs[1]);
}
ierr = MPI_Finalize();
getchar();
return 0;
}
//-------------------------------------------------------------------
// Default Constructor
//-------------------------------------------------------------------
ClpCholeskyMumps::ClpCholeskyMumps (int denseThreshold)
: ClpCholeskyBase(denseThreshold)
{
mumps_ = (DMUMPS_STRUC_C*)malloc(sizeof(DMUMPS_STRUC_C));
type_ = 16;
mumps_->n = 0;
mumps_->nz = 0;
mumps_->a = NULL;
mumps_->jcn = NULL;
mumps_->irn = NULL;
mumps_->job = JOB_INIT;//initialize mumps
mumps_->par = 1;//working host for sequential version
mumps_->sym = 2;//general symmetric matrix
mumps_->comm_fortran = USE_COMM_WORLD;
int myid, ierr;
int justName;
ierr = MPI_Init(&justName, NULL);
ierr = MPI_Comm_rank(MPI_COMM_WORLD, &myid);
assert (!ierr);
dmumps_c(mumps_);
#define ICNTL(I) icntl[(I)-1] /* macro s.t. indices match documentation */
#define CNTL(I) cntl[(I)-1] /* macro s.t. indices match documentation */
mumps_->ICNTL(5) = 1; // say compressed format
mumps_->ICNTL(4) = 2; // log messages
mumps_->ICNTL(24) = 1; // Deal with zeros on diagonal
mumps_->CNTL(3) = 1.0e-20; // drop if diagonal less than this
// output off
mumps_->ICNTL(1) = -1;
mumps_->ICNTL(2) = -1;
mumps_->ICNTL(3) = -1;
mumps_->ICNTL(4) = 0;
}
//=============================================================================
void Amesos_Mumps::Destroy()
{
if (!NoDestroy_)
{
// destroy instance of the package
MDS.job = -2;
if (Comm().MyPID() < MaxProcs_) dmumps_c(&(MDS));
#if 0
if (IsComputeSchurComplementOK_ && (Comm().MyPID() == 0)
&& MDS.schur) {
delete [] MDS.schur;
MDS.schur = 0;
}
#endif
#ifdef HAVE_MPI
if (MUMPSComm_)
{
MPI_Comm_free( &MUMPSComm_ );
MUMPSComm_ = 0;
}
#endif
if( (verbose_ && PrintTiming_) || verbose_ == 2) PrintTiming();
if( (verbose_ && PrintStatus_) || verbose_ == 2) PrintStatus();
}
}
bool Chordal::solveSchur(Vector& rhs)
{
mumps_id.job = MUMPS_JOB_SOLVE;
mumps_id.rhs = rhs.ele;
dmumps_c(&mumps_id);
return SDPA_SUCCESS;
}
/* Uses factorization to solve. */
void
ClpCholeskyMumps::solve (double * region)
{
mumps_->rhs = region;
mumps_->job = 3; // solve
dmumps_c(mumps_);
}
//-------------------------------------------------------------------
// Destructor
//-------------------------------------------------------------------
ClpCholeskyMumps::~ClpCholeskyMumps ()
{
mumps_->job = JOB_END;
dmumps_c(mumps_); /* Terminate instance */
MPI_Finalize();
free(mumps_);
}
ESymSolverStatus MumpsSolverInterface::Solve(Index nrhs, double *rhs_vals)
{
DBG_START_METH("MumpsSolverInterface::Solve", dbg_verbosity);
DMUMPS_STRUC_C* mumps_data = (DMUMPS_STRUC_C*)mumps_ptr_;
ESymSolverStatus retval = SYMSOLVER_SUCCESS;
if (HaveIpData()) {
IpData().TimingStats().LinearSystemBackSolve().Start();
}
for (Index i = 0; i < nrhs; i++) {
Index offset = i * mumps_data->n;
mumps_data->rhs = &(rhs_vals[offset]);
mumps_data->job = 3;//solve
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Calling MUMPS-3 for solve at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
dmumps_c(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Done with MUMPS-3 for solve at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
int error = mumps_data->info[0];
if (error < 0) {
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"Error=%d returned from MUMPS in Solve.\n",
error);
retval = SYMSOLVER_FATAL_ERROR;
}
}
if (HaveIpData()) {
IpData().TimingStats().LinearSystemBackSolve().End();
}
return retval;
}
MumpsSolverInterface::MumpsSolverInterface()
{
DBG_START_METH("MumpsSolverInterface::MumpsSolverInterface()",
dbg_verbosity);
//initialize mumps
DMUMPS_STRUC_C* mumps_ = new DMUMPS_STRUC_C;
int argc=1;
char ** argv = 0;
int myid;
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &myid);
mumps_->n = 0;
mumps_->nz = 0;
mumps_->a = NULL;
mumps_->jcn = NULL;
mumps_->irn = NULL;
mumps_->job = -1;//initialize mumps
mumps_->par = 1;//working host for sequential version
mumps_->sym = 2;//general symetric matrix
mumps_->comm_fortran = USE_COMM_WORLD;
dmumps_c(mumps_);
mumps_->icntl[1] = 0;
mumps_->icntl[2] = 0;//QUIETLY!
mumps_->icntl[3] = 0;
mumps_ptr_ = (void*)mumps_;
}
bool Chordal::factorizeSchur(int m, int* diagonalIndex,
FILE* Display, FILE* fpOut)
{
// I need to adjust Schur before factorization
// to loose Numerical Error Condition
double adjustSize = PLUS_ADJUST_DIAGONAL;
for (int i=0; i<m; ++i) {
sparse_bMat_ptr->sp_ele[diagonalIndex[i]] += adjustSize;
}
mumps_id.job = MUMPS_JOB_FACTORIZE;
mumps_id.a = sparse_bMat_ptr->sp_ele;
dmumps_c(&mumps_id);
bool isSuccess = SDPA_SUCCESS;
while (mumps_id.infog[1-1] == -9) {
#if 0
rMessage("mumps icntl(14) = " << mumps_id.icntl[14-1]);
rMessage("mumps icntl(23) = " << mumps_id.icntl[23-1]);
#endif
if (Display) {
fprintf(Display,"MUMPS needs more memory space. Trying ANALYSIS phase once more\n");
}
if (fpOut) {
fprintf(fpOut, "MUMPS needs more memory space. Trying ANALYSIS phase once more\n");
}
mumps_id.icntl[14-1] += 20; // More 20% working memory space
analysisAndcountLowerNonZero(m);
mumps_id.job = MUMPS_JOB_FACTORIZE;
dmumps_c(&mumps_id);
}
if (mumps_id.infog[1-1] < 0) {
isSuccess = SDPA_FAILURE;
if (mumps_id.infog[1-1] == -10) {
rMessage("Cholesky failed by NUMERICAL ERROR");
rMessage("There are some possibilities.");
rMessage("1. SDPA terminates due to inaccuracy of numerical error");
rMessage("2. The input problem may not have (any) interior-points");
rMessage("3. Input matrices are linearly dependent");
}
else {
rMessage("Cholesky failed with Error Code "
<< mumps_id.infog[1-1]);
}
}
return isSuccess;
}
MUMPS_solver::~MUMPS_solver() {
if ( MPI_is_initialized ) {
id.job = job_end;
dmumps_c( &id ); /** Terminate instance */
ierr = MPI_Finalize();
//std::cerr << "MPI_Finalize() = " << ierr << std::endl;
}
}
void Chordal::terminate()
{
if (mumps_usage == true) {
mumps_id.job = MUMPS_JOB_END;
dmumps_c(&mumps_id);
mumps_usage = false;
}
if (sparse_bMat_ptr) {
sparse_bMat_ptr->terminate();
}
sparse_bMat_ptr = NULL;
}
MumpsSolverInterface::~MumpsSolverInterface()
{
DBG_START_METH("MumpsSolverInterface::~MumpsSolverInterface()",
dbg_verbosity);
DMUMPS_STRUC_C* mumps_ = (DMUMPS_STRUC_C*)mumps_ptr_;
mumps_->job = -2; //terminate mumps
dmumps_c(mumps_);
MPI_Finalize();
delete [] mumps_->a;
delete mumps_;
}
void Chordal::initialize(SparseMatrix* sparse_bMat_ptr)
{
// condition of sparse computation
// m_threshold < mDim,
// b_threshold < nBlock,
// aggregate_threshold >= aggrigated sparsity ratio
// extend_threshold >= extended sparsity ratio
m_threshold = 100;
b_threshold = 5;
aggregate_threshold = 0.70;
extend_threshold = 0.80;
#if FORCE_SCHUR_DENSE // DENSE computation for debugging
m_threshold = 10000000;
b_threshold = 1000000;
aggregate_threshold = 0.0;
extend_threshold = 0.0;
#endif
#if FORCE_SCHUR_SPARSE // SPARSE computation for debugging
m_threshold = 0;
b_threshold = 0;
aggregate_threshold = 2.0;
extend_threshold = 2.0;
#endif
// initialize by assuming Schur would be DENSE
best = SELECT_DENSE;
this->sparse_bMat_ptr = sparse_bMat_ptr;
// initialize MUMPS
mumps_id.job = MUMPS_JOB_INIT;
mumps_id.comm_fortran = MUMPS_USE_COMM_WORLD;
// rank 0 process participates factorizations
mumps_id.par = 1;
// Only symmetric positive definite matricies
mumps_id.sym = 1;
// No OUTPUTS
mumps_id.icntl[1-1] = -1;
mumps_id.icntl[2-1] = -1;
mumps_id.icntl[3-1] = -1;
mumps_id.icntl[4-1] = 0;
// MUMPS selects ordering automatically
mumps_id.icntl[7-1] = SELECT_MUMPS_BEST;
// for Minumum Degree Ordering
// mumps_id.icntl[7-1] = 0;
dmumps_c(&mumps_id);
mumps_usage = true;
}
int Amesos_Mumps::NumericFactorization()
{
IsNumericFactorizationOK_ = false;
if (IsSymbolicFactorizationOK_ == false)
AMESOS_CHK_ERR(SymbolicFactorization());
RedistrMatrix(true);
AMESOS_CHK_ERR(ConvertToTriplet(true));
if (Comm().NumProc() != 1)
{
if (Comm().MyPID() < MaxProcs_)
MDS.a_loc = &Val[0];
}
else
MDS.a = &Val[0];
// Request numeric factorization
MDS.job = 2;
// Perform numeric factorization
ResetTimer();
if (Comm().MyPID() < MaxProcs_) {
dmumps_c(&(MDS));
}
NumFactTime_ = AddTime("Total numeric factorization time", NumFactTime_);
int IntWrong = CheckError()?1:0 ;
int AnyWrong;
Comm().SumAll( &IntWrong, &AnyWrong, 1 ) ;
bool Wrong = AnyWrong > 0 ;
if ( Wrong ) {
AMESOS_CHK_ERR( NumericallySingularMatrixError ) ;
}
IsNumericFactorizationOK_ = true;
NumNumericFact_++;
return(0);
}
double Chordal::analysisAndcountLowerNonZero(int m)
{
mumps_id.job = MUMPS_JOB_ANALYSIS;
mumps_id.n = m;
mumps_id.nz = sparse_bMat_ptr->NonZeroCount;
mumps_id.irn = sparse_bMat_ptr->row_index;
mumps_id.jcn = sparse_bMat_ptr->column_index;
mumps_id.a = sparse_bMat_ptr->sp_ele;
// sparse_bMat_ptr->display();
// rMessage("m = " << m);
// rMessage("NonZeroCount = " << sparse_bMat_ptr->NonZeroCount);
// No OUTPUTS for analysis
mumps_id.icntl[1-1] = -1;
mumps_id.icntl[2-1] = -1;
mumps_id.icntl[3-1] = -1;
mumps_id.icntl[4-1] = 0;
// strcpy(mumps_id.write_problem,"write_problem");
dmumps_c(&mumps_id);
double lower_nonzeros = (double)mumps_id.infog[20-1];
// if lower_nonzeros is greater than 1.0e+6,
// the value infog[20-1] is lower_nonzeros*(-1)/(1.0e+6).
// we need to adjust the value.
if (lower_nonzeros < 0) {
lower_nonzeros *= (-1.0e+6);
}
#if 0
rMessage("lower_nonzeros = " << lower_nonzeros);
rMessage("Schur density = " << lower_nonzeros/((m+1)*m/2)*100 << "%" );
#endif
if (mumps_id.infog[1-1] != 0) {
rError("MUMPS ERROR " << mumps_id.infog[1-1]);
}
return lower_nonzeros;
}
/* Orders rows and saves pointer to matrix.and model */
int
ClpCholeskyMumps::order(ClpInterior * model)
{
numberRows_ = model->numberRows();
if (doKKT_) {
numberRows_ += numberRows_ + model->numberColumns();
printf("finish coding MUMPS KKT!\n");
abort();
}
rowsDropped_ = new char [numberRows_];
memset(rowsDropped_, 0, numberRows_);
numberRowsDropped_ = 0;
model_ = model;
rowCopy_ = model->clpMatrix()->reverseOrderedCopy();
const CoinBigIndex * columnStart = model_->clpMatrix()->getVectorStarts();
const int * columnLength = model_->clpMatrix()->getVectorLengths();
const int * row = model_->clpMatrix()->getIndices();
const CoinBigIndex * rowStart = rowCopy_->getVectorStarts();
const int * rowLength = rowCopy_->getVectorLengths();
const int * column = rowCopy_->getIndices();
// We need two arrays for counts
int * which = new int [numberRows_];
int * used = new int[numberRows_+1];
CoinZeroN(used, numberRows_);
int iRow;
sizeFactor_ = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
int number = 1;
// make sure diagonal exists
which[0] = iRow;
used[iRow] = 1;
if (!rowsDropped_[iRow]) {
CoinBigIndex startRow = rowStart[iRow];
CoinBigIndex endRow = rowStart[iRow] + rowLength[iRow];
for (CoinBigIndex k = startRow; k < endRow; k++) {
int iColumn = column[k];
CoinBigIndex start = columnStart[iColumn];
CoinBigIndex end = columnStart[iColumn] + columnLength[iColumn];
for (CoinBigIndex j = start; j < end; j++) {
int jRow = row[j];
if (jRow >= iRow && !rowsDropped_[jRow]) {
if (!used[jRow]) {
used[jRow] = 1;
which[number++] = jRow;
}
}
}
}
sizeFactor_ += number;
int j;
for (j = 0; j < number; j++)
used[which[j]] = 0;
}
}
delete [] which;
// NOT COMPRESSED FOR NOW ??? - Space for starts
mumps_->ICNTL(5) = 0; // say NOT compressed format
try {
choleskyStart_ = new CoinBigIndex[numberRows_+1+sizeFactor_];
} catch (...) {
// no memory
return -1;
}
// Now we have size - create arrays and fill in
try {
choleskyRow_ = new int [sizeFactor_];
} catch (...) {
// no memory
delete [] choleskyStart_;
choleskyStart_ = NULL;
return -1;
}
try {
sparseFactor_ = new double[sizeFactor_];
} catch (...) {
// no memory
delete [] choleskyRow_;
choleskyRow_ = NULL;
delete [] choleskyStart_;
choleskyStart_ = NULL;
return -1;
}
sizeFactor_ = 0;
which = choleskyRow_;
for (iRow = 0; iRow < numberRows_; iRow++) {
int number = 1;
// make sure diagonal exists
which[0] = iRow;
used[iRow] = 1;
choleskyStart_[iRow] = sizeFactor_;
if (!rowsDropped_[iRow]) {
CoinBigIndex startRow = rowStart[iRow];
CoinBigIndex endRow = rowStart[iRow] + rowLength[iRow];
for (CoinBigIndex k = startRow; k < endRow; k++) {
int iColumn = column[k];
CoinBigIndex start = columnStart[iColumn];
CoinBigIndex end = columnStart[iColumn] + columnLength[iColumn];
for (CoinBigIndex j = start; j < end; j++) {
int jRow = row[j];
if (jRow >= iRow && !rowsDropped_[jRow]) {
if (!used[jRow]) {
used[jRow] = 1;
which[number++] = jRow;
}
}
}
}
sizeFactor_ += number;
int j;
for (j = 0; j < number; j++)
used[which[j]] = 0;
// Sort
std::sort(which, which + number);
// move which on
which += number;
}
}
choleskyStart_[numberRows_] = sizeFactor_;
delete [] used;
permuteInverse_ = new int [numberRows_];
permute_ = new int[numberRows_];
// To Fortran and fake
for (iRow = 0; iRow < numberRows_ + 1; iRow++) {
int k = choleskyStart_[iRow];
int kEnd = choleskyStart_[iRow+1];
k += numberRows_ + 1;
kEnd += numberRows_ + 1;
for (; k < kEnd; k++)
choleskyStart_[k] = iRow + 1;
choleskyStart_[iRow]++;
}
mumps_->nz = sizeFactor_;
mumps_->irn = choleskyStart_ + numberRows_ + 1;
mumps_->jcn = choleskyRow_;
mumps_->a = NULL;
for (CoinBigIndex i = 0; i < sizeFactor_; i++) {
choleskyRow_[i]++;
#ifndef NDEBUG
assert (mumps_->irn[i] >= 1 && mumps_->irn[i] <= numberRows_);
assert (mumps_->jcn[i] >= 1 && mumps_->jcn[i] <= numberRows_);
#endif
}
// validate
//mumps code here
mumps_->n = numberRows_;
mumps_->nelt = numberRows_;
mumps_->eltptr = choleskyStart_;
mumps_->eltvar = choleskyRow_;
mumps_->a_elt = NULL;
mumps_->rhs = NULL;
mumps_->job = 1; // order
dmumps_c(mumps_);
mumps_->a = sparseFactor_;
if (mumps_->infog[0]) {
COIN_DETAIL_PRINT(printf("MUMPS ordering failed -error %d %d\n",
mumps_->infog[0], mumps_->infog[1]));
return 1;
} else {
double size = mumps_->infog[19];
if (size < 0)
size *= -1000000;
COIN_DETAIL_PRINT(printf("%g nonzeros, flop count %g\n", size, mumps_->rinfog[0]));
}
for (iRow = 0; iRow < numberRows_; iRow++) {
permuteInverse_[iRow] = iRow;
permute_[iRow] = iRow;
}
return 0;
}
ESymSolverStatus MumpsSolverInterface::
DetermineDependentRows(const Index* ia, const Index* ja,
std::list<Index>& c_deps)
{
DBG_START_METH("MumpsSolverInterface::DetermineDependentRows",
dbg_verbosity);
DMUMPS_STRUC_C* mumps_data = (DMUMPS_STRUC_C*)mumps_ptr_;
c_deps.clear();
ESymSolverStatus retval;
// Do the symbolic facotrization if it hasn't been done yet
if (!have_symbolic_factorization_) {
const Index mumps_permuting_scaling_orig = mumps_permuting_scaling_;
const Index mumps_scaling_orig = mumps_scaling_;
mumps_permuting_scaling_ = 0;
mumps_scaling_ = 6;
retval = SymbolicFactorization();
mumps_permuting_scaling_ = mumps_permuting_scaling_orig;
mumps_scaling_ = mumps_scaling_orig;
if (retval != SYMSOLVER_SUCCESS ) {
return retval;
}
have_symbolic_factorization_ = true;
}
// perform the factorization, in order to find dependent rows/columns
//Set flags to ask MUMPS for checking linearly dependent rows
mumps_data->icntl[23] = 1;
mumps_data->cntl[2] = mumps_dep_tol_;
mumps_data->job = 2;//numerical factorization
dump_matrix(mumps_data);
dmumps_c(mumps_data);
int error = mumps_data->info[0];
//Check for errors
if (error == -8 || error == -9) {//not enough memory
const Index trycount_max = 20;
for (int trycount=0; trycount<trycount_max; trycount++) {
Jnlst().Printf(J_WARNING, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) = %d and requires more memory, reallocating. Attempt %d\n",
error,trycount+1);
Jnlst().Printf(J_WARNING, J_LINEAR_ALGEBRA,
" Increasing icntl[13] from %d to ", mumps_data->icntl[13]);
double mem_percent = mumps_data->icntl[13];
mumps_data->icntl[13] = (Index)(2.0 * mem_percent);
Jnlst().Printf(J_WARNING, J_LINEAR_ALGEBRA, "%d.\n", mumps_data->icntl[13]);
dump_matrix(mumps_data);
dmumps_c(mumps_data);
error = mumps_data->info[0];
if (error != -8 && error != -9)
break;
}
if (error == -8 || error == -9) {
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"MUMPS was not able to obtain enough memory.\n");
// Reset flags
mumps_data->icntl[23] = 0;
return SYMSOLVER_FATAL_ERROR;
}
}
// Reset flags
mumps_data->icntl[23] = 0;
if (error < 0) {//some other error
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) =%d MUMPS failure.\n",
error);
return SYMSOLVER_FATAL_ERROR;
}
const Index n_deps = mumps_data->infog[27];
for (Index i=0; i<n_deps; i++) {
c_deps.push_back(mumps_data->pivnul_list[i]-1);
}
return SYMSOLVER_SUCCESS;
}
DMUMPS_STRUC_C* NM_MUMPS_id(NumericsMatrix* A)
{
NumericsSparseLinearSolverParams* params = NM_linearSolverParams(A);
if (!params->solver_data)
{
params->solver_data = malloc(sizeof(DMUMPS_STRUC_C));
DMUMPS_STRUC_C* mumps_id = (DMUMPS_STRUC_C*) params->solver_data;
// Initialize a MUMPS instance. Use MPI_COMM_WORLD.
mumps_id->job = JOB_INIT;
mumps_id->par = 1;
mumps_id->sym = 0;
if (NM_MPI_com(A) == MPI_COMM_WORLD)
{
mumps_id->comm_fortran = USE_COMM_WORLD;
}
else
{
mumps_id->comm_fortran = MPI_Comm_c2f(NM_MPI_com(A));
}
dmumps_c(mumps_id);
if (verbose == 1)
{
mumps_id->ICNTL(1) = -1; // Error messages, standard output stream.
mumps_id->ICNTL(2) = -1; // Diagnostics, standard output stream.
mumps_id->ICNTL(3) = -1; // Global infos, standard output stream.
mumps_id->ICNTL(11) = 1; // Error analysis
}
else if (verbose == 2)
{
mumps_id->ICNTL(1) = -1; // Error messages, standard output stream.
mumps_id->ICNTL(2) = -1; // Diagnostics, standard output stream.
mumps_id->ICNTL(3) = 6; // Global infos, standard output stream.
// mumps_id->ICNTL(4) = 4; // Errors, warnings and information on
// input, output parameters printed.
// mumps_id->ICNTL(10) = 1; // One step of iterative refinment
mumps_id->ICNTL(11) = 1; // Error analysis
}
else if (verbose >= 3)
{
mumps_id->ICNTL(1) = 6; // Error messages, standard output stream.
mumps_id->ICNTL(2) = 6; // Diagnostics, standard output stream.
mumps_id->ICNTL(3) = 6; // Global infos, standard output stream.
// mumps_id->ICNTL(4) = 4; // Errors, warnings and information on
// input, output parameters printed.
// mumps_id->ICNTL(10) = 1; // One step of iterative refinment
mumps_id->ICNTL(11) = 1; // Error analysis
}
else
{
mumps_id->ICNTL(1) = -1;
mumps_id->ICNTL(2) = -1;
mumps_id->ICNTL(3) = -1;
}
mumps_id->ICNTL(24) = 1; // Null pivot row detection see also CNTL(3) & CNTL(5)
// ok for a cube on a plane & four contact points
// computeAlartCurnierSTD != generated in this case...
//mumps_id->CNTL(3) = ...;
//mumps_id->CNTL(5) = ...;
}
DMUMPS_STRUC_C* mumps_id = (DMUMPS_STRUC_C*) params->solver_data;
mumps_id->n = (int) NM_triplet(A)->n;
mumps_id->irn = NM_MUMPS_irn(A);
mumps_id->jcn = NM_MUMPS_jcn(A);
int nz;
if (NM_sparse(A)->triplet)
{
nz = (int) NM_sparse(A)->triplet->nz;
mumps_id->nz = nz;
mumps_id->a = NM_sparse(A)->triplet->x;
}
else
{
nz = NM_linearSolverParams(A)->iWork[2 * NM_csc(A)->nzmax];
mumps_id->nz = nz;
mumps_id->a = NM_sparse(A)->csc->x;
}
return (DMUMPS_STRUC_C*) params->solver_data;
}
int NM_gesv(NumericsMatrix* A, double *b)
{
assert(A->size0 == A->size1);
int info = 1;
switch (A->storageType)
{
case NM_DENSE:
{
assert(A->matrix0);
DGESV(A->size0, 1, A->matrix0, A->size0, NM_iWork(A, A->size0), b,
A->size0, &info);
break;
}
case NM_SPARSE_BLOCK: /* sparse block -> triplet -> csc */
case NM_SPARSE:
{
switch (NM_linearSolverParams(A)->solver)
{
case NS_CS_LUSOL:
info = !cs_lusol(1, NM_csc(A), b, DBL_EPSILON);
break;
#ifdef WITH_MUMPS
case NS_MUMPS:
{
/* the mumps instance is initialized (call with job=-1) */
DMUMPS_STRUC_C* mumps_id = NM_MUMPS_id(A);
mumps_id->rhs = b;
mumps_id->job = 6;
/* compute the solution */
dmumps_c(mumps_id);
/* clean the mumps instance */
mumps_id->job = -2;
dmumps_c(mumps_id);
info = mumps_id->info[0];
if (info > 0)
{
if (verbose > 0)
{
printf("NM_gesv: MUMPS fails : info(1)=%d, info(2)=%d\n", info, mumps_id->info[1]);
}
}
if (verbose > 1)
{
printf("MUMPS : condition number %g\n", mumps_id->rinfog[9]);
printf("MUMPS : component wise scaled residual %g\n", mumps_id->rinfog[6]);
printf("MUMPS : \n");
}
/* Here we free mumps_id ... */
free(NM_linearSolverParams(A)->solver_data);
NM_linearSolverParams(A)->solver_data = NULL;
break;
}
#endif
default:
{
fprintf(stderr, "NM_gesv: unknown sparse linearsolver : %d\n", NM_linearSolverParams(A)->solver);
exit(EXIT_FAILURE);
}
}
break;
}
default:
assert (0 && "NM_gesv unknown storageType");
}
/* some time we cannot find a solution to a linear system, and its fine, for
* instance with the minFBLSA. Therefore, we should not check here for
* problems, but the calling function has to check the return code.*/
// CHECK_RETURN(info);
return info;
}
ESymSolverStatus MumpsSolverInterface::Factorization(
bool check_NegEVals, Index numberOfNegEVals)
{
DBG_START_METH("MumpsSolverInterface::Factorization", dbg_verbosity);
DMUMPS_STRUC_C* mumps_data = (DMUMPS_STRUC_C*)mumps_ptr_;
mumps_data->job = 2;//numerical factorization
dump_matrix(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Calling MUMPS-2 for numerical factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
dmumps_c(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Done with MUMPS-2 for numerical factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
int error = mumps_data->info[0];
//Check for errors
if (error == -8 || error == -9) {//not enough memory
const Index trycount_max = 20;
for (int trycount=0; trycount<trycount_max; trycount++) {
Jnlst().Printf(J_WARNING, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) = %d and requires more memory, reallocating. Attempt %d\n",
error,trycount+1);
Jnlst().Printf(J_WARNING, J_LINEAR_ALGEBRA,
" Increasing icntl[13] from %d to ", mumps_data->icntl[13]);
double mem_percent = mumps_data->icntl[13];
mumps_data->icntl[13] = (Index)(2.0 * mem_percent);
Jnlst().Printf(J_WARNING, J_LINEAR_ALGEBRA, "%d.\n", mumps_data->icntl[13]);
dump_matrix(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Calling MUMPS-2 (repeated) for numerical factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
dmumps_c(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Done with MUMPS-2 (repeated) for numerical factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
error = mumps_data->info[0];
if (error != -8 && error != -9)
break;
}
if (error == -8 || error == -9) {
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"MUMPS was not able to obtain enough memory.\n");
return SYMSOLVER_FATAL_ERROR;
}
}
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"Number of doubles for MUMPS to hold factorization (INFO(9)) = %d\n",
mumps_data->info[8]);
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"Number of integers for MUMPS to hold factorization (INFO(10)) = %d\n",
mumps_data->info[9]);
if (error == -10) {//system is singular
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) = %d matrix is singular.\n",error);
return SYMSOLVER_SINGULAR;
}
negevals_ = mumps_data->infog[11];
if (error == -13) {
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) =%d - out of memory when trying to allocate %d %s.\nIn some cases it helps to decrease the value of the option \"mumps_mem_percent\".\n",
error, mumps_data->info[1] < 0 ? -mumps_data->info[1] : mumps_data->info[1], mumps_data->info[1] < 0 ? "MB" : "bytes");
return SYMSOLVER_FATAL_ERROR;
}
if (error < 0) {//some other error
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) =%d MUMPS failure.\n",
error);
return SYMSOLVER_FATAL_ERROR;
}
if (check_NegEVals && (numberOfNegEVals!=negevals_)) {
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"In MumpsSolverInterface::Factorization: negevals_ = %d, but numberOfNegEVals = %d\n",
negevals_, numberOfNegEVals);
return SYMSOLVER_WRONG_INERTIA;
}
return SYMSOLVER_SUCCESS;
}
ESymSolverStatus MumpsSolverInterface::SymbolicFactorization()
{
DBG_START_METH("MumpsSolverInterface::SymbolicFactorization",
dbg_verbosity);
DMUMPS_STRUC_C* mumps_data = (DMUMPS_STRUC_C*)mumps_ptr_;
if (HaveIpData()) {
IpData().TimingStats().LinearSystemSymbolicFactorization().Start();
}
mumps_data->job = 1;//symbolic ordering pass
//mumps_data->icntl[1] = 6;
//mumps_data->icntl[2] = 6;//QUIETLY!
//mumps_data->icntl[3] = 4;
mumps_data->icntl[5] = mumps_permuting_scaling_;
mumps_data->icntl[6] = mumps_pivot_order_;
mumps_data->icntl[7] = mumps_scaling_;
mumps_data->icntl[9] = 0;//no iterative refinement iterations
mumps_data->icntl[12] = 1;//avoid lapack bug, ensures proper inertia
mumps_data->icntl[13] = mem_percent_; //% memory to allocate over expected
mumps_data->cntl[0] = pivtol_; // Set pivot tolerance
dump_matrix(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Calling MUMPS-1 for symbolic factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
dmumps_c(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Done with MUMPS-1 for symbolic factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
int error = mumps_data->info[0];
const int& mumps_permuting_scaling_used = mumps_data->infog[22];
const int& mumps_pivot_order_used = mumps_data->infog[6];
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"MUMPS used permuting_scaling %d and pivot_order %d.\n",
mumps_permuting_scaling_used, mumps_pivot_order_used);
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
" scaling will be %d.\n",
mumps_data->icntl[7]);
if (HaveIpData()) {
IpData().TimingStats().LinearSystemSymbolicFactorization().End();
}
//return appropriat value
if (error == -6) {//system is singular
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) = %d matrix is singular.\n",error);
return SYMSOLVER_SINGULAR;
}
if (error < 0) {
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"Error=%d returned from MUMPS in Factorization.\n",
error);
return SYMSOLVER_FATAL_ERROR;
}
return SYMSOLVER_SUCCESS;
}
int MAIN__()
{
int argc=1;
char * name = "c_example";
char ** argv ;
#else
int main(int argc, char ** argv)
{
#endif
DMUMPS_STRUC_C id;
int n = 6;
int nz = 21;
int irnL[] = {1,2,3,4,5,6,2,3,4,5,6,3,4,5,6,4,5,6,5,6,6};
int jcnL[] = {1,1,1,1,1,1,2,2,2,2,2,3,3,3,3,4,4,4,5,5,6};
int irnS[] = {1,2,3,2,3,3};
int jcnS[] = {1,1,1,2,2,3};
int *irn = irnL;
int *jcn = jcnL;
double a[21];
double rhs[6];
int myid, ierr, numP;
#if defined(MAIN_COMP)
argv = &name;
#endif
ierr = MPI_Init(&argc, &argv);
ierr = MPI_Comm_rank(MPI_COMM_WORLD, &myid);
ierr = MPI_Comm_size(MPI_COMM_WORLD, &numP);
double nump = numP * 1.0;
/* Define A and rhs */
rhs[0]=2.0;rhs[1]=2.0;rhs[2]=2.0;rhs[3]=2.0;rhs[4]=2.0;rhs[5]=2.0;
a[0]=4169.95/nump;
a[1]=0.0;
a[2]=10075.0/nump;
a[3]=-4030;
a[4]=0.0;
a[5]=0.0;
a[6]=10084.0/nump;
a[7]=-1612.0/nump;
a[8]=0.0;
a[9]=-8.9556;
a[10]=-1612;
a[11]=1354080.0/nump;
a[12]=0.0;
a[13]=1612.0;
a[14]=193440.0;
a[15]=4169.93;
a[16]=0.0;
a[17]=10075;
a[18]=10084;
a[19]=1612;
a[20]=1354000;
if (myid != 0) {
nz = 6;
a[0]=4169.95/nump;
a[1]=0.0;
a[2]=10075.0/nump;
a[3]=10084.0/nump;
a[4]=-1612.0/nump;
a[5]=1354080.0/nump;
irn = irnS;
jcn = jcnS;
}
#define ICNTL(I) icntl[(I)-1] /* macro s.t. indices match documentation */
/* Initialize a MUMPS instance. Use MPI_COMM_WORLD */
id.job=JOB_INIT; id.par=1; id.sym=2; id.comm_fortran=USE_COMM_WORLD;
/* parallel solver; distributed i/p matrix A */
id.ICNTL(5)=0; id.ICNTL(18)=3;
dmumps_c(&id);
/* Define the problem on the host */
/*
id.n = n; id.nz =nz; id.irn=irn; id.jcn=jcn;
id.a = a; id.rhs = rhs;
*/
/* parallel solver; distributed i/p matrix A */
id.ICNTL(5)=0; id.ICNTL(18)=3;
id.n = n; id.nz_loc =nz; id.irn_loc=irn; id.jcn_loc=jcn;
id.a_loc = a; id.rhs = rhs;
/* No outputs */
id.ICNTL(1)=-1; id.ICNTL(2)=-1; id.ICNTL(3)=-1; id.ICNTL(4)=0;
id.job=1;
dmumps_c(&id);
id.job=5;
dmumps_c(&id);
if (myid == 0) {
printf("Solution is : (%8.6e %8.6e %8.6e %8.6e %8.6e %8.6e \n", rhs[0],rhs[1], rhs[2], rhs[3], rhs[4], rhs[5]);
}
rhs[0]=2.0;rhs[1]=2.0;rhs[2]=2.0;rhs[3]=1.0;rhs[4]=1.0;rhs[5]=1.0;
id.job=3;
dmumps_c(&id);
/* Terminate instance */
id.job=JOB_END;
dmumps_c(&id);
if (myid == 0) {
printf("Solution is : (%8.6e %8.6e %8.6e %8.6e %8.6e %8.6e \n", rhs[0],rhs[1], rhs[2], rhs[3], rhs[4], rhs[5]);
}
ierr = MPI_Finalize();
return 0;
}
void operator() ( DMUMPS_STRUC_C& struc ) const {
dmumps_c( &struc ) ;
}
void mexFunction(int nlhs, mxArray *plhs[ ],
int nrhs, const mxArray *prhs[ ]) {
int i,j,pos;
int *ptr_int;
double *ptr_matlab;
#if MUMPS_ARITH == MUMPS_ARITH_z
double *ptri_matlab;
#endif
mwSize tmp_m,tmp_n;
/* C pointer for input parameters */
size_t inst_address;
mwSize n,m,ne, netrue ;
int job;
mwIndex *irn_in,*jcn_in;
/* variable for multiple and sparse rhs */
int posrhs;
mwSize nbrhs,ldrhs, nz_rhs;
mwIndex *irhs_ptr, *irhs_sparse;
double *rhs_sparse;
#if MUMPS_ARITH == MUMPS_ARITH_z
double *im_rhs_sparse;
#endif
DMUMPS_STRUC_C *dmumps_par;
int dosolve = 0;
int donullspace = 0;
int doanalysis = 0;
int dofactorize = 0;
EXTRACT_FROM_MATLAB_TOVAL(JOB,job);
doanalysis = (job == 1 || job == 4 || job == 6);
dofactorize = (job == 2 || job == 4 || job == 5 || job == 6);
dosolve = (job == 3 || job == 5 || job == 6);
if(job == -1){
DMUMPS_alloc(&dmumps_par);
EXTRACT_FROM_MATLAB_TOVAL(SYM,dmumps_par->sym);
dmumps_par->job = -1;
dmumps_par->par = 1;
dmumps_c(dmumps_par);
dmumps_par->nz = -1;
dmumps_par->nz_alloc = -1;
}else{
EXTRACT_FROM_MATLAB_TOVAL(INST,inst_address);
ptr_int = (int *) inst_address;
dmumps_par = (DMUMPS_STRUC_C *) ptr_int;
if(job == -2){
dmumps_par->job = -2;
dmumps_c(dmumps_par);
/* If colsca/rowsca were freed by MUMPS,
dmumps_par->colsca/rowsca are now null.
Application of MYFREE in call below thus ok */
DMUMPS_free(&dmumps_par);
}else{
/* check of input arguments */
n = mxGetN(A_IN);
m = mxGetM(A_IN);
if (!mxIsSparse(A_IN) || n != m )
mexErrMsgTxt("Input matrix must be a sparse square matrix");
jcn_in = mxGetJc(A_IN);
ne = jcn_in[n];
irn_in = mxGetIr(A_IN);
dmumps_par->n = (int)n;
if(dmumps_par->n != n)
mexErrMsgTxt("Input is too big; will not work...barfing out\n");
if(dmumps_par->sym != 0)
netrue = (n+ne)/2;
else
netrue = ne;
if(dmumps_par->nz_alloc < netrue || dmumps_par->nz_alloc >= 2*netrue){
MYFREE(dmumps_par->jcn);
MYFREE(dmumps_par->irn);
MYFREE(dmumps_par->a);
MYMALLOC((dmumps_par->jcn),(int)netrue,int);
MYMALLOC((dmumps_par->irn),(int)netrue,int);
MYMALLOC((dmumps_par->a),(int)netrue,double2);
dmumps_par->nz_alloc = (int)netrue;
if (dmumps_par->nz_alloc != netrue)
mexErrMsgTxt("Input is too big; will not work...barfing out\n");
}
if(dmumps_par->sym == 0){
/* if analysis already performed then we only need to read
numerical values
Note that we suppose that matlab did not change the internal
format of the matrix between the 2 calls */
if(doanalysis){
/* || dmumps_par->info[22] == 0 */
for(i=0;i<dmumps_par->n;i++){
for(j=jcn_in[i];j<jcn_in[i+1];j++){
(dmumps_par->jcn)[j] = i+1;
(dmumps_par->irn)[j] = ((int)irn_in[j])+1;
}
}
}
dmumps_par->nz = (int)ne;
if( dmumps_par->nz != ne)
mexErrMsgTxt("Input is too big; will not work...barfing out\n");
#if MUMPS_ARITH == MUMPS_ARITH_z
ptr_matlab = mxGetPr(A_IN);
for(i=0;i<dmumps_par->nz;i++){
((dmumps_par->a)[i]).r = ptr_matlab[i];
}
ptr_matlab = mxGetPi(A_IN);
if(ptr_matlab){
for(i=0;i<dmumps_par->nz;i++){
((dmumps_par->a)[i]).i = ptr_matlab[i];
}
}else{
for(i=0;i<dmumps_par->nz;i++){
((dmumps_par->a)[i]).i = 0.0;
}
}
#else
ptr_matlab = mxGetPr(A_IN);
for(i=0;i<dmumps_par->nz;i++){
(dmumps_par->a)[i] = ptr_matlab[i];
}
#endif
}else{
/* in the symmetric case we do not need to check doanalysis */
pos = 0;
ptr_matlab = mxGetPr(A_IN);
#if MUMPS_ARITH == MUMPS_ARITH_z
ptri_matlab = mxGetPi(A_IN);
#endif
for(i=0;i<dmumps_par->n;i++){
for(j=jcn_in[i];j<jcn_in[i+1];j++){
if(irn_in[j] >= i){
if(pos >= netrue)
mexErrMsgTxt("Input matrix must be symmetric");
(dmumps_par->jcn)[pos] = i+1;
(dmumps_par->irn)[pos] = (int)irn_in[j]+1;
#if MUMPS_ARITH == MUMPS_ARITH_z
((dmumps_par->a)[pos]).r = ptr_matlab[j];
if(ptri_matlab){
((dmumps_par->a)[pos]).i = ptri_matlab[j];
}else{
((dmumps_par->a)[pos]).i = 0.0;
}
#else
(dmumps_par->a)[pos] = ptr_matlab[j];
#endif
pos++;
}
}
}
dmumps_par->nz = pos;
}
EXTRACT_FROM_MATLAB_TOVAL(JOB,dmumps_par->job);
EXTRACT_FROM_MATLAB_TOARR(ICNTL_IN,dmumps_par->icntl,int,40);
EXTRACT_FROM_MATLAB_TOARR(CNTL_IN,dmumps_par->cntl,double,15);
EXTRACT_FROM_MATLAB_TOPTR(PERM_IN,(dmumps_par->perm_in),int,((int)n));
/* colsca and rowsca are treated differently: it may happen that
dmumps_par-> colsca is nonzero because it was set to a nonzero
value on output (COLSCA_OUT) from MUMPS. Unfortunately if scaling
was on output, one cannot currently provide scaling on input
afterwards without reinitializing the instance */
EXTRACT_SCALING_FROM_MATLAB_TOPTR(COLSCA_IN,(dmumps_par->colsca),(dmumps_par->colsca_from_mumps),((int)n)); /* type always double */
EXTRACT_SCALING_FROM_MATLAB_TOPTR(ROWSCA_IN,(dmumps_par->rowsca),(dmumps_par->rowsca_from_mumps),((int)n)); /* type always double */
EXTRACT_FROM_MATLAB_TOARR(KEEP_IN,dmumps_par->keep,int,500);
EXTRACT_FROM_MATLAB_TOARR(DKEEP_IN,dmumps_par->dkeep,double,230);
dmumps_par->size_schur = (int)mxGetN(VAR_SCHUR);
EXTRACT_FROM_MATLAB_TOPTR(VAR_SCHUR,(dmumps_par->listvar_schur),int,dmumps_par->size_schur);
if(!dmumps_par->listvar_schur) dmumps_par->size_schur = 0;
ptr_matlab = mxGetPr (RHS_IN);
/*
* To follow the "spirit" of the Matlab/Scilab interfaces, treat case of null
* space separately. In that case, we initialize lrhs and nrhs, automatically
* allocate the space needed, and do not rely on what is provided by the user
* in component RHS, that is not touched.
*
* Note that, at the moment, the user should not call the solution step combined
* with the factorization step when he/she sets icntl[25-1] to a non-zero value.
* Hence we suppose in the following that infog[28-1] is available and that we
* can use it.
*
* For users of scilab/matlab, it would still be nice to be able to set ICNTL(25)=-1,
* and use JOB=6. If we want to make such a feature available, we should
* call separately job=2 and job=3 even if job=5 or 6 and set nbrhs (and allocate
* space correctly) between job=2 and job=3 calls to MUMPS.
*
*/
if ( dmumps_par->icntl[25-1] == -1 && dmumps_par->infog[28-1] > 0 ) {
dmumps_par->nrhs=dmumps_par->infog[28-1];
donullspace = dosolve;
}
else if ( dmumps_par->icntl[25-1] > 0 && dmumps_par->icntl[25-1] <= dmumps_par->infog[28-1] ) {
dmumps_par->nrhs=1;
donullspace = dosolve;
}
else {
donullspace=0;
}
if (donullspace) {
nbrhs=dmumps_par->nrhs; ldrhs=n;
dmumps_par->lrhs=(int)n;
MYMALLOC((dmumps_par->rhs),((dmumps_par->n)*(dmumps_par->nrhs)),double2);
}
else if((!dosolve) || ptr_matlab[0] == -9999 ) { /* rhs not already provided, or not used */
/* Case where dosolve is true and ptr_matlab[0]=-9999, this could cause problems:
* 1/ RHS was not initialized while it should have been
* 2/ RHS was explicitely initialized to -9999 but is not allocated of the right size
*/
EXTRACT_CMPLX_FROM_MATLAB_TOPTR(RHS_IN,(dmumps_par->rhs),double,1);
}else{
//=============================================================================
int Amesos_Mumps::SymbolicFactorization()
{
// erase data if present.
if (IsSymbolicFactorizationOK_ && MDS.job != -777)
Destroy();
IsSymbolicFactorizationOK_ = false;
IsNumericFactorizationOK_ = false;
CreateTimer(Comm());
CheckParameters();
AMESOS_CHK_ERR(ConvertToTriplet(false));
#if defined(HAVE_MPI) && defined(HAVE_AMESOS_MPI_C2F)
if (MaxProcs_ != Comm().NumProc())
{
if(MUMPSComm_)
MPI_Comm_free(&MUMPSComm_);
std::vector<int> ProcsInGroup(MaxProcs_);
for (int i = 0 ; i < MaxProcs_ ; ++i)
ProcsInGroup[i] = i;
MPI_Group OrigGroup, MumpsGroup;
MPI_Comm_group(MPI_COMM_WORLD, &OrigGroup);
MPI_Group_incl(OrigGroup, MaxProcs_, &ProcsInGroup[0], &MumpsGroup);
MPI_Comm_create(MPI_COMM_WORLD, MumpsGroup, &MUMPSComm_);
#ifdef MUMPS_4_9
MDS.comm_fortran = (MUMPS_INT) MPI_Comm_c2f( MUMPSComm_);
#else
#ifndef HAVE_AMESOS_OLD_MUMPS
MDS.comm_fortran = (DMUMPS_INT) MPI_Comm_c2f( MUMPSComm_);
#else
MDS.comm_fortran = (F_INT) MPI_Comm_c2f( MUMPSComm_);
#endif
#endif
}
else
{
const Epetra_MpiComm* MpiComm = dynamic_cast<const Epetra_MpiComm*>(&Comm());
assert (MpiComm != 0);
#ifdef MUMPS_4_9
MDS.comm_fortran = (MUMPS_INT) MPI_Comm_c2f(MpiComm->GetMpiComm());
#else
#ifndef HAVE_AMESOS_OLD_MUMPS
MDS.comm_fortran = (DMUMPS_INT) MPI_Comm_c2f(MpiComm->GetMpiComm());
#else
MDS.comm_fortran = (F_INT) MPI_Comm_c2f(MpiComm->GetMpiComm());
#endif
#endif
}
#else
// This next three lines of code were required to make Amesos_Mumps work
// with Ifpack_SubdomainFilter. They is usefull in all cases
// when using MUMPS on a subdomain.
const Epetra_MpiComm* MpiComm = dynamic_cast<const Epetra_MpiComm*>(&Comm());
assert (MpiComm != 0);
MDS.comm_fortran = (MUMPS_INT) MPI_Comm_c2f(MpiComm->GetMpiComm());
// only thing I can do, use MPI_COMM_WORLD. This will work in serial as well
// Previously, the next line was uncommented, but we don't want MUMPS to work
// on the global MPI comm, but on the comm associated with the matrix
// MDS.comm_fortran = -987654;
#endif
MDS.job = -1 ; // Initialization
MDS.par = 1 ; // Host IS involved in computations
// MDS.sym = MatrixProperty_;
MDS.sym = 0; // MatrixProperty_ is unititalized. Furthermore MUMPS
// expects only half of the matrix to be provided for
// symmetric matrices. Hence setting MDS.sym to be non-zero
// indicating that the matrix is symmetric will only work
// if we change ConvertToTriplet to pass only half of the
// matrix. Bug #2331 and Bug #2332 - low priority
RedistrMatrix(true);
if (Comm().MyPID() < MaxProcs_)
{
dmumps_c(&(MDS)); // Initialize MUMPS
static_cast<void>( CheckError( ) );
}
MDS.n = Matrix().NumGlobalRows();
// fix pointers for nonzero pattern of A. Numerical values
// will be entered in PerformNumericalFactorization()
if (Comm().NumProc() != 1)
{
MDS.nz_loc = RedistrMatrix().NumMyNonzeros();
if (Comm().MyPID() < MaxProcs_)
{
MDS.irn_loc = &Row[0];
MDS.jcn_loc = &Col[0];
}
}
else
{
if (Comm().MyPID() == 0)
{
MDS.nz = Matrix().NumMyNonzeros();
MDS.irn = &Row[0];
MDS.jcn = &Col[0];
}
}
// scaling if provided by the user
if (RowSca_ != 0)
{
MDS.rowsca = RowSca_;
MDS.colsca = ColSca_;
}
// given ordering if provided by the user
if (PermIn_ != 0)
{
MDS.perm_in = PermIn_;
}
MDS.job = 1; // Request symbolic factorization
SetICNTLandCNTL();
// Perform symbolic factorization
ResetTimer();
if (Comm().MyPID() < MaxProcs_)
dmumps_c(&(MDS));
SymFactTime_ = AddTime("Total symbolic factorization time", SymFactTime_);
int IntWrong = CheckError()?1:0 ;
int AnyWrong;
Comm().SumAll( &IntWrong, &AnyWrong, 1 ) ;
bool Wrong = AnyWrong > 0 ;
if ( Wrong ) {
AMESOS_CHK_ERR( StructurallySingularMatrixError ) ;
}
IsSymbolicFactorizationOK_ = true ;
NumSymbolicFact_++;
return 0;
}
/* Factorize - filling in rowsDropped and returning number dropped */
int
ClpCholeskyMumps::factorize(const double * diagonal, int * rowsDropped)
{
const CoinBigIndex * columnStart = model_->clpMatrix()->getVectorStarts();
const int * columnLength = model_->clpMatrix()->getVectorLengths();
const int * row = model_->clpMatrix()->getIndices();
const double * element = model_->clpMatrix()->getElements();
const CoinBigIndex * rowStart = rowCopy_->getVectorStarts();
const int * rowLength = rowCopy_->getVectorLengths();
const int * column = rowCopy_->getIndices();
const double * elementByRow = rowCopy_->getElements();
int numberColumns = model_->clpMatrix()->getNumCols();
int iRow;
double * work = new double[numberRows_];
CoinZeroN(work, numberRows_);
const double * diagonalSlack = diagonal + numberColumns;
int newDropped = 0;
double largest;
//double smallest;
//perturbation
double perturbation = model_->diagonalPerturbation() * model_->diagonalNorm();
perturbation = 0.0;
perturbation = perturbation * perturbation;
if (perturbation > 1.0) {
#ifdef COIN_DEVELOP
//if (model_->model()->logLevel()&4)
std::cout << "large perturbation " << perturbation << std::endl;
#endif
perturbation = sqrt(perturbation);;
perturbation = 1.0;
}
double delta2 = model_->delta(); // add delta*delta to diagonal
delta2 *= delta2;
for (iRow = 0; iRow < numberRows_; iRow++) {
double * put = sparseFactor_ + choleskyStart_[iRow] - 1; // Fortran
int * which = choleskyRow_ + choleskyStart_[iRow] - 1; // Fortran
int number = choleskyStart_[iRow+1] - choleskyStart_[iRow];
if (!rowLength[iRow])
rowsDropped_[iRow] = 1;
if (!rowsDropped_[iRow]) {
CoinBigIndex startRow = rowStart[iRow];
CoinBigIndex endRow = rowStart[iRow] + rowLength[iRow];
work[iRow] = diagonalSlack[iRow] + delta2;
for (CoinBigIndex k = startRow; k < endRow; k++) {
int iColumn = column[k];
if (!whichDense_ || !whichDense_[iColumn]) {
CoinBigIndex start = columnStart[iColumn];
CoinBigIndex end = columnStart[iColumn] + columnLength[iColumn];
double multiplier = diagonal[iColumn] * elementByRow[k];
for (CoinBigIndex j = start; j < end; j++) {
int jRow = row[j];
if (jRow >= iRow && !rowsDropped_[jRow]) {
double value = element[j] * multiplier;
work[jRow] += value;
}
}
}
}
int j;
for (j = 0; j < number; j++) {
int jRow = which[j] - 1; // to Fortran
put[j] = work[jRow];
work[jRow] = 0.0;
}
} else {
// dropped
int j;
for (j = 1; j < number; j++) {
put[j] = 0.0;
}
put[0] = 1.0;
}
}
//check sizes
double largest2 = maximumAbsElement(sparseFactor_, sizeFactor_);
largest2 *= 1.0e-20;
largest = CoinMin(largest2, 1.0e-11);
int numberDroppedBefore = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
int dropped = rowsDropped_[iRow];
// Move to int array
rowsDropped[iRow] = dropped;
if (!dropped) {
CoinBigIndex start = choleskyStart_[iRow] - 1; // to Fortran
double diagonal = sparseFactor_[start];
if (diagonal > largest2) {
sparseFactor_[start] = CoinMax(diagonal, 1.0e-10);
} else {
sparseFactor_[start] = CoinMax(diagonal, 1.0e-10);
rowsDropped[iRow] = 2;
numberDroppedBefore++;
}
}
}
delete [] work;
// code here
mumps_->a_elt = sparseFactor_;
mumps_->rhs = NULL;
mumps_->job = 2; // factorize
dmumps_c(mumps_);
if (mumps_->infog[0]) {
COIN_DETAIL_PRINT(printf("MUMPS factorization failed -error %d %d\n",
mumps_->infog[0], mumps_->infog[1]));
}
choleskyCondition_ = 1.0;
bool cleanCholesky;
if (model_->numberIterations() < 2000)
cleanCholesky = true;
else
cleanCholesky = false;
if (cleanCholesky) {
//drop fresh makes some formADAT easier
//int oldDropped=numberRowsDropped_;
if (newDropped || numberRowsDropped_) {
//std::cout <<"Rank "<<numberRows_-newDropped<<" ( "<<
// newDropped<<" dropped)";
//if (newDropped>oldDropped)
//std::cout<<" ( "<<newDropped-oldDropped<<" dropped this time)";
//std::cout<<std::endl;
newDropped = 0;
for (int i = 0; i < numberRows_; i++) {
int dropped = rowsDropped[i];
rowsDropped_[i] = (char)dropped;
if (dropped == 2) {
//dropped this time
rowsDropped[newDropped++] = i;
rowsDropped_[i] = 0;
}
}
numberRowsDropped_ = newDropped;
newDropped = -(2 + newDropped);
}
} else {
if (newDropped) {
newDropped = 0;
for (int i = 0; i < numberRows_; i++) {
int dropped = rowsDropped[i];
rowsDropped_[i] = (char)dropped;
if (dropped == 2) {
//dropped this time
rowsDropped[newDropped++] = i;
rowsDropped_[i] = 1;
}
}
}
numberRowsDropped_ += newDropped;
if (numberRowsDropped_ && 0) {
std::cout << "Rank " << numberRows_ - numberRowsDropped_ << " ( " <<
numberRowsDropped_ << " dropped)";
if (newDropped) {
std::cout << " ( " << newDropped << " dropped this time)";
}
std::cout << std::endl;
}
}
status_ = 0;
return newDropped;
}
int Amesos_Mumps::Solve()
{
if (IsNumericFactorizationOK_ == false)
AMESOS_CHK_ERR(NumericFactorization());
Epetra_MultiVector* vecX = Problem_->GetLHS();
Epetra_MultiVector* vecB = Problem_->GetRHS();
int NumVectors = vecX->NumVectors();
if ((vecX == 0) || (vecB == 0))
AMESOS_CHK_ERR(-1);
if (NumVectors != vecB->NumVectors())
AMESOS_CHK_ERR(-1);
if (Comm().NumProc() == 1)
{
// do not import any data
for (int j = 0 ; j < NumVectors; j++)
{
ResetTimer();
MDS.job = 3; // Request solve
for (int i = 0 ; i < Matrix().NumMyRows() ; ++i)
(*vecX)[j][i] = (*vecB)[j][i];
MDS.rhs = (*vecX)[j];
dmumps_c(&(MDS)) ; // Perform solve
static_cast<void>( CheckError( ) ); // Can hang
SolveTime_ = AddTime("Total solve time", SolveTime_);
}
}
else
{
Epetra_MultiVector SerialVector(SerialMap(),NumVectors);
ResetTimer();
AMESOS_CHK_ERR(SerialVector.Import(*vecB,SerialImporter(),Insert));
VecRedistTime_ = AddTime("Total vector redistribution time", VecRedistTime_);
for (int j = 0 ; j < NumVectors; j++)
{
if (Comm().MyPID() == 0)
MDS.rhs = SerialVector[j];
// solve the linear system and take time
MDS.job = 3;
ResetTimer();
if (Comm().MyPID() < MaxProcs_)
dmumps_c(&(MDS)) ; // Perform solve
static_cast<void>( CheckError( ) ); // Can hang
SolveTime_ = AddTime("Total solve time", SolveTime_);
}
// ship solution back and take timing
ResetTimer();
AMESOS_CHK_ERR(vecX->Export(SerialVector,SerialImporter(),Insert));
VecRedistTime_ = AddTime("Total vector redistribution time", VecRedistTime_);
}
if (ComputeTrueResidual_)
ComputeTrueResidual(Matrix(), *vecX, *vecB, UseTranspose(), "Amesos_Mumps");
if (ComputeVectorNorms_)
ComputeVectorNorms(*vecX, *vecB, "Amesos_Mumps");
NumSolve_++;
return(0) ;
}
