DllExport int NumericFactor( struct SymbolicSolver_tag *sp, int n, int nnz, int *rowIndex, int *colIndex, double *value )
{
int ret = -1;
struct SymbolicSolver_tag * s = (struct SymbolicSolver_tag *) sp;
taucs_ccs_matrix * m = taucs_ccs_create(n, n, nnz, TAUCS_DOUBLE|TAUCS_LOWER|TAUCS_SYMMETRIC);
if (m == NULL) return -1;
// copy elements to matrix
memcpy(m->rowind, rowIndex, sizeof(int) * nnz);
memcpy(m->values.d, value, sizeof(double) * nnz);
memcpy(m->colptr, colIndex, sizeof(int) * (n+1));
m = taucs_ccs_permute_symmetrically(m, s->perm, s->invperm);
ret = taucs_ccs_factor_llt_numeric(m, s->factorization);
taucs_ccs_free(m);
return ret;
}
// Will create *symbolic* factorization of ATA and store it aside
// returns the id of the factor stored, to be used later to factor
// other matrices with the same zero structure
int SymbolicFactorATA(const int matrixId) {
if (matrixId >= (int)matrixArray.size() || matrixId < 0)
return -1;
// find a place for the new factor entry in the array
int fId;
for (fId = 0; fId < (int)factorArray.size(); fId++) {
if (factorArray[fId].SL == NULL) {
break;
}
}
if (fId == (int)factorArray.size()) {
factorArray.push_back(FactorEntry());
}
FactorEntry & newFactor = factorArray[fId];
// get the ATA matrix
taucs_ccs_matrix * ATA = matrixArray[matrixId]->GetATA_Copy();
if (ATA == NULL)
return -1;
// make all non-zero entries 1
for (int col = 0; col < ATA->n; ++col) {
for (int p = ATA->colptr[col]; p < ATA->colptr[col+1]; ++p) {
ATA->taucs_values[p] = 1;
}
}
// create symbolic factor and permutations:
// compute re-ordering
taucs_ccs_order(ATA, &newFactor.perm, &newFactor.invperm, "metis");
newFactor.PAP = taucs_ccs_permute_symmetrically(ATA, newFactor.perm, newFactor.invperm);
if (!newFactor.PAP) {
return -1;
}
// symbolic Cholesky factorization only
newFactor.SL = taucs_ccs_factor_llt_symbolic(newFactor.PAP);
return fId;
}
//
// functions for symbolic solver
//
DllExport struct SymbolicSolver_tag * CreaterSymbolicSolver(int n, int nnz, int *rowIndex, int *colIndex, double *value)
{
struct SymbolicSolver_tag * s = (struct SymbolicSolver_tag*) malloc(sizeof(struct SymbolicSolver_tag));
s->n = n;
s->matrix = taucs_ccs_create(n, n, nnz, TAUCS_DOUBLE|TAUCS_LOWER|TAUCS_SYMMETRIC);
s->factorization = NULL;
s->perm = (int*) malloc(sizeof(int) * n);
s->invperm = (int*) malloc(sizeof(int) * n);
s->tmp_b = (double*) malloc(sizeof(double) * n);
s->tmp_x = (double*) malloc(sizeof(double) * n);
if (s->matrix == NULL) return NULL;
if (s->perm == NULL) return NULL;
memcpy(s->matrix->colptr, colIndex, sizeof(int)*(n+1));
memcpy(s->matrix->rowind, rowIndex, sizeof(int)*nnz);
memcpy(s->matrix->values.d, value, sizeof(double)*nnz);
taucs_ccs_order(s->matrix, &s->perm, &s->invperm, "metis");
s->matrix = taucs_ccs_permute_symmetrically(s->matrix, s->perm, s->invperm);
s->factorization = taucs_ccs_factor_llt_symbolic(s->matrix);
return s;
}
/* Factorize - filling in rowsDropped and returning number dropped */
int
ClpCholeskyTaucs::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 = rowCopyT_->getVectorStarts();
const int * rowLength = rowCopyT_->getVectorLengths();
const int * column = rowCopyT_->getIndices();
const double * elementByRow = rowCopyT_->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;
//perturbation
double perturbation = model_->diagonalPerturbation() * model_->diagonalNorm();
perturbation = perturbation * perturbation;
if (perturbation > 1.0) {
//if (model_->model()->logLevel()&4)
std::cout << "large perturbation " << perturbation << std::endl;
perturbation = sqrt(perturbation);
perturbation = 1.0;
}
for (iRow = 0; iRow < numberRows_; iRow++) {
double * put = sparseFactorT_ + choleskyStartT_[iRow];
int * which = choleskyRowT_ + choleskyStartT_[iRow];
int number = choleskyStartT_[iRow+1] - choleskyStartT_[iRow];
if (!rowLength[iRow])
rowsDropped_[iRow] = 1;
if (!rowsDropped_[iRow]) {
CoinBigIndex startRow = rowStart[iRow];
CoinBigIndex endRow = rowStart[iRow] + rowLength[iRow];
work[iRow] = diagonalSlack[iRow];
for (CoinBigIndex k = startRow; k < endRow; k++) {
int iColumn = column[k];
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];
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(sparseFactorT_, sizeFactorT_);
largest2 *= 1.0e-19;
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 = choleskyStartT_[iRow];
double diagonal = sparseFactorT_[start];
if (diagonal > largest2) {
sparseFactorT_[start] = diagonal + perturbation;
} else {
sparseFactorT_[start] = diagonal + perturbation;
rowsDropped[iRow] = 2;
numberDroppedBefore++;
}
}
}
taucs_supernodal_factor_free_numeric(factorization_);
// need to permute
taucs_ccs_matrix * permuted = taucs_ccs_permute_symmetrically(matrix_, permuteInverse_, permute_);
int rCode = taucs_ccs_factor_llt_numeric(permuted, factorization_);
taucs_ccs_free(permuted);
if (rCode)
printf("return code of %d from factor\n", rCode);
delete [] work;
choleskyCondition_ = 1.0;
bool cleanCholesky;
if (model_->numberIterations() < 200)
cleanCholesky = true;
else
cleanCholesky = false;
/*
How do I find out where 1.0e100's are in cholesky?
*/
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++) {
char dropped = rowsDropped[i];
rowsDropped_[i] = dropped;
if (dropped == 2) {
//dropped this time
rowsDropped[newDropped++] = i;
rowsDropped_[i] = 0;
}
}
numberRowsDropped_ = newDropped;
newDropped = -(1 + newDropped);
}
} else {
if (newDropped) {
newDropped = 0;
for (int i = 0; i < numberRows_; i++) {
char dropped = rowsDropped[i];
int oldDropped = rowsDropped_[i];
rowsDropped_[i] = dropped;
if (dropped == 2) {
assert (!oldDropped);
//dropped this time
rowsDropped[newDropped++] = i;
rowsDropped_[i] = 1;
}
}
}
numberRowsDropped_ += newDropped;
if (numberRowsDropped_) {
std::cout << "Rank " << numberRows_ - numberRowsDropped_ << " ( " <<
numberRowsDropped_ << " dropped)";
if (newDropped) {
std::cout << " ( " << newDropped << " dropped this time)";
}
std::cout << std::endl;
}
}
status_ = 0;
return newDropped;
}
/* Orders rows and saves pointer to matrix.and model */
int
ClpCholeskyTaucs::order(ClpInterior * model)
{
numberRows_ = model->numberRows();
rowsDropped_ = new char [numberRows_];
memset(rowsDropped_, 0, numberRows_);
numberRowsDropped_ = 0;
model_ = model;
rowCopyT_ = model->clpMatrix()->reverseOrderedCopy();
const CoinBigIndex * columnStart = model_->clpMatrix()->getVectorStarts();
const int * columnLength = model_->clpMatrix()->getVectorLengths();
const int * row = model_->clpMatrix()->getIndices();
const CoinBigIndex * rowStart = rowCopyT_->getVectorStarts();
const int * rowLength = rowCopyT_->getVectorLengths();
const int * column = rowCopyT_->getIndices();
// We need two arrays for counts
int * which = new int [numberRows_];
int * used = new int[numberRows_];
CoinZeroN(used, numberRows_);
int iRow;
sizeFactorT_ = 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;
}
}
}
}
sizeFactorT_ += number;
int j;
for (j = 0; j < number; j++)
used[which[j]] = 0;
}
}
delete [] which;
// Now we have size - create arrays and fill in
matrix_ = taucs_ccs_create(numberRows_, numberRows_, sizeFactorT_,
TAUCS_DOUBLE | TAUCS_SYMMETRIC | TAUCS_LOWER);
if (!matrix_)
return 1;
// Space for starts
choleskyStartT_ = matrix_->colptr;
choleskyRowT_ = matrix_->rowind;
sparseFactorT_ = matrix_->values.d;
sizeFactorT_ = 0;
which = choleskyRowT_;
for (iRow = 0; iRow < numberRows_; iRow++) {
int number = 1;
// make sure diagonal exists
which[0] = iRow;
used[iRow] = 1;
choleskyStartT_[iRow] = sizeFactorT_;
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;
}
}
}
}
sizeFactorT_ += number;
int j;
for (j = 0; j < number; j++)
used[which[j]] = 0;
// Sort
std::sort(which, which + number);
// move which on
which += number;
}
}
choleskyStartT_[numberRows_] = sizeFactorT_;
delete [] used;
permuteInverse_ = new int [numberRows_];
permute_ = new int[numberRows_];
int * perm, *invp;
// There seem to be bugs in ordering if model too small
if (numberRows_ > 10)
taucs_ccs_order(matrix_, &perm, &invp, (const char *) "genmmd");
else
taucs_ccs_order(matrix_, &perm, &invp, (const char *) "identity");
CoinMemcpyN(perm, numberRows_, permuteInverse_);
free(perm);
CoinMemcpyN(invp, numberRows_, permute_);
free(invp);
// need to permute
taucs_ccs_matrix * permuted = taucs_ccs_permute_symmetrically(matrix_, permuteInverse_, permute_);
// symbolic
factorization_ = taucs_ccs_factor_llt_symbolic(permuted);
taucs_ccs_free(permuted);
return factorization_ ? 0 : 1;
}
int sci_taucs_chfact(char* fname, void* pvApiCtx)
{
SciErr sciErr;
int stat = 0;
int* perm = NULL;
int* invperm = NULL;
taucs_ccs_matrix *PAPT;
taucs_ccs_matrix B;
void *C = NULL;
taucs_handle_factors *pC;
SciSparse A;
int mA = 0; // rows
int nA = 0; // cols
int iNbItem = 0;
int* piNbItemRow = NULL;
int* piColPos = NULL;
double* pdblSpReal = NULL;
double* pdblSpImg = NULL;
int iComplex = 0;
int* piAddr1 = NULL;
/* Check numbers of input/output arguments */
CheckInputArgument(pvApiCtx, 1, 1);
CheckOutputArgument(pvApiCtx, 1, 1);
/* get A the sparse matrix to factorize */
sciErr = getVarAddressFromPosition(pvApiCtx, 1, &piAddr1);
if (sciErr.iErr)
{
printError(&sciErr, 0);
return 1;
}
if (isVarComplex(pvApiCtx, piAddr1))
{
iComplex = 1;
sciErr = getComplexSparseMatrix(pvApiCtx, piAddr1, &mA, &nA, &iNbItem, &piNbItemRow, &piColPos, &pdblSpReal, &pdblSpImg);
}
else
{
sciErr = getSparseMatrix(pvApiCtx, piAddr1, &mA, &nA, &iNbItem, &piNbItemRow, &piColPos, &pdblSpReal);
}
if (sciErr.iErr)
{
printError(&sciErr, 0);
return 1;
}
// fill struct sparse
A.m = mA;
A.n = nA;
A.it = iComplex;
A.nel = iNbItem;
A.mnel = piNbItemRow;
A.icol = piColPos;
A.R = pdblSpReal;
A.I = pdblSpImg;
stat = spd_sci_sparse_to_taucs_sparse(&A, &B);
if ( stat != A_PRIORI_OK )
{
if ( stat == MAT_IS_NOT_SPD )
{
freeTaucsSparse(B);
Scierror(999, _("%s: Wrong value for input argument #%d: Must be symmetric positive definite matrix."), fname, 1);
}
/* the message for the other problem (not enough memory in stk) is treated automaticaly */
return 1;
}
/* find the permutation */
taucs_ccs_genmmd(&B, &perm, &invperm);
if ( !perm )
{
freeTaucsSparse(B);
Scierror(999, _("%s: No more memory.\n") , fname);
return 1;
}
/* apply permutation */
PAPT = taucs_ccs_permute_symmetrically(&B, perm, invperm);
FREE(invperm);
freeTaucsSparse(B);
/* factor */
C = taucs_ccs_factor_llt_mf(PAPT);
taucs_ccs_free(PAPT);
if (C == NULL)
{
/* Note : an error indicator is given in the main scilab window
* (out of memory, no positive definite matrix , etc ...)
*/
Scierror(999, _("%s: An error occurred: %s\n"), fname, _("factorization"));
return 1;
}
/* put in an handle (Chol fact + perm + size) */
pC = (taucs_handle_factors*)MALLOC( sizeof(taucs_handle_factors) );
pC->p = perm;
pC->C = C;
pC->n = A.n;
/* add in the list of Chol Factors */
AddAdrToList((Adr) pC, 0, &ListCholFactors); /* FIXME add a test here .. */
/* create the scilab object to store the pointer onto the Chol handle */
sciErr = createPointer(pvApiCtx, 2, (void *)pC);
if (sciErr.iErr)
{
printError(&sciErr, 0);
return 1;
}
/* return the pointer */
AssignOutputVariable(pvApiCtx, 1) = 2;
ReturnArguments(pvApiCtx);
return 0;
}
int actual_main(int argc, char* argv[])
{
double wtime_order;
double wtime_permute;
double wtime_factor;
double wtime_solve;
/*double wtime_precond_create; omer*/
double ctime_factor;
int i;/*,j omer*/
double NormErr;
taucs_ccs_matrix* A = NULL;
taucs_ccs_matrix* PAPT = NULL;
taucs_ccs_matrix* L = NULL;
double* Xd = NULL;
double* Bd = NULL;
double* PXd = NULL;
double* PBd = NULL;
double* NXd = NULL;
float* Xs = NULL;
float* Bs = NULL;
float* PXs = NULL;
float* PBs = NULL;
float* NXs = NULL;
taucs_dcomplex* Xz = NULL;
taucs_dcomplex* Bz = NULL;
taucs_dcomplex* PXz = NULL;
taucs_dcomplex* PBz = NULL;
taucs_dcomplex* NXz = NULL;
char* ordering = "metis";
char* mat_type = "neumann";
int* perm;
int* invperm;
int precision = TAUCS_DOUBLE;
int ldlt_flag = 0;
int snmf_flag = 0;
int snll_flag = 0;
int symb_flag = 0;
int mesh2d_flag = 0,mesh2d_size = 0;
int ooc_flag = 0;
int panelize = 0;
double memory_mb = -1.0;
char* matrixfile = "/tmp/taucs.L";
taucs_io_handle* oocL = NULL;
int (*precond_fn)(void*,void* x,void* b);
void* precond_args;
/***********************************************************/
/* Read arguments: log file, matrix, memory size, ooc name */
/***********************************************************/
if ((argc == 1) ||((argc == 2) && !strncmp(argv[1],"-h",2)))
usage(argc,argv);
A = NULL;
for (i=1; i<argc; i++) {
if (!strcmp(argv[i],"-single")) precision = TAUCS_SINGLE;
if (!strcmp(argv[i],"-dcomplex")) precision = TAUCS_DCOMPLEX;
if (!strcmp(argv[i],"-ldlt")) ldlt_flag = 1;
if (!strcmp(argv[i],"-snmf")) snmf_flag = 1;
if (!strcmp(argv[i],"-snll")) snll_flag = 1;
if (!strcmp(argv[i],"-symb")) symb_flag = 1;
if (!strcmp(argv[i],"-ooc")) ooc_flag = 1;
if (!strcmp(argv[i],"-log") && i <= argc-1 ) {
i++;
taucs_logfile(argv[i]);
}
if (!strcmp(argv[i],"-panelize") && i <= argc-1) {
i++;
if (sscanf(argv[i],"%d",&panelize) != 1) {
taucs_printf("0 (smart), 1 (in-core), or 2 (single supernode) follow -panelize argument\n");
exit(1);
}
}
if (!strcmp(argv[i],"-memory") && i <= argc-1) {
i++;
if (sscanf(argv[i],"%lf",&memory_mb) != 1) {
taucs_printf("memory size in MB must follow -memory argument\n");
exit(1);
}
}
if (!strcmp(argv[i],"-matrixfile") && i <= argc-1 ) {
i++;
matrixfile = argv[i];
}
if (!strcmp(argv[i],"-ordering") && i <= argc-1 ) {
i++;
ordering = argv[i];
}
if (!strcmp(argv[i],"-mat_type") && i <= argc-1 ) {
i++;
mat_type = argv[i];
}
#if 0
if (!strcmp(argv[i],"-hb") && i <= argc-1 ) {
int nrows,ncols,nnz,j;
char fname[256];
char type[3];
i++;
for (j=0; j<256; j++) fname[j] = ' ';
strcpy(fname,argv[i]);
taucs_printf("main: reading HB matrix %s\n",argv[i]);
ireadhb_(fname,type,&nrows,&ncols,&nnz);
A = taucs_dccs_creagte(nrows,ncols,nnz);
if (type[1] == 's' || type[1] == 'S')
A->flags |= TAUCS_SYMMETRIC | TAUCS_LOWER;
dreadhb_(fname,&nrows,&ncols,&nnz,
A->colptr,A->rowind,A->values);
/* make indices 0-based */
for (j=0; j<=ncols; j++) ((A->colptr)[j])--;
for (j=0; j<nnz; j++) ((A->rowind)[j])--;
taucs_printf("main: done reading\n");
}
#endif
if (!strcmp(argv[i],"-hb") && i <= argc-1) {
i++;
taucs_printf("main: reading hb matrix %s\n",argv[i]);
switch (precision) {
case TAUCS_SINGLE:
A = taucs_ccs_read_hb (argv[i], TAUCS_SINGLE); break;
case TAUCS_DOUBLE:
A = taucs_ccs_read_hb (argv[i], TAUCS_DOUBLE); break;
case TAUCS_DCOMPLEX:
A = taucs_ccs_read_hb (argv[i], TAUCS_DCOMPLEX); break;
default:
taucs_printf("main: unknown precision\n");
exit(1);
}
taucs_printf("main: done reading\n");
}
if (!strcmp(argv[i],"-mtx") && i <= argc-1) {
i++;
taucs_printf("main: reading mtx matrix %s\n",argv[i]);
A = taucs_ccs_read_mtx (argv[i],TAUCS_SYMMETRIC | TAUCS_PATTERN);
taucs_printf("main: done reading\n");
}
if (!strcmp(argv[i],"-ijv") && i <= argc-1) {
printf(">>> ijv\n");
i++;
taucs_printf("main: reading ijv matrix %s\n",argv[i]);
switch (precision) {
case TAUCS_SINGLE:
A = taucs_ccs_read_ijv (argv[i],TAUCS_SYMMETRIC | TAUCS_SINGLE); break;
case TAUCS_DOUBLE:
A = taucs_ccs_read_ijv (argv[i],TAUCS_SYMMETRIC | TAUCS_DOUBLE); break;
case TAUCS_DCOMPLEX:
A = taucs_ccs_read_ijv (argv[i],TAUCS_HERMITIAN | TAUCS_DCOMPLEX); break;
default:
taucs_printf("main: unknown precision\n");
exit(1);
}
taucs_printf("main: done reading\n");
}
if (!strcmp(argv[i],"-ccs") && i <= argc-1) {
i++;
taucs_printf("main: reading ccs matrix %s\n",argv[i]);
A = taucs_ccs_read_ccs (argv[i],TAUCS_SYMMETRIC);
taucs_printf("main: done reading\n");
}
if (!strcmp(argv[i],"-mesh2d") && i <= argc-1) {
mesh2d_flag = 1;
taucs_printf("A is a mesh2d\n");
i++;
if (sscanf(argv[i],"%d",&mesh2d_size) != 1) {
taucs_printf("mesh size (n, where the mesh is n-by-n) must follow -mesh2d argument\n");
exit(1);
}
}
if (!strcmp(argv[i],"-mesh3d") && i <= argc-3) {
int X,Y,Z;
taucs_printf("A is a mesh3d\n");
if (sscanf(argv[i+1],"%d",&X) != 1
|| sscanf(argv[i+2],"%d",&Y) != 1
|| sscanf(argv[i+3],"%d",&Z) != 1) {
taucs_printf("mesh size (X Y Z must follow -mesh3d argument\n");
exit(1);
}
i += 3;
taucs_printf("main: creating mesh\n");
A = taucs_ccs_generate_mesh3d(X,Y,Z);
}
if (!strcmp(argv[i],"-n+rhs") && i <= argc-1) {
FILE* f;
int n,j,nnz;
i++;
taucs_printf("main: reading right-hand side %s\n",argv[i]);
f=fopen(argv[i],"r");
assert(f);
fscanf(f,"%d",&n);
Bd=(double*) malloc(n*sizeof(double));
nnz = 0;
for (j=0; j<n; j++) {
fscanf(f,"%lg",(Bd)+j);
if (Bd[j]) nnz++;
}
fclose(f);
taucs_printf("main: done reading rhs, %d nonzeros\n",nnz);
}
}
taucs_printf("Chosen Ordering: %s\n",ordering);
if (mesh2d_flag)
{
taucs_printf("Matrix type is %s\n",mat_type);
taucs_printf("Grid Size is %d\n",mesh2d_size);
A = taucs_ccs_generate_mesh2d(mesh2d_size,mat_type);
}
if (!A) {
taucs_printf("matrix argument not given or matrix file not found\n");
usage(argc,argv);
}
N = M = A->n;
/*taucs_maximize_stacksize();*/
/***********************************************************/
/* Create exact solution, compute right-hand-side */
/***********************************************************/
if (A->flags & TAUCS_SINGLE) {
if (! (Xs)) {
Xs = (float*)malloc(N*sizeof(float));
/*for(i=0; i<N; i++) (Xs)[i]=(double)random()/RAND_MAX; omer*/
for(i=0; i<N; i++) (Xs)[i]=(float)((double)rand()/RAND_MAX);
} else
taucs_printf("iter: not using a random X, already allocated\n");
if (!(Bs)) {
Bs = (float*)malloc(N*sizeof(float));
taucs_ccs_times_vec(A,Xs,Bs);
} else {
/*double zero1 = 0.0;
double nan = zero1 / zero1; omer*/
double nan = taucs_get_nan();
for(i=0; i<N; i++) Xs[i]= (float)nan;
}
NXs=(float*)malloc(N*sizeof(float));
PXs=(float*)malloc(N*sizeof(float));
PBs=(float*)malloc(N*sizeof(float));
}
if (A->flags & TAUCS_DOUBLE) {
if (! (Xd)) {
Xd =(double*)malloc(N*sizeof(double));
/*for(i=0; i<N; i++) (Xd)[i]=(double)rand()/RAND_MAX; omer*/
for(i=0; i<N; i++) (Xd)[i]=(float)((double)rand()/RAND_MAX);
} else
taucs_printf("iter: not using a random X, already allocated\n");
if (!(Bd)) {
Bd =(double*)malloc(N*sizeof(double));
taucs_ccs_times_vec(A,Xd,Bd);
} else {
/*double zero1 = 0.0;
double nan = zero1 / zero1; omer*/
double nan = taucs_get_nan();
for(i=0; i<N; i++) Xd[i]= (float)nan;
}
NXd=(double*)malloc(N*sizeof(double));
PXd=(double*)malloc(N*sizeof(double));
PBd=(double*)malloc(N*sizeof(double));
}
if (A->flags & TAUCS_DCOMPLEX) {
if (!(Xz)) {
double* p;
taucs_printf("direct: creating a random dcomplex X\n");
Xz =(taucs_dcomplex*)malloc(N*sizeof(taucs_dcomplex));
p = (double*) Xz;
for(i=0; i<2*N; i++) p[i] = (double)rand()/RAND_MAX;
} else
taucs_printf("iter: not using a random X, already allocated\n");
if (!(Bz)) {
Bz =(taucs_dcomplex*)malloc(N*sizeof(taucs_dcomplex));
taucs_ccs_times_vec(A,Xz,Bz);
} else {
double* p;
/*double zero1 = 0.0;
double nan = zero1 / zero1; omer*/
double nan = taucs_get_nan();
p = (double*) Xz;
for(i=0; i<2*N; i++) p[i] = nan;
}
NXz=(taucs_dcomplex*)malloc(N*sizeof(taucs_dcomplex));
PXz=(taucs_dcomplex*)malloc(N*sizeof(taucs_dcomplex));
PBz=(taucs_dcomplex*)malloc(N*sizeof(taucs_dcomplex));
}
/***********************************************************/
/* Compute column ordering */
/***********************************************************/
/***********************************************************/
/* factor */
/***********************************************************/
{
int n;
double unit;
n = A->n;
unit = (n-1.)+n;
wtime_order = taucs_wtime();
taucs_ccs_order(A,&perm,&invperm,ordering);
wtime_order = taucs_wtime() - wtime_order;
taucs_printf("\tOrdering time = % 10.3f seconds\n",wtime_order);
if (!perm) {
taucs_printf("\tOrdering Failed\n");
exit(1);
}
if (0) {
int i;
FILE* f;
f=fopen("p.ijv","w");
for (i=0; i<n; i++) fprintf(f,"%d\n",perm[i]+1);
fclose(f);
}
if (A->flags & TAUCS_SYMMETRIC || A->flags & TAUCS_HERMITIAN) {
wtime_permute = taucs_wtime();
PAPT = taucs_ccs_permute_symmetrically(A,perm,invperm);
wtime_permute = taucs_wtime() - wtime_permute;
taucs_printf("\tPermute time = % 10.3f seconds\n",wtime_permute);
}
wtime_factor = taucs_wtime();
ctime_factor = taucs_ctime();
if (ldlt_flag) {
L = taucs_ccs_factor_ldlt(PAPT);
precond_args = L;
precond_fn = taucs_ccs_solve_ldlt;
} else if (snmf_flag) {
/*taucs_ccs_matrix* C;*/
L = taucs_ccs_factor_llt_mf(PAPT);
precond_args = L;
precond_fn = taucs_supernodal_solve_llt;
{
taucs_ccs_matrix* C;
C = taucs_supernodal_factor_to_ccs(L);
/*taucs_ccs_write_ijv(PAPT,"PAPT.ijv");*/
/*C->flags = TAUCS_DCOMPLEX | TAUCS_TRIANGULAR | TAUCS_LOWER;*/
precond_args = C;
precond_fn = taucs_ccs_solve_llt;
/*taucs_ccs_write_ijv(C,"L.ijv");*/
/*
{
int i;
double* diag = taucs_supernodal_factor_get_diag(L);
for (i=0; i<C->n; i++) {
printf("%.2le\n",diag[i]);
}
}
*/
}
} else if (ooc_flag) {
if (A->flags & TAUCS_SYMMETRIC || A->flags & TAUCS_HERMITIAN) {
#define TESTING
#ifdef TESTING
int taucs_ooc_factor_llt_panelchoice(taucs_ccs_matrix* A,
taucs_io_handle* handle,
double memory,
int panelization_method);
/*int c; omer*/
oocL = taucs_io_create_multifile(matrixfile);
assert(oocL);
if (memory_mb == -1.0) memory_mb = taucs_available_memory_size()/1048576.0;
taucs_ooc_factor_llt_panelchoice(PAPT, oocL, memory_mb*1048576.0,panelize);
precond_args = oocL;
precond_fn = taucs_ooc_solve_llt;
#else
/*int c;*/
oocL = taucs_io_create_multifile(matrixfile);
assert(oocL);
if (memory_mb == -1.0) memory_mb = taucs_available_memory_size()/1048576.0;
taucs_ooc_factor_llt(PAPT, oocL, memory_mb*1048576.0);
precond_args = oocL;
precond_fn = taucs_ooc_solve_llt;
#endif
} else {
if (memory_mb == -1.0) memory_mb = taucs_available_memory_size()/1048576.0;
oocL = taucs_io_create_multifile(matrixfile);
taucs_ooc_factor_lu(A, perm, oocL, memory_mb*1048576.0);
precond_args = matrixfile;
precond_fn = NULL;
}
} else if (snll_flag) {
L = taucs_ccs_factor_llt_ll(PAPT);
precond_args = L;
precond_fn = taucs_supernodal_solve_llt;
} else if (symb_flag) {
L = taucs_ccs_factor_llt_symbolic(PAPT);
taucs_ccs_factor_llt_numeric(PAPT,L); /* should check error code */
precond_args = L;
precond_fn = taucs_supernodal_solve_llt;
} else {
L = taucs_ccs_factor_llt(PAPT,0.0,0);
precond_args = L;
precond_fn = taucs_ccs_solve_llt;
}
wtime_factor = taucs_wtime() - wtime_factor;
ctime_factor = taucs_ctime() - ctime_factor;
taucs_printf("\tFactor time = % 10.3f seconds ",wtime_factor);
taucs_printf("(%.3f cpu time)\n",ctime_factor);
}
if (!L && !ooc_flag /* no L in ooc */) {
taucs_printf("\tFactorization Failed\n");
exit(1);
}
/*taucs_ccs_write_ijv(PAPT,"A.ijv",1);*/ /* 1 = complete the upper part */
/*taucs_ccs_write_ijv(L,"L.ijv",0);*/
/***********************************************************/
/* solve */
/***********************************************************/
if (!L) {
taucs_printf("FACTORIZATION FAILED!\n");
exit(1);
}
if (A->flags & TAUCS_SYMMETRIC || A->flags & TAUCS_HERMITIAN) {
if (A->flags & TAUCS_DOUBLE)
taucs_vec_permute(A->n,A->flags,Bd,PBd,perm);
if (A->flags & TAUCS_SINGLE)
taucs_vec_permute(A->n,A->flags,Bs,PBs,perm);
if (A->flags & TAUCS_DCOMPLEX)
taucs_vec_permute(A->n,A->flags,Bz,PBz,perm);
wtime_solve = taucs_wtime();
if (A->flags & TAUCS_DOUBLE)
precond_fn(precond_args,PXd,PBd); /* direct solver */
if (A->flags & TAUCS_SINGLE)
precond_fn(precond_args,PXs,PBs); /* direct solver */
if (A->flags & TAUCS_DCOMPLEX)
precond_fn(precond_args,PXz,PBz); /* direct solver */
#ifdef TAUCS_CONFIG_SINGLE
if (A->flags & TAUCS_SINGLE) {
taucs_sccs_times_vec_dacc(PAPT,PXs,NXs);
for(i=0; i<(A->n); i++) NXs[i] -= PBs[i];
precond_fn(precond_args,PBs,NXs); /* direct solver */
for(i=0; i<(A->n); i++) PXs[i] -= PBs[i];
}
#endif
wtime_solve = taucs_wtime() - wtime_solve;
taucs_printf("\tSolve time = % 10.3f seconds\n",wtime_solve);
if (A->flags & TAUCS_DOUBLE)
taucs_vec_ipermute(A->n,A->flags,PXd,NXd,perm);
if (A->flags & TAUCS_SINGLE)
taucs_vec_ipermute(A->n,A->flags,PXs,NXs,perm);
if (A->flags & TAUCS_DCOMPLEX)
taucs_vec_ipermute(A->n,A->flags,PXz,NXz,perm);
} else {
taucs_ooc_solve_lu(oocL, NXd, Bd);
}
/***********************************************************/
/* delete out-of-core matrices */
/***********************************************************/
if (ooc_flag) {
taucs_io_delete(oocL);
/*taucs_io_close(oocL);*/
}
/***********************************************************/
/* Compute norm of forward error */
/***********************************************************/
if (A->flags & TAUCS_SINGLE) {
float snrm2_();
int one = 1;
NormErr = 0.0;
for(i=0; i<N; i++) NormErr = max(NormErr,fabs((NXs[i]-Xs[i])/Xs[i]));
for(i=0; i<N; i++) PXs[i] = NXs[i]-Xs[i];
taucs_printf("main: max relative error = %1.6e, 2-norm relative error %.2e \n",
NormErr,
snrm2_(&(A->n),PXs,&one)/snrm2_(&(A->n),Xs,&one));
}
if (A->flags & TAUCS_DOUBLE) {
double dnrm2_();
int one = 1;
NormErr = 0.0;
for(i=0; i<N; i++) NormErr = max(NormErr,fabs((NXd[i]-Xd[i])/Xd[i]));
for(i=0; i<N; i++) PXd[i] = NXd[i]-Xd[i];
taucs_printf("main: max relative error = %1.6e, 2-norm relative error %.2e \n",
NormErr,
dnrm2_(&(A->n),PXd,&one)/dnrm2_(&(A->n),Xd,&one));
}
#ifdef TAUCS_CONFIG_DCOMPLEX
if (A->flags & TAUCS_DCOMPLEX) {
double dznrm2_();
int one = 1;
double* pX = (double*) Xz;
double* pNX = (double*) NXz;
double* pPX = (double*) PXz;
taucs_dcomplex zzero = taucs_zzero_const;
taucs_dcomplex zone = taucs_zone_const;
taucs_dcomplex zmone = taucs_zneg(taucs_zone_const);
NormErr = 0.0;
/*
for(i=0; i<N; i++) NormErr = max(NormErr,fabs((NXd[i]-Xd[i])/Xd[i]));
*/
/*for(i=0; i<N; i++) PXd[i] = NXd[i]-Xd[i];*/
/*for(i=0; i<N; i++) PXz[i] = taucs_add(NXz[i],taucs_neg(Xz[i]));*/
zscal_(&(A->n),&zzero,pPX,&one);
zaxpy_(&(A->n),&zone ,pNX,&one,pPX,&one);
zaxpy_(&(A->n),&zmone,pX ,&one,pPX,&one);
taucs_printf("main: max relative error = %1.6e, 2-norm relative error %.2e \n",
NormErr,
dznrm2_(&(A->n),PXz,&one)/dznrm2_(&(A->n),Xz,&one));
}
#endif
/***********************************************************/
/* Exit */
/***********************************************************/
taucs_printf("main: done\n");
return 0;
}
