void TPZStructMatrixCS::Assemble(TPZMatrix<STATE> & stiffness, TPZFMatrix<STATE> & rhs,TPZAutoPointer<TPZGuiInterface> guiInterface){
ass_stiff.start();
if (fEquationFilter.IsActive()) {
int64_t neqcondense = fEquationFilter.NActiveEquations();
#ifdef PZDEBUG
if (stiffness.Rows() != neqcondense) {
DebugStop();
}
#endif
TPZFMatrix<STATE> rhsloc(neqcondense,rhs.Cols(),0.);
if(this->fNumThreads){
this->MultiThread_Assemble(stiffness,rhsloc,guiInterface);
}
else{
this->Serial_Assemble(stiffness,rhsloc,guiInterface);
}
fEquationFilter.Scatter(rhsloc, rhs);
}
else
{
if(this->fNumThreads){
this->MultiThread_Assemble(stiffness,rhs,guiInterface);
}
else{
this->Serial_Assemble(stiffness,rhs,guiInterface);
}
}
ass_stiff.stop();
}
inline void TPZQuadraticQuad::GradX(const TPZFMatrix<REAL> &nodes,TPZVec<T> &loc, TPZFMatrix<T> &gradx){
gradx.Resize(3,2);
gradx.Zero();
int nrow = nodes.Rows();
int ncol = nodes.Cols();
#ifdef PZDEBUG
if(nrow != 3 || ncol != 8){
std::cout << "Objects of incompatible lengths, gradient cannot be computed." << std::endl;
std::cout << "nodes matrix must be 3x8." << std::endl;
DebugStop();
}
#endif
TPZFNMatrix<3,T> phi(NNodes,1);
TPZFNMatrix<6,T> dphi(2,NNodes);
TShape(loc,phi,dphi);
for(int i = 0; i < NNodes; i++)
{
for(int j = 0; j < 3; j++)
{
gradx(j,0) += nodes.GetVal(j,i)*dphi(0,i);
gradx(j,1) += nodes.GetVal(j,i)*dphi(1,i);
}
}
}
void TPZMatRed<TVar,TSideMatrix>::SetF(const TPZFMatrix<TVar> & F)
{
int64_t FCols=F.Cols(),c,r,r1;
fF0.Redim(fDim0,FCols);
fF1.Redim(fDim1,FCols);
for(c=0; c<FCols; c++){
r1=0;
for(r=0; r<fDim0; r++){
fF0.PutVal( r,c,F.GetVal(r,c) ) ;
}
//aqui r=fDim0
for( ;r<fDim0+fDim1; r++){
fF1.PutVal( r1++,c,F.GetVal(r,c) );
}
}
#ifdef LOG4CXX
if (logger->isDebugEnabled()) {
std::stringstream sout;
F.Print("F Input",sout);
fF0.Print("fF0 Initialized",sout);
fF1.Print("fF1 Initialized",sout);
LOGPZ_DEBUG(logger, sout.str())
}
void TPZArtDiff::ContributeFastestImplDiff_dim(TPZFMatrix<REAL> &jacinv, TPZVec<STATE> &sol, TPZFMatrix<STATE> &dsol, TPZFMatrix<REAL> &phi, TPZFMatrix<REAL> &dphi, TPZFMatrix<STATE> &ek, TPZFMatrix<STATE> &ef, REAL weight, REAL timeStep, REAL deltaX)
{
REAL delta = Delta(deltaX, sol);
REAL constant = /*-*/ weight * delta * timeStep;
REAL buff;
TPZVec<TPZVec<STATE> > TauDiv;
TPZVec<TPZDiffMatrix<STATE> > dTauDiv;
PrepareFastestDiff<dim>( jacinv, sol, dsol, phi, dphi, TauDiv, dTauDiv);
int i, j, k, l;
int nshape = dphi.Cols();
int nstate = dim + 2;
int neq = nstate * nshape;
// ODotProduct speeded up
for(l=0;l<nshape;l++)
for(i=0;i<nstate;i++)
for(k=0;k<dim;k++)
{
buff = dphi(k,l) * constant;
ef(i+l*nstate,0) += buff * TauDiv[k][i];
for(j=0;j<neq;j++)
ek(i+l*nstate,j) -= buff * dTauDiv[k](i,j);
}
}
void TPZErrorIndicator::Sort(TPZFMatrix &error, TPZFMatrix &perm) {
int i,j,k;
int imin = 0;
int imax = error.Rows();
perm.Resize(imax,error.Cols());
for (i=0;i<imax;i++)
for (j=0;j<error.Cols();j++) perm(i,j) = i;
for(i=imin; i<imax; i++) {
for(j=i+1; j<imax; j++) {
for (k=0;k<error.Cols();k++){
if(error((int)perm(i,k)) < error((int)perm(j,k))) {
int kp = (int) perm(i,k);
perm(i,k) = perm(j,k);
perm(j,k) = kp;
}
}
}
}
}
REAL TPZIncNavierStokesKEps::Dot(TPZFMatrix<REAL> &A, TPZFMatrix<REAL> &B){
REAL sum = 0.;
int i, j, rows, cols;
rows = A.Rows();
cols = A.Cols();
for(i = 0; i < rows; i++){
for(j = 0; j < cols; j++){
sum += A(i,j) * B(i,j);
}
}
return sum;
}
/** This function creates a Visualization Tool Kit (VTK) file that allow to visualization of the value of a matrix passed as parameter */
void VisualMatrixVTK(TPZFMatrix<REAL> & matrix, const std::string &outfilename)
{
const int nelx = matrix.Cols();
const int nely = matrix.Rows();
const int neltotal = nelx * nely;
int i,j;
ofstream out(outfilename.c_str());
out << "# vtk DataFile Version 3.0\n";
out << "Generated by PZ\n";
out << "ASCII\n";
out << "DATASET RECTILINEAR_GRID\n";
out << "DIMENSIONS " << (nelx+1) << " " << (nely+1) << " 1\n";
out << "X_COORDINATES " << nelx+1 << " float\n";
for (i=0; i<=nelx; i++) {
out << i << " ";
}
out << std::endl;
out << "Y_COORDINATES " << nely+1 << " float\n";
for (j=0; j<=nely; j++) {
out << j << " ";
}
out << std::endl;
out << "Z_COORDINATES " << 1 << " float\n0.\n";
out << "CELL_DATA " << nelx*nely << std::endl;
out << "SCALARS mat_value float 1\n";
out << "LOOKUP_TABLE default\n";
const REAL *elem = &matrix(0,0);
for (i=0; i<neltotal; i++) {
out << *(elem+i) << std::endl;
}
/*
# vtk DataFile Version 3.0
Cube example
ASCII
DATASET RECTILINEAR_GRID
DIMENSIONS 3 3 1
X_COORDINATES 3 float
0. 1. 2.
Y_COORDINATES 3 float
0. 1. 2.
Z_COORDINATES 1 float
0.
CELL_DATA 4
SCALARS toto float 1
LOOKUP_TABLE default
1
2
3
4
*/
}
void TPZDohrMatrix<TVar,TSubStruct>::MultAddTBB(const TPZFMatrix<TVar> &x,const TPZFMatrix<TVar> &y, TPZFMatrix<TVar> &z,
const TVar alpha,const TVar beta,const int opt) const
{
#ifdef USING_TBB
if ((!opt && this->Cols() != x.Rows()) || this->Rows() != x.Rows())
this->Error( "Operator* <matrixs with incompatible dimensions>" );
if(x.Cols() != y.Cols() || x.Cols() != z.Cols() || x.Rows() != y.Rows() || x.Rows() != z.Rows()) {
this->Error ("TPZFMatrix::MultiplyAdd incompatible dimensions\n");
}
this->PrepareZ(y,z,beta,opt);
unsigned int nglob = fGlobal.size();
TPZAutoPointer<TPZDohrAssembleList<TVar> > assemblelist = new TPZDohrAssembleList<TVar>(nglob,z,this->fAssembly);
ParallelAssembleTaskMatrix<TVar,TSubStruct> multwork(x,alpha,fAssembly,assemblelist);
typename std::list<TPZAutoPointer<TSubStruct> >::const_iterator iter;
int isub=0;
for (iter=fGlobal.begin(); iter!=fGlobal.end(); iter++,isub++) {
TPZDohrThreadMultData<TSubStruct> data(isub,*iter);
multwork.addWorkItem(data);
}
TPZVec<pthread_t> AllThreads(1);
multwork.run_parallel_for(pzenviroment.fSubstructurePartitioner);
PZ_PTHREAD_CREATE(&AllThreads[0], 0, TPZDohrAssembleList<TVar>::Assemble,
assemblelist.operator->(), __FUNCTION__);
void *result;
PZ_PTHREAD_JOIN(AllThreads[0], &result, __FUNCTION__);
#endif
}
/** This function creates a Data Explorer file that allow to visualization of the value of a matrix passed as parameter */
void VisualMatrixDX(TPZFMatrix<REAL> & matrix, const std::string &outfilename)
{
const int nelx = matrix.Cols();
const int nely = matrix.Rows();
const int neltotal = nelx * nely;
int i,j;
ofstream out(outfilename.c_str());
out << "# Graphical Visualization of Matrix." << endl;
out << "# Positions as the indexes of the matrix, beginning by column." << endl;
out << "# The number of elements in x direction correspond to the number of the columns of the matrix." << endl;
out << "# The number of elements in y direction correspond to the number of the rows of the matrix." << endl;
out << "object 1 class gridpositions counts " << nelx+1 << " " << nely +1 << endl;
out << "origin 0. 0." << endl;
out << "delta 1. 0." << endl;
out << "delta 0. 1." << endl;
out << "attribute \"dep\" string \"positions\"" << endl;
out << endl;
out << "object 2 class gridconnections counts " << nelx+1 << " " << nely +1 << endl;
out << "attribute \"element type\" string \"quads\"" << endl;
out << "attribute \"ref\" string \"positions\"" << endl;
out.precision(5);
out << "object 3 class array type float rank 0 items " << neltotal << " data follows" << endl;
for (i = 0; i < nelx; i++) {
for(j=0; j< nely ; j++) out << matrix(i,j) << endl;
}
out << "attribute \"dep\" string \"connections\" " << endl;
out << endl;
out << "object 4 class field" << endl;
out << "component \"data\" value 3" << endl;
out << "component \"positions\" value 1" << endl;
out << "component \"connections\" value 2" << endl;
out << "attribute \"name\" string \"Matrix\"" << endl;
out << endl;
out << "end" << endl;
out.close();
cout << "Data Explorer file " << outfilename << " was created with success!\n";
}
void TPZArtDiff::ContributeExplDiff(int dim, TPZFMatrix<REAL> &jacinv, TPZVec<STATE> &sol, TPZFMatrix<STATE> &dsol, TPZFMatrix<REAL> &dphix, TPZFMatrix<STATE> &ef, REAL weight, REAL timeStep, REAL deltaX)
{
REAL delta = Delta(deltaX, sol);
REAL constant = /*-*/ weight * delta * timeStep;
TPZVec<TPZVec<STATE> > TauDiv;
PrepareFastDiff(dim, jacinv, sol, dsol, dphix, TauDiv, NULL);
int i, k, l;
int nshape = dphix.Cols();
int nstate = dim + 2;
// ODotProduct speeded up
for(l=0;l<nshape;l++)
for(i=0;i<nstate;i++)
for(k=0;k<dim;k++)
ef(i+l*nstate,0) += dphix(k,l) * TauDiv[k][i] * constant;
}
void TPZArtDiff::Divergent(TPZFMatrix<STATE> &dsol,
TPZFMatrix<REAL> & dphi,
TPZVec<TPZDiffMatrix<STATE> > & Ai,
TPZVec<STATE> & Div,
TPZDiffMatrix<STATE> * dDiv)
{
int nstate = Ai[0].Cols();
int dim = nstate - 2;
int nshape = dphi.Cols();
Div.Resize(nstate);
Div.Fill(0.);
int i, j, k;
// computing the divergent:
// A.du/dx + B.du/dy + C.du/dz
for(k=0;k<dim;k++)
for(i=0;i<nstate; i++)
for(j=0;j<nstate;j++)
{
Div[i]+=Ai[k](i,j)*dsol(k,j);
}
if(!dDiv)return;
// computing an approximation to the divergent derivative:
// dDiv/dUj ~= A.d2U/dUidx + B.d2U/dUidy + C.d2U/dUidz
dDiv->Redim(nstate, nstate * nshape);
int l;
REAL buff;
for(l=0;l<nshape;l++)
for(j=0;j<nstate;j++)
for(i=0;i<nstate; i++)
{
buff =0.;
for(k=0;k<dim;k++)
{
buff+=Ai[k](i,j)*dphi(k,l);
}
dDiv->operator()(i,j+l*nstate)=buff;
}
}
void TPZArtDiff::Divergent(TPZFMatrix<STATE> &dsol,
TPZFMatrix<REAL> & phi,
TPZFMatrix<REAL> & dphi,
TPZVec<TPZDiffMatrix<T> > & Ai,
TPZVec<STATE> & Div,
TPZDiffMatrix<STATE> * dDiv)
{
int nstate = Ai[0].Cols();
int dim = nstate - 2;
int nshape = dphi.Cols();
Div.Resize(nstate);
Div = (STATE(0.));
int i, j, k;
// computing the divergent:
// A.du/dx + B.du/dy + C.du/dz
for(k=0;k<dim;k++)
for(i=0;i<nstate; i++)
for(j=0;j<nstate;j++)
{
Div[i] += Ai[k](i,j).val() * dsol(k,j);
}
if(!dDiv)return;
// computing an approximation to the divergent derivative:
// dDiv/dUj ~= A.d2U/dUidx + B.d2U/dUidy + C.d2U/dUidz
dDiv->Redim(nstate, nstate * nshape);
int l;
REAL buff;
for(l=0;l<nshape;l++)
for(j=0;j<nstate;j++)
for(i=0;i<nstate; i++)
{
buff =0.;
for(k=0;k<dim;k++)
{
buff+=Ai[k](i,j).val()*dphi(k,l);
}
dDiv->operator()(i,j+l*nstate)=buff;
}
TPZVec<T> ADiv(nstate);
T temp;
for( k = 0; k < dim; k++)
{
//Ai[k].Multiply(Div, ADiv);
for(i = 0; i < nstate; i++)
{
temp = T(0.);
for(j = 0; j < nstate; j++)
temp += Ai[k](i,j) * (T)dsol(k,j);//[j];
ADiv[i] = temp;
}
for(l=0;l<nshape;l++)
for(j=0;j<nstate;j++)
for(i=0;i<nstate; i++)
dDiv->operator()(i,j+l*nstate) +=
ADiv[i]./*fastAccessDx*/dx(j) * phi(l,0);
}
}
/** @brief Computing operator for the parallel for. */
void operator()(const blocked_range<size_t>& range) const
{
for(size_t i=range.begin(); i!=range.end(); ++i )
{
TPZDohrThreadMultData<TSubStruct> runner = mWorkItems[i];
TPZFMatrix<TVar> xlocal;
fAssembly->Extract(runner.fisub,*(fInput),xlocal);
TPZAutoPointer<TPZDohrAssembleItem<TVar> > assembleItem = new TPZDohrAssembleItem<TVar>(runner.fisub,xlocal.Rows(),xlocal.Cols());
runner.fSub->ContributeKULocal(fAlpha,xlocal,assembleItem->fAssembleData);
fAssemblyStructure->AddItem(assembleItem);
}
}
void TPZSkylMatrix<TVar>::MultAdd(const TPZFMatrix<TVar> &x,const TPZFMatrix<TVar> &y, TPZFMatrix<TVar> &z,
const TVar alpha,const TVar beta ,const int opt,const int stride ) const {
// Computes z = beta * y + alpha * opt(this)*x
// z and x cannot overlap in memory
if ((!opt && this->Cols()*stride != x.Rows()) || this->Rows()*stride != x.Rows())
TPZMatrix<TVar>::Error(__PRETTY_FUNCTION__," <matrixs with incompatible dimensions>" );
if(z.Rows() != x.Rows() || z.Cols() != x.Cols()) z.Redim(x.Rows(),x.Cols());
if(x.Cols() != y.Cols() || x.Cols() != z.Cols() || x.Rows() != y.Rows() || x.Rows() != z.Rows()) {
cout << "x.Cols = " << x.Cols() << " y.Cols()"<< y.Cols() << " z.Cols() " << z.Cols() << " x.Rows() " << x.Rows() << " y.Rows() "<< y.Rows() << " z.Rows() "<< z.Rows() << endl;
TPZMatrix<TVar>::Error(__PRETTY_FUNCTION__," incompatible dimensions\n");
}
this->PrepareZ(y,z,beta,opt,stride);
int rows = this->Rows();
int xcols = x.Cols();
int ic, r;
for (ic = 0; ic < xcols; ic++) {
for( r = 0 ; r < rows ; r++ ) {
int offset = Size(r);
TVar val = 0.;
const TVar *p = &x.g((r-offset+1)*stride,ic);
TVar *diag = fElem[r] + offset-1;
TVar *diaglast = fElem[r];
while( diag > diaglast ) {
val += *diag-- * *p;
p += stride;
}
if( diag == diaglast ) val += *diag * *p;
z(r*stride,ic) += val*alpha;
TVar *zp = &z((r-offset+1)*stride,ic);
val = x.g(r*stride,ic);
diag = fElem[r] + offset-1;
while( diag > diaglast ) {
*zp += alpha * *diag-- * val;
zp += stride;
}
}
}
}
void TPZGeoTriangle::VecHdiv(TPZFMatrix<REAL> & coord, TPZFMatrix<REAL> & fNormalVec,TPZVec<int> &fVectorSide) {
if(coord.Rows()!=3)
{
cout<< "Erro na dimens�o das linhas de coord"<< endl;
}
if(coord.Cols()!=3)
{
cout<< "Erro na dimens�o das colunas de coord"<< endl;
}
TPZVec<REAL> p1(3), p2(3), p3(3),result(3);
for(int j=0; j<3; j++)
{
p1[j]=coord(j,0);
p2[j]=coord(j,1);
p3[j]=coord(j,2);
}
fNormalVec.Resize(14, 3);
fVectorSide.Resize(14);
int count=0;
//primeira face
for(int j=0; j<3; j++) //v0
{
fNormalVec(0,j) = coord(j,0)- coord(j,2);
}
fVectorSide[count]=0;
count++;
for(int j=0; j<3; j++) //v1
{
fNormalVec(1,j) = coord(j,1)- coord(j,2);
}
fVectorSide[count]=1;
count++;
//v2
ComputeNormal(p1,p2,p3,result);
fNormalVec(2,0) = -result[0];
fNormalVec(2,1) = -result[1];
fNormalVec(2,2) = -result[2];
fVectorSide[count]=3;
count++;
//segunda face
for(int j=0; j<3; j++) //v3
{
fNormalVec(3,j) = coord(j,1)- coord(j,0);
}
fVectorSide[count]=1;
count++;
for(int j=0; j<3; j++) //v4
{
fNormalVec(4,j) = coord(j,2)- coord(j,0);
}
fVectorSide[count]=2;
count++;
//v5
ComputeNormal(p2,p3,p1,result);
fNormalVec(5,0) = -result[0];
fNormalVec(5,1) = -result[1];
fNormalVec(5,2) = -result[2];
fVectorSide[count]=4;
count++;
//terceira face
for(int j=0; j<3; j++) //v6
{
fNormalVec(6,j) = coord(j,2)- coord(j,1);
}
fVectorSide[count]=2;
count++;
for(int j=0; j<3; j++) //v7
{
fNormalVec(7,j) = coord(j,0)- coord(j,1);
}
fVectorSide[count]=0;
count++;
//v8
ComputeNormal(p3,p1,p2,result);
fNormalVec(8,0) = -result[0];
fNormalVec(8,1) = -result[1];
fNormalVec(8,2) = -result[2];
fVectorSide[count]=5;
count++;
// internos tangentes
for(int j=0; j<3; j++) //v9
{
fNormalVec(9,j) = coord(j,1)- coord(j,0);
}
fVectorSide[count]=3;
count++;
for(int j=0; j<3; j++) //v10
{
fNormalVec(10,j) = coord(j,2)- coord(j,1);
}
fVectorSide[count]=4;
count++;
for(int j=0; j<3; j++) //v11
{
fNormalVec(11,j) = coord(j,0)- coord(j,2);
}
fVectorSide[count]=5;
count++;
//internos meio
TPZVec<REAL> midle(3,0.);
midle[0]=(1./3.)*(coord(0,2)+coord(0,0)+coord(0,1));
midle[1]=(1./3.)*(coord(1,2)+coord(1,0)+coord(1,1));
midle[2]=(1./3.)*(coord(2,2)+coord(2,0)+coord(2,1));
TPZFMatrix<REAL> jacobian;
TPZFMatrix<REAL> axes;
REAL detjac;
TPZFMatrix<REAL> jacinv;
Jacobian(coord,midle,jacobian,axes,detjac,jacinv);
fNormalVec(12,0)=axes(0,0);
fNormalVec(12,1)=axes(0,1);
fNormalVec(12,2)=axes(0,2);
fNormalVec(13,0)=axes(1,0);
fNormalVec(13,1)=axes(1,1);
fNormalVec(13,2)=axes(1,2);
fVectorSide[count]=6;
fVectorSide[count+1]=6;
//normaliza��o
for(int k=0; k<14; k++)
{
REAL temp=0.;
temp=sqrt( fNormalVec(k,0)*fNormalVec(k,0) + fNormalVec(k,1)*fNormalVec(k,1) + fNormalVec(k,2)*fNormalVec(k,2));
fNormalVec(k,0) *=1./temp;
fNormalVec(k,1) *=1./temp;
}
// produto normal == 1
for(int kk=0; kk<3; kk++)
{
REAL temp1=0.;
REAL temp2=0.;
temp1 = fNormalVec(kk*3,0)*fNormalVec(kk*3+2,0) + fNormalVec(kk*3,1)*fNormalVec(kk*3+2,1);
temp2 = fNormalVec(kk*3+1,0)*fNormalVec(kk*3+2,0) + fNormalVec(kk*3+1,1)*fNormalVec(kk*3+2,1);
fNormalVec(kk*3,0) *=1./temp1;
fNormalVec(kk*3,1) *=1./temp1;
fNormalVec(kk*3+1,0) *=1./temp2;
fNormalVec(kk*3+1,1) *=1./temp2;
}
#ifdef LOG4CXX
{
std::stringstream sout;
fNormalVec.Print("fNormalVec", sout);
LOGPZ_DEBUG(logger,sout.str())
}
#endif
}
void TPZSkylMatrix<TVar>::SolveSOR(int & numiterations,const TPZFMatrix<TVar> &F,
TPZFMatrix<TVar> &result, TPZFMatrix<TVar> *residual, TPZFMatrix<TVar> &scratch,const TVar overrelax,
TVar &tol,const int FromCurrent,const int direction) {
if(residual == &F) {
cout << "TPZMatrix::SolveSOR called with residual and F equal, no solution\n";
return;
}
TVar res = 2*tol+1.;;
if(residual) res = Norm(*residual);
if(!FromCurrent) {
result.Zero();
}
int r = this->Dim();
int c = F.Cols();
int i,ifirst = 0, ilast = r, iinc = 1;
if(direction == -1) {
ifirst = r-1;
ilast = 0;
iinc = -1;
}
int it;
for(it=0; it<numiterations && res > tol; it++) {
res = 0.;
scratch = F;
for(int ic=0; ic<c; ic++) {
if(direction == 1) {
//
// compute the upper triangular part first and put into the scractch vector
//
for(i=ifirst; i!=ilast; i+= iinc) {
//TPZColuna *mydiag = &fDiag[i];
int offset = Size(i);
TVar val;
TVar *diag;
TVar *diaglast = fElem[i];
TVar *scratchp = &scratch(i-offset+1,ic);
val = result(i,ic);
diag = fElem[i] + offset-1;
int lastid = diag-diaglast;
int id;
for(id=0; id<=lastid; id++) *(scratchp+id) -= *(diag-id) * val;
/* codeguard fix
while( diag >= diaglast ) *scratchp++ -= *diag-- * val;
*/
}
//
// perform the SOR operation
//
for(i=ifirst; i!=ilast; i+= iinc) {
//TPZColuna *mydiag = &fDiag[i];
int offset = Size(i);
TVar val = scratch(i,ic);
TVar *p = &result(i-offset+1,ic);
TVar *diag = fElem[i] + offset-1;
TVar *diaglast = fElem[i];
while( diag > diaglast ) val -= *diag-- * *p++;
res += val*val;
result(i,ic) += val*overrelax/ *diag;
}
} else {
//
// the direction is upward
//
// put the lower triangular part of the multiplication into the scratch vector
//
for(i=ifirst; i!=ilast; i+= iinc) {
//TPZColuna *mydiag = &fDiag[i];
int offset = Size(i);
TVar val = scratch(i,ic);
TVar *p = &result(i-offset+1,ic);
TVar *diag = fElem[i] + offset-1;
TVar *diaglast = fElem[i];
while( diag > diaglast ) val -= *diag-- * *p++;
// res += val*val;
scratch(i,ic) = val;
}
//
// perform the SOR operation
//
for(i=ifirst; i!=ilast; i+= iinc) {
//TPZColuna *mydiag = &fDiag[i];
int offset = Size(i);
// REAL val = scratch(i,ic);
TVar *diag;
TVar *diaglast = fElem[i];
TVar *scratchp = &scratch(i-offset+1,ic);
//val= result(i,ic);
TVar val = scratch(i,ic);
val -= *diaglast * result(i,ic);
res += val*val;
val = overrelax * val / *diaglast;
result(i,ic) += val;
val = result(i,ic);
diag = fElem[i] + offset-1;
while( diag > diaglast ) *scratchp++ -= *diag-- * val;
}
}
}
res = sqrt(res);
}
if(residual) {
this->Residual(result,F,*residual);
}
numiterations = it;
tol = res;
}
void TPZTransform::SetMatrix(TPZFMatrix<REAL> &mult, TPZFMatrix<REAL> &sum) {
fRow = mult.Rows();
fCol = mult.Cols();
fMult = mult;
fSum = sum;
}
void TPZDohrMatrix<TVar,TSubStruct>::MultAdd(const TPZFMatrix<TVar> &x,const TPZFMatrix<TVar> &y, TPZFMatrix<TVar> &z,
const TVar alpha,const TVar beta,const int opt) const
{
#ifdef USING_TBB
MultAddTBB(x, y, z, alpha, beta, opt);
return;
#endif
TPZfTime mult;
if ((!opt && this->Cols() != x.Rows()) || this->Rows() != x.Rows())
this->Error( "Operator* <matrixs with incompatible dimensions>" );
if(x.Cols() != y.Cols() || x.Cols() != z.Cols() || x.Rows() != y.Rows() || x.Rows() != z.Rows()) {
this->Error ("TPZFMatrix::MultiplyAdd incompatible dimensions\n");
}
this->PrepareZ(y,z,beta,opt);
typename SubsList::const_iterator iter;
int isub = 0;
if (fNumThreads == 0) {
for (iter=fGlobal.begin();iter!=fGlobal.end();iter++,isub++) {
if(0)
{
TPZPersistenceManager::OpenWrite("dohr.txt");
TPZPersistenceManager::WriteToFile(fAssembly.operator ->());
TPZPersistenceManager::WriteToFile(&x);
TPZAutoPointer<TSubStruct> point(*iter);
TPZPersistenceManager::WriteToFile(point.operator ->());
TPZPersistenceManager::CloseWrite();
}
TPZFMatrix<TVar> xlocal,zlocal;
fAssembly->Extract(isub,x,xlocal);
zlocal.Redim(xlocal.Rows(),xlocal.Cols());
(*iter)->ContributeKULocal(alpha,xlocal,zlocal);
fAssembly->Assemble(isub,zlocal,z);
// z.Print("Resultado intermediario");
}
}
else {
unsigned int nglob = fGlobal.size();
TPZAutoPointer<TPZDohrAssembleList<TVar> > assemblelist = new TPZDohrAssembleList<TVar>(nglob,z,this->fAssembly);
TPZDohrThreadMultList<TVar,TSubStruct> multwork(x,alpha,fAssembly,assemblelist);
typename std::list<TPZAutoPointer<TSubStruct> >::const_iterator iter;
int isub=0;
for (iter=fGlobal.begin(); iter!=fGlobal.end(); iter++,isub++) {
TPZDohrThreadMultData<TSubStruct> data(isub,*iter);
multwork.AddItem(data);
}
TPZVec<pthread_t> AllThreads(fNumThreads+1);
int i;
for (i=0; i<fNumThreads; i++) {
PZ_PTHREAD_CREATE(&AllThreads[i+1], 0, (TPZDohrThreadMultList<TVar,TSubStruct>::ThreadWork),
&multwork, __FUNCTION__);
}
//sleep(1);
PZ_PTHREAD_CREATE(&AllThreads[0], 0, TPZDohrAssembleList<TVar>::Assemble,
assemblelist.operator->(), __FUNCTION__);
for (i=0; i<fNumThreads+1; i++) {
void *result;
PZ_PTHREAD_JOIN(AllThreads[i], &result, __FUNCTION__);
}
}
tempo.fMultiply.Push(mult.ReturnTimeDouble());
}
void TPZStepSolver<TVar>::Solve(const TPZFMatrix<TVar> &F, TPZFMatrix<TVar> &result, TPZFMatrix<TVar> *residual){
if(!this->Matrix()) {
cout << "TPZMatrixSolver::Solve called without a matrix pointer\n";
DebugStop();
}
TPZAutoPointer<TPZMatrix<TVar> > mat = this->Matrix();
// update the matrix to which the preconditioner refers
if(fPrecond)
{
fPrecond->UpdateFrom(this->Matrix());
}
if(result.Rows() != mat->Rows() || result.Cols() != F.Cols()) {
result.Redim(mat->Rows(),F.Cols());
}
if(this->fScratch.Rows() != result.Rows() || this->fScratch.Cols() != result.Cols()) {
this->fScratch.Redim(result.Rows(),result.Cols());
}
TVar tol = fTol;
int numiterations = fNumIterations;
switch(fSolver) {
case TPZStepSolver::ENoSolver:
default:
cout << "TPZMatrixSolver::Solve called without initialized solver, Jacobi used\n";
SetJacobi(1,0.,0);
case TPZStepSolver::EJacobi:
// cout << "fScratch dimension " << fScratch.Rows() << ' ' << fScratch.Cols() << endl;
mat->SolveJacobi(numiterations,F,result,residual,this->fScratch,tol,fFromCurrent);
break;
case TPZStepSolver::ESOR:
mat->SolveSOR(numiterations,F,result,residual,this->fScratch,fOverRelax,tol,fFromCurrent);
break;
case TPZStepSolver::ESSOR:
mat->SolveSSOR(numiterations,F,result,residual,this->fScratch,fOverRelax,tol,fFromCurrent);
break;
case TPZStepSolver::ECG:
mat->SolveCG(numiterations,*fPrecond,F,result,residual,tol,fFromCurrent);
#ifdef LOG4CXX
{
std::stringstream sout;
sout << "Number of equations " << mat->Rows() << std::endl;
sout << "Number of CG iterations " << numiterations << " tol = " << tol;
LOGPZ_DEBUG(logger,sout.str().c_str());
}
#endif
break;
case TPZStepSolver::EGMRES: {
TPZFMatrix<TVar> H(fNumVectors+1,fNumVectors+1,0.);
mat->SolveGMRES(numiterations,*fPrecond,H,fNumVectors,F,result,residual,tol,fFromCurrent);
if(numiterations == fNumIterations || tol >= fTol)
{
std::cout << "GMRes tolerance was not achieved : numiter " << numiterations <<
" tol " << tol << endl;
}
#ifdef LOG4CXX
{
std::stringstream sout;
sout << "Number of GMRES iterations " << numiterations << " tol = " << tol;
LOGPZ_DEBUG(logger,sout.str().c_str());
}
#endif
}
break;
case TPZStepSolver::EBICGSTAB:
mat->SolveBICGStab(numiterations, *fPrecond, F, result,residual,tol,fFromCurrent);
if(numiterations == fNumIterations || tol >= fTol)
{
std::cout << "BiCGStab tolerance was not achieved : numiter " << numiterations <<
" tol " << tol << endl;
}
#ifdef LOG4CXX
{
std::stringstream sout;
sout << "Number of BiCGStab iterations " << numiterations << " tol = " << tol;
LOGPZ_DEBUG(logger,sout.str().c_str());
}
#endif
break;
case TPZStepSolver::EDirect:
result = F;
mat->SolveDirect(result,fDecompose,fSingular);
if(residual) residual->Redim(F.Rows(),F.Cols());
break;
case TPZStepSolver::EMultiply:
mat->Multiply(F,result);
if(residual) mat->Residual(result,F,*residual);
}
}
void *TPZDohrThreadMultList<TVar,TSubStruct>::ThreadWork(void *ptr)
{
TPZDohrThreadMultList<TVar,TSubStruct> *myptr = (TPZDohrThreadMultList<TVar,TSubStruct> *) ptr;
TPZDohrThreadMultData<TSubStruct> runner = myptr->PopItem();
while (runner.IsValid()) {
TPZFMatrix<TVar> xlocal;
myptr->fAssembly->Extract(runner.fisub,*(myptr->fInput),xlocal);
TPZAutoPointer<TPZDohrAssembleItem<TVar> > assembleItem = new TPZDohrAssembleItem<TVar>(runner.fisub,xlocal.Rows(),xlocal.Cols());
runner.fSub->ContributeKULocal(myptr->fAlpha,xlocal,assembleItem->fAssembleData);
myptr->fAssemblyStructure->AddItem(assembleItem);
runner = myptr->PopItem();
}
return ptr;
}
