void ElementIntegral
::createOneFormTransformationMatrix(const CellJacobianBatch& JTrans,
const CellJacobianBatch& JVol) const
{
TimeMonitor timer(transCreationTimer());
Tabs tab;
SUNDANCE_MSG2(transformVerb(),
tab << "ElementIntegral creating linear form trans matrices");
int maxDim = JTrans.cellDim();
if (transformationMatrixIsValid(alpha())) return;
transformationMatrixIsValid(alpha()) = true;
int flops = JTrans.numCells() * maxDim + JTrans.numCells();
G(alpha()).resize(JTrans.numCells() * JTrans.cellDim());
int k = 0;
double* GPtr = &(G(alpha())[0]);
for (int c=0; c<JTrans.numCells(); c++)
{
Array<double> invJ;
JTrans.getInvJ(c, invJ);
double detJ = fabs(JVol.detJ()[c]);
for (int gamma=0; gamma<maxDim; gamma++, k++)
{
GPtr[k] = detJ*invJ[alpha() + maxDim * gamma];
}
}
addFlops(flops);
}
void PeriodicSingleCellMesh1D::getJacobians(int cellDim, const Array<int>& cellLID,
CellJacobianBatch& jBatch) const
{
TEUCHOS_TEST_FOR_EXCEPTION(cellDim < 0 || cellDim > spatialDim(), std::logic_error,
"cellDim=" << cellDim
<< " is not in expected range [0, " << spatialDim()
<< "]");
jBatch.resize(cellLID.size(), spatialDim(), cellDim);
int nCells = cellLID.size();
TEUCHOS_TEST_FOR_EXCEPT(nCells != 1);
if (cellDim==0)
{
for (int i=0; i<nCells; i++)
{
double* detJ = jBatch.detJ(i);
*detJ = 1.0;
}
}
else
{
for (int i=0; i<nCells; i++)
{
double* J = jBatch.jVals(i);
J[0] = fabs(xMin_-xMax_);
}
}
}
void PeanoMesh3D::getJacobians(int cellDim, const Array<int>& cellLID,
CellJacobianBatch& jBatch) const
{
//printf("cellDim:%d _uniqueResolution:%f ",cellDim, _uniqueResolution);
SUNDANCE_VERB_HIGH("getJacobians()");
TEUCHOS_TEST_FOR_EXCEPTION(cellDim < 0 || cellDim > spatialDim(), std::logic_error,
"cellDim=" << cellDim << " is not in expected range [0, " << spatialDim() << "]");
int nCells = cellLID.size();
int tmp_index , tmp;
int tmp_index1 , tmp_index2;
Point pnt(0.0,0.0,0.0);
Point pnt1(0.0,0.0,0.0);
Point pnt2(0.0,0.0,0.0);
jBatch.resize(cellLID.size(), spatialDim(), cellDim);
if (cellDim < spatialDim()) // they need the Jacobian of a lower dinemsional element
{
//printf("PeanoMesh3D::getJacobians() cellDim < spatialDim() \n");
for (int i=0; i<nCells; i++)
{
//printf("PeanoMesh3D::getJacobian() cellDim < spatialDim() cellDim:%d , ret:%f \n",cellDim , _uniqueResolution);
double* detJ = jBatch.detJ(i);
switch(cellDim)
{
case 0: *detJ = 1.0;
break;
case 1:
tmp_index = this->facetLID(cellDim, cellLID[i] , 0 , 0 , tmp );
tmp_index1= this->facetLID(cellDim, cellLID[i] , 0 , 1 , tmp );
pnt = nodePosition(tmp_index);
pnt1 = nodePosition(tmp_index1);
pnt = pnt1 - pnt;
*detJ = sqrt(pnt * pnt); // the length of the edge
break;
case 2:{
tmp_index = this->facetLID(cellDim, cellLID[i] , 0 , 0 , tmp );
tmp_index1= this->facetLID(cellDim, cellLID[i] , 0 , 1 , tmp );
tmp_index2= this->facetLID(cellDim, cellLID[i] , 0 , 2 , tmp );
pnt = nodePosition(tmp_index);
pnt1 = nodePosition(tmp_index1);
pnt2 = nodePosition(tmp_index2);
Point directedArea = cross( pnt1 - pnt , pnt2 - pnt );
*detJ = sqrt(directedArea * directedArea); // the are of the face
break;}
default:
TEUCHOS_TEST_FOR_EXCEPTION(true, std::logic_error, "impossible switch value "
"cellDim=" << cellDim << " in PeanoMesh3D::getJacobians()");
}
}
}else{ // they request the complete Jacoby matrix for this bunch of elements
//Array<double> J(cellDim*cellDim);
SUNDANCE_VERB_HIGH("cellDim == spatialDim()");
for (unsigned int i=0; i<(unsigned int)cellLID.size(); i++)
{
//printf("PeanoMesh3D::getJacobian() cellDim == spatialDim() cellDim:%d , ret:%f \n",cellDim , _uniqueResolution);
double* J = jBatch.jVals(i);
switch(cellDim)
{
case 3:
J[0] = _peanoMesh->returnResolution(0);
J[1] = 0.0; J[2] = 0.0; J[3] = 0.0;
J[4] = _peanoMesh->returnResolution(1);
J[5] = 0.0; J[6] = 0.0; J[7] = 0.0;
J[8] = _peanoMesh->returnResolution(2); // the Jacobi of the tet
break;
default:
TEUCHOS_TEST_FOR_EXCEPTION(true, std::logic_error, "impossible switch value "
"cellDim=" << cellDim
<< " in PeanoMesh3D::getJacobians()");
}
}
}
}
void MaximalQuadratureIntegral::transformTwoForm(const CellJacobianBatch& JTrans,
const CellJacobianBatch& JVol,
const Array<int>& facetIndex,
const RCP<Array<int> >& cellLIDs,
const double* const coeff,
RCP<Array<double> >& A) const
{
TimeMonitor timer(maxCellQuadrature2Timer());
Tabs tabs;
TEUCHOS_TEST_FOR_EXCEPTION(order() != 2, std::logic_error,
"MaximalQuadratureIntegral::transformTwoForm() called for form "
"of order " << order());
SUNDANCE_MSG2(integrationVerb(), tabs << "doing one form by quadrature");
int nQuad = quadWeights_.size();
const Array<int>* cellLID = cellLIDs.get();
/* If the derivative orders are zero, the only thing to be done
* is to multiply by the cell's Jacobian determinant and sum over the
* quad points */
if (testDerivOrder() == 0 && unkDerivOrder() == 0)
{
double* aPtr = &((*A)[0]);
double* coeffPtr = (double*) coeff;
int offset = 0 ;
const Array<double>& w = W_;
if (globalCurve().isCurveValid())
{
int fc = 0;
Array<double> quadWeightsTmp = quadWeights_;
Array<Point> quadPointsTmp = quadPts_;
bool isCutByCurve;
for (int c=0; c<JVol.numCells(); c++, offset+=nNodes())
{
double detJ = fabs(JVol.detJ()[c]);
quadWeightsTmp = quadWeights_;
quadPointsTmp = quadPts_;
/* call the special integration routine */
quad_.getAdaptedWeights(cellType(), dim(), (*cellLID)[c] , fc ,
mesh() , globalCurve() , quadPointsTmp , quadWeightsTmp , isCutByCurve );
if (isCutByCurve)
{
Array<double> wi;
wi.resize(nQuad * nNodesTest() *nNodesUnk() ); //recalculate the special weights
for (int ii = 0 ; ii < wi.size() ; ii++ ) wi[ii] = 0.0;
/* Good coding practice: always use braces { } when nesting loops even if
* there's only one line. Otherwise, if someone inserts a new line
* (e.g., a print statement) it can totally change the code logic */
for (int nt = 0 ; nt < nNodesTest() ; nt++)
{
for (int nu=0; nu<nNodesUnk(); nu++)
{
for (int q=0 ; q < quadWeightsTmp.size() ; q++)
{
//Indexing: unkNode + nNodesUnk()*(testNode + nNodesTest()*(unkDerivDir + nRefDerivUnk()*(testDerivDir + nRefDerivTest()*q)))
wi[nu + nNodesUnk()*(nt + nNodesTest()*q)] +=
chop(quadWeightsTmp[q] * W_ACI_F2_[q][0][nt][0][nu]);
}
}
}
for (int q=0; q<nQuad; q++, coeffPtr++)
{
double f = (*coeffPtr)*detJ;
for (int n=0; n<nNodes(); n++)
{
aPtr[offset+n] += f*wi[n + nNodes()*q];
}
}
}// end isCutByCurve
else
{
for (int q=0; q<nQuad; q++, coeffPtr++)
{
double f = (*coeffPtr)*detJ;
for (int n=0; n<nNodes(); n++)
{
aPtr[offset+n] += f*w[n + nNodes()*q];
}
}
}
}
}
else // No ACI
{
for (int c=0; c<JVol.numCells(); c++, offset+=nNodes())
{
double detJ = fabs(JVol.detJ()[c]);
for (int q=0; q<nQuad; q++, coeffPtr++)
{
double f = (*coeffPtr)*detJ;
for (int n=0; n<nNodes(); n++)
{
aPtr[offset+n] += f*w[n + nNodes()*q];
}
}
}
}
addFlops( JVol.numCells() * (1 + nQuad * (1 + 2*nNodes())) );
}
else
{
createTwoFormTransformationMatrix(JTrans, JVol);
double* GPtr;
if (testDerivOrder() == 0)
{
GPtr = &(G(beta())[0]);
SUNDANCE_MSG2(transformVerb(),
Tabs() << "transformation matrix=" << G(beta()));
}
else if (unkDerivOrder() == 0)
{
GPtr = &(G(alpha())[0]);
SUNDANCE_MSG2(transformVerb(),
Tabs() << "transformation matrix=" << G(alpha()));
}
else
{
GPtr = &(G(alpha(), beta())[0]);
SUNDANCE_MSG2(transformVerb(),
Tabs() << "transformation matrix="
<< G(alpha(),beta()));
}
transformSummingFirst(JTrans.numCells(), facetIndex, cellLIDs, GPtr, coeff, A);
}
}
void MaximalQuadratureIntegral::transformOneForm(const CellJacobianBatch& JTrans,
const CellJacobianBatch& JVol,
const Array<int>& facetIndex,
const RCP<Array<int> >& cellLIDs,
const double* const coeff,
RCP<Array<double> >& A) const
{
TimeMonitor timer(maxCellQuadrature1Timer());
Tabs tabs;
TEUCHOS_TEST_FOR_EXCEPTION(order() != 1, std::logic_error,
"MaximalQuadratureIntegral::transformOneForm() called for form "
"of order " << order());
SUNDANCE_MSG2(integrationVerb(), tabs << "doing one form by quadrature");
int flops = 0;
const Array<int>* cellLID = cellLIDs.get();
int nQuad = quadWeights_.size();
/* If the derivative order is zero, the only thing to be done
* is to multiply by the cell's Jacobian determinant and sum over the
* quad points */
if (testDerivOrder() == 0)
{
double* aPtr = &((*A)[0]);
SUNDANCE_MSG5(integrationVerb(), tabs << "input A = ");
if (integrationVerb() >= 5) writeTable(Out::os(), tabs, *A, 6);
double* coeffPtr = (double*) coeff;
int offset = 0 ;
const Array<double>& w = W_;
if (globalCurve().isCurveValid()) /* ----- ACI logic ---- */
{
Array<double> quadWeightsTmp = quadWeights_;
Array<Point> quadPointsTmp = quadPts_;
bool isCutByCurve;
for (int c=0; c<JVol.numCells(); c++, offset+=nNodes())
{
Tabs tab2;
double detJ = fabs(JVol.detJ()[c]);
int fc = 0;
SUNDANCE_MSG4(integrationVerb(), tab2 << "c=" << c << " detJ=" << detJ);
/* call the special integration routine */
quad_.getAdaptedWeights(cellType(), dim(), (*cellLID)[c] , fc ,
mesh() , globalCurve() , quadPointsTmp , quadWeightsTmp , isCutByCurve );
if (isCutByCurve)
{
Array<double> wi;
wi.resize(nQuad * nNodes()); //recalculate the special weights
for (int ii = 0 ; ii < wi.size() ; ii++ ) wi[ii] = 0.0;
for (int n = 0 ; n < nNodes() ; n++)
{
for (int q=0 ; q < quadWeightsTmp.size() ; q++)
{
//Indexing: testNode + nNodesTest()*(testDerivDir + nRefDerivTest()*q)
wi[n + nNodes()*q] +=
chop(quadWeightsTmp[q] * W_ACI_F1_[q][0][n]);
}
}
// if it is cut by curve then use this vector
for (int q=0; q<nQuad; q++, coeffPtr++)
{
double f = (*coeffPtr)*detJ;
for (int n=0; n<nNodes(); n++)
{
aPtr[offset+n] += f*wi[n + nNodes()*q];
}
}
} // end isCutByCurve
else
{
for (int q=0; q<nQuad; q++, coeffPtr++)
{
double f = (*coeffPtr)*detJ;
for (int n=0; n<nNodes(); n++)
{
aPtr[offset+n] += f*w[n + nNodes()*q];
}
}
}
if (integrationVerb() >= 4)
{
Out::os() << tab2 << "integration results on cell:" << std::endl;
Out::os() << tab2 << setw(10) << "n" << setw(12) << "I_n" << std::endl;
for (int n=0; n<nNodes(); n++)
{
Out::os() << tab2 << setw(10) << n
<< setw(12) << setprecision(6) << aPtr[offset+n] << std::endl;
}
}
}
}
else /* -------- No ACI -------- */
{
for (int c=0; c<JVol.numCells(); c++, offset+=nNodes())
{
Tabs tab2;
double detJ = fabs(JVol.detJ()[c]);
SUNDANCE_MSG4(integrationVerb(), tab2 << "c=" << c << " detJ=" << detJ);
for (int q=0; q<nQuad; q++, coeffPtr++)
{
Tabs tab3;
double f = (*coeffPtr)*detJ;
SUNDANCE_MSG4(integrationVerb(), tab3 << "q=" << q << " coeff=" <<
*coeffPtr << " coeff*detJ=" << f);
for (int n=0; n<nNodes(); n++)
{
Tabs tab4;
SUNDANCE_MSG4(integrationVerb(), tab4 << "n=" << n << " w=" <<
w[n + nNodes()*q]);
aPtr[offset+n] += f*w[n + nNodes()*q];
}
}
if (integrationVerb() >= 4)
{
Out::os() << tab2 << "integration results on cell:" << std::endl;
Out::os() << tab2 << setw(10) << "n" << setw(12) << "I_n" << std::endl;
for (int n=0; n<nNodes(); n++)
{
Out::os() << tab2 << setw(10) << n
<< setw(12) << setprecision(6) << aPtr[offset+n] << std::endl;
}
}
}
}
SUNDANCE_MSG5(integrationVerb(), tabs << "output A = ");
if (integrationVerb() >= 5) writeTable(Out::os(), tabs, *A, 6);
}
else
{
/* If the derivative order is nonzero, then we have to do a transformation. */
createOneFormTransformationMatrix(JTrans, JVol);
SUNDANCE_MSG4(transformVerb(),
Tabs() << "transformation matrix=" << G(alpha()));
double* GPtr = &(G(alpha())[0]);
transformSummingFirst(JVol.numCells(), facetIndex, cellLIDs, GPtr, coeff, A);
}
addFlops(flops);
}
void MaximalQuadratureIntegral
::transformZeroForm(const CellJacobianBatch& JTrans,
const CellJacobianBatch& JVol,
const Array<int>& isLocalFlag,
const Array<int>& facetIndex,
const RCP<Array<int> >& cellLIDs,
const double* const coeff,
RCP<Array<double> >& A) const
{
TimeMonitor timer(maxCellQuadrature0Timer());
Tabs tabs;
SUNDANCE_MSG1(integrationVerb(), tabs << "doing zero form by quadrature");
TEUCHOS_TEST_FOR_EXCEPTION(order() != 0, std::logic_error,
"MaximalQuadratureIntegral::transformZeroForm() called "
"for form of order " << order());
TEUCHOS_TEST_FOR_EXCEPTION( (int) isLocalFlag.size() != 0
&& (int) isLocalFlag.size() != JVol.numCells(),
std::runtime_error,
"mismatch between isLocalFlag.size()=" << isLocalFlag.size()
<< " and JVol.numCells()=" << JVol.numCells());
bool checkLocalFlag = (int) isLocalFlag.size() != 0;
const Array<int>* cellLID = cellLIDs.get();
int nQuad = quadWeights_.size();
double& a = (*A)[0];
SUNDANCE_MSG5(integrationVerb(), tabs << "input A=");
if (integrationVerb() >= 5) writeTable(Out::os(), tabs, *A, 6);
double* coeffPtr = (double*) coeff;
const Array<double>& w = W_;
if (globalCurve().isCurveValid()) /* ---------- ACI ------------- */
{
Array<double> quadWeightsTmp = quadWeights_;
Array<Point> quadPointsTmp = quadPts_;
int fc = 0;
bool isCutByCurve;
for (int c=0; c<JVol.numCells(); c++)
{
if (checkLocalFlag && !isLocalFlag[c])
{
coeffPtr += nQuad;
continue;
}
double detJ = fabs(JVol.detJ()[c]);
quad_.getAdaptedWeights(cellType(), dim(), (*cellLID)[c], fc ,mesh(),
globalCurve(), quadPointsTmp, quadWeightsTmp, isCutByCurve);
/* if we have special weights then do the same as before */
if (isCutByCurve)
{
for (int q=0; q<nQuad; q++, coeffPtr++)
{
a += quadWeightsTmp[q]*(*coeffPtr)*detJ;
}
} // end cut by curve
else
{
for (int q=0; q<nQuad; q++, coeffPtr++)
{
a += w[q]*(*coeffPtr)*detJ;
}
}
}
}
else /* --------- No ACI ------------- */
{
for (int c=0; c<JVol.numCells(); c++)
{
if (checkLocalFlag && !isLocalFlag[c])
{
coeffPtr += nQuad;
continue;
}
double detJ = fabs(JVol.detJ()[c]);
for (int q=0; q<nQuad; q++, coeffPtr++)
{
a += w[q]*(*coeffPtr)*detJ;
}
}
}
SUNDANCE_MSG5(integrationVerb(), tabs << "output A = ");
if (integrationVerb() >= 5) writeTable(Out::os(), tabs, *A, 6);
SUNDANCE_MSG1(integrationVerb(), tabs << "done zero form");
}
void ElementIntegral
::createTwoFormTransformationMatrix(const CellJacobianBatch& JTrans,
const CellJacobianBatch& JVol) const
{
TimeMonitor timer(transCreationTimer());
Tabs tab;
int flops = 0;
int maxDim = JTrans.cellDim();
int cellDim = JVol.cellDim();
if (testDerivOrder() == 1 && unkDerivOrder() == 1)
{
Tabs tab2;
if (transformationMatrixIsValid(alpha(), beta())) return;
transformationMatrixIsValid(alpha(), beta()) = true;
G(alpha(), beta()).resize(JTrans.numCells() * JTrans.cellDim() * JTrans.cellDim());
double* GPtr = &(G(alpha(),beta())[0]);
int k = 0;
for (int c=0; c<JTrans.numCells(); c++)
{
static Array<double> invJ;
JTrans.getInvJ(c, invJ);
double detJ = fabs(JVol.detJ()[c]);
for (int gamma=0; gamma<maxDim; gamma++)
{
for (int delta=0; delta<maxDim; delta++, k++)
{
GPtr[k] = detJ*invJ[alpha() + gamma*maxDim]
* invJ[beta() + maxDim*delta];
}
}
}
flops = 2 * JTrans.numCells() * maxDim * maxDim + JTrans.numCells();
}
else if (testDerivOrder() == 1 && unkDerivOrder() == 0)
{
if (transformationMatrixIsValid(alpha())) return;
transformationMatrixIsValid(alpha()) = true;
G(alpha()).resize(JTrans.numCells() * JTrans.cellDim());
int k = 0;
double* GPtr = &(G(alpha())[0]);
for (int c=0; c<JTrans.numCells(); c++)
{
static Array<double> invJ;
JTrans.getInvJ(c, invJ);
double detJ = fabs(JVol.detJ()[c]);
for (int gamma=0; gamma<maxDim; gamma++,k++)
{
GPtr[k] = detJ*invJ[alpha() + maxDim * gamma];
}
}
flops = JTrans.numCells() * maxDim + JTrans.numCells();
}
else
{
if (transformationMatrixIsValid(beta())) return;
transformationMatrixIsValid(beta()) = true;
G(beta()).resize(JTrans.numCells() * JTrans.cellDim());
int k = 0;
double* GPtr = &(G(beta())[0]);
for (int c=0; c<JTrans.numCells(); c++)
{
static Array<double> invJ;
JTrans.getInvJ(c, invJ);
double detJ = fabs(JVol.detJ()[c]);
for (int gamma=0; gamma<maxDim; gamma++,k++)
{
GPtr[k] = detJ*invJ[beta() + maxDim * gamma];
}
}
flops = JTrans.numCells() * maxDim + JTrans.numCells();
}
addFlops(flops);
}
void RefIntegral::transformTwoForm(const CellJacobianBatch& JTrans,
const CellJacobianBatch& JVol,
const Array<int>& facetIndex,
const RCP<Array<int> >& cellLIDs,
const double& coeff,
RCP<Array<double> >& A) const
{
TimeMonitor timer(ref2IntegrationTimer());
TEUCHOS_TEST_FOR_EXCEPTION(order() != 2, std::logic_error,
"RefIntegral::transformTwoForm() called for form "
"of order " << order());
Tabs tabs;
SUNDANCE_MSG1(transformVerb(), tabs << "doing two form by reference");
int nfc = nFacetCases();
SUNDANCE_MSG1(transformVerb(), tabs << "doing two form by reference ... ");
/* If the derivative orders are zero, the only transformation to be done
* is to multiply by the cell's Jacobian determinant */
if (testDerivOrder() == 0 && unkDerivOrder() == 0)
{
if (globalCurve().isCurveValid())
{ /* ----------- ACI logic ------------ */
Array<double> quadWeightsTmp = quadWeights_;
Array<Point> quadPointsTmp = quadPts_;
bool isCutByCurve = false;
double* aPtr = &((*A)[0]);
int count = 0;
for (int c=0; c<JVol.numCells(); c++)
{
int fc = 0;
if (nFacetCases() != 1) fc = facetIndex[c];
/* ---------- ACI ----------- */
/* call the special integration routine */
quadWeightsTmp = quadWeights_;
quadPointsTmp = quadPts_;
quad_.getAdaptedWeights(cellType(), dim(), (*cellLIDs)[c] , fc ,
mesh(), globalCurve(), quadPointsTmp, quadWeightsTmp, isCutByCurve);
if (isCutByCurve){
Array<double> w;
int ci = 0;
w.resize(nNodesTest()*nNodesUnk()); //recalculate the special weights
for (int nt = 0 ; nt < nNodesTest() ; nt++)
for(int nu=0 ; nu < nNodesUnk() ; nu++ , ci++){
w[ci] = 0.0;
for (int q=0 ; q < quadWeightsTmp.size() ; q++)
w[ci] += chop( quadWeightsTmp[q] * W_ACI_F2_[fc][q][0][nt][0][nu] );
}
// do the integration here
double detJ = coeff * fabs(JVol.detJ()[c]);
for (int n=0; n<nNodes(); n++, count++)
{
aPtr[count] += detJ*w[n];
}
}
else
{
const Array<double>& w = W_[fc];
double detJ = coeff * fabs(JVol.detJ()[c]);
for (int n=0; n<nNodes(); n++, count++)
{
aPtr[count] += detJ*w[n];
}
}
}
}
else /* ---------- NO ACI logic----------- */
{
double* aPtr = &((*A)[0]);
int count = 0;
for (int c=0; c<JVol.numCells(); c++)
{
int fc = 0;
if (nFacetCases() != 1) fc = facetIndex[c];
const Array<double>& w = W_[fc];
double detJ = coeff * fabs(JVol.detJ()[c]);
for (int n=0; n<nNodes(); n++, count++)
{
aPtr[count] += detJ*w[n];
}
}
}
addFlops(JVol.numCells() * (nNodes() + 1));
}
else
{
/* If the derivative order is nonzero, then we have to do a transformation.
* If we're also on a cell of dimension lower than maximal, we need to refer
* to the facet index of the facet being integrated. */
int nCells = JVol.numCells();
double one = 1.0;
int nTransRows = nRefDerivUnk()*nRefDerivTest();
createTwoFormTransformationMatrix(JTrans, JVol);
double* GPtr;
if (testDerivOrder() == 0)
{
GPtr = &(G(beta())[0]);
SUNDANCE_MSG2(transformVerb(),
Tabs() << "transformation matrix=" << G(beta()));
}
else if (unkDerivOrder() == 0)
{
GPtr = &(G(alpha())[0]);
SUNDANCE_MSG2(transformVerb(),
Tabs() << "transformation matrix=" << G(alpha()));
}
else
{
GPtr = &(G(alpha(), beta())[0]);
SUNDANCE_MSG2(transformVerb(),
Tabs() << "transformation matrix="
<< G(alpha(),beta()));
}
int nNodes0 = nNodes();
if (nFacetCases()==1)
{
/* if we're on a maximal cell, we can do transformations
* for all cells in a single batch.
*/
if (globalCurve().isCurveValid())
{ /* ---------- ACI logic ----------- */
Array<double> quadWeightsTmp = quadWeights_;
Array<Point> quadPointsTmp = quadPts_;
bool isCutByCurve = false;
for (int c=0; c<JVol.numCells(); c++)
{
int fc = 0;
if (nfc != 1) fc = facetIndex[c];
double* aPtr = &((*A)[c*nNodes0]);
double* gPtr = &(GPtr[c*nTransRows]);
int oneI = 1;
/* call the special integration routine */
//SUNDANCE_MSG1(transformVerb(), tabs << "before quad_.getAdaptedWeights");
quadWeightsTmp = quadWeights_;
quadPointsTmp = quadPts_;
quad_.getAdaptedWeights(cellType(), dim(), (*cellLIDs)[c], fc ,
mesh(),globalCurve(), quadPointsTmp, quadWeightsTmp, isCutByCurve);
//SUNDANCE_MSG1(transformVerb(), tabs << "after quad_.getAdaptedWeights");
if (isCutByCurve){
Array<double> w;
w.resize(nNodesUnk()*nNodesTest()*nRefDerivUnk()*nRefDerivTest());
for ( int i = 0 ; i < w.size() ; i++) w[i] = 0.0;
//recalculate the special weights
for (int t=0; t<nRefDerivTest(); t++){
for (int nt=0; nt<nNodesTest(); nt++)
for (int u=0; u<nRefDerivUnk(); u++)
for (int nu=0; nu<nNodesUnk(); nu++)
for (int q=0 ; q < quadWeightsTmp.size() ; q++)
// unkNode + nNodesUnk()*testNode + nNodes()*(unkDerivDir + nRefDerivUnk()*testDerivDir)
w[nu + nNodesUnk()*nt + nNodes()*(u + nRefDerivUnk()*t)] +=
chop(quadWeightsTmp[q]*W_ACI_F2_[0][q][t][nt][u][nu]);
}
::dgemm_("N", "N", &nNodes0, &oneI , &nTransRows, &coeff, &(w[0]),
&nNodes0, &(gPtr[0]), &nTransRows, &one,
&(aPtr[0]), &nNodes0);
}else{
::dgemm_("N", "N", &nNodes0, &oneI , &nTransRows, &coeff, &(W_[0][0]),
&nNodes0, &(gPtr[0]), &nTransRows, &one,
&(aPtr[0]), &nNodes0);
}
} // end from the for loop over the cells
}
else /* ---------- NO ACI ----------- */
{
::dgemm_("N", "N", &nNodes0, &nCells, &nTransRows, &coeff, &(W_[0][0]),
&nNodes0, GPtr, &nTransRows, &one,
&((*A)[0]), &nNodes0);
}
}
else
{
/* If we're on a lower-dimensional cell and have to transform,
* we've got to do each transformation using a different facet case */
if (globalCurve().isCurveValid())
{ /* ---------- ACI logic ----------- */
int oneI = 1;
Array<double> quadWeightsTmp = quadWeights_;
Array<Point> quadPointsTmp = quadPts_;
bool isCutByCurve = false;
for (int c=0; c<JVol.numCells(); c++)
{
int fc = 0;
if (nfc != 1) fc = facetIndex[c];
double* aPtr = &((*A)[c*nNodes0]);
double* gPtr = &(GPtr[c*nTransRows]);
SUNDANCE_MSG2(integrationVerb(),
tabs << "c=" << c << ", facet case=" << fc
<< " W=" << W_[fc]);
/* call the special integration routine */
quadWeightsTmp = quadWeights_;
quadPointsTmp = quadPts_;
quad_.getAdaptedWeights(cellType(), dim(), (*cellLIDs)[c], fc ,
mesh(), globalCurve(), quadPointsTmp, quadWeightsTmp, isCutByCurve);
if (isCutByCurve){
Array<double> w;
w.resize(nNodesUnk()*nNodesTest()*nRefDerivUnk()*nRefDerivTest());
for ( int i = 0 ; i < w.size() ; i++) w[i] = 0.0;
//recalculate the special weights
for (int t=0; t<nRefDerivTest(); t++){
for (int nt=0; nt<nNodesTest(); nt++)
for (int u=0; u<nRefDerivUnk(); u++)
for (int nu=0; nu<nNodesUnk(); nu++)
for (int q=0 ; q < quadWeightsTmp.size() ; q++)
// unkNode + nNodesUnk()*testNode + nNodes()*(unkDerivDir + nRefDerivUnk()*testDerivDir)
w[nu + nNodesUnk()*nt + nNodes()*(u + nRefDerivUnk()*t)] +=
chop( quadWeightsTmp[q]*W_ACI_F2_[fc][q][t][nt][u][nu] );
}
::dgemm_("N", "N", &nNodes0, &oneI , &nTransRows, &coeff, &(w[0]),
&nNodes0, &(gPtr[0]), &nTransRows, &one,
&(aPtr[0]), &nNodes0);
}else{
::dgemm_("N", "N", &nNodes0, &oneI , &nTransRows, &coeff, &(W_[fc][0]),
&nNodes0, &(gPtr[0]), &nTransRows, &one,
&(aPtr[0]), &nNodes0);
}
}
}
else /* ---------- NO ACI ----------- */
{
int N = 1;
for (int c=0; c<JVol.numCells(); c++)
{
int fc = 0;
if (nfc != 1) fc = facetIndex[c];
double* aPtr = &((*A)[c*nNodes0]);
double* gPtr = &(GPtr[c*nTransRows]);
SUNDANCE_MSG2(integrationVerb(),
tabs << "c=" << c << ", facet case=" << fc
<< " W=" << W_[fc]);
::dgemm_("N", "N", &nNodes0, &N, &nTransRows, &coeff, &(W_[fc][0]),
&nNodes0, gPtr, &nTransRows, &one,
aPtr, &nNodes0);
}
}
}// from else of (nFacetCases()==1)
addFlops(2 * nNodes0 * nCells * nTransRows);
}
}
void RefIntegral::transformZeroForm(const CellJacobianBatch& JVol,
const Array<int>& isLocalFlag,
const RCP<Array<int> >& cellLIDs,
const double& coeff,
RCP<Array<double> >& A) const
{
TimeMonitor timer(ref0IntegrationTimer());
TEUCHOS_TEST_FOR_EXCEPTION(order() != 0, std::logic_error,
"RefIntegral::transformZeroForm() called "
"for form of order " << order());
Tabs tabs;
SUNDANCE_MSG1(integrationVerb(), tabs << "doing zero form by reference");
double& a = (*A)[0];
int flops = 0;
const Array<int>* cellLID = cellLIDs.get();
/* if we don't need to check whether elements are local, we
* can streamline the loop. This will be the case when
* we are evaluating a functional but not its gradient */
double w = coeff * W_[0][0];
if ((int) isLocalFlag.size()==0)
{
if (globalCurve().isCurveValid())
{ /* ---------- ACI logic ----------- */
Array<double> quadWeightsTmp = quadWeights_;
Array<Point> quadPointsTmp = quadPts_;
bool isCutByCurve;
for (int c=0; c<JVol.numCells(); c++)
{
int fc = 0;
/* call the special integration routine */
quadWeightsTmp = quadWeights_;
quadPointsTmp = quadPts_;
quad_.getAdaptedWeights(cellType(), dim(), (*cellLID)[c], fc ,mesh(),
globalCurve(), quadPointsTmp, quadWeightsTmp, isCutByCurve);
/* if we have special weights then do the same as before */
if (isCutByCurve){
double sumweights = 0;
for (int j=0; j < quadWeightsTmp.size(); j++) sumweights += chop(quadWeightsTmp[j]);
flops+=3+quadWeightsTmp.size(); //Todo: the curve stuff not counted
a += coeff * sumweights * fabs(JVol.detJ()[c]);
} else {
flops+=2; //Todo: the curve stuff not counted
a += w * fabs(JVol.detJ()[c]);
}
}
}
else /* -------- NO ACI logic ------- */
{
for (int c=0; c<JVol.numCells(); c++)
{
flops+=2;
a += w * fabs(JVol.detJ()[c]);
}
}
}
else
{
TEUCHOS_TEST_FOR_EXCEPTION( (int) isLocalFlag.size() != JVol.numCells(),
std::runtime_error,
"mismatch between isLocalFlag.size()="
<< isLocalFlag.size()
<< " and JVol.numCells()=" << JVol.numCells());
int fc = 0;
if (globalCurve().isCurveValid())
{ /* ---------- ACI logic ----------- */
Array<double> quadWeightsTmp = quadWeights_;
Array<Point> quadPointsTmp = quadPts_;
bool isCutByCurve;
for (int c=0; c<JVol.numCells(); c++)
{
if (isLocalFlag[c])
{
/* call the special integration routine */
quadWeightsTmp = quadWeights_;
quadPointsTmp = quadPts_;
quad_.getAdaptedWeights(cellType(), dim(), (*cellLID)[c], fc , mesh(),
globalCurve(), quadPointsTmp, quadWeightsTmp, isCutByCurve);
/* if we do not have special weights then do the same as before */
if (isCutByCurve){
double sumweights = 0;
for (int j=0; j < quadWeightsTmp.size(); j++) sumweights += chop(quadWeightsTmp[j]);
flops+=3+quadWeightsTmp.size(); //Todo: the curve stuff not counted
a += coeff * sumweights * fabs(JVol.detJ()[c]);
} else {
flops+=2; //Todo: the curve stuff not counted
a += w * fabs(JVol.detJ()[c]);
}
}
}
}
else /* ---------- NO ACI logic ----------- */
{
for (int c=0; c<JVol.numCells(); c++)
{
if (isLocalFlag[c])
{
flops+=2;
a += w * fabs(JVol.detJ()[c]);
}
}
}
}
addFlops(flops);
}