void TrilinosFEVbrMatrix::get_values(BlockAccumulator& values)
{
cf3_assert(m_is_created);
Epetra_SerialDenseMatrix **val;
int* colindices;
int blockrowsize;
int dummyneq;
int hits=0;
const int numblocks=values.indices.size();
const int rowoffset=(numblocks-1)*m_neq;
const int neqneq=m_neq*m_neq;
if (m_converted_indices.size()<numblocks) m_converted_indices.resize(numblocks);
for (int i=0; i<(const int)numblocks; i++) m_converted_indices[i]=m_p2m[values.indices[i]];
int* idxs=(int*)&m_converted_indices[0];
values.mat.setConstant(0.);
for (int irow=0; irow<(const int)numblocks; irow++)
{
if (idxs[irow]<m_blockrow_size)
{
hits=0;
TRILINOS_ASSERT(m_mat->ExtractMyBlockRowView(idxs[irow],dummyneq,blockrowsize,colindices,val));
for (int j=0; j<(const int)blockrowsize; j++)
{
for (int icol=0; icol<(const int)numblocks; icol++)
if (colindices[j]==idxs[icol])
{
double *emv=val[j][0].A();
int col_idx = icol*m_neq;
for (double* l=emv; emv<(const double*)(l+neqneq); ++col_idx)
{
int row_idx = irow*m_neq;
for (double* m=emv; emv<(const double*)(m+m_neq);)
values.mat(row_idx++, col_idx) = *emv++;
}
hits++;
//break; // this is arguable, can be that the blockaccumulator fills to the same id more than once (repetitive same id entries in values.indices)
}
if (hits==numblocks) break;
}
}
}
}
void TrilinosCrsMatrix::get_values(BlockAccumulator& values)
{
cf3_assert(m_is_created);
values.mat.setZero();
const Uint nb_nodes = values.indices.size();
const int num_entries = nb_nodes*m_neq;
int extracted_num_entries;
Real* extracted_values;
int* extracted_indices;
cf3_assert(values.mat.rows() == num_entries);
std::map<int, int> reverse_idx_map;
// Convert the index vector
for(Uint i = 0; i != nb_nodes; ++i)
{
const Uint local_start_idx = values.indices[i]*m_neq;
for(int j = 0; j != m_neq; ++j)
{
m_converted_indices[i*m_neq+j] = m_p2m[local_start_idx+j];
reverse_idx_map[m_p2m[local_start_idx+j]] = i*m_neq + j;
}
}
// get the values
for(Uint i = 0; i != nb_nodes; ++i)
{
for(int j = 0; j != m_neq; ++j)
{
if(m_converted_indices[i*m_neq+j] >= m_num_my_elements)
continue;
TRILINOS_THROW(m_mat->ExtractMyRowView(m_converted_indices[i*m_neq+j], extracted_num_entries, extracted_values, extracted_indices));
for(int k = 0; k != extracted_num_entries; ++k)
{
const std::map<int,int>::const_iterator it = reverse_idx_map.find(extracted_indices[k]);
if(it != reverse_idx_map.end())
values.mat(i*m_neq+j, it->second) = extracted_values[k];
}
}
}
}
void TrilinosMatrix::add_values(const BlockAccumulator& values)
{
/* TRILINOS-ADVICED
cf3_assert(m_is_created);
const int numblocks=values.indices.size();
if (m_converted_indices.size()<numblocks) m_converted_indices.resize(numblocks);
for (int i=0; i<(const int)numblocks; i++) m_converted_indices[i]=m_p2m[values.indices[i]];
int* idxs=(int*)&m_converted_indices[0];
for (int irow=0; irow<(const int)numblocks; irow++)
if (idxs[irow]<m_blockrow_size)
{
TRILINOS_ASSERT(m_mat->BeginSumIntoMyValues(idxs[irow],numblocks,idxs));
for (int icol=0; icol<numblocks; icol++)
TRILINOS_ASSERT(m_mat->SubmitBlockEntry((double*)values.mat.data()+irow*m_neq+icol*m_neq*m_neq*numblocks,numblocks*m_neq,m_neq,m_neq));
TRILINOS_ASSERT(m_mat->EndSubmitEntries());
}
*/
/* MANUAL FILL
cf3_assert(m_is_created);
Epetra_SerialDenseMatrix **val;
int* colindices;
int blockrowsize;
int dummyneq;
int hits=0;
const int numblocks=values.indices.size();
if (m_converted_indices.size()<numblocks) m_converted_indices.resize(numblocks);
for (int i=0; i<(const int)numblocks; i++) m_converted_indices[i]=m_p2m[values.indices[i]];
int* idxs=(int*)&m_converted_indices[0];
for (int irow=0; irow<(const int)numblocks; irow++)
{
if (idxs[irow]<m_blockrow_size)
{
TRILINOS_ASSERT(m_mat->ExtractMyBlockRowView(idxs[irow],dummyneq,blockrowsize,colindices,val));
for (int j=0; j<(const int)blockrowsize; j++)
for (int k=0; k<(const int)numblocks; k++)
if (colindices[j]==idxs[k])
for (int l=0; l<(const int)m_neq; l++)
for (int m=0; m<(const int)m_neq; m++)
val[j][0](l,m)+=values.mat(irow*m_neq+l,k*m_neq+m);
hits++;
if (hits==numblocks) break;
}
}
*/
/* FINAL OPTIMIZED */
cf3_assert(m_is_created);
Epetra_SerialDenseMatrix **val;
int* colindices;
int blockrowsize;
int dummyneq;
int hits=0;
const int numblocks=values.indices.size();
const int rowoffset=(numblocks-1)*m_neq;
const int neqneq=m_neq*m_neq;
if (m_converted_indices.size()<numblocks) m_converted_indices.resize(numblocks);
for (int i=0; i<(const int)numblocks; i++) m_converted_indices[i]=m_p2m[values.indices[i]];
int* idxs=(int*)&m_converted_indices[0];
for (int irow=0; irow<(const int)numblocks; irow++)
{
if (idxs[irow]<m_blockrow_size)
{
hits=0;
TRILINOS_ASSERT(m_mat->ExtractMyBlockRowView(idxs[irow],dummyneq,blockrowsize,colindices,val));
for (int j=0; j<(const int)blockrowsize; j++)
{
for (int icol=0; icol<(const int)numblocks; icol++)
if (colindices[j]==idxs[icol])
{
double *emv=val[j][0].A();
double *bav=(double*)&values.mat(irow*m_neq,icol*m_neq);
for (double* l=emv; emv<(const double*)(l+neqneq); bav+=rowoffset)
for (double* m=emv; emv<(const double*)(m+m_neq);)
*emv++ += *bav++;
hits++;
//break; // this is arguable, can be that the blockaccumulator fills to the same id more than once (repetitive same id entries in values.indices)
}
if (hits==numblocks) break;
}
}
}
}
