Page MenuHomec4science

PerAtomQuantity-inl.h
No OneTemporary

File Metadata

Created
Thu, Nov 7, 14:43

PerAtomQuantity-inl.h

#ifndef PER_ATOM_QUANTITY_INL_H
#define PER_ATOM_QUANTITY_INL_H
#include <string>
namespace ATC {
//--------------------------------------------------------
//--------------------------------------------------------
// Class PerAtomQuantity
//--------------------------------------------------------
//--------------------------------------------------------
//--------------------------------------------------------
// Constructor
//--------------------------------------------------------
template <typename T>
PerAtomQuantity<T>::PerAtomQuantity(ATC_Method * atc,
int nCols,
AtomType atomType) :
MatrixDependencyManager<DenseMatrix, T>(),
atc_(atc,atomType),
lammpsInterface_(LammpsInterface::instance()),
atomType_(atomType),
nCols_(nCols),
quantityToLammps_(atc_.atc_to_lammps_map()),
lammpsScalar_(NULL),
lammpsVector_(NULL)
{
// do nothing
}
//--------------------------------------------------------
// Destructor
//--------------------------------------------------------
template <typename T>
PerAtomQuantity<T>::~PerAtomQuantity()
{
if (lammpsScalar_) lammpsInterface_->destroy_array(lammpsScalar_);
if (lammpsVector_) lammpsInterface_->destroy_array(lammpsVector_);
}
//--------------------------------------------------------
// set_lammps_to_quantity
//--------------------------------------------------------
template <typename T>
void PerAtomQuantity<T>::set_lammps_to_quantity() const
{
const DenseMatrix<T> & myQuantity(this->quantity_); // necessary to access quantity_ this way because of templating
if (myQuantity.nRows()>0) {
// full matrix copy
if (atomType_ == ALL || atomType_ == PROC_GHOST) {
if (nCols_==1) { // scalar
T * lammpsQuantity = this->lammps_scalar();
for (int i = 0; i < atc_.nlocal_total(); i++)
lammpsQuantity[i] = myQuantity(i,0);
}
else{ // vector
T ** lammpsQuantity = this->lammps_vector();
for (int i = 0; i < atc_.nlocal_total(); i++)
for (int j = 0; j < nCols_; j++)
lammpsQuantity[i][j] = myQuantity(i,j);
}
}
// map quantities
else {
int atomIndex;
const Array<int> & quantityToLammps = atc_.atc_to_lammps_map();
if (nCols_==1) { // scalar
T * lammpsQuantity = this->lammps_scalar();
for (int i = 0; i < myQuantity.nRows(); i++) {
atomIndex = quantityToLammps(i);
lammpsQuantity[atomIndex] = myQuantity(i,0);
}
}
else{ // vector
T ** lammpsQuantity = this->lammps_vector();
for (int i = 0; i < myQuantity.nRows(); i++) {
atomIndex = quantityToLammps(i);
for (int j = 0; j < nCols_; j++) {
lammpsQuantity[atomIndex][j] = myQuantity(i,j);
}
}
}
}
}
}
//--------------------------------------------------------
// set_quantity_to_lammps
//--------------------------------------------------------
template <typename T>
void PerAtomQuantity<T>::set_quantity_to_lammps() const
{
DenseMatrix<T> & myQuantity(this->quantity_);
if (myQuantity.nRows()>0) {
// full matrix copy
// in the case where processor ghosts are in the quantity, don't set them back
if (atomType_ == ALL || atomType_ == PROC_GHOST) {
if (nCols_==1) { // scalar
const T * lammpsQuantity = this->lammps_scalar();
for (int i = 0; i < myQuantity.nRows(); i++) {
myQuantity(i,0) = lammpsQuantity[i];
}
}
else {
const T * const * lammpsQuantity = this->lammps_vector();
for (int i = 0; i < myQuantity.nRows(); i++) {
for (int j = 0; j < nCols_; j++) {
myQuantity(i,j) = lammpsQuantity[i][j];
}
}
}
}
// map quantities
else {
const Array<int> & quantityToLammps = atc_.atc_to_lammps_map();
int atomIndex;
if (nCols_==1) { // scalar
const T * lammpsQuantity = this->lammps_scalar();
for (int i = 0; i < myQuantity.nRows(); i++) {
atomIndex = quantityToLammps(i);
myQuantity(i,0) = lammpsQuantity[atomIndex];
}
}
else {
const T * const * lammpsQuantity = this->lammps_vector();
for (int i = 0; i < myQuantity.nRows(); i++) {
atomIndex = quantityToLammps(i);
for (int j = 0; j < nCols_; j++) {
myQuantity(i,j) = lammpsQuantity[atomIndex][j];
}
}
}
}
}
}
//-----------------------------------------------------------------
// pack values in local atom-based arrays for exchange with another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomQuantity<T>::pack_exchange(int i, double *buffer)
{
if (nCols_ == 1)
buffer[0] = static_cast<double>(this->lammps_scalar()[i]);
else
for (int j = 0; j < nCols_; j++) {
T ** lammpsVector = this->lammps_vector();
buffer[j] = static_cast<double>(lammpsVector[i][j]);
}
return nCols_;
}
//-----------------------------------------------------------------
// unpack values in local atom-based arrays from exchange with another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomQuantity<T>::unpack_exchange(int i, double *buffer)
{
if (nCols_ == 1)
this->lammps_scalar()[i] = static_cast<T>(buffer[0]);
else
for (int j = 0; j < nCols_; j++) {
T ** lammpsVector = this->lammps_vector();
lammpsVector[i][j] = static_cast<T>(buffer[j]);
}
return nCols_;
}
//-----------------------------------------------------------------
// pack values in local atom-based arrays for passing to ghosts on another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomQuantity<T>::pack_comm(int index, double *buf,
int pbc_flag, int *pbc)
{
if (this->need_reset()) this->reset();
DenseMatrix<T> & myQuantity(this->quantity_);
for (int k = 0; k < nCols_; k++) {
buf[k] = static_cast<double>(myQuantity(index,k));
}
return nCols_;
}
//-----------------------------------------------------------------
// unpack values in local atom-based arrays for passing to ghosts on another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomQuantity<T>::unpack_comm(int index, double *buf)
{
DenseMatrix<T> & myQuantity(this->quantity_);
for (int k = 0; k < nCols_; k++) {
myQuantity(index,k) = static_cast<T>(buf[k]);
}
this->propagate_reset();
return nCols_;
}
//-----------------------------------------------------------------
// allocate local atom-based arrays
//-----------------------------------------------------------------
template <typename T>
void PerAtomQuantity<T>::grow_lammps_array(int nmax, const std::string & tag)
{
if (nCols_ == 1)
this->lammpsScalar_ = lammpsInterface_->grow_array(this->lammpsScalar_,nmax,tag.c_str());
else
this->lammpsVector_ = lammpsInterface_->grow_array(this->lammpsVector_,nmax,nCols_,tag.c_str());
}
//-----------------------------------------------------------------
// copy values within local atom-based arrays
//-----------------------------------------------------------------
template <typename T>
void PerAtomQuantity<T>::copy_lammps_array(int i, int j)
{
if (nCols_ == 1) {
T * lammpsScalar = this->lammps_scalar();
lammpsScalar[j] = lammpsScalar[i];
}
else {
T ** lammpsVector = this->lammps_vector();
for (int k = 0; k < nCols_; k++)
lammpsVector[j][k] = lammpsVector[i][k];
}
}
//--------------------------------------------------------
//--------------------------------------------------------
// Class LammpsAtomQuantity
//--------------------------------------------------------
//--------------------------------------------------------
//--------------------------------------------------------
// reset
// reset is only true if set_quantity was used,
// so this syncs the quantity back with lammps
//--------------------------------------------------------
template <typename T>
void LammpsAtomQuantity<T>::reset() const
{
if (this->need_reset()) {
PerAtomQuantity<T>::reset();
this->set_quantity_to_lammps();
}
}
//-----------------------------------------------------------------
// pack values in local atom-based arrays for passing to ghosts on another proc
//-----------------------------------------------------------------
template <typename T>
int LammpsAtomQuantity<T>::pack_comm(int index, double *buf,
int pbc_flag, int *pbc)
{
if (this->need_reset()) this->reset();
int bufIdx = 0;
if (this->nCols_ == 1) {
T * lammpsQuantity = this->lammps_scalar();
buf[bufIdx++] = double(lammpsQuantity[index]);
}
else {
T ** lammpsQuantity = this->lammps_vector();
for (int k = 0; k < this->nCols_; k++)
buf[bufIdx++] = double(lammpsQuantity[index][k]);
}
return bufIdx;
}
//-----------------------------------------------------------------
// unpack values in local atom-based arrays for passing to ghosts on another proc
//-----------------------------------------------------------------
template <typename T>
int LammpsAtomQuantity<T>::unpack_comm(int index, double *buf)
{
DenseMatrix<T> & myQuantity(this->quantity_);
int bufIdx = 0;
if (this->nCols_ == 1) {
T * lammpsQuantity = this->lammps_scalar();
myQuantity(index,0) = T(buf[bufIdx++]);
lammpsQuantity[index] = myQuantity(index,0);
}
else {
T ** lammpsQuantity = this->lammps_vector();
for (int k = 0; k < this->nCols_; k++) {
myQuantity(index,k) = T(buf[bufIdx++]);
lammpsQuantity[index][k] = myQuantity(index,k);
}
}
this->propagate_reset();
return bufIdx;
}
//--------------------------------------------------------
//--------------------------------------------------------
// Class ProtectedMappedAtomQuantity
//--------------------------------------------------------
//--------------------------------------------------------
//--------------------------------------------------------
// set_lammps_to_quantity
//--------------------------------------------------------
template <typename T>
void ProtectedMappedAtomQuantity<T>::set_lammps_to_quantity() const
{
this->reset();
const DenseMatrix<T> & myQuantity(this->quantity_); // necessary to access quantity_ this way because of templating
int nCols = myQuantity.nCols();
const INT_ARRAY & atomMap(atomMap_->quantity());
if (myQuantity.nRows()>0) {
// full matrix copy
if (PerAtomQuantity<T>::atomType_ == ALL || PerAtomQuantity<T>::atomType_ == PROC_GHOST) {
if (nCols==1) { // scalar
T * lammpsQuantity = ProtectedAtomQuantity<T>::lammps_scalar();
for (int i = 0; i < PerAtomQuantity<T>::atc_.nlocal_total(); i++) {
int idx = atomMap(i,0);
if (idx > -1) {
lammpsQuantity[i] = myQuantity(idx,0);
}
}
}
else{ // vector
T ** lammpsQuantity = ProtectedAtomQuantity<T>::lammps_vector();
for (int i = 0; i < PerAtomQuantity<T>::atc_.nlocal_total(); i++) {
int idx = atomMap(i,0);
if (idx > -1) {
for (int j = 0; j < nCols; j++) {
lammpsQuantity[i][j] = myQuantity(idx,j);
}
}
}
}
}
// map quantities
else {
int atomIndex;
const Array<int> & quantityToLammps = (PerAtomQuantity<T>::atc_).atc_to_lammps_map();
if (nCols==1) { // scalar
T * lammpsQuantity = ProtectedAtomQuantity<T>::lammps_scalar();
for (int i = 0; i < atomMap.nRows(); i++) {
int idx = atomMap(i,0);
if (idx > -1) {
atomIndex = quantityToLammps(i);
lammpsQuantity[atomIndex] = myQuantity(idx,0);
}
}
}
else{ // vector
T ** lammpsQuantity = ProtectedAtomQuantity<T>::lammps_vector();
for (int i = 0; i < atomMap.nRows(); i++) {
int idx = atomMap(i,0);
if (idx > -1) {
atomIndex = quantityToLammps(i);
for (int j = 0; j < nCols; j++) {
lammpsQuantity[atomIndex][j] = myQuantity(idx,j);
}
}
}
}
}
}
}
//--------------------------------------------------------
// set_quantity_to_lammps
//--------------------------------------------------------
template <typename T>
void ProtectedMappedAtomQuantity<T>::set_quantity_to_lammps() const
{
DenseMatrix<T> & myQuantity(this->quantity_);
int nCols = myQuantity.nCols();
const INT_ARRAY & atomMap(atomMap_->quantity());
if (myQuantity.nRows()>0) {
// full matrix copy
// in the case where processor ghosts are in the quantity, don't set them back
if (PerAtomQuantity<T>::atomType_ == ALL || PerAtomQuantity<T>::atomType_ == PROC_GHOST) {
if (nCols==1) { // scalar
const T * lammpsQuantity = ProtectedAtomQuantity<T>::lammps_scalar();
for (int i = 0; i < PerAtomQuantity<T>::atc_.nlocal_total(); i++) {
int idx = atomMap(i,0);
if (idx > -1) {
myQuantity(idx,0) = lammpsQuantity[i];
}
}
}
else {
const T * const * lammpsQuantity = ProtectedAtomQuantity<T>::lammps_vector();
for (int i = 0; i < PerAtomQuantity<T>::atc_.nlocal_total(); i++) {
int idx = atomMap(i,0);
if (idx > -1) {
for (int j = 0; j < nCols; j++)
myQuantity(idx,j) = lammpsQuantity[i][j];
}
}
}
}
// map quantities
else {
int atomIndex;
const Array<int> & quantityToLammps = (PerAtomQuantity<T>::atc_).atc_to_lammps_map();
if (nCols==1) { // scalar
const T * lammpsQuantity = ProtectedAtomQuantity<T>::lammps_scalar();
for (int i = 0; i < atomMap.nRows(); i++) {
int idx = atomMap(i,0);
if (idx > -1) {
atomIndex = quantityToLammps(i);
myQuantity(idx,0) = lammpsQuantity[atomIndex];
}
}
}
else {
const T * const * lammpsQuantity = ProtectedAtomQuantity<T>::lammps_vector();
for (int i = 0; i < atomMap.nRows(); i++) {
int idx = atomMap(i,0);
if (idx > -1) {
atomIndex = quantityToLammps(i);
for (int j = 0; j < nCols; j++) {
myQuantity(idx,j) = lammpsQuantity[atomIndex][j];
}
}
}
}
}
}
}
//--------------------------------------------------------
//--------------------------------------------------------
// Class PerAtomDiagonalMatrix
//--------------------------------------------------------
//--------------------------------------------------------
//--------------------------------------------------------
// Constructor
//--------------------------------------------------------
template <typename T>
PerAtomDiagonalMatrix<T>::PerAtomDiagonalMatrix(ATC_Method * atc,
AtomType atomType) :
MatrixDependencyManager<DiagonalMatrix, T>(),
atc_(atc,atomType),
lammpsInterface_(LammpsInterface::instance()),
atomType_(atomType),
quantityToLammps_(atc_.atc_to_lammps_map()),
lammpsScalar_(NULL)
{
// do nothing
}
//--------------------------------------------------------
// Destructor
//--------------------------------------------------------
template <typename T>
PerAtomDiagonalMatrix<T>::~PerAtomDiagonalMatrix()
{
if (lammpsScalar_) lammpsInterface_->destroy_array(lammpsScalar_);
}
//--------------------------------------------------------
// set_lammps_to_quantity
//--------------------------------------------------------
template <typename T>
void PerAtomDiagonalMatrix<T>::set_lammps_to_quantity() const
{
const DiagonalMatrix<T> & myQuantity(this->quantity_); // necessary to access quantity_ this way because of templating
if (myQuantity.size()>0) {
// full matrix copy
if (atomType_ == ALL || atomType_ == PROC_GHOST) {
T * lammpsQuantity = this->lammps_scalar();
for (int i = 0; i < atc_.nlocal_total(); i++) {
lammpsQuantity[i] = myQuantity(i,i);
}
}
// map quantities
else {
int atomIndex;
const Array<int> & quantityToLammps = atc_.atc_to_lammps_map();
T * lammpsQuantity = this->lammps_scalar();
for (int i = 0; i < myQuantity.nRows(); i++) {
atomIndex = quantityToLammps(i);
lammpsQuantity[atomIndex] = myQuantity(i,i);
}
}
}
}
//--------------------------------------------------------
// set_quantity_to_lammps
//--------------------------------------------------------
template <typename T>
void PerAtomDiagonalMatrix<T>::set_quantity_to_lammps() const
{
DiagonalMatrix<T> & myQuantity(this->quantity_);
if (myQuantity.size()>0) {
// full matrix copy
// in the case where processor ghosts are in the quantity, don't set them back
if (atomType_ == ALL || atomType_ == PROC_GHOST) {
const T * lammpsQuantity = this->lammps_scalar();
for (int i = 0; i < myQuantity.nRows(); i++) {
myQuantity(i,i) = lammpsQuantity[i];
}
}
// map quantities
else {
int atomIndex;
const Array<int> & quantityToLammps = atc_.atc_to_lammps_map();
const T * lammpsQuantity = this->lammps_scalar();
for (int i = 0; i < myQuantity.nRows(); i++) {
atomIndex = quantityToLammps(i);
myQuantity(i,i) = lammpsQuantity[atomIndex];
}
}
}
}
//-----------------------------------------------------------------
// pack values in local atom-based arrays for exchange with another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomDiagonalMatrix<T>::pack_exchange(int i, double *buffer)
{
buffer[0] = static_cast<double>(lammps_scalar()[i]);
return 1;
}
//-----------------------------------------------------------------
// unpack values in local atom-based arrays from exchange with another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomDiagonalMatrix<T>::unpack_exchange(int i, double *buffer)
{
lammps_scalar()[i] = static_cast<T>(buffer[0]);
return 1;
}
//-----------------------------------------------------------------
// pack values in local atom-based arrays for passing to ghosts on another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomDiagonalMatrix<T>::pack_comm(int index, double *buf,
int pbc_flag, int *pbc)
{
if (this->need_reset()) this->reset();
DiagonalMatrix<T> & myQuantity(this->quantity_);
buf[0] = static_cast<double>(myQuantity(index,index));
return 1;
}
//-----------------------------------------------------------------
// unpack values in local atom-based arrays for passing to ghosts on another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomDiagonalMatrix<T>::unpack_comm(int index, double *buf)
{
DiagonalMatrix<T> & myQuantity(this->quantity_);
myQuantity(index,index) = static_cast<T>(buf[0]);
this->propagate_reset();
return 1;
}
//-----------------------------------------------------------------
// allocate local atom-based arrays
//-----------------------------------------------------------------
template <typename T>
void PerAtomDiagonalMatrix<T>::grow_lammps_array(int nmax, const std::string & tag)
{
this->lammpsScalar_ = lammpsInterface_->grow_array(this->lammpsScalar_,nmax,tag.c_str());
}
//-----------------------------------------------------------------
// copy values within local atom-based arrays
//-----------------------------------------------------------------
template <typename T>
void PerAtomDiagonalMatrix<T>::copy_lammps_array(int i, int j)
{
T * lammpsScalar = this->lammps_scalar();
lammpsScalar[j] = lammpsScalar[i];
}
//--------------------------------------------------------
//--------------------------------------------------------
// Class PerAtomSparseMatrix
//--------------------------------------------------------
//--------------------------------------------------------
//--------------------------------------------------------
// Constructor
//--------------------------------------------------------
template <typename T>
PerAtomSparseMatrix<T>::PerAtomSparseMatrix(ATC_Method * atc,
int nCols,
int maxEntriesPerRow,
AtomType atomType) :
MatrixDependencyManager<SparseMatrix, T>(),
atc_(atc,atomType),
lammpsInterface_(LammpsInterface::instance()),
atomType_(atomType),
nCols_(nCols),
maxEntriesPerRow_(maxEntriesPerRow),
quantityToLammps_(atc_.atc_to_lammps_map()),
lammpsVector_(NULL),
lammpsColIndices_(NULL)
{
// do nothing
}
//--------------------------------------------------------
// Destructor
//--------------------------------------------------------
template <typename T>
PerAtomSparseMatrix<T>::~PerAtomSparseMatrix()
{
if (lammpsVector_) lammpsInterface_->destroy_array(lammpsVector_);
if (lammpsColIndices_) lammpsInterface_->destroy_array(lammpsColIndices_);
}
//--------------------------------------------------------
// set_lammps_to_quantity
//--------------------------------------------------------
template <typename T>
void PerAtomSparseMatrix<T>::set_lammps_to_quantity() const
{
const SparseMatrix<T> & myQuantity(this->quantity_); // necessary to access quantity_ this way because of templating
if (myQuantity.nRows()>0) {
// full matrix copy
if (atomType_ == ALL || atomType_ == PROC_GHOST) {
T ** lammpsQuantity = this->lammps_vector();
int ** lammpsColIndices = this->lammps_column_indices();
for (int i = 0; i < atc_.nlocal_total(); i++) {
myQuantity.row(i,_values_,_colIndices_);
for (int j = 0; j < _values_.size(); j++) {
lammpsQuantity[i][j] = _values_(j);
lammpsColIndices[i][j] = _colIndices_(j);
}
for (int j = _values_.size(); j < maxEntriesPerRow_; j++) {
lammpsColIndices[i][j] = -1;
}
}
}
// map quantities
else {
int atomIndex;
const Array<int> & quantityToLammps = atc_.atc_to_lammps_map();
T ** lammpsQuantity = this->lammps_vector();
int ** lammpsColIndices = this->lammps_column_indices();
for (int i = 0; i < myQuantity.nRows(); i++) {
myQuantity.row(i,_values_,_colIndices_);
atomIndex = quantityToLammps(i);
for (int j = 0; j < _values_.size(); j++) {
lammpsQuantity[atomIndex][j] = _values_(j);
lammpsColIndices[atomIndex][j] = _colIndices_(j);
}
for (int j = _values_.size(); j < maxEntriesPerRow_; j++) {
lammpsQuantity[atomIndex][j] = 0;
lammpsColIndices[atomIndex][j] = -1;
}
}
}
}
}
//--------------------------------------------------------
// set_quantity_to_lammps
//--------------------------------------------------------
template <typename T>
void PerAtomSparseMatrix<T>::set_quantity_to_lammps() const
{
SparseMatrix<T> & myQuantity(this->quantity_);
if (myQuantity.nRows()>0) {
// full matrix copy
// in the case where processor ghosts are in the quantity, don't set them back
if (atomType_ == ALL || atomType_ == PROC_GHOST) {
const T * const * lammpsQuantity = this->lammps_vector();
const int * const * lammpsColIndices = this->lammps_column_indices();
for (int i = 0; i < myQuantity.nRows(); i++) {
for (int j = 0; j < maxEntriesPerRow_; j++) {
if (lammpsColIndices[i][j] < 0) {
break;
}
myQuantity.set(i,lammpsColIndices[i][j],lammpsQuantity[i][j]);
}
}
}
// map quantities
else {
int atomIndex;
const Array<int> & quantityToLammps = atc_.atc_to_lammps_map();
const T * const * lammpsQuantity = this->lammps_vector();
const int * const * lammpsColIndices = this->lammps_column_indices();
for (int i = 0; i < myQuantity.nRows(); i++) {
atomIndex = quantityToLammps(i);
for (int j = 0; j < maxEntriesPerRow_; j++) {
if (lammpsColIndices[atomIndex][j] < 0) {
break;
}
myQuantity.set(i,lammpsColIndices[atomIndex][j],lammpsQuantity[atomIndex][j]);
}
}
}
myQuantity.compress();
}
}
//-----------------------------------------------------------------
// pack values in local atom-based arrays for exchange with another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomSparseMatrix<T>::pack_exchange(int i, double *buffer)
{
int idx = 0;
T ** lammpsVector = this->lammps_vector();
for (int j = 0; j < maxEntriesPerRow_; j++) {
buffer[idx++] = static_cast<double>(lammpsVector[i][j]);
}
int ** lammpsColIndices = this->lammps_column_indices();
for (int j = 0; j < maxEntriesPerRow_; j++) {
buffer[idx++] = static_cast<double>(lammpsColIndices[i][j]);
}
return 2*maxEntriesPerRow_;
}
//-----------------------------------------------------------------
// unpack values in local atom-based arrays from exchange with another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomSparseMatrix<T>::unpack_exchange(int i, double *buffer)
{
int idx = 0;
T ** lammpsVector = this->lammps_vector();
for (int j = 0; j < maxEntriesPerRow_; j++) {
lammpsVector[i][j] = static_cast<T>(buffer[idx++]);
}
int ** lammpsColIndices = this->lammps_column_indices();
for (int j = 0; j < maxEntriesPerRow_; j++) {
lammpsColIndices[i][j] = static_cast<int>(buffer[idx++]);
}
return 2*maxEntriesPerRow_;
}
//-----------------------------------------------------------------
// pack values in local atom-based arrays for passing to ghosts on another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomSparseMatrix<T>::pack_comm(int index, double *buf,
int pbc_flag, int *pbc)
{
return 0;
}
//-----------------------------------------------------------------
// unpack values in local atom-based arrays for passing to ghosts on another proc
//-----------------------------------------------------------------
template <typename T>
int PerAtomSparseMatrix<T>::unpack_comm(int index, double *buf)
{
return 0;
}
//-----------------------------------------------------------------
// allocate local atom-based arrays
//-----------------------------------------------------------------
template <typename T>
void PerAtomSparseMatrix<T>::grow_lammps_array(int nmax, const std::string & tag)
{
this->lammpsVector_ = lammpsInterface_->grow_array(this->lammpsVector_,nmax,maxEntriesPerRow_,tag.c_str());
std::string myString(tag+std::string("Columns"));
this->lammpsColIndices_ = lammpsInterface_->grow_array(this->lammpsColIndices_,nmax,maxEntriesPerRow_,myString.c_str());
}
//-----------------------------------------------------------------
// copy values within local atom-based arrays
//-----------------------------------------------------------------
template <typename T>
void PerAtomSparseMatrix<T>::copy_lammps_array(int i, int j)
{
T ** lammpsVector = this->lammps_vector();
int ** lammpsColIndices = this->lammps_column_indices();
for (int k = 0; k < maxEntriesPerRow_; k++) {
lammpsVector[j][k] = lammpsVector[i][k];
lammpsColIndices[j][k] = lammpsColIndices[i][k];
}
}
}
#endif

Event Timeline