ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Cleaned up and re-arranged the functions to reflect the order of calling in a time step

Browse Source
This commit is contained in:
Trung Nguyen
2021-09-11 01:00:58 -05:00
parent 7f5a82dc54
commit c765861851
3 changed files with 493 additions and 530 deletions
Download Patch File Download Diff File Expand all files Collapse all files

105
lib/gpu/lal_amoeba.cpp
View File

@ -126,49 +126,6 @@ double AmoebaT::host_memory_usage() const {
return this->host_memory_usage_atomic()+sizeof(Amoeba<numtyp,acctyp>);
}
// ---------------------------------------------------------------------------
// Calculate the polar real-space term, returning tep
// ---------------------------------------------------------------------------
template <class numtyp, class acctyp>
int AmoebaT::polar_real(const int eflag, const int vflag) {
// Compute the block size and grid size to keep all cores busy
const int BX=this->block_size();
int GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/
(BX/this->_threads_per_atom)));
int _nall=this->atom->nall();
int ainum=this->ans->inum();
int nbor_pitch=this->nbor->nbor_pitch();
this->time_pair.start();
// Build the short neighbor list if needed
if (!this->short_nbor_avail) {
this->k_short_nbor.set_size(GX,BX);
this->k_short_nbor.run(&this->atom->x, &this->nbor->dev_nbor,
&this->_nbor_data->begin(),
&this->dev_short_nbor, &_off2, &ainum,
&nbor_pitch, &this->_threads_per_atom);
this->short_nbor_avail = true;
}
this->k_polar.set_size(GX,BX);
this->k_polar.run(&this->atom->x, &this->atom->extra, &damping, &sp_polar,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->dev_short_nbor,
&this->ans->force, &this->ans->engv, &this->_tep,
&eflag, &vflag, &ainum, &_nall, &nbor_pitch,
&this->_threads_per_atom,
&_aewald, &_felec, &_off2, &_polar_dscale, &_polar_uscale);
this->time_pair.stop();
// Signal that short nbor list is not avail for the next time step
// do it here because polar_real() is the last kernel in a time step at this point
this->short_nbor_avail = false;
return GX;
}
// ---------------------------------------------------------------------------
// Calculate the real-space permanent field, returning field and fieldp
// ---------------------------------------------------------------------------
@ -182,13 +139,13 @@ int AmoebaT::udirect2b(const int eflag, const int vflag) {
const int BX=this->block_size();
int GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/(BX/this->_threads_per_atom)));
// Build the short neighbor list if needed
// Build the short neighbor list if not done yet
if (!this->short_nbor_avail) {
this->k_short_nbor.set_size(GX,BX);
this->k_short_nbor.run(&this->atom->x, &this->nbor->dev_nbor,
&this->_nbor_data->begin(),
&this->dev_short_nbor, &_off2, &ainum,
&nbor_pitch, &this->_threads_per_atom);
&this->_nbor_data->begin(),
&this->dev_short_nbor, &_off2, &ainum,
&nbor_pitch, &this->_threads_per_atom);
this->short_nbor_avail = true;
}
@ -219,9 +176,20 @@ int AmoebaT::umutual2b(const int eflag, const int vflag) {
int nbor_pitch=this->nbor->nbor_pitch();
this->time_pair.start();
// Build the short neighbor list if not done yet
if (!this->short_nbor_avail) {
this->k_short_nbor.set_size(GX,BX);
this->k_short_nbor.run(&this->atom->x, &this->nbor->dev_nbor,
&this->_nbor_data->begin(),
&this->dev_short_nbor, &_off2, &ainum,
&nbor_pitch, &this->_threads_per_atom);
this->short_nbor_avail = true;
}
this->k_umutual2b.set_size(GX,BX);
this->k_umutual2b.run(&this->atom->x, &this->atom->extra, &damping, &sp_polar,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->dev_short_nbor,
&this->_fieldp, &ainum, &_nall, &nbor_pitch,
&this->_threads_per_atom, &_aewald, &_off2,
&_polar_dscale, &_polar_uscale);
@ -230,5 +198,48 @@ int AmoebaT::umutual2b(const int eflag, const int vflag) {
return GX;
}
// ---------------------------------------------------------------------------
// Calculate the polar real-space term, returning tep
// ---------------------------------------------------------------------------
template <class numtyp, class acctyp>
int AmoebaT::polar_real(const int eflag, const int vflag) {
// Compute the block size and grid size to keep all cores busy
const int BX=this->block_size();
int GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/
(BX/this->_threads_per_atom)));
int _nall=this->atom->nall();
int ainum=this->ans->inum();
int nbor_pitch=this->nbor->nbor_pitch();
this->time_pair.start();
// Build the short neighbor list if not done yet
if (!this->short_nbor_avail) {
this->k_short_nbor.set_size(GX,BX);
this->k_short_nbor.run(&this->atom->x, &this->nbor->dev_nbor,
&this->_nbor_data->begin(),
&this->dev_short_nbor, &_off2, &ainum,
&nbor_pitch, &this->_threads_per_atom);
this->short_nbor_avail = true;
}
this->k_polar.set_size(GX,BX);
this->k_polar.run(&this->atom->x, &this->atom->extra, &damping, &sp_polar,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->dev_short_nbor,
&this->ans->force, &this->ans->engv, &this->_tep,
&eflag, &vflag, &ainum, &_nall, &nbor_pitch,
&this->_threads_per_atom,
&_aewald, &_felec, &_off2, &_polar_dscale, &_polar_uscale);
this->time_pair.stop();
// Signal that short nbor list is not avail for the next time step
// do it here because polar_real() is the last kernel in a time step at this point
this->short_nbor_avail = false;
return GX;
}
template class Amoeba<PRECISION,ACC_PRECISION>;
}