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

Feb2021 GPU Package Update - GPU Package Files

Browse Source
This commit is contained in:
Michael Brown
2021-02-15 08:20:50 -08:00
parent 16004e8f45
commit e7e2d2323b
345 changed files with 13424 additions and 7708 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
lib/gpu/lal_lj.cpp
View File

@ -51,16 +51,31 @@ int LJT::init(const int ntypes,
const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
const double gpu_split, FILE *_screen) {
const int max_shared_types=this->device->max_shared_types();
int onetype=0;
#ifdef USE_OPENCL
if (maxspecial==0)
for (int i=1; i<ntypes; i++)
for (int j=i; j<ntypes; j++)
if (host_cutsq[i][j]>0) {
if (onetype>0)
onetype=-1;
else if (onetype==0)
onetype=i*max_shared_types+j;
}
if (onetype<0) onetype=0;
#endif
int success;
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,
_screen,lj,"k_lj");
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,
gpu_split,_screen,lj,"k_lj",onetype);
if (success!=0)
return success;
// If atom type constants fit in shared memory use fast kernel
int lj_types=ntypes;
shared_types=false;
int max_shared_types=this->device->max_shared_types();
if (lj_types<=max_shared_types && this->_block_size>=max_shared_types) {
lj_types=max_shared_types;
shared_types=true;
@ -130,20 +145,9 @@ double LJT::host_memory_usage() const {
// Calculate energies, forces, and torques
// ---------------------------------------------------------------------------
template <class numtyp, class acctyp>
void LJT::loop(const bool _eflag, const bool _vflag) {
int LJT::loop(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 eflag, vflag;
if (_eflag)
eflag=1;
else
eflag=0;
if (_vflag)
vflag=1;
else
vflag=0;
int GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/
(BX/this->_threads_per_atom)));
@ -151,8 +155,8 @@ void LJT::loop(const bool _eflag, const bool _vflag) {
int nbor_pitch=this->nbor->nbor_pitch();
this->time_pair.start();
if (shared_types) {
this->k_pair_fast.set_size(GX,BX);
this->k_pair_fast.run(&this->atom->x, &lj1, &lj3, &sp_lj,
this->k_pair_sel->set_size(GX,BX);
this->k_pair_sel->run(&this->atom->x, &lj1, &lj3, &sp_lj,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->ans->force, &this->ans->engv, &eflag,
&vflag, &ainum, &nbor_pitch,
@ -165,6 +169,7 @@ void LJT::loop(const bool _eflag, const bool _vflag) {
&ainum, &nbor_pitch, &this->_threads_per_atom);
}
this->time_pair.stop();
return GX;
}
template class LJ<PRECISION,ACC_PRECISION>;