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

git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@15248 f3b2605a-c512-4ea7-a41b-209d697bcdaa

Browse Source
This commit is contained in:
sjplimp
2016-07-01 23:27:26 +00:00
parent 8366b35459
commit 9656958169
245 changed files with 4890 additions and 4832 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
lib/gpu/lal_lj_expand.cpp
View File

@ -9,7 +9,7 @@
This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
__________________________________________________________________________
begin :
begin :
email : ibains@nvidia.com
***************************************************************************/
@ -36,7 +36,7 @@ template <class numtyp, class acctyp>
LJExpandT::~LJExpand() {
clear();
}
template <class numtyp, class acctyp>
int LJExpandT::bytes_per_atom(const int max_nbors) const {
return this->bytes_per_atom_atomic(max_nbors);
@ -76,11 +76,11 @@ int LJExpandT::init(const int ntypes, double **host_cutsq,
lj1.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack4(ntypes,lj_types,lj1,host_write,host_lj1,host_lj2,
host_cutsq, host_shift);
host_cutsq, host_shift);
lj3.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack4(ntypes,lj_types,lj3,host_write,host_lj3,host_lj4,
host_offset);
host_offset);
UCL_H_Vec<double> dview;
sp_lj.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
@ -97,17 +97,17 @@ void LJExpandT::reinit(const int ntypes, double **host_cutsq,
double **host_lj1, double **host_lj2,
double **host_lj3, double **host_lj4,
double **host_offset, double **host_shift) {
// Allocate a host write buffer for data initialization
UCL_H_Vec<numtyp> host_write(_lj_types*_lj_types*32,*(this->ucl_device),
UCL_WRITE_ONLY);
for (int i=0; i<_lj_types*_lj_types; i++)
host_write[i]=0.0;
this->atom->type_pack4(ntypes,_lj_types,lj1,host_write,host_lj1,host_lj2,
host_cutsq, host_shift);
this->atom->type_pack4(ntypes,_lj_types,lj3,host_write,host_lj3,host_lj4,
host_offset);
}
@ -146,7 +146,7 @@ void LJExpandT::loop(const bool _eflag, const bool _vflag) {
vflag=1;
else
vflag=0;
int GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/
(BX/this->_threads_per_atom)));
@ -155,15 +155,15 @@ void LJExpandT::loop(const bool _eflag, const bool _vflag) {
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_fast.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,
&this->ans->force, &this->ans->engv, &eflag,
&vflag, &ainum, &nbor_pitch,
&this->_threads_per_atom);
} else {
this->k_pair.set_size(GX,BX);
this->k_pair.run(&this->atom->x, &lj1, &lj3, &_lj_types, &sp_lj,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
this->k_pair.run(&this->atom->x, &lj1, &lj3, &_lj_types, &sp_lj,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->ans->force, &this->ans->engv, &eflag, &vflag,
&ainum, &nbor_pitch, &this->_threads_per_atom);
}