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@12146 f3b2605a-c512-4ea7-a41b-209d697bcdaa

Browse Source
This commit is contained in:
sjplimp
2014-06-27 17:25:22 +00:00
parent f745f24bcd
commit 67ae64329e
24 changed files with 422 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
lib/gpu/lal_coul_long.cpp
View File

@ -43,21 +43,19 @@ int CoulLongT::bytes_per_atom(const int max_nbors) const {
}
template <class numtyp, class acctyp>
int CoulLongT::init(const int nlocal, const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
const double gpu_split, FILE *_screen,
const double host_cut_coulsq, double *host_special_coul,
const double qqrd2e, const double g_ewald) {
int CoulLongT::init(const int ntypes, double **host_scale,
const int nlocal, const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
const double gpu_split, FILE *_screen,
const double host_cut_coulsq, double *host_special_coul,
const double qqrd2e, const double g_ewald) {
int success;
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,
gpu_split,_screen,coul_long,"k_coul_long");
if (success!=0)
return success;
// we don't have atom types for coulomb only,
// but go with the minimum so that we can use
// the same infrastructure as lj/cut/coul/long/gpu.
int lj_types=1;
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) {
@ -69,13 +67,16 @@ int CoulLongT::init(const int nlocal, const int nall, const int max_nbors,
// 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;
lj1.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
lj3.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
scale.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack1(ntypes,lj_types,scale,host_write,host_scale);
sp_cl.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
for (int i=0; i<4; i++) {
host_write[i]=host_special_coul[i];
@ -87,10 +88,18 @@ int CoulLongT::init(const int nlocal, const int nall, const int max_nbors,
_g_ewald=g_ewald;
_allocated=true;
this->_max_bytes=lj1.row_bytes()+lj3.row_bytes()+sp_cl.row_bytes();
this->_max_bytes=lj1.row_bytes()+lj3.row_bytes()+scale.row_bytes()+
sp_cl.row_bytes();
return 0;
}
template <class numtyp, class acctyp>
void CoulLongT::reinit(const int ntypes, double **host_scale) {
UCL_H_Vec<numtyp> hscale(_lj_types*_lj_types,*(this->ucl_device),
UCL_WRITE_ONLY);
this->atom->type_pack1(ntypes,_lj_types,scale,hscale,host_scale);
}
template <class numtyp, class acctyp>
void CoulLongT::clear() {
if (!_allocated)
@ -99,6 +108,7 @@ void CoulLongT::clear() {
lj1.clear();
lj3.clear();
scale.clear();
sp_cl.clear();
this->clear_atomic();
}
@ -134,7 +144,7 @@ void CoulLongT::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_cl,
this->k_pair_fast.run(&this->atom->x, &scale, &sp_cl,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->ans->force, &this->ans->engv,
&eflag, &vflag, &ainum, &nbor_pitch,
@ -142,7 +152,7 @@ void CoulLongT::loop(const bool _eflag, const bool _vflag) {
&this->_threads_per_atom);
} else {
this->k_pair.set_size(GX,BX);
this->k_pair.run(&this->atom->x, &lj1, &lj3, &_lj_types, &sp_cl,
this->k_pair.run(&this->atom->x, &scale, &_lj_types, &sp_cl,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->ans->force, &this->ans->engv, &eflag, &vflag,
&ainum, &nbor_pitch, &this->atom->q, &_cut_coulsq,