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@6053 f3b2605a-c512-4ea7-a41b-209d697bcdaa

Browse Source
This commit is contained in:
sjplimp
2011-05-02 15:02:52 +00:00
parent 2be078632d
commit 5f799182b3
70 changed files with 4489 additions and 2253 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
lib/gpu/cmmc_long_gpu_memory.cpp
View File

@ -43,26 +43,30 @@ int CMML_GPU_MemoryT::bytes_per_atom(const int max_nbors) const {
}
template <class numtyp, class acctyp>
bool CMML_GPU_MemoryT::init(const int ntypes, double **host_cutsq,
int **host_cg_type, double **host_lj1,
double **host_lj2, double **host_lj3,
double **host_lj4, double **host_offset,
double *host_special_lj, const int nlocal,
const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
const double gpu_split, FILE *_screen,
double **host_cut_ljsq,
const double host_cut_coulsq,
double *host_special_coul, const double qqrd2e,
const double g_ewald) {
this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,
_screen,cmmc_long_gpu_kernel);
int CMML_GPU_MemoryT::init(const int ntypes, double **host_cutsq,
int **host_cg_type, double **host_lj1,
double **host_lj2, double **host_lj3,
double **host_lj4, double **host_offset,
double *host_special_lj, const int nlocal,
const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
const double gpu_split, FILE *_screen,
double **host_cut_ljsq,
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,cmmc_long_gpu_kernel);
if (success!=0)
return success;
// If atom type constants fit in shared memory use fast kernel
int lj_types=ntypes;
shared_types=false;
if (lj_types<=MAX_SHARED_TYPES && this->_block_size>=MAX_SHARED_TYPES) {
lj_types=MAX_SHARED_TYPES;
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;
}
_lj_types=lj_types;
@ -95,7 +99,7 @@ bool CMML_GPU_MemoryT::init(const int ntypes, double **host_cutsq,
_allocated=true;
this->_max_bytes=lj1.row_bytes()+lj3.row_bytes()+sp_lj.row_bytes();
return true;
return 0;
}
template <class numtyp, class acctyp>
@ -133,9 +137,10 @@ void CMML_GPU_MemoryT::loop(const bool _eflag, const bool _vflag) {
else
vflag=0;
int GX=static_cast<int>(ceil(static_cast<double>(this->atom->inum())/BX));
int GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/
(BX/this->_threads_per_atom)));
int ainum=this->atom->inum();
int ainum=this->ans->inum();
int anall=this->atom->nall();
int nbor_pitch=this->nbor->nbor_pitch();
this->time_pair.start();
@ -144,19 +149,21 @@ void CMML_GPU_MemoryT::loop(const bool _eflag, const bool _vflag) {
this->k_pair_fast.run(&this->atom->dev_x.begin(), &lj1.begin(),
&lj3.begin(), &sp_lj.begin(),
&this->nbor->dev_nbor.begin(),
&this->atom->dev_ans.begin(),
&this->atom->dev_engv.begin(), &eflag, &vflag,
&this->_nbor_data->begin(),
&this->ans->dev_ans.begin(),
&this->ans->dev_engv.begin(), &eflag, &vflag,
&ainum, &anall, &nbor_pitch,
&this->atom->dev_q.begin(), &_cut_coulsq,
&_qqrd2e, &_g_ewald);
&_qqrd2e, &_g_ewald, &this->_threads_per_atom);
} else {
this->k_pair.set_size(GX,BX);
this->k_pair.run(&this->atom->dev_x.begin(), &lj1.begin(), &lj3.begin(),
&_lj_types, &sp_lj.begin(), &this->nbor->dev_nbor.begin(),
&this->atom->dev_ans.begin(),
&this->atom->dev_engv.begin(), &eflag, &vflag, &ainum,
&this->_nbor_data->begin(), &this->ans->dev_ans.begin(),
&this->ans->dev_engv.begin(), &eflag, &vflag, &ainum,
&anall, &nbor_pitch, &this->atom->dev_q.begin(),
&_cut_coulsq, &_qqrd2e, &_g_ewald);
&_cut_coulsq, &_qqrd2e, &_g_ewald,
&this->_threads_per_atom);
}
this->time_pair.stop();
}