Homec4science
Diffusion lammps (lammm-master) Browse

lammps/lib/gpu/lj96_cut_gpu_memory.cpp
84b5b0bef886lammm-master

Last Change
Raw File

lj96_cut_gpu_memory.cpp
View Options

/* ----------------------------------------------------------------------
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
http://lammps.sandia.gov, Sandia National Laboratories
Steve Plimpton, sjplimp@sandia.gov
Copyright (2003) Sandia Corporation. Under the terms of Contract
DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
certain rights in this software. This software is distributed under
the GNU General Public License.
See the README file in the top-level LAMMPS directory.
------------------------------------------------------------------------- */
/* ----------------------------------------------------------------------
Contributing authors: Mike Brown (ORNL), brownw@ornl.gov
------------------------------------------------------------------------- */
#ifdef USE_OPENCL
#include "lj96_cut_gpu_cl.h"
#else
#include "lj96_cut_gpu_ptx.h"
#endif
#include "lj96_cut_gpu_memory.h"
#include <cassert>
#define LJ96_GPU_MemoryT LJ96_GPU_Memory<numtyp, acctyp>
extern PairGPUDevice<PRECISION,ACC_PRECISION> pair_gpu_device;
template <class numtyp, class acctyp>
LJ96_GPU_MemoryT::LJ96_GPU_Memory() : AtomicGPUMemory<numtyp,acctyp>(), _allocated(false) {
}
template <class numtyp, class acctyp>
LJ96_GPU_MemoryT::~LJ96_GPU_Memory() {
clear();
}
template <class numtyp, class acctyp>
int LJ96_GPU_MemoryT::bytes_per_atom(const int max_nbors) const {
return this->bytes_per_atom_atomic(max_nbors);
}
template <class numtyp, class acctyp>
int LJ96_GPU_MemoryT::init(const int ntypes,
double **host_cutsq, 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) {
int success;
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,
_screen,lj96_cut_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;
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;
// Allocate a host write buffer for data initialization
UCL_H_Vec<numtyp> host_write(lj_types*lj_types*32,*(this->ucl_device),
UCL_WRITE_OPTIMIZED);
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);
this->atom->type_pack4(ntypes,lj_types,lj1,host_write,host_lj1,host_lj2,
host_cutsq);
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);
UCL_H_Vec<double> dview;
sp_lj.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
dview.view(host_special_lj,4,*(this->ucl_device));
ucl_copy(sp_lj,dview,false);
_allocated=true;
this->_max_bytes=lj1.row_bytes()+lj3.row_bytes()+sp_lj.row_bytes();
return 0;
}
template <class numtyp, class acctyp>
void LJ96_GPU_MemoryT::clear() {
if (!_allocated)
return;
_allocated=false;
lj1.clear();
lj3.clear();
sp_lj.clear();
this->clear_atomic();
}
template <class numtyp, class acctyp>
double LJ96_GPU_MemoryT::host_memory_usage() const {
return this->host_memory_usage_atomic()+sizeof(LJ96_GPU_Memory<numtyp,acctyp>);
}
// ---------------------------------------------------------------------------
// Calculate energies, forces, and torques
// ---------------------------------------------------------------------------
template <class numtyp, class acctyp>
void LJ96_GPU_MemoryT::loop(const bool _eflag, const bool _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)));
int ainum=this->ans->inum();
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->dev_x.begin(), &lj1.begin(),
&lj3.begin(), &sp_lj.begin(),
&this->nbor->dev_nbor.begin(),
&this->_nbor_data->begin(),
&this->ans->dev_ans.begin(),
&this->ans->dev_engv.begin(), &eflag, &vflag,
&ainum, &nbor_pitch, &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->_nbor_data->begin(), &this->ans->dev_ans.begin(),
&this->ans->dev_engv.begin(), &eflag, &vflag, &ainum,
&nbor_pitch, &this->_threads_per_atom);
}
this->time_pair.stop();
}
template class LJ96_GPU_Memory<PRECISION,ACC_PRECISION>;
c4science · Help