Homec4science
Diffusion lammps (efficient_neuronet) Browse

lammps/src/KOKKOS/pair_coul_cut_kokkos.cpp
ba6abbd9d3fcefficient_neuronet

Last Change
Raw File

pair_coul_cut_kokkos.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.
------------------------------------------------------------------------- */
#include <math.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "pair_coul_cut_kokkos.h"
#include "kokkos.h"
#include "atom_kokkos.h"
#include "comm.h"
#include "force.h"
#include "neighbor.h"
#include "neigh_list.h"
#include "neigh_request.h"
#include "update.h"
#include "integrate.h"
#include "respa.h"
#include "math_const.h"
#include "memory.h"
#include "error.h"
#include "atom_masks.h"
using namespace LAMMPS_NS;
using namespace MathConst;
#define KOKKOS_CUDA_MAX_THREADS 256
#define KOKKOS_CUDA_MIN_BLOCKS 8
/* ---------------------------------------------------------------------- */
template<class DeviceType>
PairCoulCutKokkos<DeviceType>::PairCoulCutKokkos(LAMMPS *lmp) : PairCoulCut(lmp)
{
atomKK = (AtomKokkos *) atom;
execution_space = ExecutionSpaceFromDevice<DeviceType>::space;
datamask_read = X_MASK | F_MASK | TYPE_MASK | Q_MASK | ENERGY_MASK | VIRIAL_MASK;
datamask_modify = F_MASK | ENERGY_MASK | VIRIAL_MASK;
cutsq = NULL;
}
/* ---------------------------------------------------------------------- */
template<class DeviceType>
PairCoulCutKokkos<DeviceType>::~PairCoulCutKokkos()
{
if (allocated)
memory->destroy_kokkos(k_cutsq, cutsq);
}
/* ---------------------------------------------------------------------- */
template<class DeviceType>
void PairCoulCutKokkos<DeviceType>::cleanup_copy() {
// WHY needed: this prevents parent copy from deallocating any arrays
allocated = 0;
cutsq = NULL;
eatom = NULL;
vatom = NULL;
}
/* ---------------------------------------------------------------------- */
template<class DeviceType>
void PairCoulCutKokkos<DeviceType>::compute(int eflag_in, int vflag_in)
{
eflag = eflag_in;
vflag = vflag_in;
if (neighflag == FULL || neighflag == FULLCLUSTER) no_virial_fdotr_compute = 1;
if (eflag || vflag) ev_setup(eflag,vflag);
else evflag = vflag_fdotr = 0;
atomKK->sync(execution_space,datamask_read);
k_cutsq.template sync<DeviceType>();
k_cut_ljsq.template sync<DeviceType>();
k_cut_coulsq.template sync<DeviceType>();
k_params.template sync<DeviceType>();
if (eflag || vflag) atomKK->modified(execution_space,datamask_modify);
else atomKK->modified(execution_space,F_MASK);
x = atomKK->k_x.view<DeviceType>();
c_x = atomKK->k_x.view<DeviceType>();
f = atomKK->k_f.view<DeviceType>();
q = atomKK->k_q.view<DeviceType>();
type = atomKK->k_type.view<DeviceType>();
nlocal = atom->nlocal;
nall = atom->nlocal + atom->nghost;
newton_pair = force->newton_pair;
special_lj[0] = force->special_lj[0];
special_lj[1] = force->special_lj[1];
special_lj[2] = force->special_lj[2];
special_lj[3] = force->special_lj[3];
special_coul[0] = force->special_coul[0];
special_coul[1] = force->special_coul[1];
special_coul[2] = force->special_coul[2];
special_coul[3] = force->special_coul[3];
qqrd2e = force->qqrd2e;
// loop over neighbors of my atoms
EV_FLOAT ev = pair_compute<PairCoulCutKokkos<DeviceType>,void >
(this,(NeighListKokkos<DeviceType>*)list);
DeviceType::fence();
if (eflag) eng_coul += ev.ecoul;
if (vflag_global) {
virial[0] += ev.v[0];
virial[1] += ev.v[1];
virial[2] += ev.v[2];
virial[3] += ev.v[3];
virial[4] += ev.v[4];
virial[5] += ev.v[5];
}
if (vflag_fdotr) pair_virial_fdotr_compute(this);
}
template<class DeviceType>
template<bool STACKPARAMS, class Specialisation>
KOKKOS_INLINE_FUNCTION
F_FLOAT PairCoulCutKokkos<DeviceType>::
compute_fcoul(const F_FLOAT& rsq, const int& i, const int&j, const int& itype,
const int& jtype, const F_FLOAT& factor_coul, const F_FLOAT& qtmp) const {
const F_FLOAT r2inv = 1.0/rsq;
const F_FLOAT rinv = sqrt(r2inv);
F_FLOAT forcecoul;
forcecoul = qqrd2e*(STACKPARAMS?m_params[itype][jtype].scale:params(itype,jtype).scale)*
qtmp *q(j) *rinv;
return factor_coul*forcecoul*r2inv;
}
template<class DeviceType>
template<bool STACKPARAMS, class Specialisation>
KOKKOS_INLINE_FUNCTION
F_FLOAT PairCoulCutKokkos<DeviceType>::
compute_ecoul(const F_FLOAT& rsq, const int& i, const int&j, const int& itype,
const int& jtype, const F_FLOAT& factor_coul, const F_FLOAT& qtmp) const {
const F_FLOAT r2inv = 1.0/rsq;
const F_FLOAT rinv = sqrt(r2inv);
return factor_coul*qqrd2e * (STACKPARAMS?m_params[itype][jtype].scale:params(itype,jtype).scale)
* qtmp *q(j)*rinv;
}
/* ----------------------------------------------------------------------
allocate all arrays
------------------------------------------------------------------------- */
template<class DeviceType>
void PairCoulCutKokkos<DeviceType>::allocate()
{
PairCoulCut::allocate();
int n = atom->ntypes;
memory->destroy(cutsq);
memory->create_kokkos(k_cutsq,cutsq,n+1,n+1,"pair:cutsq");
d_cutsq = k_cutsq.template view<DeviceType>();
k_cut_ljsq = typename ArrayTypes<DeviceType>::tdual_ffloat_2d("pair:cut_ljsq",n+1,n+1);
d_cut_ljsq = k_cut_ljsq.template view<DeviceType>();
k_cut_coulsq = typename ArrayTypes<DeviceType>::tdual_ffloat_2d("pair:cut_coulsq",n+1,n+1);
d_cut_coulsq = k_cut_coulsq.template view<DeviceType>();
k_params = Kokkos::DualView<params_coul**,Kokkos::LayoutRight,DeviceType>("PairCoulCut::params",n+1,n+1);
params = k_params.d_view;
}
/* ----------------------------------------------------------------------
global settings
------------------------------------------------------------------------- */
template<class DeviceType>
void PairCoulCutKokkos<DeviceType>::settings(int narg, char **arg)
{
// \todo check what should be the limit on narg
if (narg > 2) error->all(FLERR,"Illegal pair_style command");
PairCoulCut::settings(1,arg);
}
/* ----------------------------------------------------------------------
init specific to this pair style
------------------------------------------------------------------------- */
template<class DeviceType>
void PairCoulCutKokkos<DeviceType>::init_style()
{
PairCoulCut::init_style();
// irequest = neigh request made by parent class
neighflag = lmp->kokkos->neighflag;
int irequest = neighbor->nrequest - 1;
neighbor->requests[irequest]->
kokkos_host = Kokkos::Impl::is_same<DeviceType,LMPHostType>::value &&
!Kokkos::Impl::is_same<DeviceType,LMPDeviceType>::value;
neighbor->requests[irequest]->
kokkos_device = Kokkos::Impl::is_same<DeviceType,LMPDeviceType>::value;
if (neighflag == FULL) {
neighbor->requests[irequest]->full = 1;
neighbor->requests[irequest]->half = 0;
neighbor->requests[irequest]->full_cluster = 0;
} else if (neighflag == HALF || neighflag == HALFTHREAD) {
neighbor->requests[irequest]->full = 0;
neighbor->requests[irequest]->half = 1;
neighbor->requests[irequest]->full_cluster = 0;
} else {
error->all(FLERR,"Cannot use chosen neighbor list style with coul/cut/kk");
}
}
/* ----------------------------------------------------------------------
init for one type pair i,j and corresponding j,i
------------------------------------------------------------------------- */
template<class DeviceType>
double PairCoulCutKokkos<DeviceType>::init_one(int i, int j)
{
double cutone = PairCoulCut::init_one(i,j);
k_params.h_view(i,j).scale = scale[i][j];
k_params.h_view(i,j).cutsq = cutone*cutone;
k_params.h_view(j,i) = k_params.h_view(i,j);
if(i<MAX_TYPES_STACKPARAMS+1 && j<MAX_TYPES_STACKPARAMS+1) {
m_params[i][j] = m_params[j][i] = k_params.h_view(i,j);
m_cutsq[j][i] = m_cutsq[i][j] = cutone*cutone;
m_cut_ljsq[j][i] = m_cut_ljsq[i][j] = cutone*cutone;
m_cut_coulsq[j][i] = m_cut_coulsq[i][j] = cutone*cutone;
}
k_cutsq.h_view(i,j) = cutone*cutone;
k_cutsq.template modify<LMPHostType>();
k_cut_ljsq.h_view(i,j) = cutone*cutone;
k_cut_ljsq.template modify<LMPHostType>();
k_cut_coulsq.h_view(i,j) = cutone*cutone;
k_cut_coulsq.template modify<LMPHostType>();
k_params.template modify<LMPHostType>();
return cutone;
}
namespace LAMMPS_NS {
template class PairCoulCutKokkos<LMPDeviceType>;
#ifdef KOKKOS_HAVE_CUDA
template class PairCoulCutKokkos<LMPHostType>;
#endif
}
c4science · Help