git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@8690 f3b2605a-c512-4ea7-a41b-209d697bcdaa
This commit is contained in:
sjplimp
298
src/GPU/pair_born_coul_long_gpu.cpp
Normal file
298
src/GPU/pair_born_coul_long_gpu.cpp
Normal file
@ -0,0 +1,298 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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 author: Trung Dac Nguyen (ORNL)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include "lmptype.h"
|
||||
#include "math.h"
|
||||
#include "stdio.h"
|
||||
#include "stdlib.h"
|
||||
#include "pair_born_coul_long_gpu.h"
|
||||
#include "atom.h"
|
||||
#include "atom_vec.h"
|
||||
#include "comm.h"
|
||||
#include "force.h"
|
||||
#include "neighbor.h"
|
||||
#include "neigh_list.h"
|
||||
#include "integrate.h"
|
||||
#include "math_const.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "string.h"
|
||||
#include "kspace.h"
|
||||
#include "gpu_extra.h"
|
||||
|
||||
#define EWALD_F 1.12837917
|
||||
#define EWALD_P 0.3275911
|
||||
#define A1 0.254829592
|
||||
#define A2 -0.284496736
|
||||
#define A3 1.421413741
|
||||
#define A4 -1.453152027
|
||||
#define A5 1.061405429
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
int borncl_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
|
||||
double **host_born1, double **host_born2,
|
||||
double **host_born3, double **host_a,
|
||||
double **host_c, double **host_d,
|
||||
double **sigma, double **offset, double *special_lj,
|
||||
const int inum, const int nall, const int max_nbors,
|
||||
const int maxspecial, const double cell_size,
|
||||
int &gpu_mode, FILE *screen, double **host_cut_ljsq,
|
||||
double host_cut_coulsq, double *host_special_coul,
|
||||
const double qqrd2e, const double g_ewald);
|
||||
void borncl_gpu_clear();
|
||||
int** borncl_gpu_compute_n(const int ago, const int inum_full, const int nall,
|
||||
double **host_x, int *host_type, double *sublo,
|
||||
double *subhi, int *tag, int **nspecial,
|
||||
int **special, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
int **ilist, int **jnum, const double cpu_time,
|
||||
bool &success, double *host_q, double *boxlo,
|
||||
double *prd);
|
||||
void borncl_gpu_compute(const int ago, const int inum_full, const int nall,
|
||||
double **host_x, int *host_type, int *ilist, int *numj,
|
||||
int **firstneigh, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
const double cpu_time, bool &success, double *host_q,
|
||||
const int nlocal, double *boxlo, double *prd);
|
||||
double borncl_gpu_bytes();
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
using namespace MathConst;
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairBornCoulLongGPU::PairBornCoulLongGPU(LAMMPS *lmp) :
|
||||
PairBornCoulLong(lmp), gpu_mode(GPU_FORCE)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairBornCoulLongGPU::~PairBornCoulLongGPU()
|
||||
{
|
||||
borncl_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairBornCoulLongGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
int *ilist, *numneigh, **firstneigh;
|
||||
if (gpu_mode != GPU_FORCE) {
|
||||
inum = atom->nlocal;
|
||||
firstneigh = borncl_gpu_compute_n(neighbor->ago, inum, nall, atom->x,
|
||||
atom->type, domain->sublo, domain->subhi,
|
||||
atom->tag, atom->nspecial, atom->special,
|
||||
eflag, vflag, eflag_atom, vflag_atom,
|
||||
host_start, &ilist, &numneigh, cpu_time,
|
||||
success, atom->q, domain->boxlo,
|
||||
domain->prd);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
borncl_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
|
||||
ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, cpu_time, success, atom->q,
|
||||
atom->nlocal, domain->boxlo, domain->prd);
|
||||
}
|
||||
if (!success)
|
||||
error->one(FLERR,"Out of memory on GPGPU");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairBornCoulLongGPU::init_style()
|
||||
{
|
||||
if (!atom->q_flag)
|
||||
error->all(FLERR,
|
||||
"Pair style born/coul/long/gpu requires atom attribute q");
|
||||
if (force->newton_pair)
|
||||
error->all(FLERR,
|
||||
"Cannot use newton pair with born/coul/long/gpu pair style");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double maxcut = -1.0;
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
|
||||
cut = init_one(i,j);
|
||||
cut *= cut;
|
||||
if (cut > maxcut)
|
||||
maxcut = cut;
|
||||
cutsq[i][j] = cutsq[j][i] = cut;
|
||||
} else
|
||||
cutsq[i][j] = cutsq[j][i] = 0.0;
|
||||
}
|
||||
}
|
||||
double cell_size = sqrt(maxcut) + neighbor->skin;
|
||||
|
||||
cut_coulsq = cut_coul * cut_coul;
|
||||
|
||||
// insure use of KSpace long-range solver, set g_ewald
|
||||
|
||||
if (force->kspace == NULL)
|
||||
error->all(FLERR,"Pair style is incompatible with KSpace style");
|
||||
g_ewald = force->kspace->g_ewald;
|
||||
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
int success = borncl_gpu_init(atom->ntypes+1, cutsq, rhoinv,
|
||||
born1, born2, born3, a, c, d, sigma,
|
||||
offset, force->special_lj, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen, cut_ljsq,
|
||||
cut_coulsq, force->special_coul,
|
||||
force->qqrd2e, g_ewald);
|
||||
|
||||
GPU_EXTRA::check_flag(success,error,world);
|
||||
|
||||
if (gpu_mode == GPU_FORCE) {
|
||||
int irequest = neighbor->request(this);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairBornCoulLongGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + borncl_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairBornCoulLongGPU::cpu_compute(int start, int inum, int eflag,
|
||||
int vflag, int *ilist, int *numneigh,
|
||||
int **firstneigh)
|
||||
{
|
||||
int i,j,ii,jj,jnum,itype,jtype,itable;
|
||||
double qtmp,xtmp,ytmp,ztmp,delx,dely,delz,evdwl,ecoul,fpair;
|
||||
double fraction,table;
|
||||
double r,rexp,r2inv,r6inv,forcecoul,forceborn,factor_coul,factor_lj;
|
||||
double grij,expm2,prefactor,t,erfc;
|
||||
int *jlist;
|
||||
double rsq;
|
||||
|
||||
evdwl = ecoul = 0.0;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
double *q = atom->q;
|
||||
int *type = atom->type;
|
||||
double *special_coul = force->special_coul;
|
||||
double *special_lj = force->special_lj;
|
||||
double qqrd2e = force->qqrd2e;
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
qtmp = q[i];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
factor_lj = special_lj[sbmask(j)];
|
||||
factor_coul = special_coul[sbmask(j)];
|
||||
j &= NEIGHMASK;
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
jtype = type[j];
|
||||
|
||||
if (rsq < cutsq[itype][jtype]) {
|
||||
r2inv = 1.0/rsq;
|
||||
r = sqrt(rsq);
|
||||
|
||||
if (rsq < cut_coulsq) {
|
||||
grij = g_ewald * r;
|
||||
expm2 = exp(-grij*grij);
|
||||
t = 1.0 / (1.0 + EWALD_P*grij);
|
||||
erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;
|
||||
prefactor = qqrd2e * qtmp*q[j]/r;
|
||||
forcecoul = prefactor * (erfc + EWALD_F*grij*expm2);
|
||||
if (factor_coul < 1.0) forcecoul -= (1.0-factor_coul)*prefactor;
|
||||
} else forcecoul = 0.0;
|
||||
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
r6inv = r2inv*r2inv*r2inv;
|
||||
rexp = exp((sigma[itype][jtype]-r)*rhoinv[itype][jtype]);
|
||||
forceborn = born1[itype][jtype]*r*rexp - born2[itype][jtype]*r6inv
|
||||
+ born3[itype][jtype]*r2inv*r6inv;
|
||||
} else forceborn = 0.0;
|
||||
|
||||
fpair = (forcecoul + factor_lj*forceborn) * r2inv;
|
||||
|
||||
f[i][0] += delx*fpair;
|
||||
f[i][1] += dely*fpair;
|
||||
f[i][2] += delz*fpair;
|
||||
|
||||
if (eflag) {
|
||||
if (rsq < cut_coulsq) {
|
||||
ecoul = prefactor*erfc;
|
||||
if (factor_coul < 1.0) ecoul -= (1.0-factor_coul)*prefactor;
|
||||
} else ecoul = 0.0;
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
evdwl = a[itype][jtype]*rexp - c[itype][jtype]*r6inv
|
||||
+ d[itype][jtype]*r6inv*r2inv - offset[itype][jtype];
|
||||
evdwl *= factor_lj;
|
||||
} else evdwl = 0.0;
|
||||
}
|
||||
|
||||
if (evflag) ev_tally_full(i,evdwl,ecoul,fpair,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
62
src/GPU/pair_born_coul_long_gpu.h
Normal file
62
src/GPU/pair_born_coul_long_gpu.h
Normal file
@ -0,0 +1,62 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(born/coul/long/gpu,PairBornCoulLongGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_BORN_COUL_LONG_GPU_H
|
||||
#define LMP_PAIR_BORN_COUL_LONG_GPU_H
|
||||
|
||||
#include "pair_born_coul_long.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairBornCoulLongGPU : public PairBornCoulLong {
|
||||
public:
|
||||
PairBornCoulLongGPU(LAMMPS *lmp);
|
||||
~PairBornCoulLongGPU();
|
||||
void cpu_compute(int, int, int, int, int *, int *, int **);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
int *gpulist;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/* ERROR/WARNING messages:
|
||||
|
||||
E: Out of memory on GPGPU
|
||||
|
||||
UNDOCUMENTED
|
||||
|
||||
E: Cannot use newton pair with born/coul/long/gpu pair style
|
||||
|
||||
UNDOCUMENTED
|
||||
|
||||
E: Pair style born/coul/long/gpu requires atom attribute q
|
||||
|
||||
The atom style defined does not have this attribute.
|
||||
|
||||
*/
|
||||
287
src/GPU/pair_born_coul_wolf_gpu.cpp
Normal file
287
src/GPU/pair_born_coul_wolf_gpu.cpp
Normal file
@ -0,0 +1,287 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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: Trung Dac Nguyen (ORNL)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include "lmptype.h"
|
||||
#include "math.h"
|
||||
#include "stdio.h"
|
||||
#include "stdlib.h"
|
||||
#include "pair_born_coul_wolf_gpu.h"
|
||||
#include "atom.h"
|
||||
#include "atom_vec.h"
|
||||
#include "comm.h"
|
||||
#include "force.h"
|
||||
#include "neighbor.h"
|
||||
#include "neigh_list.h"
|
||||
#include "integrate.h"
|
||||
#include "math_const.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "string.h"
|
||||
#include "gpu_extra.h"
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
int borncw_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
|
||||
double **host_born1, double **host_born2,
|
||||
double **host_born3, double **host_a, double **host_c,
|
||||
double **host_d, double **sigma, double **offset,
|
||||
double *special_lj, const int inum,
|
||||
const int nall, const int max_nbors, const int maxspecial,
|
||||
const double cell_size, int &gpu_mode, FILE *screen,
|
||||
double **host_cut_ljsq, double host_cut_coulsq,
|
||||
double *host_special_coul, const double qqrd2e,
|
||||
const double alf, const double e_shift, const double f_shift);
|
||||
void borncw_gpu_clear();
|
||||
int ** borncw_gpu_compute_n(const int ago, const int inum_full, const int nall,
|
||||
double **host_x, int *host_type, double *sublo,
|
||||
double *subhi, int *tag, int **nspecial,
|
||||
int **special, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
int **ilist, int **jnum, const double cpu_time,
|
||||
bool &success, double *host_q, double *boxlo,
|
||||
double *prd);
|
||||
void borncw_gpu_compute(const int ago, const int inum_full, const int nall,
|
||||
double **host_x, int *host_type, int *ilist, int *numj,
|
||||
int **firstneigh, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
const double cpu_time, bool &success, double *host_q,
|
||||
const int nlocal, double *boxlo, double *prd);
|
||||
double borncw_gpu_bytes();
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
using namespace MathConst;
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairBornCoulWolfGPU::PairBornCoulWolfGPU(LAMMPS *lmp) : PairBornCoulWolf(lmp),
|
||||
gpu_mode(GPU_FORCE)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairBornCoulWolfGPU::~PairBornCoulWolfGPU()
|
||||
{
|
||||
borncw_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairBornCoulWolfGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
int *ilist, *numneigh, **firstneigh;
|
||||
if (gpu_mode != GPU_FORCE) {
|
||||
inum = atom->nlocal;
|
||||
firstneigh = borncw_gpu_compute_n(neighbor->ago, inum, nall,
|
||||
atom->x, atom->type, domain->sublo,
|
||||
domain->subhi, atom->tag, atom->nspecial,
|
||||
atom->special, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start,
|
||||
&ilist, &numneigh, cpu_time, success,
|
||||
atom->q, domain->boxlo, domain->prd);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
borncw_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
|
||||
ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, cpu_time, success, atom->q,
|
||||
atom->nlocal, domain->boxlo, domain->prd);
|
||||
}
|
||||
if (!success)
|
||||
error->one(FLERR,"Out of memory on GPGPU");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairBornCoulWolfGPU::init_style()
|
||||
{
|
||||
if (force->newton_pair)
|
||||
error->all(FLERR,
|
||||
"Cannot use newton pair with born/coul/wolf/gpu pair style");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double maxcut = -1.0;
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
|
||||
cut = init_one(i,j);
|
||||
cut *= cut;
|
||||
if (cut > maxcut)
|
||||
maxcut = cut;
|
||||
cutsq[i][j] = cutsq[j][i] = cut;
|
||||
} else
|
||||
cutsq[i][j] = cutsq[j][i] = 0.0;
|
||||
}
|
||||
}
|
||||
double cell_size = sqrt(maxcut) + neighbor->skin;
|
||||
|
||||
cut_coulsq = cut_coul * cut_coul;
|
||||
|
||||
double e_shift = erfc(alf*cut_coul)/cut_coul;
|
||||
double f_shift = -(e_shift+ 2.0*alf/MY_PIS * exp(-alf*alf*cut_coul*cut_coul)) /
|
||||
cut_coul;
|
||||
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
int success = borncw_gpu_init(atom->ntypes+1, cutsq, rhoinv,
|
||||
born1, born2, born3, a, c, d, sigma, offset,
|
||||
force->special_lj, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen, cut_ljsq,
|
||||
cut_coulsq, force->special_coul, force->qqrd2e,
|
||||
alf, e_shift, f_shift);
|
||||
GPU_EXTRA::check_flag(success,error,world);
|
||||
|
||||
if (gpu_mode == GPU_FORCE) {
|
||||
int irequest = neighbor->request(this);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairBornCoulWolfGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + borncw_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairBornCoulWolfGPU::cpu_compute(int start, int inum, int eflag, int vflag,
|
||||
int *ilist, int *numneigh,
|
||||
int **firstneigh) {
|
||||
int i,j,ii,jj,jnum,itype,jtype;
|
||||
double xtmp,ytmp,ztmp,qtmp,delx,dely,delz,evdwl,ecoul,fpair;
|
||||
double rsq,r2inv,r6inv,forcecoul,forceborn,factor_coul,factor_lj;
|
||||
double prefactor;
|
||||
double r,rexp;
|
||||
int *jlist;
|
||||
double erfcc,erfcd,v_sh,dvdrr,e_self,e_shift,f_shift,qisq;
|
||||
|
||||
evdwl = ecoul = 0.0;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
double *q = atom->q;
|
||||
int *type = atom->type;
|
||||
int nlocal = atom->nlocal;
|
||||
double *special_coul = force->special_coul;
|
||||
double *special_lj = force->special_lj;
|
||||
double qqrd2e = force->qqrd2e;
|
||||
|
||||
qisq = qtmp*qtmp;
|
||||
e_self = -(e_shift/2.0 + alf/MY_PIS) * qisq*qqrd2e;
|
||||
if (evflag) ev_tally(i,i,nlocal,0,0.0,e_self,0.0,0.0,0.0,0.0);
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
qtmp = q[i];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
factor_lj = special_lj[sbmask(j)];
|
||||
factor_coul = special_coul[sbmask(j)];
|
||||
j &= NEIGHMASK;
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
jtype = type[j];
|
||||
|
||||
if (rsq < cutsq[itype][jtype]) {
|
||||
r2inv = 1.0/rsq;
|
||||
|
||||
if (rsq < cut_coulsq) {
|
||||
r = sqrt(rsq);
|
||||
prefactor = qqrd2e*qtmp*q[j]/r;
|
||||
erfcc = erfc(alf*r);
|
||||
erfcd = exp(-alf*alf*r*r);
|
||||
v_sh = (erfcc - e_shift*r) * prefactor;
|
||||
dvdrr = (erfcc/rsq + 2.0*alf/MY_PIS * erfcd/r) + f_shift;
|
||||
forcecoul = dvdrr*rsq*prefactor;
|
||||
if (factor_coul < 1.0) forcecoul -= (1.0-factor_coul)*prefactor;
|
||||
} else forcecoul = 0.0;
|
||||
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
r6inv = r2inv*r2inv*r2inv;
|
||||
r = sqrt(rsq);
|
||||
rexp = exp((sigma[itype][jtype]-r)*rhoinv[itype][jtype]);
|
||||
forceborn = born1[itype][jtype]*r*rexp - born2[itype][jtype]*r6inv +
|
||||
born3[itype][jtype]*r2inv*r6inv;
|
||||
} else forceborn = 0.0;
|
||||
|
||||
fpair = (factor_coul*forcecoul + factor_lj*forceborn) * r2inv;
|
||||
|
||||
f[i][0] += delx*fpair;
|
||||
f[i][1] += dely*fpair;
|
||||
f[i][2] += delz*fpair;
|
||||
|
||||
if (eflag) {
|
||||
if (rsq < cut_coulsq)
|
||||
ecoul = v_sh;
|
||||
else ecoul = 0.0;
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
evdwl = a[itype][jtype]*rexp - c[itype][jtype]*r6inv +
|
||||
d[itype][jtype]*r6inv*r2inv - offset[itype][jtype];
|
||||
evdwl *= factor_lj;
|
||||
} else evdwl = 0.0;
|
||||
}
|
||||
|
||||
if (evflag) ev_tally_full(i,evdwl,ecoul,fpair,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
62
src/GPU/pair_born_coul_wolf_gpu.h
Normal file
62
src/GPU/pair_born_coul_wolf_gpu.h
Normal file
@ -0,0 +1,62 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(born/coul/wolf/gpu,PairBornCoulWolfGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_BORN_COUL_WOLF_GPU_H
|
||||
#define LMP_PAIR_BORN_COUL_WOLF_GPU_H
|
||||
|
||||
#include "pair_born_coul_wolf.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairBornCoulWolfGPU : public PairBornCoulWolf {
|
||||
public:
|
||||
PairBornCoulWolfGPU(LAMMPS *lmp);
|
||||
~PairBornCoulWolfGPU();
|
||||
void cpu_compute(int, int, int, int, int *, int *, int **);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
int *gpulist;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/* ERROR/WARNING messages:
|
||||
|
||||
E: Out of memory on GPGPU
|
||||
|
||||
UNDOCUMENTED
|
||||
|
||||
E: Cannot use newton pair with born/coul/wolf/gpu pair style
|
||||
|
||||
UNDOCUMENTED
|
||||
|
||||
E: Pair style born/coul/wolf/gpu requires atom attribute q
|
||||
|
||||
The atom style defined does not have this attribute.
|
||||
|
||||
*/
|
||||
233
src/GPU/pair_born_gpu.cpp
Normal file
233
src/GPU/pair_born_gpu.cpp
Normal file
@ -0,0 +1,233 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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 author: Trung Dac Nguyen (ORNL)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include "lmptype.h"
|
||||
#include "math.h"
|
||||
#include "stdio.h"
|
||||
#include "stdlib.h"
|
||||
#include "pair_born_gpu.h"
|
||||
#include "atom.h"
|
||||
#include "atom_vec.h"
|
||||
#include "comm.h"
|
||||
#include "force.h"
|
||||
#include "neighbor.h"
|
||||
#include "neigh_list.h"
|
||||
#include "integrate.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "string.h"
|
||||
#include "gpu_extra.h"
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
int born_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
|
||||
double **host_born1, double **host_born2, double **host_born3,
|
||||
double **host_a, double **host_c, double **host_d,
|
||||
double **host_sigma, double **offset, double *special_lj,
|
||||
const int inum, const int nall, const int max_nbors,
|
||||
const int maxspecial, const double cell_size,
|
||||
int &gpu_mode, FILE *screen);
|
||||
void born_gpu_clear();
|
||||
int ** born_gpu_compute_n(const int ago, const int inum_full,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
double *sublo, double *subhi, int *tag, int **nspecial,
|
||||
int **special, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
int **ilist, int **jnum, const double cpu_time,
|
||||
bool &success);
|
||||
void born_gpu_compute(const int ago, const int inum_full, const int nall,
|
||||
double **host_x, int *host_type, int *ilist, int *numj,
|
||||
int **firstneigh, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
const double cpu_time, bool &success);
|
||||
double born_gpu_bytes();
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairBornGPU::PairBornGPU(LAMMPS *lmp) : PairBorn(lmp), gpu_mode(GPU_FORCE)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairBornGPU::~PairBornGPU()
|
||||
{
|
||||
born_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairBornGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
int *ilist, *numneigh, **firstneigh;
|
||||
if (gpu_mode != GPU_FORCE) {
|
||||
inum = atom->nlocal;
|
||||
firstneigh = born_gpu_compute_n(neighbor->ago, inum, nall,
|
||||
atom->x, atom->type, domain->sublo,
|
||||
domain->subhi, atom->tag, atom->nspecial,
|
||||
atom->special, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start,
|
||||
&ilist, &numneigh, cpu_time, success);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
born_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
|
||||
ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, cpu_time, success);
|
||||
}
|
||||
if (!success)
|
||||
error->one(FLERR,"Out of memory on GPGPU");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairBornGPU::init_style()
|
||||
{
|
||||
if (force->newton_pair)
|
||||
error->all(FLERR,"Cannot use newton pair with born/gpu pair style");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double maxcut = -1.0;
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
|
||||
cut = init_one(i,j);
|
||||
cut *= cut;
|
||||
if (cut > maxcut)
|
||||
maxcut = cut;
|
||||
cutsq[i][j] = cutsq[j][i] = cut;
|
||||
} else
|
||||
cutsq[i][j] = cutsq[j][i] = 0.0;
|
||||
}
|
||||
}
|
||||
double cell_size = sqrt(maxcut) + neighbor->skin;
|
||||
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
int success = born_gpu_init(atom->ntypes+1, cutsq, rhoinv,
|
||||
born1, born2, born3, a, c, d, sigma,
|
||||
offset, force->special_lj, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen);
|
||||
GPU_EXTRA::check_flag(success,error,world);
|
||||
|
||||
if (gpu_mode == GPU_FORCE) {
|
||||
int irequest = neighbor->request(this);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairBornGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + born_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairBornGPU::cpu_compute(int start, int inum, int eflag, int vflag,
|
||||
int *ilist, int *numneigh, int **firstneigh) {
|
||||
int i,j,ii,jj,jnum,itype,jtype;
|
||||
double xtmp,ytmp,ztmp,delx,dely,delz,evdwl,fpair;
|
||||
double rsq,r2inv,r6inv,forceborn,factor_lj;
|
||||
double r,rexp;
|
||||
int *jlist;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
int *type = atom->type;
|
||||
double *special_lj = force->special_lj;
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
factor_lj = special_lj[sbmask(j)];
|
||||
j &= NEIGHMASK;
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
jtype = type[j];
|
||||
|
||||
if (rsq < cutsq[itype][jtype]) {
|
||||
r2inv = 1.0/rsq;
|
||||
r6inv = r2inv*r2inv*r2inv;
|
||||
r = sqrt(rsq);
|
||||
rexp = exp((sigma[itype][jtype]-r)*rhoinv[itype][jtype]);
|
||||
forceborn = born1[itype][jtype]*r*rexp - born2[itype][jtype]*r6inv +
|
||||
born3[itype][jtype]*r2inv*r6inv;
|
||||
fpair = factor_lj*forceborn*r2inv;
|
||||
|
||||
f[i][0] += delx*fpair;
|
||||
f[i][1] += dely*fpair;
|
||||
f[i][2] += delz*fpair;
|
||||
|
||||
if (eflag) {
|
||||
evdwl = a[itype][jtype]*rexp - c[itype][jtype]*r6inv +
|
||||
d[itype][jtype]*r6inv*r2inv - offset[itype][jtype];
|
||||
evdwl *= factor_lj;
|
||||
}
|
||||
|
||||
if (evflag) ev_tally_full(i,evdwl,0.0,fpair,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
58
src/GPU/pair_born_gpu.h
Normal file
58
src/GPU/pair_born_gpu.h
Normal file
@ -0,0 +1,58 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(born/gpu,PairBornGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_BORN_GPU_H
|
||||
#define LMP_PAIR_BORN_GPU_H
|
||||
|
||||
#include "pair_born.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairBornGPU : public PairBorn {
|
||||
public:
|
||||
PairBornGPU(LAMMPS *lmp);
|
||||
~PairBornGPU();
|
||||
void cpu_compute(int, int, int, int, int *, int *, int **);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
int *gpulist;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/* ERROR/WARNING messages:
|
||||
|
||||
E: Out of memory on GPGPU
|
||||
|
||||
UNDOCUMENTED
|
||||
|
||||
E: Cannot use newton pair with born/gpu pair style
|
||||
|
||||
UNDOCUMENTED
|
||||
|
||||
*/
|
||||
307
src/GPU/pair_colloid_gpu.cpp
Normal file
307
src/GPU/pair_colloid_gpu.cpp
Normal file
@ -0,0 +1,307 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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 author: Trung Dac Nguyen (ORNL)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include "lmptype.h"
|
||||
#include "math.h"
|
||||
#include "stdio.h"
|
||||
#include "stdlib.h"
|
||||
#include "pair_colloid_gpu.h"
|
||||
#include "atom.h"
|
||||
#include "atom_vec.h"
|
||||
#include "comm.h"
|
||||
#include "force.h"
|
||||
#include "neighbor.h"
|
||||
#include "neigh_list.h"
|
||||
#include "integrate.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "string.h"
|
||||
#include "gpu_extra.h"
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
int colloid_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
|
||||
double **host_lj2, double **host_lj3, double **host_lj4,
|
||||
double **offset, double *special_lj, double **host_a12,
|
||||
double **host_a1, double **host_a2, double **host_d1,
|
||||
double **host_d2, double **host_sigma3, double **host_sigma6,
|
||||
int **host_form, const int nlocal,
|
||||
const int nall, const int max_nbors, const int maxspecial,
|
||||
const double cell_size, int &gpu_mode, FILE *screen);
|
||||
void colloid_gpu_clear();
|
||||
int ** colloid_gpu_compute_n(const int ago, const int inum,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
double *sublo, double *subhi, int *tag, int **nspecial,
|
||||
int **special, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
int **ilist, int **jnum,
|
||||
const double cpu_time, bool &success);
|
||||
void colloid_gpu_compute(const int ago, const int inum, const int nall,
|
||||
double **host_x, int *host_type, int *ilist, int *numj,
|
||||
int **firstneigh, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
const double cpu_time, bool &success);
|
||||
double colloid_gpu_bytes();
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairColloidGPU::PairColloidGPU(LAMMPS *lmp) : PairColloid(lmp), gpu_mode(GPU_FORCE)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairColloidGPU::~PairColloidGPU()
|
||||
{
|
||||
colloid_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairColloidGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
int *ilist, *numneigh, **firstneigh;
|
||||
if (gpu_mode != GPU_FORCE) {
|
||||
inum = atom->nlocal;
|
||||
firstneigh = colloid_gpu_compute_n(neighbor->ago, inum, nall,
|
||||
atom->x, atom->type, domain->sublo,
|
||||
domain->subhi, atom->tag, atom->nspecial,
|
||||
atom->special, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start,
|
||||
&ilist, &numneigh, cpu_time, success);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
colloid_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
|
||||
ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, cpu_time, success);
|
||||
}
|
||||
if (!success)
|
||||
error->one(FLERR,"Insufficient memory on accelerator");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairColloidGPU::init_style()
|
||||
{
|
||||
if (force->newton_pair)
|
||||
error->all(FLERR,"Cannot use newton pair with colloid/gpu pair style");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double maxcut = -1.0;
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
|
||||
cut = init_one(i,j);
|
||||
cut *= cut;
|
||||
if (cut > maxcut)
|
||||
maxcut = cut;
|
||||
cutsq[i][j] = cutsq[j][i] = cut;
|
||||
} else
|
||||
cutsq[i][j] = cutsq[j][i] = 0.0;
|
||||
}
|
||||
}
|
||||
double cell_size = sqrt(maxcut) + neighbor->skin;
|
||||
|
||||
int **_form = NULL;
|
||||
int n=atom->ntypes;
|
||||
memory->create(_form,n+1,n+1,"colloid/gpu:_form");
|
||||
for (int i = 1; i <= n; i++) {
|
||||
for (int j = 1; j <= n; j++) {
|
||||
if (form[i][j] == SMALL_SMALL) _form[i][j] = 0;
|
||||
else if (form[i][j] == SMALL_LARGE) _form[i][j] = 1;
|
||||
else if (form[i][j] == LARGE_LARGE) _form[i][j] = 2;
|
||||
}
|
||||
}
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
int success = colloid_gpu_init(atom->ntypes+1, cutsq, lj1, lj2, lj3, lj4,
|
||||
offset, force->special_lj, a12, a1, a2,
|
||||
d1, d2, sigma3, sigma6, _form, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen);
|
||||
memory->destroy(_form);
|
||||
GPU_EXTRA::check_flag(success,error,world);
|
||||
|
||||
if (gpu_mode == GPU_FORCE) {
|
||||
int irequest = neighbor->request(this);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairColloidGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + colloid_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairColloidGPU::cpu_compute(int start, int inum, int eflag, int vflag,
|
||||
int *ilist, int *numneigh, int **firstneigh)
|
||||
{
|
||||
int i,j,ii,jj,jnum,itype,jtype;
|
||||
double xtmp,ytmp,ztmp,delx,dely,delz,evdwl,fpair;
|
||||
double r,rsq,r2inv,r6inv,forcelj,factor_lj;
|
||||
double c1,c2,fR,dUR,dUA;
|
||||
double K[9],h[4],g[4];
|
||||
int *jlist;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
int *type = atom->type;
|
||||
double *special_lj = force->special_lj;
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
factor_lj = special_lj[sbmask(j)];
|
||||
j &= NEIGHMASK;
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
jtype = type[j];
|
||||
|
||||
if (rsq >= cutsq[itype][jtype]) continue;
|
||||
|
||||
switch (form[itype][jtype]) {
|
||||
case SMALL_SMALL:
|
||||
r2inv = 1.0/rsq;
|
||||
r6inv = r2inv*r2inv*r2inv;
|
||||
forcelj = r6inv * (lj1[itype][jtype]*r6inv - lj2[itype][jtype]);
|
||||
fpair = factor_lj*forcelj*r2inv;
|
||||
if (eflag)
|
||||
evdwl = r6inv*(r6inv*lj3[itype][jtype]-lj4[itype][jtype]) -
|
||||
offset[itype][jtype];
|
||||
break;
|
||||
|
||||
case SMALL_LARGE:
|
||||
c2 = a2[itype][jtype];
|
||||
K[1] = c2*c2;
|
||||
K[2] = rsq;
|
||||
K[0] = K[1] - rsq;
|
||||
K[4] = rsq*rsq;
|
||||
K[3] = K[1] - K[2];
|
||||
K[3] *= K[3]*K[3];
|
||||
K[6] = K[3]*K[3];
|
||||
fR = sigma3[itype][jtype]*a12[itype][jtype]*c2*K[1]/K[3];
|
||||
fpair = 4.0/15.0*fR*factor_lj *
|
||||
(2.0*(K[1]+K[2]) * (K[1]*(5.0*K[1]+22.0*K[2])+5.0*K[4]) *
|
||||
sigma6[itype][jtype]/K[6]-5.0) / K[0];
|
||||
if (eflag)
|
||||
evdwl = 2.0/9.0*fR *
|
||||
(1.0-(K[1]*(K[1]*(K[1]/3.0+3.0*K[2])+4.2*K[4])+K[2]*K[4]) *
|
||||
sigma6[itype][jtype]/K[6]) - offset[itype][jtype];
|
||||
if (rsq <= K[1])
|
||||
error->one(FLERR,"Overlapping small/large in pair colloid");
|
||||
break;
|
||||
|
||||
case LARGE_LARGE:
|
||||
r = sqrt(rsq);
|
||||
c1 = a1[itype][jtype];
|
||||
c2 = a2[itype][jtype];
|
||||
K[0] = c1*c2;
|
||||
K[1] = c1+c2;
|
||||
K[2] = c1-c2;
|
||||
K[3] = K[1]+r;
|
||||
K[4] = K[1]-r;
|
||||
K[5] = K[2]+r;
|
||||
K[6] = K[2]-r;
|
||||
K[7] = 1.0/(K[3]*K[4]);
|
||||
K[8] = 1.0/(K[5]*K[6]);
|
||||
g[0] = pow(K[3],-7.0);
|
||||
g[1] = pow(K[4],-7.0);
|
||||
g[2] = pow(K[5],-7.0);
|
||||
g[3] = pow(K[6],-7.0);
|
||||
h[0] = ((K[3]+5.0*K[1])*K[3]+30.0*K[0])*g[0];
|
||||
h[1] = ((K[4]+5.0*K[1])*K[4]+30.0*K[0])*g[1];
|
||||
h[2] = ((K[5]+5.0*K[2])*K[5]-30.0*K[0])*g[2];
|
||||
h[3] = ((K[6]+5.0*K[2])*K[6]-30.0*K[0])*g[3];
|
||||
g[0] *= 42.0*K[0]/K[3]+6.0*K[1]+K[3];
|
||||
g[1] *= 42.0*K[0]/K[4]+6.0*K[1]+K[4];
|
||||
g[2] *= -42.0*K[0]/K[5]+6.0*K[2]+K[5];
|
||||
g[3] *= -42.0*K[0]/K[6]+6.0*K[2]+K[6];
|
||||
|
||||
fR = a12[itype][jtype]*sigma6[itype][jtype]/r/37800.0;
|
||||
evdwl = fR * (h[0]-h[1]-h[2]+h[3]);
|
||||
dUR = evdwl/r + 5.0*fR*(g[0]+g[1]-g[2]-g[3]);
|
||||
dUA = -a12[itype][jtype]/3.0*r*((2.0*K[0]*K[7]+1.0)*K[7] +
|
||||
(2.0*K[0]*K[8]-1.0)*K[8]);
|
||||
fpair = factor_lj * (dUR+dUA)/r;
|
||||
if (eflag)
|
||||
evdwl += a12[itype][jtype]/6.0 *
|
||||
(2.0*K[0]*(K[7]+K[8])-log(K[8]/K[7])) - offset[itype][jtype];
|
||||
if (r <= K[1])
|
||||
error->one(FLERR,"Overlapping large/large in pair colloid");
|
||||
break;
|
||||
}
|
||||
|
||||
if (eflag) evdwl *= factor_lj;
|
||||
|
||||
f[i][0] += delx*fpair;
|
||||
f[i][1] += dely*fpair;
|
||||
f[i][2] += delz*fpair;
|
||||
|
||||
if (evflag) ev_tally_full(i,evdwl,0.0,fpair,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
59
src/GPU/pair_colloid_gpu.h
Normal file
59
src/GPU/pair_colloid_gpu.h
Normal file
@ -0,0 +1,59 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(colloid/gpu,PairColloidGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_COLLOID_GPU_H
|
||||
#define LMP_PAIR_COLLOID_GPU_H
|
||||
|
||||
#include "pair_colloid.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairColloidGPU : public PairColloid {
|
||||
public:
|
||||
PairColloidGPU(LAMMPS *lmp);
|
||||
~PairColloidGPU();
|
||||
void cpu_compute(int, int, int, int, int *, int *, int **);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
int *gpulist;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/* ERROR/WARNING messages:
|
||||
|
||||
E: Insufficient memory on accelerator
|
||||
|
||||
There is insufficient memory on one of the devices specified for the gpu
|
||||
package
|
||||
|
||||
E: Cannot use newton pair with colloid/gpu pair style
|
||||
|
||||
Self-explanatory.
|
||||
|
||||
*/
|
||||
367
src/GPU/pair_dipole_cut_gpu.cpp
Normal file
367
src/GPU/pair_dipole_cut_gpu.cpp
Normal file
@ -0,0 +1,367 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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 author: Trung Dac Nguyen (ORNL)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include "lmptype.h"
|
||||
#include "math.h"
|
||||
#include "stdio.h"
|
||||
#include "stdlib.h"
|
||||
#include "pair_dipole_cut_gpu.h"
|
||||
#include "atom.h"
|
||||
#include "atom_vec.h"
|
||||
#include "comm.h"
|
||||
#include "force.h"
|
||||
#include "neighbor.h"
|
||||
#include "neigh_list.h"
|
||||
#include "integrate.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "string.h"
|
||||
#include "gpu_extra.h"
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
int dpl_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
|
||||
double **host_lj2, double **host_lj3, double **host_lj4,
|
||||
double **offset, double *special_lj, const int nlocal,
|
||||
const int nall, const int max_nbors, const int maxspecial,
|
||||
const double cell_size, int &gpu_mode, FILE *screen,
|
||||
double **host_cut_ljsq, double **host_cut_coulsq,
|
||||
double *host_special_coul, const double qqrd2e);
|
||||
void dpl_gpu_clear();
|
||||
int ** dpl_gpu_compute_n(const int ago, const int inum,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
double *sublo, double *subhi, int *tag,
|
||||
int **nspecial, int **special, const bool eflag,
|
||||
const bool vflag, const bool eatom, const bool vatom,
|
||||
int &host_start, int **ilist, int **jnum,
|
||||
const double cpu_time, bool &success,
|
||||
double *host_q, double **host_mu,
|
||||
double *boxlo, double *prd);
|
||||
void dpl_gpu_compute(const int ago, const int inum,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
int *ilist, int *numj, int **firstneigh,
|
||||
const bool eflag, const bool vflag, const bool eatom,
|
||||
const bool vatom, int &host_start, const double cpu_time,
|
||||
bool &success, double *host_q, double **host_mu,
|
||||
const int nlocal, double *boxlo, double *prd);
|
||||
double dpl_gpu_bytes();
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairDipoleCutGPU::PairDipoleCutGPU(LAMMPS *lmp) : PairDipoleCut(lmp),
|
||||
gpu_mode(GPU_FORCE)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairDipoleCutGPU::~PairDipoleCutGPU()
|
||||
{
|
||||
dpl_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairDipoleCutGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
int *ilist, *numneigh, **firstneigh;
|
||||
if (gpu_mode != GPU_FORCE) {
|
||||
inum = atom->nlocal;
|
||||
firstneigh = dpl_gpu_compute_n(neighbor->ago, inum, nall, atom->x,
|
||||
atom->type, domain->sublo, domain->subhi,
|
||||
atom->tag, atom->nspecial, atom->special,
|
||||
eflag, vflag, eflag_atom, vflag_atom,
|
||||
host_start, &ilist, &numneigh, cpu_time,
|
||||
success, atom->q, atom->mu, domain->boxlo,
|
||||
domain->prd);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
dpl_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
|
||||
ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, cpu_time, success, atom->q,
|
||||
atom->mu, atom->nlocal, domain->boxlo, domain->prd);
|
||||
}
|
||||
if (!success)
|
||||
error->one(FLERR,"Insufficient memory on accelerator");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairDipoleCutGPU::init_style()
|
||||
{
|
||||
if (!atom->q_flag || !atom->mu_flag || !atom->torque_flag)
|
||||
error->all(FLERR,"Pair dipole/cut/gpu requires atom attributes q, mu, torque");
|
||||
|
||||
if (force->newton_pair)
|
||||
error->all(FLERR,"Cannot use newton pair with dipole/cut/gpu pair style");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double maxcut = -1.0;
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
|
||||
cut = init_one(i,j);
|
||||
cut *= cut;
|
||||
if (cut > maxcut)
|
||||
maxcut = cut;
|
||||
cutsq[i][j] = cutsq[j][i] = cut;
|
||||
} else
|
||||
cutsq[i][j] = cutsq[j][i] = 0.0;
|
||||
}
|
||||
}
|
||||
double cell_size = sqrt(maxcut) + neighbor->skin;
|
||||
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
int success = dpl_gpu_init(atom->ntypes+1, cutsq, lj1, lj2, lj3, lj4,
|
||||
offset, force->special_lj, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen, cut_ljsq, cut_coulsq,
|
||||
force->special_coul, force->qqrd2e);
|
||||
GPU_EXTRA::check_flag(success,error,world);
|
||||
|
||||
if (gpu_mode == GPU_FORCE) {
|
||||
int irequest = neighbor->request(this);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairDipoleCutGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + dpl_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairDipoleCutGPU::cpu_compute(int start, int inum, int eflag, int vflag,
|
||||
int *ilist, int *numneigh,
|
||||
int **firstneigh)
|
||||
{
|
||||
int i,j,ii,jj,jnum,itype,jtype;
|
||||
double qtmp,xtmp,ytmp,ztmp,delx,dely,delz,evdwl,ecoul,fx,fy,fz;
|
||||
double rsq,rinv,r2inv,r6inv,r3inv,r5inv,r7inv;
|
||||
double forcecoulx,forcecouly,forcecoulz,crossx,crossy,crossz;
|
||||
double tixcoul,tiycoul,tizcoul,tjxcoul,tjycoul,tjzcoul;
|
||||
double fq,pdotp,pidotr,pjdotr,pre1,pre2,pre3,pre4;
|
||||
double forcelj,factor_coul,factor_lj;
|
||||
int *jlist;
|
||||
|
||||
evdwl = ecoul = 0.0;
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
double *q = atom->q;
|
||||
double **mu = atom->mu;
|
||||
double **torque = atom->torque;
|
||||
int *type = atom->type;
|
||||
double *special_coul = force->special_coul;
|
||||
double *special_lj = force->special_lj;
|
||||
double qqrd2e = force->qqrd2e;
|
||||
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
qtmp = q[i];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
factor_lj = special_lj[sbmask(j)];
|
||||
factor_coul = special_coul[sbmask(j)];
|
||||
j &= NEIGHMASK;
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
jtype = type[j];
|
||||
|
||||
if (rsq < cutsq[itype][jtype]) {
|
||||
r2inv = 1.0/rsq;
|
||||
rinv = sqrt(r2inv);
|
||||
|
||||
// atom can have both a charge and dipole
|
||||
// i,j = charge-charge, dipole-dipole, dipole-charge, or charge-dipole
|
||||
|
||||
forcecoulx = forcecouly = forcecoulz = 0.0;
|
||||
tixcoul = tiycoul = tizcoul = 0.0;
|
||||
tjxcoul = tjycoul = tjzcoul = 0.0;
|
||||
|
||||
if (rsq < cut_coulsq[itype][jtype]) {
|
||||
|
||||
if (qtmp != 0.0 && q[j] != 0.0) {
|
||||
r3inv = r2inv*rinv;
|
||||
pre1 = qtmp*q[j]*r3inv;
|
||||
|
||||
forcecoulx += pre1*delx;
|
||||
forcecouly += pre1*dely;
|
||||
forcecoulz += pre1*delz;
|
||||
}
|
||||
|
||||
if (mu[i][3] > 0.0 && mu[j][3] > 0.0) {
|
||||
r3inv = r2inv*rinv;
|
||||
r5inv = r3inv*r2inv;
|
||||
r7inv = r5inv*r2inv;
|
||||
|
||||
pdotp = mu[i][0]*mu[j][0] + mu[i][1]*mu[j][1] + mu[i][2]*mu[j][2];
|
||||
pidotr = mu[i][0]*delx + mu[i][1]*dely + mu[i][2]*delz;
|
||||
pjdotr = mu[j][0]*delx + mu[j][1]*dely + mu[j][2]*delz;
|
||||
|
||||
pre1 = 3.0*r5inv*pdotp - 15.0*r7inv*pidotr*pjdotr;
|
||||
pre2 = 3.0*r5inv*pjdotr;
|
||||
pre3 = 3.0*r5inv*pidotr;
|
||||
pre4 = -1.0*r3inv;
|
||||
|
||||
forcecoulx += pre1*delx + pre2*mu[i][0] + pre3*mu[j][0];
|
||||
forcecouly += pre1*dely + pre2*mu[i][1] + pre3*mu[j][1];
|
||||
forcecoulz += pre1*delz + pre2*mu[i][2] + pre3*mu[j][2];
|
||||
|
||||
crossx = pre4 * (mu[i][1]*mu[j][2] - mu[i][2]*mu[j][1]);
|
||||
crossy = pre4 * (mu[i][2]*mu[j][0] - mu[i][0]*mu[j][2]);
|
||||
crossz = pre4 * (mu[i][0]*mu[j][1] - mu[i][1]*mu[j][0]);
|
||||
|
||||
tixcoul += crossx + pre2 * (mu[i][1]*delz - mu[i][2]*dely);
|
||||
tiycoul += crossy + pre2 * (mu[i][2]*delx - mu[i][0]*delz);
|
||||
tizcoul += crossz + pre2 * (mu[i][0]*dely - mu[i][1]*delx);
|
||||
tjxcoul += -crossx + pre3 * (mu[j][1]*delz - mu[j][2]*dely);
|
||||
tjycoul += -crossy + pre3 * (mu[j][2]*delx - mu[j][0]*delz);
|
||||
tjzcoul += -crossz + pre3 * (mu[j][0]*dely - mu[j][1]*delx);
|
||||
}
|
||||
|
||||
if (mu[i][3] > 0.0 && q[j] != 0.0) {
|
||||
r3inv = r2inv*rinv;
|
||||
r5inv = r3inv*r2inv;
|
||||
pidotr = mu[i][0]*delx + mu[i][1]*dely + mu[i][2]*delz;
|
||||
pre1 = 3.0*q[j]*r5inv * pidotr;
|
||||
pre2 = q[j]*r3inv;
|
||||
|
||||
forcecoulx += pre2*mu[i][0] - pre1*delx;
|
||||
forcecouly += pre2*mu[i][1] - pre1*dely;
|
||||
forcecoulz += pre2*mu[i][2] - pre1*delz;
|
||||
tixcoul += pre2 * (mu[i][1]*delz - mu[i][2]*dely);
|
||||
tiycoul += pre2 * (mu[i][2]*delx - mu[i][0]*delz);
|
||||
tizcoul += pre2 * (mu[i][0]*dely - mu[i][1]*delx);
|
||||
}
|
||||
|
||||
if (mu[j][3] > 0.0 && qtmp != 0.0) {
|
||||
r3inv = r2inv*rinv;
|
||||
r5inv = r3inv*r2inv;
|
||||
pjdotr = mu[j][0]*delx + mu[j][1]*dely + mu[j][2]*delz;
|
||||
pre1 = 3.0*qtmp*r5inv * pjdotr;
|
||||
pre2 = qtmp*r3inv;
|
||||
|
||||
forcecoulx += pre1*delx - pre2*mu[j][0];
|
||||
forcecouly += pre1*dely - pre2*mu[j][1];
|
||||
forcecoulz += pre1*delz - pre2*mu[j][2];
|
||||
tjxcoul += -pre2 * (mu[j][1]*delz - mu[j][2]*dely);
|
||||
tjycoul += -pre2 * (mu[j][2]*delx - mu[j][0]*delz);
|
||||
tjzcoul += -pre2 * (mu[j][0]*dely - mu[j][1]*delx);
|
||||
}
|
||||
}
|
||||
|
||||
// LJ interaction
|
||||
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
r6inv = r2inv*r2inv*r2inv;
|
||||
forcelj = r6inv * (lj1[itype][jtype]*r6inv - lj2[itype][jtype]);
|
||||
forcelj *= factor_lj * r2inv;
|
||||
} else forcelj = 0.0;
|
||||
|
||||
// total force
|
||||
|
||||
fq = factor_coul*qqrd2e;
|
||||
fx = fq*forcecoulx + delx*forcelj;
|
||||
fy = fq*forcecouly + dely*forcelj;
|
||||
fz = fq*forcecoulz + delz*forcelj;
|
||||
|
||||
// force & torque accumulation
|
||||
|
||||
f[i][0] += fx;
|
||||
f[i][1] += fy;
|
||||
f[i][2] += fz;
|
||||
torque[i][0] += fq*tixcoul;
|
||||
torque[i][1] += fq*tiycoul;
|
||||
torque[i][2] += fq*tizcoul;
|
||||
|
||||
if (eflag) {
|
||||
if (rsq < cut_coulsq[itype][jtype]) {
|
||||
ecoul = qtmp*q[j]*rinv;
|
||||
if (mu[i][3] > 0.0 && mu[j][3] > 0.0)
|
||||
ecoul += r3inv*pdotp - 3.0*r5inv*pidotr*pjdotr;
|
||||
if (mu[i][3] > 0.0 && q[j] != 0.0)
|
||||
ecoul += -q[j]*r3inv*pidotr;
|
||||
if (mu[j][3] > 0.0 && qtmp != 0.0)
|
||||
ecoul += qtmp*r3inv*pjdotr;
|
||||
ecoul *= factor_coul*qqrd2e;
|
||||
} else ecoul = 0.0;
|
||||
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
evdwl = r6inv*(lj3[itype][jtype]*r6inv-lj4[itype][jtype]) -
|
||||
offset[itype][jtype];
|
||||
evdwl *= factor_lj;
|
||||
} else evdwl = 0.0;
|
||||
}
|
||||
|
||||
if (evflag) ev_tally_xyz_full(i,evdwl,ecoul,fx,fy,fz,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
63
src/GPU/pair_dipole_cut_gpu.h
Normal file
63
src/GPU/pair_dipole_cut_gpu.h
Normal file
@ -0,0 +1,63 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(dipole/cut/gpu,PairDipoleCutGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_DIPOLE_CUT_GPU_H
|
||||
#define LMP_PAIR_DIPOLE_CUT_GPU_H
|
||||
|
||||
#include "pair_dipole_cut.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairDipoleCutGPU : public PairDipoleCut {
|
||||
public:
|
||||
PairDipoleCutGPU(LAMMPS *lmp);
|
||||
~PairDipoleCutGPU();
|
||||
void cpu_compute(int, int, int, int, int *, int *, int **);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
int *gpulist;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/* ERROR/WARNING messages:
|
||||
|
||||
E: Insufficient memory on accelerator
|
||||
|
||||
There is insufficient memory on one of the devices specified for the gpu
|
||||
package
|
||||
|
||||
E: Pair style dipole/cut/gpu requires atom attribute q
|
||||
|
||||
The atom style defined does not have this attribute.
|
||||
|
||||
E: Cannot use newton pair with dipole/cut/gpu pair style
|
||||
|
||||
Self-explanatory.
|
||||
|
||||
*/
|
||||
395
src/GPU/pair_dipole_sf_gpu.cpp
Normal file
395
src/GPU/pair_dipole_sf_gpu.cpp
Normal file
@ -0,0 +1,395 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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 author: Trung Dac Nguyen (ORNL)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include "lmptype.h"
|
||||
#include "math.h"
|
||||
#include "stdio.h"
|
||||
#include "stdlib.h"
|
||||
#include "pair_dipole_sf_gpu.h"
|
||||
#include "atom.h"
|
||||
#include "atom_vec.h"
|
||||
#include "comm.h"
|
||||
#include "force.h"
|
||||
#include "neighbor.h"
|
||||
#include "neigh_list.h"
|
||||
#include "integrate.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "string.h"
|
||||
#include "gpu_extra.h"
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
int dplsf_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
|
||||
double **host_lj2, double **host_lj3, double **host_lj4,
|
||||
double *special_lj, const int nlocal,
|
||||
const int nall, const int max_nbors, const int maxspecial,
|
||||
const double cell_size, int &gpu_mode, FILE *screen,
|
||||
double **host_cut_ljsq, double **host_cut_coulsq,
|
||||
double *host_special_coul, const double qqrd2e);
|
||||
void dplsf_gpu_clear();
|
||||
int ** dplsf_gpu_compute_n(const int ago, const int inum,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
double *sublo, double *subhi, int *tag, int **nspecial,
|
||||
int **special, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
int **ilist, int **jnum, const double cpu_time,
|
||||
bool &success, double *host_q, double **host_mu,
|
||||
double *boxlo, double *prd);
|
||||
void dplsf_gpu_compute(const int ago, const int inum,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
int *ilist, int *numj, int **firstneigh,
|
||||
const bool eflag, const bool vflag, const bool eatom,
|
||||
const bool vatom, int &host_start, const double cpu_time,
|
||||
bool &success, double *host_q, double **host_mu, const int nlocal,
|
||||
double *boxlo, double *prd);
|
||||
double dplsf_gpu_bytes();
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairDipoleSFGPU::PairDipoleSFGPU(LAMMPS *lmp) : PairDipoleSF(lmp),
|
||||
gpu_mode(GPU_FORCE)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairDipoleSFGPU::~PairDipoleSFGPU()
|
||||
{
|
||||
dplsf_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairDipoleSFGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
int *ilist, *numneigh, **firstneigh;
|
||||
if (gpu_mode != GPU_FORCE) {
|
||||
inum = atom->nlocal;
|
||||
firstneigh = dplsf_gpu_compute_n(neighbor->ago, inum, nall, atom->x,
|
||||
atom->type, domain->sublo, domain->subhi,
|
||||
atom->tag, atom->nspecial, atom->special,
|
||||
eflag, vflag, eflag_atom, vflag_atom,
|
||||
host_start, &ilist, &numneigh, cpu_time,
|
||||
success, atom->q, atom->mu, domain->boxlo,
|
||||
domain->prd);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
dplsf_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
|
||||
ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, cpu_time, success, atom->q,
|
||||
atom->mu, atom->nlocal, domain->boxlo, domain->prd);
|
||||
}
|
||||
if (!success)
|
||||
error->one(FLERR,"Insufficient memory on accelerator");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairDipoleSFGPU::init_style()
|
||||
{
|
||||
if (!atom->q_flag || !atom->mu_flag || !atom->torque_flag)
|
||||
error->all(FLERR,"Pair dipole/sf/gpu requires atom attributes q, mu, torque");
|
||||
|
||||
if (force->newton_pair)
|
||||
error->all(FLERR,"Cannot use newton pair with dipole/sf/gpu pair style");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double maxcut = -1.0;
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
|
||||
cut = init_one(i,j);
|
||||
cut *= cut;
|
||||
if (cut > maxcut)
|
||||
maxcut = cut;
|
||||
cutsq[i][j] = cutsq[j][i] = cut;
|
||||
} else
|
||||
cutsq[i][j] = cutsq[j][i] = 0.0;
|
||||
}
|
||||
}
|
||||
double cell_size = sqrt(maxcut) + neighbor->skin;
|
||||
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
int success = dplsf_gpu_init(atom->ntypes+1, cutsq, lj1, lj2, lj3, lj4,
|
||||
force->special_lj, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen, cut_ljsq, cut_coulsq,
|
||||
force->special_coul, force->qqrd2e);
|
||||
GPU_EXTRA::check_flag(success,error,world);
|
||||
|
||||
if (gpu_mode == GPU_FORCE) {
|
||||
int irequest = neighbor->request(this);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairDipoleSFGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + dplsf_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairDipoleSFGPU::cpu_compute(int start, int inum, int eflag, int vflag,
|
||||
int *ilist, int *numneigh,
|
||||
int **firstneigh)
|
||||
{
|
||||
int i,j,ii,jj,jnum,itype,jtype;
|
||||
double qtmp,xtmp,ytmp,ztmp,delx,dely,delz,evdwl,ecoul,fx,fy,fz;
|
||||
double rsq,rinv,r2inv,r6inv,r3inv,r5inv;
|
||||
double forcecoulx,forcecouly,forcecoulz,crossx,crossy,crossz;
|
||||
double tixcoul,tiycoul,tizcoul,tjxcoul,tjycoul,tjzcoul;
|
||||
double fq,pdotp,pidotr,pjdotr,pre1,pre2,pre3,pre4;
|
||||
double forcelj,factor_coul,factor_lj;
|
||||
double presf,afac,bfac,pqfac,qpfac,forceljcut,forceljsf;
|
||||
double aforcecoulx,aforcecouly,aforcecoulz;
|
||||
double bforcecoulx,bforcecouly,bforcecoulz;
|
||||
double rcutlj2inv, rcutcoul2inv,rcutlj6inv;
|
||||
int *jlist;
|
||||
|
||||
evdwl = ecoul = 0.0;
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
double *q = atom->q;
|
||||
double **mu = atom->mu;
|
||||
double **torque = atom->torque;
|
||||
int *type = atom->type;
|
||||
double *special_coul = force->special_coul;
|
||||
double *special_lj = force->special_lj;
|
||||
double qqrd2e = force->qqrd2e;
|
||||
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
qtmp = q[i];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
factor_lj = special_lj[sbmask(j)];
|
||||
factor_coul = special_coul[sbmask(j)];
|
||||
j &= NEIGHMASK;
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
jtype = type[j];
|
||||
|
||||
if (rsq < cutsq[itype][jtype]) {
|
||||
r2inv = 1.0/rsq;
|
||||
rinv = sqrt(r2inv);
|
||||
|
||||
// atom can have both a charge and dipole
|
||||
// i,j = charge-charge, dipole-dipole, dipole-charge, or charge-dipole
|
||||
|
||||
forcecoulx = forcecouly = forcecoulz = 0.0;
|
||||
tixcoul = tiycoul = tizcoul = 0.0;
|
||||
tjxcoul = tjycoul = tjzcoul = 0.0;
|
||||
|
||||
if (rsq < cut_coulsq[itype][jtype]) {
|
||||
|
||||
if (qtmp != 0.0 && q[j] != 0.0) {
|
||||
pre1 = qtmp*q[j]*rinv*(r2inv-1.0/cut_coulsq[itype][jtype]);
|
||||
|
||||
forcecoulx += pre1*delx;
|
||||
forcecouly += pre1*dely;
|
||||
forcecoulz += pre1*delz;
|
||||
}
|
||||
|
||||
if (mu[i][3] > 0.0 && mu[j][3] > 0.0) {
|
||||
r3inv = r2inv*rinv;
|
||||
r5inv = r3inv*r2inv;
|
||||
rcutcoul2inv=1.0/cut_coulsq[itype][jtype];
|
||||
|
||||
pdotp = mu[i][0]*mu[j][0] + mu[i][1]*mu[j][1] + mu[i][2]*mu[j][2];
|
||||
pidotr = mu[i][0]*delx + mu[i][1]*dely + mu[i][2]*delz;
|
||||
pjdotr = mu[j][0]*delx + mu[j][1]*dely + mu[j][2]*delz;
|
||||
|
||||
afac = 1.0 - rsq*rsq * rcutcoul2inv*rcutcoul2inv;
|
||||
pre1 = afac * ( pdotp - 3.0 * r2inv * pidotr * pjdotr );
|
||||
aforcecoulx = pre1*delx;
|
||||
aforcecouly = pre1*dely;
|
||||
aforcecoulz = pre1*delz;
|
||||
|
||||
bfac = 1.0 - 4.0*rsq*sqrt(rsq)*rcutcoul2inv*sqrt(rcutcoul2inv) +
|
||||
3.0*rsq*rsq*rcutcoul2inv*rcutcoul2inv;
|
||||
presf = 2.0 * r2inv * pidotr * pjdotr;
|
||||
bforcecoulx = bfac * (pjdotr*mu[i][0]+pidotr*mu[j][0]-presf*delx);
|
||||
bforcecouly = bfac * (pjdotr*mu[i][1]+pidotr*mu[j][1]-presf*dely);
|
||||
bforcecoulz = bfac * (pjdotr*mu[i][2]+pidotr*mu[j][2]-presf*delz);
|
||||
|
||||
forcecoulx += 3.0 * r5inv * ( aforcecoulx + bforcecoulx );
|
||||
forcecouly += 3.0 * r5inv * ( aforcecouly + bforcecouly );
|
||||
forcecoulz += 3.0 * r5inv * ( aforcecoulz + bforcecoulz );
|
||||
|
||||
pre2 = 3.0 * bfac * r5inv * pjdotr;
|
||||
pre3 = 3.0 * bfac * r5inv * pidotr;
|
||||
pre4 = -bfac * r3inv;
|
||||
|
||||
crossx = pre4 * (mu[i][1]*mu[j][2] - mu[i][2]*mu[j][1]);
|
||||
crossy = pre4 * (mu[i][2]*mu[j][0] - mu[i][0]*mu[j][2]);
|
||||
crossz = pre4 * (mu[i][0]*mu[j][1] - mu[i][1]*mu[j][0]);
|
||||
|
||||
tixcoul += crossx + pre2 * (mu[i][1]*delz - mu[i][2]*dely);
|
||||
tiycoul += crossy + pre2 * (mu[i][2]*delx - mu[i][0]*delz);
|
||||
tizcoul += crossz + pre2 * (mu[i][0]*dely - mu[i][1]*delx);
|
||||
tjxcoul += -crossx + pre3 * (mu[j][1]*delz - mu[j][2]*dely);
|
||||
tjycoul += -crossy + pre3 * (mu[j][2]*delx - mu[j][0]*delz);
|
||||
tjzcoul += -crossz + pre3 * (mu[j][0]*dely - mu[j][1]*delx);
|
||||
}
|
||||
|
||||
if (mu[i][3] > 0.0 && q[j] != 0.0) {
|
||||
r3inv = r2inv*rinv;
|
||||
r5inv = r3inv*r2inv;
|
||||
rcutcoul2inv=1.0/cut_coulsq[itype][jtype];
|
||||
pre1 = 3.0 * q[j] * r5inv * pidotr * (1-rsq*rcutcoul2inv);
|
||||
pqfac = 1.0 - 3.0*rsq*rcutcoul2inv +
|
||||
2.0*rsq*sqrt(rsq)*rcutcoul2inv*sqrt(rcutcoul2inv);
|
||||
pre2 = q[j] * r3inv * pqfac;
|
||||
|
||||
forcecoulx += pre2*mu[i][0] - pre1*delx;
|
||||
forcecouly += pre2*mu[i][1] - pre1*dely;
|
||||
forcecoulz += pre2*mu[i][2] - pre1*delz;
|
||||
tixcoul += pre2 * (mu[i][1]*delz - mu[i][2]*dely);
|
||||
tiycoul += pre2 * (mu[i][2]*delx - mu[i][0]*delz);
|
||||
tizcoul += pre2 * (mu[i][0]*dely - mu[i][1]*delx);
|
||||
}
|
||||
|
||||
if (mu[j][3] > 0.0 && qtmp != 0.0) {
|
||||
r3inv = r2inv*rinv;
|
||||
r5inv = r3inv*r2inv;
|
||||
rcutcoul2inv=1.0/cut_coulsq[itype][jtype];
|
||||
pre1 = 3.0 * qtmp * r5inv * pjdotr * (1-rsq*rcutcoul2inv);
|
||||
qpfac = 1.0 - 3.0*rsq*rcutcoul2inv +
|
||||
2.0*rsq*sqrt(rsq)*rcutcoul2inv*sqrt(rcutcoul2inv);
|
||||
pre2 = qtmp * r3inv * qpfac;
|
||||
|
||||
forcecoulx += pre1*delx - pre2*mu[j][0];
|
||||
forcecouly += pre1*dely - pre2*mu[j][1];
|
||||
forcecoulz += pre1*delz - pre2*mu[j][2];
|
||||
tjxcoul += -pre2 * (mu[j][1]*delz - mu[j][2]*dely);
|
||||
tjycoul += -pre2 * (mu[j][2]*delx - mu[j][0]*delz);
|
||||
tjzcoul += -pre2 * (mu[j][0]*dely - mu[j][1]*delx);
|
||||
}
|
||||
}
|
||||
|
||||
// LJ interaction
|
||||
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
r6inv = r2inv*r2inv*r2inv;
|
||||
forceljcut = r6inv*(lj1[itype][jtype]*r6inv-lj2[itype][jtype])*r2inv;
|
||||
|
||||
rcutlj2inv = 1.0 / cut_ljsq[itype][jtype];
|
||||
rcutlj6inv = rcutlj2inv * rcutlj2inv * rcutlj2inv;
|
||||
forceljsf = (lj1[itype][jtype]*rcutlj6inv - lj2[itype][jtype]) *
|
||||
rcutlj6inv * rcutlj2inv;
|
||||
|
||||
forcelj = factor_lj * (forceljcut - forceljsf);
|
||||
} else forcelj = 0.0;
|
||||
|
||||
// total force
|
||||
|
||||
fq = factor_coul*qqrd2e;
|
||||
fx = fq*forcecoulx + delx*forcelj;
|
||||
fy = fq*forcecouly + dely*forcelj;
|
||||
fz = fq*forcecoulz + delz*forcelj;
|
||||
|
||||
// force & torque accumulation
|
||||
|
||||
f[i][0] += fx;
|
||||
f[i][1] += fy;
|
||||
f[i][2] += fz;
|
||||
torque[i][0] += fq*tixcoul;
|
||||
torque[i][1] += fq*tiycoul;
|
||||
torque[i][2] += fq*tizcoul;
|
||||
|
||||
if (eflag) {
|
||||
if (rsq < cut_coulsq[itype][jtype]) {
|
||||
ecoul = qtmp*q[j]*rinv*
|
||||
pow((1.0-sqrt(rsq)/sqrt(cut_coulsq[itype][jtype])),2);
|
||||
if (mu[i][3] > 0.0 && mu[j][3] > 0.0)
|
||||
ecoul += bfac * (r3inv*pdotp - 3.0*r5inv*pidotr*pjdotr);
|
||||
if (mu[i][3] > 0.0 && q[j] != 0.0)
|
||||
ecoul += -q[j]*r3inv * pqfac * pidotr;
|
||||
if (mu[j][3] > 0.0 && qtmp != 0.0)
|
||||
ecoul += qtmp*r3inv * qpfac * pjdotr;
|
||||
ecoul *= factor_coul*qqrd2e;
|
||||
} else ecoul = 0.0;
|
||||
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
evdwl = r6inv*(lj3[itype][jtype]*r6inv-lj4[itype][jtype]) +
|
||||
rcutlj6inv*(6*lj3[itype][jtype]*rcutlj6inv-3*lj4[itype][jtype])*
|
||||
rsq*rcutlj2inv +
|
||||
rcutlj6inv*(-7*lj3[itype][jtype]*rcutlj6inv+4*lj4[itype][jtype]);
|
||||
evdwl *= factor_lj;
|
||||
} else evdwl = 0.0;
|
||||
}
|
||||
|
||||
if (evflag) ev_tally_xyz_full(i,evdwl,ecoul,
|
||||
fx,fy,fz,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
63
src/GPU/pair_dipole_sf_gpu.h
Normal file
63
src/GPU/pair_dipole_sf_gpu.h
Normal file
@ -0,0 +1,63 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(dipole/sf/gpu,PairDipoleSFGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_DIPOLE_SF_GPU_H
|
||||
#define LMP_PAIR_DIPOLE_SF_GPU_H
|
||||
|
||||
#include "pair_dipole_sf.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairDipoleSFGPU : public PairDipoleSF {
|
||||
public:
|
||||
PairDipoleSFGPU(LAMMPS *lmp);
|
||||
~PairDipoleSFGPU();
|
||||
void cpu_compute(int, int, int, int, int *, int *, int **);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
int *gpulist;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/* ERROR/WARNING messages:
|
||||
|
||||
E: Insufficient memory on accelerator
|
||||
|
||||
There is insufficient memory on one of the devices specified for the gpu
|
||||
package
|
||||
|
||||
E: Pair style dipole/cut/gpu requires atom attribute q
|
||||
|
||||
The atom style defined does not have this attribute.
|
||||
|
||||
E: Cannot use newton pair with dipole/cut/gpu pair style
|
||||
|
||||
Self-explanatory.
|
||||
|
||||
*/
|
||||
226
src/GPU/pair_gauss_gpu.cpp
Normal file
226
src/GPU/pair_gauss_gpu.cpp
Normal file
@ -0,0 +1,226 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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 author: Trung Dac Nguyen (ORNL)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include "lmptype.h"
|
||||
#include "math.h"
|
||||
#include "stdio.h"
|
||||
#include "stdlib.h"
|
||||
#include "pair_gauss_gpu.h"
|
||||
#include "atom.h"
|
||||
#include "atom_vec.h"
|
||||
#include "comm.h"
|
||||
#include "force.h"
|
||||
#include "neighbor.h"
|
||||
#include "neigh_list.h"
|
||||
#include "integrate.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "string.h"
|
||||
#include "gpu_extra.h"
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
int gauss_gpu_init(const int ntypes, double **cutsq, double **host_a,
|
||||
double **b, double **offset, double *special_lj, const int nlocal,
|
||||
const int nall, const int max_nbors, const int maxspecial,
|
||||
const double cell_size, int &gpu_mode, FILE *screen);
|
||||
void gauss_gpu_clear();
|
||||
int ** gauss_gpu_compute_n(const int ago, const int inum,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
double *sublo, double *subhi, int *tag, int **nspecial,
|
||||
int **special, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
int **ilist, int **jnum,
|
||||
const double cpu_time, bool &success);
|
||||
void gauss_gpu_compute(const int ago, const int inum, const int nall,
|
||||
double **host_x, int *host_type, int *ilist, int *numj,
|
||||
int **firstneigh, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
const double cpu_time, bool &success);
|
||||
double gauss_gpu_bytes();
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairGaussGPU::PairGaussGPU(LAMMPS *lmp) : PairGauss(lmp), gpu_mode(GPU_FORCE)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairGaussGPU::~PairGaussGPU()
|
||||
{
|
||||
gauss_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairGaussGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
int *ilist, *numneigh, **firstneigh;
|
||||
if (gpu_mode != GPU_FORCE) {
|
||||
inum = atom->nlocal;
|
||||
firstneigh = gauss_gpu_compute_n(neighbor->ago, inum, nall,
|
||||
atom->x, atom->type, domain->sublo,
|
||||
domain->subhi, atom->tag, atom->nspecial,
|
||||
atom->special, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start,
|
||||
&ilist, &numneigh, cpu_time, success);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
gauss_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
|
||||
ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, cpu_time, success);
|
||||
}
|
||||
if (!success)
|
||||
error->one(FLERR,"Insufficient memory on accelerator");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairGaussGPU::init_style()
|
||||
{
|
||||
if (force->newton_pair)
|
||||
error->all(FLERR,"Cannot use newton pair with lj/cut/gpu pair style");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double maxcut = -1.0;
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
|
||||
cut = init_one(i,j);
|
||||
cut *= cut;
|
||||
if (cut > maxcut)
|
||||
maxcut = cut;
|
||||
cutsq[i][j] = cutsq[j][i] = cut;
|
||||
} else
|
||||
cutsq[i][j] = cutsq[j][i] = 0.0;
|
||||
}
|
||||
}
|
||||
double cell_size = sqrt(maxcut) + neighbor->skin;
|
||||
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
int success = gauss_gpu_init(atom->ntypes+1, cutsq, a, b,
|
||||
offset, force->special_lj, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen);
|
||||
GPU_EXTRA::check_flag(success,error,world);
|
||||
|
||||
if (gpu_mode == GPU_FORCE) {
|
||||
int irequest = neighbor->request(this);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairGaussGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + gauss_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairGaussGPU::cpu_compute(int start, int inum, int eflag, int vflag,
|
||||
int *ilist, int *numneigh, int **firstneigh) {
|
||||
int i,j,ii,jj,jnum,itype,jtype;
|
||||
double xtmp,ytmp,ztmp,delx,dely,delz,evdwl,fpair;
|
||||
double r,rsq,r2inv,forcelj,factor_lj;
|
||||
int *jlist;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
int *type = atom->type;
|
||||
double *special_lj = force->special_lj;
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
factor_lj = special_lj[sbmask(j)];
|
||||
j &= NEIGHMASK;
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
jtype = type[j];
|
||||
|
||||
if (rsq < cutsq[itype][jtype]) {
|
||||
r2inv = 1.0/rsq;
|
||||
r = sqrt(rsq);
|
||||
forcelj = - 2.0*a[itype][jtype]*b[itype][jtype] * rsq *
|
||||
exp(-b[itype][jtype]*rsq);
|
||||
fpair = factor_lj*forcelj*r2inv;
|
||||
|
||||
f[i][0] += delx*fpair;
|
||||
f[i][1] += dely*fpair;
|
||||
f[i][2] += delz*fpair;
|
||||
|
||||
if (eflag) {
|
||||
evdwl = -(a[itype][jtype]*exp(-b[itype][jtype]*rsq) -
|
||||
offset[itype][jtype]);
|
||||
evdwl *= factor_lj;
|
||||
}
|
||||
|
||||
if (evflag) ev_tally_full(i,evdwl,0.0,fpair,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
59
src/GPU/pair_gauss_gpu.h
Normal file
59
src/GPU/pair_gauss_gpu.h
Normal file
@ -0,0 +1,59 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(gauss/gpu,PairGaussGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_GAUSS_GPU_H
|
||||
#define LMP_PAIR_GAUSS_GPU_H
|
||||
|
||||
#include "pair_gauss.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairGaussGPU : public PairGauss {
|
||||
public:
|
||||
PairGaussGPU(LAMMPS *lmp);
|
||||
~PairGaussGPU();
|
||||
void cpu_compute(int, int, int, int, int *, int *, int **);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
int *gpulist;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/* ERROR/WARNING messages:
|
||||
|
||||
E: Insufficient memory on accelerator
|
||||
|
||||
There is insufficient memory on one of the devices specified for the gpu
|
||||
package
|
||||
|
||||
E: Cannot use newton pair with gauss/gpu pair style
|
||||
|
||||
Self-explanatory.
|
||||
|
||||
*/
|
||||
266
src/GPU/pair_lj_cut_coul_debye_gpu.cpp
Normal file
266
src/GPU/pair_lj_cut_coul_debye_gpu.cpp
Normal file
@ -0,0 +1,266 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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 author: Trung Dac Nguyen (ORNL)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include "lmptype.h"
|
||||
#include "math.h"
|
||||
#include "stdio.h"
|
||||
#include "stdlib.h"
|
||||
#include "pair_lj_cut_coul_debye_gpu.h"
|
||||
#include "atom.h"
|
||||
#include "atom_vec.h"
|
||||
#include "comm.h"
|
||||
#include "force.h"
|
||||
#include "neighbor.h"
|
||||
#include "neigh_list.h"
|
||||
#include "integrate.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "string.h"
|
||||
#include "gpu_extra.h"
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
int ljcd_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
|
||||
double **host_lj2, double **host_lj3, double **host_lj4,
|
||||
double **offset, double *special_lj, const int nlocal,
|
||||
const int nall, const int max_nbors, const int maxspecial,
|
||||
const double cell_size, int &gpu_mode, FILE *screen,
|
||||
double **host_cut_ljsq, double **host_cut_coulsq,
|
||||
double *host_special_coul, const double qqrd2e,
|
||||
const double kappa);
|
||||
void ljcd_gpu_clear();
|
||||
int ** ljcd_gpu_compute_n(const int ago, const int inum, const int nall,
|
||||
double **host_x, int *host_type,
|
||||
double *sublo, double *subhi, int *tag, int **nspecial,
|
||||
int **special, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
int **ilist, int **jnum, const double cpu_time,
|
||||
bool &success, double *host_q, double *boxlo,
|
||||
double *prd);
|
||||
void ljcd_gpu_compute(const int ago, const int inum, const int nall,
|
||||
double **host_x, int *host_type,
|
||||
int *ilist, int *numj, int **firstneigh,
|
||||
const bool eflag, const bool vflag, const bool eatom,
|
||||
const bool vatom, int &host_start, const double cpu_time,
|
||||
bool &success, double *host_q, const int nlocal,
|
||||
double *boxlo, double *prd);
|
||||
double ljcd_gpu_bytes();
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairLJCutCoulDebyeGPU::PairLJCutCoulDebyeGPU(LAMMPS *lmp) :
|
||||
PairLJCutCoulDebye(lmp), gpu_mode(GPU_FORCE)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairLJCutCoulDebyeGPU::~PairLJCutCoulDebyeGPU()
|
||||
{
|
||||
ljcd_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairLJCutCoulDebyeGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
int *ilist, *numneigh, **firstneigh;
|
||||
if (gpu_mode != GPU_FORCE) {
|
||||
inum = atom->nlocal;
|
||||
firstneigh = ljcd_gpu_compute_n(neighbor->ago, inum, nall, atom->x,
|
||||
atom->type, domain->sublo, domain->subhi,
|
||||
atom->tag, atom->nspecial, atom->special,
|
||||
eflag, vflag, eflag_atom, vflag_atom,
|
||||
host_start, &ilist, &numneigh, cpu_time,
|
||||
success, atom->q, domain->boxlo,
|
||||
domain->prd);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
ljcd_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
|
||||
ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, cpu_time, success, atom->q,
|
||||
atom->nlocal, domain->boxlo, domain->prd);
|
||||
}
|
||||
if (!success)
|
||||
error->one(FLERR,"Insufficient memory on accelerator");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairLJCutCoulDebyeGPU::init_style()
|
||||
{
|
||||
if (!atom->q_flag)
|
||||
error->all(FLERR,"Pair style lj/cut/coul/debye/gpu requires atom attribute q");
|
||||
|
||||
if (force->newton_pair)
|
||||
error->all(FLERR,"Cannot use newton pair with lj/cut/coul/debye/gpu pair style");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double maxcut = -1.0;
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
|
||||
cut = init_one(i,j);
|
||||
cut *= cut;
|
||||
if (cut > maxcut)
|
||||
maxcut = cut;
|
||||
cutsq[i][j] = cutsq[j][i] = cut;
|
||||
} else
|
||||
cutsq[i][j] = cutsq[j][i] = 0.0;
|
||||
}
|
||||
}
|
||||
double cell_size = sqrt(maxcut) + neighbor->skin;
|
||||
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
int success = ljcd_gpu_init(atom->ntypes+1, cutsq, lj1, lj2, lj3, lj4,
|
||||
offset, force->special_lj, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen, cut_ljsq,
|
||||
cut_coulsq, force->special_coul,
|
||||
force->qqrd2e, kappa);
|
||||
GPU_EXTRA::check_flag(success,error,world);
|
||||
|
||||
if (gpu_mode == GPU_FORCE) {
|
||||
int irequest = neighbor->request(this);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairLJCutCoulDebyeGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + ljcd_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairLJCutCoulDebyeGPU::cpu_compute(int start, int inum, int eflag,
|
||||
int vflag, int *ilist,
|
||||
int *numneigh, int **firstneigh)
|
||||
{
|
||||
int i,j,ii,jj,jnum,itype,jtype;
|
||||
double qtmp,xtmp,ytmp,ztmp,delx,dely,delz,evdwl,ecoul,fpair;
|
||||
double rsq,r2inv,r6inv,forcecoul,forcelj,factor_coul,factor_lj;
|
||||
double r,rinv,screening;
|
||||
int *jlist;
|
||||
|
||||
evdwl = ecoul = 0.0;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
double *q = atom->q;
|
||||
int *type = atom->type;
|
||||
double *special_coul = force->special_coul;
|
||||
double *special_lj = force->special_lj;
|
||||
double qqrd2e = force->qqrd2e;
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
qtmp = q[i];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
factor_lj = special_lj[sbmask(j)];
|
||||
factor_coul = special_coul[sbmask(j)];
|
||||
j &= NEIGHMASK;
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
jtype = type[j];
|
||||
|
||||
if (rsq < cutsq[itype][jtype]) {
|
||||
r2inv = 1.0/rsq;
|
||||
|
||||
if (rsq < cut_coulsq[itype][jtype]) {
|
||||
r = sqrt(rsq);
|
||||
rinv = 1.0/r;
|
||||
screening = exp(-kappa*r);
|
||||
forcecoul = qqrd2e * qtmp*q[j] * screening * (kappa + rinv);
|
||||
} else forcecoul = 0.0;
|
||||
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
r6inv = r2inv*r2inv*r2inv;
|
||||
forcelj = r6inv * (lj1[itype][jtype]*r6inv - lj2[itype][jtype]);
|
||||
} else forcelj = 0.0;
|
||||
|
||||
fpair = (factor_coul*forcecoul + factor_lj*forcelj) * r2inv;
|
||||
|
||||
f[i][0] += delx*fpair;
|
||||
f[i][1] += dely*fpair;
|
||||
f[i][2] += delz*fpair;
|
||||
|
||||
if (eflag) {
|
||||
if (rsq < cut_coulsq[itype][jtype])
|
||||
ecoul = factor_coul * qqrd2e * qtmp*q[j] * rinv * screening;
|
||||
else ecoul = 0.0;
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
evdwl = r6inv*(lj3[itype][jtype]*r6inv-lj4[itype][jtype]) -
|
||||
offset[itype][jtype];
|
||||
evdwl *= factor_lj;
|
||||
} else evdwl = 0.0;
|
||||
}
|
||||
|
||||
if (evflag) ev_tally_full(i,evdwl,ecoul,fpair,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
63
src/GPU/pair_lj_cut_coul_debye_gpu.h
Normal file
63
src/GPU/pair_lj_cut_coul_debye_gpu.h
Normal file
@ -0,0 +1,63 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(lj/cut/coul/debye/gpu,PairLJCutCoulDebyeGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_LJ_CUT_COUL_DEBYE_GPU_H
|
||||
#define LMP_PAIR_LJ_CUT_COUL_DEBYE_GPU_H
|
||||
|
||||
#include "pair_lj_cut_coul_debye.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairLJCutCoulDebyeGPU : public PairLJCutCoulDebye {
|
||||
public:
|
||||
PairLJCutCoulDebyeGPU(LAMMPS *lmp);
|
||||
~PairLJCutCoulDebyeGPU();
|
||||
void cpu_compute(int, int, int, int, int *, int *, int **);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
int *gpulist;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/* ERROR/WARNING messages:
|
||||
|
||||
E: Insufficient memory on accelerator
|
||||
|
||||
There is insufficient memory on one of the devices specified for the gpu
|
||||
package
|
||||
|
||||
E: Pair style lj/cut/coul/debye/gpu requires atom attribute q
|
||||
|
||||
The atom style defined does not have this attribute.
|
||||
|
||||
E: Cannot use newton pair with lj/cut/coul/debye/gpu pair style
|
||||
|
||||
Self-explanatory.
|
||||
|
||||
*/
|
||||
241
src/GPU/pair_yukawa_colloid_gpu.cpp
Normal file
241
src/GPU/pair_yukawa_colloid_gpu.cpp
Normal file
@ -0,0 +1,241 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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 author: Trung Dac Nguyen (ORNL)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include "lmptype.h"
|
||||
#include "math.h"
|
||||
#include "stdio.h"
|
||||
#include "stdlib.h"
|
||||
#include "pair_yukawa_colloid_gpu.h"
|
||||
#include "atom.h"
|
||||
#include "atom_vec.h"
|
||||
#include "comm.h"
|
||||
#include "force.h"
|
||||
#include "neighbor.h"
|
||||
#include "neigh_list.h"
|
||||
#include "integrate.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "string.h"
|
||||
#include "gpu_extra.h"
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
int ykcolloid_gpu_init(const int ntypes, double **cutsq, double **host_a,
|
||||
double **host_offset, double *special_lj, const int inum,
|
||||
const int nall, const int max_nbors, const int maxspecial,
|
||||
const double cell_size, int &gpu_mode, FILE *screen,
|
||||
const double kappa);
|
||||
void ykcolloid_gpu_clear();
|
||||
int ** ykcolloid_gpu_compute_n(const int ago, const int inum_full,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
double *sublo, double *subhi, int *tag, int **nspecial,
|
||||
int **special, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
int **ilist, int **jnum, const double cpu_time,
|
||||
bool &success, double *host_rad);
|
||||
void ykcolloid_gpu_compute(const int ago, const int inum_full,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
int *ilist, int *numj, int **firstneigh,
|
||||
const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
const double cpu_time, bool &success, double *host_rad);
|
||||
double ykcolloid_gpu_bytes();
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairYukawaColloidGPU::PairYukawaColloidGPU(LAMMPS *lmp) : PairYukawaColloid(lmp),
|
||||
gpu_mode(GPU_FORCE)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairYukawaColloidGPU::~PairYukawaColloidGPU()
|
||||
{
|
||||
ykcolloid_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairYukawaColloidGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
int *ilist, *numneigh, **firstneigh;
|
||||
if (gpu_mode != GPU_FORCE) {
|
||||
inum = atom->nlocal;
|
||||
firstneigh = ykcolloid_gpu_compute_n(neighbor->ago, inum, nall,
|
||||
atom->x, atom->type,
|
||||
domain->sublo,
|
||||
domain->subhi, atom->tag,
|
||||
atom->nspecial, atom->special,
|
||||
eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, &ilist,
|
||||
&numneigh, cpu_time,
|
||||
success, atom->radius);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
ykcolloid_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
|
||||
ilist, numneigh, firstneigh, eflag, vflag,
|
||||
eflag_atom, vflag_atom, host_start, cpu_time,
|
||||
success, atom->radius);
|
||||
}
|
||||
if (!success)
|
||||
error->one(FLERR,"Insufficient memory on accelerator");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairYukawaColloidGPU::init_style()
|
||||
{
|
||||
if (!atom->sphere_flag)
|
||||
error->all(FLERR,"Pair yukawa/colloid/gpu requires atom style sphere");
|
||||
|
||||
if (force->newton_pair)
|
||||
error->all(FLERR,"Cannot use newton pair with yukawa/colloid/gpu pair style");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double maxcut = -1.0;
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
|
||||
cut = init_one(i,j);
|
||||
cut *= cut;
|
||||
if (cut > maxcut)
|
||||
maxcut = cut;
|
||||
cutsq[i][j] = cutsq[j][i] = cut;
|
||||
} else
|
||||
cutsq[i][j] = cutsq[j][i] = 0.0;
|
||||
}
|
||||
}
|
||||
double cell_size = sqrt(maxcut) + neighbor->skin;
|
||||
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
int success = ykcolloid_gpu_init(atom->ntypes+1, cutsq, a,
|
||||
offset, force->special_lj, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen, kappa);
|
||||
GPU_EXTRA::check_flag(success,error,world);
|
||||
|
||||
if (gpu_mode == GPU_FORCE) {
|
||||
int irequest = neighbor->request(this);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairYukawaColloidGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + ykcolloid_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairYukawaColloidGPU::cpu_compute(int start, int inum, int eflag,
|
||||
int vflag, int *ilist, int *numneigh,
|
||||
int **firstneigh) {
|
||||
int i,j,ii,jj,jnum,itype,jtype;
|
||||
double xtmp,ytmp,ztmp,delx,dely,delz,evdwl,fpair,radi,radj;
|
||||
double r,rsq,r2inv,rinv,screening,forceyukawa,factor;
|
||||
int *jlist;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
int *type = atom->type;
|
||||
double *radius = atom->radius;
|
||||
double *special_lj = force->special_lj;
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
radi = radius[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
factor = special_lj[sbmask(j)];
|
||||
j &= NEIGHMASK;
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
jtype = type[j];
|
||||
radj = radius[j];
|
||||
|
||||
if (rsq < cutsq[itype][jtype]) {
|
||||
r2inv = 1.0/rsq;
|
||||
r = sqrt(rsq);
|
||||
rinv = 1.0/r;
|
||||
screening = exp(-kappa*(r-(radi+radj)));
|
||||
forceyukawa = a[itype][jtype] * screening;
|
||||
|
||||
fpair = factor*forceyukawa * rinv;
|
||||
|
||||
f[i][0] += delx*fpair;
|
||||
f[i][1] += dely*fpair;
|
||||
f[i][2] += delz*fpair;
|
||||
|
||||
if (eflag) {
|
||||
evdwl = a[itype][jtype]/kappa * screening - offset[itype][jtype];
|
||||
evdwl *= factor;
|
||||
}
|
||||
|
||||
if (evflag) ev_tally_full(i,evdwl,0.0,fpair,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
59
src/GPU/pair_yukawa_colloid_gpu.h
Normal file
59
src/GPU/pair_yukawa_colloid_gpu.h
Normal file
@ -0,0 +1,59 @@
|
||||
/* ----------------------------------------------------------------------
|
||||
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.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(yukawa/colloid/gpu,PairYukawaColloidGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_YUKAWA_COLLOID_GPU_H
|
||||
#define LMP_PAIR_YUKAWA_COLLOID_GPU_H
|
||||
|
||||
#include "pair_yukawa_colloid.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairYukawaColloidGPU : public PairYukawaColloid {
|
||||
public:
|
||||
PairYukawaColloidGPU(LAMMPS *lmp);
|
||||
~PairYukawaColloidGPU();
|
||||
void cpu_compute(int, int, int, int, int *, int *, int **);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
int *gpulist;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/* ERROR/WARNING messages:
|
||||
|
||||
E: Insufficient memory on accelerator
|
||||
|
||||
There is insufficient memory on one of the devices specified for the gpu
|
||||
package
|
||||
|
||||
E: Cannot use newton pair with yukawa/colloid/gpu pair style
|
||||
|
||||
Self-explanatory.
|
||||
|
||||
*/
|
||||
Reference in New Issue
Block a user