ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@10052 f3b2605a-c512-4ea7-a41b-209d697bcdaa

Browse Source
This commit is contained in:
sjplimp
2013-06-08 21:35:06 +00:00
parent eec6088ce7
commit 00fe31ddd4
12 changed files with 2298 additions and 2298 deletions
Download Patch File Download Diff File Expand all files Collapse all files

148
src/DIPOLE/pair_lj_cut_dipole_cut.h
View File

@ -1,74 +1,74 @@
/* ----------------------------------------------------------------------
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/dipole/cut,PairLJCutDipoleCut)
#else
#ifndef LMP_PAIR_LJ_CUT_DIPOLE_CUT_H
#define LMP_PAIR_LJ_CUT_DIPOLE_CUT_H
#include "pair.h"
namespace LAMMPS_NS {
class PairLJCutDipoleCut : public Pair {
public:
PairLJCutDipoleCut(class LAMMPS *);
virtual ~PairLJCutDipoleCut();
virtual void compute(int, int);
void settings(int, char **);
void coeff(int, char **);
void init_style();
double init_one(int, int);
void write_restart(FILE *);
void read_restart(FILE *);
void write_restart_settings(FILE *);
void read_restart_settings(FILE *);
protected:
double cut_lj_global,cut_coul_global;
double **cut_lj,**cut_ljsq;
double **cut_coul,**cut_coulsq;
double **epsilon,**sigma;
double **lj1,**lj2,**lj3,**lj4,**offset;
void allocate();
};
}
#endif
#endif
/* ERROR/WARNING messages:
E: Incorrect args in pair_style command
Self-explanatory.
E: Cannot (yet) use 'electron' units with dipoles
This feature is not yet supported.
E: Incorrect args for pair coefficients
Self-explanatory. Check the input script or data file.
E: Pair dipole/cut requires atom attributes q, mu, torque
The atom style defined does not have these attributes.
*/
/* ----------------------------------------------------------------------
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/dipole/cut,PairLJCutDipoleCut)
#else
#ifndef LMP_PAIR_LJ_CUT_DIPOLE_CUT_H
#define LMP_PAIR_LJ_CUT_DIPOLE_CUT_H
#include "pair.h"
namespace LAMMPS_NS {
class PairLJCutDipoleCut : public Pair {
public:
PairLJCutDipoleCut(class LAMMPS *);
virtual ~PairLJCutDipoleCut();
virtual void compute(int, int);
void settings(int, char **);
void coeff(int, char **);
void init_style();
double init_one(int, int);
void write_restart(FILE *);
void read_restart(FILE *);
void write_restart_settings(FILE *);
void read_restart_settings(FILE *);
protected:
double cut_lj_global,cut_coul_global;
double **cut_lj,**cut_ljsq;
double **cut_coul,**cut_coulsq;
double **epsilon,**sigma;
double **lj1,**lj2,**lj3,**lj4,**offset;
void allocate();
};
}
#endif
#endif
/* ERROR/WARNING messages:
E: Incorrect args in pair_style command
Self-explanatory.
E: Cannot (yet) use 'electron' units with dipoles
This feature is not yet supported.
E: Incorrect args for pair coefficients
Self-explanatory. Check the input script or data file.
E: Pair dipole/cut requires atom attributes q, mu, torque
The atom style defined does not have these attributes.
*/

740
src/GPU/pair_lj_cut_dipole_cut_gpu.cpp
View File

@ -1,370 +1,370 @@
/* ----------------------------------------------------------------------
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_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;
/* ---------------------------------------------------------------------- */
PairLJCutDipoleCutGPU::PairLJCutDipoleCutGPU(LAMMPS *lmp) : PairLJCutDipoleCut(lmp),
gpu_mode(GPU_FORCE)
{
respa_enable = 0;
cpu_time = 0.0;
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
}
/* ----------------------------------------------------------------------
free all arrays
------------------------------------------------------------------------- */
PairLJCutDipoleCutGPU::~PairLJCutDipoleCutGPU()
{
dpl_gpu_clear();
}
/* ---------------------------------------------------------------------- */
void PairLJCutDipoleCutGPU::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 PairLJCutDipoleCutGPU::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");
if (strcmp(update->unit_style,"electron") == 0)
error->all(FLERR,"Cannot (yet) use 'electron' units with dipoles");
// 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 PairLJCutDipoleCutGPU::memory_usage()
{
double bytes = Pair::memory_usage();
return bytes + dpl_gpu_bytes();
}
/* ---------------------------------------------------------------------- */
void PairLJCutDipoleCutGPU::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);
}
}
}
}
/* ----------------------------------------------------------------------
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_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;
/* ---------------------------------------------------------------------- */
PairLJCutDipoleCutGPU::PairLJCutDipoleCutGPU(LAMMPS *lmp) : PairLJCutDipoleCut(lmp),
gpu_mode(GPU_FORCE)
{
respa_enable = 0;
cpu_time = 0.0;
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
}
/* ----------------------------------------------------------------------
free all arrays
------------------------------------------------------------------------- */
PairLJCutDipoleCutGPU::~PairLJCutDipoleCutGPU()
{
dpl_gpu_clear();
}
/* ---------------------------------------------------------------------- */
void PairLJCutDipoleCutGPU::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 PairLJCutDipoleCutGPU::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");
if (strcmp(update->unit_style,"electron") == 0)
error->all(FLERR,"Cannot (yet) use 'electron' units with dipoles");
// 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 PairLJCutDipoleCutGPU::memory_usage()
{
double bytes = Pair::memory_usage();
return bytes + dpl_gpu_bytes();
}
/* ---------------------------------------------------------------------- */
void PairLJCutDipoleCutGPU::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);
}
}
}
}

134
src/GPU/pair_lj_cut_dipole_cut_gpu.h
View File

@ -1,67 +1,67 @@
/* ----------------------------------------------------------------------
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/dipole/cut/gpu,PairLJCutDipoleCutGPU)
#else
#ifndef LMP_PAIR_LJ_CUT_DIPOLE_CUT_GPU_H
#define LMP_PAIR_LJ_CUT_DIPOLE_CUT_GPU_H
#include "pair_lj_cut_dipole_cut.h"
namespace LAMMPS_NS {
class PairLJCutDipoleCutGPU : public PairLJCutDipoleCut {
public:
PairLJCutDipoleCutGPU(LAMMPS *lmp);
~PairLJCutDipoleCutGPU();
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 dipole/cut/gpu requires atom attributes q, mu, torque
The atom style defined does not have this attribute.
E: Cannot use newton pair with dipole/cut/gpu pair style
Self-explanatory.
E: Cannot (yet) use 'electron' units with dipoles
This feature is not yet supported.
*/
/* ----------------------------------------------------------------------
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/dipole/cut/gpu,PairLJCutDipoleCutGPU)
#else
#ifndef LMP_PAIR_LJ_CUT_DIPOLE_CUT_GPU_H
#define LMP_PAIR_LJ_CUT_DIPOLE_CUT_GPU_H
#include "pair_lj_cut_dipole_cut.h"
namespace LAMMPS_NS {
class PairLJCutDipoleCutGPU : public PairLJCutDipoleCut {
public:
PairLJCutDipoleCutGPU(LAMMPS *lmp);
~PairLJCutDipoleCutGPU();
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 dipole/cut/gpu requires atom attributes q, mu, torque
The atom style defined does not have this attribute.
E: Cannot use newton pair with dipole/cut/gpu pair style
Self-explanatory.
E: Cannot (yet) use 'electron' units with dipoles
This feature is not yet supported.
*/

800
src/GPU/pair_lj_sf_dipole_sf_gpu.cpp
View File

@ -1,400 +1,400 @@
/* ----------------------------------------------------------------------
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_sf_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;
/* ---------------------------------------------------------------------- */
PairLJSFDipoleSFGPU::PairLJSFDipoleSFGPU(LAMMPS *lmp) : PairLJSFDipoleSF(lmp),
gpu_mode(GPU_FORCE)
{
respa_enable = 0;
cpu_time = 0.0;
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
}
/* ----------------------------------------------------------------------
free all arrays
------------------------------------------------------------------------- */
PairLJSFDipoleSFGPU::~PairLJSFDipoleSFGPU()
{
dplsf_gpu_clear();
}
/* ---------------------------------------------------------------------- */
void PairLJSFDipoleSFGPU::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 PairLJSFDipoleSFGPU::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");
if (strcmp(update->unit_style,"electron") == 0)
error->all(FLERR,"Cannot (yet) use 'electron' units with dipoles");
// 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 PairLJSFDipoleSFGPU::memory_usage()
{
double bytes = Pair::memory_usage();
return bytes + dplsf_gpu_bytes();
}
/* ---------------------------------------------------------------------- */
void PairLJSFDipoleSFGPU::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;
pidotr = mu[i][0]*delx + mu[i][1]*dely + mu[i][2]*delz;
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;
pjdotr = mu[j][0]*delx + mu[j][1]*dely + mu[j][2]*delz;
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);
}
}
}
}
/* ----------------------------------------------------------------------
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_sf_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;
/* ---------------------------------------------------------------------- */
PairLJSFDipoleSFGPU::PairLJSFDipoleSFGPU(LAMMPS *lmp) : PairLJSFDipoleSF(lmp),
gpu_mode(GPU_FORCE)
{
respa_enable = 0;
cpu_time = 0.0;
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
}
/* ----------------------------------------------------------------------
free all arrays
------------------------------------------------------------------------- */
PairLJSFDipoleSFGPU::~PairLJSFDipoleSFGPU()
{
dplsf_gpu_clear();
}
/* ---------------------------------------------------------------------- */
void PairLJSFDipoleSFGPU::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 PairLJSFDipoleSFGPU::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");
if (strcmp(update->unit_style,"electron") == 0)
error->all(FLERR,"Cannot (yet) use 'electron' units with dipoles");
// 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 PairLJSFDipoleSFGPU::memory_usage()
{
double bytes = Pair::memory_usage();
return bytes + dplsf_gpu_bytes();
}
/* ---------------------------------------------------------------------- */
void PairLJSFDipoleSFGPU::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;
pidotr = mu[i][0]*delx + mu[i][1]*dely + mu[i][2]*delz;
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;
pjdotr = mu[j][0]*delx + mu[j][1]*dely + mu[j][2]*delz;
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);
}
}
}
}

134
src/GPU/pair_lj_sf_dipole_sf_gpu.h
View File

@ -1,67 +1,67 @@
/* ----------------------------------------------------------------------
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/sf/dipole/sf/gpu,PairLJSFDipoleSFGPU)
#else
#ifndef LMP_PAIR_LJ_SF_DIPOLE_SF_GPU_H
#define LMP_PAIR_LJ_SF_DIPOLE_SF_GPU_H
#include "pair_lj_sf_dipole_sf.h"
namespace LAMMPS_NS {
class PairLJSFDipoleSFGPU : public PairLJSFDipoleSF {
public:
PairLJSFDipoleSFGPU(LAMMPS *lmp);
~PairLJSFDipoleSFGPU();
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.
E: Cannot (yet) use 'electron' units with dipoles
This feature is not yet supported.
*/
/* ----------------------------------------------------------------------
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/sf/dipole/sf/gpu,PairLJSFDipoleSFGPU)
#else
#ifndef LMP_PAIR_LJ_SF_DIPOLE_SF_GPU_H
#define LMP_PAIR_LJ_SF_DIPOLE_SF_GPU_H
#include "pair_lj_sf_dipole_sf.h"
namespace LAMMPS_NS {
class PairLJSFDipoleSFGPU : public PairLJSFDipoleSF {
public:
PairLJSFDipoleSFGPU(LAMMPS *lmp);
~PairLJSFDipoleSFGPU();
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.
E: Cannot (yet) use 'electron' units with dipoles
This feature is not yet supported.
*/

1038
src/KSPACE/pppm_disp_tip4p.cpp
View File

File diff suppressed because it is too large Load Diff

108
src/USER-MISC/pair_lj_sf_dipole_sf.h
View File

@ -1,54 +1,54 @@
/* ----------------------------------------------------------------------
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/sf/dipole/sf,PairLJSFDipoleSF)
#else
#ifndef LMP_PAIR_LJ_SF_DIPOLE_SF_H
#define LMP_PAIR_LJ_SF_DIPOLE_SF_H
#include "pair.h"
namespace LAMMPS_NS {
class PairLJSFDipoleSF : public Pair {
public:
PairLJSFDipoleSF(class LAMMPS *);
virtual ~PairLJSFDipoleSF();
virtual void compute(int, int);
void settings(int, char **);
void coeff(int, char **);
void init_style();
double init_one(int, int);
void write_restart(FILE *);
void read_restart(FILE *);
void write_restart_settings(FILE *);
void read_restart_settings(FILE *);
protected:
double cut_lj_global,cut_coul_global;
double **cut_lj,**cut_ljsq;
double **cut_coul,**cut_coulsq;
double **epsilon,**sigma;
double **lj1,**lj2,**lj3,**lj4;
void allocate();
};
}
#endif
#endif
/* ----------------------------------------------------------------------
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/sf/dipole/sf,PairLJSFDipoleSF)
#else
#ifndef LMP_PAIR_LJ_SF_DIPOLE_SF_H
#define LMP_PAIR_LJ_SF_DIPOLE_SF_H
#include "pair.h"
namespace LAMMPS_NS {
class PairLJSFDipoleSF : public Pair {
public:
PairLJSFDipoleSF(class LAMMPS *);
virtual ~PairLJSFDipoleSF();
virtual void compute(int, int);
void settings(int, char **);
void coeff(int, char **);
void init_style();
double init_one(int, int);
void write_restart(FILE *);
void read_restart(FILE *);
void write_restart_settings(FILE *);
void read_restart_settings(FILE *);
protected:
double cut_lj_global,cut_coul_global;
double **cut_lj,**cut_ljsq;
double **cut_coul,**cut_coulsq;
double **epsilon,**sigma;
double **lj1,**lj2,**lj3,**lj4;
void allocate();
};
}
#endif
#endif

92
src/USER-OMP/msm_omp.h
View File

@ -1,46 +1,46 @@
/* -*- c++ -*- ----------------------------------------------------------
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 KSPACE_CLASS
KSpaceStyle(msm/omp,MSMOMP)
#else
#ifndef LMP_MSM_OMP_H
#define LMP_MSM_OMP_H
#include "msm.h"
#include "thr_omp.h"
namespace LAMMPS_NS {
class MSMOMP : public MSM, public ThrOMP {
public:
MSMOMP(class LAMMPS *, int, char **);
virtual ~MSMOMP () {};
protected:
virtual void direct(int);
virtual void compute(int,int);
private:
template <int, int, int> void direct_eval(int);
template <int> void direct_peratom(int);
};
}
#endif
#endif
/* -*- c++ -*- ----------------------------------------------------------
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 KSPACE_CLASS
KSpaceStyle(msm/omp,MSMOMP)
#else
#ifndef LMP_MSM_OMP_H
#define LMP_MSM_OMP_H
#include "msm.h"
#include "thr_omp.h"
namespace LAMMPS_NS {
class MSMOMP : public MSM, public ThrOMP {
public:
MSMOMP(class LAMMPS *, int, char **);
virtual ~MSMOMP () {};
protected:
virtual void direct(int);
virtual void compute(int,int);
private:
template <int, int, int> void direct_eval(int);
template <int> void direct_peratom(int);
};
}
#endif
#endif

572
src/USER-OMP/pair_lj_cut_dipole_cut_omp.cpp
View File

@ -1,286 +1,286 @@
/* ----------------------------------------------------------------------
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
http://lammps.sandia.gov, Sandia National Laboratories
Steve Plimpton, sjplimp@sandia.gov
This software is distributed under the GNU General Public License.
See the README file in the top-level LAMMPS directory.
------------------------------------------------------------------------- */
/* ----------------------------------------------------------------------
Contributing author: Axel Kohlmeyer (Temple U)
------------------------------------------------------------------------- */
#include "math.h"
#include "pair_lj_cut_dipole_cut_omp.h"
#include "atom.h"
#include "comm.h"
#include "force.h"
#include "neighbor.h"
#include "neigh_list.h"
#include "suffix.h"
using namespace LAMMPS_NS;
/* ---------------------------------------------------------------------- */
PairLJCutDipoleCutOMP::PairLJCutDipoleCutOMP(LAMMPS *lmp) :
PairLJCutDipoleCut(lmp), ThrOMP(lmp, THR_PAIR)
{
suffix_flag |= Suffix::OMP;
respa_enable = 0;
}
/* ---------------------------------------------------------------------- */
void PairLJCutDipoleCutOMP::compute(int eflag, int vflag)
{
if (eflag || vflag) {
ev_setup(eflag,vflag);
} else evflag = vflag_fdotr = 0;
const int nall = atom->nlocal + atom->nghost;
const int nthreads = comm->nthreads;
const int inum = list->inum;
#if defined(_OPENMP)
#pragma omp parallel default(none) shared(eflag,vflag)
#endif
{
int ifrom, ito, tid;
loop_setup_thr(ifrom, ito, tid, inum, nthreads);
ThrData *thr = fix->get_thr(tid);
ev_setup_thr(eflag, vflag, nall, eatom, vatom, thr);
if (evflag) {
if (eflag) {
if (force->newton_pair) eval<1,1,1>(ifrom, ito, thr);
else eval<1,1,0>(ifrom, ito, thr);
} else {
if (force->newton_pair) eval<1,0,1>(ifrom, ito, thr);
else eval<1,0,0>(ifrom, ito, thr);
}
} else {
if (force->newton_pair) eval<0,0,1>(ifrom, ito, thr);
else eval<0,0,0>(ifrom, ito, thr);
}
reduce_thr(this, eflag, vflag, thr);
} // end of omp parallel region
}
template <int EVFLAG, int EFLAG, int NEWTON_PAIR>
void PairLJCutDipoleCutOMP::eval(int iifrom, int iito, ThrData * const thr)
{
int i,j,ii,jj,jnum,itype,jtype;
double xtmp,ytmp,ztmp,qtmp,delx,dely,delz,evdwl,ecoul;
double rsq,rinv,r2inv,r6inv,r3inv,r5inv,r7inv,fx,fy,fz;
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 *ilist,*jlist,*numneigh,**firstneigh;
evdwl = 0.0;
const dbl3_t * _noalias const x = (dbl3_t *) atom->x[0];
dbl3_t * _noalias const f = (dbl3_t *) thr->get_f()[0];
double * const * const torque = thr->get_torque();
const double * _noalias const q = atom->q;
const dbl4_t * _noalias const mu = (dbl4_t *) atom->mu[0];
const int * _noalias const type = atom->type;
const int nlocal = atom->nlocal;
const double * _noalias const special_coul = force->special_coul;
const double * _noalias const special_lj = force->special_lj;
const double qqrd2e = force->qqrd2e;
double fxtmp,fytmp,fztmp,t1tmp,t2tmp,t3tmp;
ilist = list->ilist;
numneigh = list->numneigh;
firstneigh = list->firstneigh;
// loop over neighbors of my atoms
for (ii = iifrom; ii < iito; ++ii) {
i = ilist[ii];
xtmp = x[i].x;
ytmp = x[i].y;
ztmp = x[i].z;
qtmp = q[i];
itype = type[i];
jlist = firstneigh[i];
jnum = numneigh[i];
fxtmp=fytmp=fztmp=t1tmp=t2tmp=t3tmp=0.0;
for (jj = 0; jj < jnum; jj++) {
j = jlist[jj];
factor_coul = special_coul[sbmask(j)];
factor_lj = special_lj[sbmask(j)];
j &= NEIGHMASK;
delx = xtmp - x[j].x;
dely = ytmp - x[j].y;
delz = ztmp - x[j].z;
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].w > 0.0 && mu[j].w > 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
r7inv = r5inv*r2inv;
pdotp = mu[i].x*mu[j].x + mu[i].y*mu[j].y + mu[i].z*mu[j].z;
pidotr = mu[i].x*delx + mu[i].y*dely + mu[i].z*delz;
pjdotr = mu[j].x*delx + mu[j].y*dely + mu[j].z*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].x + pre3*mu[j].x;
forcecouly += pre1*dely + pre2*mu[i].y + pre3*mu[j].y;
forcecoulz += pre1*delz + pre2*mu[i].z + pre3*mu[j].z;
crossx = pre4 * (mu[i].y*mu[j].z - mu[i].z*mu[j].y);
crossy = pre4 * (mu[i].z*mu[j].x - mu[i].x*mu[j].z);
crossz = pre4 * (mu[i].x*mu[j].y - mu[i].y*mu[j].x);
tixcoul += crossx + pre2 * (mu[i].y*delz - mu[i].z*dely);
tiycoul += crossy + pre2 * (mu[i].z*delx - mu[i].x*delz);
tizcoul += crossz + pre2 * (mu[i].x*dely - mu[i].y*delx);
tjxcoul += -crossx + pre3 * (mu[j].y*delz - mu[j].z*dely);
tjycoul += -crossy + pre3 * (mu[j].z*delx - mu[j].x*delz);
tjzcoul += -crossz + pre3 * (mu[j].x*dely - mu[j].y*delx);
}
if (mu[i].w > 0.0 && q[j] != 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
pidotr = mu[i].x*delx + mu[i].y*dely + mu[i].z*delz;
pre1 = 3.0*q[j]*r5inv * pidotr;
pre2 = q[j]*r3inv;
forcecoulx += pre2*mu[i].x - pre1*delx;
forcecouly += pre2*mu[i].y - pre1*dely;
forcecoulz += pre2*mu[i].z - pre1*delz;
tixcoul += pre2 * (mu[i].y*delz - mu[i].z*dely);
tiycoul += pre2 * (mu[i].z*delx - mu[i].x*delz);
tizcoul += pre2 * (mu[i].x*dely - mu[i].y*delx);
}
if (mu[j].w > 0.0 && qtmp != 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
pjdotr = mu[j].x*delx + mu[j].y*dely + mu[j].z*delz;
pre1 = 3.0*qtmp*r5inv * pjdotr;
pre2 = qtmp*r3inv;
forcecoulx += pre1*delx - pre2*mu[j].x;
forcecouly += pre1*dely - pre2*mu[j].y;
forcecoulz += pre1*delz - pre2*mu[j].z;
tjxcoul += -pre2 * (mu[j].y*delz - mu[j].z*dely);
tjycoul += -pre2 * (mu[j].z*delx - mu[j].x*delz);
tjzcoul += -pre2 * (mu[j].x*dely - mu[j].y*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
fxtmp += fx;
fytmp += fy;
fztmp += fz;
t1tmp += fq*tixcoul;
t2tmp += fq*tiycoul;
t3tmp += fq*tizcoul;
if (NEWTON_PAIR || j < nlocal) {
f[j].x -= fx;
f[j].y -= fy;
f[j].z -= fz;
torque[j][0] += fq*tjxcoul;
torque[j][1] += fq*tjycoul;
torque[j][2] += fq*tjzcoul;
}
if (EFLAG) {
if (rsq < cut_coulsq[itype][jtype]) {
ecoul = qtmp*q[j]*rinv;
if (mu[i].w > 0.0 && mu[j].w > 0.0)
ecoul += r3inv*pdotp - 3.0*r5inv*pidotr*pjdotr;
if (mu[i].w > 0.0 && q[j] != 0.0)
ecoul += -q[j]*r3inv*pidotr;
if (mu[j].w > 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_thr(this,i,j,nlocal,NEWTON_PAIR,
evdwl,ecoul,fx,fy,fz,delx,dely,delz,thr);
}
}
f[i].x += fxtmp;
f[i].y += fytmp;
f[i].z += fztmp;
torque[i][0] += t1tmp;
torque[i][1] += t2tmp;
torque[i][2] += t3tmp;
}
}
/* ---------------------------------------------------------------------- */
double PairLJCutDipoleCutOMP::memory_usage()
{
double bytes = memory_usage_thr();
bytes += PairLJCutDipoleCut::memory_usage();
return bytes;
}
/* ----------------------------------------------------------------------
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
http://lammps.sandia.gov, Sandia National Laboratories
Steve Plimpton, sjplimp@sandia.gov
This software is distributed under the GNU General Public License.
See the README file in the top-level LAMMPS directory.
------------------------------------------------------------------------- */
/* ----------------------------------------------------------------------
Contributing author: Axel Kohlmeyer (Temple U)
------------------------------------------------------------------------- */
#include "math.h"
#include "pair_lj_cut_dipole_cut_omp.h"
#include "atom.h"
#include "comm.h"
#include "force.h"
#include "neighbor.h"
#include "neigh_list.h"
#include "suffix.h"
using namespace LAMMPS_NS;
/* ---------------------------------------------------------------------- */
PairLJCutDipoleCutOMP::PairLJCutDipoleCutOMP(LAMMPS *lmp) :
PairLJCutDipoleCut(lmp), ThrOMP(lmp, THR_PAIR)
{
suffix_flag |= Suffix::OMP;
respa_enable = 0;
}
/* ---------------------------------------------------------------------- */
void PairLJCutDipoleCutOMP::compute(int eflag, int vflag)
{
if (eflag || vflag) {
ev_setup(eflag,vflag);
} else evflag = vflag_fdotr = 0;
const int nall = atom->nlocal + atom->nghost;
const int nthreads = comm->nthreads;
const int inum = list->inum;
#if defined(_OPENMP)
#pragma omp parallel default(none) shared(eflag,vflag)
#endif
{
int ifrom, ito, tid;
loop_setup_thr(ifrom, ito, tid, inum, nthreads);
ThrData *thr = fix->get_thr(tid);
ev_setup_thr(eflag, vflag, nall, eatom, vatom, thr);
if (evflag) {
if (eflag) {
if (force->newton_pair) eval<1,1,1>(ifrom, ito, thr);
else eval<1,1,0>(ifrom, ito, thr);
} else {
if (force->newton_pair) eval<1,0,1>(ifrom, ito, thr);
else eval<1,0,0>(ifrom, ito, thr);
}
} else {
if (force->newton_pair) eval<0,0,1>(ifrom, ito, thr);
else eval<0,0,0>(ifrom, ito, thr);
}
reduce_thr(this, eflag, vflag, thr);
} // end of omp parallel region
}
template <int EVFLAG, int EFLAG, int NEWTON_PAIR>
void PairLJCutDipoleCutOMP::eval(int iifrom, int iito, ThrData * const thr)
{
int i,j,ii,jj,jnum,itype,jtype;
double xtmp,ytmp,ztmp,qtmp,delx,dely,delz,evdwl,ecoul;
double rsq,rinv,r2inv,r6inv,r3inv,r5inv,r7inv,fx,fy,fz;
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 *ilist,*jlist,*numneigh,**firstneigh;
evdwl = 0.0;
const dbl3_t * _noalias const x = (dbl3_t *) atom->x[0];
dbl3_t * _noalias const f = (dbl3_t *) thr->get_f()[0];
double * const * const torque = thr->get_torque();
const double * _noalias const q = atom->q;
const dbl4_t * _noalias const mu = (dbl4_t *) atom->mu[0];
const int * _noalias const type = atom->type;
const int nlocal = atom->nlocal;
const double * _noalias const special_coul = force->special_coul;
const double * _noalias const special_lj = force->special_lj;
const double qqrd2e = force->qqrd2e;
double fxtmp,fytmp,fztmp,t1tmp,t2tmp,t3tmp;
ilist = list->ilist;
numneigh = list->numneigh;
firstneigh = list->firstneigh;
// loop over neighbors of my atoms
for (ii = iifrom; ii < iito; ++ii) {
i = ilist[ii];
xtmp = x[i].x;
ytmp = x[i].y;
ztmp = x[i].z;
qtmp = q[i];
itype = type[i];
jlist = firstneigh[i];
jnum = numneigh[i];
fxtmp=fytmp=fztmp=t1tmp=t2tmp=t3tmp=0.0;
for (jj = 0; jj < jnum; jj++) {
j = jlist[jj];
factor_coul = special_coul[sbmask(j)];
factor_lj = special_lj[sbmask(j)];
j &= NEIGHMASK;
delx = xtmp - x[j].x;
dely = ytmp - x[j].y;
delz = ztmp - x[j].z;
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].w > 0.0 && mu[j].w > 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
r7inv = r5inv*r2inv;
pdotp = mu[i].x*mu[j].x + mu[i].y*mu[j].y + mu[i].z*mu[j].z;
pidotr = mu[i].x*delx + mu[i].y*dely + mu[i].z*delz;
pjdotr = mu[j].x*delx + mu[j].y*dely + mu[j].z*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].x + pre3*mu[j].x;
forcecouly += pre1*dely + pre2*mu[i].y + pre3*mu[j].y;
forcecoulz += pre1*delz + pre2*mu[i].z + pre3*mu[j].z;
crossx = pre4 * (mu[i].y*mu[j].z - mu[i].z*mu[j].y);
crossy = pre4 * (mu[i].z*mu[j].x - mu[i].x*mu[j].z);
crossz = pre4 * (mu[i].x*mu[j].y - mu[i].y*mu[j].x);
tixcoul += crossx + pre2 * (mu[i].y*delz - mu[i].z*dely);
tiycoul += crossy + pre2 * (mu[i].z*delx - mu[i].x*delz);
tizcoul += crossz + pre2 * (mu[i].x*dely - mu[i].y*delx);
tjxcoul += -crossx + pre3 * (mu[j].y*delz - mu[j].z*dely);
tjycoul += -crossy + pre3 * (mu[j].z*delx - mu[j].x*delz);
tjzcoul += -crossz + pre3 * (mu[j].x*dely - mu[j].y*delx);
}
if (mu[i].w > 0.0 && q[j] != 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
pidotr = mu[i].x*delx + mu[i].y*dely + mu[i].z*delz;
pre1 = 3.0*q[j]*r5inv * pidotr;
pre2 = q[j]*r3inv;
forcecoulx += pre2*mu[i].x - pre1*delx;
forcecouly += pre2*mu[i].y - pre1*dely;
forcecoulz += pre2*mu[i].z - pre1*delz;
tixcoul += pre2 * (mu[i].y*delz - mu[i].z*dely);
tiycoul += pre2 * (mu[i].z*delx - mu[i].x*delz);
tizcoul += pre2 * (mu[i].x*dely - mu[i].y*delx);
}
if (mu[j].w > 0.0 && qtmp != 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
pjdotr = mu[j].x*delx + mu[j].y*dely + mu[j].z*delz;
pre1 = 3.0*qtmp*r5inv * pjdotr;
pre2 = qtmp*r3inv;
forcecoulx += pre1*delx - pre2*mu[j].x;
forcecouly += pre1*dely - pre2*mu[j].y;
forcecoulz += pre1*delz - pre2*mu[j].z;
tjxcoul += -pre2 * (mu[j].y*delz - mu[j].z*dely);
tjycoul += -pre2 * (mu[j].z*delx - mu[j].x*delz);
tjzcoul += -pre2 * (mu[j].x*dely - mu[j].y*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
fxtmp += fx;
fytmp += fy;
fztmp += fz;
t1tmp += fq*tixcoul;
t2tmp += fq*tiycoul;
t3tmp += fq*tizcoul;
if (NEWTON_PAIR || j < nlocal) {
f[j].x -= fx;
f[j].y -= fy;
f[j].z -= fz;
torque[j][0] += fq*tjxcoul;
torque[j][1] += fq*tjycoul;
torque[j][2] += fq*tjzcoul;
}
if (EFLAG) {
if (rsq < cut_coulsq[itype][jtype]) {
ecoul = qtmp*q[j]*rinv;
if (mu[i].w > 0.0 && mu[j].w > 0.0)
ecoul += r3inv*pdotp - 3.0*r5inv*pidotr*pjdotr;
if (mu[i].w > 0.0 && q[j] != 0.0)
ecoul += -q[j]*r3inv*pidotr;
if (mu[j].w > 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_thr(this,i,j,nlocal,NEWTON_PAIR,
evdwl,ecoul,fx,fy,fz,delx,dely,delz,thr);
}
}
f[i].x += fxtmp;
f[i].y += fytmp;
f[i].z += fztmp;
torque[i][0] += t1tmp;
torque[i][1] += t2tmp;
torque[i][2] += t3tmp;
}
}
/* ---------------------------------------------------------------------- */
double PairLJCutDipoleCutOMP::memory_usage()
{
double bytes = memory_usage_thr();
bytes += PairLJCutDipoleCut::memory_usage();
return bytes;
}

96
src/USER-OMP/pair_lj_cut_dipole_cut_omp.h
View File

@ -1,48 +1,48 @@
/* -*- c++ -*- ----------------------------------------------------------
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: Axel Kohlmeyer (Temple U)
------------------------------------------------------------------------- */
#ifdef PAIR_CLASS
PairStyle(lj/cut/dipole/cut/omp,PairLJCutDipoleCutOMP)
#else
#ifndef LMP_PAIR_LJ_CUT_DIPOLE_CUT_OMP_H
#define LMP_PAIR_LJ_CUT_DIPOLE_CUT_OMP_H
#include "pair_lj_cut_dipole_cut.h"
#include "thr_omp.h"
namespace LAMMPS_NS {
class PairLJCutDipoleCutOMP : public PairLJCutDipoleCut, public ThrOMP {
public:
PairLJCutDipoleCutOMP(class LAMMPS *);
virtual void compute(int, int);
virtual double memory_usage();
private:
template <int EVFLAG, int EFLAG, int NEWTON_PAIR>
void eval(int ifrom, int ito, ThrData * const thr);
};
}
#endif
#endif
/* -*- c++ -*- ----------------------------------------------------------
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: Axel Kohlmeyer (Temple U)
------------------------------------------------------------------------- */
#ifdef PAIR_CLASS
PairStyle(lj/cut/dipole/cut/omp,PairLJCutDipoleCutOMP)
#else
#ifndef LMP_PAIR_LJ_CUT_DIPOLE_CUT_OMP_H
#define LMP_PAIR_LJ_CUT_DIPOLE_CUT_OMP_H
#include "pair_lj_cut_dipole_cut.h"
#include "thr_omp.h"
namespace LAMMPS_NS {
class PairLJCutDipoleCutOMP : public PairLJCutDipoleCut, public ThrOMP {
public:
PairLJCutDipoleCutOMP(class LAMMPS *);
virtual void compute(int, int);
virtual double memory_usage();
private:
template <int EVFLAG, int EFLAG, int NEWTON_PAIR>
void eval(int ifrom, int ito, ThrData * const thr);
};
}
#endif
#endif

638
src/USER-OMP/pair_lj_sf_dipole_sf_omp.cpp
View File

@ -1,319 +1,319 @@
/* ----------------------------------------------------------------------
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
http://lammps.sandia.gov, Sandia National Laboratories
Steve Plimpton, sjplimp@sandia.gov
This software is distributed under the GNU General Public License.
See the README file in the top-level LAMMPS directory.
------------------------------------------------------------------------- */
/* ----------------------------------------------------------------------
Contributing author: Axel Kohlmeyer (Temple U)
------------------------------------------------------------------------- */
#include "math.h"
#include "pair_lj_sf_dipole_sf_omp.h"
#include "atom.h"
#include "comm.h"
#include "force.h"
#include "neighbor.h"
#include "neigh_list.h"
#include "suffix.h"
using namespace LAMMPS_NS;
/* ---------------------------------------------------------------------- */
PairLJSFDipoleSFOMP::PairLJSFDipoleSFOMP(LAMMPS *lmp) :
PairLJSFDipoleSF(lmp), ThrOMP(lmp, THR_PAIR)
{
suffix_flag |= Suffix::OMP;
respa_enable = 0;
}
/* ---------------------------------------------------------------------- */
void PairLJSFDipoleSFOMP::compute(int eflag, int vflag)
{
if (eflag || vflag) {
ev_setup(eflag,vflag);
} else evflag = vflag_fdotr = 0;
const int nall = atom->nlocal + atom->nghost;
const int nthreads = comm->nthreads;
const int inum = list->inum;
#if defined(_OPENMP)
#pragma omp parallel default(none) shared(eflag,vflag)
#endif
{
int ifrom, ito, tid;
loop_setup_thr(ifrom, ito, tid, inum, nthreads);
ThrData *thr = fix->get_thr(tid);
ev_setup_thr(eflag, vflag, nall, eatom, vatom, thr);
if (evflag) {
if (eflag) {
if (force->newton_pair) eval<1,1,1>(ifrom, ito, thr);
else eval<1,1,0>(ifrom, ito, thr);
} else {
if (force->newton_pair) eval<1,0,1>(ifrom, ito, thr);
else eval<1,0,0>(ifrom, ito, thr);
}
} else {
if (force->newton_pair) eval<0,0,1>(ifrom, ito, thr);
else eval<0,0,0>(ifrom, ito, thr);
}
reduce_thr(this, eflag, vflag, thr);
} // end of omp parallel region
}
template <int EVFLAG, int EFLAG, int NEWTON_PAIR>
void PairLJSFDipoleSFOMP::eval(int iifrom, int iito, ThrData * const thr)
{
int i,j,ii,jj,jnum,itype,jtype;
double xtmp,ytmp,ztmp,qtmp,delx,dely,delz,evdwl,ecoul;
double rsq,rinv,r2inv,r6inv,r3inv,r5inv,fx,fy,fz;
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 *ilist,*jlist,*numneigh,**firstneigh;
evdwl = 0.0;
const dbl3_t * _noalias const x = (dbl3_t *) atom->x[0];
dbl3_t * _noalias const f = (dbl3_t *) thr->get_f()[0];
double * const * const torque = thr->get_torque();
const double * _noalias const q = atom->q;
const dbl4_t * _noalias const mu = (dbl4_t *) atom->mu[0];
const int * _noalias const type = atom->type;
const int nlocal = atom->nlocal;
const double * _noalias const special_coul = force->special_coul;
const double * _noalias const special_lj = force->special_lj;
const double qqrd2e = force->qqrd2e;
double fxtmp,fytmp,fztmp,t1tmp,t2tmp,t3tmp;
ilist = list->ilist;
numneigh = list->numneigh;
firstneigh = list->firstneigh;
// loop over neighbors of my atoms
for (ii = iifrom; ii < iito; ++ii) {
i = ilist[ii];
xtmp = x[i].x;
ytmp = x[i].y;
ztmp = x[i].z;
qtmp = q[i];
itype = type[i];
jlist = firstneigh[i];
jnum = numneigh[i];
fxtmp=fytmp=fztmp=t1tmp=t2tmp=t3tmp=0.0;
for (jj = 0; jj < jnum; jj++) {
j = jlist[jj];
factor_coul = special_coul[sbmask(j)];
factor_lj = special_lj[sbmask(j)];
j &= NEIGHMASK;
delx = xtmp - x[j].x;
dely = ytmp - x[j].y;
delz = ztmp - x[j].z;
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
// 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].w > 0.0 && mu[j].w > 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
rcutcoul2inv=1.0/cut_coulsq[itype][jtype];
pdotp = mu[i].x*mu[j].x + mu[i].y*mu[j].y + mu[i].z*mu[j].z;
pidotr = mu[i].x*delx + mu[i].y*dely + mu[i].z*delz;
pjdotr = mu[j].x*delx + mu[j].y*dely + mu[j].z*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].x+pidotr*mu[j].x-presf*delx);
bforcecouly = bfac * (pjdotr*mu[i].y+pidotr*mu[j].y-presf*dely);
bforcecoulz = bfac * (pjdotr*mu[i].z+pidotr*mu[j].z-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].y*mu[j].z - mu[i].z*mu[j].y);
crossy = pre4 * (mu[i].z*mu[j].x - mu[i].x*mu[j].z);
crossz = pre4 * (mu[i].x*mu[j].y - mu[i].y*mu[j].x);
tixcoul += crossx + pre2 * (mu[i].y*delz - mu[i].z*dely);
tiycoul += crossy + pre2 * (mu[i].z*delx - mu[i].x*delz);
tizcoul += crossz + pre2 * (mu[i].x*dely - mu[i].y*delx);
tjxcoul += -crossx + pre3 * (mu[j].y*delz - mu[j].z*dely);
tjycoul += -crossy + pre3 * (mu[j].z*delx - mu[j].x*delz);
tjzcoul += -crossz + pre3 * (mu[j].x*dely - mu[j].y*delx);
}
if (mu[i].w > 0.0 && q[j] != 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
pidotr = mu[i].x*delx + mu[i].y*dely + mu[i].z*delz;
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].x - pre1*delx;
forcecouly += pre2*mu[i].y - pre1*dely;
forcecoulz += pre2*mu[i].z - pre1*delz;
tixcoul += pre2 * (mu[i].y*delz - mu[i].z*dely);
tiycoul += pre2 * (mu[i].z*delx - mu[i].x*delz);
tizcoul += pre2 * (mu[i].x*dely - mu[i].y*delx);
}
if (mu[j].w > 0.0 && qtmp != 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
pjdotr = mu[j].x*delx + mu[j].y*dely + mu[j].z*delz;
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].x;
forcecouly += pre1*dely - pre2*mu[j].y;
forcecoulz += pre1*delz - pre2*mu[j].z;
tjxcoul += -pre2 * (mu[j].y*delz - mu[j].z*dely);
tjycoul += -pre2 * (mu[j].z*delx - mu[j].x*delz);
tjzcoul += -pre2 * (mu[j].x*dely - mu[j].y*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
fxtmp += fx;
fytmp += fy;
fztmp += fz;
t1tmp += fq*tixcoul;
t2tmp += fq*tiycoul;
t3tmp += fq*tizcoul;
if (NEWTON_PAIR || j < nlocal) {
f[j].x -= fx;
f[j].y -= fy;
f[j].z -= fz;
torque[j][0] += fq*tjxcoul;
torque[j][1] += fq*tjycoul;
torque[j][2] += fq*tjzcoul;
}
if (EFLAG) {
if (rsq < cut_coulsq[itype][jtype]) {
ecoul = (1.0-sqrt(rsq)/sqrt(cut_coulsq[itype][jtype]));
ecoul *= ecoul;
ecoul *= qtmp * q[j] * rinv;
if (mu[i].w > 0.0 && mu[j].w > 0.0)
ecoul += bfac * (r3inv*pdotp - 3.0*r5inv*pidotr*pjdotr);
if (mu[i].w > 0.0 && q[j] != 0.0)
ecoul += -q[j] * r3inv * pqfac * pidotr;
if (mu[j].w > 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_thr(this,i,j,nlocal,NEWTON_PAIR,
evdwl,ecoul,fx,fy,fz,delx,dely,delz,thr);
}
}
f[i].x += fxtmp;
f[i].y += fytmp;
f[i].z += fztmp;
torque[i][0] += t1tmp;
torque[i][1] += t2tmp;
torque[i][2] += t3tmp;
}
}
/* ---------------------------------------------------------------------- */
double PairLJSFDipoleSFOMP::memory_usage()
{
double bytes = memory_usage_thr();
bytes += PairLJSFDipoleSF::memory_usage();
return bytes;
}
/* ----------------------------------------------------------------------
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
http://lammps.sandia.gov, Sandia National Laboratories
Steve Plimpton, sjplimp@sandia.gov
This software is distributed under the GNU General Public License.
See the README file in the top-level LAMMPS directory.
------------------------------------------------------------------------- */
/* ----------------------------------------------------------------------
Contributing author: Axel Kohlmeyer (Temple U)
------------------------------------------------------------------------- */
#include "math.h"
#include "pair_lj_sf_dipole_sf_omp.h"
#include "atom.h"
#include "comm.h"
#include "force.h"
#include "neighbor.h"
#include "neigh_list.h"
#include "suffix.h"
using namespace LAMMPS_NS;
/* ---------------------------------------------------------------------- */
PairLJSFDipoleSFOMP::PairLJSFDipoleSFOMP(LAMMPS *lmp) :
PairLJSFDipoleSF(lmp), ThrOMP(lmp, THR_PAIR)
{
suffix_flag |= Suffix::OMP;
respa_enable = 0;
}
/* ---------------------------------------------------------------------- */
void PairLJSFDipoleSFOMP::compute(int eflag, int vflag)
{
if (eflag || vflag) {
ev_setup(eflag,vflag);
} else evflag = vflag_fdotr = 0;
const int nall = atom->nlocal + atom->nghost;
const int nthreads = comm->nthreads;
const int inum = list->inum;
#if defined(_OPENMP)
#pragma omp parallel default(none) shared(eflag,vflag)
#endif
{
int ifrom, ito, tid;
loop_setup_thr(ifrom, ito, tid, inum, nthreads);
ThrData *thr = fix->get_thr(tid);
ev_setup_thr(eflag, vflag, nall, eatom, vatom, thr);
if (evflag) {
if (eflag) {
if (force->newton_pair) eval<1,1,1>(ifrom, ito, thr);
else eval<1,1,0>(ifrom, ito, thr);
} else {
if (force->newton_pair) eval<1,0,1>(ifrom, ito, thr);
else eval<1,0,0>(ifrom, ito, thr);
}
} else {
if (force->newton_pair) eval<0,0,1>(ifrom, ito, thr);
else eval<0,0,0>(ifrom, ito, thr);
}
reduce_thr(this, eflag, vflag, thr);
} // end of omp parallel region
}
template <int EVFLAG, int EFLAG, int NEWTON_PAIR>
void PairLJSFDipoleSFOMP::eval(int iifrom, int iito, ThrData * const thr)
{
int i,j,ii,jj,jnum,itype,jtype;
double xtmp,ytmp,ztmp,qtmp,delx,dely,delz,evdwl,ecoul;
double rsq,rinv,r2inv,r6inv,r3inv,r5inv,fx,fy,fz;
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 *ilist,*jlist,*numneigh,**firstneigh;
evdwl = 0.0;
const dbl3_t * _noalias const x = (dbl3_t *) atom->x[0];
dbl3_t * _noalias const f = (dbl3_t *) thr->get_f()[0];
double * const * const torque = thr->get_torque();
const double * _noalias const q = atom->q;
const dbl4_t * _noalias const mu = (dbl4_t *) atom->mu[0];
const int * _noalias const type = atom->type;
const int nlocal = atom->nlocal;
const double * _noalias const special_coul = force->special_coul;
const double * _noalias const special_lj = force->special_lj;
const double qqrd2e = force->qqrd2e;
double fxtmp,fytmp,fztmp,t1tmp,t2tmp,t3tmp;
ilist = list->ilist;
numneigh = list->numneigh;
firstneigh = list->firstneigh;
// loop over neighbors of my atoms
for (ii = iifrom; ii < iito; ++ii) {
i = ilist[ii];
xtmp = x[i].x;
ytmp = x[i].y;
ztmp = x[i].z;
qtmp = q[i];
itype = type[i];
jlist = firstneigh[i];
jnum = numneigh[i];
fxtmp=fytmp=fztmp=t1tmp=t2tmp=t3tmp=0.0;
for (jj = 0; jj < jnum; jj++) {
j = jlist[jj];
factor_coul = special_coul[sbmask(j)];
factor_lj = special_lj[sbmask(j)];
j &= NEIGHMASK;
delx = xtmp - x[j].x;
dely = ytmp - x[j].y;
delz = ztmp - x[j].z;
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
// 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].w > 0.0 && mu[j].w > 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
rcutcoul2inv=1.0/cut_coulsq[itype][jtype];
pdotp = mu[i].x*mu[j].x + mu[i].y*mu[j].y + mu[i].z*mu[j].z;
pidotr = mu[i].x*delx + mu[i].y*dely + mu[i].z*delz;
pjdotr = mu[j].x*delx + mu[j].y*dely + mu[j].z*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].x+pidotr*mu[j].x-presf*delx);
bforcecouly = bfac * (pjdotr*mu[i].y+pidotr*mu[j].y-presf*dely);
bforcecoulz = bfac * (pjdotr*mu[i].z+pidotr*mu[j].z-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].y*mu[j].z - mu[i].z*mu[j].y);
crossy = pre4 * (mu[i].z*mu[j].x - mu[i].x*mu[j].z);
crossz = pre4 * (mu[i].x*mu[j].y - mu[i].y*mu[j].x);
tixcoul += crossx + pre2 * (mu[i].y*delz - mu[i].z*dely);
tiycoul += crossy + pre2 * (mu[i].z*delx - mu[i].x*delz);
tizcoul += crossz + pre2 * (mu[i].x*dely - mu[i].y*delx);
tjxcoul += -crossx + pre3 * (mu[j].y*delz - mu[j].z*dely);
tjycoul += -crossy + pre3 * (mu[j].z*delx - mu[j].x*delz);
tjzcoul += -crossz + pre3 * (mu[j].x*dely - mu[j].y*delx);
}
if (mu[i].w > 0.0 && q[j] != 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
pidotr = mu[i].x*delx + mu[i].y*dely + mu[i].z*delz;
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].x - pre1*delx;
forcecouly += pre2*mu[i].y - pre1*dely;
forcecoulz += pre2*mu[i].z - pre1*delz;
tixcoul += pre2 * (mu[i].y*delz - mu[i].z*dely);
tiycoul += pre2 * (mu[i].z*delx - mu[i].x*delz);
tizcoul += pre2 * (mu[i].x*dely - mu[i].y*delx);
}
if (mu[j].w > 0.0 && qtmp != 0.0) {
r3inv = r2inv*rinv;
r5inv = r3inv*r2inv;
pjdotr = mu[j].x*delx + mu[j].y*dely + mu[j].z*delz;
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].x;
forcecouly += pre1*dely - pre2*mu[j].y;
forcecoulz += pre1*delz - pre2*mu[j].z;
tjxcoul += -pre2 * (mu[j].y*delz - mu[j].z*dely);
tjycoul += -pre2 * (mu[j].z*delx - mu[j].x*delz);
tjzcoul += -pre2 * (mu[j].x*dely - mu[j].y*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
fxtmp += fx;
fytmp += fy;
fztmp += fz;
t1tmp += fq*tixcoul;
t2tmp += fq*tiycoul;
t3tmp += fq*tizcoul;
if (NEWTON_PAIR || j < nlocal) {
f[j].x -= fx;
f[j].y -= fy;
f[j].z -= fz;
torque[j][0] += fq*tjxcoul;
torque[j][1] += fq*tjycoul;
torque[j][2] += fq*tjzcoul;
}
if (EFLAG) {
if (rsq < cut_coulsq[itype][jtype]) {
ecoul = (1.0-sqrt(rsq)/sqrt(cut_coulsq[itype][jtype]));
ecoul *= ecoul;
ecoul *= qtmp * q[j] * rinv;
if (mu[i].w > 0.0 && mu[j].w > 0.0)
ecoul += bfac * (r3inv*pdotp - 3.0*r5inv*pidotr*pjdotr);
if (mu[i].w > 0.0 && q[j] != 0.0)
ecoul += -q[j] * r3inv * pqfac * pidotr;
if (mu[j].w > 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_thr(this,i,j,nlocal,NEWTON_PAIR,
evdwl,ecoul,fx,fy,fz,delx,dely,delz,thr);
}
}
f[i].x += fxtmp;
f[i].y += fytmp;
f[i].z += fztmp;
torque[i][0] += t1tmp;
torque[i][1] += t2tmp;
torque[i][2] += t3tmp;
}
}
/* ---------------------------------------------------------------------- */
double PairLJSFDipoleSFOMP::memory_usage()
{
double bytes = memory_usage_thr();
bytes += PairLJSFDipoleSF::memory_usage();
return bytes;
}

96
src/USER-OMP/pair_lj_sf_dipole_sf_omp.h
View File

@ -1,48 +1,48 @@
/* -*- c++ -*- ----------------------------------------------------------
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: Axel Kohlmeyer (Temple U)
------------------------------------------------------------------------- */
#ifdef PAIR_CLASS
PairStyle(lj/sf/dipole/sf/omp,PairLJSFDipoleSFOMP)
#else
#ifndef LMP_PAIR_LJ_SF_DIPOLE_SF_OMP_H
#define LMP_PAIR_LJ_SF_DIPOLE_SF_OMP_H
#include "pair_lj_sf_dipole_sf.h"
#include "thr_omp.h"
namespace LAMMPS_NS {
class PairLJSFDipoleSFOMP : public PairLJSFDipoleSF, public ThrOMP {
public:
PairLJSFDipoleSFOMP(class LAMMPS *);
virtual void compute(int, int);
virtual double memory_usage();
private:
template <int EVFLAG, int EFLAG, int NEWTON_PAIR>
void eval(int ifrom, int ito, ThrData * const thr);
};
}
#endif
#endif
/* -*- c++ -*- ----------------------------------------------------------
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: Axel Kohlmeyer (Temple U)
------------------------------------------------------------------------- */
#ifdef PAIR_CLASS
PairStyle(lj/sf/dipole/sf/omp,PairLJSFDipoleSFOMP)
#else
#ifndef LMP_PAIR_LJ_SF_DIPOLE_SF_OMP_H
#define LMP_PAIR_LJ_SF_DIPOLE_SF_OMP_H
#include "pair_lj_sf_dipole_sf.h"
#include "thr_omp.h"
namespace LAMMPS_NS {
class PairLJSFDipoleSFOMP : public PairLJSFDipoleSF, public ThrOMP {
public:
PairLJSFDipoleSFOMP(class LAMMPS *);
virtual void compute(int, int);
virtual double memory_usage();
private:
template <int EVFLAG, int EFLAG, int NEWTON_PAIR>
void eval(int ifrom, int ito, ThrData * const thr);
};
}
#endif
#endif