void PPPMTIP4POMP::fieldforce()
{
// loop over my charges, interpolate electric field from nearby grid points
// (nx,ny,nz) = global coords of grid pt to "lower left" of charge
// (dx,dy,dz) = distance to "lower left" grid pt
// (mx,my,mz) = global coords of moving stencil pt
// ek = 3 components of E-field on particle
const double * const q = atom->q;
const double * const * const x = atom->x;
const int * const type = atom->type;
const int nthreads = comm->nthreads;
const int nlocal = atom->nlocal;
const double qqrd2e = force->qqrd2e;
#if defined(_OPENMP)
#pragma omp parallel default(none)
#endif
{
#if defined(_OPENMP)
// each thread works on a fixed chunk of atoms.
const int tid = omp_get_thread_num();
const int inum = nlocal;
const int idelta = 1 + inum/nthreads;
const int ifrom = tid*idelta;
const int ito = ((ifrom + idelta) > inum) ? inum : ifrom + idelta;
#else
const int ifrom = 0;
const int ito = nlocal;
const int tid = 0;
#endif
ThrData *thr = fix->get_thr(tid);
double * const * const f = thr->get_f();
FFT_SCALAR * const * const r1d = static_cast<FFT_SCALAR **>(thr->get_rho1d());
int l,m,n,nx,ny,nz,mx,my,mz;
FFT_SCALAR dx,dy,dz,x0,y0,z0;
FFT_SCALAR ekx,eky,ekz;
int iH1,iH2;
double xM[3], fx,fy,fz;
double ddotf, rOMx, rOMy, rOMz, f1x, f1y, f1z;
// this if protects against having more threads than local atoms
if (ifrom < nlocal) {
for (int i = ifrom; i < ito; i++) {
if (type[i] == typeO) {
find_M(i,iH1,iH2,xM);
} else {
xM[0] = x[i][0];
xM[1] = x[i][1];
xM[2] = x[i][2];
}
nx = part2grid[i][0];
ny = part2grid[i][1];
nz = part2grid[i][2];
dx = nx+shiftone - (xM[0]-boxlo[0])*delxinv;
dy = ny+shiftone - (xM[1]-boxlo[1])*delyinv;
dz = nz+shiftone - (xM[2]-boxlo[2])*delzinv;
compute_rho1d_thr(r1d,dx,dy,dz);
ekx = eky = ekz = ZEROF;
for (n = nlower; n <= nupper; n++) {
mz = n+nz;
z0 = r1d[2][n];
for (m = nlower; m <= nupper; m++) {
my = m+ny;
y0 = z0*r1d[1][m];
for (l = nlower; l <= nupper; l++) {
mx = l+nx;
x0 = y0*r1d[0][l];
ekx -= x0*vdx_brick[mz][my][mx];
eky -= x0*vdy_brick[mz][my][mx];
ekz -= x0*vdz_brick[mz][my][mx];
}
}
}
// convert E-field to force
const double qfactor = qqrd2e*scale*q[i];
if (type[i] != typeO) {
f[i][0] += qfactor*ekx;
f[i][1] += qfactor*eky;
f[i][2] += qfactor*ekz;
} else {
fx = qfactor * ekx;
fy = qfactor * eky;
fz = qfactor * ekz;
find_M(i,iH1,iH2,xM);
rOMx = xM[0] - x[i][0];
rOMy = xM[1] - x[i][1];
rOMz = xM[2] - x[i][2];
ddotf = (rOMx * fx + rOMy * fy + rOMz * fz) / (qdist * qdist);
f1x = ddotf * rOMx;
f1y = ddotf * rOMy;
f1z = ddotf * rOMz;
f[i][0] += fx - alpha * (fx - f1x);
f[i][1] += fy - alpha * (fy - f1y);
f[i][2] += fz - alpha * (fz - f1z);
f[iH1][0] += 0.5*alpha*(fx - f1x);
f[iH1][1] += 0.5*alpha*(fy - f1y);
f[iH1][2] += 0.5*alpha*(fz - f1z);
f[iH2][0] += 0.5*alpha*(fx - f1x);
f[iH2][1] += 0.5*alpha*(fy - f1y);
f[iH2][2] += 0.5*alpha*(fz - f1z);
}
}
}
}
}
void PPPMTIP4POMP::make_rho()
{
const double * const q = atom->q;
const double * const * const x = atom->x;
const int * const type = atom->type;
const int nthreads = comm->nthreads;
const int nlocal = atom->nlocal;
#if defined(_OPENMP)
#pragma omp parallel default(none)
#endif
{
#if defined(_OPENMP)
// each thread works on a fixed chunk of atoms.
const int tid = omp_get_thread_num();
const int inum = nlocal;
const int idelta = 1 + inum/nthreads;
const int ifrom = tid*idelta;
const int ito = ((ifrom + idelta) > inum) ? inum : ifrom + idelta;
#else
const int tid = 0;
const int ifrom = 0;
const int ito = nlocal;
#endif
int l,m,n,nx,ny,nz,mx,my,mz,iH1,iH2;
FFT_SCALAR dx,dy,dz,x0,y0,z0;
double xM[3];
// set up clear 3d density array
const int nzoffs = (nzhi_out-nzlo_out+1)*tid;
FFT_SCALAR * const * const * const db = &(density_brick[nzoffs]);
memset(&(db[nzlo_out][nylo_out][nxlo_out]),0,ngrid*sizeof(FFT_SCALAR));
ThrData *thr = fix->get_thr(tid);
FFT_SCALAR * const * const r1d = static_cast<FFT_SCALAR **>(thr->get_rho1d());
// loop over my charges, add their contribution to nearby grid points
// (nx,ny,nz) = global coords of grid pt to "lower left" of charge
// (dx,dy,dz) = distance to "lower left" grid pt
// (mx,my,mz) = global coords of moving stencil pt
// this if protects against having more threads than local atoms
if (ifrom < nlocal) {
for (int i = ifrom; i < ito; i++) {
if (type[i] == typeO) {
find_M(i,iH1,iH2,xM);
} else {
xM[0] = x[i][0];
xM[1] = x[i][1];
xM[2] = x[i][2];
}
nx = part2grid[i][0];
ny = part2grid[i][1];
nz = part2grid[i][2];
dx = nx+shiftone - (xM[0]-boxlo[0])*delxinv;
dy = ny+shiftone - (xM[1]-boxlo[1])*delyinv;
dz = nz+shiftone - (xM[2]-boxlo[2])*delzinv;
compute_rho1d_thr(r1d,dx,dy,dz);
z0 = delvolinv * q[i];
for (n = nlower; n <= nupper; n++) {
mz = n+nz;
y0 = z0*r1d[2][n];
for (m = nlower; m <= nupper; m++) {
my = m+ny;
x0 = y0*r1d[1][m];
for (l = nlower; l <= nupper; l++) {
mx = l+nx;
db[mz][my][mx] += x0*r1d[0][l];
}
}
}
}
}
#if defined(_OPENMP)
// reduce 3d density array
if (nthreads > 1) {
data_reduce_fft(&(density_brick[nzlo_out][nylo_out][nxlo_out]),ngrid,nthreads,1,tid);
}
#endif
}
}