Homec4science
Diffusion lammps (master) Browse

lammps/src/USER-OMP/pppm_cg_omp.cpp
9a9291036d88master

Last Change
Raw File

pppm_cg_omp.cpp
View Options

/* ----------------------------------------------------------------------
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
http://lammps.sandia.gov, Sandia National Laboratories
Steve Plimpton, sjplimp@sandia.gov
Copyright (2003) Sandia Corporation. Under the terms of Contract
DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
certain rights in this software. This software is distributed under
the GNU General Public License.
See the README file in the top-level LAMMPS directory.
------------------------------------------------------------------------- */
/* ----------------------------------------------------------------------
Contributing author: Axel Kohlmeyer (Temple U)
------------------------------------------------------------------------- */
#include "pppm_cg_omp.h"
#include "atom.h"
#include "comm.h"
#include "domain.h"
#include "force.h"
#include "memory.h"
#include "math_const.h"
#include <string.h>
#include <math.h>
#include "suffix.h"
using namespace LAMMPS_NS;
using namespace MathConst;
#define SMALLQ 0.00001
#if defined(FFT_SINGLE)
#define ZEROF 0.0f
#else
#define ZEROF 0.0
#endif
#define EPS_HOC 1.0e-7
/* ---------------------------------------------------------------------- */
PPPMCGOMP::PPPMCGOMP(LAMMPS *lmp, int narg, char **arg) :
PPPMCG(lmp, narg, arg), ThrOMP(lmp, THR_KSPACE)
{
suffix_flag |= Suffix::OMP;
}
/* ----------------------------------------------------------------------
allocate memory that depends on # of K-vectors and order
------------------------------------------------------------------------- */
void PPPMCGOMP::allocate()
{
PPPMCG::allocate();
const int nthreads = comm->nthreads;
#if defined(_OPENMP)
#pragma omp parallel default(none)
#endif
{
#if defined(_OPENMP)
const int tid = omp_get_thread_num();
#else
const int tid = 0;
#endif
FFT_SCALAR **rho1d_thr;
memory->create2d_offset(rho1d_thr,3,-order/2,order/2,"pppm:rho1d_thr");
ThrData *thr = fix->get_thr(tid);
thr->init_pppm(static_cast<void *>(rho1d_thr));
}
const int nzend = (nzhi_out-nzlo_out+1)*nthreads + nzlo_out -1;
// reallocate density brick, so it fits our needs
memory->destroy3d_offset(density_brick,nzlo_out,nylo_out,nxlo_out);
memory->create3d_offset(density_brick,nzlo_out,nzend,nylo_out,nyhi_out,
nxlo_out,nxhi_out,"pppm:density_brick");
}
// NOTE: special version of reduce_data for FFT_SCALAR data type.
// reduce per thread data into the first part of the data
// array that is used for the non-threaded parts and reset
// the temporary storage to 0.0. this routine depends on
// multi-dimensional arrays like force stored in this order
// x1,y1,z1,x2,y2,z2,...
// we need to post a barrier to wait until all threads are done
// writing to the array.
static void data_reduce_fft(FFT_SCALAR *dall, int nall, int nthreads, int ndim, int tid)
{
#if defined(_OPENMP)
// NOOP in non-threaded execution.
if (nthreads == 1) return;
#pragma omp barrier
{
const int nvals = ndim*nall;
const int idelta = nvals/nthreads + 1;
const int ifrom = tid*idelta;
const int ito = ((ifrom + idelta) > nvals) ? nvals : (ifrom + idelta);
// this if protects against having more threads than atoms
if (ifrom < nall) {
for (int m = ifrom; m < ito; ++m) {
for (int n = 1; n < nthreads; ++n) {
dall[m] += dall[n*nvals + m];
dall[n*nvals + m] = 0.0;
}
}
}
}
#else
// NOOP in non-threaded execution.
return;
#endif
}
/* ----------------------------------------------------------------------
free memory that depends on # of K-vectors and order
------------------------------------------------------------------------- */
void PPPMCGOMP::deallocate()
{
PPPMCG::deallocate();
for (int i=0; i < comm->nthreads; ++i) {
ThrData * thr = fix->get_thr(i);
FFT_SCALAR ** rho1d_thr = static_cast<FFT_SCALAR **>(thr->get_rho1d());
memory->destroy2d_offset(rho1d_thr,-order/2);
}
}
/* ----------------------------------------------------------------------
adjust PPPMCG coeffs, called initially and whenever volume has changed
------------------------------------------------------------------------- */
void PPPMCGOMP::setup()
{
int i,j,k,n;
double *prd;
// volume-dependent factors
// adjust z dimension for 2d slab PPPM
// z dimension for 3d PPPM is zprd since slab_volfactor = 1.0
if (triclinic == 0) prd = domain->prd;
else prd = domain->prd_lamda;
const double xprd = prd[0];
const double yprd = prd[1];
const double zprd = prd[2];
const double zprd_slab = zprd*slab_volfactor;
volume = xprd * yprd * zprd_slab;
delxinv = nx_pppm/xprd;
delyinv = ny_pppm/yprd;
delzinv = nz_pppm/zprd_slab;
delvolinv = delxinv*delyinv*delzinv;
const double unitkx = (2.0*MY_PI/xprd);
const double unitky = (2.0*MY_PI/yprd);
const double unitkz = (2.0*MY_PI/zprd_slab);
// fkx,fky,fkz for my FFT grid pts
double per;
for (i = nxlo_fft; i <= nxhi_fft; i++) {
per = i - nx_pppm*(2*i/nx_pppm);
fkx[i] = unitkx*per;
}
for (i = nylo_fft; i <= nyhi_fft; i++) {
per = i - ny_pppm*(2*i/ny_pppm);
fky[i] = unitky*per;
}
for (i = nzlo_fft; i <= nzhi_fft; i++) {
per = i - nz_pppm*(2*i/nz_pppm);
fkz[i] = unitkz*per;
}
// virial coefficients
double sqk,vterm;
n = 0;
for (k = nzlo_fft; k <= nzhi_fft; k++) {
for (j = nylo_fft; j <= nyhi_fft; j++) {
for (i = nxlo_fft; i <= nxhi_fft; i++) {
sqk = fkx[i]*fkx[i] + fky[j]*fky[j] + fkz[k]*fkz[k];
if (sqk == 0.0) {
vg[n][0] = 0.0;
vg[n][1] = 0.0;
vg[n][2] = 0.0;
vg[n][3] = 0.0;
vg[n][4] = 0.0;
vg[n][5] = 0.0;
} else {
vterm = -2.0 * (1.0/sqk + 0.25/(g_ewald*g_ewald));
vg[n][0] = 1.0 + vterm*fkx[i]*fkx[i];
vg[n][1] = 1.0 + vterm*fky[j]*fky[j];
vg[n][2] = 1.0 + vterm*fkz[k]*fkz[k];
vg[n][3] = vterm*fkx[i]*fky[j];
vg[n][4] = vterm*fkx[i]*fkz[k];
vg[n][5] = vterm*fky[j]*fkz[k];
}
n++;
}
}
}
// modified (Hockney-Eastwood) Coulomb Green's function
const int nbx = static_cast<int> ((g_ewald*xprd/(MY_PI*nx_pppm)) *
pow(-log(EPS_HOC),0.25));
const int nby = static_cast<int> ((g_ewald*yprd/(MY_PI*ny_pppm)) *
pow(-log(EPS_HOC),0.25));
const int nbz = static_cast<int> ((g_ewald*zprd_slab/(MY_PI*nz_pppm)) *
pow(-log(EPS_HOC),0.25));
const double form = 1.0;
#if defined(_OPENMP)
#pragma omp parallel default(none)
#endif
{
int tid,nn,nnfrom,nnto,nx,ny,nz,k,l,m;
double snx,sny,snz,sqk;
double argx,argy,argz,wx,wy,wz,sx,sy,sz,qx,qy,qz;
double sum1,dot1,dot2;
double numerator,denominator;
const double gew2 = -4.0*g_ewald*g_ewald;
const int nnx = nxhi_fft-nxlo_fft+1;
const int nny = nyhi_fft-nylo_fft+1;
loop_setup_thr(nnfrom, nnto, tid, nfft, comm->nthreads);
for (m = nzlo_fft; m <= nzhi_fft; m++) {
const double fkzm = fkz[m];
snz = sin(0.5*fkzm*zprd_slab/nz_pppm);
snz *= snz;
for (l = nylo_fft; l <= nyhi_fft; l++) {
const double fkyl = fky[l];
sny = sin(0.5*fkyl*yprd/ny_pppm);
sny *= sny;
for (k = nxlo_fft; k <= nxhi_fft; k++) {
/* only compute the part designated to this thread */
nn = k-nxlo_fft + nnx*(l-nylo_fft + nny*(m-nzlo_fft));
if ((nn < nnfrom) || (nn >=nnto)) continue;
const double fkxk = fkx[k];
snx = sin(0.5*fkxk*xprd/nx_pppm);
snx *= snx;
sqk = fkxk*fkxk + fkyl*fkyl + fkzm*fkzm;
if (sqk != 0.0) {
numerator = form*MY_4PI/sqk;
denominator = gf_denom(snx,sny,snz);
const double dorder = static_cast<double>(order);
sum1 = 0.0;
for (nx = -nbx; nx <= nbx; nx++) {
qx = fkxk + unitkx*nx_pppm*nx;
sx = exp(qx*qx/gew2);
wx = 1.0;
argx = 0.5*qx*xprd/nx_pppm;
if (argx != 0.0) wx = pow(sin(argx)/argx,dorder);
wx *=wx;
for (ny = -nby; ny <= nby; ny++) {
qy = fkyl + unitky*ny_pppm*ny;
sy = exp(qy*qy/gew2);
wy = 1.0;
argy = 0.5*qy*yprd/ny_pppm;
if (argy != 0.0) wy = pow(sin(argy)/argy,dorder);
wy *= wy;
for (nz = -nbz; nz <= nbz; nz++) {
qz = fkzm + unitkz*nz_pppm*nz;
sz = exp(qz*qz/gew2);
wz = 1.0;
argz = 0.5*qz*zprd_slab/nz_pppm;
if (argz != 0.0) wz = pow(sin(argz)/argz,dorder);
wz *= wz;
dot1 = fkxk*qx + fkyl*qy + fkzm*qz;
dot2 = qx*qx+qy*qy+qz*qz;
sum1 += (dot1/dot2) * sx*sy*sz * wx*wy*wz;
}
}
}
greensfn[nn] = numerator*sum1/denominator;
} else greensfn[nn] = 0.0;
}
}
}
}
}
/* ----------------------------------------------------------------------
run the regular toplevel compute method from plain PPPPMCG
which will have individual methods replaced by our threaded
versions and then call the obligatory force reduction.
------------------------------------------------------------------------- */
void PPPMCGOMP::compute(int eflag, int vflag)
{
PPPMCG::compute(eflag,vflag);
#if defined(_OPENMP)
#pragma omp parallel default(none) shared(eflag,vflag)
#endif
{
#if defined(_OPENMP)
const int tid = omp_get_thread_num();
#else
const int tid = 0;
#endif
ThrData *thr = fix->get_thr(tid);
reduce_thr(this, eflag, vflag, thr);
} // end of omp parallel region
}
/* ----------------------------------------------------------------------
create discretized "density" on section of global grid due to my particles
density(x,y,z) = charge "density" at grid points of my 3d brick
(nxlo:nxhi,nylo:nyhi,nzlo:nzhi) is extent of my brick (including ghosts)
in global grid
------------------------------------------------------------------------- */
void PPPMCGOMP::make_rho()
{
const double * const q = atom->q;
const double * const * const x = atom->x;
const int nthreads = comm->nthreads;
#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 = num_charged;
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 = num_charged;
#endif
int i,j,l,m,n,nx,ny,nz,mx,my,mz;
FFT_SCALAR dx,dy,dz,x0,y0,z0;
// 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 charged local atoms
if (ifrom < num_charged) {
for (j = ifrom; j < ito; j++) {
i = is_charged[j];
nx = part2grid[i][0];
ny = part2grid[i][1];
nz = part2grid[i][2];
dx = nx+shiftone - (x[i][0]-boxlo[0])*delxinv;
dy = ny+shiftone - (x[i][1]-boxlo[1])*delyinv;
dz = nz+shiftone - (x[i][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
}
}
/* ----------------------------------------------------------------------
interpolate from grid to get electric field & force on my particles
------------------------------------------------------------------------- */
void PPPMCGOMP::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 nthreads = comm->nthreads;
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 = num_charged;
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 = num_charged;
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 i,j,l,m,n,nx,ny,nz,mx,my,mz;
FFT_SCALAR dx,dy,dz,x0,y0,z0;
FFT_SCALAR ekx,eky,ekz;
// this if protects against having more threads than charged local atoms
if (ifrom < num_charged) {
for (j = ifrom; j < ito; j++) {
i = is_charged[j];
nx = part2grid[i][0];
ny = part2grid[i][1];
nz = part2grid[i][2];
dx = nx+shiftone - (x[i][0]-boxlo[0])*delxinv;
dy = ny+shiftone - (x[i][1]-boxlo[1])*delyinv;
dz = nz+shiftone - (x[i][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];
f[i][0] += qfactor*ekx;
f[i][1] += qfactor*eky;
f[i][2] += qfactor*ekz;
}
}
}
}
/* ----------------------------------------------------------------------
interpolate from grid to get per-atom energy/virial
------------------------------------------------------------------------- */
void PPPMCGOMP::fieldforce_peratom()
{
// loop over my charges, interpolate 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
const double * const q = atom->q;
const double * const * const x = atom->x;
const int nthreads = comm->nthreads;
#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 = num_charged;
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 = num_charged;
const int tid = 0;
#endif
ThrData *thr = fix->get_thr(tid);
FFT_SCALAR * const * const r1d = static_cast<FFT_SCALAR **>(thr->get_rho1d());
int i,j,l,m,n,nx,ny,nz,mx,my,mz;
FFT_SCALAR dx,dy,dz,x0,y0,z0;
FFT_SCALAR u,v0,v1,v2,v3,v4,v5;
// this if protects against having more threads than charged local atoms
if (ifrom < num_charged) {
for (j = ifrom; j < ito; j++) {
i = is_charged[j];
nx = part2grid[i][0];
ny = part2grid[i][1];
nz = part2grid[i][2];
dx = nx+shiftone - (x[i][0]-boxlo[0])*delxinv;
dy = ny+shiftone - (x[i][1]-boxlo[1])*delyinv;
dz = nz+shiftone - (x[i][2]-boxlo[2])*delzinv;
compute_rho1d_thr(r1d,dx,dy,dz);
u = v0 = v1 = v2 = v3 = v4 = v5 = 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];
if (eflag_atom) u += x0*u_brick[mz][my][mx];
if (vflag_atom) {
v0 += x0*v0_brick[mz][my][mx];
v1 += x0*v1_brick[mz][my][mx];
v2 += x0*v2_brick[mz][my][mx];
v3 += x0*v3_brick[mz][my][mx];
v4 += x0*v4_brick[mz][my][mx];
v5 += x0*v5_brick[mz][my][mx];
}
}
}
}
if (eflag_atom) eatom[i] += q[i]*u;
if (vflag_atom) {
vatom[i][0] += v0;
vatom[i][1] += v1;
vatom[i][2] += v2;
vatom[i][3] += v3;
vatom[i][4] += v4;
vatom[i][5] += v5;
}
}
}
}
}
/* ----------------------------------------------------------------------
charge assignment into rho1d
dx,dy,dz = distance of particle from "lower left" grid point
------------------------------------------------------------------------- */
void PPPMCGOMP::compute_rho1d_thr(FFT_SCALAR * const * const r1d, const FFT_SCALAR &dx,
const FFT_SCALAR &dy, const FFT_SCALAR &dz)
{
int k,l;
FFT_SCALAR r1,r2,r3;
for (k = (1-order)/2; k <= order/2; k++) {
r1 = r2 = r3 = ZEROF;
for (l = order-1; l >= 0; l--) {
r1 = rho_coeff[l][k] + r1*dx;
r2 = rho_coeff[l][k] + r2*dy;
r3 = rho_coeff[l][k] + r3*dz;
}
r1d[0][k] = r1;
r1d[1][k] = r2;
r1d[2][k] = r3;
}
}
c4science · Help