SIZE wci_dpcm_get_blocks(
SIZE IN rSize
)
{
SIZE rBlocks;
rBlocks.cx = CELLS( rSize.cx, DPCM_BLOCK_SIZE_X );
rBlocks.cy = CELLS( rSize.cy, DPCM_BLOCK_SIZE_Y );
return rBlocks;
}
hword_t hopc_tagsize(hopc_runtime *r, htag tag) {
switch(tag) {
case HOPCTASKTAG:
return CELLS(sizeof(hopc_task));
}
return r->tagdata[tag].size;
}
void adjust()
{
int k;
for (k=0; k<ncells; ++k) {
cell *p;
int i,typ;
real x,y,z;
p = cell_array + CELLS(k);
for (i=0; i<p->n; ++i) {
x = ORT(p,i,X)-0.1;
y = ORT(p,i,Y)-0.1;
z = ORT(p,i,Z)-0.1;
typ = VSORTE(p,i);
/* fix the type of the first atom */
if (FABS(x+2.*gmin.x) < 0.0001 && FABS(y+2.*gmin.y) < 0.0001 &&
FABS(z+2.*gmin.z) < 0.0001)
{
typ=0;
SORTE (p,i) = typ;
VSORTE(p,i) = typ;
}
num_sort[typ]++;
}
}
}
void calc_forces(int steps)
{
int n, k;
real tmpvec1[8], tmpvec2[8] = {0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0};
/* fill the buffer cells */
if ((steps == steps_min) || (0 == steps % BUFSTEP)) setup_buffers();
send_cells(copy_cell,pack_cell,unpack_cell);
/* clear global accumulation variables */
tot_pot_energy = 0.0;
virial = 0.0;
vir_xx = 0.0;
vir_yy = 0.0;
vir_zz = 0.0;
vir_yz = 0.0;
vir_zx = 0.0;
vir_xy = 0.0;
nfc++;
/* clear per atom accumulation variables */
#ifdef _OPENMP
#pragma omp parallel for
#endif
for (k=0; k<nallcells; ++k) {
int i;
cell *p;
p = cell_array + k;
for (i=0; i<p->n; ++i) {
KRAFT(p,i,X) = 0.0;
KRAFT(p,i,Y) = 0.0;
KRAFT(p,i,Z) = 0.0;
#ifdef UNIAX
DREH_MOMENT(p,i,X) = 0.0;
DREH_MOMENT(p,i,Y) = 0.0;
DREH_MOMENT(p,i,Z) = 0.0;
#endif
#if defined(STRESS_TENS)
PRESSTENS(p,i,xx) = 0.0;
PRESSTENS(p,i,yy) = 0.0;
PRESSTENS(p,i,zz) = 0.0;
PRESSTENS(p,i,yz) = 0.0;
PRESSTENS(p,i,zx) = 0.0;
PRESSTENS(p,i,xy) = 0.0;
#endif
#ifndef MONOLJ
POTENG(p,i) = 0.0;
#endif
#ifdef NNBR
NBANZ(p,i) = 0;
#endif
#ifdef CNA
if (cna)
MARK(p,i) = 0;
#endif
#ifdef COVALENT
NEIGH(p,i)->n = 0;
#endif
#ifdef EAM2
EAM_RHO(p,i) = 0.0; /* zero host electron density at atom site */
#ifdef EEAM
EAM_P(p,i) = 0.0; /* zero host electron density at atom site */
#endif
#endif
}
}
#ifdef RIGID
/* clear total forces */
if ( nsuperatoms>0 )
for(k=0; k<nsuperatoms; k++) {
superforce[k].x = 0.0;
superforce[k].y = 0.0;
superforce[k].z = 0.0;
}
#endif
/* What follows is the standard one-cpu force
loop acting on our local data cells */
/* compute forces for all pairs of cells */
for (n=0; n<nlists; ++n) {
#ifdef _OPENMP
#pragma omp parallel for schedule(runtime) \
reduction(+:tot_pot_energy,virial,vir_xx,vir_yy,vir_zz,vir_yz,vir_zx,vir_xy)
#endif
for (k=0; k<npairs[n]; ++k) {
vektor pbc;
pair *P;
P = pairs[n] + k;
pbc.x = P->ipbc[0]*box_x.x + P->ipbc[1]*box_y.x + P->ipbc[2]*box_z.x;
pbc.y = P->ipbc[0]*box_x.y + P->ipbc[1]*box_y.y + P->ipbc[2]*box_z.y;
pbc.z = P->ipbc[0]*box_x.z + P->ipbc[1]*box_y.z + P->ipbc[2]*box_z.z;
do_forces(cell_array + P->np, cell_array + P->nq, pbc,
&tot_pot_energy, &virial, &vir_xx, &vir_yy, &vir_zz,
&vir_yz, &vir_zx, &vir_xy);
}
}
#ifdef COVALENT
/* complete neighbor tables for remaining pairs of cells */
for (n=0; n<nlists; ++n) {
#ifdef _OPENMP
#pragma omp parallel for schedule(runtime)
#endif
for (k=npairs[n]; k<npairs2[n]; ++k) {
vektor pbc;
pair *P;
P = pairs[n] + k;
pbc.x = P->ipbc[0]*box_x.x + P->ipbc[1]*box_y.x + P->ipbc[2]*box_z.x;
pbc.y = P->ipbc[0]*box_x.y + P->ipbc[1]*box_y.y + P->ipbc[2]*box_z.y;
pbc.z = P->ipbc[0]*box_x.z + P->ipbc[1]*box_y.z + P->ipbc[2]*box_z.z;
do_neightab(cell_array + P->np, cell_array + P->nq, pbc);
}
}
#ifndef CNA
/* second force loop for covalent systems */
/* does not work correctly - different threads may write to same variables
#ifdef _OPENMP
#pragma omp parallel for schedule(runtime) \
reduction(+:tot_pot_energy,virial,vir_xx,vir_yy,vir_zz,vir_yz,vir_zx,vir_xy)
#endif
*/
for (k=0; k<ncells; ++k) {
do_forces2(cell_array + CELLS(k),
&tot_pot_energy, &virial, &vir_xx, &vir_yy, &vir_zz,
&vir_yz, &vir_zx, &vir_xy);
}
#endif
#endif /* COVALENT */
#ifndef AR
/* If we don't use actio=reactio accross the cpus, we have do do
the force loop also on the other half of the neighbours for the
cells on the surface of the CPU */
/* compute forces for remaining pairs of cells */
for (n=0; n<nlists; ++n) {
/* does not work correctly - different threads may write to same variables
#ifdef _OPENMP
#pragma omp parallel for schedule(runtime)
#endif
*/
for (k=npairs[n]; k<npairs2[n]; ++k) {
vektor pbc;
pair *P;
P = pairs[n] + k;
pbc.x = P->ipbc[0]*box_x.x + P->ipbc[1]*box_y.x + P->ipbc[2]*box_z.x;
pbc.y = P->ipbc[0]*box_x.y + P->ipbc[1]*box_y.y + P->ipbc[2]*box_z.y;
pbc.z = P->ipbc[0]*box_x.z + P->ipbc[1]*box_y.z + P->ipbc[2]*box_z.z;
/* potential energy and virial are already complete; */
/* to avoid double counting, we update only the dummy tmpvec2 */
do_forces(cell_array + P->np, cell_array + P->nq, pbc,
tmpvec2, tmpvec2+1, tmpvec2+2, tmpvec2+3, tmpvec2+4,
tmpvec2+5, tmpvec2+6, tmpvec2+7);
}
}
#endif /* not AR */
#ifdef EAM2
#ifdef AR
/* collect host electron density */
send_forces(add_rho,pack_rho,unpack_add_rho);
#endif
/* compute embedding energy and its derivative */
do_embedding_energy();
/* distribute derivative of embedding energy */
send_cells(copy_dF,pack_dF,unpack_dF);
/* second EAM2 loop over all cells pairs */
for (n=0; n<nlists; ++n) {
#ifdef _OPENMP
#pragma omp parallel for schedule(runtime) \
reduction(+:virial,vir_xx,vir_yy,vir_zz,vir_yz,vir_zx,vir_xy)
#endif
for (k=0; k<npairs[n]; ++k) {
vektor pbc;
pair *P;
P = pairs[n]+k;
pbc.x = P->ipbc[0]*box_x.x + P->ipbc[1]*box_y.x + P->ipbc[2]*box_z.x;
pbc.y = P->ipbc[0]*box_x.y + P->ipbc[1]*box_y.y + P->ipbc[2]*box_z.y;
pbc.z = P->ipbc[0]*box_x.z + P->ipbc[1]*box_y.z + P->ipbc[2]*box_z.z;
do_forces_eam2(cell_array + P->np, cell_array + P->nq, pbc,
&virial, &vir_xx, &vir_yy, &vir_zz, &vir_yz, &vir_zx, &vir_xy);
}
}
#ifndef AR
/* If we don't use actio=reactio accross the cpus, we have do do
the force loop also on the other half of the neighbours for the
cells on the surface of the CPU */
/* compute forces for remaining pairs of cells */
for (n=0; n<nlists; ++n) {
#ifdef _OPENMP
#pragma omp parallel for schedule(runtime)
#endif
for (k=npairs[n]; k<npairs2[n]; ++k) {
vektor pbc;
pair *P;
P = pairs[n]+k;
pbc.x = P->ipbc[0]*box_x.x + P->ipbc[1]*box_y.x + P->ipbc[2]*box_z.x;
pbc.y = P->ipbc[0]*box_x.y + P->ipbc[1]*box_y.y + P->ipbc[2]*box_z.y;
pbc.z = P->ipbc[0]*box_x.z + P->ipbc[1]*box_y.z + P->ipbc[2]*box_z.z;
/* potential energy and virial are already complete; */
/* to avoid double counting, we update only the dummy tmpvec2 */
do_forces_eam2(cell_array + P->np, cell_array + P->nq, pbc,
tmpvec2, tmpvec2+1, tmpvec2+2, tmpvec2+3, tmpvec2+4,
tmpvec2+5, tmpvec2+6, tmpvec2+7);
}
}
#endif /* not AR */
#endif /* EAM2 */
/* sum up results of different CPUs */
tmpvec1[0] = tot_pot_energy;
tmpvec1[1] = virial;
tmpvec1[2] = vir_xx;
tmpvec1[3] = vir_yy;
tmpvec1[4] = vir_zz;
tmpvec1[5] = vir_yz;
tmpvec1[6] = vir_zx;
tmpvec1[7] = vir_xy;
MPI_Allreduce( tmpvec1, tmpvec2, 8, REAL, MPI_SUM, cpugrid);
tot_pot_energy = tmpvec2[0];
virial = tmpvec2[1];
vir_xx = tmpvec2[2];
vir_yy = tmpvec2[3];
vir_zz = tmpvec2[4];
vir_yz = tmpvec2[5];
vir_zx = tmpvec2[6];
vir_xy = tmpvec2[7];
#ifdef AR
send_forces(add_forces,pack_forces,unpack_forces);
#endif
}
