/** Initializes the (inverse) grid constant \ref ifcs_p3m_struct::a (\ref
ifcs_p3m_struct::ai) and the cutoff for charge assignment \ref
ifcs_p3m_struct::cao_cut, which has to be done by \ref ifcs_p3m_init_charges
once and by \ref ifcs_p3m_scaleby_box_l whenever the \ref box_l
changed. */
static void ifcs_p3m_prepare_a_ai_cao_cut(ifcs_p3m_data_struct *d) {
P3M_DEBUG(printf(" ifcs_p3m_prepare_a_ai_cao_cut() started... \n"));
for (fcs_int i=0; i<3; i++) {
d->ai[i] = (fcs_float)d->grid[i]/d->box_l[i];
d->a[i] = 1.0/d->ai[i];
d->cao_cut[i] = 0.5*d->a[i]*d->cao;
}
P3M_DEBUG(printf(" ifcs_p3m_prepare_a_ai_cao_cut() finished. \n"));
}
/** Prepare the data structures and constants of the P3M algorithm.
All parameters have to be set. */
void ifcs_p3m_prepare(ifcs_p3m_data_struct *d, fcs_int max_charges) {
P3M_DEBUG(printf(" ifcs_p3m_prepare() started... \n"));
P3M_DEBUG(printf(" system parameters: box_l=(%" FCS_LMOD_FLOAT "f, %" FCS_LMOD_FLOAT "f, %" FCS_LMOD_FLOAT "f)\n", \
d->box_l[0], d->box_l[1], d->box_l[2]));
P3M_DEBUG(printf( \
" p3m params: r_cut=%" FCS_LMOD_FLOAT "f, grid=%d, cao=%d, alpha=%" FCS_LMOD_FLOAT "f" \
", grid_off=(%" FCS_LMOD_FLOAT "f,%" FCS_LMOD_FLOAT "f,%" FCS_LMOD_FLOAT "f)\n", \
d->r_cut, d->grid[0], d->cao, d->alpha, \
d->grid_off[0], d->grid_off[1], d->grid_off[2]));
/* initializes the (inverse) grid constant d->a
(d->ai) and the cutoff for charge assignment
d->cao_cut */
ifcs_p3m_prepare_a_ai_cao_cut(d);
ifcs_p3m_calc_local_ca_grid(d);
ifcs_p3m_calc_send_grid(d);
P3M_DEBUG(ifcs_p3m_print_local_grid(d->local_grid));
P3M_DEBUG(ifcs_p3m_print_send_grid(d->sm));
d->send_grid = (fcs_float *) realloc(d->send_grid, sizeof(fcs_float)*d->sm.max);
d->recv_grid = (fcs_float *) realloc(d->recv_grid, sizeof(fcs_float)*d->sm.max);
P3M_INFO(printf(" Interpolating charge assignement function...\n"));
ifcs_p3m_interpolate_charge_assignment_function(d);
/* position offset for calc. of first gridpoint */
d->pos_shift = (fcs_float)((d->cao-1)/2) - (d->cao%2)/2.0;
P3M_DEBUG(printf(" pos_shift = %" FCS_LMOD_FLOAT "f\n",d->pos_shift));
/* FFT */
P3M_INFO(printf(" Preparing FFTs...\n"));
ifcs_fft_prepare(&d->fft, &d->comm,
&d->rs_grid, &d->ks_grid,
d->local_grid.dim,d->local_grid.margin,
d->grid, d->grid_off,
&d->ks_pnum);
/* k-space part */
ifcs_p3m_calc_differential_operator(d);
P3M_INFO(printf(" Calculating influence function...\n"));
#if !defined(P3M_INTERLACE) && defined(P3M_IK)
ifcs_p3m_calc_influence_function_ik(d);
#elif defined(P3M_INTERLACE) && defined(P3M_IK)
ifcs_p3m_calc_influence_function_iki(d);
#else
ifcs_p3m_calc_influence_function_adi(d);
#endif
P3M_DEBUG(printf(" ifcs_p3m_prepare() finished.\n"));
}
/** Calculates the properties of the send/recv sub-grides of the local
* FFT grid. In order to calculate the recv sub-grides there is a
* communication of the margins between neighbouring nodes. */
static void ifcs_p3m_calc_send_grid(ifcs_p3m_data_struct *d) {
fcs_int i,j, evenodd;
fcs_int done[3]={0,0,0};
MPI_Status status;
P3M_DEBUG(printf(" ifcs_p3m_calc_send_grid() started... \n"));
/* send grids */
for (i=0; i<3; i++) {
for (j=0; j<3; j++) {
/* left */
d->sm.s_ld[i*2][j] = 0 + done[j]*d->local_grid.margin[j*2];
if (j==i) d->sm.s_ur[i*2][j] = d->local_grid.margin[j*2];
else d->sm.s_ur[i*2][j] = d->local_grid.dim[j]-done[j]*d->local_grid.margin[(j*2)+1];
/* right */
if (j==i) d->sm.s_ld[(i*2)+1][j] = d->local_grid.in_ur[j];
else d->sm.s_ld[(i*2)+1][j] = 0 + done[j]*d->local_grid.margin[j*2];
d->sm.s_ur[(i*2)+1][j] = d->local_grid.dim[j] - done[j]*d->local_grid.margin[(j*2)+1];
}
done[i]=1;
}
d->sm.max=0;
for (i=0; i<6; i++) {
d->sm.s_size[i] = 1;
for (j=0; j<3; j++) {
d->sm.s_dim[i][j] = d->sm.s_ur[i][j]-d->sm.s_ld[i][j];
d->sm.s_size[i] *= d->sm.s_dim[i][j];
}
if (d->sm.s_size[i]>d->sm.max) d->sm.max=d->sm.s_size[i];
}
/* communication */
for (i=0; i<6; i++) {
if (i%2 == 0) j = i+1;
else j = i-1;
if (d->comm.node_neighbors[i] != d->comm.rank) {
/* two step communication: first all even positions than all odd */
for (evenodd=0; evenodd<2; evenodd++) {
if ((d->comm.node_pos[i/2]+evenodd)%2 == 0) {
P3M_DEBUG(printf(" %d: sending local_grid.margin to %d\n", \
d->comm.rank, d->comm.node_neighbors[i]));
MPI_Send(&(d->local_grid.margin[i]), 1, FCS_MPI_INT,
d->comm.node_neighbors[i], 0, d->comm.mpicomm);
} else {
P3M_DEBUG(printf(" %d: receiving local_grid.margin from %d\n", \
d->comm.rank, d->comm.node_neighbors[j]));
MPI_Recv(&(d->local_grid.r_margin[j]), 1, FCS_MPI_INT,
d->comm.node_neighbors[j], 0, d->comm.mpicomm, &status);
}
}
} else {
d->local_grid.r_margin[j] = d->local_grid.margin[i];
}
}
/* /\* communication *\/ */
/* for (i = 0; i < 3; i++) { */
/* /\* upshift *\/ */
/* MPI_Sendrecv(&(d->local_grid.margin[2*i]), 1, FCS_MPI_INT, */
/* d->comm.node_neighbors[2*i+1], 0, */
/* &(d->local_grid.r_margin[2*i]), 1, FCS_MPI_INT, */
/* d->comm.node_neighbors[2*i], 0, */
/* d->comm.mpicomm, &status); */
/* /\* downshift *\/ */
/* MPI_Sendrecv(&(d->local_grid.margin[2*i+1]), 1, FCS_MPI_INT, */
/* d->comm.node_neighbors[2*i], 0, */
/* &(d->local_grid.r_margin[2*i+1]), 1, FCS_MPI_INT, */
/* d->comm.node_neighbors[2*i+1], 0, */
/* d->comm.mpicomm, &status); */
/* } */
/* recv grids */
for (i=0; i<3; i++)
for (j=0; j<3; j++) {
if (j==i) {
d->sm.r_ld[ i*2 ][j] = d->sm.s_ld[ i*2 ][j] + d->local_grid.margin[2*j];
d->sm.r_ur[ i*2 ][j] = d->sm.s_ur[ i*2 ][j] + d->local_grid.r_margin[2*j];
d->sm.r_ld[(i*2)+1][j] = d->sm.s_ld[(i*2)+1][j] - d->local_grid.r_margin[(2*j)+1];
d->sm.r_ur[(i*2)+1][j] = d->sm.s_ur[(i*2)+1][j] - d->local_grid.margin[(2*j)+1];
} else {
d->sm.r_ld[ i*2 ][j] = d->sm.s_ld[ i*2 ][j];
d->sm.r_ur[ i*2 ][j] = d->sm.s_ur[ i*2 ][j];
d->sm.r_ld[(i*2)+1][j] = d->sm.s_ld[(i*2)+1][j];
d->sm.r_ur[(i*2)+1][j] = d->sm.s_ur[(i*2)+1][j];
}
}
for (i=0; i<6; i++) {
d->sm.r_size[i] = 1;
for (j=0;j<3;j++) {
d->sm.r_dim[i][j] = d->sm.r_ur[i][j]-d->sm.r_ld[i][j];
d->sm.r_size[i] *= d->sm.r_dim[i][j];
}
if (d->sm.r_size[i]>d->sm.max) d->sm.max=d->sm.r_size[i];
}
P3M_DEBUG(printf(" ifcs_p3m_calc_send_grid() finished. \n"));
}
/** Calculates properties of the local FFT grid for the
charge assignment process. */
static void ifcs_p3m_calc_local_ca_grid(ifcs_p3m_data_struct *d) {
fcs_int i;
fcs_int ind[3];
/* total skin size */
fcs_float full_skin[3];
P3M_DEBUG(printf(" ifcs_p3m_calc_local_ca_grid() started... \n"));
for(i=0;i<3;i++)
full_skin[i]= d->cao_cut[i]+d->skin+d->additional_grid[i];
/* inner left down grid point (global index) */
for(i=0;i<3;i++)
d->local_grid.in_ld[i] =
(fcs_int)ceil(d->comm.my_left[i]*d->ai[i]-d->grid_off[i]);
/* inner up right grid point (global index) */
for(i=0;i<3;i++)
d->local_grid.in_ur[i] =
(fcs_int)floor(d->comm.my_right[i]*d->ai[i]-d->grid_off[i]);
/* correct roundof errors at boundary */
for(i=0;i<3;i++) {
if (fcs_float_is_zero((d->comm.my_right[i] * d->ai[i] - d->grid_off[i])
- d->local_grid.in_ur[i]))
d->local_grid.in_ur[i]--;
if (fcs_float_is_zero(1.0+(d->comm.my_left[i] * d->ai[i] - d->grid_off[i])
- d->local_grid.in_ld[i]))
d->local_grid.in_ld[i]--;
}
/* inner grid dimensions */
for(i=0; i<3; i++)
d->local_grid.inner[i] = d->local_grid.in_ur[i] - d->local_grid.in_ld[i] + 1;
/* index of left down grid point in global grid */
for(i=0; i<3; i++)
d->local_grid.ld_ind[i] =
(fcs_int)ceil((d->comm.my_left[i]-full_skin[i])*d->ai[i]-d->grid_off[i]);
/* spatial position of left down grid point */
ifcs_p3m_calc_lm_ld_pos(d);
/* left down margin */
for(i=0;i<3;i++)
d->local_grid.margin[i*2] = d->local_grid.in_ld[i]-d->local_grid.ld_ind[i];
/* up right grid point */
for(i=0;i<3;i++)
ind[i] =
(fcs_int)floor((d->comm.my_right[i]+full_skin[i])*d->ai[i]-d->grid_off[i]);
/* correct roundof errors at up right boundary */
for(i=0;i<3;i++)
if (((d->comm.my_right[i]+full_skin[i])*d->ai[i]-d->grid_off[i])-ind[i]==0)
ind[i]--;
/* up right margin */
for(i=0;i<3;i++) d->local_grid.margin[(i*2)+1] = ind[i] - d->local_grid.in_ur[i];
/* grid dimension */
d->local_grid.size=1;
for(i=0;i<3;i++) {
d->local_grid.dim[i] = ind[i] - d->local_grid.ld_ind[i] + 1;
d->local_grid.size *= d->local_grid.dim[i];
}
/* reduce inner grid indices from global to local */
for(i=0;i<3;i++)
d->local_grid.in_ld[i] = d->local_grid.margin[i*2];
for(i=0;i<3;i++)
d->local_grid.in_ur[i] = d->local_grid.margin[i*2]+d->local_grid.inner[i];
d->local_grid.q_2_off = d->local_grid.dim[2] - d->cao;
d->local_grid.q_21_off = d->local_grid.dim[2] * (d->local_grid.dim[1] - d->cao);
P3M_DEBUG(printf(" ifcs_p3m_calc_local_ca_grid() finished. \n"));
}
void Communication::prepare(p3m_float box_l[3]) {
P3M_DEBUG(printf( " P3M::Communication::prepare() started...\n"));
/* Test whether the communicator is cartesian and correct dimensionality */
bool comm_is_cart = false;
int status;
MPI_Topo_test(mpicomm_orig, &status);
if (status == MPI_CART) {
/* Communicator is cartesian, so test dimensionality */
int ndims;
MPI_Cartdim_get(mpicomm_orig, &ndims);
if (ndims == 3) {
/* Correct dimensionality, so get grid and test periodicity */
int periodicity[3];
MPI_Cart_get(mpicomm_orig, 3, node_grid,
periodicity, node_pos);
if (periodicity[0] && periodicity[1] && periodicity[2]) {
/* If periodicity is correct, we can just use this communicator */
mpicomm = mpicomm_orig;
/* get the rank */
MPI_Comm_rank(mpicomm, &rank);
comm_is_cart = true;
}
}
}
/* otherwise, we have to set up the cartesian communicator */
if (!comm_is_cart) {
P3M_DEBUG(printf( " Setting up cartesian communicator...\n"));
node_grid[0] = 0;
node_grid[1] = 0;
node_grid[2] = 0;
/* compute node grid */
MPI_Dims_create(size, 3, node_grid);
#ifdef P3M_ENABLE_INFO
if (onMaster())
printf(" node_grid=%dx%dx%d\n", node_grid[0], node_grid[1], node_grid[2]);
#endif
/* create communicator */
int periodicity[3] = {1, 1, 1};
MPI_Cart_create(mpicomm_orig, 3, node_grid,
periodicity, 1, &mpicomm);
/* get the rank */
MPI_Comm_rank(mpicomm, &rank);
/* get node pos */
MPI_Cart_coords(mpicomm, rank, 3, node_pos);
}
/* fetch neighborhood info */
for (int dir = 0; dir < 3; dir++) {
MPI_Cart_shift(mpicomm, dir, 1,
&node_neighbors[2*dir],
&node_neighbors[2*dir+1]);
P3M_DEBUG_LOCAL(printf( " %d: dir=%d: n1=%d n2=%d\n", rank, dir, \
node_neighbors[2*dir], \
node_neighbors[2*dir+1]));
}
/* init local points */
for (int i=0; i< 3; i++) {
local_box_l[i] = 0.0;
my_left[i] = 0.0;
my_right[i] = 0.0;
}
/* compute box limits */
for(p3m_int i = 0; i < 3; i++) {
local_box_l[i] = box_l[i]/(p3m_float)node_grid[i];
my_left[i] = node_pos[i] *local_box_l[i];
my_right[i] = (node_pos[i]+1)*local_box_l[i];
}
P3M_DEBUG(printf(" local_box_l=" F3FLOAT "\n" \
" my_left=" F3FLOAT "\n" \
" my_right=" F3FLOAT "\n", \
local_box_l[0], \
local_box_l[1], \
local_box_l[2], \
my_left[0], my_left[1], my_left[2], \
my_right[0], my_right[1], my_right[2] \
));
P3M_DEBUG(printf(" P3M::Communication::prepare() finished.\n"));
}
