/** Init cell interactions for cell system domain decomposition.
* initializes the interacting neighbor cell list of a cell The
* created list of interacting neighbor cells is used by the verlet
* algorithm (see verlet.c) to build the verlet lists.
*/
void dd_init_cell_interactions()
{
int m,n,o,p,q,r,ind1,ind2,c_cnt=0,n_cnt;
/* initialize cell neighbor structures */
dd.cell_inter = (IA_Neighbor_List *) realloc(dd.cell_inter,local_cells.n*sizeof(IA_Neighbor_List));
for(m=0; m<local_cells.n; m++) {
dd.cell_inter[m].nList = NULL;
dd.cell_inter[m].n_neighbors=0;
}
/* loop all local cells */
DD_LOCAL_CELLS_LOOP(m,n,o) {
dd.cell_inter[c_cnt].nList = (IA_Neighbor *) realloc(dd.cell_inter[c_cnt].nList, CELLS_MAX_NEIGHBORS*sizeof(IA_Neighbor));
dd.cell_inter[c_cnt].n_neighbors = CELLS_MAX_NEIGHBORS;
n_cnt=0;
ind1 = get_linear_index(m,n,o,dd.ghost_cell_grid);
/* loop all neighbor cells */
for(p=o-1; p<=o+1; p++)
for(q=n-1; q<=n+1; q++)
for(r=m-1; r<=m+1; r++) {
ind2 = get_linear_index(r,q,p,dd.ghost_cell_grid);
if(ind2 >= ind1) {
dd.cell_inter[c_cnt].nList[n_cnt].cell_ind = ind2;
dd.cell_inter[c_cnt].nList[n_cnt].pList = &cells[ind2];
init_pairList(&dd.cell_inter[c_cnt].nList[n_cnt].vList);
n_cnt++;
}
}
c_cnt++;
}
void Lattice::get_data_for_local_index(index_t* ind, void** data) {
index_t index_in_halogrid[3];
index_in_halogrid[0] = ind[0]+this->halo_size;
index_in_halogrid[1] = ind[1]+this->halo_size;
index_in_halogrid[2] = ind[2]+this->halo_size;
(*data) = ((char*)this->_data) + get_linear_index(index_in_halogrid[0], index_in_halogrid[1], index_in_halogrid[2], this->halo_grid)*this->element_size;
}
/** Calculate temperature of the LB fluid.
* \param result Fluid temperature
*/
void lb_calc_fluid_temp(double *result) {
int x, y, z, index;
double local_rho, local_j2;
double temp = 0.0;
for (x=1; x<=lblattice.grid[0]; x++) {
for (y=1; y<=lblattice.grid[1]; y++) {
for (z=1; z<=lblattice.grid[2]; z++) {
index = get_linear_index(x,y,z,lblattice.halo_grid);
lb_calc_local_j(&lbfluid[index]);
lb_calc_local_rho(&lbfluid[index]);
local_rho = *lbfluid[index].rho;
local_j2 = scalar(lbfluid[index].j,lbfluid[index].j);
temp += local_j2;
}
}
}
temp *= 1./(lbpar.rho*lblattice.grid_volume*lbpar.tau*lbpar.tau*pow(lblattice.agrid,4));
MPI_Reduce(&temp, result, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
}
/** Calculate momentum of the LB fluid.
* \param result Fluid momentum
*/
void lb_calc_fluid_momentum(double *result) {
int x, y, z, index;
double j[3], momentum[3] = { 0.0, 0.0, 0.0 };
for (x=1; x<=lblattice.grid[0]; x++) {
for (y=1; y<=lblattice.grid[1]; y++) {
for (z=1; z<=lblattice.grid[2]; z++) {
index = get_linear_index(x,y,z,lblattice.halo_grid);
lb_calc_local_j(index,j);
momentum[0] += j[0] + lbfields[index].force[0];
momentum[1] += j[1] + lbfields[index].force[1];
momentum[2] += j[2] + lbfields[index].force[2];
}
}
}
momentum[0] *= lblattice.agrid/lbpar.tau;
momentum[1] *= lblattice.agrid/lbpar.tau;
momentum[2] *= lblattice.agrid/lbpar.tau;
MPI_Reduce(momentum, result, 3, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
}
/** Initialize a planar boundary specified by a wall constraint.
* @param plane The \ref Constraint_wall struct describing the boundary.
*/
static void lb_init_constraint_wall(Constraint_wall* plane) {
int x, y, z;
double pos[3], dist;
for (x=0;x<lblattice.halo_grid[0];x++) {
for (y=0;y<lblattice.halo_grid[1];y++) {
for (z=0;z<lblattice.halo_grid[2];z++) {
pos[0] = my_left[0] + (x-1)*lblattice.agrid;
pos[1] = my_left[1] + (y-1)*lblattice.agrid;
pos[2] = my_left[2] + (z-1)*lblattice.agrid;
dist = scalar(pos,plane->n) - plane->d;
if (fabs(dist) < lblattice.agrid) {
//printf("%d %d %d\n",x,y,z);
lbfluid[get_linear_index(x,y,z,lblattice.halo_grid)].boundary = 1;
}
}
}
}
}
int Lattice::global_pos_to_lattice_index_checked(double pos[3], int* index) {
int i;
for (i=0; i<3; i++)
if (fabs(fmod(pos[i]-this->offset[i],this->agrid[i])) > ROUND_ERROR_PREC)
return ES_ERROR;
int ind[3];
for (i=0; i<3; i++)
ind[i] = (int) round((pos[i]-this->offset[i])/this->agrid[i]);
*index = get_linear_index(this->halo_size + ind[0], this->halo_size + ind[1], this->halo_size + ind[2], this->halo_grid);
return ES_OK;
}
//int Lattice::init(double *agrid, double* offset, int halo_size, size_t dim) {
int Lattice::init(double *agrid, double* offset, int halo_size, size_t dim) {
this->dim=dim;
/* determine the number of local lattice nodes */
for (int d=0; d<3; d++) {
this->agrid[d] = agrid[d];
this->global_grid[d] = (int)dround(box_l[d]/agrid[d]);
this->offset[d]=offset[d];
this->local_index_offset[d]=(int) ceil((my_left[d]-this->offset[d])/this->agrid[d]);
this->local_offset[d] = this->offset[d] +
this->local_index_offset[d]*this->agrid[d];
this->grid[d] = (int) ceil ( ( my_right[d] - this->local_offset[d]-ROUND_ERROR_PREC )
/ this->agrid[d]);
}
// sanity checks
for (int dir=0;dir<3;dir++) {
// check if local_box_l is compatible with lattice spacing
if (fabs(local_box_l[dir]-this->grid[dir]*agrid[dir]) > ROUND_ERROR_PREC*box_l[dir]) {
char *errtxt = runtime_error(256);
ERROR_SPRINTF(errtxt, \
"{097 Lattice spacing agrid[%d]=%f " \
"is incompatible with local_box_l[%d]=%f " \
"(box_l[%d]=%f node_grid[%d]=%d)} ", \
dir, agrid[dir], \
dir, local_box_l[dir], \
dir, box_l[dir], \
dir, node_grid[dir]);
}
}
this->element_size = this->dim*sizeof(double);
LATTICE_TRACE(fprintf(stderr,"%d: box_l (%.3f,%.3f,%.3f) grid (%d,%d,%d) node_neighbors (%d,%d,%d,%d,%d,%d)\n",this_node,local_box_l[0],local_box_l[1],local_box_l[2],this->grid[0],this->grid[1],this->grid[2],node_neighbors[0],node_neighbors[1],node_neighbors[2],node_neighbors[3],node_neighbors[4],node_neighbors[5]));
this->halo_size = halo_size;
/* determine the number of total nodes including halo */
this->halo_grid[0] = this->grid[0] + 2*halo_size ;
this->halo_grid[1] = this->grid[1] + 2*halo_size ;
this->halo_grid[2] = this->grid[2] + 2*halo_size ;
this->grid_volume = this->grid[0]*this->grid[1]*this->grid[2] ;
this->halo_grid_volume = this->halo_grid[0]*this->halo_grid[1]*this->halo_grid[2] ;
this->halo_grid_surface = this->halo_grid_volume - this->grid_volume ;
this->halo_offset = get_linear_index(halo_size,halo_size,halo_size,this->halo_grid) ;
this->interpolation_type = INTERPOLATION_LINEAR;
allocate_memory();
return ES_OK;
}
void Lattice::interpolate_linear(double* pos, double* value) {
int left_halo_index[3];
double d[3];
if (this->halo_size <= 0) {
char* c = runtime_error(128);
ERROR_SPRINTF(c, "Error in interpolate_linear: halo size is 0");
return;
}
for (int dim = 0; dim<3; dim++) {
left_halo_index[dim]=(int) floor((pos[dim]-this->local_offset[dim])/this->agrid[dim]) + this->halo_size;
d[dim]=((pos[dim]-this->local_offset[dim])/this->agrid[dim] - floor((pos[dim]-this->local_offset[dim])/this->agrid[dim]));
if (left_halo_index[dim] < 0 || left_halo_index[dim] >= this->halo_grid[dim]) {
char* c = runtime_error(128);
ERROR_SPRINTF(c, "Error in interpolate_linear: Particle out of range");
return;
}
}
double w[8];
index_t index[8];
w[0] = (1-d[0])*(1-d[1])*(1-d[2]);
index[0]=get_linear_index( left_halo_index[0], left_halo_index[1], left_halo_index[2], this->halo_grid);
w[1] = ( +d[0])*(1-d[1])*(1-d[2]);
index[1]=get_linear_index( left_halo_index[0]+1, left_halo_index[1], left_halo_index[2], this->halo_grid);
w[2] = (1-d[0])*( +d[1])*(1-d[2]);
index[2]=get_linear_index( left_halo_index[0], left_halo_index[1]+1, left_halo_index[2], this->halo_grid);
w[3] = ( +d[0])*( +d[1])*(1-d[2]);
index[3]=get_linear_index( left_halo_index[0]+1, left_halo_index[1]+1, left_halo_index[2], this->halo_grid);
w[4] = (1-d[0])*(1-d[1])*( +d[2]);
index[4]=get_linear_index( left_halo_index[0], left_halo_index[1], left_halo_index[2]+1, this->halo_grid);
w[5] = ( +d[0])*(1-d[1])*( +d[2]);
index[5]=get_linear_index( left_halo_index[0]+1, left_halo_index[1], left_halo_index[2]+1, this->halo_grid);
w[6] = (1-d[0])*( +d[1])*( +d[2]);
index[6]=get_linear_index( left_halo_index[0], left_halo_index[1]+1, left_halo_index[2]+1, this->halo_grid);
w[7] = ( +d[0])*( +d[1])*( +d[2]);
index[7]=get_linear_index( left_halo_index[0]+1, left_halo_index[1]+1, left_halo_index[2]+1, this->halo_grid);
for (unsigned int i = 0; i<this->dim; i++) {
value[i] = 0;
}
double* local_value;
for (unsigned int i=0; i<8; i++) {
get_data_for_linear_index(index[i], (void**) &local_value);
for (unsigned int j = 0; j<this->dim; j++) {
value[j]+=w[i]*local_value[j];
}
}
}
/** Caclulate mass of the LB fluid.
* \param result Fluid mass
*/
void lb_calc_fluid_mass(double *result) {
int x, y, z, index;
double mass = 0.0;
for (x=1; x<=lblattice.grid[0]; x++) {
for (y=1; y<=lblattice.grid[1]; y++) {
for (z=1; z<=lblattice.grid[2]; z++) {
index = get_linear_index(x,y,z,lblattice.halo_grid);
lb_calc_local_rho(&lbfluid[index]);
mass += *lbfluid[index].rho;
}
}
}
MPI_Reduce(&mass, result, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
}
/** Caclulate mass of the LB fluid.
* \param result Fluid mass
*/
void lb_calc_fluid_mass(double *result) {
int x, y, z, index;
double sum_rho=0.0, rho=0.0;
for (x=1; x<=lblattice.grid[0]; x++) {
for (y=1; y<=lblattice.grid[1]; y++) {
for (z=1; z<=lblattice.grid[2]; z++) {
index = get_linear_index(x,y,z,lblattice.halo_grid);
lb_calc_local_rho(index,&rho);
//fprintf(stderr,"(%d,%d,%d) %e\n",x,y,z,rho);
sum_rho += rho;
}
}
}
MPI_Reduce(&sum_rho, result, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
}
/** Fill a communication cell pointer list. Fill the cell pointers of
all cells which are inside a rectangular subgrid of the 3D cell
grid (\ref DomainDecomposition::ghost_cell_grid) starting from the
lower left corner lc up to the high top corner hc. The cell
pointer list part_lists must already be large enough.
\param part_lists List of cell pointers to store the result.
\param lc lower left corner of the subgrid.
\param hc high up corner of the subgrid.
*/
int dd_fill_comm_cell_lists(Cell **part_lists, int lc[3], int hc[3])
{
int i,m,n,o,c=0;
/* sanity check */
for(i=0; i<3; i++) {
if(lc[i]<0 || lc[i] >= dd.ghost_cell_grid[i]) return 0;
if(hc[i]<0 || hc[i] >= dd.ghost_cell_grid[i]) return 0;
if(lc[i] > hc[i]) return 0;
}
for(o=lc[0]; o<=hc[0]; o++)
for(n=lc[1]; n<=hc[1]; n++)
for(m=lc[2]; m<=hc[2]; m++) {
i = get_linear_index(o,n,m,dd.ghost_cell_grid);
CELL_TRACE(fprintf(stderr,"%d: dd_fill_comm_cell_list: add cell %d\n",this_node,i));
part_lists[c] = &cells[i];
c++;
}
return c;
}
void Lattice::map_position_to_lattice(const double pos[3], index_t node_index[8], double delta[6]) {
int dir,ind[3] ;
double lpos, rel;
/* determine the elementary lattice cell containing the particle
and the relative position of the particle in this cell */
for (dir=0;dir<3;dir++) {
lpos = pos[dir] - my_left[dir];
rel = lpos/this->agrid[dir] + 0.5; // +1 for halo offset
ind[dir] = (int)floor(rel);
/* surrounding elementary cell is not completely inside this box,
adjust if this is due to round off errors */
if (ind[dir] < 0) {
if (fabs(rel) < ROUND_ERROR_PREC) {
ind[dir] = 0; // TODO
} else {
fprintf(stderr,"%d: map_position_to_lattice: position (%f,%f,%f) not inside a local plaquette in dir %d ind[dir]=%d rel=%f lpos=%f.\n",this_node,pos[0],pos[1],pos[2],dir,ind[dir],rel,lpos);
}
}
else if (ind[dir] > this->grid[dir]) {
if (lpos - local_box_l[dir] < ROUND_ERROR_PREC*local_box_l[dir])
ind[dir] = this->grid[dir];
else
fprintf(stderr,"%d: map_position_to_lattice: position (%f,%f,%f) not inside a local plaquette in dir %d ind[dir]=%d rel=%f lpos=%f.\n",this_node,pos[0],pos[1],pos[2],dir,ind[dir],rel,lpos);
}
delta[3+dir] = rel - ind[dir]; // delta_x/a
delta[dir] = 1.0 - delta[3+dir];
}
node_index[0] = get_linear_index(ind[0],ind[1],ind[2],this->halo_grid);
node_index[1] = node_index[0] + 1;
node_index[2] = node_index[0] + this->halo_grid[0];
node_index[3] = node_index[0] + this->halo_grid[0] + 1;
node_index[4] = node_index[0] + this->halo_grid[0]*this->halo_grid[1];
node_index[5] = node_index[4] + 1;
node_index[6] = node_index[4] + this->halo_grid[0];
node_index[7] = node_index[4] + this->halo_grid[0] + 1;
}
/** Calculate temperature of the LB fluid.
* \param result Fluid temperature
*/
void lb_calc_fluid_temp(double *result) {
int x, y, z, index;
double rho, j[3];
double temp = 0.0;
for (x=1; x<=lblattice.grid[0]; x++) {
for (y=1; y<=lblattice.grid[1]; y++) {
for (z=1; z<=lblattice.grid[2]; z++) {
index = get_linear_index(x,y,z,lblattice.halo_grid);
lb_calc_local_fields(index, &rho, j, NULL);
temp += scalar(j,j);
}
}
}
temp *= 1./(3.*lbpar.rho*lblattice.grid_volume*lbpar.tau*lbpar.tau*lblattice.agrid);
MPI_Reduce(&temp, result, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
}
/** Initialize lattice.
*
* This function initializes the variables describing the lattice
* layout. Important: The lattice data is <em>not</em> allocated here!
*
* \param lattice pointer to the lattice
* \param agrid lattice spacing
* \param tau time step for lattice dynamics
*/
void init_lattice(Lattice *lattice, double agrid, double tau) {
int dir;
/* determine the number of local lattice nodes */
lattice->grid[0] = local_box_l[0]/agrid;
lattice->grid[1] = local_box_l[1]/agrid;
lattice->grid[2] = local_box_l[2]/agrid;
/* sanity checks */
for (dir=0;dir<3;dir++) {
/* check if local_box_l is compatible with lattice spacing */
if (fabs(local_box_l[dir]-lattice->grid[dir]*agrid) > ROUND_ERROR_PREC*box_l[dir]) {
char *errtxt = runtime_error(128);
ERROR_SPRINTF(errtxt, "{097 Lattice spacing agrid=%f is incompatible with local_box_l[%d]=%f (box_l[%d]=%f node_grid[%d]=%d) %f} ",agrid,dir,local_box_l[dir],dir,box_l[dir],dir,node_grid[dir],local_box_l[dir]-lattice->grid[dir]*agrid);
return;
}
}
/* set the lattice spacing */
lattice->agrid = agrid ;
lattice->tau = tau ;
LATTICE_TRACE(fprintf(stderr,"%d: box_l (%.3f,%.3f,%.3f) grid (%d,%d,%d) node_neighbors (%d,%d,%d,%d,%d,%d)\n",this_node,local_box_l[0],local_box_l[1],local_box_l[2],lattice->grid[0],lattice->grid[1],lattice->grid[2],node_neighbors[0],node_neighbors[1],node_neighbors[2],node_neighbors[3],node_neighbors[4],node_neighbors[5]));
/* determine the number of total nodes including halo */
lattice->halo_grid[0] = lattice->grid[0] + 2 ;
lattice->halo_grid[1] = lattice->grid[1] + 2 ;
lattice->halo_grid[2] = lattice->grid[2] + 2 ;
lattice->grid_volume = lattice->grid[0]*lattice->grid[1]*lattice->grid[2] ;
lattice->halo_grid_volume = lattice->halo_grid[0]*lattice->halo_grid[1]*lattice->halo_grid[2] ;
lattice->halo_grid_surface = lattice->halo_grid_volume - lattice->grid_volume ;
lattice->halo_offset = get_linear_index(1,1,1,lattice->halo_grid) ;
}
/** Calculate a velocity profile for the LB fluid. */
void lb_calc_velocity_profile(double *velprof, int vcomp, int pdir, int x1, int x2) {
int index, dir[3];
double local_rho, local_j;
/* \todo generalize and parallelize */
dir[(pdir+1)%3] = x1;
dir[(pdir+2)%3] = x2;
for (dir[pdir]=1;dir[pdir]<=lblattice.grid[pdir];dir[pdir]++) {
index = get_linear_index(dir[0],dir[1],dir[2],lblattice.halo_grid);
lb_calc_local_j(&lbfluid[index]);
lb_calc_local_rho(&lbfluid[index]);
local_rho = *lbfluid[index].rho;
local_j = lbfluid[index].j[vcomp];
if (local_j == 0) {
velprof[dir[pdir]-1] = 0.0;
} else {
velprof[dir[pdir]-1] = local_j/local_rho * lblattice.agrid/lbpar.tau;
}
}
}
void lb_bounce_back() {
#ifdef D3Q19
#ifndef PULL
int k,i,l;
int yperiod = lblattice.halo_grid[0];
int zperiod = lblattice.halo_grid[0]*lblattice.halo_grid[1];
int next[19];
int x,y,z;
double population_shift;
double modes[19];
next[0] = 0; // ( 0, 0, 0) =
next[1] = 1; // ( 1, 0, 0) +
next[2] = - 1; // (-1, 0, 0)
next[3] = yperiod; // ( 0, 1, 0) +
next[4] = - yperiod; // ( 0,-1, 0)
next[5] = zperiod; // ( 0, 0, 1) +
next[6] = - zperiod; // ( 0, 0,-1)
next[7] = (1+yperiod); // ( 1, 1, 0) +
next[8] = - (1+yperiod); // (-1,-1, 0)
next[9] = (1-yperiod); // ( 1,-1, 0)
next[10] = - (1-yperiod); // (-1, 1, 0) +
next[11] = (1+zperiod); // ( 1, 0, 1) +
next[12] = - (1+zperiod); // (-1, 0,-1)
next[13] = (1-zperiod); // ( 1, 0,-1)
next[14] = - (1-zperiod); // (-1, 0, 1) +
next[15] = (yperiod+zperiod); // ( 0, 1, 1) +
next[16] = - (yperiod+zperiod); // ( 0,-1,-1)
next[17] = (yperiod-zperiod); // ( 0, 1,-1)
next[18] = - (yperiod-zperiod); // ( 0,-1, 1) +
int reverse[] = { 0, 2, 1, 4, 3, 6, 5, 8, 7, 10, 9, 12, 11, 14, 13, 16, 15, 18, 17 };
/* bottom-up sweep */
// for (k=lblattice.halo_offset;k<lblattice.halo_grid_volume;k++) {
for (z=0; z<lblattice.grid[2]+2; z++) {
for (y=0; y<lblattice.grid[1]+2; y++) {
for (x=0; x<lblattice.grid[0]+2; x++) {
k= get_linear_index(x,y,z,lblattice.halo_grid);
if (lbfields[k].boundary) {
lb_calc_modes(k, modes);
for (i=0; i<19; i++) {
population_shift=0;
for (l=0; l<3; l++) {
population_shift-=lbpar.agrid*lbpar.agrid*lbpar.agrid*lbpar.agrid*lbpar.agrid*lbpar.rho[0]*2*lbmodel.c[i][l]*lbmodel.w[i]*lb_boundaries[lbfields[k].boundary-1].velocity[l]/lbmodel.c_sound_sq;
}
if ( x-lbmodel.c[i][0] > 0 && x -lbmodel.c[i][0] < lblattice.grid[0]+1 &&
y-lbmodel.c[i][1] > 0 && y -lbmodel.c[i][1] < lblattice.grid[1]+1 &&
z-lbmodel.c[i][2] > 0 && z -lbmodel.c[i][2] < lblattice.grid[2]+1) {
if ( !lbfields[k-next[i]].boundary ) {
for (l=0; l<3; l++) {
lb_boundaries[lbfields[k].boundary-1].force[l]+=(2*lbfluid[1][i][k]+population_shift)*lbmodel.c[i][l];
}
lbfluid[1][reverse[i]][k-next[i]] = lbfluid[1][i][k]+ population_shift;
}
else {
lbfluid[1][reverse[i]][k-next[i]] = lbfluid[1][i][k] = 0.0;
}
}
}
}
}
}
}
#else
#error Bounce back boundary conditions are only implemented for PUSH scheme!
#endif
#else
#error Bounce back boundary conditions are only implemented for D3Q19!
#endif
}
void Lattice::interpolate_linear_gradient(double* pos, double* value) {
int left_halo_index[3];
double d[3];
if (this->halo_size <= 0) {
runtimeErrorMsg() << "Error in interpolate_linear: halo size is 0";
return;
}
for (int dim = 0; dim<3; dim++) {
left_halo_index[dim]=(int) floor((pos[dim]-this->local_offset[dim])/this->agrid[dim]) + this->halo_size;
d[dim]=((pos[dim]-this->local_offset[dim])/this->agrid[dim] - floor((pos[dim]-this->local_offset[dim])/this->agrid[dim]));
if (left_halo_index[dim] < 0 || left_halo_index[dim] >= this->halo_grid[dim]) {
runtimeErrorMsg() <<"Error in interpolate_linear: Particle out of range";
return;
}
}
index_t index;
double* local_value;
for (unsigned int i = 0; i<3*this->dim; i++) {
value[i] = 0;
}
index=get_linear_index( left_halo_index[0], left_halo_index[1], left_halo_index[2], this->halo_grid);
for (unsigned int i = 0; i<this->dim; i++) {
get_data_for_linear_index(index, (void**) &local_value);
value[3*i ]+= ( -1 )*(1-d[1])*(1-d[2]) * local_value[i] / this->agrid[0];
value[3*i+1]+= (1-d[0])*( -1 )*(1-d[2]) * local_value[i] / this->agrid[1];
value[3*i+2]+= (1-d[0])*(1-d[1])*( -1 ) * local_value[i] / this->agrid[2];
}
index=get_linear_index( left_halo_index[0]+1, left_halo_index[1], left_halo_index[2], this->halo_grid);
for (unsigned int i = 0; i<this->dim; i++) {
get_data_for_linear_index(index, (void**) &local_value);
value[3*i ]+= ( +1 )*(1-d[1])*(1-d[2]) * local_value[i] / this->agrid[0];
value[3*i+1]+= ( +d[0])*( -1 )*(1-d[2]) * local_value[i] / this->agrid[1];
value[3*i+2]+= ( +d[0])*(1-d[1])*( -1 ) * local_value[i] / this->agrid[2];
}
index=get_linear_index( left_halo_index[0], left_halo_index[1]+1, left_halo_index[2], this->halo_grid);
for (unsigned int i = 0; i<this->dim; i++) {
get_data_for_linear_index(index, (void**) &local_value);
value[3*i ]+= ( -1 )*( +d[1])*(1-d[2]) * local_value[i] / this->agrid[0];
value[3*i+1]+= (1-d[0])*( +1 )*(1-d[2]) * local_value[i] / this->agrid[1];
value[3*i+2]+= (1-d[0])*( +d[1])*( -1 ) * local_value[i] / this->agrid[2];
}
index=get_linear_index( left_halo_index[0]+1, left_halo_index[1]+1, left_halo_index[2], this->halo_grid);
for (unsigned int i = 0; i<this->dim; i++) {
get_data_for_linear_index(index, (void**) &local_value);
value[3*i ]+= ( +1 )*( +d[1])*(1-d[2]) * local_value[i] / this->agrid[0];
value[3*i+1]+= ( +d[0])*( +1 )*(1-d[2]) * local_value[i] / this->agrid[1];
value[3*i+2]+= ( +d[0])*( +d[1])*( -1 ) * local_value[i] / this->agrid[2];
}
index=get_linear_index( left_halo_index[0] , left_halo_index[1] , left_halo_index[2] + 1, this->halo_grid);
for (unsigned int i = 0; i<this->dim; i++) {
get_data_for_linear_index(index, (void**) &local_value);
value[3*i ]+= ( -1 )*(1-d[1])*( +d[2]) * local_value[i] / this->agrid[0];
value[3*i+1]+= (1-d[0])*( -1 )*( +d[2]) * local_value[i] / this->agrid[1];
value[3*i+2]+= (1-d[0])*(1-d[1])*( +1 ) * local_value[i] / this->agrid[2];
}
index=get_linear_index( left_halo_index[0]+1, left_halo_index[1], left_halo_index[2]+1, this->halo_grid);
for (unsigned int i = 0; i<this->dim; i++) {
get_data_for_linear_index(index, (void**) &local_value);
value[3*i ]+= ( +1 )*(1-d[1])*( +d[2]) * local_value[i] / this->agrid[0];
value[3*i+1]+= ( +d[0])*( -1 )*( +d[2]) * local_value[i] / this->agrid[1];
value[3*i+2]+= ( +d[0])*(1-d[1])*( +1 ) * local_value[i] / this->agrid[2];
}
index=get_linear_index( left_halo_index[0], left_halo_index[1]+1, left_halo_index[2]+1, this->halo_grid);
for (unsigned int i = 0; i<this->dim; i++) {
get_data_for_linear_index(index, (void**) &local_value);
value[3*i ]+= ( -1 )*( +d[1])*( +d[2]) * local_value[i] / this->agrid[0];
value[3*i+1]+= (1-d[0])*( +1 )*( +d[2]) * local_value[i] / this->agrid[1];
value[3*i+2]+= (1-d[0])*( +d[1])*( +1 ) * local_value[i] / this->agrid[2];
}
index=get_linear_index( left_halo_index[0]+1, left_halo_index[1]+1, left_halo_index[2]+1, this->halo_grid);
for (unsigned int i = 0; i<this->dim; i++) {
get_data_for_linear_index(index, (void**) &local_value);
value[3*i ]+= ( +1 )*( +d[1])*( +d[2]) * local_value[i] / this->agrid[0];
value[3*i+1]+= ( +d[0])*( +1 )*( +d[2]) * local_value[i] / this->agrid[1];
value[3*i+2]+= ( +d[0])*( +d[1])*( +1 ) * local_value[i] / this->agrid[2];
}
}
/** Calculate a velocity profile for the LB fluid. */
void lb_calc_velprof(double *result, int *params) {
int index, dir[3], grid[3];
int newroot=0, subrank, involved=0;
double rho, j[3], *velprof;
MPI_Comm slice_comm = NULL;
MPI_Status status;
if (this_node != 0) params = malloc(4*sizeof(int));
MPI_Bcast(params, 4, MPI_INT, 0, MPI_COMM_WORLD);
int vcomp = params[0];
int pdir = params[1];
int x1 = params[2];
int x2 = params[3];
dir[pdir] = 0;
dir[(pdir+1)%3] = x1;
dir[(pdir+2)%3] = x2;
//fprintf(stderr,"%d: (%d,%d,%d)\n",this_node,dir[0],dir[1],dir[2]);
newroot = map_lattice_to_node(&lblattice, dir, grid);
map_node_array(this_node, node_pos);
//fprintf(stderr,"%d: newroot=%d (%d,%d,%d)\n",this_node,newroot,grid[0],grid[1],grid[2]);
if ( (grid[(pdir+1)%3] == node_pos[(pdir+1)%3])
&& (grid[(pdir+2)%3] == node_pos[(pdir+2)%3]) ) {
involved = 1;
}
MPI_Comm_split(MPI_COMM_WORLD, involved, this_node, &slice_comm);
MPI_Comm_rank(slice_comm, &subrank);
if (this_node == newroot)
result = realloc(result,box_l[pdir]/lblattice.agrid*sizeof(double));
//fprintf(stderr,"%d (%d,%d): result=%p vcomp=%d pdir=%d x1=%d x2=%d involved=%d\n",this_node,subrank,newroot,result,vcomp,pdir,x1,x2,involved);
if (involved) {
velprof = malloc(lblattice.grid[pdir]*sizeof(double));
//dir[(pdir+1)%3] += 1;
//dir[(pdir+2)%3] += 1;
for (dir[pdir]=1;dir[pdir]<=lblattice.grid[pdir];dir[pdir]++) {
index = get_linear_index(dir[0],dir[1],dir[2],lblattice.halo_grid);
lb_calc_local_fields(index, &rho, j, NULL);
//fprintf(stderr,"%p %d %.12e %.12e %d\n",lbfluid[0],index,rho,j[0],vcomp);
if (rho < ROUND_ERROR_PREC) {
velprof[dir[pdir]-1] = 0.0;
} else {
//velprof[dir[pdir]-1] = local_j / (SQR(lbpar.agrid)*lbpar.tau);
velprof[dir[pdir]-1] = j[vcomp]/rho * lblattice.agrid/lbpar.tau;
//fprintf(stderr,"%f %f %f\n",velprof[dir[pdir]-1],local_j,local_rho);
}
//if (dir[pdir]==lblattice.grid[pdir]) {
// int i;
// fprintf(stderr,"(%d,%d,%d)\n",dir[0],dir[1],dir[2]);
// fprintf(stderr,"%d\n",lbfluid[index].boundary);
// for (i=0;i<lbmodel.n_veloc;i++) fprintf(stderr,"local_n[%p][%d]=%.12e\n",lbfluid[index].n,i,lbfluid[index].n[i]+lbmodel.coeff[i][0]*lbpar.rho);
// fprintf(stderr,"local_rho=%e\n",local_rho);
// fprintf(stderr,"local_j=%e\n",local_j);
//}
}
MPI_Gather(velprof, lblattice.grid[pdir], MPI_DOUBLE, result, lblattice.grid[pdir], MPI_DOUBLE, newroot, slice_comm);
free(velprof);
}
MPI_Comm_free(&slice_comm);
if (newroot != 0) {
if (this_node == newroot) {
MPI_Send(result, lblattice.grid[pdir]*node_grid[pdir], MPI_DOUBLE, 0, REQ_VELPROF, MPI_COMM_WORLD);
free(result);
}
if (this_node == 0) {
//fprintf(stderr,"%d: I'm just here!\n",this_node);
MPI_Recv(result, lblattice.grid[pdir]*node_grid[pdir], MPI_DOUBLE, newroot, REQ_VELPROF, MPI_COMM_WORLD, &status);
//fprintf(stderr,"%d: And now I'm here!\n",this_node);
}
}
if (this_node !=0) free(params);
}
/** Calculate a density profile of the fluid. */
void lb_calc_densprof(double *result, int *params) {
int index, dir[3], grid[3];
int newroot=0, subrank, involved=0;
double *profile;
MPI_Comm slice_comm = NULL;
MPI_Status status;
if (this_node !=0) params = malloc(3*sizeof(int));
MPI_Bcast(params, 3, MPI_INT, 0, MPI_COMM_WORLD);
int pdir = params[0];
int x1 = params[1];
int x2 = params[2];
dir[pdir] = 0;
dir[(pdir+1)%3] = x1;
dir[(pdir+2)%3] = x2;
newroot = map_lattice_to_node(&lblattice, dir, grid);
map_node_array(this_node, node_pos);
if ( (grid[(pdir+1)%3] == node_pos[(pdir+1)%3])
&& (grid[(pdir+2)%3] == node_pos[(pdir+2)%3]) ) {
involved = 1;
}
MPI_Comm_split(MPI_COMM_WORLD, involved, this_node, &slice_comm);
MPI_Comm_rank(slice_comm, &subrank);
if (this_node == newroot)
result = realloc(result,box_l[pdir]/lblattice.agrid*sizeof(double));
if (involved) {
profile = malloc(lblattice.grid[pdir]*sizeof(double));
//dir[(pdir+1)%3] += 1;
//dir[(pdir+2)%3] += 1;
for (dir[pdir]=1;dir[pdir]<=lblattice.grid[pdir];dir[pdir]++) {
index = get_linear_index(dir[0],dir[1],dir[2],lblattice.halo_grid);
lb_calc_local_rho(index,&profile[dir[pdir]-1]);
//profile[dir[pdir]-1] = *lbfluid[index].rho;
//if (dir[pdir]==lblattice.grid[pdir]) {
// int i;
// fprintf(stderr,"(%d,%d,%d)\n",dir[0],dir[1],dir[2]);
// fprintf(stderr,"%d\n",lbfluid[index].boundary);
// for (i=0;i<lbmodel.n_veloc;i++) fprintf(stderr,"local_n[%p][%d]=%.12e\n",lbfluid[index].n,i,lbfluid[index].n[i]+lbmodel.coeff[i][0]*lbpar.rho);
// fprintf(stderr,"local_rho=%e\n",*lbfluid[index].rho);
//}
}
MPI_Gather(profile, lblattice.grid[pdir], MPI_DOUBLE, result, lblattice.grid[pdir], MPI_DOUBLE, 0, slice_comm);
free(profile);
}
MPI_Comm_free(&slice_comm);
if (newroot != 0) {
if (this_node == newroot) {
MPI_Send(result, lblattice.grid[pdir]*node_grid[pdir], MPI_DOUBLE, 0, REQ_DENSPROF, MPI_COMM_WORLD);
free(result);
}
if (this_node == 0) {
MPI_Recv(result, lblattice.grid[pdir]*node_grid[pdir], MPI_DOUBLE, newroot, REQ_DENSPROF, MPI_COMM_WORLD, &status);
}
}
if (this_node != 0) free(params);
}
void Lattice::set_data_for_local_grid_index(index_t* ind, void* data) {
memmove(((char*)this->_data) + get_linear_index(ind[0]+this->halo_size, ind[1]+this->halo_size, ind[2]+this->halo_size, this->halo_grid)*this->element_size, data, this->element_size);
}
/** Initialize boundary conditions for all constraints in the system. */
void lb_init_boundaries() {
int n, x, y, z, node_domain_position[3], offset[3];
char *errtxt;
double pos[3], dist, dist_tmp=0.0, dist_vec[3];
int the_boundary=-1;
map_node_array(this_node, node_domain_position);
offset[0] = node_domain_position[0]*lblattice.grid[0];
offset[1] = node_domain_position[1]*lblattice.grid[1];
offset[2] = node_domain_position[2]*lblattice.grid[2];
for (n=0;n<lblattice.halo_grid_volume;n++) {
lbfields[n].boundary = 0;
}
if (lblattice.halo_grid_volume==0)
return;
for (z=0; z<lblattice.grid[2]+2; z++) {
for (y=0; y<lblattice.grid[1]+2; y++) {
for (x=0; x<lblattice.grid[0]+2; x++) {
pos[0] = (offset[0]+(x-1))*lblattice.agrid;
pos[1] = (offset[1]+(y-1))*lblattice.agrid;
pos[2] = (offset[2]+(z-1))*lblattice.agrid;
dist = 1e99;
for (n=0;n<n_lb_boundaries;n++) {
switch (lb_boundaries[n].type) {
case LB_BOUNDARY_WAL:
calculate_wall_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.wal, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_SPH:
calculate_sphere_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.sph, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_CYL:
calculate_cylinder_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.cyl, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_POR:
calculate_pore_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.pore, &dist_tmp, dist_vec);
break;
default:
errtxt = runtime_error(128);
ERROR_SPRINTF(errtxt, "{109 lbboundary type %d not implemented in lb_init_boundaries()\n", lb_boundaries[n].type);
}
// if (abs(dist) > abs(dist_tmp) || n == 0) {
if (dist_tmp<dist) { //If you try to create a wall of finite thickness ...|xxx|..., it makes every node a wall node! We still leave it like that, since it allows for corners without problems. We will add a box type to allow for walls of finite thickness. (Georg Rempfer, Stefan Kesselheim, 05.10.2011)
dist = dist_tmp;
the_boundary = n;
}
}
if (dist <= 0 && n_lb_boundaries > 0) {
lbfields[get_linear_index(x,y,z,lblattice.halo_grid)].boundary = the_boundary+1;
} else {
lbfields[get_linear_index(x,y,z,lblattice.halo_grid)].boundary=0;
}
}
}
}
}
void Lattice::get_data_for_halo_index(index_t* ind, void** data) {
(*data) = ((char*)this->_data) + get_linear_index(ind[0], ind[1], ind[2], this->halo_grid)*this->element_size;
}
int fft_find_comm_groups(int grid1[3], int grid2[3], int *node_list1, int *node_list2,
int *group, int *pos, int *my_pos)
{
int i;
/* communication group cell size on grid1 and grid2 */
int s1[3], s2[3];
/* The communication group cells build the same super grid on grid1 and grid2 */
int ds[3];
/* communication group size */
int g_size=1;
/* comm. group cell index */
int gi[3];
/* position of a node in a grid */
int p1[3], p2[3];
/* node identity */
int n;
/* this_node position in the communication group. */
int c_pos=-1;
/* flag for group identification */
int my_group=0;
FFT_TRACE(fprintf(stderr,"%d: fft_find_comm_groups:\n",this_node));
FFT_TRACE(fprintf(stderr,"%d: for grid1=(%d,%d,%d) and grids=(%d,%d,%d)\n",
this_node,grid1[0],grid1[1],grid1[2],grid2[0],grid2[1],grid2[2]));
/* calculate dimension of comm. group cells for both grids */
if( (grid1[0]*grid1[1]*grid1[2]) != (grid2[0]*grid2[1]*grid2[2]) ) return -1; /* unlike number of nodes */
for(i=0;i<3;i++) {
s1[i] = grid1[i] / grid2[i];
if(s1[i] == 0) s1[i] = 1;
else if(grid1[i] != grid2[i]*s1[i]) return -1; /* grids do not match!!! */
s2[i] = grid2[i] / grid1[i];
if(s2[i] == 0) s2[i] = 1;
else if(grid2[i] != grid1[i]*s2[i]) return -1; /* grids do not match!!! */
ds[i] = grid2[i] / s2[i];
g_size *= s2[i];
}
/* calc node_list2 */
/* loop through all comm. group cells */
for(gi[2] = 0; gi[2] < ds[2]; gi[2]++)
for(gi[1] = 0; gi[1] < ds[1]; gi[1]++)
for(gi[0] = 0; gi[0] < ds[0]; gi[0]++) {
/* loop through all nodes in that comm. group cell */
for(i=0;i<g_size;i++) {
p1[0] = (gi[0]*s1[0]) + (i%s1[0]);
p1[1] = (gi[1]*s1[1]) + ((i/s1[0])%s1[1]);
p1[2] = (gi[2]*s1[2]) + (i/(s1[0]*s1[1]));
p2[0] = (gi[0]*s2[0]) + (i%s2[0]);
p2[1] = (gi[1]*s2[1]) + ((i/s2[0])%s2[1]);
p2[2] = (gi[2]*s2[2]) + (i/(s2[0]*s2[1]));
n = node_list1[ get_linear_index(p1[0],p1[1],p1[2],grid1) ];
node_list2[ get_linear_index(p2[0],p2[1],p2[2],grid2) ] = n ;
pos[3*n+0] = p2[0]; pos[3*n+1] = p2[1]; pos[3*n+2] = p2[2];
if(my_group==1) group[i] = n;
if(n==this_node && my_group==0) {
my_group = 1;
c_pos = i;
my_pos[0] = p2[0]; my_pos[1] = p2[1]; my_pos[2] = p2[2];
i=-1; /* restart the loop */
}
}
my_group=0;
}
/* permute comm. group according to the nodes position in the group */
/* This is necessary to have matching node pairs during communication! */
while( c_pos>0 ) {
n=group[g_size-1];
for(i=g_size-1; i>0; i--) group[i] = group[i-1];
group[0] = n;
c_pos--;
}
return g_size;
}
/** Initialize boundary conditions for all constraints in the system. */
void lb_init_boundaries() {
int n, x, y, z;
//char *errtxt;
double pos[3], dist, dist_tmp=0.0, dist_vec[3];
if (lattice_switch & LATTICE_LB_GPU) {
#if defined (LB_GPU) && defined (LB_BOUNDARIES_GPU)
int number_of_boundnodes = 0;
int *host_boundary_node_list= (int*)Utils::malloc(sizeof(int));
int *host_boundary_index_list= (int*)Utils::malloc(sizeof(int));
size_t size_of_index;
int boundary_number = -1; // the number the boundary will actually belong to.
#ifdef EK_BOUNDARIES
ekfloat *host_wallcharge_species_density = NULL;
float node_wallcharge = 0.0f;
int wallcharge_species = -1, charged_boundaries = 0;
int node_charged = 0;
for(n = 0; n < int(n_lb_boundaries); n++)
lb_boundaries[n].net_charge = 0.0;
if (ek_initialized)
{
host_wallcharge_species_density = (ekfloat*) Utils::malloc(ek_parameters.number_of_nodes * sizeof(ekfloat));
for(n = 0; n < int(n_lb_boundaries); n++) {
if(lb_boundaries[n].charge_density != 0.0) {
charged_boundaries = 1;
break;
}
}
if (pdb_charge_lattice) {
charged_boundaries = 1;
}
for(n = 0; n < int(ek_parameters.number_of_species); n++)
if(ek_parameters.valency[n] != 0.0) {
wallcharge_species = n;
break;
}
if(wallcharge_species == -1 && charged_boundaries) {
runtimeErrorMsg() <<"no charged species available to create wall charge\n";
}
}
#endif
for(z=0; z<int(lbpar_gpu.dim_z); z++) {
for(y=0; y<int(lbpar_gpu.dim_y); y++) {
for (x=0; x<int(lbpar_gpu.dim_x); x++) {
pos[0] = (x+0.5)*lbpar_gpu.agrid;
pos[1] = (y+0.5)*lbpar_gpu.agrid;
pos[2] = (z+0.5)*lbpar_gpu.agrid;
dist = 1e99;
#ifdef EK_BOUNDARIES
if (ek_initialized)
{
host_wallcharge_species_density[ek_parameters.dim_y*ek_parameters.dim_x*z + ek_parameters.dim_x*y + x] = 0.0f;
node_charged = 0;
node_wallcharge = 0.0f;
}
#endif
for (n=0; n < n_lb_boundaries; n++) {
switch (lb_boundaries[n].type) {
case LB_BOUNDARY_WAL:
calculate_wall_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.wal, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_SPH:
calculate_sphere_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.sph, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_CYL:
calculate_cylinder_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.cyl, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_RHOMBOID:
calculate_rhomboid_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.rhomboid, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_POR:
calculate_pore_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.pore, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_STOMATOCYTE:
calculate_stomatocyte_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.stomatocyte, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_HOLLOW_CONE:
calculate_hollow_cone_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.hollow_cone, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_SPHEROCYLINDER:
calculate_spherocylinder_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.spherocyl, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_VOXEL: // voxel data do not need dist
//calculate_voxel_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.voxel, &dist_tmp, dist_vec);
dist_tmp=1e99;
break;
default:
runtimeErrorMsg() <<"lbboundary type "<< lb_boundaries[n].type << " not implemented in lb_init_boundaries()\n";
}
if (dist > dist_tmp || n == 0) {
dist = dist_tmp;
boundary_number = n;
}
#ifdef EK_BOUNDARIES
if (ek_initialized)
{
if(dist_tmp <= 0 && lb_boundaries[n].charge_density != 0.0f) {
node_charged = 1;
node_wallcharge += lb_boundaries[n].charge_density * ek_parameters.agrid*ek_parameters.agrid*ek_parameters.agrid;
lb_boundaries[n].net_charge += lb_boundaries[n].charge_density * ek_parameters.agrid*ek_parameters.agrid*ek_parameters.agrid;
}
}
#endif
}
#ifdef EK_BOUNDARIES
if(pdb_boundary_lattice &&
pdb_boundary_lattice[ek_parameters.dim_y*ek_parameters.dim_x*z + ek_parameters.dim_x*y + x]) {
dist = -1;
boundary_number = n_lb_boundaries; // Makes sure that boundary_number is not used by a constraint
}
#endif
if (dist <= 0 && boundary_number >= 0 && (n_lb_boundaries > 0 || pdb_boundary_lattice)) {
size_of_index = (number_of_boundnodes+1)*sizeof(int);
host_boundary_node_list = (int *) Utils::realloc(host_boundary_node_list, size_of_index);
host_boundary_index_list = (int *) Utils::realloc(host_boundary_index_list, size_of_index);
host_boundary_node_list[number_of_boundnodes] = x + lbpar_gpu.dim_x*y + lbpar_gpu.dim_x*lbpar_gpu.dim_y*z;
host_boundary_index_list[number_of_boundnodes] = boundary_number + 1;
number_of_boundnodes++;
//printf("boundindex %i: \n", number_of_boundnodes);
}
#ifdef EK_BOUNDARIES
if (ek_initialized)
{
ek_parameters.number_of_boundary_nodes = number_of_boundnodes;
if(wallcharge_species != -1) {
if(pdb_charge_lattice &&
pdb_charge_lattice[ek_parameters.dim_y*ek_parameters.dim_x*z + ek_parameters.dim_x*y + x] != 0.0f) {
node_charged = 1;
node_wallcharge += pdb_charge_lattice[ek_parameters.dim_y*ek_parameters.dim_x*z + ek_parameters.dim_x*y + x];
}
if(node_charged)
host_wallcharge_species_density[ek_parameters.dim_y*ek_parameters.dim_x*z + ek_parameters.dim_x*y + x] = node_wallcharge / ek_parameters.valency[wallcharge_species];
else if(dist <= 0)
host_wallcharge_species_density[ek_parameters.dim_y*ek_parameters.dim_x*z + ek_parameters.dim_x*y + x] = 0.0f;
else
host_wallcharge_species_density[ek_parameters.dim_y*ek_parameters.dim_x*z + ek_parameters.dim_x*y + x] = ek_parameters.density[wallcharge_species] * ek_parameters.agrid*ek_parameters.agrid*ek_parameters.agrid;
}
}
#endif
}
}
}
/**call of cuda fkt*/
float* boundary_velocity = (float *) Utils::malloc(3*(n_lb_boundaries+1)*sizeof(float));
for (n=0; n<n_lb_boundaries; n++) {
boundary_velocity[3*n+0]=lb_boundaries[n].velocity[0];
boundary_velocity[3*n+1]=lb_boundaries[n].velocity[1];
boundary_velocity[3*n+2]=lb_boundaries[n].velocity[2];
}
boundary_velocity[3*n_lb_boundaries+0] = 0.0f;
boundary_velocity[3*n_lb_boundaries+1] = 0.0f;
boundary_velocity[3*n_lb_boundaries+2] = 0.0f;
if (n_lb_boundaries || pdb_boundary_lattice)
lb_init_boundaries_GPU(n_lb_boundaries, number_of_boundnodes, host_boundary_node_list, host_boundary_index_list, boundary_velocity);
free(boundary_velocity);
free(host_boundary_node_list);
free(host_boundary_index_list);
#ifdef EK_BOUNDARIES
if (ek_initialized)
{
ek_init_species_density_wallcharge(host_wallcharge_species_density, wallcharge_species);
free(host_wallcharge_species_density);
}
#endif
#endif /* defined (LB_GPU) && defined (LB_BOUNDARIES_GPU) */
}
else {
#if defined (LB) && defined (LB_BOUNDARIES)
int node_domain_position[3], offset[3];
int the_boundary=-1;
map_node_array(this_node, node_domain_position);
offset[0] = node_domain_position[0]*lblattice.grid[0];
offset[1] = node_domain_position[1]*lblattice.grid[1];
offset[2] = node_domain_position[2]*lblattice.grid[2];
for (n=0;n<lblattice.halo_grid_volume;n++) {
lbfields[n].boundary = 0;
}
if (lblattice.halo_grid_volume==0)
return;
for (z=0; z<lblattice.grid[2]+2; z++) {
for (y=0; y<lblattice.grid[1]+2; y++) {
for (x=0; x<lblattice.grid[0]+2; x++) {
pos[0] = (offset[0]+(x-0.5))*lblattice.agrid[0];
pos[1] = (offset[1]+(y-0.5))*lblattice.agrid[1];
pos[2] = (offset[2]+(z-0.5))*lblattice.agrid[2];
dist = 1e99;
for (n=0;n<n_lb_boundaries;n++) {
switch (lb_boundaries[n].type) {
case LB_BOUNDARY_WAL:
calculate_wall_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.wal, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_SPH:
calculate_sphere_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.sph, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_CYL:
calculate_cylinder_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.cyl, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_RHOMBOID:
calculate_rhomboid_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.rhomboid, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_POR:
calculate_pore_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.pore, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_STOMATOCYTE:
calculate_stomatocyte_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.stomatocyte, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_HOLLOW_CONE:
calculate_hollow_cone_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.hollow_cone, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_VOXEL: // voxel data do not need dist
dist_tmp=1e99;
//calculate_voxel_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.voxel, &dist_tmp, dist_vec);
break;
default:
runtimeErrorMsg() <<"lbboundary type " << lb_boundaries[n].type << " not implemented in lb_init_boundaries()\n";
}
if (dist_tmp<dist || n == 0) {
dist = dist_tmp;
the_boundary = n;
}
}
if (dist <= 0 && the_boundary >= 0 && n_lb_boundaries > 0) {
lbfields[get_linear_index(x,y,z,lblattice.halo_grid)].boundary = the_boundary+1;
//printf("boundindex %i: \n", get_linear_index(x,y,z,lblattice.halo_grid));
}
else {
lbfields[get_linear_index(x,y,z,lblattice.halo_grid)].boundary = 0;
}
}
}
}
//printf("init voxels\n\n");
// SET VOXEL BOUNDARIES DIRECTLY
int xxx,yyy,zzz=0;
char line[80];
for (n=0;n<n_lb_boundaries;n++) {
switch (lb_boundaries[n].type) {
case LB_BOUNDARY_VOXEL:
//lbfields[get_linear_index(lb_boundaries[n].c.voxel.pos[0],lb_boundaries[n].c.voxel.pos[1],lb_boundaries[n].c.voxel.pos[2],lblattice.halo_grid)].boundary = n+1;
FILE *fp;
//fp=fopen("/home/mgusenbauer/Daten/Copy/DUK/GentlePump/Optimierer/NSvsLBM/geometry_files/bottleneck_fine_voxel_data_d20_converted_noMirror.csv", "r");
//fp=fopen("/home/mgusenbauer/Daten/Copy/DUK/GentlePump/Optimierer/NSvsLBM/geometry_files/bottleneck_fine_voxel_data_d80_converted_noMirror.csv", "r");
//fp=fopen("/home/mgusenbauer/Daten/Copy/DUK/GentlePump/Optimierer/NSvsLBM/geometry_files/bottleneck_fine_voxel_data_d80_converted.csv", "r");
fp=fopen(lb_boundaries[n].c.voxel.filename, "r");
while(fgets(line, 80, fp) != NULL)
{
/* get a line, up to 80 chars from fp, done if NULL */
sscanf (line, "%d %d %d", &xxx,&yyy,&zzz);
//printf("%d %d %d\n", xxx,yyy,zzz);
//lbfields[get_linear_index(xxx,yyy+30,zzz,lblattice.halo_grid)].boundary = n+1;
lbfields[get_linear_index(xxx,yyy,zzz,lblattice.halo_grid)].boundary = n+1;
}
fclose(fp);
break;
default:
break;
}
}
// CHECK FOR BOUNDARY NEIGHBOURS AND SET FLUID NORMAL VECTOR
//int neighbours = {0,0,0,0,0,0};
//int x=0,y=0,z=0;
//double nn[]={0.0,0.0,0.0,0.0,0.0,0.0};
//for (n=0;n<n_lb_boundaries;n++) {
//switch (lb_boundaries[n].type) {
//case LB_BOUNDARY_VOXEL:
//x=lb_boundaries[n].c.voxel.pos[0];
//y=lb_boundaries[n].c.voxel.pos[1];
//z=lb_boundaries[n].c.voxel.pos[2];
//if(((x-1) >= 0) && (lbfields[get_linear_index(x-1,y,z,lblattice.halo_grid)].boundary == 0)) nn[0] = -1.0;//neighbours[0] = -1;
//if(((x+1) <= lblattice.grid[0]) && (lbfields[get_linear_index(x+1,y,z,lblattice.halo_grid)].boundary == 0)) nn[1] = 1.0;//neighbours[1] = 1;
////printf("%.0lf %.0lf ",nn[0],nn[1]);
//lb_boundaries[n].c.voxel.n[0] = nn[0]+nn[1];
////nn=0.0;
//if(((y-1) >= 0) && (lbfields[get_linear_index(x,y-1,z,lblattice.halo_grid)].boundary == 0)) nn[2] = -1.0;//neighbours[2] = -1;
//if(((y+1) <= lblattice.grid[1]) && (lbfields[get_linear_index(x,y+1,z,lblattice.halo_grid)].boundary == 0)) nn[3] = 1.0;//neighbours[3] = 1;
////printf("%.0lf %.0lf ",nn[2],nn[3]);
//lb_boundaries[n].c.voxel.n[1] = nn[2]+nn[3];
////nn=0.0;
//if(((z-1) >= 0) && (lbfields[get_linear_index(x,y,z-1,lblattice.halo_grid)].boundary == 0)) nn[4] = -1.0;//neighbours[4] = -1;
//if(((z+1) <= lblattice.grid[2]) && (lbfields[get_linear_index(x,y,z+1,lblattice.halo_grid)].boundary == 0)) nn[5] = 1.0;//neighbours[5]= 1;
////printf("%.0lf %.0lf ",nn[4],nn[5]);
//lb_boundaries[n].c.voxel.n[2] = nn[4]+nn[5];
//nn[0]=0.0,nn[1]=0.0,nn[2]=0.0,nn[3]=0.0,nn[4]=0.0,nn[5]=0.0;
////printf("t %d pos: %.0lf %.0lf %.0lf, fluid normal %.0lf %.0lf %.0lf\n",n, x,y,z,lb_boundaries[n].c.voxel.normal[0],lb_boundaries[n].c.voxel.normal[1],lb_boundaries[n].c.voxel.normal[2]);
////printf("boundaries: %d %d %d %d %d %d\n",lbfields[get_linear_index(x-1,y,z,lblattice.halo_grid)].boundary,lbfields[get_linear_index(x+1,y,z,lblattice.halo_grid)].boundary,lbfields[get_linear_index(x,y-1,z,lblattice.halo_grid)].boundary,lbfields[get_linear_index(x,y+1,z,lblattice.halo_grid)].boundary,lbfields[get_linear_index(x,y,z-1,lblattice.halo_grid)].boundary,lbfields[get_linear_index(x,y,z+1,lblattice.halo_grid)].boundary);
//break;
//default:
//break;
//}
//}
//// DO THE SAME FOR THE CONSTRAINTS: CONSTRAINTS MUST BE SET AND THE SAME AS LB_BOUNDARY !!!
//for(n=0;n<n_constraints;n++) {
//switch(constraints[n].type) {
//case CONSTRAINT_VOXEL:
//x=constraints[n].c.voxel.pos[0];
//y=constraints[n].c.voxel.pos[1];
//z=constraints[n].c.voxel.pos[2];
//if(((x-1) >= 0) && (lbfields[get_linear_index(x-1,y,z,lblattice.halo_grid)].boundary == 0)) nn[0] = -1.0;//neighbours[0] = -1;
//if(((x+1) <= lblattice.grid[0]) && (lbfields[get_linear_index(x+1,y,z,lblattice.halo_grid)].boundary == 0)) nn[1] = 1.0;//neighbours[1] = 1;
////printf("%.0lf %.0lf ",nn[0],nn[1]);
//constraints[n].c.voxel.n[0] = nn[0]+nn[1];
////nn=0.0;
//if(((y-1) >= 0) && (lbfields[get_linear_index(x,y-1,z,lblattice.halo_grid)].boundary == 0)) nn[2] = -1.0;//neighbours[2] = -1;
//if(((y+1) <= lblattice.grid[1]) && (lbfields[get_linear_index(x,y+1,z,lblattice.halo_grid)].boundary == 0)) nn[3] = 1.0;//neighbours[3] = 1;
////printf("%.0lf %.0lf ",nn[2],nn[3]);
//constraints[n].c.voxel.n[1] = nn[2]+nn[3];
////nn=0.0;
//if(((z-1) >= 0) && (lbfields[get_linear_index(x,y,z-1,lblattice.halo_grid)].boundary == 0)) nn[4] = -1.0;//neighbours[4] = -1;
//if(((z+1) <= lblattice.grid[2]) && (lbfields[get_linear_index(x,y,z+1,lblattice.halo_grid)].boundary == 0)) nn[5] = 1.0;//neighbours[5]= 1;
////printf("%.0lf %.0lf ",nn[4],nn[5]);
//constraints[n].c.voxel.n[2] = nn[4]+nn[5];
//nn[0]=0.0,nn[1]=0.0,nn[2]=0.0,nn[3]=0.0,nn[4]=0.0,nn[5]=0.0;
//break;
//default:
//break;
//}
//}
//#ifdef VOXEL_BOUNDARIES
/*
for (z=0; z<lblattice.grid[2]+2; z++) {
for (y=0; y<lblattice.grid[1]+2; y++) {
for (x=0; x<lblattice.grid[0]+2; x++) {
lbfields[get_linear_index(x,y,z,lblattice.halo_grid)].boundary = 1;
}
}
}
static const char filename[] = "/home/mgusenbauer/Daten/Copy/DUK/GentlePump/Optimierer/voxels/stl/data_final.csv";
FILE *file = fopen ( filename, "r" );
int coords[3];
printf("start new\n");
if ( file != NULL ){
char line [ 128 ]; // or other suitable maximum line size
while ( fgets ( line, sizeof line, file ) != NULL ) {// read a line
//fputs ( line, stdout ); // write the line
//coords = line.Split(' ').Select(n => Convert.ToInt32(n)).ToArray();
//printf("readline: %s\n",line);
int i;
sscanf(line, "%d %d %d", &coords[0],&coords[1],&coords[2]);
//printf("%d %d %d\n", coords[0],coords[1],coords[2]);
lbfields[get_linear_index(coords[0]+5,coords[1]+5,coords[2]+5,lblattice.halo_grid)].boundary = 0;
}
fclose ( file );
}
printf("end new\n");
*/
#endif
}
}
/** Init cell interactions for the Lees-Edwards cell system.
* initializes the interacting neighbor cell list of a cell. The
* created list of interacting neighbor cells is used by the verlet
* algorithm (see verlet.cpp) to build the verlet lists.
*/
void le_dd_init_cell_interactions()
{
int m,n,o,p,q,r,ind1,ind2,c_cnt=0,n_cnt=0;
int extra_cells = 0;
/* initialize cell neighbor structures */
dd.cell_inter = (IA_Neighbor_List *) realloc(dd.cell_inter,local_cells.n*sizeof(IA_Neighbor_List));
for(m=0; m<local_cells.n; m++) {
dd.cell_inter[m].nList = NULL;
dd.cell_inter[m].n_neighbors=0;
}
/* loop over non-ghost cells */
for(o=1; o<=dd.cell_grid[2]; o++) {
for(n=1; n<=dd.cell_grid[1]; n++) {
for(m=1; m<=dd.cell_grid[0]; m++) {
/* plenty for most cases */
dd.cell_inter[c_cnt].nList = (IA_Neighbor *) realloc(dd.cell_inter[c_cnt].nList, 14*sizeof(IA_Neighbor));
n_cnt=0;
ind1 = get_linear_index(m,n,o,dd.ghost_cell_grid);
/* loop all 'conventional' neighbor cells */
for(p=o-1; p<=o+1; p++) { /*z-loop*/
for(q=n-1; q<=n+1; q++) { /*y-loop*/
for(r=m-1; r<=m+2; r++) { /*x-loop*/
/* Extra neighbours in x only for some cases */
if( (q == 0 && node_pos[1] == 0)
|| (q == dd.cell_grid[1]+1 && node_pos[1] == node_grid[1]-1) ){
extra_cells++;
dd.cell_inter[c_cnt].nList = (IA_Neighbor *) realloc(dd.cell_inter[c_cnt].nList, (extra_cells+14)*sizeof(IA_Neighbor));
}else{
if( r == m + 2 )
continue;
}
ind2 = get_linear_index(r,q,p,dd.ghost_cell_grid);
if(ind2 >= ind1) {
dd.cell_inter[c_cnt].nList[n_cnt].cell_ind = ind2;
dd.cell_inter[c_cnt].nList[n_cnt].pList = &cells[ind2];
init_pairList(&dd.cell_inter[c_cnt].nList[n_cnt].vList);
#ifdef LE_DEBUG
dd.cell_inter[c_cnt].nList[n_cnt].my_pos[0] = my_left[0] + r * dd.cell_size[0];
dd.cell_inter[c_cnt].nList[n_cnt].my_pos[1] = my_left[1] + q * dd.cell_size[1];
dd.cell_inter[c_cnt].nList[n_cnt].my_pos[2] = my_left[2] + p * dd.cell_size[2];
#endif
n_cnt++;
}
}
}
}
dd.cell_inter[c_cnt].n_neighbors = n_cnt;
c_cnt++;
}
}
}
#ifdef LE_DEBUG
FILE *cells_fp;
char cLogName[64];
int c,nn,this_n;
double myPos[3];
sprintf(cLogName, "cells_map%i.dat", this_node);
cells_fp = fopen(cLogName,"w");
/* print out line segments showing the vector from each cell to each neighbour cell*/
for(c=0;c<c_cnt;c++){
myPos[0] = my_left[0] + dd.cell_size[0] * ( 1 + c % dd.cell_grid[0] );
myPos[1] = my_left[1] + dd.cell_size[1] * ( 1 + (c / dd.cell_grid[0]) % dd.cell_grid[1]);
myPos[2] = my_left[2] + dd.cell_size[2] * ( 1 + (c / (dd.cell_grid[0] * dd.cell_grid[1])));
for(nn=0;nn<dd.cell_inter[c].n_neighbors;nn++){
this_n = dd.cell_inter[c].nList[nn].cell_ind;
fprintf(cells_fp,"%i %i %i %f %f %f %f %f %f\n",c,nn,this_n,
myPos[0], myPos[1], myPos[2],
dd.cell_inter[c].nList[nn].my_pos[0],
dd.cell_inter[c].nList[nn].my_pos[1],
dd.cell_inter[c].nList[nn].my_pos[2]);
}
}
fclose(cells_fp);
#endif
}
/** Initialize boundary conditions for all constraints in the system. */
void lb_init_boundaries() {
int n, x, y, z;
char *errtxt;
double pos[3], dist, dist_tmp=0.0, dist_vec[3];
if (lattice_switch & LATTICE_LB_GPU) {
#if defined (LB_GPU) && defined (LB_BOUNDARIES_GPU)
int number_of_boundnodes = 0;
int *host_boundary_node_list= (int*)malloc(sizeof(int));
int *host_boundary_index_list= (int*)malloc(sizeof(int));
size_t size_of_index;
int boundary_number = -1; // the number the boundary will actually belong to.
for(z=0; z<lbpar_gpu.dim_z; z++) {
for(y=0; y<lbpar_gpu.dim_y; y++) {
for (x=0; x<lbpar_gpu.dim_x; x++) {
pos[0] = (x+0.5)*lbpar_gpu.agrid;
pos[1] = (y+0.5)*lbpar_gpu.agrid;
pos[2] = (z+0.5)*lbpar_gpu.agrid;
dist = 1e99;
for (n=0;n<n_lb_boundaries;n++) {
switch (lb_boundaries[n].type) {
case LB_BOUNDARY_WAL:
calculate_wall_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.wal, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_SPH:
calculate_sphere_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.sph, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_CYL:
calculate_cylinder_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.cyl, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_RHOMBOID:
calculate_rhomboid_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.rhomboid, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_POR:
calculate_pore_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.pore, &dist_tmp, dist_vec);
break;
default:
errtxt = runtime_error(128);
ERROR_SPRINTF(errtxt, "{109 lbboundary type %d not implemented in lb_init_boundaries()\n", lb_boundaries[n].type);
}
if (dist > dist_tmp || n == 0) {
dist = dist_tmp;
boundary_number = n;
}
}
if (dist <= 0 && boundary_number >= 0 && n_lb_boundaries > 0) {
size_of_index = (number_of_boundnodes+1)*sizeof(int);
host_boundary_node_list = realloc(host_boundary_node_list, size_of_index);
host_boundary_index_list = realloc(host_boundary_index_list, size_of_index);
host_boundary_node_list[number_of_boundnodes] = x + lbpar_gpu.dim_x*y + lbpar_gpu.dim_x*lbpar_gpu.dim_y*z;
host_boundary_index_list[number_of_boundnodes] = boundary_number + 1;
number_of_boundnodes++;
// printf("boundindex %i: \n", number_of_boundnodes);
}
}
}
}
/**call of cuda fkt*/
float* boundary_velocity = malloc(3*n_lb_boundaries*sizeof(float));
for (n=0; n<n_lb_boundaries; n++) {
boundary_velocity[3*n+0]=lb_boundaries[n].velocity[0];
boundary_velocity[3*n+1]=lb_boundaries[n].velocity[1];
boundary_velocity[3*n+2]=lb_boundaries[n].velocity[2];
}
if (n_lb_boundaries)
lb_init_boundaries_GPU(n_lb_boundaries, number_of_boundnodes, host_boundary_node_list, host_boundary_index_list, boundary_velocity);
free(boundary_velocity);
free(host_boundary_node_list);
free(host_boundary_index_list);
#endif
} else {
#if defined (LB) && defined (LB_BOUNDARIES)
int node_domain_position[3], offset[3];
int the_boundary=-1;
map_node_array(this_node, node_domain_position);
offset[0] = node_domain_position[0]*lblattice.grid[0];
offset[1] = node_domain_position[1]*lblattice.grid[1];
offset[2] = node_domain_position[2]*lblattice.grid[2];
for (n=0;n<lblattice.halo_grid_volume;n++) {
lbfields[n].boundary = 0;
}
if (lblattice.halo_grid_volume==0)
return;
for (z=0; z<lblattice.grid[2]+2; z++) {
for (y=0; y<lblattice.grid[1]+2; y++) {
for (x=0; x<lblattice.grid[0]+2; x++) {
pos[0] = (offset[0]+(x-0.5))*lblattice.agrid;
pos[1] = (offset[1]+(y-0.5))*lblattice.agrid;
pos[2] = (offset[2]+(z-0.5))*lblattice.agrid;
dist = 1e99;
for (n=0;n<n_lb_boundaries;n++) {
switch (lb_boundaries[n].type) {
case LB_BOUNDARY_WAL:
calculate_wall_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.wal, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_SPH:
calculate_sphere_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.sph, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_CYL:
calculate_cylinder_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.cyl, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_RHOMBOID:
calculate_rhomboid_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.rhomboid, &dist_tmp, dist_vec);
break;
case LB_BOUNDARY_POR:
calculate_pore_dist((Particle*) NULL, pos, (Particle*) NULL, &lb_boundaries[n].c.pore, &dist_tmp, dist_vec);
break;
default:
errtxt = runtime_error(128);
ERROR_SPRINTF(errtxt, "{109 lbboundary type %d not implemented in lb_init_boundaries()\n", lb_boundaries[n].type);
}
if (dist_tmp<dist || n == 0) {
dist = dist_tmp;
the_boundary = n;
}
}
if (dist <= 0 && the_boundary >= 0 && n_lb_boundaries > 0) {
lbfields[get_linear_index(x,y,z,lblattice.halo_grid)].boundary = the_boundary+1;
//printf("boundindex %i: \n", get_linear_index(x,y,z,lblattice.halo_grid));
}
else {
lbfields[get_linear_index(x,y,z,lblattice.halo_grid)].boundary = 0;
}
}
}
}
#endif
}
}
int dfft_init(double **data,
int *local_mesh_dim, int *local_mesh_margin,
int* global_mesh_dim, double *global_mesh_off,
int *ks_pnum)
{
int i,j;
/* helpers */
int mult[3];
int n_grid[4][3]; /* The four node grids. */
int my_pos[4][3]; /* The position of this_node in the node grids. */
int *n_id[4]; /* linear node identity lists for the node grids. */
int *n_pos[4]; /* positions of nodes in the node grids. */
/* FFTW WISDOM stuff. */
char wisdom_file_name[255];
FILE *wisdom_file;
int wisdom_status;
FFT_TRACE(fprintf(stderr,"%d: dipolar dfft_init():\n",this_node));
dfft.max_comm_size=0; dfft.max_mesh_size=0;
for(i=0;i<4;i++) {
n_id[i] = (int *) malloc(1*n_nodes*sizeof(int));
n_pos[i] = (int *) malloc(3*n_nodes*sizeof(int));
}
/* === node grids === */
/* real space node grid (n_grid[0]) */
for(i=0;i<3;i++) {
n_grid[0][i] = node_grid[i];
my_pos[0][i] = node_pos[i];
}
for(i=0;i<n_nodes;i++) {
map_node_array(i,&(n_pos[0][3*i+0]));
n_id[0][get_linear_index( n_pos[0][3*i+0],n_pos[0][3*i+1],n_pos[0][3*i+2], n_grid[0])] = i;
}
/* FFT node grids (n_grid[1 - 3]) */
calc_2d_grid(n_nodes,n_grid[1]);
/* resort n_grid[1] dimensions if necessary */
dfft.plan[1].row_dir = map_3don2d_grid(n_grid[0], n_grid[1], mult);
dfft.plan[0].n_permute = 0;
for(i=1;i<4;i++) dfft.plan[i].n_permute = (dfft.plan[1].row_dir+i)%3;
for(i=0;i<3;i++) {
n_grid[2][i] = n_grid[1][(i+1)%3];
n_grid[3][i] = n_grid[1][(i+2)%3];
}
dfft.plan[2].row_dir = (dfft.plan[1].row_dir-1)%3;
dfft.plan[3].row_dir = (dfft.plan[1].row_dir-2)%3;
/* === communication groups === */
/* copy local mesh off real space charge assignment grid */
for(i=0;i<3;i++) dfft.plan[0].new_mesh[i] = local_mesh_dim[i];
for(i=1; i<4;i++) {
dfft.plan[i].g_size=fft_find_comm_groups(n_grid[i-1], n_grid[i], n_id[i-1], n_id[i],
dfft.plan[i].group, n_pos[i], my_pos[i]);
if(dfft.plan[i].g_size==-1) {
/* try permutation */
j = n_grid[i][(dfft.plan[i].row_dir+1)%3];
n_grid[i][(dfft.plan[i].row_dir+1)%3] = n_grid[i][(dfft.plan[i].row_dir+2)%3];
n_grid[i][(dfft.plan[i].row_dir+2)%3] = j;
dfft.plan[i].g_size=fft_find_comm_groups(n_grid[i-1], n_grid[i], n_id[i-1], n_id[i],
dfft.plan[i].group, n_pos[i], my_pos[i]);
if(dfft.plan[i].g_size==-1) {
fprintf(stderr,"%d: dipolar INTERNAL ERROR: fft_find_comm_groups error\n", this_node);
errexit();
}
}
dfft.plan[i].send_block = (int *)realloc(dfft.plan[i].send_block, 6*dfft.plan[i].g_size*sizeof(int));
dfft.plan[i].send_size = (int *)realloc(dfft.plan[i].send_size, 1*dfft.plan[i].g_size*sizeof(int));
dfft.plan[i].recv_block = (int *)realloc(dfft.plan[i].recv_block, 6*dfft.plan[i].g_size*sizeof(int));
dfft.plan[i].recv_size = (int *)realloc(dfft.plan[i].recv_size, 1*dfft.plan[i].g_size*sizeof(int));
dfft.plan[i].new_size = fft_calc_local_mesh(my_pos[i], n_grid[i], global_mesh_dim,
global_mesh_off, dfft.plan[i].new_mesh,
dfft.plan[i].start);
permute_ifield(dfft.plan[i].new_mesh,3,-(dfft.plan[i].n_permute));
permute_ifield(dfft.plan[i].start,3,-(dfft.plan[i].n_permute));
dfft.plan[i].n_ffts = dfft.plan[i].new_mesh[0]*dfft.plan[i].new_mesh[1];
/* === send/recv block specifications === */
for(j=0; j<dfft.plan[i].g_size; j++) {
int k, node;
/* send block: this_node to comm-group-node i (identity: node) */
node = dfft.plan[i].group[j];
dfft.plan[i].send_size[j]
= fft_calc_send_block(my_pos[i-1], n_grid[i-1], &(n_pos[i][3*node]), n_grid[i],
global_mesh_dim, global_mesh_off, &(dfft.plan[i].send_block[6*j]));
permute_ifield(&(dfft.plan[i].send_block[6*j]),3,-(dfft.plan[i-1].n_permute));
permute_ifield(&(dfft.plan[i].send_block[6*j+3]),3,-(dfft.plan[i-1].n_permute));
if(dfft.plan[i].send_size[j] > dfft.max_comm_size)
dfft.max_comm_size = dfft.plan[i].send_size[j];
/* First plan send blocks have to be adjusted, since the CA grid
may have an additional margin outside the actual domain of the
node */
if(i==1) {
for(k=0;k<3;k++)
dfft.plan[1].send_block[6*j+k ] += local_mesh_margin[2*k];
}
/* recv block: this_node from comm-group-node i (identity: node) */
dfft.plan[i].recv_size[j]
= fft_calc_send_block(my_pos[i], n_grid[i], &(n_pos[i-1][3*node]), n_grid[i-1],
global_mesh_dim, global_mesh_off,&(dfft.plan[i].recv_block[6*j]));
permute_ifield(&(dfft.plan[i].recv_block[6*j]),3,-(dfft.plan[i].n_permute));
permute_ifield(&(dfft.plan[i].recv_block[6*j+3]),3,-(dfft.plan[i].n_permute));
if(dfft.plan[i].recv_size[j] > dfft.max_comm_size)
dfft.max_comm_size = dfft.plan[i].recv_size[j];
}
for(j=0;j<3;j++) dfft.plan[i].old_mesh[j] = dfft.plan[i-1].new_mesh[j];
if(i==1)
dfft.plan[i].element = 1;
else {
dfft.plan[i].element = 2;
for(j=0; j<dfft.plan[i].g_size; j++) {
dfft.plan[i].send_size[j] *= 2;
dfft.plan[i].recv_size[j] *= 2;
}
}
/* DEBUG */
for(j=0;j<n_nodes;j++) {
/* MPI_Barrier(comm_cart); */
if(j==this_node) FFT_TRACE(fft_print_fft_plan(dfft.plan[i]));
}
}
/* Factor 2 for complex fields */
dfft.max_comm_size *= 2;
dfft.max_mesh_size = (local_mesh_dim[0]*local_mesh_dim[1]*local_mesh_dim[2]);
for(i=1;i<4;i++)
if(2*dfft.plan[i].new_size > dfft.max_mesh_size) dfft.max_mesh_size = 2*dfft.plan[i].new_size;
FFT_TRACE(fprintf(stderr,"%d: dfft.max_comm_size = %d, dfft.max_mesh_size = %d\n",
this_node,dfft.max_comm_size,dfft.max_mesh_size));
/* === pack function === */
for(i=1;i<4;i++) {
dfft.plan[i].pack_function = fft_pack_block_permute2;
FFT_TRACE(fprintf(stderr,"%d: forw plan[%d] permute 2 \n",this_node,i));
}
(*ks_pnum)=6;
if(dfft.plan[1].row_dir==2) {
dfft.plan[1].pack_function = fft_pack_block;
FFT_TRACE(fprintf(stderr,"%d: forw plan[%d] permute 0 \n",this_node,1));
(*ks_pnum)=4;
}
else if(dfft.plan[1].row_dir==1) {
dfft.plan[1].pack_function = fft_pack_block_permute1;
FFT_TRACE(fprintf(stderr,"%d: forw plan[%d] permute 1 \n",this_node,1));
(*ks_pnum)=5;
}
/* Factor 2 for complex numbers */
dfft.send_buf = (double *)realloc(dfft.send_buf, dfft.max_comm_size*sizeof(double));
dfft.recv_buf = (double *)realloc(dfft.recv_buf, dfft.max_comm_size*sizeof(double));
(*data) = (double *)realloc((*data), dfft.max_mesh_size*sizeof(double));
dfft.data_buf = (double *)realloc(dfft.data_buf, dfft.max_mesh_size*sizeof(double));
if(!(*data) || !dfft.data_buf || !dfft.recv_buf || !dfft.send_buf) {
fprintf(stderr,"%d: Could not allocate FFT data arays\n",this_node);
errexit();
}
fftw_complex *c_data = (fftw_complex *) (*data);
/* === FFT Routines (Using FFTW / RFFTW package)=== */
for(i=1;i<4;i++) {
dfft.plan[i].dir = FFTW_FORWARD;
/* FFT plan creation.
Attention: destroys contents of c_data/data and c_data_buf/data_buf. */
wisdom_status = FFTW_FAILURE;
sprintf(wisdom_file_name,"dfftw3_1d_wisdom_forw_n%d.file",
dfft.plan[i].new_mesh[2]);
if( (wisdom_file=fopen(wisdom_file_name,"r"))!=NULL ) {
wisdom_status = fftw_import_wisdom_from_file(wisdom_file);
fclose(wisdom_file);
}
if(dfft.init_tag==1) fftw_destroy_plan(dfft.plan[i].our_fftw_plan);
//printf("dfft.plan[%d].n_ffts=%d\n",i,dfft.plan[i].n_ffts);
dfft.plan[i].our_fftw_plan =
fftw_plan_many_dft(1,&dfft.plan[i].new_mesh[2],dfft.plan[i].n_ffts,
c_data,NULL,1,dfft.plan[i].new_mesh[2],
c_data,NULL,1,dfft.plan[i].new_mesh[2],
dfft.plan[i].dir,FFTW_PATIENT);
if( wisdom_status == FFTW_FAILURE &&
(wisdom_file=fopen(wisdom_file_name,"w"))!=NULL ) {
fftw_export_wisdom_to_file(wisdom_file);
fclose(wisdom_file);
}
dfft.plan[i].fft_function = fftw_execute;
}
/* === The BACK Direction === */
/* this is needed because slightly different functions are used */
for(i=1;i<4;i++) {
dfft.back[i].dir = FFTW_BACKWARD;
wisdom_status = FFTW_FAILURE;
sprintf(wisdom_file_name,"dfftw3_1d_wisdom_back_n%d.file",
dfft.plan[i].new_mesh[2]);
if( (wisdom_file=fopen(wisdom_file_name,"r"))!=NULL ) {
wisdom_status = fftw_import_wisdom_from_file(wisdom_file);
fclose(wisdom_file);
}
if(dfft.init_tag==1) fftw_destroy_plan(dfft.back[i].our_fftw_plan);
dfft.back[i].our_fftw_plan =
fftw_plan_many_dft(1,&dfft.plan[i].new_mesh[2],dfft.plan[i].n_ffts,
c_data,NULL,1,dfft.plan[i].new_mesh[2],
c_data,NULL,1,dfft.plan[i].new_mesh[2],
dfft.back[i].dir,FFTW_PATIENT);
if( wisdom_status == FFTW_FAILURE &&
(wisdom_file=fopen(wisdom_file_name,"w"))!=NULL ) {
fftw_export_wisdom_to_file(wisdom_file);
fclose(wisdom_file);
}
dfft.back[i].fft_function = fftw_execute;
dfft.back[i].pack_function = fft_pack_block_permute1;
FFT_TRACE(fprintf(stderr,"%d: back plan[%d] permute 1 \n",this_node,i));
}
if(dfft.plan[1].row_dir==2) {
dfft.back[1].pack_function = fft_pack_block;
FFT_TRACE(fprintf(stderr,"%d: back plan[%d] permute 0 \n",this_node,1));
}
else if(dfft.plan[1].row_dir==1) {
dfft.back[1].pack_function = fft_pack_block_permute2;
FFT_TRACE(fprintf(stderr,"%d: back plan[%d] permute 2 \n",this_node,1));
}
dfft.init_tag=1;
/* free(data); */
for(i=0;i<4;i++) { free(n_id[i]); free(n_pos[i]); }
return dfft.max_mesh_size;
}