int main(int argc,char *argv[])
{
FILE *fp;
int nnn[] = { 8, 10, 12, 15, 16, 18, 20, 24, 25, 27, 30, 32, 36, 40,
45, 48, 50, 54, 60, 64, 72, 75, 80, 81, 90, 100 };
#define NNN asize(nnn)
int *niter;
int i,j,n,nit,ntot,n3;
double t,nflop;
double *rt,*ct;
t_complex ***g;
real ***h;
snew(rt,NNN);
snew(ct,NNN);
snew(niter,NNN);
for(i=0; (i<NNN); i++) {
n = nnn[i];
fprintf(stderr,"\rReal %d ",n);
if (n < 16)
niter[i] = 100;
else if (n < 26)
niter[i] = 50;
else if (n < 51)
niter[i] = 10;
else
niter[i] = 5;
nit = niter[i];
h = mk_rgrid(n+2,n,n);
start_time();
for(j=0; (j<nit); j++) {
testrft(stdout,h,n,n,n,(j==0));
}
update_time();
rt[i] = node_time();
free_rgrid(h,n,n);
fprintf(stderr,"\rComplex %d ",n);
g = mk_cgrid(n,n,n);
start_time();
for(j=0; (j<nit); j++) {
testfft(stdout,g,n,n,n,(j==0));
}
update_time();
ct[i] = node_time();
free_cgrid(g,n,n);
}
fprintf(stderr,"\n");
fp=xvgropen("timing.xvg","FFT timings per grid point","n","t (s)");
for(i=0; (i<NNN); i++) {
n3 = 2*niter[i]*nnn[i]*nnn[i]*nnn[i];
fprintf(fp,"%10d %10g %10g\n",nnn[i],rt[i]/n3,ct[i]/n3);
}
gmx_fio_fclose(fp);
return 0;
}
real do_opt_pppm(FILE *log, bool bVerbose,
t_inputrec *ir, int natoms,
rvec x[], rvec f[],
real charge[], rvec box,
real phi[], t_commrec *cr,
t_nrnb *nrnb, rvec beta,
t_fftgrid *grid, bool bOld)
{
real ***ghat;
int nx,ny,nz;
real ener;
ener = 0.0;
fprintf(log,"Generating Ghat function\n");
nx = ir->nkx;
ny = ir->nky;
nz = ir->nkz;
ghat = mk_rgrid(nx,ny,nz);
mk_ghat(NULL,nx,ny,nz,ghat,box,ir->rcoulomb_switch,ir->rcoulomb,TRUE,bOld);
/* pr_scalar_gk("generghat.xvg",nx,ny,nz,box,ghat); */
/* Now start the actual PPPM procedure.
* First step: spreading the charges over the grid.
*/
/* Make the grid empty */
clear_fftgrid(grid);
spread_q(log,bVerbose,0,natoms,x,charge,box,grid,nrnb);
/* Second step: solving the poisson equation in Fourier space */
solve_pppm(log,cr,grid,ghat,box,bVerbose,nrnb);
/* Third and last step: gather the forces, energies and potential
* from the grid.
*/
ener=gather_f(log,bVerbose,0,natoms,x,f,charge,box,phi,grid,beta,nrnb);
free_rgrid(ghat,nx,ny);
return ener;
}
void init_pppm(FILE *log,t_commrec *cr,t_nsborder *nsb,
bool bVerbose,bool bOld,rvec box,char *ghatfn,t_inputrec *ir)
{
int nx,ny,nz,m,porder;
ivec grids;
real r1,rc;
const real tol = 1e-5;
rvec spacing;
#ifdef WITHOUT_FFTW
fatal_error(0,"PPPM used, but GROMACS was compiled without FFTW support!\n");
#endif
if (cr != NULL) {
if (cr->nnodes > 1)
fprintf(log,"Initializing parallel PPPM.\n");
}
fprintf(log,"Will use the PPPM algorithm for long-range electrostatics\n");
if (!fexist(ghatfn)) {
beta[XX]=beta[YY]=beta[ZZ]= 1.85;
nx = ir->nkx;
ny = ir->nky;
nz = ir->nkz;
fprintf(log,"Generating Ghat function\n");
fprintf(log,"Grid size is %d x %d x %d\n",nx,ny,nz);
if ((nx < 4) || (ny < 4) || (nz < 4))
fatal_error(0,"Grid must be at least 4 points in all directions");
ghat = mk_rgrid(nx,ny,nz);
mk_ghat(NULL,nx,ny,nz,ghat,box,ir->rcoulomb_switch,ir->rcoulomb,TRUE,bOld);
if (bVerbose)
pr_scalar_gk("generghat.xvg",nx,ny,nz,box,ghat);
}
else {
fprintf(stderr,"Reading Ghat function from %s\n",ghatfn);
ghat = rd_ghat(log,ghatfn,grids,spacing,beta,&porder,&r1,&rc);
/* Check whether cut-offs correspond */
if ((fabs(r1-ir->rcoulomb_switch)>tol) || (fabs(rc-ir->rcoulomb)>tol)) {
fprintf(log,"rcoulomb_switch = %10.3e rcoulomb = %10.3e"
" r1 = %10.3e rc = %10.3e\n",
ir->rcoulomb_switch,ir->rcoulomb,r1,rc);
fflush(log);
fatal_error(0,"Cut-off lengths in tpb file and Ghat file %s "
"do not match\nCheck your log file!",ghatfn);
}
/* Check whether boxes correspond */
for(m=0; (m<DIM); m++)
if (fabs(box[m]-grids[m]*spacing[m]) > tol) {
pr_rvec(log,0,"box",box,DIM);
pr_rvec(log,0,"grid-spacing",spacing,DIM);
pr_ivec(log,0,"grid size",grids,DIM);
fflush(log);
fatal_error(0,"Box sizes in tpb file and Ghat file %s do not match\n"
"Check your log file!",ghatfn);
}
if (porder != 2)
fatal_error(0,"porder = %d, should be 2 in %s",porder,ghatfn);
nx = grids[XX];
ny = grids[YY];
nz = grids[ZZ];
if (bVerbose)
pr_scalar_gk("optimghat.xvg",nx,ny,nz,box,ghat);
}
/* Now setup the FFT things */
grid = mk_fftgrid(log,PAR(cr),nx,ny,nz,ir->bOptFFT);
}
int gmx_pppm_init(FILE *log, t_commrec *cr,
const output_env_t oenv, gmx_bool bVerbose,
gmx_bool bOld, matrix box,
char *ghatfn, t_inputrec *ir,
gmx_bool bReproducible)
{
int nx,ny,nz,m,porder;
ivec grids;
real r1,rc;
const real tol = 1e-5;
rvec box_diag,spacing;
#ifdef DISABLE_PPPM
gmx_fatal(FARGS,"PPPM is not functional in the current version, we plan to implement PPPM through a small modification of the PME code.");
return -1;
#else
#ifdef GMX_WITHOUT_FFTW
gmx_fatal(FARGS,"PPPM used, but GROMACS was compiled without FFTW support!\n");
#endif
if (log) {
if (cr != NULL) {
if (cr->nnodes > 1)
fprintf(log,"Initializing parallel PPPM.\n");
}
fprintf(log,
"Will use the PPPM algorithm for long-range electrostatics\n");
}
for(m=0; m<DIM; m++)
box_diag[m] = box[m][m];
if (!gmx_fexist(ghatfn)) {
beta[XX]=beta[YY]=beta[ZZ]= 1.85;
nx = ir->nkx;
ny = ir->nky;
nz = ir->nkz;
if (log) {
fprintf(log,"Generating Ghat function\n");
fprintf(log,"Grid size is %d x %d x %d\n",nx,ny,nz);
}
if ((nx < 4) || (ny < 4) || (nz < 4))
gmx_fatal(FARGS,"Grid must be at least 4 points in all directions");
ghat = mk_rgrid(nx,ny,nz);
mk_ghat(NULL,nx,ny,nz,ghat,box_diag,
ir->rcoulomb_switch,ir->rcoulomb,TRUE,bOld);
if (bVerbose)
pr_scalar_gk("generghat.xvg",oenv,nx,ny,nz,box_diag,ghat);
}
else {
fprintf(stderr,"Reading Ghat function from %s\n",ghatfn);
ghat = rd_ghat(log,oenv,ghatfn,grids,spacing,beta,&porder,&r1,&rc);
/* Check whether cut-offs correspond */
if ((fabs(r1-ir->rcoulomb_switch)>tol) || (fabs(rc-ir->rcoulomb)>tol)) {
if (log) {
fprintf(log,"rcoulomb_switch = %10.3e rcoulomb = %10.3e"
" r1 = %10.3e rc = %10.3e\n",
ir->rcoulomb_switch,ir->rcoulomb,r1,rc);
fflush(log);
}
gmx_fatal(FARGS,"Cut-off lengths in tpb file and Ghat file %s "
"do not match\nCheck your log file!",ghatfn);
}
/* Check whether boxes correspond */
for(m=0; (m<DIM); m++)
if (fabs(box_diag[m]-grids[m]*spacing[m]) > tol) {
if (log) {
pr_rvec(log,0,"box",box_diag,DIM,TRUE);
pr_rvec(log,0,"grid-spacing",spacing,DIM,TRUE);
pr_ivec(log,0,"grid size",grids,DIM,TRUE);
fflush(log);
}
gmx_fatal(FARGS,"Box sizes in tpb file and Ghat file %s do not match\n"
"Check your log file!",ghatfn);
}
if (porder != 2)
gmx_fatal(FARGS,"porder = %d, should be 2 in %s",porder,ghatfn);
nx = grids[XX];
ny = grids[YY];
nz = grids[ZZ];
if (bVerbose)
pr_scalar_gk("optimghat.xvg",oenv,nx,ny,nz,box_diag,ghat);
}
/* Now setup the FFT things */
#ifdef GMX_MPI
cr->mpi_comm_mygroup=cr->mpi_comm_mysim;
#endif
grid = mk_fftgrid(nx,ny,nz,NULL,NULL,cr,bReproducible);
return 0;
#endif
}
