real do_pppm(FILE *log, bool bVerbose,
rvec x[], rvec f[],
real charge[], rvec box,
real phi[], t_commrec *cr,
t_nsborder *nsb, t_nrnb *nrnb)
{
real ener;
int start,nr;
start = START(nsb);
nr = HOMENR(nsb);
/* Make the grid empty */
clear_fftgrid(grid);
/* First step: spreading the charges over the grid. */
spread_q(log,bVerbose,start,nr,x,charge,box,grid,nrnb);
/* In the parallel code we have to sum the grids from neighbouring nodes */
if (PAR(cr))
sum_qgrid(cr,nsb,grid,TRUE);
/* Second step: solving the poisson equation in Fourier space */
solve_pppm(log,cr,grid,ghat,box,bVerbose,nrnb);
/* In the parallel code we have to sum once again... */
if (PAR(cr))
sum_qgrid(cr,nsb,grid,FALSE);
/* Third and last step: gather the forces, energies and potential
* from the grid.
*/
ener=gather_f(log,bVerbose,start,nr,x,f,charge,box,phi,grid,beta,nrnb);
return ener;
}
static void do_my_pme(FILE *fp,real tm,gmx_bool bVerbose,t_inputrec *ir,
rvec x[],rvec xbuf[],rvec f[],
real charge[],real qbuf[],real qqbuf[],
matrix box,gmx_bool bSort,
t_commrec *cr,t_nsborder *nsb,t_nrnb *nrnb,
t_block *excl,real qtot,
t_forcerec *fr,int index[],FILE *fp_xvg,
int ngroups,unsigned short cENER[])
{
real ener,vcorr,q,xx,dvdl=0,vdip,vcharge;
tensor vir,vir_corr,vir_tot;
rvec mu_tot[2];
int i,m,ii,ig,jg;
real **epme,*qptr;
/* Initiate local variables */
fr->f_el_recip = f;
clear_mat(vir);
clear_mat(vir_corr);
if (ngroups > 1) {
fprintf(fp,"There are %d energy groups\n",ngroups);
snew(epme,ngroups);
for(i=0; (i<ngroups); i++)
snew(epme[i],ngroups);
}
/* Put x is in the box, this part needs to be parallellized properly */
/*put_atoms_in_box(box,nsb->natoms,x);*/
/* Here sorting of X (and q) is done.
* Alternatively, one could just put the atoms in one of the
* cr->nnodes slabs. That is much cheaper than sorting.
*/
for(i=0; (i<nsb->natoms); i++)
index[i] = i;
if (bSort) {
xptr = x;
qsort(index,nsb->natoms,sizeof(index[0]),comp_xptr);
xptr = NULL; /* To trap unintentional use of the ptr */
}
/* After sorting we only need the part that is to be computed on
* this processor. We also compute the mu_tot here (system dipole)
*/
clear_rvec(mu_tot[0]);
for(i=START(nsb); (i<START(nsb)+HOMENR(nsb)); i++) {
ii = index[i];
q = charge[ii];
qbuf[i] = q;
for(m=0; (m<DIM); m++) {
xx = x[ii][m];
xbuf[i][m] = xx;
mu_tot[0][m] += q*xx;
}
clear_rvec(f[ii]);
}
copy_rvec(mu_tot[0],mu_tot[1]);
if (debug) {
pr_rvec(debug,0,"qbuf",qbuf,nsb->natoms,TRUE);
pr_rvecs(debug,0,"xbuf",xbuf,nsb->natoms);
pr_rvecs(debug,0,"box",box,DIM);
}
for(ig=0; (ig<ngroups); ig++) {
for(jg=ig; (jg<ngroups); jg++) {
if (ngroups > 1) {
for(i=START(nsb); (i<START(nsb)+HOMENR(nsb)); i++) {
if ((cENER[i] == ig) || (cENER[i] == jg))
qqbuf[i] = qbuf[i];
else
qqbuf[i] = 0;
}
qptr = qqbuf;
}
else
qptr = qbuf;
ener = do_pme(fp,bVerbose,ir,xbuf,f,qptr,qptr,box,cr,
nsb,nrnb,vir,fr->ewaldcoeff,FALSE,0,&dvdl,FALSE);
vcorr = ewald_LRcorrection(fp,nsb,cr,fr,qptr,qptr,excl,xbuf,box,mu_tot,
ir->ewald_geometry,ir->epsilon_surface,
0,&dvdl,&vdip,&vcharge);
gmx_sum(1,&ener,cr);
gmx_sum(1,&vcorr,cr);
if (ngroups > 1)
epme[ig][jg] = ener+vcorr;
}
}
if (ngroups > 1) {
if (fp_xvg)
fprintf(fp_xvg,"%10.3f",tm);
for(ig=0; (ig<ngroups); ig++) {
for(jg=ig; (jg<ngroups); jg++) {
if (ig != jg)
epme[ig][jg] -= epme[ig][ig]+epme[jg][jg];
if (fp_xvg)
fprintf(fp_xvg," %12.5e",epme[ig][jg]);
}
}
if (fp_xvg)
fprintf(fp_xvg,"\n");
}
else {
fprintf(fp,"Time: %10.3f Energy: %12.5e Correction: %12.5e Total: %12.5e\n",
tm,ener,vcorr,ener+vcorr);
if (fp_xvg)
fprintf(fp_xvg,"%10.3f %12.5e %12.5e %12.5e\n",tm,ener+vcorr,vdip,vcharge);
if (bVerbose) {
m_add(vir,vir_corr,vir_tot);
gmx_sum(9,vir_tot[0],cr);
pr_rvecs(fp,0,"virial",vir_tot,DIM);
}
fflush(fp);
}
}
real do_ewald(FILE *log, bool bVerbose,
t_inputrec *ir,
rvec x[], rvec f[],
real charge[], rvec box,
t_commrec *cr, t_nsborder *nsb,
matrix lrvir, real ewaldcoeff)
{
static bool bFirst = TRUE;
static int nx,ny,nz,kmax;
static cvec **eir;
static t_complex *tab_xy,*tab_qxyz;
real factor=-1.0/(4*ewaldcoeff*ewaldcoeff);
real energy;
rvec lll;
int lowiy,lowiz,ix,iy,iz,n;
real tmp,cs,ss,ak,akv,mx,my,mz,m2;
if (bFirst) {
if (bVerbose)
fprintf(log,"Will do ordinary reciprocal space Ewald sum.\n");
if (cr != NULL) {
if (cr->nnodes > 1 || cr->nthreads>1)
fatal_error(0,"No parallel Ewald. Use PME instead.\n");
}
nx = ir->nkx+1;
ny = ir->nky+1;
nz = ir->nkz+1;
kmax = max(nx,max(ny,nz));
snew(eir,kmax);
for(n=0;n<kmax;n++)
snew(eir[n],HOMENR(nsb));
snew(tab_xy,HOMENR(nsb));
snew(tab_qxyz,HOMENR(nsb));
bFirst = FALSE;
}
clear_mat(lrvir);
calc_lll(box,lll);
/* make tables for the structure factor parts */
tabulate_eir(HOMENR(nsb),x,kmax,eir,lll);
lowiy=0;
lowiz=1;
energy=0;
for(ix=0;ix<nx;ix++) {
mx=ix*lll[XX];
for(iy=lowiy;iy<ny;iy++) {
my=iy*lll[YY];
if(iy>=0)
for(n=0;n<HOMENR(nsb);n++)
tab_xy[n]=cmul(eir[ix][n][XX],eir[iy][n][YY]);
else
for(n=0;n<HOMENR(nsb);n++)
tab_xy[n]=conjmul(eir[ix][n][XX],eir[-iy][n][YY]);
for(iz=lowiz;iz<nz;iz++) {
mz=iz*lll[ZZ];
m2=mx*mx+my*my+mz*mz;
ak=exp(m2*factor)/m2;
akv=2.0*ak*(1.0/m2-factor);
if(iz>=0)
for(n=0;n<HOMENR(nsb);n++)
tab_qxyz[n]=rcmul(charge[n],cmul(tab_xy[n],eir[iz][n][ZZ]));
else
for(n=0;n<HOMENR(nsb);n++)
tab_qxyz[n]=rcmul(charge[n],conjmul(tab_xy[n],eir[-iz][n][ZZ]));
cs=ss=0;
for(n=0;n<HOMENR(nsb);n++) {
cs+=tab_qxyz[n].re;
ss+=tab_qxyz[n].im;
}
energy+=ak*(cs*cs+ss*ss);
tmp=akv*(cs*cs+ss*ss);
lrvir[XX][XX]-=tmp*mx*mx;
lrvir[XX][YY]-=tmp*mx*my;
lrvir[XX][ZZ]-=tmp*mx*mz;
lrvir[YY][YY]-=tmp*my*my;
lrvir[YY][ZZ]-=tmp*my*mz;
lrvir[ZZ][ZZ]-=tmp*mz*mz;
for(n=0;n<HOMENR(nsb);n++) {
tmp=ak*(cs*tab_qxyz[n].im-ss*tab_qxyz[n].re);
f[n][XX]+=tmp*mx;
f[n][YY]+=tmp*my;
f[n][ZZ]+=tmp*mz;
}
lowiz=1-nz;
}
lowiy=1-ny;
}
}
tmp=4.0*M_PI/(box[XX]*box[YY]*box[ZZ])*ONE_4PI_EPS0;
for(n=0;n<HOMENR(nsb);n++) {
f[n][XX]*=2*tmp;
f[n][YY]*=2*tmp;
f[n][ZZ]*=2*tmp;
}
lrvir[XX][XX]=-0.5*tmp*(lrvir[XX][XX]+energy);
lrvir[XX][YY]=-0.5*tmp*(lrvir[XX][YY]);
lrvir[XX][ZZ]=-0.5*tmp*(lrvir[XX][ZZ]);
lrvir[YY][YY]=-0.5*tmp*(lrvir[YY][YY]+energy);
lrvir[YY][ZZ]=-0.5*tmp*(lrvir[YY][ZZ]);
lrvir[ZZ][ZZ]=-0.5*tmp*(lrvir[ZZ][ZZ]+energy);
lrvir[YY][XX]=lrvir[XX][YY];
lrvir[ZZ][XX]=lrvir[XX][ZZ];
lrvir[ZZ][YY]=lrvir[YY][ZZ];
energy*=tmp;
return energy;
}
void init_md(t_commrec *cr,t_inputrec *ir,tensor box,real *t,real *t0,
real *lambda,real *lam0,real *SAfactor,
t_nrnb *mynrnb,bool *bTYZ,t_topology *top,
int nfile,t_filenm fnm[],char **traj,
char **xtc_traj,int *fp_ene,
FILE **fp_dgdl,t_mdebin **mdebin,t_groups *grps,
tensor force_vir,tensor pme_vir,
tensor shake_vir,t_mdatoms *mdatoms,rvec mu_tot,
bool *bNEMD,t_vcm **vcm,t_nsborder *nsb)
{
bool bBHAM,b14,bLR,bLJLR;
int i;
/* Initial values */
*t = *t0 = ir->init_t;
if (ir->efep != efepNO) {
*lambda = *lam0 = ir->init_lambda;
}
else {
*lambda = *lam0 = 0.0;
}
if (ir->bSimAnn) {
*SAfactor = 1.0 - *t0/ir->zero_temp_time;
if (*SAfactor < 0)
*SAfactor = 0;
} else
*SAfactor = 1.0;
init_nrnb(mynrnb);
/* Check Environment variables & other booleans */
#ifdef SPEC_CPU
*bTYZ = FALSE;
#else
*bTYZ=getenv("TYZ") != NULL;
#endif
set_pot_bools(ir,top,&bLR,&bLJLR,&bBHAM,&b14);
if (nfile != -1) {
/* Filenames */
*traj = ftp2fn(efTRN,nfile,fnm);
*xtc_traj = ftp2fn(efXTC,nfile,fnm);
#ifndef SPEC_CPU
if (MASTER(cr)) {
*fp_ene = open_enx(ftp2fn(efENX,nfile,fnm),"w");
if ((fp_dgdl != NULL) && ir->efep!=efepNO)
*fp_dgdl =
xvgropen(opt2fn("-dgdl",nfile,fnm),
"dG/d\\8l\\4","Time (ps)",
"dG/d\\8l\\4 (kJ mol\\S-1\\N nm\\S-2\\N \\8l\\4\\S-1\\N)");
} else
#endif
*fp_ene = -1;
*mdebin = init_mdebin(*fp_ene,grps,&(top->atoms),&(top->idef),
bLR,bLJLR,bBHAM,b14,ir->efep!=efepNO,ir->epc,
ir->eDispCorr,(TRICLINIC(ir->compress) || TRICLINIC(box)),
(ir->etc==etcNOSEHOOVER),cr);
}
/* Initiate variables */
clear_mat(force_vir);
clear_mat(pme_vir);
clear_mat(shake_vir);
clear_rvec(mu_tot);
/* Set initial values for invmass etc. */
init_mdatoms(mdatoms,*lambda,TRUE);
*vcm = init_vcm(stdlog,top,cr,mdatoms,START(nsb),HOMENR(nsb),ir->nstcomm);
debug_gmx();
*bNEMD = (ir->opts.ngacc > 1) || (norm(ir->opts.acc[0]) > 0);
if (ir->eI == eiSD)
init_sd_consts(ir->opts.ngtc,ir->opts.tau_t,ir->delta_t);
}
void do_shakefirst(FILE *log,bool bTYZ,real lambda,real ener[],
t_parm *parm,t_nsborder *nsb,t_mdatoms *md,
rvec x[],rvec vold[],rvec buf[],rvec f[],
rvec v[],t_graph *graph,t_commrec *cr,t_nrnb *nrnb,
t_groups *grps,t_forcerec *fr,t_topology *top,
t_edsamyn *edyn,t_pull *pulldata)
{
int i,m,start,homenr,end,step;
tensor shake_vir;
double mass,tmass,vcm[4];
real dt=parm->ir.delta_t;
real dt_1;
if (count_constraints(top,cr)) {
start = START(nsb);
homenr = HOMENR(nsb);
end = start+homenr;
if (debug)
fprintf(debug,"vcm: start=%d, homenr=%d, end=%d\n",start,homenr,end);
/* Do a first SHAKE to reset particles... */
step = -2;
if(log)
fprintf(log,"\nConstraining the starting coordinates (step %d)\n",step);
clear_mat(shake_vir);
update(nsb->natoms,start,homenr,step,lambda,&ener[F_DVDL],
parm,1.0,md,x,graph,
NULL,NULL,vold,NULL,x,top,grps,shake_vir,cr,nrnb,bTYZ,
FALSE,edyn,pulldata,FALSE);
/* Compute coordinates at t=-dt, store them in buf */
/* for(i=0; (i<nsb->natoms); i++) {*/
for(i=start; (i<end); i++) {
for(m=0; (m<DIM); m++) {
f[i][m]=x[i][m];
buf[i][m]=x[i][m]-dt*v[i][m];
}
}
/* Shake the positions at t=-dt with the positions at t=0
* as reference coordinates.
*/
step = -1;
if(log)
fprintf(log,"\nConstraining the coordinates at t0-dt (step %d)\n",step);
clear_mat(shake_vir);
update(nsb->natoms,start,homenr,
step,lambda,&ener[F_DVDL],parm,1.0,md,f,graph,
NULL,NULL,vold,NULL,buf,top,grps,shake_vir,cr,nrnb,bTYZ,FALSE,
edyn,pulldata,FALSE);
/* Compute the velocities at t=-dt/2 using the coordinates at
* t=-dt and t=0
* Compute velocity of center of mass and total mass
*/
for(m=0; (m<4); m++)
vcm[m] = 0;
dt_1=1.0/dt;
for(i=start; (i<end); i++) {
/*for(i=0; (i<nsb->natoms); i++) {*/
mass = md->massA[i];
for(m=0; (m<DIM); m++) {
v[i][m]=(x[i][m]-f[i][m])*dt_1;
vcm[m] += v[i][m]*mass;
}
vcm[3] += mass;
}
/* Compute the global sum of vcm */
if (debug)
fprintf(debug,"vcm: %8.3f %8.3f %8.3f,"
" total mass = %12.5e\n",vcm[XX],vcm[YY],vcm[ZZ],vcm[3]);
if (PAR(cr))
gmx_sumd(4,vcm,cr);
tmass = vcm[3];
for(m=0; (m<DIM); m++)
vcm[m] /= tmass;
if (debug)
fprintf(debug,"vcm: %8.3f %8.3f %8.3f,"
" total mass = %12.5e\n",vcm[XX],vcm[YY],vcm[ZZ],tmass);
/* Now we have the velocity of center of mass, let's remove it */
for(i=start; (i<end); i++) {
for(m=0; (m<DIM); m++)
v[i][m] -= vcm[m];
}
}
}
void do_force(FILE *log,t_commrec *cr,t_commrec *mcr,
t_parm *parm,t_nsborder *nsb,tensor vir_part,tensor pme_vir,
int step,t_nrnb *nrnb,t_topology *top,t_groups *grps,
rvec x[],rvec v[],rvec f[],rvec buf[],
t_mdatoms *mdatoms,real ener[],t_fcdata *fcd,bool bVerbose,
real lambda,t_graph *graph,
bool bNS,bool bNBFonly,t_forcerec *fr, rvec mu_tot,
bool bGatherOnly)
{
static rvec box_size;
static real dvdl_lr = 0;
int cg0,cg1,i,j;
int start,homenr;
static real mu_and_q[DIM+1];
real qsum;
start = START(nsb);
homenr = HOMENR(nsb);
cg0 = CG0(nsb);
cg1 = CG1(nsb);
update_forcerec(log,fr,parm->box);
/* Calculate total (local) dipole moment in a temporary common array.
* This makes it possible to sum them over nodes faster.
*/
calc_mu_and_q(nsb,x,mdatoms->chargeT,mu_and_q,mu_and_q+DIM);
if (fr->ePBC != epbcNONE) {
/* Compute shift vectors every step, because of pressure coupling! */
if (parm->ir.epc != epcNO)
calc_shifts(parm->box,box_size,fr->shift_vec);
if (bNS) {
put_charge_groups_in_box(log,cg0,cg1,parm->box,box_size,
&(top->blocks[ebCGS]),x,fr->cg_cm);
inc_nrnb(nrnb,eNR_RESETX,homenr);
} else if (parm->ir.eI==eiSteep || parm->ir.eI==eiCG)
unshift_self(graph,parm->box,x);
}
else if (bNS)
calc_cgcm(log,cg0,cg1,&(top->blocks[ebCGS]),x,fr->cg_cm);
if (bNS) {
inc_nrnb(nrnb,eNR_CGCM,cg1-cg0);
if (PAR(cr))
move_cgcm(log,cr,fr->cg_cm,nsb->workload);
if (debug)
pr_rvecs(debug,0,"cgcm",fr->cg_cm,nsb->cgtotal);
}
/* Communicate coordinates and sum dipole and net charge if necessary */
if (PAR(cr)) {
move_x(log,cr->left,cr->right,x,nsb,nrnb);
gmx_sum(DIM+1,mu_and_q,cr);
}
for(i=0;i<DIM;i++)
mu_tot[i]=mu_and_q[i];
qsum=mu_and_q[DIM];
/* Reset energies */
reset_energies(&(parm->ir.opts),grps,fr,bNS,ener);
if (bNS) {
if (fr->ePBC != epbcNONE)
/* Calculate intramolecular shift vectors to make molecules whole */
mk_mshift(log,graph,parm->box,x);
/* Reset long range forces if necessary */
if (fr->bTwinRange) {
clear_rvecs(nsb->natoms,fr->f_twin);
clear_rvecs(SHIFTS,fr->fshift_twin);
}
/* Do the actual neighbour searching and if twin range electrostatics
* also do the calculation of long range forces and energies.
*/
dvdl_lr = 0;
ns(log,fr,x,f,parm->box,grps,&(parm->ir.opts),top,mdatoms,
cr,nrnb,nsb,step,lambda,&dvdl_lr);
}
/* Reset PME/Ewald forces if necessary */
if (EEL_LR(fr->eeltype))
clear_rvecs(homenr,fr->f_pme+start);
/* Copy long range forces into normal buffers */
if (fr->bTwinRange) {
for(i=0; i<nsb->natoms; i++)
copy_rvec(fr->f_twin[i],f[i]);
for(i=0; i<SHIFTS; i++)
copy_rvec(fr->fshift_twin[i],fr->fshift[i]);
}
else {
clear_rvecs(nsb->natoms,f);
clear_rvecs(SHIFTS,fr->fshift);
}
/* Compute the forces */
force(log,step,fr,&(parm->ir),&(top->idef),nsb,cr,mcr,nrnb,grps,mdatoms,
top->atoms.grps[egcENER].nr,&(parm->ir.opts),
x,f,ener,fcd,bVerbose,parm->box,lambda,graph,&(top->atoms.excl),
bNBFonly,pme_vir,mu_tot,qsum,bGatherOnly);
/* Take long range contribution to free energy into account */
ener[F_DVDL] += dvdl_lr;
#ifdef DEBUG
if (bNS)
print_nrnb(log,nrnb);
#endif
/* The short-range virial from surrounding boxes */
clear_mat(vir_part);
calc_vir(log,SHIFTS,fr->shift_vec,fr->fshift,vir_part);
inc_nrnb(nrnb,eNR_VIRIAL,SHIFTS);
if (debug)
pr_rvecs(debug,0,"vir_shifts",vir_part,DIM);
/* Compute forces due to electric field */
calc_f_el(start,homenr,mdatoms->chargeT,f,parm->ir.ex);
/* When using PME/Ewald we compute the long range virial (pme_vir) there.
* otherwise we do it based on long range forces from twin range
* cut-off based calculation (or not at all).
*/
/* Communicate the forces */
if (PAR(cr))
move_f(log,cr->left,cr->right,f,buf,nsb,nrnb);
}
void ns(FILE *fp,
t_forcerec *fr,
rvec x[],
rvec f[],
matrix box,
t_groups *grps,
t_grpopts *opts,
t_topology *top,
t_mdatoms *md,
t_commrec *cr,
t_nrnb *nrnb,
t_nsborder *nsb,
int step,
real lambda,
real *dvdlambda)
{
static bool bFirst=TRUE;
static int nDNL;
char *ptr;
int nsearch;
if (bFirst) {
#ifdef SPEC_CPU
ptr = NULL;
#else
ptr=getenv("DUMPNL");
#endif
if (ptr) {
nDNL=atoi(ptr);
fprintf(fp,"nDNL = %d\n",nDNL);
} else
nDNL=0;
/* Allocate memory for the neighbor lists */
init_neighbor_list(fp,fr,HOMENR(nsb));
bFirst=FALSE;
}
if (fr->bTwinRange)
fr->nlr=0;
/* Whether or not we do dynamic load balancing,
* workload contains the proper numbers of charge groups
* to be searched.
*/
if (cr->nodeid == 0)
fr->cg0=0;
else
fr->cg0=nsb->workload[cr->nodeid-1];
fr->hcg=nsb->workload[cr->nodeid];
nsearch = search_neighbours(fp,fr,x,box,top,grps,cr,nsb,nrnb,md,
lambda,dvdlambda);
if (debug)
fprintf(debug,"nsearch = %d\n",nsearch);
/* Check whether we have to do dynamic load balancing */
/*if ((nsb->nstDlb > 0) && (mod(step,nsb->nstDlb) == 0))
count_nb(cr,nsb,&(top->blocks[ebCGS]),nns,fr->nlr,
&(top->idef),opts->ngener);
*/
if (nDNL > 0)
dump_nblist(fp,fr,nDNL);
}
static void check_solvent(FILE *fp,t_topology *top,t_forcerec *fr,
t_mdatoms *md,t_nsborder *nsb)
{
/* This routine finds out whether a charge group can be used as
* solvent in the innerloops. The routine should be called once
* at the beginning of the MD program.
*/
t_block *cgs,*excl,*mols;
atom_id *cgid;
int i,j,m,j0,j1,nj,k,aj,ak,tjA,tjB,nl_m,nl_n,nl_o;
int warncount;
bool bOneCG;
bool *bAllExcl,bAE,bOrder;
bool *bHaveLJ,*bHaveCoul;
cgs = &(top->blocks[ebCGS]);
excl = &(top->atoms.excl);
mols = &(top->blocks[ebMOLS]);
if (fp)
fprintf(fp,"Going to determine what solvent types we have.\n");
snew(fr->solvent_type,cgs->nr+1);
snew(fr->mno_index,(cgs->nr+1)*3);
/* Generate charge group number for all atoms */
cgid = make_invblock(cgs,cgs->nra);
warncount=0;
/* Loop over molecules */
if (fp)
fprintf(fp,"There are %d molecules, %d charge groups and %d atoms\n",
mols->nr,cgs->nr,cgs->nra);
for(i=0; (i<mols->nr); i++) {
/* Set boolean that determines whether the molecules consists of one CG */
bOneCG = TRUE;
/* Set some counters */
j0 = mols->index[i];
j1 = mols->index[i+1];
nj = j1-j0;
for(j=j0+1; (j<j1); j++) {
bOneCG = bOneCG && (cgid[mols->a[j]] == cgid[mols->a[j-1]]);
}
if (fr->bSolvOpt && bOneCG && nj>1) {
/* Check whether everything is excluded */
snew(bAllExcl,nj);
bAE = TRUE;
/* Loop over all atoms in molecule */
for(j=j0; (j<j1) && bAE; j++) {
/* Set a flag for each atom in the molecule that determines whether
* it is excluded or not
*/
for(k=0; (k<nj); k++)
bAllExcl[k] = FALSE;
/* Now check all the exclusions of this atom */
for(k=excl->index[j]; (k<excl->index[j+1]); k++) {
ak = excl->a[k];
/* Consistency and range check */
if ((ak < j0) || (ak >= j1))
fatal_error(0,"Exclusion outside molecule? ak = %d, j0 = %d, j1 = 5d, mol is %d",ak,j0,j1,i);
bAllExcl[ak-j0] = TRUE;
}
/* Now sum up the booleans */
for(k=0; (k<nj); k++)
bAE = bAE && bAllExcl[k];
}
if (bAE) {
snew(bHaveCoul,nj);
snew(bHaveLJ,nj);
for(j=j0; (j<j1); j++) {
/* Check for coulomb */
aj = mols->a[j];
bHaveCoul[j-j0] = ( (fabs(top->atoms.atom[aj].q ) > GMX_REAL_MIN) ||
(fabs(top->atoms.atom[aj].qB) > GMX_REAL_MIN));
/* Check for LJ. */
tjA = top->atoms.atom[aj].type;
tjB = top->atoms.atom[aj].typeB;
bHaveLJ[j-j0] = FALSE;
for(k=0; (k<fr->ntype); k++) {
if (fr->bBHAM)
bHaveLJ[j-j0] = (bHaveLJ[j-j0] ||
(fabs(BHAMA(fr->nbfp,fr->ntype,tjA,k)) > GMX_REAL_MIN) ||
(fabs(BHAMB(fr->nbfp,fr->ntype,tjA,k)) > GMX_REAL_MIN) ||
(fabs(BHAMC(fr->nbfp,fr->ntype,tjA,k)) > GMX_REAL_MIN) ||
(fabs(BHAMA(fr->nbfp,fr->ntype,tjB,k)) > GMX_REAL_MIN) ||
(fabs(BHAMB(fr->nbfp,fr->ntype,tjB,k)) > GMX_REAL_MIN) ||
(fabs(BHAMC(fr->nbfp,fr->ntype,tjB,k)) > GMX_REAL_MIN));
else
bHaveLJ[j-j0] = (bHaveLJ[j-j0] ||
(fabs(C6(fr->nbfp,fr->ntype,tjA,k)) > GMX_REAL_MIN) ||
(fabs(C12(fr->nbfp,fr->ntype,tjA,k)) > GMX_REAL_MIN) ||
(fabs(C6(fr->nbfp,fr->ntype,tjB,k)) > GMX_REAL_MIN) ||
(fabs(C12(fr->nbfp,fr->ntype,tjB,k)) > GMX_REAL_MIN));
}
}
/* Now we have determined what particles have which interactions
* In the case of water-like molecules we only check for the number
* of particles and the LJ, not for the Coulomb. Let's just assume
* that the water loops are faster than the MNO loops anyway. DvdS
*/
/* No - there's another problem: To optimize the water
* innerloop assumes the charge of the first i atom is constant
* qO, and charge on atoms 2/3 is constant qH. /EL
*/
/* I won't write any altivec versions of the general solvent inner
* loops. Thus, when USE_PPC_ALTIVEC is defined it is faster
* to use the normal loops instead of the MNO solvent version. /EL
*/
aj=mols->a[j0];
if((nj==3) && bHaveCoul[0] && bHaveLJ[0] &&
!bHaveLJ[1] && !bHaveLJ[2] &&
fabs(top->atoms.atom[aj+1].q - top->atoms.atom[aj+2].q) < GMX_REAL_MIN)
fr->solvent_type[cgid[aj]] = esolWATER;
else {
#ifdef USE_PPC_ALTIVEC
fr->solvent_type[cgid[aj]] = esolNO;
#else
/* Time to compute M & N & O */
for(k=0; (k<nj) && (bHaveLJ[k] && bHaveCoul[k]); k++)
;
nl_n = k;
for(; (k<nj) && (!bHaveLJ[k] && bHaveCoul[k]); k++)
;
nl_o = k;
for(; (k<nj) && (bHaveLJ[k] && !bHaveCoul[k]); k++)
;
nl_m = k;
/* Now check whether we're at the end of the pack */
bOrder = FALSE;
for(; (k<nj); k++)
bOrder = bOrder || (bHaveLJ[k] || bHaveCoul[k]);
if (bOrder) {
/* If we have a solvent molecule with LJC everywhere, then
* we shouldn't issue a warning. Only if we suspect something
* could be better.
*/
if (nl_n != nj) {
warncount++;
if(warncount<11 && fp)
fprintf(fp,"The order in molecule %d could be optimized"
" for better performance\n",i);
if(warncount==10 && fp)
fprintf(fp,"(More than 10 molecules where the order can be optimized)\n");
}
nl_m = nl_n = nl_o = nj;
}
fr->mno_index[cgid[aj]*3] = nl_m;
fr->mno_index[cgid[aj]*3+1] = nl_n;
fr->mno_index[cgid[aj]*3+2] = nl_o;
fr->solvent_type[cgid[aj]] = esolMNO;
#endif /* MNO solvent if not using altivec */
}
/* Last check for perturbed atoms */
for(j=j0; (j<j1); j++)
if (md->bPerturbed[mols->a[j]])
fr->solvent_type[cgid[mols->a[j0]]] = esolNO;
sfree(bHaveLJ);
sfree(bHaveCoul);
}
else {
/* Turn off solvent optimization for all cg's in the molecule,
* here there is only one.
*/
fr->solvent_type[cgid[mols->a[j0]]] = esolNO;
}
sfree(bAllExcl);
}
else {
/* Turn off solvent optimization for all cg's in the molecule */
for(j=mols->index[i]; (j<mols->index[i+1]); j++) {
fr->solvent_type[cgid[mols->a[j]]] = esolNO;
}
}
}
if (debug) {
for(i=0; (i<cgs->nr); i++)
fprintf(debug,"MNO: cg = %5d, m = %2d, n = %2d, o = %2d\n",
i,fr->mno_index[3*i],fr->mno_index[3*i+1],fr->mno_index[3*i+2]);
}
/* Now compute the number of solvent molecules, could be merged with code above */
fr->nMNOMol = 0;
fr->nWatMol = 0;
for(m=0; m<3; m++)
fr->nMNOav[m] = 0;
for(i=0; i<mols->nr; i++) {
j = mols->a[mols->index[i]];
if (j>=START(nsb) && j<START(nsb)+HOMENR(nsb)) {
if (fr->solvent_type[cgid[j]] == esolMNO) {
fr->nMNOMol++;
for(m=0; m<3; m++)
fr->nMNOav[m] += fr->mno_index[3*cgid[j]+m];
}
else if (fr->solvent_type[cgid[j]] == esolWATER)
fr->nWatMol++;
}
}
if (fr->nMNOMol > 0)
for(m=0; (m<3); m++)
fr->nMNOav[m] /= fr->nMNOMol;
sfree(cgid);
if (fp) {
fprintf(fp,"There are %d optimized solvent molecules on node %d\n",
fr->nMNOMol,nsb->nodeid);
if (fr->nMNOMol > 0)
fprintf(fp," aver. nr. of atoms per molecule: vdwc %.1f coul %.1f vdw %.1f\n",
fr->nMNOav[1],fr->nMNOav[2]-fr->nMNOav[1],fr->nMNOav[0]-fr->nMNOav[2]);
fprintf(fp,"There are %d optimized water molecules on node %d\n",
fr->nWatMol,nsb->nodeid);
}
}