void do_pbc_first(FILE *log,t_parm *parm,rvec box_size,t_forcerec *fr,
t_graph *graph,rvec x[])
{
char *pbcenv;
if(log)
fprintf(log,"Removing pbc first time\n");
calc_shifts(parm->box,box_size,fr->shift_vec);
mk_mshift(log,graph,parm->box,x);
#ifdef SPEC_CPU
pbcenv = NULL;
#else
pbcenv = getenv ("NOPBC");
#endif
if (pbcenv == NULL)
shift_self(graph,parm->box,x);
else
{
if(log)
fprintf(log,"Not doing first shift_self\n");
}
if(log)
fprintf(log,"Done rmpbc\n");
}
void gmx_rmpbc(gmx_rmpbc_t gpbc, int natoms, matrix box, rvec x[])
{
int ePBC;
t_graph *gr;
ePBC = gmx_rmpbc_ePBC(gpbc, box);
gr = gmx_rmpbc_get_graph(gpbc, ePBC, natoms);
if (gr != NULL)
{
mk_mshift(stdout, gr, ePBC, box, x);
shift_self(gr, box, x);
}
}
void gmx_rmpbc_trxfr(gmx_rmpbc_t gpbc, t_trxframe *fr)
{
int ePBC;
t_graph *gr;
if (fr->bX && fr->bBox)
{
ePBC = gmx_rmpbc_ePBC(gpbc, fr->box);
gr = gmx_rmpbc_get_graph(gpbc, ePBC, fr->natoms);
if (gr != NULL)
{
mk_mshift(stdout, gr, ePBC, fr->box, fr->x);
shift_self(gr, fr->box, fr->x);
}
}
}
static void low_do_pbc_mtop(FILE *fplog,int ePBC,matrix box,
gmx_mtop_t *mtop,rvec x[],
bool bFirst)
{
t_graph *graph;
int mb,as,mol;
gmx_molblock_t *molb;
if (bFirst && fplog)
fprintf(fplog,"Removing pbc first time\n");
snew(graph,1);
as = 0;
for(mb=0; mb<mtop->nmolblock; mb++) {
molb = &mtop->molblock[mb];
if (molb->natoms_mol == 1 ||
(!bFirst && mtop->moltype[molb->type].cgs.nr == 1)) {
/* Just one atom or charge group in the molecule, no PBC required */
as += molb->nmol*molb->natoms_mol;
} else {
/* Pass NULL iso fplog to avoid graph prints for each molecule type */
mk_graph_ilist(NULL,mtop->moltype[molb->type].ilist,
0,molb->natoms_mol,FALSE,FALSE,graph);
for(mol=0; mol<molb->nmol; mol++) {
mk_mshift(fplog,graph,ePBC,box,x+as);
shift_self(graph,box,x+as);
/* The molecule is whole now.
* We don't need the second mk_mshift call as in do_pbc_first,
* since we no longer need this graph.
*/
as += molb->natoms_mol;
}
done_graph(graph);
}
}
sfree(graph);
}
void do_pbc_first(FILE *fplog,matrix box,t_forcerec *fr,
t_graph *graph,rvec x[])
{
if (fplog)
fprintf(fplog,"Removing pbc first time\n");
calc_shifts(box,fr->shift_vec);
if (graph) {
mk_mshift(fplog,graph,fr->ePBC,box,x);
if (gmx_debug_at)
p_graph(debug,"do_pbc_first 1",graph);
shift_self(graph,box,x);
/* By doing an extra mk_mshift the molecules that are broken
* because they were e.g. imported from another software
* will be made whole again. Such are the healing powers
* of GROMACS.
*/
mk_mshift(fplog,graph,fr->ePBC,box,x);
if (gmx_debug_at)
p_graph(debug,"do_pbc_first 2",graph);
}
if (fplog)
fprintf(fplog,"Done rmpbc\n");
}
int relax_shell_flexcon(FILE *fplog, t_commrec *cr, gmx_bool bVerbose,
gmx_int64_t mdstep, t_inputrec *inputrec,
gmx_bool bDoNS, int force_flags,
gmx_localtop_t *top,
gmx_constr_t constr,
gmx_enerdata_t *enerd, t_fcdata *fcd,
t_state *state, rvec f[],
tensor force_vir,
t_mdatoms *md,
t_nrnb *nrnb, gmx_wallcycle_t wcycle,
t_graph *graph,
gmx_groups_t *groups,
struct gmx_shellfc *shfc,
t_forcerec *fr,
gmx_bool bBornRadii,
double t, rvec mu_tot,
gmx_bool *bConverged,
gmx_vsite_t *vsite,
FILE *fp_field)
{
int nshell;
t_shell *shell;
t_idef *idef;
rvec *pos[2], *force[2], *acc_dir = NULL, *x_old = NULL;
real Epot[2], df[2];
rvec dx;
real sf_dir, invdt;
real ftol, xiH, xiS, dum = 0;
char sbuf[22];
gmx_bool bCont, bInit;
int nat, dd_ac0, dd_ac1 = 0, i;
int start = 0, homenr = md->homenr, end = start+homenr, cg0, cg1;
int nflexcon, g, number_steps, d, Min = 0, count = 0;
#define Try (1-Min) /* At start Try = 1 */
bCont = (mdstep == inputrec->init_step) && inputrec->bContinuation;
bInit = (mdstep == inputrec->init_step) || shfc->bRequireInit;
ftol = inputrec->em_tol;
number_steps = inputrec->niter;
nshell = shfc->nshell;
shell = shfc->shell;
nflexcon = shfc->nflexcon;
idef = &top->idef;
if (DOMAINDECOMP(cr))
{
nat = dd_natoms_vsite(cr->dd);
if (nflexcon > 0)
{
dd_get_constraint_range(cr->dd, &dd_ac0, &dd_ac1);
nat = max(nat, dd_ac1);
}
}
else
{
nat = state->natoms;
}
if (nat > shfc->x_nalloc)
{
/* Allocate local arrays */
shfc->x_nalloc = over_alloc_dd(nat);
for (i = 0; (i < 2); i++)
{
srenew(shfc->x[i], shfc->x_nalloc);
srenew(shfc->f[i], shfc->x_nalloc);
}
}
for (i = 0; (i < 2); i++)
{
pos[i] = shfc->x[i];
force[i] = shfc->f[i];
}
/* When we had particle decomposition, this code only worked with
* PD when all particles involved with each shell were in the same
* charge group. Not sure if this is still relevant. */
if (bDoNS && inputrec->ePBC != epbcNONE && !DOMAINDECOMP(cr))
{
/* This is the only time where the coordinates are used
* before do_force is called, which normally puts all
* charge groups in the box.
*/
cg0 = 0;
cg1 = top->cgs.nr;
put_charge_groups_in_box(fplog, cg0, cg1, fr->ePBC, state->box,
&(top->cgs), state->x, fr->cg_cm);
if (graph)
{
mk_mshift(fplog, graph, fr->ePBC, state->box, state->x);
}
}
/* After this all coordinate arrays will contain whole molecules */
if (graph)
{
shift_self(graph, state->box, state->x);
}
if (nflexcon)
{
if (nat > shfc->flex_nalloc)
{
shfc->flex_nalloc = over_alloc_dd(nat);
srenew(shfc->acc_dir, shfc->flex_nalloc);
srenew(shfc->x_old, shfc->flex_nalloc);
}
acc_dir = shfc->acc_dir;
x_old = shfc->x_old;
for (i = 0; i < homenr; i++)
{
for (d = 0; d < DIM; d++)
{
shfc->x_old[i][d] =
state->x[start+i][d] - state->v[start+i][d]*inputrec->delta_t;
}
}
}
/* Do a prediction of the shell positions */
if (shfc->bPredict && !bCont)
{
predict_shells(fplog, state->x, state->v, inputrec->delta_t, nshell, shell,
md->massT, NULL, bInit);
}
/* do_force expected the charge groups to be in the box */
if (graph)
{
unshift_self(graph, state->box, state->x);
}
/* Calculate the forces first time around */
if (gmx_debug_at)
{
pr_rvecs(debug, 0, "x b4 do_force", state->x + start, homenr);
}
do_force(fplog, cr, inputrec, mdstep, nrnb, wcycle, top, groups,
state->box, state->x, &state->hist,
force[Min], force_vir, md, enerd, fcd,
state->lambda, graph,
fr, vsite, mu_tot, t, fp_field, NULL, bBornRadii,
(bDoNS ? GMX_FORCE_NS : 0) | force_flags);
sf_dir = 0;
if (nflexcon)
{
init_adir(fplog, shfc,
constr, idef, inputrec, cr, dd_ac1, mdstep, md, start, end,
shfc->x_old-start, state->x, state->x, force[Min],
shfc->acc_dir-start,
fr->bMolPBC, state->box, state->lambda, &dum, nrnb);
for (i = start; i < end; i++)
{
sf_dir += md->massT[i]*norm2(shfc->acc_dir[i-start]);
}
}
Epot[Min] = enerd->term[F_EPOT];
df[Min] = rms_force(cr, shfc->f[Min], nshell, shell, nflexcon, &sf_dir, &Epot[Min]);
df[Try] = 0;
if (debug)
{
fprintf(debug, "df = %g %g\n", df[Min], df[Try]);
}
if (gmx_debug_at)
{
pr_rvecs(debug, 0, "force0", force[Min], md->nr);
}
if (nshell+nflexcon > 0)
{
/* Copy x to pos[Min] & pos[Try]: during minimization only the
* shell positions are updated, therefore the other particles must
* be set here.
*/
memcpy(pos[Min], state->x, nat*sizeof(state->x[0]));
memcpy(pos[Try], state->x, nat*sizeof(state->x[0]));
}
if (bVerbose && MASTER(cr))
{
print_epot(stdout, mdstep, 0, Epot[Min], df[Min], nflexcon, sf_dir);
}
if (debug)
{
fprintf(debug, "%17s: %14.10e\n",
interaction_function[F_EKIN].longname, enerd->term[F_EKIN]);
fprintf(debug, "%17s: %14.10e\n",
interaction_function[F_EPOT].longname, enerd->term[F_EPOT]);
fprintf(debug, "%17s: %14.10e\n",
interaction_function[F_ETOT].longname, enerd->term[F_ETOT]);
fprintf(debug, "SHELLSTEP %s\n", gmx_step_str(mdstep, sbuf));
}
/* First check whether we should do shells, or whether the force is
* low enough even without minimization.
*/
*bConverged = (df[Min] < ftol);
for (count = 1; (!(*bConverged) && (count < number_steps)); count++)
{
if (vsite)
{
construct_vsites(vsite, pos[Min], inputrec->delta_t, state->v,
idef->iparams, idef->il,
fr->ePBC, fr->bMolPBC, cr, state->box);
}
if (nflexcon)
{
init_adir(fplog, shfc,
constr, idef, inputrec, cr, dd_ac1, mdstep, md, start, end,
x_old-start, state->x, pos[Min], force[Min], acc_dir-start,
fr->bMolPBC, state->box, state->lambda, &dum, nrnb);
directional_sd(pos[Min], pos[Try], acc_dir-start, start, end,
fr->fc_stepsize);
}
/* New positions, Steepest descent */
shell_pos_sd(pos[Min], pos[Try], force[Min], nshell, shell, count);
/* do_force expected the charge groups to be in the box */
if (graph)
{
unshift_self(graph, state->box, pos[Try]);
}
if (gmx_debug_at)
{
pr_rvecs(debug, 0, "RELAX: pos[Min] ", pos[Min] + start, homenr);
pr_rvecs(debug, 0, "RELAX: pos[Try] ", pos[Try] + start, homenr);
}
/* Try the new positions */
do_force(fplog, cr, inputrec, 1, nrnb, wcycle,
top, groups, state->box, pos[Try], &state->hist,
force[Try], force_vir,
md, enerd, fcd, state->lambda, graph,
fr, vsite, mu_tot, t, fp_field, NULL, bBornRadii,
force_flags);
if (gmx_debug_at)
{
pr_rvecs(debug, 0, "RELAX: force[Min]", force[Min] + start, homenr);
pr_rvecs(debug, 0, "RELAX: force[Try]", force[Try] + start, homenr);
}
sf_dir = 0;
if (nflexcon)
{
init_adir(fplog, shfc,
constr, idef, inputrec, cr, dd_ac1, mdstep, md, start, end,
x_old-start, state->x, pos[Try], force[Try], acc_dir-start,
fr->bMolPBC, state->box, state->lambda, &dum, nrnb);
for (i = start; i < end; i++)
{
sf_dir += md->massT[i]*norm2(acc_dir[i-start]);
}
}
Epot[Try] = enerd->term[F_EPOT];
df[Try] = rms_force(cr, force[Try], nshell, shell, nflexcon, &sf_dir, &Epot[Try]);
if (debug)
{
fprintf(debug, "df = %g %g\n", df[Min], df[Try]);
}
if (debug)
{
if (gmx_debug_at)
{
pr_rvecs(debug, 0, "F na do_force", force[Try] + start, homenr);
}
if (gmx_debug_at)
{
fprintf(debug, "SHELL ITER %d\n", count);
dump_shells(debug, pos[Try], force[Try], ftol, nshell, shell);
}
}
if (bVerbose && MASTER(cr))
{
print_epot(stdout, mdstep, count, Epot[Try], df[Try], nflexcon, sf_dir);
}
*bConverged = (df[Try] < ftol);
if ((df[Try] < df[Min]))
{
if (debug)
{
fprintf(debug, "Swapping Min and Try\n");
}
if (nflexcon)
{
/* Correct the velocities for the flexible constraints */
invdt = 1/inputrec->delta_t;
for (i = start; i < end; i++)
{
for (d = 0; d < DIM; d++)
{
state->v[i][d] += (pos[Try][i][d] - pos[Min][i][d])*invdt;
}
}
}
Min = Try;
}
else
{
decrease_step_size(nshell, shell);
}
}
if (MASTER(cr) && !(*bConverged))
{
/* Note that the energies and virial are incorrect when not converged */
if (fplog)
{
fprintf(fplog,
"step %s: EM did not converge in %d iterations, RMS force %.3f\n",
gmx_step_str(mdstep, sbuf), number_steps, df[Min]);
}
fprintf(stderr,
"step %s: EM did not converge in %d iterations, RMS force %.3f\n",
gmx_step_str(mdstep, sbuf), number_steps, df[Min]);
}
/* Copy back the coordinates and the forces */
memcpy(state->x, pos[Min], nat*sizeof(state->x[0]));
memcpy(f, force[Min], nat*sizeof(f[0]));
return count;
}
void do_force_lowlevel(FILE *fplog, gmx_large_int_t step,
t_forcerec *fr, t_inputrec *ir,
t_idef *idef, t_commrec *cr,
t_nrnb *nrnb, gmx_wallcycle_t wcycle,
t_mdatoms *md,
t_grpopts *opts,
rvec x[], history_t *hist,
rvec f[],
rvec f_longrange[],
gmx_enerdata_t *enerd,
t_fcdata *fcd,
gmx_mtop_t *mtop,
gmx_localtop_t *top,
gmx_genborn_t *born,
t_atomtypes *atype,
gmx_bool bBornRadii,
matrix box,
t_lambda *fepvals,
real *lambda,
t_graph *graph,
t_blocka *excl,
rvec mu_tot[],
int flags,
float *cycles_pme)
{
int i, j, status;
int donb_flags;
gmx_bool bDoEpot, bSepDVDL, bSB;
int pme_flags;
matrix boxs;
rvec box_size;
real Vsr, Vlr, Vcorr = 0;
t_pbc pbc;
real dvdgb;
char buf[22];
double clam_i, vlam_i;
real dvdl_dum[efptNR], dvdl, dvdl_nb[efptNR], lam_i[efptNR];
real dvdlsum;
#ifdef GMX_MPI
double t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
#endif
#define PRINT_SEPDVDL(s, v, dvdlambda) if (bSepDVDL) {fprintf(fplog, sepdvdlformat, s, v, dvdlambda); }
GMX_MPE_LOG(ev_force_start);
set_pbc(&pbc, fr->ePBC, box);
/* reset free energy components */
for (i = 0; i < efptNR; i++)
{
dvdl_nb[i] = 0;
dvdl_dum[i] = 0;
}
/* Reset box */
for (i = 0; (i < DIM); i++)
{
box_size[i] = box[i][i];
}
bSepDVDL = (fr->bSepDVDL && do_per_step(step, ir->nstlog));
debug_gmx();
/* do QMMM first if requested */
if (fr->bQMMM)
{
enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr, md);
}
if (bSepDVDL)
{
fprintf(fplog, "Step %s: non-bonded V and dVdl for node %d:\n",
gmx_step_str(step, buf), cr->nodeid);
}
/* Call the short range functions all in one go. */
GMX_MPE_LOG(ev_do_fnbf_start);
#ifdef GMX_MPI
/*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
#define TAKETIME FALSE
if (TAKETIME)
{
MPI_Barrier(cr->mpi_comm_mygroup);
t0 = MPI_Wtime();
}
#endif
if (ir->nwall)
{
/* foreign lambda component for walls */
dvdl = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
enerd->grpp.ener[egLJSR], nrnb);
PRINT_SEPDVDL("Walls", 0.0, dvdl);
enerd->dvdl_lin[efptVDW] += dvdl;
}
/* If doing GB, reset dvda and calculate the Born radii */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsNONBONDED);
for (i = 0; i < born->nr; i++)
{
fr->dvda[i] = 0;
}
if (bBornRadii)
{
calc_gb_rad(cr, fr, ir, top, atype, x, &(fr->gblist), born, md, nrnb);
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
}
where();
/* We only do non-bonded calculation with group scheme here, the verlet
* calls are done from do_force_cutsVERLET(). */
if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
{
donb_flags = 0;
/* Add short-range interactions */
donb_flags |= GMX_NONBONDED_DO_SR;
if (flags & GMX_FORCE_FORCES)
{
donb_flags |= GMX_NONBONDED_DO_FORCE;
}
if (flags & GMX_FORCE_ENERGY)
{
donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
}
if (flags & GMX_FORCE_DO_LR)
{
donb_flags |= GMX_NONBONDED_DO_LR;
}
wallcycle_sub_start(wcycle, ewcsNONBONDED);
do_nonbonded(cr, fr, x, f, f_longrange, md, excl,
&enerd->grpp, box_size, nrnb,
lambda, dvdl_nb, -1, -1, donb_flags);
/* If we do foreign lambda and we have soft-core interactions
* we have to recalculate the (non-linear) energies contributions.
*/
if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
{
for (i = 0; i < enerd->n_lambda; i++)
{
for (j = 0; j < efptNR; j++)
{
lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
}
reset_foreign_enerdata(enerd);
do_nonbonded(cr, fr, x, f, f_longrange, md, excl,
&(enerd->foreign_grpp), box_size, nrnb,
lam_i, dvdl_dum, -1, -1,
(donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
sum_epot(&ir->opts, &(enerd->foreign_grpp), enerd->foreign_term);
enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
}
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
where();
}
/* If we are doing GB, calculate bonded forces and apply corrections
* to the solvation forces */
/* MRS: Eventually, many need to include free energy contribution here! */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsBONDED);
calc_gb_forces(cr, md, born, top, atype, x, f, fr, idef,
ir->gb_algorithm, ir->sa_algorithm, nrnb, bBornRadii, &pbc, graph, enerd);
wallcycle_sub_stop(wcycle, ewcsBONDED);
}
#ifdef GMX_MPI
if (TAKETIME)
{
t1 = MPI_Wtime();
fr->t_fnbf += t1-t0;
}
#endif
if (fepvals->sc_alpha != 0)
{
enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
}
else
{
enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
}
if (fepvals->sc_alpha != 0)
/* even though coulomb part is linear, we already added it, beacuse we
need to go through the vdw calculation anyway */
{
enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
}
else
{
enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
}
Vsr = 0;
if (bSepDVDL)
{
for (i = 0; i < enerd->grpp.nener; i++)
{
Vsr +=
(fr->bBHAM ?
enerd->grpp.ener[egBHAMSR][i] :
enerd->grpp.ener[egLJSR][i])
+ enerd->grpp.ener[egCOULSR][i] + enerd->grpp.ener[egGB][i];
}
dvdlsum = dvdl_nb[efptVDW] + dvdl_nb[efptCOUL];
PRINT_SEPDVDL("VdW and Coulomb SR particle-p.", Vsr, dvdlsum);
}
debug_gmx();
GMX_MPE_LOG(ev_do_fnbf_finish);
if (debug)
{
pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
}
/* Shift the coordinates. Must be done before bonded forces and PPPM,
* but is also necessary for SHAKE and update, therefore it can NOT
* go when no bonded forces have to be evaluated.
*/
/* Here sometimes we would not need to shift with NBFonly,
* but we do so anyhow for consistency of the returned coordinates.
*/
if (graph)
{
shift_self(graph, box, x);
if (TRICLINIC(box))
{
inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
}
else
{
inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
}
}
/* Check whether we need to do bondeds or correct for exclusions */
if (fr->bMolPBC &&
((flags & GMX_FORCE_BONDED)
|| EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype)))
{
/* Since all atoms are in the rectangular or triclinic unit-cell,
* only single box vector shifts (2 in x) are required.
*/
set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
}
debug_gmx();
if (flags & GMX_FORCE_BONDED)
{
GMX_MPE_LOG(ev_calc_bonds_start);
wallcycle_sub_start(wcycle, ewcsBONDED);
calc_bonds(fplog, cr->ms,
idef, x, hist, f, fr, &pbc, graph, enerd, nrnb, lambda, md, fcd,
DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL, atype, born,
flags,
fr->bSepDVDL && do_per_step(step, ir->nstlog), step);
/* Check if we have to determine energy differences
* at foreign lambda's.
*/
if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) &&
idef->ilsort != ilsortNO_FE)
{
if (idef->ilsort != ilsortFE_SORTED)
{
gmx_incons("The bonded interactions are not sorted for free energy");
}
for (i = 0; i < enerd->n_lambda; i++)
{
reset_foreign_enerdata(enerd);
for (j = 0; j < efptNR; j++)
{
lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
}
calc_bonds_lambda(fplog, idef, x, fr, &pbc, graph, &(enerd->foreign_grpp), enerd->foreign_term, nrnb, lam_i, md,
fcd, DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL);
sum_epot(&ir->opts, &(enerd->foreign_grpp), enerd->foreign_term);
enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
}
}
debug_gmx();
GMX_MPE_LOG(ev_calc_bonds_finish);
wallcycle_sub_stop(wcycle, ewcsBONDED);
}
where();
*cycles_pme = 0;
if (EEL_FULL(fr->eeltype))
{
bSB = (ir->nwall == 2);
if (bSB)
{
copy_mat(box, boxs);
svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
box_size[ZZ] *= ir->wall_ewald_zfac;
}
clear_mat(fr->vir_el_recip);
if (fr->bEwald)
{
Vcorr = 0;
dvdl = 0;
/* With the Verlet scheme exclusion forces are calculated
* in the non-bonded kernel.
*/
/* The TPI molecule does not have exclusions with the rest
* of the system and no intra-molecular PME grid contributions
* will be calculated in gmx_pme_calc_energy.
*/
if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
ir->ewald_geometry != eewg3D ||
ir->epsilon_surface != 0)
{
int nthreads, t;
wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
if (fr->n_tpi > 0)
{
gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
}
nthreads = gmx_omp_nthreads_get(emntBonded);
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (t = 0; t < nthreads; t++)
{
int s, e, i;
rvec *fnv;
tensor *vir;
real *Vcorrt, *dvdlt;
if (t == 0)
{
fnv = fr->f_novirsum;
vir = &fr->vir_el_recip;
Vcorrt = &Vcorr;
dvdlt = &dvdl;
}
else
{
fnv = fr->f_t[t].f;
vir = &fr->f_t[t].vir;
Vcorrt = &fr->f_t[t].Vcorr;
dvdlt = &fr->f_t[t].dvdl[efptCOUL];
for (i = 0; i < fr->natoms_force; i++)
{
clear_rvec(fnv[i]);
}
clear_mat(*vir);
}
*dvdlt = 0;
*Vcorrt =
ewald_LRcorrection(fplog,
fr->excl_load[t], fr->excl_load[t+1],
cr, t, fr,
md->chargeA,
md->nChargePerturbed ? md->chargeB : NULL,
ir->cutoff_scheme != ecutsVERLET,
excl, x, bSB ? boxs : box, mu_tot,
ir->ewald_geometry,
ir->epsilon_surface,
fnv, *vir,
lambda[efptCOUL], dvdlt);
}
if (nthreads > 1)
{
reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
fr->vir_el_recip,
&Vcorr, efptCOUL, &dvdl,
nthreads, fr->f_t);
}
wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
}
if (fr->n_tpi == 0)
{
Vcorr += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
&dvdl, fr->vir_el_recip);
}
PRINT_SEPDVDL("Ewald excl./charge/dip. corr.", Vcorr, dvdl);
enerd->dvdl_lin[efptCOUL] += dvdl;
}
status = 0;
Vlr = 0;
dvdl = 0;
switch (fr->eeltype)
{
case eelPME:
case eelPMESWITCH:
case eelPMEUSER:
case eelPMEUSERSWITCH:
case eelP3M_AD:
if (cr->duty & DUTY_PME)
{
assert(fr->n_tpi >= 0);
if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
{
pme_flags = GMX_PME_SPREAD_Q | GMX_PME_SOLVE;
if (flags & GMX_FORCE_FORCES)
{
pme_flags |= GMX_PME_CALC_F;
}
if (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY))
{
pme_flags |= GMX_PME_CALC_ENER_VIR;
}
if (fr->n_tpi > 0)
{
/* We don't calculate f, but we do want the potential */
pme_flags |= GMX_PME_CALC_POT;
}
wallcycle_start(wcycle, ewcPMEMESH);
status = gmx_pme_do(fr->pmedata,
md->start, md->homenr - fr->n_tpi,
x, fr->f_novirsum,
md->chargeA, md->chargeB,
bSB ? boxs : box, cr,
DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
nrnb, wcycle,
fr->vir_el_recip, fr->ewaldcoeff,
&Vlr, lambda[efptCOUL], &dvdl,
pme_flags);
*cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
/* We should try to do as little computation after
* this as possible, because parallel PME synchronizes
* the nodes, so we want all load imbalance of the rest
* of the force calculation to be before the PME call.
* DD load balancing is done on the whole time of
* the force call (without PME).
*/
}
if (fr->n_tpi > 0)
{
/* Determine the PME grid energy of the test molecule
* with the PME grid potential of the other charges.
*/
gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
x + md->homenr - fr->n_tpi,
md->chargeA + md->homenr - fr->n_tpi,
&Vlr);
}
PRINT_SEPDVDL("PME mesh", Vlr, dvdl);
}
break;
case eelEWALD:
Vlr = do_ewald(fplog, FALSE, ir, x, fr->f_novirsum,
md->chargeA, md->chargeB,
box_size, cr, md->homenr,
fr->vir_el_recip, fr->ewaldcoeff,
lambda[efptCOUL], &dvdl, fr->ewald_table);
PRINT_SEPDVDL("Ewald long-range", Vlr, dvdl);
break;
default:
gmx_fatal(FARGS, "No such electrostatics method implemented %s",
eel_names[fr->eeltype]);
}
if (status != 0)
{
gmx_fatal(FARGS, "Error %d in long range electrostatics routine %s",
status, EELTYPE(fr->eeltype));
}
/* Note that with separate PME nodes we get the real energies later */
enerd->dvdl_lin[efptCOUL] += dvdl;
enerd->term[F_COUL_RECIP] = Vlr + Vcorr;
if (debug)
{
fprintf(debug, "Vlr = %g, Vcorr = %g, Vlr_corr = %g\n",
Vlr, Vcorr, enerd->term[F_COUL_RECIP]);
pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
}
}
else
{
if (EEL_RF(fr->eeltype))
{
/* With the Verlet scheme exclusion forces are calculated
* in the non-bonded kernel.
*/
if (ir->cutoff_scheme != ecutsVERLET && fr->eeltype != eelRF_NEC)
{
dvdl = 0;
enerd->term[F_RF_EXCL] =
RF_excl_correction(fplog, fr, graph, md, excl, x, f,
fr->fshift, &pbc, lambda[efptCOUL], &dvdl);
}
enerd->dvdl_lin[efptCOUL] += dvdl;
PRINT_SEPDVDL("RF exclusion correction",
enerd->term[F_RF_EXCL], dvdl);
}
}
where();
debug_gmx();
if (debug)
{
print_nrnb(debug, nrnb);
}
debug_gmx();
#ifdef GMX_MPI
if (TAKETIME)
{
t2 = MPI_Wtime();
MPI_Barrier(cr->mpi_comm_mygroup);
t3 = MPI_Wtime();
fr->t_wait += t3-t2;
if (fr->timesteps == 11)
{
fprintf(stderr, "* PP load balancing info: node %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
cr->nodeid, gmx_step_str(fr->timesteps, buf),
100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
(fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
}
fr->timesteps++;
}
#endif
if (debug)
{
pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
}
GMX_MPE_LOG(ev_force_finish);
}
void do_force_lowlevel(t_forcerec *fr, t_inputrec *ir,
t_idef *idef, t_commrec *cr,
t_nrnb *nrnb, gmx_wallcycle_t wcycle,
t_mdatoms *md,
rvec x[], history_t *hist,
rvec f[],
rvec f_longrange[],
gmx_enerdata_t *enerd,
t_fcdata *fcd,
gmx_localtop_t *top,
gmx_genborn_t *born,
gmx_bool bBornRadii,
matrix box,
t_lambda *fepvals,
real *lambda,
t_graph *graph,
t_blocka *excl,
rvec mu_tot[],
int flags,
float *cycles_pme)
{
int i, j;
int donb_flags;
gmx_bool bSB;
int pme_flags;
matrix boxs;
rvec box_size;
t_pbc pbc;
real dvdl_dum[efptNR], dvdl_nb[efptNR];
#ifdef GMX_MPI
double t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
#endif
set_pbc(&pbc, fr->ePBC, box);
/* reset free energy components */
for (i = 0; i < efptNR; i++)
{
dvdl_nb[i] = 0;
dvdl_dum[i] = 0;
}
/* Reset box */
for (i = 0; (i < DIM); i++)
{
box_size[i] = box[i][i];
}
debug_gmx();
/* do QMMM first if requested */
if (fr->bQMMM)
{
enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr);
}
/* Call the short range functions all in one go. */
#ifdef GMX_MPI
/*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
#define TAKETIME FALSE
if (TAKETIME)
{
MPI_Barrier(cr->mpi_comm_mygroup);
t0 = MPI_Wtime();
}
#endif
if (ir->nwall)
{
/* foreign lambda component for walls */
real dvdl_walls = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
enerd->grpp.ener[egLJSR], nrnb);
enerd->dvdl_lin[efptVDW] += dvdl_walls;
}
/* If doing GB, reset dvda and calculate the Born radii */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsNONBONDED);
for (i = 0; i < born->nr; i++)
{
fr->dvda[i] = 0;
}
if (bBornRadii)
{
calc_gb_rad(cr, fr, ir, top, x, &(fr->gblist), born, md, nrnb);
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
}
where();
/* We only do non-bonded calculation with group scheme here, the verlet
* calls are done from do_force_cutsVERLET(). */
if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
{
donb_flags = 0;
/* Add short-range interactions */
donb_flags |= GMX_NONBONDED_DO_SR;
/* Currently all group scheme kernels always calculate (shift-)forces */
if (flags & GMX_FORCE_FORCES)
{
donb_flags |= GMX_NONBONDED_DO_FORCE;
}
if (flags & GMX_FORCE_VIRIAL)
{
donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
}
if (flags & GMX_FORCE_ENERGY)
{
donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
}
if (flags & GMX_FORCE_DO_LR)
{
donb_flags |= GMX_NONBONDED_DO_LR;
}
wallcycle_sub_start(wcycle, ewcsNONBONDED);
do_nonbonded(fr, x, f, f_longrange, md, excl,
&enerd->grpp, nrnb,
lambda, dvdl_nb, -1, -1, donb_flags);
/* If we do foreign lambda and we have soft-core interactions
* we have to recalculate the (non-linear) energies contributions.
*/
if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
{
for (i = 0; i < enerd->n_lambda; i++)
{
real lam_i[efptNR];
for (j = 0; j < efptNR; j++)
{
lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
}
reset_foreign_enerdata(enerd);
do_nonbonded(fr, x, f, f_longrange, md, excl,
&(enerd->foreign_grpp), nrnb,
lam_i, dvdl_dum, -1, -1,
(donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
}
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
where();
}
/* If we are doing GB, calculate bonded forces and apply corrections
* to the solvation forces */
/* MRS: Eventually, many need to include free energy contribution here! */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsLISTED);
calc_gb_forces(cr, md, born, top, x, f, fr, idef,
ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
wallcycle_sub_stop(wcycle, ewcsLISTED);
}
#ifdef GMX_MPI
if (TAKETIME)
{
t1 = MPI_Wtime();
fr->t_fnbf += t1-t0;
}
#endif
if (fepvals->sc_alpha != 0)
{
enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
}
else
{
enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
}
if (fepvals->sc_alpha != 0)
/* even though coulomb part is linear, we already added it, beacuse we
need to go through the vdw calculation anyway */
{
enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
}
else
{
enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
}
debug_gmx();
if (debug)
{
pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
}
/* Shift the coordinates. Must be done before listed forces and PPPM,
* but is also necessary for SHAKE and update, therefore it can NOT
* go when no listed forces have to be evaluated.
*
* The shifting and PBC code is deliberately not timed, since with
* the Verlet scheme it only takes non-zero time with triclinic
* boxes, and even then the time is around a factor of 100 less
* than the next smallest counter.
*/
/* Here sometimes we would not need to shift with NBFonly,
* but we do so anyhow for consistency of the returned coordinates.
*/
if (graph)
{
shift_self(graph, box, x);
if (TRICLINIC(box))
{
inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
}
else
{
inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
}
}
/* Check whether we need to do listed interactions or correct for exclusions */
if (fr->bMolPBC &&
((flags & GMX_FORCE_LISTED)
|| EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype)))
{
/* TODO There are no electrostatics methods that require this
transformation, when using the Verlet scheme, so update the
above conditional. */
/* Since all atoms are in the rectangular or triclinic unit-cell,
* only single box vector shifts (2 in x) are required.
*/
set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
}
debug_gmx();
do_force_listed(wcycle, box, ir->fepvals, cr->ms,
idef, (const rvec *) x, hist, f, fr,
&pbc, graph, enerd, nrnb, lambda, md, fcd,
DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL,
flags);
where();
*cycles_pme = 0;
clear_mat(fr->vir_el_recip);
clear_mat(fr->vir_lj_recip);
/* Do long-range electrostatics and/or LJ-PME, including related short-range
* corrections.
*/
if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
{
int status = 0;
real Vlr_q = 0, Vlr_lj = 0, Vcorr_q = 0, Vcorr_lj = 0;
real dvdl_long_range_q = 0, dvdl_long_range_lj = 0;
bSB = (ir->nwall == 2);
if (bSB)
{
copy_mat(box, boxs);
svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
box_size[ZZ] *= ir->wall_ewald_zfac;
}
if (EEL_PME_EWALD(fr->eeltype) || EVDW_PME(fr->vdwtype))
{
real dvdl_long_range_correction_q = 0;
real dvdl_long_range_correction_lj = 0;
/* With the Verlet scheme exclusion forces are calculated
* in the non-bonded kernel.
*/
/* The TPI molecule does not have exclusions with the rest
* of the system and no intra-molecular PME grid
* contributions will be calculated in
* gmx_pme_calc_energy.
*/
if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
ir->ewald_geometry != eewg3D ||
ir->epsilon_surface != 0)
{
int nthreads, t;
wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
if (fr->n_tpi > 0)
{
gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
}
nthreads = fr->nthread_ewc;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (t = 0; t < nthreads; t++)
{
try
{
tensor *vir_q, *vir_lj;
real *Vcorrt_q, *Vcorrt_lj, *dvdlt_q, *dvdlt_lj;
if (t == 0)
{
vir_q = &fr->vir_el_recip;
vir_lj = &fr->vir_lj_recip;
Vcorrt_q = &Vcorr_q;
Vcorrt_lj = &Vcorr_lj;
dvdlt_q = &dvdl_long_range_correction_q;
dvdlt_lj = &dvdl_long_range_correction_lj;
}
else
{
vir_q = &fr->ewc_t[t].vir_q;
vir_lj = &fr->ewc_t[t].vir_lj;
Vcorrt_q = &fr->ewc_t[t].Vcorr_q;
Vcorrt_lj = &fr->ewc_t[t].Vcorr_lj;
dvdlt_q = &fr->ewc_t[t].dvdl[efptCOUL];
dvdlt_lj = &fr->ewc_t[t].dvdl[efptVDW];
clear_mat(*vir_q);
clear_mat(*vir_lj);
}
*dvdlt_q = 0;
*dvdlt_lj = 0;
/* Threading is only supported with the Verlet cut-off
* scheme and then only single particle forces (no
* exclusion forces) are calculated, so we can store
* the forces in the normal, single fr->f_novirsum array.
*/
ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
cr, t, fr,
md->chargeA, md->chargeB,
md->sqrt_c6A, md->sqrt_c6B,
md->sigmaA, md->sigmaB,
md->sigma3A, md->sigma3B,
md->nChargePerturbed || md->nTypePerturbed,
ir->cutoff_scheme != ecutsVERLET,
excl, x, bSB ? boxs : box, mu_tot,
ir->ewald_geometry,
ir->epsilon_surface,
fr->f_novirsum, *vir_q, *vir_lj,
Vcorrt_q, Vcorrt_lj,
lambda[efptCOUL], lambda[efptVDW],
dvdlt_q, dvdlt_lj);
}
GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
}
if (nthreads > 1)
{
reduce_thread_energies(fr->vir_el_recip, fr->vir_lj_recip,
&Vcorr_q, &Vcorr_lj,
&dvdl_long_range_correction_q,
&dvdl_long_range_correction_lj,
nthreads, fr->ewc_t);
}
wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
}
void do_force_lowlevel(FILE *fplog, gmx_large_int_t step,
t_forcerec *fr, t_inputrec *ir,
t_idef *idef, t_commrec *cr,
t_nrnb *nrnb, gmx_wallcycle_t wcycle,
t_mdatoms *md,
t_grpopts *opts,
rvec x[], history_t *hist,
rvec f[],
gmx_enerdata_t *enerd,
t_fcdata *fcd,
gmx_mtop_t *mtop,
gmx_localtop_t *top,
gmx_genborn_t *born,
t_atomtypes *atype,
gmx_bool bBornRadii,
matrix box,
real lambda,
t_graph *graph,
t_blocka *excl,
rvec mu_tot[],
int flags,
float *cycles_pme)
{
int i,status;
int donb_flags;
gmx_bool bDoEpot,bSepDVDL,bSB;
int pme_flags;
matrix boxs;
rvec box_size;
real dvdlambda,Vsr,Vlr,Vcorr=0,vdip,vcharge;
t_pbc pbc;
real dvdgb;
char buf[22];
gmx_enerdata_t ed_lam;
double lam_i;
real dvdl_dum;
#ifdef GMX_MPI
double t0=0.0,t1,t2,t3; /* time measurement for coarse load balancing */
#endif
#define PRINT_SEPDVDL(s,v,dvdl) if (bSepDVDL) fprintf(fplog,sepdvdlformat,s,v,dvdl);
GMX_MPE_LOG(ev_force_start);
set_pbc(&pbc,fr->ePBC,box);
/* Reset box */
for(i=0; (i<DIM); i++)
{
box_size[i]=box[i][i];
}
bSepDVDL=(fr->bSepDVDL && do_per_step(step,ir->nstlog));
debug_gmx();
/* do QMMM first if requested */
if(fr->bQMMM)
{
enerd->term[F_EQM] = calculate_QMMM(cr,x,f,fr,md);
}
if (bSepDVDL)
{
fprintf(fplog,"Step %s: non-bonded V and dVdl for node %d:\n",
gmx_step_str(step,buf),cr->nodeid);
}
/* Call the short range functions all in one go. */
GMX_MPE_LOG(ev_do_fnbf_start);
dvdlambda = 0;
#ifdef GMX_MPI
/*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
#define TAKETIME FALSE
if (TAKETIME)
{
MPI_Barrier(cr->mpi_comm_mygroup);
t0=MPI_Wtime();
}
#endif
if (ir->nwall)
{
dvdlambda = do_walls(ir,fr,box,md,x,f,lambda,
enerd->grpp.ener[egLJSR],nrnb);
PRINT_SEPDVDL("Walls",0.0,dvdlambda);
enerd->dvdl_lin += dvdlambda;
}
/* If doing GB, reset dvda and calculate the Born radii */
if (ir->implicit_solvent)
{
/* wallcycle_start(wcycle,ewcGB); */
for(i=0; i<born->nr; i++)
{
fr->dvda[i]=0;
}
if(bBornRadii)
{
calc_gb_rad(cr,fr,ir,top,atype,x,&(fr->gblist),born,md,nrnb);
}
/* wallcycle_stop(wcycle, ewcGB); */
}
where();
donb_flags = 0;
if (flags & GMX_FORCE_FORCES)
{
donb_flags |= GMX_DONB_FORCES;
}
do_nonbonded(cr,fr,x,f,md,excl,
fr->bBHAM ?
enerd->grpp.ener[egBHAMSR] :
enerd->grpp.ener[egLJSR],
enerd->grpp.ener[egCOULSR],
enerd->grpp.ener[egGB],box_size,nrnb,
lambda,&dvdlambda,-1,-1,donb_flags);
/* If we do foreign lambda and we have soft-core interactions
* we have to recalculate the (non-linear) energies contributions.
*/
if (ir->n_flambda > 0 && (flags & GMX_FORCE_DHDL) && ir->sc_alpha != 0)
{
init_enerdata(mtop->groups.grps[egcENER].nr,ir->n_flambda,&ed_lam);
for(i=0; i<enerd->n_lambda; i++)
{
lam_i = (i==0 ? lambda : ir->flambda[i-1]);
dvdl_dum = 0;
reset_enerdata(&ir->opts,fr,TRUE,&ed_lam,FALSE);
do_nonbonded(cr,fr,x,f,md,excl,
fr->bBHAM ?
ed_lam.grpp.ener[egBHAMSR] :
ed_lam.grpp.ener[egLJSR],
ed_lam.grpp.ener[egCOULSR],
enerd->grpp.ener[egGB], box_size,nrnb,
lam_i,&dvdl_dum,-1,-1,
GMX_DONB_FOREIGNLAMBDA);
sum_epot(&ir->opts,&ed_lam);
enerd->enerpart_lambda[i] += ed_lam.term[F_EPOT];
}
destroy_enerdata(&ed_lam);
}
where();
/* If we are doing GB, calculate bonded forces and apply corrections
* to the solvation forces */
if (ir->implicit_solvent) {
calc_gb_forces(cr,md,born,top,atype,x,f,fr,idef,
ir->gb_algorithm,ir->sa_algorithm,nrnb,bBornRadii,&pbc,graph,enerd);
}
#ifdef GMX_MPI
if (TAKETIME)
{
t1=MPI_Wtime();
fr->t_fnbf += t1-t0;
}
#endif
if (ir->sc_alpha != 0)
{
enerd->dvdl_nonlin += dvdlambda;
}
else
{
enerd->dvdl_lin += dvdlambda;
}
Vsr = 0;
if (bSepDVDL)
{
for(i=0; i<enerd->grpp.nener; i++)
{
Vsr +=
(fr->bBHAM ?
enerd->grpp.ener[egBHAMSR][i] :
enerd->grpp.ener[egLJSR][i])
+ enerd->grpp.ener[egCOULSR][i] + enerd->grpp.ener[egGB][i];
}
}
PRINT_SEPDVDL("VdW and Coulomb SR particle-p.",Vsr,dvdlambda);
debug_gmx();
GMX_MPE_LOG(ev_do_fnbf_finish);
if (debug)
{
pr_rvecs(debug,0,"fshift after SR",fr->fshift,SHIFTS);
}
/* Shift the coordinates. Must be done before bonded forces and PPPM,
* but is also necessary for SHAKE and update, therefore it can NOT
* go when no bonded forces have to be evaluated.
*/
/* Here sometimes we would not need to shift with NBFonly,
* but we do so anyhow for consistency of the returned coordinates.
*/
if (graph)
{
shift_self(graph,box,x);
if (TRICLINIC(box))
{
inc_nrnb(nrnb,eNR_SHIFTX,2*graph->nnodes);
}
else
{
inc_nrnb(nrnb,eNR_SHIFTX,graph->nnodes);
}
}
/* Check whether we need to do bondeds or correct for exclusions */
if (fr->bMolPBC &&
((flags & GMX_FORCE_BONDED)
|| EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype)))
{
/* Since all atoms are in the rectangular or triclinic unit-cell,
* only single box vector shifts (2 in x) are required.
*/
set_pbc_dd(&pbc,fr->ePBC,cr->dd,TRUE,box);
}
debug_gmx();
if (flags & GMX_FORCE_BONDED)
{
GMX_MPE_LOG(ev_calc_bonds_start);
calc_bonds(fplog,cr->ms,
idef,x,hist,f,fr,&pbc,graph,enerd,nrnb,lambda,md,fcd,
DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL, atype, born,
fr->bSepDVDL && do_per_step(step,ir->nstlog),step);
/* Check if we have to determine energy differences
* at foreign lambda's.
*/
if (ir->n_flambda > 0 && (flags & GMX_FORCE_DHDL) &&
idef->ilsort != ilsortNO_FE)
{
if (idef->ilsort != ilsortFE_SORTED)
{
gmx_incons("The bonded interactions are not sorted for free energy");
}
init_enerdata(mtop->groups.grps[egcENER].nr,ir->n_flambda,&ed_lam);
for(i=0; i<enerd->n_lambda; i++)
{
lam_i = (i==0 ? lambda : ir->flambda[i-1]);
dvdl_dum = 0;
reset_enerdata(&ir->opts,fr,TRUE,&ed_lam,FALSE);
calc_bonds_lambda(fplog,
idef,x,fr,&pbc,graph,&ed_lam,nrnb,lam_i,md,
fcd,
DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL);
sum_epot(&ir->opts,&ed_lam);
enerd->enerpart_lambda[i] += ed_lam.term[F_EPOT];
}
destroy_enerdata(&ed_lam);
}
debug_gmx();
GMX_MPE_LOG(ev_calc_bonds_finish);
}
where();
*cycles_pme = 0;
if (EEL_FULL(fr->eeltype))
{
bSB = (ir->nwall == 2);
if (bSB)
{
copy_mat(box,boxs);
svmul(ir->wall_ewald_zfac,boxs[ZZ],boxs[ZZ]);
box_size[ZZ] *= ir->wall_ewald_zfac;
}
clear_mat(fr->vir_el_recip);
if (fr->bEwald)
{
if (fr->n_tpi == 0)
{
dvdlambda = 0;
Vcorr = ewald_LRcorrection(fplog,md->start,md->start+md->homenr,
cr,fr,
md->chargeA,
md->nChargePerturbed ? md->chargeB : NULL,
excl,x,bSB ? boxs : box,mu_tot,
ir->ewald_geometry,
ir->epsilon_surface,
lambda,&dvdlambda,&vdip,&vcharge);
PRINT_SEPDVDL("Ewald excl./charge/dip. corr.",Vcorr,dvdlambda);
enerd->dvdl_lin += dvdlambda;
}
else
{
if (ir->ewald_geometry != eewg3D || ir->epsilon_surface != 0)
{
gmx_fatal(FARGS,"TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
}
/* The TPI molecule does not have exclusions with the rest
* of the system and no intra-molecular PME grid contributions
* will be calculated in gmx_pme_calc_energy.
*/
Vcorr = 0;
}
}
else
{
Vcorr = shift_LRcorrection(fplog,md->start,md->homenr,cr,fr,
md->chargeA,excl,x,TRUE,box,
fr->vir_el_recip);
}
dvdlambda = 0;
status = 0;
switch (fr->eeltype)
{
case eelPPPM:
status = gmx_pppm_do(fplog,fr->pmedata,FALSE,x,fr->f_novirsum,
md->chargeA,
box_size,fr->phi,cr,md->start,md->homenr,
nrnb,ir->pme_order,&Vlr);
break;
case eelPME:
case eelPMESWITCH:
case eelPMEUSER:
case eelPMEUSERSWITCH:
if (cr->duty & DUTY_PME)
{
if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
{
pme_flags = GMX_PME_SPREAD_Q | GMX_PME_SOLVE;
if (flags & GMX_FORCE_FORCES)
{
pme_flags |= GMX_PME_CALC_F;
}
if (flags & GMX_FORCE_VIRIAL)
{
pme_flags |= GMX_PME_CALC_ENER_VIR;
}
if (fr->n_tpi > 0)
{
/* We don't calculate f, but we do want the potential */
pme_flags |= GMX_PME_CALC_POT;
}
wallcycle_start(wcycle,ewcPMEMESH);
status = gmx_pme_do(fr->pmedata,
md->start,md->homenr - fr->n_tpi,
x,fr->f_novirsum,
md->chargeA,md->chargeB,
bSB ? boxs : box,cr,
DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
nrnb,wcycle,
fr->vir_el_recip,fr->ewaldcoeff,
&Vlr,lambda,&dvdlambda,
pme_flags);
*cycles_pme = wallcycle_stop(wcycle,ewcPMEMESH);
/* We should try to do as little computation after
* this as possible, because parallel PME synchronizes
* the nodes, so we want all load imbalance of the rest
* of the force calculation to be before the PME call.
* DD load balancing is done on the whole time of
* the force call (without PME).
*/
}
if (fr->n_tpi > 0)
{
/* Determine the PME grid energy of the test molecule
* with the PME grid potential of the other charges.
*/
gmx_pme_calc_energy(fr->pmedata,fr->n_tpi,
x + md->homenr - fr->n_tpi,
md->chargeA + md->homenr - fr->n_tpi,
&Vlr);
}
PRINT_SEPDVDL("PME mesh",Vlr,dvdlambda);
}
else
{
/* Energies and virial are obtained later from the PME nodes */
/* but values have to be zeroed out here */
Vlr=0.0;
}
break;
case eelEWALD:
Vlr = do_ewald(fplog,FALSE,ir,x,fr->f_novirsum,
md->chargeA,md->chargeB,
box_size,cr,md->homenr,
fr->vir_el_recip,fr->ewaldcoeff,
lambda,&dvdlambda,fr->ewald_table);
PRINT_SEPDVDL("Ewald long-range",Vlr,dvdlambda);
break;
default:
Vlr = 0;
gmx_fatal(FARGS,"No such electrostatics method implemented %s",
eel_names[fr->eeltype]);
}
if (status != 0)
{
gmx_fatal(FARGS,"Error %d in long range electrostatics routine %s",
status,EELTYPE(fr->eeltype));
}
enerd->dvdl_lin += dvdlambda;
enerd->term[F_COUL_RECIP] = Vlr + Vcorr;
if (debug)
{
fprintf(debug,"Vlr = %g, Vcorr = %g, Vlr_corr = %g\n",
Vlr,Vcorr,enerd->term[F_COUL_RECIP]);
pr_rvecs(debug,0,"vir_el_recip after corr",fr->vir_el_recip,DIM);
pr_rvecs(debug,0,"fshift after LR Corrections",fr->fshift,SHIFTS);
}
}
else
{
if (EEL_RF(fr->eeltype))
{
dvdlambda = 0;
if (fr->eeltype != eelRF_NEC)
{
enerd->term[F_RF_EXCL] =
RF_excl_correction(fplog,fr,graph,md,excl,x,f,
fr->fshift,&pbc,lambda,&dvdlambda);
}
enerd->dvdl_lin += dvdlambda;
PRINT_SEPDVDL("RF exclusion correction",
enerd->term[F_RF_EXCL],dvdlambda);
}
}
where();
debug_gmx();
if (debug)
{
print_nrnb(debug,nrnb);
}
debug_gmx();
#ifdef GMX_MPI
if (TAKETIME)
{
t2=MPI_Wtime();
MPI_Barrier(cr->mpi_comm_mygroup);
t3=MPI_Wtime();
fr->t_wait += t3-t2;
if (fr->timesteps == 11)
{
fprintf(stderr,"* PP load balancing info: node %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
cr->nodeid, gmx_step_str(fr->timesteps,buf),
100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
(fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
}
fr->timesteps++;
}
#endif
if (debug)
{
pr_rvecs(debug,0,"fshift after bondeds",fr->fshift,SHIFTS);
}
GMX_MPE_LOG(ev_force_finish);
}
void do_force_lowlevel(t_forcerec *fr, t_inputrec *ir,
t_idef *idef, t_commrec *cr,
t_nrnb *nrnb, gmx_wallcycle_t wcycle,
t_mdatoms *md,
rvec x[], history_t *hist,
rvec f[],
rvec f_longrange[],
gmx_enerdata_t *enerd,
t_fcdata *fcd,
gmx_localtop_t *top,
gmx_genborn_t *born,
gmx_bool bBornRadii,
matrix box,
t_lambda *fepvals,
real *lambda,
t_graph *graph,
t_blocka *excl,
rvec mu_tot[],
int flags,
float *cycles_pme)
{
int i, j;
int donb_flags;
gmx_bool bSB;
int pme_flags;
matrix boxs;
rvec box_size;
t_pbc pbc;
real dvdl_dum[efptNR], dvdl_nb[efptNR];
#ifdef GMX_MPI
double t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
#endif
set_pbc(&pbc, fr->ePBC, box);
/* reset free energy components */
for (i = 0; i < efptNR; i++)
{
dvdl_nb[i] = 0;
dvdl_dum[i] = 0;
}
/* Reset box */
for (i = 0; (i < DIM); i++)
{
box_size[i] = box[i][i];
}
debug_gmx();
/* do QMMM first if requested */
if (fr->bQMMM)
{
enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr);
}
/* Call the short range functions all in one go. */
#ifdef GMX_MPI
/*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
#define TAKETIME FALSE
if (TAKETIME)
{
MPI_Barrier(cr->mpi_comm_mygroup);
t0 = MPI_Wtime();
}
#endif
if (ir->nwall)
{
/* foreign lambda component for walls */
real dvdl_walls = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
enerd->grpp.ener[egLJSR], nrnb);
enerd->dvdl_lin[efptVDW] += dvdl_walls;
}
/* If doing GB, reset dvda and calculate the Born radii */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsNONBONDED);
for (i = 0; i < born->nr; i++)
{
fr->dvda[i] = 0;
}
if (bBornRadii)
{
calc_gb_rad(cr, fr, ir, top, x, &(fr->gblist), born, md, nrnb);
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
}
where();
/* We only do non-bonded calculation with group scheme here, the verlet
* calls are done from do_force_cutsVERLET(). */
if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
{
donb_flags = 0;
/* Add short-range interactions */
donb_flags |= GMX_NONBONDED_DO_SR;
/* Currently all group scheme kernels always calculate (shift-)forces */
if (flags & GMX_FORCE_FORCES)
{
donb_flags |= GMX_NONBONDED_DO_FORCE;
}
if (flags & GMX_FORCE_VIRIAL)
{
donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
}
if (flags & GMX_FORCE_ENERGY)
{
donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
}
if (flags & GMX_FORCE_DO_LR)
{
donb_flags |= GMX_NONBONDED_DO_LR;
}
wallcycle_sub_start(wcycle, ewcsNONBONDED);
do_nonbonded(fr, x, f, f_longrange, md, excl,
&enerd->grpp, nrnb,
lambda, dvdl_nb, -1, -1, donb_flags);
/* If we do foreign lambda and we have soft-core interactions
* we have to recalculate the (non-linear) energies contributions.
*/
if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
{
for (i = 0; i < enerd->n_lambda; i++)
{
real lam_i[efptNR];
for (j = 0; j < efptNR; j++)
{
lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
}
reset_foreign_enerdata(enerd);
do_nonbonded(fr, x, f, f_longrange, md, excl,
&(enerd->foreign_grpp), nrnb,
lam_i, dvdl_dum, -1, -1,
(donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
}
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
where();
}
/* If we are doing GB, calculate bonded forces and apply corrections
* to the solvation forces */
/* MRS: Eventually, many need to include free energy contribution here! */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsLISTED);
calc_gb_forces(cr, md, born, top, x, f, fr, idef,
ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
wallcycle_sub_stop(wcycle, ewcsLISTED);
}
#ifdef GMX_MPI
if (TAKETIME)
{
t1 = MPI_Wtime();
fr->t_fnbf += t1-t0;
}
#endif
if (fepvals->sc_alpha != 0)
{
enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
}
else
{
enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
}
if (fepvals->sc_alpha != 0)
/* even though coulomb part is linear, we already added it, beacuse we
need to go through the vdw calculation anyway */
{
enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
}
else
{
enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
}
debug_gmx();
if (debug)
{
pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
}
/* Shift the coordinates. Must be done before listed forces and PPPM,
* but is also necessary for SHAKE and update, therefore it can NOT
* go when no listed forces have to be evaluated.
*
* The shifting and PBC code is deliberately not timed, since with
* the Verlet scheme it only takes non-zero time with triclinic
* boxes, and even then the time is around a factor of 100 less
* than the next smallest counter.
*/
/* Here sometimes we would not need to shift with NBFonly,
* but we do so anyhow for consistency of the returned coordinates.
*/
if (graph)
{
shift_self(graph, box, x);
if (TRICLINIC(box))
{
inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
}
else
{
inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
}
}
/* Check whether we need to do listed interactions or correct for exclusions */
if (fr->bMolPBC &&
((flags & GMX_FORCE_LISTED)
|| EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype)))
{
/* TODO There are no electrostatics methods that require this
transformation, when using the Verlet scheme, so update the
above conditional. */
/* Since all atoms are in the rectangular or triclinic unit-cell,
* only single box vector shifts (2 in x) are required.
*/
set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
}
debug_gmx();
do_force_listed(wcycle, box, ir->fepvals, cr->ms,
idef, (const rvec *) x, hist, f, fr,
&pbc, graph, enerd, nrnb, lambda, md, fcd,
DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL,
flags);
where();
*cycles_pme = 0;
clear_mat(fr->vir_el_recip);
clear_mat(fr->vir_lj_recip);
/* Do long-range electrostatics and/or LJ-PME, including related short-range
* corrections.
*/
if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
{
int status = 0;
real Vlr_q = 0, Vlr_lj = 0, Vcorr_q = 0, Vcorr_lj = 0;
real dvdl_long_range_q = 0, dvdl_long_range_lj = 0;
bSB = (ir->nwall == 2);
if (bSB)
{
copy_mat(box, boxs);
svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
box_size[ZZ] *= ir->wall_ewald_zfac;
}
if (EEL_PME_EWALD(fr->eeltype) || EVDW_PME(fr->vdwtype))
{
real dvdl_long_range_correction_q = 0;
real dvdl_long_range_correction_lj = 0;
/* With the Verlet scheme exclusion forces are calculated
* in the non-bonded kernel.
*/
/* The TPI molecule does not have exclusions with the rest
* of the system and no intra-molecular PME grid
* contributions will be calculated in
* gmx_pme_calc_energy.
*/
if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
ir->ewald_geometry != eewg3D ||
ir->epsilon_surface != 0)
{
int nthreads, t;
wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
if (fr->n_tpi > 0)
{
gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
}
nthreads = gmx_omp_nthreads_get(emntBonded);
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (t = 0; t < nthreads; t++)
{
int i;
rvec *fnv;
tensor *vir_q, *vir_lj;
real *Vcorrt_q, *Vcorrt_lj, *dvdlt_q, *dvdlt_lj;
if (t == 0)
{
fnv = fr->f_novirsum;
vir_q = &fr->vir_el_recip;
vir_lj = &fr->vir_lj_recip;
Vcorrt_q = &Vcorr_q;
Vcorrt_lj = &Vcorr_lj;
dvdlt_q = &dvdl_long_range_correction_q;
dvdlt_lj = &dvdl_long_range_correction_lj;
}
else
{
fnv = fr->f_t[t].f;
vir_q = &fr->f_t[t].vir_q;
vir_lj = &fr->f_t[t].vir_lj;
Vcorrt_q = &fr->f_t[t].Vcorr_q;
Vcorrt_lj = &fr->f_t[t].Vcorr_lj;
dvdlt_q = &fr->f_t[t].dvdl[efptCOUL];
dvdlt_lj = &fr->f_t[t].dvdl[efptVDW];
for (i = 0; i < fr->natoms_force; i++)
{
clear_rvec(fnv[i]);
}
clear_mat(*vir_q);
clear_mat(*vir_lj);
}
*dvdlt_q = 0;
*dvdlt_lj = 0;
ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
cr, t, fr,
md->chargeA, md->chargeB,
md->sqrt_c6A, md->sqrt_c6B,
md->sigmaA, md->sigmaB,
md->sigma3A, md->sigma3B,
md->nChargePerturbed || md->nTypePerturbed,
ir->cutoff_scheme != ecutsVERLET,
excl, x, bSB ? boxs : box, mu_tot,
ir->ewald_geometry,
ir->epsilon_surface,
fnv, *vir_q, *vir_lj,
Vcorrt_q, Vcorrt_lj,
lambda[efptCOUL], lambda[efptVDW],
dvdlt_q, dvdlt_lj);
}
if (nthreads > 1)
{
reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
fr->vir_el_recip, fr->vir_lj_recip,
&Vcorr_q, &Vcorr_lj,
&dvdl_long_range_correction_q,
&dvdl_long_range_correction_lj,
nthreads, fr->f_t);
}
wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
}
if (EEL_PME_EWALD(fr->eeltype) && fr->n_tpi == 0)
{
/* This is not in a subcounter because it takes a
negligible and constant-sized amount of time */
Vcorr_q += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
&dvdl_long_range_correction_q,
fr->vir_el_recip);
}
enerd->dvdl_lin[efptCOUL] += dvdl_long_range_correction_q;
enerd->dvdl_lin[efptVDW] += dvdl_long_range_correction_lj;
if ((EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype)) && (cr->duty & DUTY_PME))
{
/* Do reciprocal PME for Coulomb and/or LJ. */
assert(fr->n_tpi >= 0);
if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
{
pme_flags = GMX_PME_SPREAD | GMX_PME_SOLVE;
if (EEL_PME(fr->eeltype))
{
pme_flags |= GMX_PME_DO_COULOMB;
}
if (EVDW_PME(fr->vdwtype))
{
pme_flags |= GMX_PME_DO_LJ;
}
if (flags & GMX_FORCE_FORCES)
{
pme_flags |= GMX_PME_CALC_F;
}
if (flags & GMX_FORCE_VIRIAL)
{
pme_flags |= GMX_PME_CALC_ENER_VIR;
}
if (fr->n_tpi > 0)
{
/* We don't calculate f, but we do want the potential */
pme_flags |= GMX_PME_CALC_POT;
}
wallcycle_start(wcycle, ewcPMEMESH);
status = gmx_pme_do(fr->pmedata,
0, md->homenr - fr->n_tpi,
x, fr->f_novirsum,
md->chargeA, md->chargeB,
md->sqrt_c6A, md->sqrt_c6B,
md->sigmaA, md->sigmaB,
bSB ? boxs : box, cr,
DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
nrnb, wcycle,
fr->vir_el_recip, fr->ewaldcoeff_q,
fr->vir_lj_recip, fr->ewaldcoeff_lj,
&Vlr_q, &Vlr_lj,
lambda[efptCOUL], lambda[efptVDW],
&dvdl_long_range_q, &dvdl_long_range_lj, pme_flags);
*cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
if (status != 0)
{
gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
}
/* We should try to do as little computation after
* this as possible, because parallel PME synchronizes
* the nodes, so we want all load imbalance of the
* rest of the force calculation to be before the PME
* call. DD load balancing is done on the whole time
* of the force call (without PME).
*/
}
if (fr->n_tpi > 0)
{
if (EVDW_PME(ir->vdwtype))
{
gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
}
/* Determine the PME grid energy of the test molecule
* with the PME grid potential of the other charges.
*/
gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
x + md->homenr - fr->n_tpi,
md->chargeA + md->homenr - fr->n_tpi,
&Vlr_q);
}
}
}
if (!EEL_PME(fr->eeltype) && EEL_PME_EWALD(fr->eeltype))
{
Vlr_q = do_ewald(ir, x, fr->f_novirsum,
md->chargeA, md->chargeB,
box_size, cr, md->homenr,
fr->vir_el_recip, fr->ewaldcoeff_q,
lambda[efptCOUL], &dvdl_long_range_q, fr->ewald_table);
}
/* Note that with separate PME nodes we get the real energies later */
enerd->dvdl_lin[efptCOUL] += dvdl_long_range_q;
enerd->dvdl_lin[efptVDW] += dvdl_long_range_lj;
enerd->term[F_COUL_RECIP] = Vlr_q + Vcorr_q;
enerd->term[F_LJ_RECIP] = Vlr_lj + Vcorr_lj;
if (debug)
{
fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
Vlr_q, Vcorr_q, enerd->term[F_COUL_RECIP]);
pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
Vlr_lj, Vcorr_lj, enerd->term[F_LJ_RECIP]);
pr_rvecs(debug, 0, "vir_lj_recip after corr", fr->vir_lj_recip, DIM);
}
}
else
{
/* Is there a reaction-field exclusion correction needed? */
if (EEL_RF(fr->eeltype) && eelRF_NEC != fr->eeltype)
{
/* With the Verlet scheme, exclusion forces are calculated
* in the non-bonded kernel.
*/
if (ir->cutoff_scheme != ecutsVERLET)
{
real dvdl_rf_excl = 0;
enerd->term[F_RF_EXCL] =
RF_excl_correction(fr, graph, md, excl, x, f,
fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
}
}
}
where();
debug_gmx();
if (debug)
{
print_nrnb(debug, nrnb);
}
debug_gmx();
#ifdef GMX_MPI
if (TAKETIME)
{
t2 = MPI_Wtime();
MPI_Barrier(cr->mpi_comm_mygroup);
t3 = MPI_Wtime();
fr->t_wait += t3-t2;
if (fr->timesteps == 11)
{
char buf[22];
fprintf(stderr, "* PP load balancing info: rank %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
cr->nodeid, gmx_step_str(fr->timesteps, buf),
100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
(fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
}
fr->timesteps++;
}
#endif
if (debug)
{
pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
}
}
void mc_pscale(t_inputrec *ir,matrix mu,
matrix box,matrix box_rel,
int start,int nr_atoms,
rvec x[],unsigned short cFREEZE[],
t_nrnb *nrnb,t_block *mols,rvec *xcm,t_graph *graph)
{
ivec *nFreeze=ir->opts.nFreeze;
int n,m,d,g=0,natoms;
real dv;
rvec dxcm,dx,*xs;
snew(xs,nr_atoms);
for(n=start;n<nr_atoms;n++)
{
copy_rvec(x[n],xs[n]);
}
shift_self(graph,box,xs);
for (d=0; d<DIM; d++) {
box[d][XX] = mu[XX][XX]*box[d][XX]+mu[YY][XX]*box[d][YY]+mu[ZZ][XX]*box[d][ZZ];
box[d][YY] = mu[YY][YY]*box[d][YY]+mu[ZZ][YY]*box[d][ZZ];
box[d][ZZ] = mu[ZZ][ZZ]*box[d][ZZ];
}
preserve_box_shape(ir,box_rel,box);
/* Scale the positions */
shift_self(graph,box,x);
for (n=0; n<mols->nr; n++) {
clear_rvec(dxcm);
natoms = mols->index[n+1]-mols->index[n];
dxcm[XX] = mu[XX][XX]*xcm[n][XX]+mu[YY][XX]*xcm[n][YY]+mu[ZZ][XX]*xcm[n][ZZ];
dxcm[YY] = mu[YY][YY]*xcm[n][YY]+mu[ZZ][YY]*xcm[n][ZZ];
dxcm[ZZ] = mu[ZZ][ZZ]*xcm[n][ZZ];
for(m=mols->index[n];m<mols->index[n+1];m++) {
//rvec_sub(x[m],xcm[n],dx);
rvec_sub(dxcm,xcm[n],dx);
rvec_add(xs[m],dx,x[m]);
}
}
unshift_self(graph,box,x);
/*for (n=start; n<start+nr_atoms; n++) {
x[n][XX] = mu[XX][XX]*x[n][XX]+mu[YY][XX]*x[n][YY]+mu[ZZ][XX]*x[n][ZZ];
x[n][YY] = mu[YY][YY]*x[n][YY]+mu[ZZ][YY]*x[n][ZZ];
x[n][ZZ] = mu[ZZ][ZZ]*x[n][ZZ];
}*/
/* (un)shifting should NOT be done after this,
* since the box vectors might have changed
*/
inc_nrnb(nrnb,eNR_PCOUPL,nr_atoms);
sfree(xs);
}
void construct_dummies(FILE *log,rvec x[],t_nrnb *nrnb,real dt,
rvec *v,t_idef *idef,t_graph *graph,t_commrec *cr,
matrix box,t_comm_dummies *dummycomm)
{
rvec xd,vv;
real a1,b1,c1,inv_dt;
int i,ii,nra,nrd,tp,ftype;
t_iatom adum,ai,aj,ak,al;
t_iatom *ia;
t_iparams *ip;
/* Molecules always whole, but I'm not sure whether
* the periodicity and shift are guaranteed to be consistent
* between different nodes when running e.g. polymers in
* parallel. In this special case we thus unshift/shift, but
* only when necessary. This is to make sure the coordinates
* we move don't end up a box away...
*/
if (dummycomm) {
unshift_self(graph,box,x);
move_construct_x(dummycomm,x,cr);
shift_self(graph,box,x);
}
ip = idef->iparams;
if (v)
inv_dt = 1.0/dt;
else
inv_dt = 1.0;
for(ftype=0; (ftype<F_NRE); ftype++) {
if (interaction_function[ftype].flags & IF_DUMMY) {
nra = interaction_function[ftype].nratoms;
nrd = idef->il[ftype].nr;
ia = idef->il[ftype].iatoms;
for(i=0; (i<nrd); ) {
tp = ia[0];
assert(ftype == idef->functype[tp]);
/* The dummy and constructing atoms */
adum = ia[1];
ai = ia[2];
aj = ia[3];
/* Constants for constructing dummies */
a1 = ip[tp].dummy.a;
/* Copy the old position */
copy_rvec(x[adum],xd);
/* Construct the dummy depending on type */
switch (ftype) {
case F_DUMMY2:
constr_dum2(x[ai],x[aj],x[adum],a1);
break;
case F_DUMMY3:
ak = ia[4];
b1 = ip[tp].dummy.b;
constr_dum3(x[ai],x[aj],x[ak],x[adum],a1,b1);
break;
case F_DUMMY3FD:
ak = ia[4];
b1 = ip[tp].dummy.b;
constr_dum3FD(x[ai],x[aj],x[ak],x[adum],a1,b1);
break;
case F_DUMMY3FAD:
ak = ia[4];
b1 = ip[tp].dummy.b;
constr_dum3FAD(x[ai],x[aj],x[ak],x[adum],a1,b1);
break;
case F_DUMMY3OUT:
ak = ia[4];
b1 = ip[tp].dummy.b;
c1 = ip[tp].dummy.c;
constr_dum3OUT(x[ai],x[aj],x[ak],x[adum],a1,b1,c1);
break;
case F_DUMMY4FD:
ak = ia[4];
al = ia[5];
b1 = ip[tp].dummy.b;
c1 = ip[tp].dummy.c;
constr_dum4FD(x[ai],x[aj],x[ak],x[al],x[adum],a1,b1,c1);
break;
default:
fatal_error(0,"No such dummy type %d in %s, line %d",
ftype,__FILE__,__LINE__);
}
if (v) {
/* Calculate velocity of dummy... */
rvec_sub(x[adum],xd,vv);
svmul(inv_dt,vv,v[adum]);
}
/* Increment loop variables */
i += nra+1;
ia += nra+1;
}
}
}
if (dummycomm) {
unshift_self(graph,box,x);
move_dummy_xv(dummycomm,x,NULL,cr);
shift_self(graph,box,x); /* maybe not necessary */
}
}
int main(int argc,char *argv[])
{
static char *desc[] = {
"testlr tests the PPPM and Ewald method for the",
"long range electrostatics problem."
};
static t_filenm fnm[] = {
{ efTPX, NULL, NULL, ffREAD },
{ efHAT, "-g", "ghat", ffOPTRD },
{ efOUT, "-o", "rho", ffOPTWR },
{ efOUT, "-op", "lr-pb", ffOPTWR },
{ efOUT, "-of", "lr-four", ffOPTWR },
{ efOUT, "-opt", "tot-pb", ffOPTWR },
{ efOUT, "-oft", "tot-four", ffOPTWR },
{ efOUT, "-fin", "lr-four", ffOPTWR },
{ efEPS, "-es", "sr", ffOPTWR },
{ efEPS, "-elf", "lr-four", ffOPTWR },
{ efEPS, "-etf", "tot-four", ffOPTWR },
{ efEPS, "-qr", "qk-real", ffOPTWR },
{ efEPS, "-qi", "qk-im", ffOPTWR },
{ efEPS, "-elp", "lr-pb", ffOPTWR },
{ efEPS, "-etp", "tot-pb", ffOPTWR },
{ efEPS, "-rho", "rho", ffOPTWR },
{ efEPS, "-qq", "charge", ffOPTWR },
{ efXVG, "-gt", "gk-tab", ffOPTWR },
{ efXVG, "-fcorr","fcorr", ffWRITE },
{ efXVG, "-pcorr","pcorr", ffWRITE },
{ efXVG, "-ftotcorr","ftotcorr", ffWRITE },
{ efXVG, "-ptotcorr","ptotcorr", ffWRITE },
{ efLOG, "-l", "fptest", ffWRITE },
{ efXVG, "-gr", "spread", ffOPTWR },
{ efPDB, "-pf", "pqr-four", ffOPTWR },
{ efPDB, "-phitot", "pppm-phitot", ffOPTWR }
};
#define NFILE asize(fnm)
FILE *log;
t_topology top;
t_tpxheader stath;
t_inputrec ir;
t_block *excl;
t_forcerec *fr;
t_commrec *cr;
t_mdatoms *mdatoms;
t_graph *graph;
int i,step,nre,natoms,nmol;
rvec *x,*f_sr,*f_excl,*f_four,*f_pppm,*f_pois,box_size,hbox;
matrix box;
real t,lambda,vsr,*charge,*phi_f,*phi_pois,*phi_s,*phi_p3m,*rho;
static bool bFour=FALSE,bVerbose=FALSE,bGGhat=FALSE,bPPPM=TRUE,
bPoisson=FALSE,bOld=FALSE,bOldEwald=TRUE;
static int nprocs = 1;
static t_pargs pa[] = {
{ "-np", FALSE, etINT, &nprocs, "Do it in parallel" },
{ "-ewald", FALSE, etBOOL, &bFour, "Do an Ewald solution"},
{ "-pppm", FALSE, etBOOL, &bPPPM, "Do a PPPM solution" },
{ "-poisson",FALSE, etBOOL, &bPoisson,"Do a Poisson solution" },
{ "-v", FALSE, etBOOL, &bVerbose,"Verbose on"},
{ "-ghat", FALSE, etBOOL, &bGGhat, "Generate Ghat function"},
{ "-old", FALSE, etBOOL, &bOld, "Use old function types"},
{ "-oldewald",FALSE,etBOOL, &bOldEwald,"Use old Ewald code"}
};
CopyRight(stderr,argv[0]);
parse_common_args(&argc,argv,PCA_CAN_TIME | PCA_CAN_VIEW,
NFILE,fnm,asize(pa),pa,asize(desc),desc,0,NULL);
if (nprocs > 1) {
cr = init_par(&argc,argv);
open_log(ftp2fn(efLOG,NFILE,fnm),cr);
log = stdlog;
}
else {
cr = init_par(&argc,argv);
log = ftp2FILE(efLOG,NFILE,fnm,"w");
stdlog = log; }
/* Read topology and coordinates */
read_tpxheader(ftp2fn(efTPX,NFILE,fnm),&stath,FALSE);
snew(x,stath.natoms);
snew(f_sr,stath.natoms);
snew(f_excl,stath.natoms);
snew(f_four,stath.natoms);
snew(f_pppm,stath.natoms);
snew(f_pois,stath.natoms);
read_tpx(ftp2fn(efTPX,NFILE,fnm),&step,&t,&lambda,&ir,
box,&natoms,x,NULL,NULL,&top);
excl=&(top.atoms.excl);
nmol=top.blocks[ebMOLS].nr;
/* Allocate space for potential, charges and rho (charge density) */
snew(charge,stath.natoms);
snew(phi_f,stath.natoms);
snew(phi_p3m,stath.natoms);
snew(phi_pois,stath.natoms);
snew(phi_s,stath.natoms);
snew(rho,stath.natoms);
/* Set the charges */
for(i=0; (i<natoms); i++)
charge[i]=top.atoms.atom[i].q;
/* Make a simple box vector instead of tensor */
for(i=0; (i<DIM); i++)
box_size[i]=box[i][i];
/* Set some constants */
fr = mk_forcerec();
mdatoms = atoms2md(&(top.atoms),FALSE,FALSE);
set_LRconsts(log,ir.rcoulomb_switch,ir.rcoulomb,box_size,fr);
init_forcerec(log,fr,&ir,&(top.blocks[ebMOLS]),cr,
&(top.blocks[ebCGS]),&(top.idef),mdatoms,box,FALSE);
calc_shifts(box,box_size,fr->shift_vec,FALSE);
/* Periodicity stuff */
graph = mk_graph(&(top.idef),top.atoms.nr,FALSE,FALSE);
shift_self(graph,fr->shift_vec,x);
calc_LRcorrections(log,0,natoms,ir.rcoulomb_switch,
ir.rcoulomb,charge,excl,x,f_excl,bOld);
pr_f("f_excl.dat",natoms,f_excl);
/* Compute the short range potential */
put_atoms_in_box(natoms,box,x);
vsr=phi_sr(log,natoms,x,charge,ir.rcoulomb,
ir.rcoulomb_switch,box_size,phi_s,excl,f_sr,bOld);
pr_f("f_sr.dat",natoms,f_sr);
/* Plot the short range potential in a matrix */
calc_ener(log,"Short Range",TRUE,nmol,natoms,phi_s,charge,excl);
if (bFour)
test_four(log,NFILE,fnm,&(top.atoms),&ir,x,f_four,box_size,charge,phi_f,
phi_s,nmol,cr,bOld,bOldEwald);
if (bPPPM)
test_pppm(log,bVerbose,bGGhat,opt2fn("-g",NFILE,fnm),
&(top.atoms),&ir,x,f_pppm,charge,box_size,phi_p3m,phi_s,nmol,
cr,bOld,&(top.blocks[ebCGS]));
if (bPoisson)
test_poisson(log,bVerbose,
&(top.atoms),&ir,x,f_pois,charge,box_size,phi_pois,
phi_s,nmol,cr,bFour,f_four,phi_f,bOld);
if (bPPPM && bFour)
analyse_diff(log,"PPPM",
top.atoms.nr,f_four,f_pppm,phi_f,phi_p3m,phi_s,
opt2fn("-fcorr",NFILE,fnm),
opt2fn("-pcorr",NFILE,fnm),
opt2fn("-ftotcorr",NFILE,fnm),
opt2fn("-ptotcorr",NFILE,fnm));
if (bPoisson && bFour)
analyse_diff(log,"Poisson",
top.atoms.nr,f_four,f_pois,phi_f,phi_pois,phi_s,
opt2fn("-fcorr",NFILE,fnm),
opt2fn("-pcorr",NFILE,fnm),
opt2fn("-ftotcorr",NFILE,fnm),
opt2fn("-ptotcorr",NFILE,fnm));
gmx_fio_fclose(log);
thanx(stderr);
return 0;
}
void force(FILE *fp, int step,
t_forcerec *fr, t_inputrec *ir,
t_idef *idef, t_nsborder *nsb,
t_commrec *cr, t_commrec *mcr,
t_nrnb *nrnb,
t_groups *grps, t_mdatoms *md,
int ngener, t_grpopts *opts,
rvec x[], rvec f[],
real epot[], t_fcdata *fcd,
bool bVerbose, matrix box,
real lambda, t_graph *graph,
t_block *excl, bool bNBFonly,
matrix lr_vir, rvec mu_tot,
real qsum, bool bGatherOnly)
{
int i,nit;
bool bDoEpot;
rvec box_size;
real Vlr,Vcorr=0;
/* Reset box */
for(i=0; (i<DIM); i++)
box_size[i]=box[i][i];
bDoEpot=((fr->nmol > 0) && (fr->nstcalc > 0) && (mod(step,fr->nstcalc)==0));
/* Reset epot... */
if (bDoEpot)
for(i=0; (i<fr->nmol); i++)
fr->mol_epot[i]=0.0;
debug_gmx();
/* Call the short range functions all in one go. */
do_fnbf(fp,cr,fr,x,f,md,
fr->bBHAM ? grps->estat.ee[egBHAM] : grps->estat.ee[egLJ],
grps->estat.ee[egCOUL],box_size,nrnb,
lambda,&epot[F_DVDL],FALSE,-1);
debug_gmx();
if (debug)
pr_rvecs(debug,0,"fshift after SR",fr->fshift,SHIFTS);
/* Shift the coordinates. Must be done before bonded forces and PPPM,
* but is also necessary for SHAKE and update, therefore it can NOT
* go when no bonded forces have to be evaluated.
*/
if (debug && 0)
p_graph(debug,"DeBUGGGG",graph);
/* Check whether we need to do bondeds */
if (!bNBFonly) {
shift_self(graph,box,x);
if (debug && 0) {
fprintf(debug,"BBBBBBBBBBBBBBBB\n");
fprintf(debug,"%5d\n",graph->nnodes);
for(i=graph->start; (i<=graph->end); i++)
fprintf(debug,"%5d%5s%5s%5d%8.3f%8.3f%8.3f\n",
i,"A","B",i,x[i][XX],x[i][YY],x[i][ZZ]);
fprintf(debug,"%10.5f%10.5f%10.5f\n",
box[XX][XX],box[YY][YY],box[ZZ][ZZ]);
}
if (TRICLINIC(box))
inc_nrnb(nrnb,eNR_SHIFTX,2*graph->nnodes);
else
inc_nrnb(nrnb,eNR_SHIFTX,graph->nnodes);
debug_gmx();
}
if (EEL_LR(fr->eeltype)) {
switch (fr->eeltype) {
case eelPPPM:
Vlr = do_pppm(fp,FALSE,x,fr->f_pme,md->chargeT,
box_size,fr->phi,cr,nsb,nrnb);
break;
case eelPOISSON:
Vlr = do_poisson(fp,FALSE,ir,md->nr,x,fr->f_pme,md->chargeT,
box_size,fr->phi,cr,nrnb,&nit,TRUE);
break;
case eelPME:
Vlr = do_pme(fp,FALSE,ir,x,fr->f_pme,md->chargeT,
box,cr,nsb,nrnb,lr_vir,fr->ewaldcoeff,bGatherOnly);
break;
case eelEWALD:
Vlr = do_ewald(fp,FALSE,ir,x,fr->f_pme,md->chargeT,
box_size,cr,nsb,lr_vir,fr->ewaldcoeff);
break;
default:
Vlr = 0;
fatal_error(0,"No such electrostatics method implemented %s",
eel_names[fr->eeltype]);
}
if(fr->bEwald)
Vcorr =
ewald_LRcorrection(fp,nsb,cr,fr,md->chargeT,excl,x,box,mu_tot,qsum,
ir->ewald_geometry,ir->epsilon_surface,lr_vir);
else
Vcorr = shift_LRcorrection(fp,nsb,cr,fr,md->chargeT,excl,x,TRUE,box,lr_vir);
epot[F_LR] = Vlr + Vcorr;
if (debug)
fprintf(debug,"Vlr = %g, Vcorr = %g, Vlr_corr = %g\n",
Vlr,Vcorr,epot[F_LR]);
if (debug) {
pr_rvecs(debug,0,"lr_vir after corr",lr_vir,DIM);
pr_rvecs(debug,0,"fshift after LR Corrections",fr->fshift,SHIFTS);
}
}
debug_gmx();
if (debug)
print_nrnb(debug,nrnb);
debug_gmx();
if (!bNBFonly) {
calc_bonds(fp,cr,mcr,
idef,x,f,fr,graph,epot,nrnb,box,lambda,md,
opts->ngener,grps->estat.ee[egLJ14],grps->estat.ee[egCOUL14],
fcd,step,fr->bSepDVDL && do_per_step(step,ir->nstlog));
debug_gmx();
}
if (debug)
pr_rvecs(debug,0,"fshift after bondeds",fr->fshift,SHIFTS);
for(i=0; (i<F_EPOT); i++)
if (i != F_DISRES)
epot[F_EPOT]+=epot[i];
}
