* To help us fund GROMACS development, we humbly ask that you cite
* the research papers on the package. Check out http://www.gromacs.org.
*/
-#ifdef HAVE_CONFIG_H
-#include <config.h>
-#endif
+#include "gmxpre.h"
-#include <stdlib.h>
+#include "config.h"
-#include "typedefs.h"
-#include "gromacs/math/vec.h"
-#include "vcm.h"
-#include "mdebin.h"
-#include "nrnb.h"
-#include "calcmu.h"
-#include "vsite.h"
-#include "update.h"
-#include "ns.h"
-#include "mdrun.h"
-#include "md_support.h"
-#include "md_logging.h"
-#include "network.h"
-#include "names.h"
-#include "force.h"
-#include "disre.h"
-#include "orires.h"
-#include "pme.h"
-#include "mdatoms.h"
-#include "repl_ex.h"
-#include "deform.h"
-#include "qmmm.h"
-#include "domdec.h"
-#include "domdec_network.h"
-#include "coulomb.h"
-#include "constr.h"
-#include "shellfc.h"
-#include "gromacs/gmxpreprocess/compute_io.h"
-#include "checkpoint.h"
-#include "gromacs/topology/mtop_util.h"
-#include "sighandler.h"
-#include "txtdump.h"
-#include "gromacs/utility/cstringutil.h"
-#include "pme_loadbal.h"
-#include "bondf.h"
-#include "membed.h"
-#include "types/nlistheuristics.h"
-#include "types/iteratedconstraints.h"
-#include "nbnxn_cuda_data_mgmt.h"
+#include <stdlib.h>
#include "gromacs/fileio/confio.h"
#include "gromacs/fileio/mdoutf.h"
#include "gromacs/fileio/trnio.h"
#include "gromacs/fileio/trxio.h"
#include "gromacs/fileio/xtcio.h"
+#include "gromacs/gmxpreprocess/compute_io.h"
#include "gromacs/imd/imd.h"
+#include "gromacs/legacyheaders/bonded-threading.h"
+#include "gromacs/legacyheaders/calcmu.h"
+#include "gromacs/legacyheaders/checkpoint.h"
+#include "gromacs/legacyheaders/constr.h"
+#include "gromacs/legacyheaders/coulomb.h"
+#include "gromacs/legacyheaders/disre.h"
+#include "gromacs/legacyheaders/domdec.h"
+#include "gromacs/legacyheaders/domdec_network.h"
+#include "gromacs/legacyheaders/force.h"
+#include "gromacs/legacyheaders/md_logging.h"
+#include "gromacs/legacyheaders/md_support.h"
+#include "gromacs/legacyheaders/mdatoms.h"
+#include "gromacs/legacyheaders/mdebin.h"
+#include "gromacs/legacyheaders/mdrun.h"
+#include "gromacs/legacyheaders/names.h"
+#include "gromacs/legacyheaders/network.h"
+#include "gromacs/legacyheaders/nrnb.h"
+#include "gromacs/legacyheaders/ns.h"
+#include "gromacs/legacyheaders/orires.h"
+#include "gromacs/legacyheaders/pme.h"
+#include "gromacs/legacyheaders/qmmm.h"
+#include "gromacs/legacyheaders/shellfc.h"
+#include "gromacs/legacyheaders/sighandler.h"
+#include "gromacs/legacyheaders/txtdump.h"
+#include "gromacs/legacyheaders/typedefs.h"
+#include "gromacs/legacyheaders/update.h"
+#include "gromacs/legacyheaders/vcm.h"
+#include "gromacs/legacyheaders/vsite.h"
+#include "gromacs/legacyheaders/types/iteratedconstraints.h"
+#include "gromacs/legacyheaders/types/nlistheuristics.h"
+#include "gromacs/math/vec.h"
+#include "gromacs/mdlib/nbnxn_cuda/nbnxn_cuda_data_mgmt.h"
#include "gromacs/pbcutil/mshift.h"
#include "gromacs/pbcutil/pbc.h"
#include "gromacs/pulling/pull.h"
#include "gromacs/swap/swapcoords.h"
#include "gromacs/timing/wallcycle.h"
#include "gromacs/timing/walltime_accounting.h"
+#include "gromacs/topology/mtop_util.h"
+#include "gromacs/utility/cstringutil.h"
#include "gromacs/utility/gmxmpi.h"
#include "gromacs/utility/smalloc.h"
+#include "deform.h"
+#include "membed.h"
+#include "pme_loadbal.h"
+#include "repl_ex.h"
+
#ifdef GMX_FAHCORE
#include "corewrap.h"
#endif
gmx_int64_t *step_rel, t_inputrec *ir,
gmx_wallcycle_t wcycle, t_nrnb *nrnb,
gmx_walltime_accounting_t walltime_accounting,
- nbnxn_cuda_ptr_t cu_nbv)
+ struct nonbonded_verlet_t *nbv)
{
char sbuf[STEPSTRSIZE];
md_print_warn(cr, fplog, "step %s: resetting all time and cycle counters\n",
gmx_step_str(step, sbuf));
- if (cu_nbv)
- {
- nbnxn_cuda_reset_timings(cu_nbv);
- }
+ nbnxn_cuda_reset_timings(nbv);
wallcycle_stop(wcycle, ewcRUN);
wallcycle_reset_all(wcycle);
gmx_shellfc_t shellfc;
int count, nconverged = 0;
double tcount = 0;
- gmx_bool bConverged = TRUE, bOK, bSumEkinhOld, bExchanged, bNeedRepartition;
+ gmx_bool bConverged = TRUE, bSumEkinhOld, bDoReplEx, bExchanged, bNeedRepartition;
gmx_bool bResetCountersHalfMaxH = FALSE;
gmx_bool bVV, bIterativeCase, bFirstIterate, bTemp, bPres, bTrotter;
gmx_bool bUpdateDoLR;
&(state_global->fep_state), lam0,
nrnb, top_global, &upd,
nfile, fnm, &outf, &mdebin,
- force_vir, shake_vir, mu_tot, &bSimAnn, &vcm, Flags);
+ force_vir, shake_vir, mu_tot, &bSimAnn, &vcm, Flags, wcycle);
clear_mat(total_vir);
clear_mat(pres);
debug_gmx();
/* Check for polarizable models and flexible constraints */
- shellfc = init_shell_flexcon(fplog, fr->cutoff_scheme == ecutsVERLET,
+ shellfc = init_shell_flexcon(fplog,
top_global, n_flexible_constraints(constr),
(ir->bContinuation ||
(DOMAINDECOMP(cr) && !MASTER(cr))) ?
NULL : state_global->x);
-
+ if (shellfc && ir->nstcalcenergy != 1)
+ {
+ gmx_fatal(FARGS, "You have nstcalcenergy set to a value (%d) that is different from 1.\nThis is not supported in combinations with shell particles.\nPlease make a new tpr file.", ir->nstcalcenergy);
+ }
+ if (shellfc && DOMAINDECOMP(cr))
+ {
+ gmx_fatal(FARGS, "Shell particles are not implemented with domain decomposition, use a single rank");
+ }
if (shellfc && ir->eI == eiNM)
{
/* Currently shells don't work with Normal Modes */
set_constraints(constr, top, ir, mdatoms, cr);
}
- if (repl_ex_nst > 0)
- {
- /* We need to be sure replica exchange can only occur
- * when the energies are current */
- check_nst_param(fplog, cr, "nstcalcenergy", ir->nstcalcenergy,
- "repl_ex_nst", &repl_ex_nst);
- /* This check needs to happen before inter-simulation
- * signals are initialized, too */
- }
if (repl_ex_nst > 0 && MASTER(cr))
{
repl_ex = init_replica_exchange(fplog, cr->ms, state_global, ir,
*/
if ((Flags & MD_TUNEPME) &&
EEL_PME(fr->eeltype) &&
- ( (fr->cutoff_scheme == ecutsVERLET && fr->nbv->bUseGPU) || !(cr->duty & DUTY_PME)) &&
+ ( use_GPU(fr->nbv) || !(cr->duty & DUTY_PME)) &&
!bRerunMD)
{
pme_loadbal_init(&pme_loadbal, ir, state->box, fr->ic, fr->pmedata);
&& (ir->bExpanded) && (step > 0) && (!bStartingFromCpt));
}
+ bDoReplEx = ((repl_ex_nst > 0) && (step > 0) && !bLastStep &&
+ do_per_step(step, repl_ex_nst));
+
if (bSimAnn)
{
update_annealing_target_temp(&(ir->opts), t);
do_ene = (do_per_step(step, ir->nstenergy) || bLastStep);
- if (do_ene || do_log)
+ if (do_ene || do_log || bDoReplEx)
{
bCalcVir = TRUE;
bCalcEner = TRUE;
if (bVV && !bStartingFromCpt && !bRerunMD)
/* ############### START FIRST UPDATE HALF-STEP FOR VV METHODS############### */
{
+ wallcycle_start(wcycle, ewcUPDATE);
if (ir->eI == eiVV && bInitStep)
{
/* if using velocity verlet with full time step Ekin,
}
}
- bOK = TRUE;
if (!bRerunMD || rerun_fr.bV || bForceUpdate) /* Why is rerun_fr.bV here? Unclear. */
{
+ wallcycle_stop(wcycle, ewcUPDATE);
update_constraints(fplog, step, NULL, ir, ekind, mdatoms,
state, fr->bMolPBC, graph, f,
&top->idef, shake_vir,
cr, nrnb, wcycle, upd, constr,
TRUE, bCalcVir, vetanew);
+ wallcycle_start(wcycle, ewcUPDATE);
if (bCalcVir && bUpdateDoLR && ir->nstcalclr > 1)
{
/* Correct the virial for multiple time stepping */
m_sub(shake_vir, fr->vir_twin_constr, shake_vir);
}
-
- if (!bOK)
- {
- gmx_fatal(FARGS, "Constraint error: Shake, Lincs or Settle could not solve the constrains");
- }
-
}
else if (graph)
{
So we need information from the last step in the first half of the integration */
if (bGStat || do_per_step(step-1, nstglobalcomm))
{
+ wallcycle_stop(wcycle, ewcUPDATE);
compute_globals(fplog, gstat, cr, ir, fr, ekind, state, state_global, mdatoms, nrnb, vcm,
wcycle, enerd, force_vir, shake_vir, total_vir, pres, mu_tot,
constr, NULL, FALSE, state->box,
time step kinetic energy for the pressure (always true now, since we want accurate statistics).
b) If we are using EkinAveEkin for the kinetic energy for the temperature control, we still feed in
EkinAveVel because it's needed for the pressure */
+ wallcycle_start(wcycle, ewcUPDATE);
}
/* temperature scaling and pressure scaling to produce the extended variables at t+dt */
if (!bInitStep)
{
if (bExchanged)
{
-
+ wallcycle_stop(wcycle, ewcUPDATE);
/* We need the kinetic energy at minus the half step for determining
* the full step kinetic energy and possibly for T-coupling.*/
/* This may not be quite working correctly yet . . . . */
constr, NULL, FALSE, state->box,
top_global, &bSumEkinhOld,
CGLO_RERUNMD | CGLO_GSTAT | CGLO_TEMPERATURE);
+ wallcycle_start(wcycle, ewcUPDATE);
}
}
}
{
copy_rvecn(cbuf, state->v, 0, state->natoms);
}
+ wallcycle_stop(wcycle, ewcUPDATE);
}
/* MRS -- now done iterating -- compute the conserved quantity */
do_md_trajectory_writing(fplog, cr, nfile, fnm, step, step_rel, t,
ir, state, state_global, top_global, fr,
outf, mdebin, ekind, f, f_global,
- wcycle, &nchkpt,
+ &nchkpt,
bCPT, bRerunMD, bLastStep, (Flags & MD_CONFOUT),
bSumEkinhOld);
/* Check if IMD step and do IMD communication, if bIMD is TRUE. */
*/
copy_mat(state->box, lastbox);
- bOK = TRUE;
dvdl_constr = 0;
if (!(bRerunMD && !rerun_fr.bV && !bForceUpdate))
FALSE, bCalcVir,
state->veta);
}
- if (!bOK)
- {
- gmx_fatal(FARGS, "Constraint error: Shake, Lincs or Settle could not solve the constrains");
- }
- if (fr->bSepDVDL && fplog && do_log)
- {
- gmx_print_sepdvdl(fplog, "Constraint dV/dl", 0.0, dvdl_constr);
- }
if (bVV)
{
/* this factor or 2 correction is necessary
/* Replica exchange */
bExchanged = FALSE;
- if ((repl_ex_nst > 0) && (step > 0) && !bLastStep &&
- do_per_step(step, repl_ex_nst))
+ if (bDoReplEx)
{
bExchanged = replica_exchange(fplog, cr, repl_ex,
state_global, enerd,
}
dd_bcast(cr->dd, sizeof(gmx_bool), &bPMETuneRunning);
+ if (bPMETuneRunning &&
+ use_GPU(fr->nbv) && DOMAINDECOMP(cr) &&
+ !(cr->duty & DUTY_PME))
+ {
+ /* Lock DLB=auto to off (does nothing when DLB=yes/no).
+ * With GPUs + separate PME ranks, we don't want DLB.
+ * This could happen when we scan coarse grids and
+ * it would then never be turned off again.
+ * This would hurt performance at the final, optimal
+ * grid spacing, where DLB almost never helps.
+ * Also, DLB can limit the cut-off for PME tuning.
+ */
+ dd_dlb_set_lock(cr->dd, TRUE);
+ }
+
if (bPMETuneRunning || step_rel > ir->nstlist*50)
{
bPMETuneTry = FALSE;
{
calc_enervirdiff(NULL, ir->eDispCorr, fr);
}
+
+ if (!bPMETuneRunning &&
+ DOMAINDECOMP(cr) &&
+ dd_dlb_is_locked(cr->dd))
+ {
+ /* Unlock the DLB=auto, DLB is allowed to activate
+ * (but we don't expect it to activate in most cases).
+ */
+ dd_dlb_set_lock(cr->dd, FALSE);
+ }
}
cycles_pmes = 0;
}
{
/* Reset all the counters related to performance over the run */
reset_all_counters(fplog, cr, step, &step_rel, ir, wcycle, nrnb, walltime_accounting,
- fr->nbv != NULL && fr->nbv->bUseGPU ? fr->nbv->cu_nbv : NULL);
+ use_GPU(fr->nbv) ? fr->nbv : NULL);
wcycle_set_reset_counters(wcycle, -1);
if (!(cr->duty & DUTY_PME))
{
/* End of main MD loop */
debug_gmx();
+ /* Closing TNG files can include compressing data. Therefore it is good to do that
+ * before stopping the time measurements. */
+ mdoutf_tng_close(outf);
+
/* Stop measuring walltime */
walltime_accounting_end(walltime_accounting);
if (pme_loadbal != NULL)
{
pme_loadbal_done(pme_loadbal, cr, fplog,
- fr->nbv != NULL && fr->nbv->bUseGPU);
+ use_GPU(fr->nbv));
}
if (shellfc && fplog)