2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2004, The GROMACS development team.
6 * Copyright (c) 2011-2019,2020, by the GROMACS development team, led by
7 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
8 * and including many others, as listed in the AUTHORS file in the
9 * top-level source directory and at http://www.gromacs.org.
11 * GROMACS is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public License
13 * as published by the Free Software Foundation; either version 2.1
14 * of the License, or (at your option) any later version.
16 * GROMACS is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with GROMACS; if not, see
23 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
24 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
26 * If you want to redistribute modifications to GROMACS, please
27 * consider that scientific software is very special. Version
28 * control is crucial - bugs must be traceable. We will be happy to
29 * consider code for inclusion in the official distribution, but
30 * derived work must not be called official GROMACS. Details are found
31 * in the README & COPYING files - if they are missing, get the
32 * official version at http://www.gromacs.org.
34 * To help us fund GROMACS development, we humbly ask that you cite
35 * the research papers on the package. Check out http://www.gromacs.org.
39 * \brief Implements the integrator for normal molecular dynamics simulations
41 * \author David van der Spoel <david.vanderspoel@icm.uu.se>
42 * \ingroup module_mdrun
55 #include "gromacs/applied_forces/awh/awh.h"
56 #include "gromacs/commandline/filenm.h"
57 #include "gromacs/domdec/collect.h"
58 #include "gromacs/domdec/dlbtiming.h"
59 #include "gromacs/domdec/domdec.h"
60 #include "gromacs/domdec/domdec_network.h"
61 #include "gromacs/domdec/domdec_struct.h"
62 #include "gromacs/domdec/gpuhaloexchange.h"
63 #include "gromacs/domdec/mdsetup.h"
64 #include "gromacs/domdec/partition.h"
65 #include "gromacs/essentialdynamics/edsam.h"
66 #include "gromacs/ewald/pme_load_balancing.h"
67 #include "gromacs/ewald/pme_pp.h"
68 #include "gromacs/fileio/trxio.h"
69 #include "gromacs/gmxlib/network.h"
70 #include "gromacs/gmxlib/nrnb.h"
71 #include "gromacs/gpu_utils/device_stream_manager.h"
72 #include "gromacs/gpu_utils/gpu_utils.h"
73 #include "gromacs/imd/imd.h"
74 #include "gromacs/listed_forces/listed_forces.h"
75 #include "gromacs/math/functions.h"
76 #include "gromacs/math/invertmatrix.h"
77 #include "gromacs/math/vec.h"
78 #include "gromacs/math/vectypes.h"
79 #include "gromacs/mdlib/checkpointhandler.h"
80 #include "gromacs/mdlib/compute_io.h"
81 #include "gromacs/mdlib/constr.h"
82 #include "gromacs/mdlib/coupling.h"
83 #include "gromacs/mdlib/ebin.h"
84 #include "gromacs/mdlib/enerdata_utils.h"
85 #include "gromacs/mdlib/energyoutput.h"
86 #include "gromacs/mdlib/expanded.h"
87 #include "gromacs/mdlib/force.h"
88 #include "gromacs/mdlib/force_flags.h"
89 #include "gromacs/mdlib/forcerec.h"
90 #include "gromacs/mdlib/freeenergyparameters.h"
91 #include "gromacs/mdlib/md_support.h"
92 #include "gromacs/mdlib/mdatoms.h"
93 #include "gromacs/mdlib/mdoutf.h"
94 #include "gromacs/mdlib/membed.h"
95 #include "gromacs/mdlib/resethandler.h"
96 #include "gromacs/mdlib/sighandler.h"
97 #include "gromacs/mdlib/simulationsignal.h"
98 #include "gromacs/mdlib/stat.h"
99 #include "gromacs/mdlib/stophandler.h"
100 #include "gromacs/mdlib/tgroup.h"
101 #include "gromacs/mdlib/trajectory_writing.h"
102 #include "gromacs/mdlib/update.h"
103 #include "gromacs/mdlib/update_constrain_gpu.h"
104 #include "gromacs/mdlib/vcm.h"
105 #include "gromacs/mdlib/vsite.h"
106 #include "gromacs/mdrunutility/handlerestart.h"
107 #include "gromacs/mdrunutility/multisim.h"
108 #include "gromacs/mdrunutility/printtime.h"
109 #include "gromacs/mdtypes/awh_history.h"
110 #include "gromacs/mdtypes/awh_params.h"
111 #include "gromacs/mdtypes/commrec.h"
112 #include "gromacs/mdtypes/df_history.h"
113 #include "gromacs/mdtypes/energyhistory.h"
114 #include "gromacs/mdtypes/fcdata.h"
115 #include "gromacs/mdtypes/forcebuffers.h"
116 #include "gromacs/mdtypes/forcerec.h"
117 #include "gromacs/mdtypes/group.h"
118 #include "gromacs/mdtypes/inputrec.h"
119 #include "gromacs/mdtypes/interaction_const.h"
120 #include "gromacs/mdtypes/md_enums.h"
121 #include "gromacs/mdtypes/mdatom.h"
122 #include "gromacs/mdtypes/mdrunoptions.h"
123 #include "gromacs/mdtypes/multipletimestepping.h"
124 #include "gromacs/mdtypes/observableshistory.h"
125 #include "gromacs/mdtypes/pullhistory.h"
126 #include "gromacs/mdtypes/simulation_workload.h"
127 #include "gromacs/mdtypes/state.h"
128 #include "gromacs/mdtypes/state_propagator_data_gpu.h"
129 #include "gromacs/modularsimulator/energydata.h"
130 #include "gromacs/nbnxm/gpu_data_mgmt.h"
131 #include "gromacs/nbnxm/nbnxm.h"
132 #include "gromacs/pbcutil/pbc.h"
133 #include "gromacs/pulling/output.h"
134 #include "gromacs/pulling/pull.h"
135 #include "gromacs/swap/swapcoords.h"
136 #include "gromacs/timing/wallcycle.h"
137 #include "gromacs/timing/walltime_accounting.h"
138 #include "gromacs/topology/atoms.h"
139 #include "gromacs/topology/idef.h"
140 #include "gromacs/topology/mtop_util.h"
141 #include "gromacs/topology/topology.h"
142 #include "gromacs/trajectory/trajectoryframe.h"
143 #include "gromacs/utility/basedefinitions.h"
144 #include "gromacs/utility/cstringutil.h"
145 #include "gromacs/utility/fatalerror.h"
146 #include "gromacs/utility/logger.h"
147 #include "gromacs/utility/real.h"
148 #include "gromacs/utility/smalloc.h"
150 #include "legacysimulator.h"
151 #include "replicaexchange.h"
155 # include "corewrap.h"
158 using gmx::SimulationSignaller;
160 void gmx::LegacySimulator::do_md()
162 // TODO Historically, the EM and MD "integrators" used different
163 // names for the t_inputrec *parameter, but these must have the
164 // same name, now that it's a member of a struct. We use this ir
165 // alias to avoid a large ripple of nearly useless changes.
166 // t_inputrec is being replaced by IMdpOptionsProvider, so this
167 // will go away eventually.
168 t_inputrec* ir = inputrec;
169 int64_t step, step_rel;
170 double t, t0 = ir->init_t;
171 gmx_bool bGStatEveryStep, bGStat, bCalcVir, bCalcEnerStep, bCalcEner;
172 gmx_bool bNS = FALSE, bNStList, bStopCM, bFirstStep, bInitStep, bLastStep = FALSE;
173 gmx_bool bDoDHDL = FALSE, bDoFEP = FALSE, bDoExpanded = FALSE;
174 gmx_bool do_ene, do_log, do_verbose;
175 gmx_bool bMasterState;
176 unsigned int force_flags;
177 tensor force_vir = { { 0 } }, shake_vir = { { 0 } }, total_vir = { { 0 } }, pres = { { 0 } };
180 matrix pressureCouplingMu, M;
181 gmx_repl_ex_t repl_ex = nullptr;
182 gmx_global_stat_t gstat;
183 gmx_shellfc_t* shellfc;
184 gmx_bool bSumEkinhOld, bDoReplEx, bExchanged, bNeedRepartition;
185 gmx_bool bTemp, bPres, bTrotter;
187 std::vector<RVec> cbuf;
193 real saved_conserved_quantity = 0;
196 char sbuf[STEPSTRSIZE], sbuf2[STEPSTRSIZE];
198 /* PME load balancing data for GPU kernels */
199 gmx_bool bPMETune = FALSE;
200 gmx_bool bPMETunePrinting = FALSE;
202 bool bInteractiveMDstep = false;
204 /* Domain decomposition could incorrectly miss a bonded
205 interaction, but checking for that requires a global
206 communication stage, which does not otherwise happen in DD
207 code. So we do that alongside the first global energy reduction
208 after a new DD is made. These variables handle whether the
209 check happens, and the result it returns. */
210 bool shouldCheckNumberOfBondedInteractions = false;
211 int totalNumberOfBondedInteractions = -1;
213 SimulationSignals signals;
214 // Most global communnication stages don't propagate mdrun
215 // signals, and will use this object to achieve that.
216 SimulationSignaller nullSignaller(nullptr, nullptr, nullptr, false, false);
218 if (!mdrunOptions.writeConfout)
220 // This is on by default, and the main known use case for
221 // turning it off is for convenience in benchmarking, which is
222 // something that should not show up in the general user
227 "The -noconfout functionality is deprecated, and may be removed in a "
231 /* md-vv uses averaged full step velocities for T-control
232 md-vv-avek uses averaged half step velocities for T-control (but full step ekin for P control)
233 md uses averaged half step kinetic energies to determine temperature unless defined otherwise by GMX_EKIN_AVE_VEL; */
234 bTrotter = (EI_VV(ir->eI)
235 && (inputrecNptTrotter(ir) || inputrecNphTrotter(ir) || inputrecNvtTrotter(ir)));
237 const bool bRerunMD = false;
239 int nstglobalcomm = computeGlobalCommunicationPeriod(mdlog, ir, cr);
240 bGStatEveryStep = (nstglobalcomm == 1);
242 const SimulationGroups* groups = &top_global->groups;
244 std::unique_ptr<EssentialDynamics> ed = nullptr;
245 if (opt2bSet("-ei", nfile, fnm))
247 /* Initialize essential dynamics sampling */
248 ed = init_edsam(mdlog, opt2fn_null("-ei", nfile, fnm), opt2fn("-eo", nfile, fnm), top_global,
249 ir, cr, constr, state_global, observablesHistory, oenv, startingBehavior);
251 else if (observablesHistory->edsamHistory)
254 "The checkpoint is from a run with essential dynamics sampling, "
255 "but the current run did not specify the -ei option. "
256 "Either specify the -ei option to mdrun, or do not use this checkpoint file.");
259 int* fep_state = MASTER(cr) ? &state_global->fep_state : nullptr;
260 gmx::ArrayRef<real> lambda = MASTER(cr) ? state_global->lambda : gmx::ArrayRef<real>();
261 initialize_lambdas(fplog, *ir, MASTER(cr), fep_state, lambda);
262 Update upd(*ir, deform);
263 const bool doSimulatedAnnealing = initSimulatedAnnealing(ir, &upd);
264 const bool useReplicaExchange = (replExParams.exchangeInterval > 0);
266 const t_fcdata& fcdata = *fr->fcdata;
268 bool simulationsShareState = false;
269 int nstSignalComm = nstglobalcomm;
271 // TODO This implementation of ensemble orientation restraints is nasty because
272 // a user can't just do multi-sim with single-sim orientation restraints.
273 bool usingEnsembleRestraints =
274 (fcdata.disres->nsystems > 1) || ((ms != nullptr) && (fcdata.orires->nr != 0));
275 bool awhUsesMultiSim = (ir->bDoAwh && ir->awhParams->shareBiasMultisim && (ms != nullptr));
277 // Replica exchange, ensemble restraints and AWH need all
278 // simulations to remain synchronized, so they need
279 // checkpoints and stop conditions to act on the same step, so
280 // the propagation of such signals must take place between
281 // simulations, not just within simulations.
282 // TODO: Make algorithm initializers set these flags.
283 simulationsShareState = useReplicaExchange || usingEnsembleRestraints || awhUsesMultiSim;
285 if (simulationsShareState)
287 // Inter-simulation signal communication does not need to happen
288 // often, so we use a minimum of 200 steps to reduce overhead.
289 const int c_minimumInterSimulationSignallingInterval = 200;
290 nstSignalComm = ((c_minimumInterSimulationSignallingInterval + nstglobalcomm - 1) / nstglobalcomm)
295 if (startingBehavior != StartingBehavior::RestartWithAppending)
297 pleaseCiteCouplingAlgorithms(fplog, *ir);
300 init_mdoutf(fplog, nfile, fnm, mdrunOptions, cr, outputProvider, mdModulesNotifier, ir,
301 top_global, oenv, wcycle, startingBehavior, simulationsShareState, ms);
302 gmx::EnergyOutput energyOutput(mdoutf_get_fp_ene(outf), top_global, ir, pull_work,
303 mdoutf_get_fp_dhdl(outf), false, startingBehavior, mdModulesNotifier);
305 gstat = global_stat_init(ir);
307 /* Check for polarizable models and flexible constraints */
308 shellfc = init_shell_flexcon(fplog, top_global, constr ? constr->numFlexibleConstraints() : 0,
309 ir->nstcalcenergy, DOMAINDECOMP(cr));
312 double io = compute_io(ir, top_global->natoms, *groups, energyOutput.numEnergyTerms(), 1);
313 if ((io > 2000) && MASTER(cr))
315 fprintf(stderr, "\nWARNING: This run will generate roughly %.0f Mb of data\n\n", io);
319 // Local state only becomes valid now.
320 std::unique_ptr<t_state> stateInstance;
324 gmx_localtop_t top(top_global->ffparams);
326 auto mdatoms = mdAtoms->mdatoms();
328 const auto& simulationWork = runScheduleWork->simulationWork;
329 const bool useGpuForPme = simulationWork.useGpuPme;
330 const bool useGpuForNonbonded = simulationWork.useGpuNonbonded;
331 const bool useGpuForBufferOps = simulationWork.useGpuBufferOps;
332 const bool useGpuForUpdate = simulationWork.useGpuUpdate;
334 ForceBuffers f(fr->useMts, ((useGpuForNonbonded && useGpuForBufferOps) || useGpuForUpdate)
335 ? PinningPolicy::PinnedIfSupported
336 : PinningPolicy::CannotBePinned);
337 if (DOMAINDECOMP(cr))
339 stateInstance = std::make_unique<t_state>();
340 state = stateInstance.get();
341 dd_init_local_state(cr->dd, state_global, state);
343 /* Distribute the charge groups over the nodes from the master node */
344 dd_partition_system(fplog, mdlog, ir->init_step, cr, TRUE, 1, state_global, *top_global, ir,
345 imdSession, pull_work, state, &f, mdAtoms, &top, fr, vsite, constr,
346 nrnb, nullptr, FALSE);
347 shouldCheckNumberOfBondedInteractions = true;
348 upd.setNumAtoms(state->natoms);
352 state_change_natoms(state_global, state_global->natoms);
353 /* Copy the pointer to the global state */
354 state = state_global;
356 /* Generate and initialize new topology */
357 mdAlgorithmsSetupAtomData(cr, ir, *top_global, &top, fr, &f, mdAtoms, constr, vsite, shellfc);
359 upd.setNumAtoms(state->natoms);
362 std::unique_ptr<UpdateConstrainGpu> integrator;
364 StatePropagatorDataGpu* stateGpu = fr->stateGpu;
366 // TODO: the assertions below should be handled by UpdateConstraintsBuilder.
369 GMX_RELEASE_ASSERT(!DOMAINDECOMP(cr) || ddUsesUpdateGroups(*cr->dd) || constr == nullptr
370 || constr->numConstraintsTotal() == 0,
371 "Constraints in domain decomposition are only supported with update "
372 "groups if using GPU update.\n");
373 GMX_RELEASE_ASSERT(ir->eConstrAlg != econtSHAKE || constr == nullptr
374 || constr->numConstraintsTotal() == 0,
375 "SHAKE is not supported with GPU update.");
376 GMX_RELEASE_ASSERT(useGpuForPme || (useGpuForNonbonded && simulationWork.useGpuBufferOps),
377 "Either PME or short-ranged non-bonded interaction tasks must run on "
378 "the GPU to use GPU update.\n");
379 GMX_RELEASE_ASSERT(ir->eI == eiMD,
380 "Only the md integrator is supported with the GPU update.\n");
382 ir->etc != etcNOSEHOOVER,
383 "Nose-Hoover temperature coupling is not supported with the GPU update.\n");
385 ir->epc == epcNO || ir->epc == epcPARRINELLORAHMAN || ir->epc == epcBERENDSEN
386 || ir->epc == epcCRESCALE,
387 "Only Parrinello-Rahman, Berendsen, and C-rescale pressure coupling are supported "
388 "with the GPU update.\n");
389 GMX_RELEASE_ASSERT(!mdatoms->haveVsites,
390 "Virtual sites are not supported with the GPU update.\n");
391 GMX_RELEASE_ASSERT(ed == nullptr,
392 "Essential dynamics is not supported with the GPU update.\n");
393 GMX_RELEASE_ASSERT(!ir->bPull || !pull_have_constraint(ir->pull),
394 "Constraints pulling is not supported with the GPU update.\n");
395 GMX_RELEASE_ASSERT(fcdata.orires->nr == 0,
396 "Orientation restraints are not supported with the GPU update.\n");
399 || (!haveFreeEnergyType(*ir, efptBONDED) && !haveFreeEnergyType(*ir, efptMASS)),
400 "Free energy perturbation of masses and constraints are not supported with the GPU "
403 if (constr != nullptr && constr->numConstraintsTotal() > 0)
407 .appendText("Updating coordinates and applying constraints on the GPU.");
411 GMX_LOG(mdlog.info).asParagraph().appendText("Updating coordinates on the GPU.");
413 GMX_RELEASE_ASSERT(fr->deviceStreamManager != nullptr,
414 "Device stream manager should be initialized in order to use GPU "
415 "update-constraints.");
417 fr->deviceStreamManager->streamIsValid(gmx::DeviceStreamType::UpdateAndConstraints),
418 "Update stream should be initialized in order to use GPU "
419 "update-constraints.");
420 integrator = std::make_unique<UpdateConstrainGpu>(
421 *ir, *top_global, fr->deviceStreamManager->context(),
422 fr->deviceStreamManager->stream(gmx::DeviceStreamType::UpdateAndConstraints),
423 stateGpu->xUpdatedOnDevice());
425 integrator->setPbc(PbcType::Xyz, state->box);
428 if (useGpuForPme || (useGpuForNonbonded && useGpuForBufferOps) || useGpuForUpdate)
430 changePinningPolicy(&state->x, PinningPolicy::PinnedIfSupported);
434 changePinningPolicy(&state->v, PinningPolicy::PinnedIfSupported);
437 // NOTE: The global state is no longer used at this point.
438 // But state_global is still used as temporary storage space for writing
439 // the global state to file and potentially for replica exchange.
440 // (Global topology should persist.)
442 update_mdatoms(mdatoms, state->lambda[efptMASS]);
446 /* Check nstexpanded here, because the grompp check was broken */
447 if (ir->expandedvals->nstexpanded % ir->nstcalcenergy != 0)
450 "With expanded ensemble, nstexpanded should be a multiple of nstcalcenergy");
452 init_expanded_ensemble(startingBehavior != StartingBehavior::NewSimulation, ir, state->dfhist);
457 EnergyData::initializeEnergyHistory(startingBehavior, observablesHistory, &energyOutput);
460 preparePrevStepPullCom(ir, pull_work, mdatoms->massT, state, state_global, cr,
461 startingBehavior != StartingBehavior::NewSimulation);
463 // TODO: Remove this by converting AWH into a ForceProvider
464 auto awh = prepareAwhModule(fplog, *ir, state_global, cr, ms,
465 startingBehavior != StartingBehavior::NewSimulation,
466 shellfc != nullptr, opt2fn("-awh", nfile, fnm), pull_work);
468 if (useReplicaExchange && MASTER(cr))
470 repl_ex = init_replica_exchange(fplog, ms, top_global->natoms, ir, replExParams);
472 /* PME tuning is only supported in the Verlet scheme, with PME for
473 * Coulomb. It is not supported with only LJ PME. */
474 bPMETune = (mdrunOptions.tunePme && EEL_PME(fr->ic->eeltype) && !mdrunOptions.reproducible
475 && ir->cutoff_scheme != ecutsGROUP);
477 pme_load_balancing_t* pme_loadbal = nullptr;
480 pme_loadbal_init(&pme_loadbal, cr, mdlog, *ir, state->box, *fr->ic, *fr->nbv, fr->pmedata,
484 if (!ir->bContinuation)
486 if (state->flags & (1U << estV))
488 auto v = makeArrayRef(state->v);
489 /* Set the velocities of vsites, shells and frozen atoms to zero */
490 for (i = 0; i < mdatoms->homenr; i++)
492 if (mdatoms->ptype[i] == eptVSite || mdatoms->ptype[i] == eptShell)
496 else if (mdatoms->cFREEZE)
498 for (m = 0; m < DIM; m++)
500 if (ir->opts.nFreeze[mdatoms->cFREEZE[i]][m])
511 /* Constrain the initial coordinates and velocities */
512 do_constrain_first(fplog, constr, ir, mdatoms->nr, mdatoms->homenr,
513 state->x.arrayRefWithPadding(), state->v.arrayRefWithPadding(),
514 state->box, state->lambda[efptBONDED]);
518 /* Construct the virtual sites for the initial configuration */
519 vsite->construct(state->x, ir->delta_t, {}, state->box);
523 if (ir->efep != efepNO)
525 /* Set free energy calculation frequency as the greatest common
526 * denominator of nstdhdl and repl_ex_nst. */
527 nstfep = ir->fepvals->nstdhdl;
530 nstfep = std::gcd(ir->expandedvals->nstexpanded, nstfep);
532 if (useReplicaExchange)
534 nstfep = std::gcd(replExParams.exchangeInterval, nstfep);
538 nstfep = std::gcd(ir->awhParams->nstSampleCoord, nstfep);
542 /* Be REALLY careful about what flags you set here. You CANNOT assume
543 * this is the first step, since we might be restarting from a checkpoint,
544 * and in that case we should not do any modifications to the state.
546 bStopCM = (ir->comm_mode != ecmNO && !ir->bContinuation);
548 // When restarting from a checkpoint, it can be appropriate to
549 // initialize ekind from quantities in the checkpoint. Otherwise,
550 // compute_globals must initialize ekind before the simulation
551 // starts/restarts. However, only the master rank knows what was
552 // found in the checkpoint file, so we have to communicate in
553 // order to coordinate the restart.
555 // TODO Consider removing this communication if/when checkpoint
556 // reading directly follows .tpr reading, because all ranks can
557 // agree on hasReadEkinState at that time.
558 bool hasReadEkinState = MASTER(cr) ? state_global->ekinstate.hasReadEkinState : false;
561 gmx_bcast(sizeof(hasReadEkinState), &hasReadEkinState, cr->mpi_comm_mygroup);
563 if (hasReadEkinState)
565 restore_ekinstate_from_state(cr, ekind, &state_global->ekinstate);
568 unsigned int cglo_flags =
569 (CGLO_TEMPERATURE | CGLO_GSTAT | (EI_VV(ir->eI) ? CGLO_PRESSURE : 0)
570 | (EI_VV(ir->eI) ? CGLO_CONSTRAINT : 0) | (hasReadEkinState ? CGLO_READEKIN : 0));
572 bSumEkinhOld = FALSE;
574 t_vcm vcm(top_global->groups, *ir);
575 reportComRemovalInfo(fplog, vcm);
577 /* To minimize communication, compute_globals computes the COM velocity
578 * and the kinetic energy for the velocities without COM motion removed.
579 * Thus to get the kinetic energy without the COM contribution, we need
580 * to call compute_globals twice.
582 for (int cgloIteration = 0; cgloIteration < (bStopCM ? 2 : 1); cgloIteration++)
584 unsigned int cglo_flags_iteration = cglo_flags;
585 if (bStopCM && cgloIteration == 0)
587 cglo_flags_iteration |= CGLO_STOPCM;
588 cglo_flags_iteration &= ~CGLO_TEMPERATURE;
590 compute_globals(gstat, cr, ir, fr, ekind, makeConstArrayRef(state->x),
591 makeConstArrayRef(state->v), state->box, mdatoms, nrnb, &vcm, nullptr,
592 enerd, force_vir, shake_vir, total_vir, pres, constr, &nullSignaller,
593 state->box, &totalNumberOfBondedInteractions, &bSumEkinhOld,
595 | (shouldCheckNumberOfBondedInteractions ? CGLO_CHECK_NUMBER_OF_BONDED_INTERACTIONS
597 if (cglo_flags_iteration & CGLO_STOPCM)
599 /* At initialization, do not pass x with acceleration-correction mode
600 * to avoid (incorrect) correction of the initial coordinates.
602 auto x = (vcm.mode == ecmLINEAR_ACCELERATION_CORRECTION) ? ArrayRef<RVec>()
603 : makeArrayRef(state->x);
604 process_and_stopcm_grp(fplog, &vcm, *mdatoms, x, makeArrayRef(state->v));
605 inc_nrnb(nrnb, eNR_STOPCM, mdatoms->homenr);
608 checkNumberOfBondedInteractions(mdlog, cr, totalNumberOfBondedInteractions, top_global, &top,
609 makeConstArrayRef(state->x), state->box,
610 &shouldCheckNumberOfBondedInteractions);
611 if (ir->eI == eiVVAK)
613 /* a second call to get the half step temperature initialized as well */
614 /* we do the same call as above, but turn the pressure off -- internally to
615 compute_globals, this is recognized as a velocity verlet half-step
616 kinetic energy calculation. This minimized excess variables, but
617 perhaps loses some logic?*/
619 compute_globals(gstat, cr, ir, fr, ekind, makeConstArrayRef(state->x),
620 makeConstArrayRef(state->v), state->box, mdatoms, nrnb, &vcm, nullptr,
621 enerd, force_vir, shake_vir, total_vir, pres, constr, &nullSignaller,
622 state->box, nullptr, &bSumEkinhOld, cglo_flags & ~CGLO_PRESSURE);
625 /* Calculate the initial half step temperature, and save the ekinh_old */
626 if (startingBehavior == StartingBehavior::NewSimulation)
628 for (i = 0; (i < ir->opts.ngtc); i++)
630 copy_mat(ekind->tcstat[i].ekinh, ekind->tcstat[i].ekinh_old);
634 /* need to make an initiation call to get the Trotter variables set, as well as other constants
635 for non-trotter temperature control */
636 auto trotter_seq = init_npt_vars(ir, state, &MassQ, bTrotter);
640 if (!ir->bContinuation)
642 if (constr && ir->eConstrAlg == econtLINCS)
644 fprintf(fplog, "RMS relative constraint deviation after constraining: %.2e\n",
647 if (EI_STATE_VELOCITY(ir->eI))
649 real temp = enerd->term[F_TEMP];
652 /* Result of Ekin averaged over velocities of -half
653 * and +half step, while we only have -half step here.
657 fprintf(fplog, "Initial temperature: %g K\n", temp);
662 fprintf(stderr, "starting mdrun '%s'\n", *(top_global->name));
665 sprintf(tbuf, "%8.1f", (ir->init_step + ir->nsteps) * ir->delta_t);
669 sprintf(tbuf, "%s", "infinite");
671 if (ir->init_step > 0)
673 fprintf(stderr, "%s steps, %s ps (continuing from step %s, %8.1f ps).\n",
674 gmx_step_str(ir->init_step + ir->nsteps, sbuf), tbuf,
675 gmx_step_str(ir->init_step, sbuf2), ir->init_step * ir->delta_t);
679 fprintf(stderr, "%s steps, %s ps.\n", gmx_step_str(ir->nsteps, sbuf), tbuf);
681 fprintf(fplog, "\n");
684 walltime_accounting_start_time(walltime_accounting);
685 wallcycle_start(wcycle, ewcRUN);
686 print_start(fplog, cr, walltime_accounting, "mdrun");
689 /* safest point to do file checkpointing is here. More general point would be immediately before integrator call */
690 int chkpt_ret = fcCheckPointParallel(cr->nodeid, NULL, 0);
693 gmx_fatal(3, __FILE__, __LINE__, "Checkpoint error on step %d\n", 0);
697 /***********************************************************
701 ************************************************************/
704 /* Skip the first Nose-Hoover integration when we get the state from tpx */
705 bInitStep = startingBehavior == StartingBehavior::NewSimulation || EI_VV(ir->eI);
706 bSumEkinhOld = FALSE;
708 bNeedRepartition = FALSE;
710 step = ir->init_step;
713 auto stopHandler = stopHandlerBuilder->getStopHandlerMD(
714 compat::not_null<SimulationSignal*>(&signals[eglsSTOPCOND]), simulationsShareState,
715 MASTER(cr), ir->nstlist, mdrunOptions.reproducible, nstSignalComm,
716 mdrunOptions.maximumHoursToRun, ir->nstlist == 0, fplog, step, bNS, walltime_accounting);
718 auto checkpointHandler = std::make_unique<CheckpointHandler>(
719 compat::make_not_null<SimulationSignal*>(&signals[eglsCHKPT]), simulationsShareState,
720 ir->nstlist == 0, MASTER(cr), mdrunOptions.writeConfout,
721 mdrunOptions.checkpointOptions.period);
723 const bool resetCountersIsLocal = true;
724 auto resetHandler = std::make_unique<ResetHandler>(
725 compat::make_not_null<SimulationSignal*>(&signals[eglsRESETCOUNTERS]),
726 !resetCountersIsLocal, ir->nsteps, MASTER(cr), mdrunOptions.timingOptions.resetHalfway,
727 mdrunOptions.maximumHoursToRun, mdlog, wcycle, walltime_accounting);
729 const DDBalanceRegionHandler ddBalanceRegionHandler(cr);
731 if (MASTER(cr) && isMultiSim(ms) && !useReplicaExchange)
733 logInitialMultisimStatus(ms, cr, mdlog, simulationsShareState, ir->nsteps, ir->init_step);
736 /* and stop now if we should */
737 bLastStep = (bLastStep || (ir->nsteps >= 0 && step_rel > ir->nsteps));
741 /* Determine if this is a neighbor search step */
742 bNStList = (ir->nstlist > 0 && step % ir->nstlist == 0);
744 if (bPMETune && bNStList)
746 // This has to be here because PME load balancing is called so early.
747 // TODO: Move to after all booleans are defined.
748 if (useGpuForUpdate && !bFirstStep)
750 stateGpu->copyCoordinatesFromGpu(state->x, AtomLocality::Local);
751 stateGpu->waitCoordinatesReadyOnHost(AtomLocality::Local);
753 /* PME grid + cut-off optimization with GPUs or PME nodes */
754 pme_loadbal_do(pme_loadbal, cr, (mdrunOptions.verbose && MASTER(cr)) ? stderr : nullptr,
755 fplog, mdlog, *ir, fr, state->box, state->x, wcycle, step, step_rel,
756 &bPMETunePrinting, simulationWork.useGpuPmePpCommunication);
759 wallcycle_start(wcycle, ewcSTEP);
761 bLastStep = (step_rel == ir->nsteps);
762 t = t0 + step * ir->delta_t;
764 // TODO Refactor this, so that nstfep does not need a default value of zero
765 if (ir->efep != efepNO || ir->bSimTemp)
767 /* find and set the current lambdas */
768 state->lambda = currentLambdas(step, *(ir->fepvals), state->fep_state);
770 bDoDHDL = do_per_step(step, ir->fepvals->nstdhdl);
771 bDoFEP = ((ir->efep != efepNO) && do_per_step(step, nstfep));
772 bDoExpanded = (do_per_step(step, ir->expandedvals->nstexpanded) && (ir->bExpanded)
776 bDoReplEx = (useReplicaExchange && (step > 0) && !bLastStep
777 && do_per_step(step, replExParams.exchangeInterval));
779 if (doSimulatedAnnealing)
781 update_annealing_target_temp(ir, t, &upd);
784 /* Stop Center of Mass motion */
785 bStopCM = (ir->comm_mode != ecmNO && do_per_step(step, ir->nstcomm));
787 /* Determine whether or not to do Neighbour Searching */
788 bNS = (bFirstStep || bNStList || bExchanged || bNeedRepartition);
790 /* Note that the stopHandler will cause termination at nstglobalcomm
791 * steps. Since this concides with nstcalcenergy, nsttcouple and/or
792 * nstpcouple steps, we have computed the half-step kinetic energy
793 * of the previous step and can always output energies at the last step.
795 bLastStep = bLastStep || stopHandler->stoppingAfterCurrentStep(bNS);
797 /* do_log triggers energy and virial calculation. Because this leads
798 * to different code paths, forces can be different. Thus for exact
799 * continuation we should avoid extra log output.
800 * Note that the || bLastStep can result in non-exact continuation
801 * beyond the last step. But we don't consider that to be an issue.
803 do_log = (do_per_step(step, ir->nstlog)
804 || (bFirstStep && startingBehavior == StartingBehavior::NewSimulation) || bLastStep);
805 do_verbose = mdrunOptions.verbose
806 && (step % mdrunOptions.verboseStepPrintInterval == 0 || bFirstStep || bLastStep);
808 if (useGpuForUpdate && !bFirstStep && bNS)
810 // Copy velocities from the GPU on search steps to keep a copy on host (device buffers are reinitialized).
811 stateGpu->copyVelocitiesFromGpu(state->v, AtomLocality::Local);
812 stateGpu->waitVelocitiesReadyOnHost(AtomLocality::Local);
813 // Copy coordinate from the GPU when needed at the search step.
814 // NOTE: The cases when coordinates needed on CPU for force evaluation are handled in sim_utils.
815 // NOTE: If the coordinates are to be written into output file they are also copied separately before the output.
816 stateGpu->copyCoordinatesFromGpu(state->x, AtomLocality::Local);
817 stateGpu->waitCoordinatesReadyOnHost(AtomLocality::Local);
820 if (bNS && !(bFirstStep && ir->bContinuation))
822 bMasterState = FALSE;
823 /* Correct the new box if it is too skewed */
824 if (inputrecDynamicBox(ir))
826 if (correct_box(fplog, step, state->box))
829 // If update is offloaded, it should be informed about the box size change
832 integrator->setPbc(PbcType::Xyz, state->box);
836 if (DOMAINDECOMP(cr) && bMasterState)
838 dd_collect_state(cr->dd, state, state_global);
841 if (DOMAINDECOMP(cr))
843 /* Repartition the domain decomposition */
844 dd_partition_system(fplog, mdlog, step, cr, bMasterState, nstglobalcomm, state_global,
845 *top_global, ir, imdSession, pull_work, state, &f, mdAtoms, &top,
846 fr, vsite, constr, nrnb, wcycle, do_verbose && !bPMETunePrinting);
847 shouldCheckNumberOfBondedInteractions = true;
848 upd.setNumAtoms(state->natoms);
852 // Allocate or re-size GPU halo exchange object, if necessary
853 if (bNS && havePPDomainDecomposition(cr) && simulationWork.useGpuHaloExchange && useGpuForNonbonded)
855 GMX_RELEASE_ASSERT(fr->deviceStreamManager != nullptr,
856 "GPU device manager has to be initialized to use GPU "
857 "version of halo exchange.");
858 constructGpuHaloExchange(mdlog, *cr, *fr->deviceStreamManager, wcycle);
861 if (MASTER(cr) && do_log)
863 gmx::EnergyOutput::printHeader(fplog, step,
864 t); /* can we improve the information printed here? */
867 if (ir->efep != efepNO)
869 update_mdatoms(mdatoms, state->lambda[efptMASS]);
875 /* We need the kinetic energy at minus the half step for determining
876 * the full step kinetic energy and possibly for T-coupling.*/
877 /* This may not be quite working correctly yet . . . . */
878 compute_globals(gstat, cr, ir, fr, ekind, makeConstArrayRef(state->x),
879 makeConstArrayRef(state->v), state->box, mdatoms, nrnb, &vcm, wcycle,
880 enerd, nullptr, nullptr, nullptr, nullptr, constr, &nullSignaller,
881 state->box, &totalNumberOfBondedInteractions, &bSumEkinhOld,
882 CGLO_GSTAT | CGLO_TEMPERATURE | CGLO_CHECK_NUMBER_OF_BONDED_INTERACTIONS);
883 checkNumberOfBondedInteractions(mdlog, cr, totalNumberOfBondedInteractions, top_global,
884 &top, makeConstArrayRef(state->x), state->box,
885 &shouldCheckNumberOfBondedInteractions);
887 clear_mat(force_vir);
889 checkpointHandler->decideIfCheckpointingThisStep(bNS, bFirstStep, bLastStep);
891 /* Determine the energy and pressure:
892 * at nstcalcenergy steps and at energy output steps (set below).
894 if (EI_VV(ir->eI) && (!bInitStep))
896 bCalcEnerStep = do_per_step(step, ir->nstcalcenergy);
897 bCalcVir = bCalcEnerStep
899 && (do_per_step(step, ir->nstpcouple) || do_per_step(step - 1, ir->nstpcouple)));
903 bCalcEnerStep = do_per_step(step, ir->nstcalcenergy);
904 bCalcVir = bCalcEnerStep || (ir->epc != epcNO && do_per_step(step, ir->nstpcouple));
906 bCalcEner = bCalcEnerStep;
908 do_ene = (do_per_step(step, ir->nstenergy) || bLastStep);
910 if (do_ene || do_log || bDoReplEx)
916 /* Do we need global communication ? */
917 bGStat = (bCalcVir || bCalcEner || bStopCM || do_per_step(step, nstglobalcomm)
918 || (EI_VV(ir->eI) && inputrecNvtTrotter(ir) && do_per_step(step - 1, nstglobalcomm)));
920 force_flags = (GMX_FORCE_STATECHANGED | ((inputrecDynamicBox(ir)) ? GMX_FORCE_DYNAMICBOX : 0)
921 | GMX_FORCE_ALLFORCES | (bCalcVir ? GMX_FORCE_VIRIAL : 0)
922 | (bCalcEner ? GMX_FORCE_ENERGY : 0) | (bDoFEP ? GMX_FORCE_DHDL : 0));
923 if (fr->useMts && !do_per_step(step, ir->nstfout))
925 force_flags |= GMX_FORCE_DO_NOT_NEED_NORMAL_FORCE;
930 /* Now is the time to relax the shells */
931 relax_shell_flexcon(fplog, cr, ms, mdrunOptions.verbose, enforcedRotation, step, ir,
932 imdSession, pull_work, bNS, force_flags, &top, constr, enerd,
933 state->natoms, state->x.arrayRefWithPadding(),
934 state->v.arrayRefWithPadding(), state->box, state->lambda,
935 &state->hist, &f.view(), force_vir, mdatoms, nrnb, wcycle, shellfc,
936 fr, runScheduleWork, t, mu_tot, vsite, ddBalanceRegionHandler);
940 /* The AWH history need to be saved _before_ doing force calculations where the AWH bias
941 is updated (or the AWH update will be performed twice for one step when continuing).
942 It would be best to call this update function from do_md_trajectory_writing but that
943 would occur after do_force. One would have to divide the update_awh function into one
944 function applying the AWH force and one doing the AWH bias update. The update AWH
945 bias function could then be called after do_md_trajectory_writing (then containing
946 update_awh_history). The checkpointing will in the future probably moved to the start
947 of the md loop which will rid of this issue. */
948 if (awh && checkpointHandler->isCheckpointingStep() && MASTER(cr))
950 awh->updateHistory(state_global->awhHistory.get());
953 /* The coordinates (x) are shifted (to get whole molecules)
955 * This is parallellized as well, and does communication too.
956 * Check comments in sim_util.c
958 do_force(fplog, cr, ms, ir, awh.get(), enforcedRotation, imdSession, pull_work, step,
959 nrnb, wcycle, &top, state->box, state->x.arrayRefWithPadding(), &state->hist,
960 &f.view(), force_vir, mdatoms, enerd, state->lambda, fr, runScheduleWork,
961 vsite, mu_tot, t, ed ? ed->getLegacyED() : nullptr,
962 (bNS ? GMX_FORCE_NS : 0) | force_flags, ddBalanceRegionHandler);
965 // VV integrators do not need the following velocity half step
966 // if it is the first step after starting from a checkpoint.
967 // That is, the half step is needed on all other steps, and
968 // also the first step when starting from a .tpr file.
969 if (EI_VV(ir->eI) && (!bFirstStep || startingBehavior == StartingBehavior::NewSimulation))
970 /* ############### START FIRST UPDATE HALF-STEP FOR VV METHODS############### */
972 rvec* vbuf = nullptr;
974 wallcycle_start(wcycle, ewcUPDATE);
975 if (ir->eI == eiVV && bInitStep)
977 /* if using velocity verlet with full time step Ekin,
978 * take the first half step only to compute the
979 * virial for the first step. From there,
980 * revert back to the initial coordinates
981 * so that the input is actually the initial step.
983 snew(vbuf, state->natoms);
984 copy_rvecn(state->v.rvec_array(), vbuf, 0,
985 state->natoms); /* should make this better for parallelizing? */
989 /* this is for NHC in the Ekin(t+dt/2) version of vv */
990 trotter_update(ir, step, ekind, enerd, state, total_vir, mdatoms, &MassQ,
991 trotter_seq, ettTSEQ1);
994 upd.update_coords(*ir, step, mdatoms, state, f.view().forceWithPadding(), fcdata, ekind,
995 M, etrtVELOCITY1, cr, constr != nullptr);
997 wallcycle_stop(wcycle, ewcUPDATE);
998 constrain_velocities(constr, do_log, do_ene, step, state, nullptr, bCalcVir, shake_vir);
999 wallcycle_start(wcycle, ewcUPDATE);
1000 /* if VV, compute the pressure and constraints */
1001 /* For VV2, we strictly only need this if using pressure
1002 * control, but we really would like to have accurate pressures
1004 * Think about ways around this in the future?
1005 * For now, keep this choice in comments.
1007 /*bPres = (ir->eI==eiVV || inputrecNptTrotter(ir)); */
1008 /*bTemp = ((ir->eI==eiVV &&(!bInitStep)) || (ir->eI==eiVVAK && inputrecNptTrotter(ir)));*/
1010 bTemp = ((ir->eI == eiVV && (!bInitStep)) || (ir->eI == eiVVAK));
1011 if (bCalcEner && ir->eI == eiVVAK)
1013 bSumEkinhOld = TRUE;
1015 /* for vv, the first half of the integration actually corresponds to the previous step.
1016 So we need information from the last step in the first half of the integration */
1017 if (bGStat || do_per_step(step - 1, nstglobalcomm))
1019 wallcycle_stop(wcycle, ewcUPDATE);
1020 compute_globals(gstat, cr, ir, fr, ekind, makeConstArrayRef(state->x),
1021 makeConstArrayRef(state->v), state->box, mdatoms, nrnb, &vcm, wcycle,
1022 enerd, force_vir, shake_vir, total_vir, pres, constr, &nullSignaller,
1023 state->box, &totalNumberOfBondedInteractions, &bSumEkinhOld,
1024 (bGStat ? CGLO_GSTAT : 0) | (bCalcEner ? CGLO_ENERGY : 0)
1025 | (bTemp ? CGLO_TEMPERATURE : 0) | (bPres ? CGLO_PRESSURE : 0)
1026 | (bPres ? CGLO_CONSTRAINT : 0) | (bStopCM ? CGLO_STOPCM : 0)
1027 | (shouldCheckNumberOfBondedInteractions ? CGLO_CHECK_NUMBER_OF_BONDED_INTERACTIONS
1030 /* explanation of above:
1031 a) We compute Ekin at the full time step
1032 if 1) we are using the AveVel Ekin, and it's not the
1033 initial step, or 2) if we are using AveEkin, but need the full
1034 time step kinetic energy for the pressure (always true now, since we want accurate statistics).
1035 b) If we are using EkinAveEkin for the kinetic energy for the temperature control, we still feed in
1036 EkinAveVel because it's needed for the pressure */
1037 checkNumberOfBondedInteractions(mdlog, cr, totalNumberOfBondedInteractions,
1038 top_global, &top, makeConstArrayRef(state->x),
1039 state->box, &shouldCheckNumberOfBondedInteractions);
1042 process_and_stopcm_grp(fplog, &vcm, *mdatoms, makeArrayRef(state->x),
1043 makeArrayRef(state->v));
1044 inc_nrnb(nrnb, eNR_STOPCM, mdatoms->homenr);
1046 wallcycle_start(wcycle, ewcUPDATE);
1048 /* temperature scaling and pressure scaling to produce the extended variables at t+dt */
1053 m_add(force_vir, shake_vir,
1054 total_vir); /* we need the un-dispersion corrected total vir here */
1055 trotter_update(ir, step, ekind, enerd, state, total_vir, mdatoms, &MassQ,
1056 trotter_seq, ettTSEQ2);
1058 /* TODO This is only needed when we're about to write
1059 * a checkpoint, because we use it after the restart
1060 * (in a kludge?). But what should we be doing if
1061 * the startingBehavior is NewSimulation or bInitStep are true? */
1062 if (inputrecNptTrotter(ir) || inputrecNphTrotter(ir))
1064 copy_mat(shake_vir, state->svir_prev);
1065 copy_mat(force_vir, state->fvir_prev);
1067 if (inputrecNvtTrotter(ir) && ir->eI == eiVV)
1069 /* update temperature and kinetic energy now that step is over - this is the v(t+dt) point */
1070 enerd->term[F_TEMP] =
1071 sum_ekin(&(ir->opts), ekind, nullptr, (ir->eI == eiVV), FALSE);
1072 enerd->term[F_EKIN] = trace(ekind->ekin);
1075 else if (bExchanged)
1077 wallcycle_stop(wcycle, ewcUPDATE);
1078 /* We need the kinetic energy at minus the half step for determining
1079 * the full step kinetic energy and possibly for T-coupling.*/
1080 /* This may not be quite working correctly yet . . . . */
1081 compute_globals(gstat, cr, ir, fr, ekind, makeConstArrayRef(state->x),
1082 makeConstArrayRef(state->v), state->box, mdatoms, nrnb, &vcm, wcycle,
1083 enerd, nullptr, nullptr, nullptr, nullptr, constr, &nullSignaller,
1084 state->box, nullptr, &bSumEkinhOld, CGLO_GSTAT | CGLO_TEMPERATURE);
1085 wallcycle_start(wcycle, ewcUPDATE);
1088 /* if it's the initial step, we performed this first step just to get the constraint virial */
1089 if (ir->eI == eiVV && bInitStep)
1091 copy_rvecn(vbuf, state->v.rvec_array(), 0, state->natoms);
1094 wallcycle_stop(wcycle, ewcUPDATE);
1097 /* compute the conserved quantity */
1100 saved_conserved_quantity = NPT_energy(ir, state, &MassQ);
1103 last_ekin = enerd->term[F_EKIN];
1105 if ((ir->eDispCorr != edispcEnerPres) && (ir->eDispCorr != edispcAllEnerPres))
1107 saved_conserved_quantity -= enerd->term[F_DISPCORR];
1109 /* sum up the foreign kinetic energy and dK/dl terms for vv. currently done every step so that dhdl is correct in the .edr */
1110 if (ir->efep != efepNO)
1112 accumulateKineticLambdaComponents(enerd, state->lambda, *ir->fepvals);
1116 /* ######## END FIRST UPDATE STEP ############## */
1117 /* ######## If doing VV, we now have v(dt) ###### */
1120 /* perform extended ensemble sampling in lambda - we don't
1121 actually move to the new state before outputting
1122 statistics, but if performing simulated tempering, we
1123 do update the velocities and the tau_t. */
1125 lamnew = ExpandedEnsembleDynamics(fplog, ir, enerd, state, &MassQ, state->fep_state,
1126 state->dfhist, step, state->v.rvec_array(), mdatoms);
1127 /* history is maintained in state->dfhist, but state_global is what is sent to trajectory and log output */
1130 copy_df_history(state_global->dfhist, state->dfhist);
1134 // Copy coordinate from the GPU for the output/checkpointing if the update is offloaded and
1135 // coordinates have not already been copied for i) search or ii) CPU force tasks.
1136 if (useGpuForUpdate && !bNS && !runScheduleWork->domainWork.haveCpuLocalForceWork
1137 && (do_per_step(step, ir->nstxout) || do_per_step(step, ir->nstxout_compressed)
1138 || checkpointHandler->isCheckpointingStep()))
1140 stateGpu->copyCoordinatesFromGpu(state->x, AtomLocality::Local);
1141 stateGpu->waitCoordinatesReadyOnHost(AtomLocality::Local);
1143 // Copy velocities if needed for the output/checkpointing.
1144 // NOTE: Copy on the search steps is done at the beginning of the step.
1145 if (useGpuForUpdate && !bNS
1146 && (do_per_step(step, ir->nstvout) || checkpointHandler->isCheckpointingStep()))
1148 stateGpu->copyVelocitiesFromGpu(state->v, AtomLocality::Local);
1149 stateGpu->waitVelocitiesReadyOnHost(AtomLocality::Local);
1151 // Copy forces for the output if the forces were reduced on the GPU (not the case on virial steps)
1152 // and update is offloaded hence forces are kept on the GPU for update and have not been
1153 // already transferred in do_force().
1154 // TODO: There should be an improved, explicit mechanism that ensures this copy is only executed
1155 // when the forces are ready on the GPU -- the same synchronizer should be used as the one
1156 // prior to GPU update.
1157 // TODO: When the output flags will be included in step workload, this copy can be combined with the
1158 // copy call in do_force(...).
1159 // NOTE: The forces should not be copied here if the vsites are present, since they were modified
1160 // on host after the D2H copy in do_force(...).
1161 if (runScheduleWork->stepWork.useGpuFBufferOps && (simulationWork.useGpuUpdate && !vsite)
1162 && do_per_step(step, ir->nstfout))
1164 stateGpu->copyForcesFromGpu(f.view().force(), AtomLocality::Local);
1165 stateGpu->waitForcesReadyOnHost(AtomLocality::Local);
1167 /* Now we have the energies and forces corresponding to the
1168 * coordinates at time t. We must output all of this before
1171 do_md_trajectory_writing(fplog, cr, nfile, fnm, step, step_rel, t, ir, state, state_global,
1172 observablesHistory, top_global, fr, outf, energyOutput, ekind,
1173 f.view().force(), checkpointHandler->isCheckpointingStep(),
1174 bRerunMD, bLastStep, mdrunOptions.writeConfout, bSumEkinhOld);
1175 /* Check if IMD step and do IMD communication, if bIMD is TRUE. */
1176 bInteractiveMDstep = imdSession->run(step, bNS, state->box, state->x.rvec_array(), t);
1178 /* kludge -- virial is lost with restart for MTTK NPT control. Must reload (saved earlier). */
1179 if (startingBehavior != StartingBehavior::NewSimulation && bFirstStep
1180 && (inputrecNptTrotter(ir) || inputrecNphTrotter(ir)))
1182 copy_mat(state->svir_prev, shake_vir);
1183 copy_mat(state->fvir_prev, force_vir);
1186 stopHandler->setSignal();
1187 resetHandler->setSignal(walltime_accounting);
1189 if (bGStat || !PAR(cr))
1191 /* In parallel we only have to check for checkpointing in steps
1192 * where we do global communication,
1193 * otherwise the other nodes don't know.
1195 checkpointHandler->setSignal(walltime_accounting);
1198 /* ######### START SECOND UPDATE STEP ################# */
1200 /* at the start of step, randomize or scale the velocities ((if vv. Restriction of Andersen
1201 controlled in preprocessing */
1203 if (ETC_ANDERSEN(ir->etc)) /* keep this outside of update_tcouple because of the extra info required to pass */
1205 gmx_bool bIfRandomize;
1206 bIfRandomize = update_randomize_velocities(ir, step, cr, mdatoms, state->v, &upd, constr);
1207 /* if we have constraints, we have to remove the kinetic energy parallel to the bonds */
1208 if (constr && bIfRandomize)
1210 constrain_velocities(constr, do_log, do_ene, step, state, nullptr, false, nullptr);
1213 /* Box is changed in update() when we do pressure coupling,
1214 * but we should still use the old box for energy corrections and when
1215 * writing it to the energy file, so it matches the trajectory files for
1216 * the same timestep above. Make a copy in a separate array.
1218 copy_mat(state->box, lastbox);
1222 wallcycle_start(wcycle, ewcUPDATE);
1223 /* UPDATE PRESSURE VARIABLES IN TROTTER FORMULATION WITH CONSTRAINTS */
1226 trotter_update(ir, step, ekind, enerd, state, total_vir, mdatoms, &MassQ, trotter_seq, ettTSEQ3);
1227 /* We can only do Berendsen coupling after we have summed
1228 * the kinetic energy or virial. Since the happens
1229 * in global_state after update, we should only do it at
1230 * step % nstlist = 1 with bGStatEveryStep=FALSE.
1235 update_tcouple(step, ir, state, ekind, &MassQ, mdatoms);
1236 update_pcouple_before_coordinates(fplog, step, ir, state, pressureCouplingMu, M, bInitStep);
1241 /* velocity half-step update */
1242 upd.update_coords(*ir, step, mdatoms, state, f.view().forceWithPadding(), fcdata, ekind,
1243 M, etrtVELOCITY2, cr, constr != nullptr);
1246 /* Above, initialize just copies ekinh into ekin,
1247 * it doesn't copy position (for VV),
1248 * and entire integrator for MD.
1251 if (ir->eI == eiVVAK)
1253 cbuf.resize(state->x.size());
1254 std::copy(state->x.begin(), state->x.end(), cbuf.begin());
1257 /* With leap-frog type integrators we compute the kinetic energy
1258 * at a whole time step as the average of the half-time step kinetic
1259 * energies of two subsequent steps. Therefore we need to compute the
1260 * half step kinetic energy also if we need energies at the next step.
1262 const bool needHalfStepKineticEnergy =
1263 (!EI_VV(ir->eI) && (do_per_step(step + 1, nstglobalcomm) || step_rel + 1 == ir->nsteps));
1265 // Parrinello-Rahman requires the pressure to be availible before the update to compute
1266 // the velocity scaling matrix. Hence, it runs one step after the nstpcouple step.
1267 const bool doParrinelloRahman = (ir->epc == epcPARRINELLORAHMAN
1268 && do_per_step(step + ir->nstpcouple - 1, ir->nstpcouple));
1270 if (useGpuForUpdate)
1272 if (bNS && (bFirstStep || DOMAINDECOMP(cr)))
1274 integrator->set(stateGpu->getCoordinates(), stateGpu->getVelocities(),
1275 stateGpu->getForces(), top.idef, *mdatoms, ekind->ngtc);
1277 // Copy data to the GPU after buffers might have being reinitialized
1278 stateGpu->copyVelocitiesToGpu(state->v, AtomLocality::Local);
1279 stateGpu->copyCoordinatesToGpu(state->x, AtomLocality::Local);
1282 // If the buffer ops were not offloaded this step, the forces are on the host and have to be copied
1283 if (!runScheduleWork->stepWork.useGpuFBufferOps)
1285 stateGpu->copyForcesToGpu(f.view().force(), AtomLocality::Local);
1288 const bool doTemperatureScaling =
1289 (ir->etc != etcNO && do_per_step(step + ir->nsttcouple - 1, ir->nsttcouple));
1291 // This applies Leap-Frog, LINCS and SETTLE in succession
1292 integrator->integrate(stateGpu->getForcesReadyOnDeviceEvent(
1293 AtomLocality::Local, runScheduleWork->stepWork.useGpuFBufferOps),
1294 ir->delta_t, true, bCalcVir, shake_vir, doTemperatureScaling,
1295 ekind->tcstat, doParrinelloRahman, ir->nstpcouple * ir->delta_t, M);
1297 // Copy velocities D2H after update if:
1298 // - Globals are computed this step (includes the energy output steps).
1299 // - Temperature is needed for the next step.
1300 if (bGStat || needHalfStepKineticEnergy)
1302 stateGpu->copyVelocitiesFromGpu(state->v, AtomLocality::Local);
1303 stateGpu->waitVelocitiesReadyOnHost(AtomLocality::Local);
1308 /* With multiple time stepping we need to do an additional normal
1309 * update step to obtain the virial, as the actual MTS integration
1310 * using an acceleration where the slow forces are multiplied by mtsFactor.
1311 * Using that acceleration would result in a virial with the slow
1312 * force contribution would be a factor mtsFactor too large.
1314 if (fr->useMts && bCalcVir && constr != nullptr)
1316 upd.update_for_constraint_virial(*ir, *mdatoms, *state, f.view().forceWithPadding(), *ekind);
1318 constrain_coordinates(constr, do_log, do_ene, step, state,
1319 upd.xp()->arrayRefWithPadding(), &dvdl_constr, bCalcVir, shake_vir);
1322 ArrayRefWithPadding<const RVec> forceCombined =
1323 (fr->useMts && step % ir->mtsLevels[1].stepFactor == 0)
1324 ? f.view().forceMtsCombinedWithPadding()
1325 : f.view().forceWithPadding();
1326 upd.update_coords(*ir, step, mdatoms, state, forceCombined, fcdata, ekind, M,
1327 etrtPOSITION, cr, constr != nullptr);
1329 wallcycle_stop(wcycle, ewcUPDATE);
1331 constrain_coordinates(constr, do_log, do_ene, step, state, upd.xp()->arrayRefWithPadding(),
1332 &dvdl_constr, bCalcVir && !fr->useMts, shake_vir);
1334 upd.update_sd_second_half(*ir, step, &dvdl_constr, mdatoms, state, cr, nrnb, wcycle,
1335 constr, do_log, do_ene);
1336 upd.finish_update(*ir, mdatoms, state, wcycle, constr != nullptr);
1339 if (ir->bPull && ir->pull->bSetPbcRefToPrevStepCOM)
1341 updatePrevStepPullCom(pull_work, state);
1344 if (ir->eI == eiVVAK)
1346 /* erase F_EKIN and F_TEMP here? */
1347 /* just compute the kinetic energy at the half step to perform a trotter step */
1348 compute_globals(gstat, cr, ir, fr, ekind, makeConstArrayRef(state->x),
1349 makeConstArrayRef(state->v), state->box, mdatoms, nrnb, &vcm, wcycle, enerd,
1350 force_vir, shake_vir, total_vir, pres, constr, &nullSignaller, lastbox,
1351 nullptr, &bSumEkinhOld, (bGStat ? CGLO_GSTAT : 0) | CGLO_TEMPERATURE);
1352 wallcycle_start(wcycle, ewcUPDATE);
1353 trotter_update(ir, step, ekind, enerd, state, total_vir, mdatoms, &MassQ, trotter_seq, ettTSEQ4);
1354 /* now we know the scaling, we can compute the positions again */
1355 std::copy(cbuf.begin(), cbuf.end(), state->x.begin());
1357 upd.update_coords(*ir, step, mdatoms, state, f.view().forceWithPadding(), fcdata, ekind,
1358 M, etrtPOSITION, cr, constr != nullptr);
1359 wallcycle_stop(wcycle, ewcUPDATE);
1361 /* do we need an extra constraint here? just need to copy out of as_rvec_array(state->v.data()) to upd->xp? */
1362 /* are the small terms in the shake_vir here due
1363 * to numerical errors, or are they important
1364 * physically? I'm thinking they are just errors, but not completely sure.
1365 * For now, will call without actually constraining, constr=NULL*/
1366 upd.finish_update(*ir, mdatoms, state, wcycle, false);
1370 /* this factor or 2 correction is necessary
1371 because half of the constraint force is removed
1372 in the vv step, so we have to double it. See
1373 the Issue #1255. It is not yet clear
1374 if the factor of 2 is exact, or just a very
1375 good approximation, and this will be
1376 investigated. The next step is to see if this
1377 can be done adding a dhdl contribution from the
1378 rattle step, but this is somewhat more
1379 complicated with the current code. Will be
1380 investigated, hopefully for 4.6.3. However,
1381 this current solution is much better than
1382 having it completely wrong.
1384 enerd->term[F_DVDL_CONSTR] += 2 * dvdl_constr;
1388 enerd->term[F_DVDL_CONSTR] += dvdl_constr;
1391 if (vsite != nullptr)
1393 wallcycle_start(wcycle, ewcVSITECONSTR);
1394 vsite->construct(state->x, ir->delta_t, state->v, state->box);
1395 wallcycle_stop(wcycle, ewcVSITECONSTR);
1398 /* ############## IF NOT VV, Calculate globals HERE ############ */
1399 /* With Leap-Frog we can skip compute_globals at
1400 * non-communication steps, but we need to calculate
1401 * the kinetic energy one step before communication.
1404 // Organize to do inter-simulation signalling on steps if
1405 // and when algorithms require it.
1406 const bool doInterSimSignal = (simulationsShareState && do_per_step(step, nstSignalComm));
1408 if (bGStat || needHalfStepKineticEnergy || doInterSimSignal)
1410 // Copy coordinates when needed to stop the CM motion.
1411 if (useGpuForUpdate && !EI_VV(ir->eI) && bStopCM)
1413 stateGpu->copyCoordinatesFromGpu(state->x, AtomLocality::Local);
1414 stateGpu->waitCoordinatesReadyOnHost(AtomLocality::Local);
1416 // Since we're already communicating at this step, we
1417 // can propagate intra-simulation signals. Note that
1418 // check_nstglobalcomm has the responsibility for
1419 // choosing the value of nstglobalcomm that is one way
1420 // bGStat becomes true, so we can't get into a
1421 // situation where e.g. checkpointing can't be
1423 bool doIntraSimSignal = true;
1424 SimulationSignaller signaller(&signals, cr, ms, doInterSimSignal, doIntraSimSignal);
1426 compute_globals(gstat, cr, ir, fr, ekind, makeConstArrayRef(state->x),
1427 makeConstArrayRef(state->v), state->box, mdatoms, nrnb, &vcm,
1428 wcycle, enerd, force_vir, shake_vir, total_vir, pres, constr,
1429 &signaller, lastbox, &totalNumberOfBondedInteractions, &bSumEkinhOld,
1430 (bGStat ? CGLO_GSTAT : 0) | (!EI_VV(ir->eI) && bCalcEner ? CGLO_ENERGY : 0)
1431 | (!EI_VV(ir->eI) && bStopCM ? CGLO_STOPCM : 0)
1432 | (!EI_VV(ir->eI) ? CGLO_TEMPERATURE : 0)
1433 | (!EI_VV(ir->eI) ? CGLO_PRESSURE : 0) | CGLO_CONSTRAINT
1434 | (shouldCheckNumberOfBondedInteractions ? CGLO_CHECK_NUMBER_OF_BONDED_INTERACTIONS
1436 checkNumberOfBondedInteractions(mdlog, cr, totalNumberOfBondedInteractions,
1437 top_global, &top, makeConstArrayRef(state->x),
1438 state->box, &shouldCheckNumberOfBondedInteractions);
1439 if (!EI_VV(ir->eI) && bStopCM)
1441 process_and_stopcm_grp(fplog, &vcm, *mdatoms, makeArrayRef(state->x),
1442 makeArrayRef(state->v));
1443 inc_nrnb(nrnb, eNR_STOPCM, mdatoms->homenr);
1445 // TODO: The special case of removing CM motion should be dealt more gracefully
1446 if (useGpuForUpdate)
1448 stateGpu->copyCoordinatesToGpu(state->x, AtomLocality::Local);
1449 // Here we block until the H2D copy completes because event sync with the
1450 // force kernels that use the coordinates on the next steps is not implemented
1451 // (not because of a race on state->x being modified on the CPU while H2D is in progress).
1452 stateGpu->waitCoordinatesCopiedToDevice(AtomLocality::Local);
1453 // If the COM removal changed the velocities on the CPU, this has to be accounted for.
1454 if (vcm.mode != ecmNO)
1456 stateGpu->copyVelocitiesToGpu(state->v, AtomLocality::Local);
1463 /* ############# END CALC EKIN AND PRESSURE ################# */
1465 /* Note: this is OK, but there are some numerical precision issues with using the convergence of
1466 the virial that should probably be addressed eventually. state->veta has better properies,
1467 but what we actually need entering the new cycle is the new shake_vir value. Ideally, we could
1468 generate the new shake_vir, but test the veta value for convergence. This will take some thought. */
1470 if (ir->efep != efepNO && !EI_VV(ir->eI))
1472 /* Sum up the foreign energy and dK/dl terms for md and sd.
1473 Currently done every step so that dH/dl is correct in the .edr */
1474 accumulateKineticLambdaComponents(enerd, state->lambda, *ir->fepvals);
1477 update_pcouple_after_coordinates(fplog, step, ir, mdatoms, pres, force_vir, shake_vir,
1478 pressureCouplingMu, state, nrnb, upd.deform(), !useGpuForUpdate);
1480 const bool doBerendsenPressureCoupling =
1481 (inputrec->epc == epcBERENDSEN && do_per_step(step, inputrec->nstpcouple));
1482 const bool doCRescalePressureCoupling =
1483 (inputrec->epc == epcCRESCALE && do_per_step(step, inputrec->nstpcouple));
1485 && (doBerendsenPressureCoupling || doCRescalePressureCoupling || doParrinelloRahman))
1487 integrator->scaleCoordinates(pressureCouplingMu);
1488 if (doCRescalePressureCoupling)
1490 matrix pressureCouplingInvMu;
1491 gmx::invertBoxMatrix(pressureCouplingMu, pressureCouplingInvMu);
1492 integrator->scaleVelocities(pressureCouplingInvMu);
1494 integrator->setPbc(PbcType::Xyz, state->box);
1497 /* ################# END UPDATE STEP 2 ################# */
1498 /* #### We now have r(t+dt) and v(t+dt/2) ############# */
1500 /* The coordinates (x) were unshifted in update */
1503 /* We will not sum ekinh_old,
1504 * so signal that we still have to do it.
1506 bSumEkinhOld = TRUE;
1511 /* ######### BEGIN PREPARING EDR OUTPUT ########### */
1513 /* use the directly determined last velocity, not actually the averaged half steps */
1514 if (bTrotter && ir->eI == eiVV)
1516 enerd->term[F_EKIN] = last_ekin;
1518 enerd->term[F_ETOT] = enerd->term[F_EPOT] + enerd->term[F_EKIN];
1520 if (integratorHasConservedEnergyQuantity(ir))
1524 enerd->term[F_ECONSERVED] = enerd->term[F_ETOT] + saved_conserved_quantity;
1528 enerd->term[F_ECONSERVED] = enerd->term[F_ETOT] + NPT_energy(ir, state, &MassQ);
1531 /* ######### END PREPARING EDR OUTPUT ########### */
1537 if (fplog && do_log && bDoExpanded)
1539 /* only needed if doing expanded ensemble */
1540 PrintFreeEnergyInfoToFile(fplog, ir->fepvals, ir->expandedvals,
1541 ir->bSimTemp ? ir->simtempvals : nullptr,
1542 state_global->dfhist, state->fep_state, ir->nstlog, step);
1546 energyOutput.addDataAtEnergyStep(
1547 bDoDHDL, bCalcEnerStep, t, mdatoms->tmass, enerd, ir->fepvals,
1548 ir->expandedvals, lastbox,
1549 PTCouplingArrays{ state->boxv, state->nosehoover_xi, state->nosehoover_vxi,
1550 state->nhpres_xi, state->nhpres_vxi },
1551 state->fep_state, shake_vir, force_vir, total_vir, pres, ekind, mu_tot, constr);
1555 energyOutput.recordNonEnergyStep();
1558 gmx_bool do_dr = do_per_step(step, ir->nstdisreout);
1559 gmx_bool do_or = do_per_step(step, ir->nstorireout);
1561 if (doSimulatedAnnealing)
1563 gmx::EnergyOutput::printAnnealingTemperatures(do_log ? fplog : nullptr, groups,
1566 if (do_log || do_ene || do_dr || do_or)
1568 energyOutput.printStepToEnergyFile(mdoutf_get_fp_ene(outf), do_ene, do_dr, do_or,
1569 do_log ? fplog : nullptr, step, t,
1570 fr->fcdata.get(), awh.get());
1572 if (do_log && ir->bDoAwh && awh->hasFepLambdaDimension())
1574 const bool isInitialOutput = false;
1575 printLambdaStateToLog(fplog, state->lambda, isInitialOutput);
1580 pull_print_output(pull_work, step, t);
1583 if (do_per_step(step, ir->nstlog))
1585 if (fflush(fplog) != 0)
1587 gmx_fatal(FARGS, "Cannot flush logfile - maybe you are out of disk space?");
1593 /* Have to do this part _after_ outputting the logfile and the edr file */
1594 /* Gets written into the state at the beginning of next loop*/
1595 state->fep_state = lamnew;
1597 else if (ir->bDoAwh && awh->needForeignEnergyDifferences(step))
1599 state->fep_state = awh->fepLambdaState();
1601 /* Print the remaining wall clock time for the run */
1602 if (isMasterSimMasterRank(ms, MASTER(cr)) && (do_verbose || gmx_got_usr_signal()) && !bPMETunePrinting)
1606 fprintf(stderr, "\n");
1608 print_time(stderr, walltime_accounting, step, ir, cr);
1611 /* Ion/water position swapping.
1612 * Not done in last step since trajectory writing happens before this call
1613 * in the MD loop and exchanges would be lost anyway. */
1614 bNeedRepartition = FALSE;
1615 if ((ir->eSwapCoords != eswapNO) && (step > 0) && !bLastStep && do_per_step(step, ir->swap->nstswap))
1618 do_swapcoords(cr, step, t, ir, swap, wcycle, as_rvec_array(state->x.data()),
1619 state->box, MASTER(cr) && mdrunOptions.verbose, bRerunMD);
1621 if (bNeedRepartition && DOMAINDECOMP(cr))
1623 dd_collect_state(cr->dd, state, state_global);
1627 /* Replica exchange */
1631 bExchanged = replica_exchange(fplog, cr, ms, repl_ex, state_global, enerd, state, step, t);
1634 if ((bExchanged || bNeedRepartition) && DOMAINDECOMP(cr))
1636 dd_partition_system(fplog, mdlog, step, cr, TRUE, 1, state_global, *top_global, ir,
1637 imdSession, pull_work, state, &f, mdAtoms, &top, fr, vsite, constr,
1638 nrnb, wcycle, FALSE);
1639 shouldCheckNumberOfBondedInteractions = true;
1640 upd.setNumAtoms(state->natoms);
1646 /* ####### SET VARIABLES FOR NEXT ITERATION IF THEY STILL NEED IT ###### */
1647 /* With all integrators, except VV, we need to retain the pressure
1648 * at the current step for coupling at the next step.
1650 if ((state->flags & (1U << estPRES_PREV))
1651 && (bGStatEveryStep || (ir->nstpcouple > 0 && step % ir->nstpcouple == 0)))
1653 /* Store the pressure in t_state for pressure coupling
1654 * at the next MD step.
1656 copy_mat(pres, state->pres_prev);
1659 /* ####### END SET VARIABLES FOR NEXT ITERATION ###### */
1661 if ((membed != nullptr) && (!bLastStep))
1663 rescale_membed(step_rel, membed, as_rvec_array(state_global->x.data()));
1666 cycles = wallcycle_stop(wcycle, ewcSTEP);
1667 if (DOMAINDECOMP(cr) && wcycle)
1669 dd_cycles_add(cr->dd, cycles, ddCyclStep);
1672 /* increase the MD step number */
1676 resetHandler->resetCounters(step, step_rel, mdlog, fplog, cr, fr->nbv.get(), nrnb,
1677 fr->pmedata, pme_loadbal, wcycle, walltime_accounting);
1679 /* If bIMD is TRUE, the master updates the IMD energy record and sends positions to VMD client */
1680 imdSession->updateEnergyRecordAndSendPositionsAndEnergies(bInteractiveMDstep, step, bCalcEner);
1682 /* End of main MD loop */
1684 /* Closing TNG files can include compressing data. Therefore it is good to do that
1685 * before stopping the time measurements. */
1686 mdoutf_tng_close(outf);
1688 /* Stop measuring walltime */
1689 walltime_accounting_end_time(walltime_accounting);
1691 if (!thisRankHasDuty(cr, DUTY_PME))
1693 /* Tell the PME only node to finish */
1694 gmx_pme_send_finish(cr);
1699 if (ir->nstcalcenergy > 0)
1701 gmx::EnergyOutput::printAnnealingTemperatures(fplog, groups, &(ir->opts));
1702 energyOutput.printAverages(fplog, groups);
1709 pme_loadbal_done(pme_loadbal, fplog, mdlog, fr->nbv->useGpu());
1712 done_shellfc(fplog, shellfc, step_rel);
1714 if (useReplicaExchange && MASTER(cr))
1716 print_replica_exchange_statistics(fplog, repl_ex);
1719 walltime_accounting_set_nsteps_done(walltime_accounting, step_rel);
1721 global_stat_destroy(gstat);