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) 2011-2019,2020,2021, 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 MD runner routine calling all integrators.
41 * \author David van der Spoel <david.vanderspoel@icm.uu.se>
42 * \ingroup module_mdrun
59 #include "gromacs/commandline/filenm.h"
60 #include "gromacs/domdec/builder.h"
61 #include "gromacs/domdec/domdec.h"
62 #include "gromacs/domdec/domdec_struct.h"
63 #include "gromacs/domdec/gpuhaloexchange.h"
64 #include "gromacs/domdec/localatomsetmanager.h"
65 #include "gromacs/domdec/partition.h"
66 #include "gromacs/ewald/ewald_utils.h"
67 #include "gromacs/ewald/pme.h"
68 #include "gromacs/ewald/pme_gpu_program.h"
69 #include "gromacs/ewald/pme_only.h"
70 #include "gromacs/ewald/pme_pp_comm_gpu.h"
71 #include "gromacs/fileio/checkpoint.h"
72 #include "gromacs/fileio/gmxfio.h"
73 #include "gromacs/fileio/oenv.h"
74 #include "gromacs/fileio/tpxio.h"
75 #include "gromacs/gmxlib/network.h"
76 #include "gromacs/gmxlib/nrnb.h"
77 #include "gromacs/gpu_utils/device_stream_manager.h"
78 #include "gromacs/hardware/cpuinfo.h"
79 #include "gromacs/hardware/detecthardware.h"
80 #include "gromacs/hardware/device_management.h"
81 #include "gromacs/hardware/hardwaretopology.h"
82 #include "gromacs/hardware/printhardware.h"
83 #include "gromacs/imd/imd.h"
84 #include "gromacs/listed_forces/disre.h"
85 #include "gromacs/listed_forces/gpubonded.h"
86 #include "gromacs/listed_forces/listed_forces.h"
87 #include "gromacs/listed_forces/orires.h"
88 #include "gromacs/math/functions.h"
89 #include "gromacs/math/utilities.h"
90 #include "gromacs/math/vec.h"
91 #include "gromacs/mdlib/boxdeformation.h"
92 #include "gromacs/mdlib/broadcaststructs.h"
93 #include "gromacs/mdlib/calc_verletbuf.h"
94 #include "gromacs/mdlib/dispersioncorrection.h"
95 #include "gromacs/mdlib/enerdata_utils.h"
96 #include "gromacs/mdlib/force.h"
97 #include "gromacs/mdlib/forcerec.h"
98 #include "gromacs/mdlib/gmx_omp_nthreads.h"
99 #include "gromacs/mdlib/gpuforcereduction.h"
100 #include "gromacs/mdlib/makeconstraints.h"
101 #include "gromacs/mdlib/md_support.h"
102 #include "gromacs/mdlib/mdatoms.h"
103 #include "gromacs/mdlib/sighandler.h"
104 #include "gromacs/mdlib/stophandler.h"
105 #include "gromacs/mdlib/tgroup.h"
106 #include "gromacs/mdlib/updategroups.h"
107 #include "gromacs/mdlib/vsite.h"
108 #include "gromacs/mdrun/mdmodules.h"
109 #include "gromacs/mdrun/simulationcontext.h"
110 #include "gromacs/mdrun/simulationinput.h"
111 #include "gromacs/mdrun/simulationinputhandle.h"
112 #include "gromacs/mdrunutility/handlerestart.h"
113 #include "gromacs/mdrunutility/logging.h"
114 #include "gromacs/mdrunutility/multisim.h"
115 #include "gromacs/mdrunutility/printtime.h"
116 #include "gromacs/mdrunutility/threadaffinity.h"
117 #include "gromacs/mdtypes/checkpointdata.h"
118 #include "gromacs/mdtypes/commrec.h"
119 #include "gromacs/mdtypes/enerdata.h"
120 #include "gromacs/mdtypes/fcdata.h"
121 #include "gromacs/mdtypes/forcerec.h"
122 #include "gromacs/mdtypes/group.h"
123 #include "gromacs/mdtypes/inputrec.h"
124 #include "gromacs/mdtypes/interaction_const.h"
125 #include "gromacs/mdtypes/md_enums.h"
126 #include "gromacs/mdtypes/mdatom.h"
127 #include "gromacs/mdtypes/mdrunoptions.h"
128 #include "gromacs/mdtypes/observableshistory.h"
129 #include "gromacs/mdtypes/simulation_workload.h"
130 #include "gromacs/mdtypes/state.h"
131 #include "gromacs/mdtypes/state_propagator_data_gpu.h"
132 #include "gromacs/modularsimulator/modularsimulator.h"
133 #include "gromacs/nbnxm/gpu_data_mgmt.h"
134 #include "gromacs/nbnxm/nbnxm.h"
135 #include "gromacs/nbnxm/pairlist_tuning.h"
136 #include "gromacs/pbcutil/pbc.h"
137 #include "gromacs/pulling/output.h"
138 #include "gromacs/pulling/pull.h"
139 #include "gromacs/pulling/pull_rotation.h"
140 #include "gromacs/restraint/manager.h"
141 #include "gromacs/restraint/restraintmdmodule.h"
142 #include "gromacs/restraint/restraintpotential.h"
143 #include "gromacs/swap/swapcoords.h"
144 #include "gromacs/taskassignment/decidegpuusage.h"
145 #include "gromacs/taskassignment/decidesimulationworkload.h"
146 #include "gromacs/taskassignment/resourcedivision.h"
147 #include "gromacs/taskassignment/taskassignment.h"
148 #include "gromacs/taskassignment/usergpuids.h"
149 #include "gromacs/timing/gpu_timing.h"
150 #include "gromacs/timing/wallcycle.h"
151 #include "gromacs/timing/wallcyclereporting.h"
152 #include "gromacs/topology/mtop_util.h"
153 #include "gromacs/trajectory/trajectoryframe.h"
154 #include "gromacs/utility/basenetwork.h"
155 #include "gromacs/utility/cstringutil.h"
156 #include "gromacs/utility/exceptions.h"
157 #include "gromacs/utility/fatalerror.h"
158 #include "gromacs/utility/filestream.h"
159 #include "gromacs/utility/gmxassert.h"
160 #include "gromacs/utility/gmxmpi.h"
161 #include "gromacs/utility/keyvaluetree.h"
162 #include "gromacs/utility/logger.h"
163 #include "gromacs/utility/loggerbuilder.h"
164 #include "gromacs/utility/mdmodulenotification.h"
165 #include "gromacs/utility/physicalnodecommunicator.h"
166 #include "gromacs/utility/pleasecite.h"
167 #include "gromacs/utility/programcontext.h"
168 #include "gromacs/utility/smalloc.h"
169 #include "gromacs/utility/stringutil.h"
171 #include "isimulator.h"
172 #include "membedholder.h"
173 #include "replicaexchange.h"
174 #include "simulatorbuilder.h"
180 /*! \brief Manage any development feature flag variables encountered
182 * The use of dev features indicated by environment variables is
183 * logged in order to ensure that runs with such features enabled can
184 * be identified from their log and standard output. Any cross
185 * dependencies are also checked, and if unsatisfied, a fatal error
188 * Note that some development features overrides are applied already here:
189 * the GPU communication flags are set to false in non-tMPI and non-CUDA builds.
191 * \param[in] mdlog Logger object.
192 * \param[in] useGpuForNonbonded True if the nonbonded task is offloaded in this run.
193 * \param[in] pmeRunMode The PME run mode for this run
194 * \returns The object populated with development feature flags.
196 static DevelopmentFeatureFlags manageDevelopmentFeatures(const gmx::MDLogger& mdlog,
197 const bool useGpuForNonbonded,
198 const PmeRunMode pmeRunMode)
200 DevelopmentFeatureFlags devFlags;
202 // Some builds of GCC 5 give false positive warnings that these
203 // getenv results are ignored when clearly they are used.
204 #pragma GCC diagnostic push
205 #pragma GCC diagnostic ignored "-Wunused-result"
207 devFlags.enableGpuBufferOps =
208 GMX_GPU_CUDA && useGpuForNonbonded && (getenv("GMX_USE_GPU_BUFFER_OPS") != nullptr);
209 devFlags.enableGpuHaloExchange = GMX_GPU_CUDA && GMX_THREAD_MPI && getenv("GMX_GPU_DD_COMMS") != nullptr;
210 devFlags.forceGpuUpdateDefault = (getenv("GMX_FORCE_UPDATE_DEFAULT_GPU") != nullptr) || GMX_FAHCORE;
211 devFlags.enableGpuPmePPComm =
212 GMX_GPU_CUDA && GMX_THREAD_MPI && getenv("GMX_GPU_PME_PP_COMMS") != nullptr;
214 #pragma GCC diagnostic pop
216 if (devFlags.enableGpuBufferOps)
218 GMX_LOG(mdlog.warning)
220 .appendTextFormatted(
221 "This run uses the 'GPU buffer ops' feature, enabled by the "
222 "GMX_USE_GPU_BUFFER_OPS environment variable.");
225 if (devFlags.forceGpuUpdateDefault)
227 GMX_LOG(mdlog.warning)
229 .appendTextFormatted(
230 "This run will default to '-update gpu' as requested by the "
231 "GMX_FORCE_UPDATE_DEFAULT_GPU environment variable. GPU update with domain "
232 "decomposition lacks substantial testing and should be used with caution.");
235 if (devFlags.enableGpuHaloExchange)
237 if (useGpuForNonbonded)
239 if (!devFlags.enableGpuBufferOps)
241 GMX_LOG(mdlog.warning)
243 .appendTextFormatted(
244 "Enabling GPU buffer operations required by GMX_GPU_DD_COMMS "
245 "(equivalent with GMX_USE_GPU_BUFFER_OPS=1).");
246 devFlags.enableGpuBufferOps = true;
248 GMX_LOG(mdlog.warning)
250 .appendTextFormatted(
251 "This run has requested the 'GPU halo exchange' feature, enabled by "
253 "GMX_GPU_DD_COMMS environment variable.");
257 GMX_LOG(mdlog.warning)
259 .appendTextFormatted(
260 "GMX_GPU_DD_COMMS environment variable detected, but the 'GPU "
261 "halo exchange' feature will not be enabled as nonbonded interactions "
262 "are not offloaded.");
263 devFlags.enableGpuHaloExchange = false;
267 if (devFlags.enableGpuPmePPComm)
269 if (pmeRunMode == PmeRunMode::GPU)
271 if (!devFlags.enableGpuBufferOps)
273 GMX_LOG(mdlog.warning)
275 .appendTextFormatted(
276 "Enabling GPU buffer operations required by GMX_GPU_PME_PP_COMMS "
277 "(equivalent with GMX_USE_GPU_BUFFER_OPS=1).");
278 devFlags.enableGpuBufferOps = true;
280 GMX_LOG(mdlog.warning)
282 .appendTextFormatted(
283 "This run uses the 'GPU PME-PP communications' feature, enabled "
284 "by the GMX_GPU_PME_PP_COMMS environment variable.");
288 std::string clarification;
289 if (pmeRunMode == PmeRunMode::Mixed)
292 "PME FFT and gather are not offloaded to the GPU (PME is running in mixed "
297 clarification = "PME is not offloaded to the GPU.";
299 GMX_LOG(mdlog.warning)
302 "GMX_GPU_PME_PP_COMMS environment variable detected, but the "
303 "'GPU PME-PP communications' feature was not enabled as "
305 devFlags.enableGpuPmePPComm = false;
312 /*! \brief Barrier for safe simultaneous thread access to mdrunner data
314 * Used to ensure that the master thread does not modify mdrunner during copy
315 * on the spawned threads. */
316 static void threadMpiMdrunnerAccessBarrier()
319 MPI_Barrier(MPI_COMM_WORLD);
323 Mdrunner Mdrunner::cloneOnSpawnedThread() const
325 auto newRunner = Mdrunner(std::make_unique<MDModules>());
327 // All runners in the same process share a restraint manager resource because it is
328 // part of the interface to the client code, which is associated only with the
329 // original thread. Handles to the same resources can be obtained by copy.
331 newRunner.restraintManager_ = std::make_unique<RestraintManager>(*restraintManager_);
334 // Copy members of master runner.
335 // \todo Replace with builder when Simulation context and/or runner phases are better defined.
336 // Ref https://gitlab.com/gromacs/gromacs/-/issues/2587 and https://gitlab.com/gromacs/gromacs/-/issues/2375
337 newRunner.hw_opt = hw_opt;
338 newRunner.filenames = filenames;
340 newRunner.hwinfo_ = hwinfo_;
341 newRunner.oenv = oenv;
342 newRunner.mdrunOptions = mdrunOptions;
343 newRunner.domdecOptions = domdecOptions;
344 newRunner.nbpu_opt = nbpu_opt;
345 newRunner.pme_opt = pme_opt;
346 newRunner.pme_fft_opt = pme_fft_opt;
347 newRunner.bonded_opt = bonded_opt;
348 newRunner.update_opt = update_opt;
349 newRunner.nstlist_cmdline = nstlist_cmdline;
350 newRunner.replExParams = replExParams;
351 newRunner.pforce = pforce;
352 // Give the spawned thread the newly created valid communicator
353 // for the simulation.
354 newRunner.libraryWorldCommunicator = MPI_COMM_WORLD;
355 newRunner.simulationCommunicator = MPI_COMM_WORLD;
357 newRunner.startingBehavior = startingBehavior;
358 newRunner.stopHandlerBuilder_ = std::make_unique<StopHandlerBuilder>(*stopHandlerBuilder_);
359 newRunner.inputHolder_ = inputHolder_;
361 threadMpiMdrunnerAccessBarrier();
366 /*! \brief The callback used for running on spawned threads.
368 * Obtains the pointer to the master mdrunner object from the one
369 * argument permitted to the thread-launch API call, copies it to make
370 * a new runner for this thread, reinitializes necessary data, and
371 * proceeds to the simulation. */
372 static void mdrunner_start_fn(const void* arg)
376 auto masterMdrunner = reinterpret_cast<const gmx::Mdrunner*>(arg);
377 /* copy the arg list to make sure that it's thread-local. This
378 doesn't copy pointed-to items, of course; fnm, cr and fplog
379 are reset in the call below, all others should be const. */
380 gmx::Mdrunner mdrunner = masterMdrunner->cloneOnSpawnedThread();
383 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
387 void Mdrunner::spawnThreads(int numThreadsToLaunch)
390 /* now spawn new threads that start mdrunner_start_fn(), while
391 the main thread returns. Thread affinity is handled later. */
392 if (tMPI_Init_fn(TRUE, numThreadsToLaunch, TMPI_AFFINITY_NONE, mdrunner_start_fn, static_cast<const void*>(this))
395 GMX_THROW(gmx::InternalError("Failed to spawn thread-MPI threads"));
398 // Give the master thread the newly created valid communicator for
400 libraryWorldCommunicator = MPI_COMM_WORLD;
401 simulationCommunicator = MPI_COMM_WORLD;
402 threadMpiMdrunnerAccessBarrier();
404 GMX_UNUSED_VALUE(numThreadsToLaunch);
405 GMX_UNUSED_VALUE(mdrunner_start_fn);
411 /*! \brief Initialize variables for Verlet scheme simulation */
412 static void prepare_verlet_scheme(FILE* fplog,
416 const gmx_mtop_t* mtop,
418 bool makeGpuPairList,
419 const gmx::CpuInfo& cpuinfo)
421 // We checked the cut-offs in grompp, but double-check here.
422 // We have PME+LJcutoff kernels for rcoulomb>rvdw.
423 if (EEL_PME_EWALD(ir->coulombtype) && ir->vdwtype == eelCUT)
425 GMX_RELEASE_ASSERT(ir->rcoulomb >= ir->rvdw,
426 "With Verlet lists and PME we should have rcoulomb>=rvdw");
430 GMX_RELEASE_ASSERT(ir->rcoulomb == ir->rvdw,
431 "With Verlet lists and no PME rcoulomb and rvdw should be identical");
433 /* For NVE simulations, we will retain the initial list buffer */
434 if (EI_DYNAMICS(ir->eI) && ir->verletbuf_tol > 0
435 && !(EI_MD(ir->eI) && ir->etc == TemperatureCoupling::No))
437 /* Update the Verlet buffer size for the current run setup */
439 /* Here we assume SIMD-enabled kernels are being used. But as currently
440 * calc_verlet_buffer_size gives the same results for 4x8 and 4x4
441 * and 4x2 gives a larger buffer than 4x4, this is ok.
443 ListSetupType listType =
444 (makeGpuPairList ? ListSetupType::Gpu : ListSetupType::CpuSimdWhenSupported);
445 VerletbufListSetup listSetup = verletbufGetSafeListSetup(listType);
447 const real rlist_new =
448 calcVerletBufferSize(*mtop, det(box), *ir, ir->nstlist, ir->nstlist - 1, -1, listSetup);
450 if (rlist_new != ir->rlist)
452 if (fplog != nullptr)
455 "\nChanging rlist from %g to %g for non-bonded %dx%d atom kernels\n\n",
458 listSetup.cluster_size_i,
459 listSetup.cluster_size_j);
461 ir->rlist = rlist_new;
465 if (nstlist_cmdline > 0 && (!EI_DYNAMICS(ir->eI) || ir->verletbuf_tol <= 0))
468 "Can not set nstlist without %s",
469 !EI_DYNAMICS(ir->eI) ? "dynamics" : "verlet-buffer-tolerance");
472 if (EI_DYNAMICS(ir->eI))
474 /* Set or try nstlist values */
475 increaseNstlist(fplog, cr, ir, nstlist_cmdline, mtop, box, makeGpuPairList, cpuinfo);
479 /*! \brief Override the nslist value in inputrec
481 * with value passed on the command line (if any)
483 static void override_nsteps_cmdline(const gmx::MDLogger& mdlog, int64_t nsteps_cmdline, t_inputrec* ir)
487 /* override with anything else than the default -2 */
488 if (nsteps_cmdline > -2)
490 char sbuf_steps[STEPSTRSIZE];
491 char sbuf_msg[STRLEN];
493 ir->nsteps = nsteps_cmdline;
494 if (EI_DYNAMICS(ir->eI) && nsteps_cmdline != -1)
497 "Overriding nsteps with value passed on the command line: %s steps, %.3g ps",
498 gmx_step_str(nsteps_cmdline, sbuf_steps),
499 fabs(nsteps_cmdline * ir->delta_t));
504 "Overriding nsteps with value passed on the command line: %s steps",
505 gmx_step_str(nsteps_cmdline, sbuf_steps));
508 GMX_LOG(mdlog.warning).asParagraph().appendText(sbuf_msg);
510 else if (nsteps_cmdline < -2)
512 gmx_fatal(FARGS, "Invalid nsteps value passed on the command line: %" PRId64, nsteps_cmdline);
514 /* Do nothing if nsteps_cmdline == -2 */
520 /*! \brief Return whether GPU acceleration of nonbondeds is supported with the given settings.
522 * If not, and if a warning may be issued, logs a warning about
523 * falling back to CPU code. With thread-MPI, only the first
524 * call to this function should have \c issueWarning true. */
525 static bool gpuAccelerationOfNonbondedIsUseful(const MDLogger& mdlog, const t_inputrec& ir, bool issueWarning)
527 bool gpuIsUseful = true;
530 if (ir.opts.ngener - ir.nwall > 1)
532 /* The GPU code does not support more than one energy group.
533 * If the user requested GPUs explicitly, a fatal error is given later.
537 "Multiple energy groups is not implemented for GPUs, falling back to the CPU. "
538 "For better performance, run on the GPU without energy groups and then do "
539 "gmx mdrun -rerun option on the trajectory with an energy group .tpr file.";
545 warning = "TPI is not implemented for GPUs.";
548 if (!gpuIsUseful && issueWarning)
550 GMX_LOG(mdlog.warning).asParagraph().appendText(warning);
556 //! Initializes the logger for mdrun.
557 static gmx::LoggerOwner buildLogger(FILE* fplog, const bool isSimulationMasterRank)
559 gmx::LoggerBuilder builder;
560 if (fplog != nullptr)
562 builder.addTargetFile(gmx::MDLogger::LogLevel::Info, fplog);
564 if (isSimulationMasterRank)
566 builder.addTargetStream(gmx::MDLogger::LogLevel::Warning, &gmx::TextOutputFile::standardError());
568 return builder.build();
571 //! Make a TaskTarget from an mdrun argument string.
572 static TaskTarget findTaskTarget(const char* optionString)
574 TaskTarget returnValue = TaskTarget::Auto;
576 if (strncmp(optionString, "auto", 3) == 0)
578 returnValue = TaskTarget::Auto;
580 else if (strncmp(optionString, "cpu", 3) == 0)
582 returnValue = TaskTarget::Cpu;
584 else if (strncmp(optionString, "gpu", 3) == 0)
586 returnValue = TaskTarget::Gpu;
590 GMX_ASSERT(false, "Option string should have been checked for sanity already");
596 //! Finish run, aggregate data to print performance info.
597 static void finish_run(FILE* fplog,
598 const gmx::MDLogger& mdlog,
600 const t_inputrec* inputrec,
602 gmx_wallcycle_t wcycle,
603 gmx_walltime_accounting_t walltime_accounting,
604 nonbonded_verlet_t* nbv,
605 const gmx_pme_t* pme,
609 double nbfs = 0, mflop = 0;
610 double elapsed_time, elapsed_time_over_all_ranks, elapsed_time_over_all_threads,
611 elapsed_time_over_all_threads_over_all_ranks;
612 /* Control whether it is valid to print a report. Only the
613 simulation master may print, but it should not do so if the run
614 terminated e.g. before a scheduled reset step. This is
615 complicated by the fact that PME ranks are unaware of the
616 reason why they were sent a pmerecvqxFINISH. To avoid
617 communication deadlocks, we always do the communication for the
618 report, even if we've decided not to write the report, because
619 how long it takes to finish the run is not important when we've
620 decided not to report on the simulation performance.
622 Further, we only report performance for dynamical integrators,
623 because those are the only ones for which we plan to
624 consider doing any optimizations. */
625 bool printReport = EI_DYNAMICS(inputrec->eI) && SIMMASTER(cr);
627 if (printReport && !walltime_accounting_get_valid_finish(walltime_accounting))
629 GMX_LOG(mdlog.warning)
631 .appendText("Simulation ended prematurely, no performance report will be written.");
636 std::unique_ptr<t_nrnb> nrnbTotalStorage;
639 nrnbTotalStorage = std::make_unique<t_nrnb>();
640 nrnb_tot = nrnbTotalStorage.get();
642 MPI_Allreduce(nrnb->n, nrnb_tot->n, eNRNB, MPI_DOUBLE, MPI_SUM, cr->mpi_comm_mysim);
650 elapsed_time = walltime_accounting_get_time_since_reset(walltime_accounting);
651 elapsed_time_over_all_threads =
652 walltime_accounting_get_time_since_reset_over_all_threads(walltime_accounting);
656 /* reduce elapsed_time over all MPI ranks in the current simulation */
657 MPI_Allreduce(&elapsed_time, &elapsed_time_over_all_ranks, 1, MPI_DOUBLE, MPI_SUM, cr->mpi_comm_mysim);
658 elapsed_time_over_all_ranks /= cr->nnodes;
659 /* Reduce elapsed_time_over_all_threads over all MPI ranks in the
660 * current simulation. */
661 MPI_Allreduce(&elapsed_time_over_all_threads,
662 &elapsed_time_over_all_threads_over_all_ranks,
671 elapsed_time_over_all_ranks = elapsed_time;
672 elapsed_time_over_all_threads_over_all_ranks = elapsed_time_over_all_threads;
677 print_flop(fplog, nrnb_tot, &nbfs, &mflop);
680 if (thisRankHasDuty(cr, DUTY_PP) && DOMAINDECOMP(cr))
682 print_dd_statistics(cr, inputrec, fplog);
685 /* TODO Move the responsibility for any scaling by thread counts
686 * to the code that handled the thread region, so that there's a
687 * mechanism to keep cycle counting working during the transition
688 * to task parallelism. */
689 int nthreads_pp = gmx_omp_nthreads_get(emntNonbonded);
690 int nthreads_pme = gmx_omp_nthreads_get(emntPME);
691 wallcycle_scale_by_num_threads(
692 wcycle, thisRankHasDuty(cr, DUTY_PME) && !thisRankHasDuty(cr, DUTY_PP), nthreads_pp, nthreads_pme);
693 auto cycle_sum(wallcycle_sum(cr, wcycle));
697 auto nbnxn_gpu_timings =
698 (nbv != nullptr && nbv->useGpu()) ? Nbnxm::gpu_get_timings(nbv->gpu_nbv) : nullptr;
699 gmx_wallclock_gpu_pme_t pme_gpu_timings = {};
701 if (pme_gpu_task_enabled(pme))
703 pme_gpu_get_timings(pme, &pme_gpu_timings);
705 wallcycle_print(fplog,
711 elapsed_time_over_all_ranks,
717 if (EI_DYNAMICS(inputrec->eI))
719 delta_t = inputrec->delta_t;
725 elapsed_time_over_all_threads_over_all_ranks,
726 elapsed_time_over_all_ranks,
727 walltime_accounting_get_nsteps_done_since_reset(walltime_accounting),
735 elapsed_time_over_all_threads_over_all_ranks,
736 elapsed_time_over_all_ranks,
737 walltime_accounting_get_nsteps_done_since_reset(walltime_accounting),
745 int Mdrunner::mdrunner()
748 t_forcerec* fr = nullptr;
749 real ewaldcoeff_q = 0;
750 real ewaldcoeff_lj = 0;
751 int nChargePerturbed = -1, nTypePerturbed = 0;
752 gmx_wallcycle_t wcycle;
753 gmx_walltime_accounting_t walltime_accounting = nullptr;
754 MembedHolder membedHolder(filenames.size(), filenames.data());
756 /* CAUTION: threads may be started later on in this function, so
757 cr doesn't reflect the final parallel state right now */
760 /* TODO: inputrec should tell us whether we use an algorithm, not a file option */
761 const bool doEssentialDynamics = opt2bSet("-ei", filenames.size(), filenames.data());
762 const bool doRerun = mdrunOptions.rerun;
764 // Handle task-assignment related user options.
765 EmulateGpuNonbonded emulateGpuNonbonded =
766 (getenv("GMX_EMULATE_GPU") != nullptr ? EmulateGpuNonbonded::Yes : EmulateGpuNonbonded::No);
768 std::vector<int> userGpuTaskAssignment;
771 userGpuTaskAssignment = parseUserTaskAssignmentString(hw_opt.userGpuTaskAssignment);
773 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
774 auto nonbondedTarget = findTaskTarget(nbpu_opt);
775 auto pmeTarget = findTaskTarget(pme_opt);
776 auto pmeFftTarget = findTaskTarget(pme_fft_opt);
777 auto bondedTarget = findTaskTarget(bonded_opt);
778 auto updateTarget = findTaskTarget(update_opt);
780 FILE* fplog = nullptr;
781 // If we are appending, we don't write log output because we need
782 // to check that the old log file matches what the checkpoint file
783 // expects. Otherwise, we should start to write log output now if
784 // there is a file ready for it.
785 if (logFileHandle != nullptr && startingBehavior != StartingBehavior::RestartWithAppending)
787 fplog = gmx_fio_getfp(logFileHandle);
789 const bool isSimulationMasterRank = findIsSimulationMasterRank(ms, simulationCommunicator);
790 gmx::LoggerOwner logOwner(buildLogger(fplog, isSimulationMasterRank));
791 gmx::MDLogger mdlog(logOwner.logger());
793 gmx_print_detected_hardware(fplog, isSimulationMasterRank && isMasterSim(ms), mdlog, hwinfo_);
795 std::vector<int> gpuIdsToUse = makeGpuIdsToUse(hwinfo_->deviceInfoList, hw_opt.gpuIdsAvailable);
796 const int numDevicesToUse = gmx::ssize(gpuIdsToUse);
798 // Print citation requests after all software/hardware printing
799 pleaseCiteGromacs(fplog);
801 // Note: legacy program logic relies on checking whether these pointers are assigned.
802 // Objects may or may not be allocated later.
803 std::unique_ptr<t_inputrec> inputrec;
804 std::unique_ptr<t_state> globalState;
806 auto partialDeserializedTpr = std::make_unique<PartialDeserializedTprFile>();
808 if (isSimulationMasterRank)
810 // Allocate objects to be initialized by later function calls.
811 /* Only the master rank has the global state */
812 globalState = std::make_unique<t_state>();
813 inputrec = std::make_unique<t_inputrec>();
815 /* Read (nearly) all data required for the simulation
816 * and keep the partly serialized tpr contents to send to other ranks later
818 applyGlobalSimulationState(
819 *inputHolder_.get(), partialDeserializedTpr.get(), globalState.get(), inputrec.get(), &mtop);
822 /* Check and update the hardware options for internal consistency */
823 checkAndUpdateHardwareOptions(
824 mdlog, &hw_opt, isSimulationMasterRank, domdecOptions.numPmeRanks, inputrec.get());
826 if (GMX_THREAD_MPI && isSimulationMasterRank)
828 bool useGpuForNonbonded = false;
829 bool useGpuForPme = false;
832 GMX_RELEASE_ASSERT(inputrec != nullptr, "Keep the compiler happy");
834 // If the user specified the number of ranks, then we must
835 // respect that, but in default mode, we need to allow for
836 // the number of GPUs to choose the number of ranks.
837 auto canUseGpuForNonbonded = buildSupportsNonbondedOnGpu(nullptr);
838 useGpuForNonbonded = decideWhetherToUseGpusForNonbondedWithThreadMpi(
841 userGpuTaskAssignment,
843 canUseGpuForNonbonded,
844 gpuAccelerationOfNonbondedIsUseful(mdlog, *inputrec, GMX_THREAD_MPI),
845 hw_opt.nthreads_tmpi);
846 useGpuForPme = decideWhetherToUseGpusForPmeWithThreadMpi(useGpuForNonbonded,
849 userGpuTaskAssignment,
852 hw_opt.nthreads_tmpi,
853 domdecOptions.numPmeRanks);
855 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
857 /* Determine how many thread-MPI ranks to start.
859 * TODO Over-writing the user-supplied value here does
860 * prevent any possible subsequent checks from working
862 hw_opt.nthreads_tmpi = get_nthreads_mpi(hwinfo_,
870 membedHolder.doMembed());
872 // Now start the threads for thread MPI.
873 spawnThreads(hw_opt.nthreads_tmpi);
874 // The spawned threads enter mdrunner() and execution of
875 // master and spawned threads joins at the end of this block.
878 GMX_RELEASE_ASSERT(ms || simulationCommunicator != MPI_COMM_NULL,
879 "Must have valid communicator unless running a multi-simulation");
880 CommrecHandle crHandle = init_commrec(simulationCommunicator);
881 t_commrec* cr = crHandle.get();
882 GMX_RELEASE_ASSERT(cr != nullptr, "Must have valid commrec");
884 PhysicalNodeCommunicator physicalNodeComm(libraryWorldCommunicator, gmx_physicalnode_id_hash());
886 // If we detected the topology on this system, double-check that it makes sense
887 if (hwinfo_->hardwareTopology->isThisSystem())
889 hardwareTopologyDoubleCheckDetection(mdlog, *hwinfo_->hardwareTopology);
894 /* now broadcast everything to the non-master nodes/threads: */
895 if (!isSimulationMasterRank)
897 // Until now, only the master rank has a non-null pointer.
898 // On non-master ranks, allocate the object that will receive data in the following call.
899 inputrec = std::make_unique<t_inputrec>();
901 init_parallel(cr->mpiDefaultCommunicator,
905 partialDeserializedTpr.get());
907 GMX_RELEASE_ASSERT(inputrec != nullptr, "All ranks should have a valid inputrec now");
908 partialDeserializedTpr.reset(nullptr);
911 !inputrec->useConstantAcceleration,
912 "Linear acceleration has been removed in GROMACS 2022, and was broken for many years "
913 "before that. Use GROMACS 4.5 or earlier if you need this feature.");
915 // Now the number of ranks is known to all ranks, and each knows
916 // the inputrec read by the master rank. The ranks can now all run
917 // the task-deciding functions and will agree on the result
918 // without needing to communicate.
919 const bool useDomainDecomposition = (PAR(cr) && !(EI_TPI(inputrec->eI) || inputrec->eI == eiNM));
921 // Note that these variables describe only their own node.
923 // Note that when bonded interactions run on a GPU they always run
924 // alongside a nonbonded task, so do not influence task assignment
925 // even though they affect the force calculation workload.
926 bool useGpuForNonbonded = false;
927 bool useGpuForPme = false;
928 bool useGpuForBonded = false;
929 bool useGpuForUpdate = false;
930 bool gpusWereDetected = hwinfo_->ngpu_compatible_tot > 0;
933 // It's possible that there are different numbers of GPUs on
934 // different nodes, which is the user's responsibility to
935 // handle. If unsuitable, we will notice that during task
937 auto canUseGpuForNonbonded = buildSupportsNonbondedOnGpu(nullptr);
938 useGpuForNonbonded = decideWhetherToUseGpusForNonbonded(
940 userGpuTaskAssignment,
942 canUseGpuForNonbonded,
943 gpuAccelerationOfNonbondedIsUseful(mdlog, *inputrec, !GMX_THREAD_MPI),
945 useGpuForPme = decideWhetherToUseGpusForPme(useGpuForNonbonded,
947 userGpuTaskAssignment,
950 cr->sizeOfDefaultCommunicator,
951 domdecOptions.numPmeRanks,
953 useGpuForBonded = decideWhetherToUseGpusForBonded(
954 useGpuForNonbonded, useGpuForPme, bondedTarget, *inputrec, mtop, domdecOptions.numPmeRanks, gpusWereDetected);
956 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
958 const PmeRunMode pmeRunMode = determinePmeRunMode(useGpuForPme, pmeFftTarget, *inputrec);
960 // Initialize development feature flags that enabled by environment variable
961 // and report those features that are enabled.
962 const DevelopmentFeatureFlags devFlags =
963 manageDevelopmentFeatures(mdlog, useGpuForNonbonded, pmeRunMode);
965 const bool useModularSimulator = checkUseModularSimulator(false,
973 membedHolder.doMembed());
976 // TODO: hide restraint implementation details from Mdrunner.
977 // There is nothing unique about restraints at this point as far as the
978 // Mdrunner is concerned. The Mdrunner should just be getting a sequence of
979 // factory functions from the SimulationContext on which to call mdModules_->add().
980 // TODO: capture all restraints into a single RestraintModule, passed to the runner builder.
981 for (auto&& restraint : restraintManager_->getRestraints())
983 auto module = RestraintMDModule::create(restraint, restraint->sites());
984 mdModules_->add(std::move(module));
987 // TODO: Error handling
988 mdModules_->assignOptionsToModules(*inputrec->params, nullptr);
989 // now that the MdModules know their options, they know which callbacks to sign up to
990 mdModules_->subscribeToSimulationSetupNotifications();
991 const auto& mdModulesNotifier = mdModules_->notifier().simulationSetupNotifications_;
993 if (inputrec->internalParameters != nullptr)
995 mdModulesNotifier.notify(*inputrec->internalParameters);
998 if (fplog != nullptr)
1000 pr_inputrec(fplog, 0, "Input Parameters", inputrec.get(), FALSE);
1001 fprintf(fplog, "\n");
1006 /* In rerun, set velocities to zero if present */
1007 if (doRerun && ((globalState->flags & (1 << estV)) != 0))
1009 // rerun does not use velocities
1013 "Rerun trajectory contains velocities. Rerun does only evaluate "
1014 "potential energy and forces. The velocities will be ignored.");
1015 for (int i = 0; i < globalState->natoms; i++)
1017 clear_rvec(globalState->v[i]);
1019 globalState->flags &= ~(1 << estV);
1022 /* now make sure the state is initialized and propagated */
1023 set_state_entries(globalState.get(), inputrec.get(), useModularSimulator);
1026 /* NM and TPI parallelize over force/energy calculations, not atoms,
1027 * so we need to initialize and broadcast the global state.
1029 if (inputrec->eI == eiNM || inputrec->eI == eiTPI)
1033 globalState = std::make_unique<t_state>();
1035 broadcastStateWithoutDynamics(
1036 cr->mpiDefaultCommunicator, DOMAINDECOMP(cr), PAR(cr), globalState.get());
1039 /* A parallel command line option consistency check that we can
1040 only do after any threads have started. */
1042 && (domdecOptions.numCells[XX] > 1 || domdecOptions.numCells[YY] > 1
1043 || domdecOptions.numCells[ZZ] > 1 || domdecOptions.numPmeRanks > 0))
1046 "The -dd or -npme option request a parallel simulation, "
1048 "but %s was compiled without threads or MPI enabled",
1049 output_env_get_program_display_name(oenv));
1050 #elif GMX_THREAD_MPI
1051 "but the number of MPI-threads (option -ntmpi) is not set or is 1");
1053 "but %s was not started through mpirun/mpiexec or only one rank was requested "
1054 "through mpirun/mpiexec",
1055 output_env_get_program_display_name(oenv));
1059 if (doRerun && (EI_ENERGY_MINIMIZATION(inputrec->eI) || eiNM == inputrec->eI))
1062 "The .mdp file specified an energy mininization or normal mode algorithm, and "
1063 "these are not compatible with mdrun -rerun");
1066 /* Object for collecting reasons for not using PME-only ranks */
1067 SeparatePmeRanksPermitted separatePmeRanksPermitted;
1069 /* Permit MDModules to notify whether they want to use PME-only ranks */
1070 mdModulesNotifier.notify(&separatePmeRanksPermitted);
1072 /* If simulation is not using PME then disable PME-only ranks */
1073 if (!(EEL_PME(inputrec->coulombtype) || EVDW_PME(inputrec->vdwtype)))
1075 separatePmeRanksPermitted.disablePmeRanks(
1076 "PME-only ranks are requested, but the system does not use PME "
1077 "for electrostatics or LJ");
1080 /* With NB GPUs we don't automatically use PME-only CPU ranks. PME ranks can
1081 * improve performance with many threads per GPU, since our OpenMP
1082 * scaling is bad, but it's difficult to automate the setup.
1084 if (useGpuForNonbonded && domdecOptions.numPmeRanks < 0)
1086 separatePmeRanksPermitted.disablePmeRanks(
1087 "PME-only CPU ranks are not automatically used when "
1088 "non-bonded interactions are computed on GPUs");
1091 /* If GPU is used for PME then only 1 PME rank is permitted */
1092 if (useGpuForPme && (domdecOptions.numPmeRanks < 0 || domdecOptions.numPmeRanks > 1))
1094 separatePmeRanksPermitted.disablePmeRanks(
1095 "PME GPU decomposition is not supported. Only one separate PME-only GPU rank "
1099 /* Disable PME-only ranks if some parts of the code requested so and it's up to GROMACS to decide */
1100 if (!separatePmeRanksPermitted.permitSeparatePmeRanks() && domdecOptions.numPmeRanks < 0)
1102 domdecOptions.numPmeRanks = 0;
1105 .appendText("Simulation will not use PME-only ranks because: "
1106 + separatePmeRanksPermitted.reasonsWhyDisabled());
1109 /* If some parts of the code could not use PME-only ranks and
1110 * user explicitly used mdrun -npme option then throw an error */
1111 if (!separatePmeRanksPermitted.permitSeparatePmeRanks() && domdecOptions.numPmeRanks > 0)
1113 gmx_fatal_collective(FARGS,
1114 cr->mpiDefaultCommunicator,
1116 "Requested -npme %d option is not viable because: %s",
1117 domdecOptions.numPmeRanks,
1118 separatePmeRanksPermitted.reasonsWhyDisabled().c_str());
1121 /* NMR restraints must be initialized before load_checkpoint,
1122 * since with time averaging the history is added to t_state.
1123 * For proper consistency check we therefore need to extend
1125 * So the PME-only nodes (if present) will also initialize
1126 * the distance restraints.
1129 /* This needs to be called before read_checkpoint to extend the state */
1130 t_disresdata* disresdata;
1131 snew(disresdata, 1);
1135 DisResRunMode::MDRun,
1136 MASTER(cr) ? DDRole::Master : DDRole::Agent,
1137 PAR(cr) ? NumRanks::Multiple : NumRanks::Single,
1142 replExParams.exchangeInterval > 0);
1144 t_oriresdata* oriresdata;
1145 snew(oriresdata, 1);
1146 init_orires(fplog, &mtop, inputrec.get(), cr, ms, globalState.get(), oriresdata);
1148 auto deform = prepareBoxDeformation(globalState != nullptr ? globalState->box : box,
1149 MASTER(cr) ? DDRole::Master : DDRole::Agent,
1150 PAR(cr) ? NumRanks::Multiple : NumRanks::Single,
1151 cr->mpi_comm_mygroup,
1155 /* We have to remember the generation's first step before reading checkpoint.
1156 This way, we can report to the F@H core both the generation's first step
1157 and the restored first step, thus making it able to distinguish between
1158 an interruption/resume and start of the n-th generation simulation.
1159 Having this information, the F@H core can correctly calculate and report
1162 int gen_first_step = 0;
1165 gen_first_step = inputrec->init_step;
1169 ObservablesHistory observablesHistory = {};
1171 auto modularSimulatorCheckpointData = std::make_unique<ReadCheckpointDataHolder>();
1172 if (startingBehavior != StartingBehavior::NewSimulation)
1174 /* Check if checkpoint file exists before doing continuation.
1175 * This way we can use identical input options for the first and subsequent runs...
1177 if (mdrunOptions.numStepsCommandline > -2)
1179 /* Temporarily set the number of steps to unlimited to avoid
1180 * triggering the nsteps check in load_checkpoint().
1181 * This hack will go away soon when the -nsteps option is removed.
1183 inputrec->nsteps = -1;
1186 // Finish applying initial simulation state information from external sources on all ranks.
1187 // Reconcile checkpoint file data with Mdrunner state established up to this point.
1188 applyLocalState(*inputHolder_.get(),
1191 domdecOptions.numCells,
1194 &observablesHistory,
1195 mdrunOptions.reproducible,
1196 mdModules_->notifier(),
1197 modularSimulatorCheckpointData.get(),
1198 useModularSimulator);
1199 // TODO: (#3652) Synchronize filesystem state, SimulationInput contents, and program
1201 // on all code paths.
1202 // Write checkpoint or provide hook to update SimulationInput.
1203 // If there was a checkpoint file, SimulationInput contains more information
1204 // than if there wasn't. At this point, we have synchronized the in-memory
1205 // state with the filesystem state only for restarted simulations. We should
1206 // be calling applyLocalState unconditionally and expect that the completeness
1207 // of SimulationInput is not dependent on its creation method.
1209 if (startingBehavior == StartingBehavior::RestartWithAppending && logFileHandle)
1211 // Now we can start normal logging to the truncated log file.
1212 fplog = gmx_fio_getfp(logFileHandle);
1213 prepareLogAppending(fplog);
1214 logOwner = buildLogger(fplog, MASTER(cr));
1215 mdlog = logOwner.logger();
1222 fcRegisterSteps(inputrec->nsteps + inputrec->init_step, gen_first_step);
1226 if (mdrunOptions.numStepsCommandline > -2)
1231 "The -nsteps functionality is deprecated, and may be removed in a future "
1233 "Consider using gmx convert-tpr -nsteps or changing the appropriate .mdp "
1236 /* override nsteps with value set on the commandline */
1237 override_nsteps_cmdline(mdlog, mdrunOptions.numStepsCommandline, inputrec.get());
1239 if (isSimulationMasterRank)
1241 copy_mat(globalState->box, box);
1246 gmx_bcast(sizeof(box), box, cr->mpiDefaultCommunicator);
1249 if (inputrec->cutoff_scheme != ecutsVERLET)
1252 "This group-scheme .tpr file can no longer be run by mdrun. Please update to the "
1253 "Verlet scheme, or use an earlier version of GROMACS if necessary.");
1255 /* Update rlist and nstlist. */
1256 /* Note: prepare_verlet_scheme is calling increaseNstlist(...), which (while attempting to
1257 * increase rlist) tries to check if the newly chosen value fits with the DD scheme. As this is
1258 * run before any DD scheme is set up, this check is never executed. See #3334 for more details.
1260 prepare_verlet_scheme(fplog,
1266 useGpuForNonbonded || (emulateGpuNonbonded == EmulateGpuNonbonded::Yes),
1269 // This builder is necessary while we have multi-part construction
1270 // of DD. Before DD is constructed, we use the existence of
1271 // the builder object to indicate that further construction of DD
1273 std::unique_ptr<DomainDecompositionBuilder> ddBuilder;
1274 if (useDomainDecomposition)
1276 ddBuilder = std::make_unique<DomainDecompositionBuilder>(
1284 positionsFromStatePointer(globalState.get()));
1288 /* PME, if used, is done on all nodes with 1D decomposition */
1289 cr->nnodes = cr->sizeOfDefaultCommunicator;
1290 cr->sim_nodeid = cr->rankInDefaultCommunicator;
1291 cr->nodeid = cr->rankInDefaultCommunicator;
1293 cr->duty = (DUTY_PP | DUTY_PME);
1295 if (inputrec->pbcType == PbcType::Screw)
1297 gmx_fatal(FARGS, "pbc=screw is only implemented with domain decomposition");
1301 // Produce the task assignment for this rank - done after DD is constructed
1302 GpuTaskAssignments gpuTaskAssignments = GpuTaskAssignmentsBuilder::build(
1304 userGpuTaskAssignment,
1306 simulationCommunicator,
1314 thisRankHasDuty(cr, DUTY_PP),
1315 // TODO cr->duty & DUTY_PME should imply that a PME
1316 // algorithm is active, but currently does not.
1317 EEL_PME(inputrec->coulombtype) && thisRankHasDuty(cr, DUTY_PME));
1319 // Get the device handles for the modules, nullptr when no task is assigned.
1321 DeviceInformation* deviceInfo = gpuTaskAssignments.initDevice(&deviceId);
1323 // timing enabling - TODO put this in gpu_utils (even though generally this is just option handling?)
1324 bool useTiming = true;
1328 /* WARNING: CUDA timings are incorrect with multiple streams.
1329 * This is the main reason why they are disabled by default.
1331 // TODO: Consider turning on by default when we can detect nr of streams.
1332 useTiming = (getenv("GMX_ENABLE_GPU_TIMING") != nullptr);
1334 else if (GMX_GPU_OPENCL)
1336 useTiming = (getenv("GMX_DISABLE_GPU_TIMING") == nullptr);
1339 // TODO Currently this is always built, yet DD partition code
1340 // checks if it is built before using it. Probably it should
1341 // become an MDModule that is made only when another module
1342 // requires it (e.g. pull, CompEl, density fitting), so that we
1343 // don't update the local atom sets unilaterally every step.
1344 LocalAtomSetManager atomSets;
1347 // TODO Pass the GPU streams to ddBuilder to use in buffer
1348 // transfers (e.g. halo exchange)
1349 cr->dd = ddBuilder->build(&atomSets);
1350 // The builder's job is done, so destruct it
1351 ddBuilder.reset(nullptr);
1352 // Note that local state still does not exist yet.
1355 // The GPU update is decided here because we need to know whether the constraints or
1356 // SETTLEs can span accross the domain borders (i.e. whether or not update groups are
1357 // defined). This is only known after DD is initialized, hence decision on using GPU
1358 // update is done so late.
1361 const bool useUpdateGroups = cr->dd ? ddUsesUpdateGroups(*cr->dd) : false;
1363 useGpuForUpdate = decideWhetherToUseGpuForUpdate(useDomainDecomposition,
1366 domdecOptions.numPmeRanks > 0,
1372 doEssentialDynamics,
1373 gmx_mtop_ftype_count(mtop, F_ORIRES) > 0,
1374 replExParams.exchangeInterval > 0,
1379 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
1381 const bool printHostName = (cr->nnodes > 1);
1382 gpuTaskAssignments.reportGpuUsage(mdlog, printHostName, useGpuForBonded, pmeRunMode, useGpuForUpdate);
1384 const bool disableNonbondedCalculation = (getenv("GMX_NO_NONBONDED") != nullptr);
1385 if (disableNonbondedCalculation)
1387 /* turn off non-bonded calculations */
1388 GMX_LOG(mdlog.warning)
1391 "Found environment variable GMX_NO_NONBONDED.\n"
1392 "Disabling nonbonded calculations.");
1395 MdrunScheduleWorkload runScheduleWork;
1397 bool useGpuDirectHalo = decideWhetherToUseGpuForHalo(devFlags,
1398 havePPDomainDecomposition(cr),
1400 useModularSimulator,
1402 EI_ENERGY_MINIMIZATION(inputrec->eI));
1404 // Also populates the simulation constant workload description.
1405 runScheduleWork.simulationWork = createSimulationWorkload(*inputrec,
1406 disableNonbondedCalculation,
1414 std::unique_ptr<DeviceStreamManager> deviceStreamManager = nullptr;
1416 if (deviceInfo != nullptr)
1418 if (DOMAINDECOMP(cr) && thisRankHasDuty(cr, DUTY_PP))
1420 dd_setup_dlb_resource_sharing(cr, deviceId);
1422 deviceStreamManager = std::make_unique<DeviceStreamManager>(
1423 *deviceInfo, havePPDomainDecomposition(cr), runScheduleWork.simulationWork, useTiming);
1426 // If the user chose a task assignment, give them some hints
1427 // where appropriate.
1428 if (!userGpuTaskAssignment.empty())
1430 gpuTaskAssignments.logPerformanceHints(mdlog, numDevicesToUse);
1435 /* After possible communicator splitting in make_dd_communicators.
1436 * we can set up the intra/inter node communication.
1438 gmx_setup_nodecomm(fplog, cr);
1444 GMX_LOG(mdlog.warning)
1446 .appendTextFormatted(
1447 "This is simulation %d out of %d running as a composite GROMACS\n"
1448 "multi-simulation job. Setup for this simulation:\n",
1449 ms->simulationIndex_,
1450 ms->numSimulations_);
1452 GMX_LOG(mdlog.warning)
1453 .appendTextFormatted("Using %d MPI %s\n",
1456 cr->nnodes == 1 ? "thread" : "threads"
1458 cr->nnodes == 1 ? "process" : "processes"
1464 // If mdrun -pin auto honors any affinity setting that already
1465 // exists. If so, it is nice to provide feedback about whether
1466 // that existing affinity setting was from OpenMP or something
1467 // else, so we run this code both before and after we initialize
1468 // the OpenMP support.
1469 gmx_check_thread_affinity_set(mdlog, &hw_opt, hwinfo_->nthreads_hw_avail, FALSE);
1470 /* Check and update the number of OpenMP threads requested */
1471 checkAndUpdateRequestedNumOpenmpThreads(
1472 &hw_opt, *hwinfo_, cr, ms, physicalNodeComm.size_, pmeRunMode, mtop, *inputrec);
1474 gmx_omp_nthreads_init(mdlog,
1476 hwinfo_->nthreads_hw_avail,
1477 physicalNodeComm.size_,
1478 hw_opt.nthreads_omp,
1479 hw_opt.nthreads_omp_pme,
1480 !thisRankHasDuty(cr, DUTY_PP));
1482 const bool bEnableFPE = gmxShouldEnableFPExceptions();
1483 // FIXME - reconcile with gmx_feenableexcept() call from CommandLineModuleManager::run()
1486 gmx_feenableexcept();
1489 /* Now that we know the setup is consistent, check for efficiency */
1490 check_resource_division_efficiency(
1491 hwinfo_, gpuTaskAssignments.thisRankHasAnyGpuTask(), mdrunOptions.ntompOptionIsSet, cr, mdlog);
1493 /* getting number of PP/PME threads on this MPI / tMPI rank.
1494 PME: env variable should be read only on one node to make sure it is
1495 identical everywhere;
1497 const int numThreadsOnThisRank = thisRankHasDuty(cr, DUTY_PP) ? gmx_omp_nthreads_get(emntNonbonded)
1498 : gmx_omp_nthreads_get(emntPME);
1499 checkHardwareOversubscription(
1500 numThreadsOnThisRank, cr->nodeid, *hwinfo_->hardwareTopology, physicalNodeComm, mdlog);
1502 // Enable Peer access between GPUs where available
1503 // Only for DD, only master PP rank needs to perform setup, and only if thread MPI plus
1504 // any of the GPU communication features are active.
1505 if (DOMAINDECOMP(cr) && MASTER(cr) && thisRankHasDuty(cr, DUTY_PP) && GMX_THREAD_MPI
1506 && (runScheduleWork.simulationWork.useGpuHaloExchange
1507 || runScheduleWork.simulationWork.useGpuPmePpCommunication))
1509 setupGpuDevicePeerAccess(gpuIdsToUse, mdlog);
1512 if (hw_opt.threadAffinity != ThreadAffinity::Off)
1514 /* Before setting affinity, check whether the affinity has changed
1515 * - which indicates that probably the OpenMP library has changed it
1516 * since we first checked).
1518 gmx_check_thread_affinity_set(mdlog, &hw_opt, hwinfo_->nthreads_hw_avail, TRUE);
1520 int numThreadsOnThisNode, intraNodeThreadOffset;
1521 analyzeThreadsOnThisNode(
1522 physicalNodeComm, numThreadsOnThisRank, &numThreadsOnThisNode, &intraNodeThreadOffset);
1524 /* Set the CPU affinity */
1525 gmx_set_thread_affinity(mdlog,
1528 *hwinfo_->hardwareTopology,
1529 numThreadsOnThisRank,
1530 numThreadsOnThisNode,
1531 intraNodeThreadOffset,
1535 if (mdrunOptions.timingOptions.resetStep > -1)
1540 "The -resetstep functionality is deprecated, and may be removed in a "
1543 wcycle = wallcycle_init(fplog, mdrunOptions.timingOptions.resetStep, cr);
1547 /* Master synchronizes its value of reset_counters with all nodes
1548 * including PME only nodes */
1549 int64_t reset_counters = wcycle_get_reset_counters(wcycle);
1550 gmx_bcast(sizeof(reset_counters), &reset_counters, cr->mpi_comm_mysim);
1551 wcycle_set_reset_counters(wcycle, reset_counters);
1554 // Membrane embedding must be initialized before we call init_forcerec()
1555 membedHolder.initializeMembed(fplog,
1562 &mdrunOptions.checkpointOptions.period);
1564 const bool thisRankHasPmeGpuTask = gpuTaskAssignments.thisRankHasPmeGpuTask();
1565 std::unique_ptr<MDAtoms> mdAtoms;
1566 std::unique_ptr<VirtualSitesHandler> vsite;
1567 std::unique_ptr<GpuBonded> gpuBonded;
1570 if (thisRankHasDuty(cr, DUTY_PP))
1572 mdModulesNotifier.notify(*cr);
1573 mdModulesNotifier.notify(&atomSets);
1574 mdModulesNotifier.notify(mtop);
1575 mdModulesNotifier.notify(inputrec->pbcType);
1576 mdModulesNotifier.notify(SimulationTimeStep{ inputrec->delta_t });
1577 /* Initiate forcerecord */
1578 fr = new t_forcerec;
1579 fr->forceProviders = mdModules_->initForceProviders();
1580 init_forcerec(fplog,
1587 opt2fn("-table", filenames.size(), filenames.data()),
1588 opt2fn("-tablep", filenames.size(), filenames.data()),
1589 opt2fns("-tableb", filenames.size(), filenames.data()),
1591 // Dirty hack, for fixing disres and orires should be made mdmodules
1592 fr->fcdata->disres = disresdata;
1593 fr->fcdata->orires = oriresdata;
1595 // Save a handle to device stream manager to use elsewhere in the code
1596 // TODO: Forcerec is not a correct place to store it.
1597 fr->deviceStreamManager = deviceStreamManager.get();
1599 if (runScheduleWork.simulationWork.useGpuPmePpCommunication && !thisRankHasDuty(cr, DUTY_PME))
1602 deviceStreamManager != nullptr,
1603 "GPU device stream manager should be valid in order to use PME-PP direct "
1606 deviceStreamManager->streamIsValid(DeviceStreamType::PmePpTransfer),
1607 "GPU PP-PME stream should be valid in order to use GPU PME-PP direct "
1609 fr->pmePpCommGpu = std::make_unique<gmx::PmePpCommGpu>(
1612 deviceStreamManager->context(),
1613 deviceStreamManager->stream(DeviceStreamType::PmePpTransfer));
1616 fr->nbv = Nbnxm::init_nb_verlet(mdlog,
1621 runScheduleWork.simulationWork.useGpuNonbonded,
1622 deviceStreamManager.get(),
1626 // TODO: Move the logic below to a GPU bonded builder
1627 if (runScheduleWork.simulationWork.useGpuBonded)
1629 GMX_RELEASE_ASSERT(deviceStreamManager != nullptr,
1630 "GPU device stream manager should be valid in order to use GPU "
1631 "version of bonded forces.");
1632 gpuBonded = std::make_unique<GpuBonded>(
1634 fr->ic->epsfac * fr->fudgeQQ,
1635 deviceStreamManager->context(),
1636 deviceStreamManager->bondedStream(havePPDomainDecomposition(cr)),
1638 fr->gpuBonded = gpuBonded.get();
1641 /* Initialize the mdAtoms structure.
1642 * mdAtoms is not filled with atom data,
1643 * as this can not be done now with domain decomposition.
1645 mdAtoms = makeMDAtoms(fplog, mtop, *inputrec, thisRankHasPmeGpuTask);
1646 if (globalState && thisRankHasPmeGpuTask)
1648 // The pinning of coordinates in the global state object works, because we only use
1649 // PME on GPU without DD or on a separate PME rank, and because the local state pointer
1650 // points to the global state object without DD.
1651 // FIXME: MD and EM separately set up the local state - this should happen in the same
1652 // function, which should also perform the pinning.
1653 changePinningPolicy(&globalState->x, pme_get_pinning_policy());
1656 /* Initialize the virtual site communication */
1657 vsite = makeVirtualSitesHandler(mtop, cr, fr->pbcType);
1659 calc_shifts(box, fr->shift_vec);
1661 /* With periodic molecules the charge groups should be whole at start up
1662 * and the virtual sites should not be far from their proper positions.
1664 if (!inputrec->bContinuation && MASTER(cr)
1665 && !(inputrec->pbcType != PbcType::No && inputrec->bPeriodicMols))
1667 /* Make molecules whole at start of run */
1668 if (fr->pbcType != PbcType::No)
1670 do_pbc_first_mtop(fplog, inputrec->pbcType, box, &mtop, globalState->x.rvec_array());
1674 /* Correct initial vsite positions are required
1675 * for the initial distribution in the domain decomposition
1676 * and for the initial shell prediction.
1678 constructVirtualSitesGlobal(mtop, globalState->x);
1682 if (EEL_PME(fr->ic->eeltype) || EVDW_PME(fr->ic->vdwtype))
1684 ewaldcoeff_q = fr->ic->ewaldcoeff_q;
1685 ewaldcoeff_lj = fr->ic->ewaldcoeff_lj;
1690 /* This is a PME only node */
1692 GMX_ASSERT(globalState == nullptr,
1693 "We don't need the state on a PME only rank and expect it to be unitialized");
1695 ewaldcoeff_q = calc_ewaldcoeff_q(inputrec->rcoulomb, inputrec->ewald_rtol);
1696 ewaldcoeff_lj = calc_ewaldcoeff_lj(inputrec->rvdw, inputrec->ewald_rtol_lj);
1699 gmx_pme_t* sepPmeData = nullptr;
1700 // This reference hides the fact that PME data is owned by runner on PME-only ranks and by forcerec on other ranks
1701 GMX_ASSERT(thisRankHasDuty(cr, DUTY_PP) == (fr != nullptr),
1702 "Double-checking that only PME-only ranks have no forcerec");
1703 gmx_pme_t*& pmedata = fr ? fr->pmedata : sepPmeData;
1705 // TODO should live in ewald module once its testing is improved
1707 // Later, this program could contain kernels that might be later
1708 // re-used as auto-tuning progresses, or subsequent simulations
1710 PmeGpuProgramStorage pmeGpuProgram;
1711 if (thisRankHasPmeGpuTask)
1714 (deviceStreamManager != nullptr),
1715 "GPU device stream manager should be initialized in order to use GPU for PME.");
1716 GMX_RELEASE_ASSERT((deviceInfo != nullptr),
1717 "GPU device should be initialized in order to use GPU for PME.");
1718 pmeGpuProgram = buildPmeGpuProgram(deviceStreamManager->context());
1721 /* Initiate PME if necessary,
1722 * either on all nodes or on dedicated PME nodes only. */
1723 if (EEL_PME(inputrec->coulombtype) || EVDW_PME(inputrec->vdwtype))
1725 if (mdAtoms && mdAtoms->mdatoms())
1727 nChargePerturbed = mdAtoms->mdatoms()->nChargePerturbed;
1728 if (EVDW_PME(inputrec->vdwtype))
1730 nTypePerturbed = mdAtoms->mdatoms()->nTypePerturbed;
1733 if (cr->npmenodes > 0)
1735 /* The PME only nodes need to know nChargePerturbed(FEP on Q) and nTypePerturbed(FEP on LJ)*/
1736 gmx_bcast(sizeof(nChargePerturbed), &nChargePerturbed, cr->mpi_comm_mysim);
1737 gmx_bcast(sizeof(nTypePerturbed), &nTypePerturbed, cr->mpi_comm_mysim);
1740 if (thisRankHasDuty(cr, DUTY_PME))
1744 // TODO: This should be in the builder.
1745 GMX_RELEASE_ASSERT(!runScheduleWork.simulationWork.useGpuPme
1746 || (deviceStreamManager != nullptr),
1747 "Device stream manager should be valid in order to use GPU "
1750 !runScheduleWork.simulationWork.useGpuPme
1751 || deviceStreamManager->streamIsValid(DeviceStreamType::Pme),
1752 "GPU PME stream should be valid in order to use GPU version of PME.");
1754 const DeviceContext* deviceContext = runScheduleWork.simulationWork.useGpuPme
1755 ? &deviceStreamManager->context()
1757 const DeviceStream* pmeStream =
1758 runScheduleWork.simulationWork.useGpuPme
1759 ? &deviceStreamManager->stream(DeviceStreamType::Pme)
1762 pmedata = gmx_pme_init(cr,
1763 getNumPmeDomains(cr->dd),
1765 nChargePerturbed != 0,
1766 nTypePerturbed != 0,
1767 mdrunOptions.reproducible,
1770 gmx_omp_nthreads_get(emntPME),
1775 pmeGpuProgram.get(),
1778 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
1783 if (EI_DYNAMICS(inputrec->eI))
1785 /* Turn on signal handling on all nodes */
1787 * (A user signal from the PME nodes (if any)
1788 * is communicated to the PP nodes.
1790 signal_handler_install();
1793 pull_t* pull_work = nullptr;
1794 if (thisRankHasDuty(cr, DUTY_PP))
1796 /* Assumes uniform use of the number of OpenMP threads */
1797 walltime_accounting = walltime_accounting_init(gmx_omp_nthreads_get(emntDefault));
1799 if (inputrec->bPull)
1801 /* Initialize pull code */
1802 pull_work = init_pull(fplog,
1803 inputrec->pull.get(),
1808 inputrec->fepvals->init_lambda);
1809 if (inputrec->pull->bXOutAverage || inputrec->pull->bFOutAverage)
1811 initPullHistory(pull_work, &observablesHistory);
1813 if (EI_DYNAMICS(inputrec->eI) && MASTER(cr))
1815 init_pull_output_files(pull_work, filenames.size(), filenames.data(), oenv, startingBehavior);
1819 std::unique_ptr<EnforcedRotation> enforcedRotation;
1822 /* Initialize enforced rotation code */
1823 enforcedRotation = init_rot(fplog,
1836 t_swap* swap = nullptr;
1837 if (inputrec->eSwapCoords != eswapNO)
1839 /* Initialize ion swapping code */
1840 swap = init_swapcoords(fplog,
1842 opt2fn_master("-swap", filenames.size(), filenames.data(), cr),
1845 &observablesHistory,
1853 /* Let makeConstraints know whether we have essential dynamics constraints. */
1854 auto constr = makeConstraints(
1855 mtop, *inputrec, pull_work, doEssentialDynamics, fplog, cr, ms, &nrnb, wcycle, fr->bMolPBC);
1857 /* Energy terms and groups */
1858 gmx_enerdata_t enerd(mtop.groups.groups[SimulationAtomGroupType::EnergyOutput].size(),
1859 inputrec->fepvals->n_lambda);
1861 // cos acceleration is only supported by md, but older tpr
1862 // files might still combine it with other integrators
1863 GMX_RELEASE_ASSERT(inputrec->cos_accel == 0.0 || inputrec->eI == eiMD,
1864 "cos_acceleration is only supported by integrator=md");
1866 /* Kinetic energy data */
1867 gmx_ekindata_t ekind;
1868 init_ekindata(fplog, &(inputrec->opts), &ekind, inputrec->cos_accel);
1870 /* Set up interactive MD (IMD) */
1871 auto imdSession = makeImdSession(inputrec.get(),
1878 MASTER(cr) ? globalState->x.rvec_array() : nullptr,
1882 mdrunOptions.imdOptions,
1885 if (DOMAINDECOMP(cr))
1887 GMX_RELEASE_ASSERT(fr, "fr was NULL while cr->duty was DUTY_PP");
1888 /* This call is not included in init_domain_decomposition mainly
1889 * because fr->cginfo_mb is set later.
1891 dd_init_bondeds(fplog,
1896 domdecOptions.checkBondedInteractions,
1900 if (runScheduleWork.simulationWork.useGpuBufferOps)
1902 fr->gpuForceReduction[gmx::AtomLocality::Local] = std::make_unique<gmx::GpuForceReduction>(
1903 deviceStreamManager->context(),
1904 deviceStreamManager->stream(gmx::DeviceStreamType::NonBondedLocal),
1906 fr->gpuForceReduction[gmx::AtomLocality::NonLocal] = std::make_unique<gmx::GpuForceReduction>(
1907 deviceStreamManager->context(),
1908 deviceStreamManager->stream(gmx::DeviceStreamType::NonBondedNonLocal),
1912 std::unique_ptr<gmx::StatePropagatorDataGpu> stateGpu;
1913 if (gpusWereDetected
1914 && ((runScheduleWork.simulationWork.useGpuPme && thisRankHasDuty(cr, DUTY_PME))
1915 || runScheduleWork.simulationWork.useGpuBufferOps))
1917 GpuApiCallBehavior transferKind = (inputrec->eI == eiMD && !doRerun && !useModularSimulator)
1918 ? GpuApiCallBehavior::Async
1919 : GpuApiCallBehavior::Sync;
1920 GMX_RELEASE_ASSERT(deviceStreamManager != nullptr,
1921 "GPU device stream manager should be initialized to use GPU.");
1922 stateGpu = std::make_unique<gmx::StatePropagatorDataGpu>(
1923 *deviceStreamManager, transferKind, pme_gpu_get_block_size(fr->pmedata), wcycle);
1924 fr->stateGpu = stateGpu.get();
1927 GMX_ASSERT(stopHandlerBuilder_, "Runner must provide StopHandlerBuilder to simulator.");
1928 SimulatorBuilder simulatorBuilder;
1930 simulatorBuilder.add(SimulatorStateData(globalState.get(), &observablesHistory, &enerd, &ekind));
1931 simulatorBuilder.add(std::move(membedHolder));
1932 simulatorBuilder.add(std::move(stopHandlerBuilder_));
1933 simulatorBuilder.add(SimulatorConfig(mdrunOptions, startingBehavior, &runScheduleWork));
1936 simulatorBuilder.add(SimulatorEnv(fplog, cr, ms, mdlog, oenv));
1937 simulatorBuilder.add(Profiling(&nrnb, walltime_accounting, wcycle));
1938 simulatorBuilder.add(ConstraintsParam(
1939 constr.get(), enforcedRotation ? enforcedRotation->getLegacyEnfrot() : nullptr, vsite.get()));
1940 // TODO: Separate `fr` to a separate add, and make the `build` handle the coupling sensibly.
1941 simulatorBuilder.add(LegacyInput(
1942 static_cast<int>(filenames.size()), filenames.data(), inputrec.get(), fr));
1943 simulatorBuilder.add(ReplicaExchangeParameters(replExParams));
1944 simulatorBuilder.add(InteractiveMD(imdSession.get()));
1945 simulatorBuilder.add(SimulatorModules(mdModules_->outputProvider(), mdModules_->notifier()));
1946 simulatorBuilder.add(CenterOfMassPulling(pull_work));
1947 // Todo move to an MDModule
1948 simulatorBuilder.add(IonSwapping(swap));
1949 simulatorBuilder.add(TopologyData(&mtop, mdAtoms.get()));
1950 simulatorBuilder.add(BoxDeformationHandle(deform.get()));
1951 simulatorBuilder.add(std::move(modularSimulatorCheckpointData));
1953 // build and run simulator object based on user-input
1954 auto simulator = simulatorBuilder.build(useModularSimulator);
1957 if (fr->pmePpCommGpu)
1959 // destroy object since it is no longer required. (This needs to be done while the GPU context still exists.)
1960 fr->pmePpCommGpu.reset();
1963 if (inputrec->bPull)
1965 finish_pull(pull_work);
1967 finish_swapcoords(swap);
1971 GMX_RELEASE_ASSERT(pmedata, "pmedata was NULL while cr->duty was not DUTY_PP");
1973 walltime_accounting = walltime_accounting_init(gmx_omp_nthreads_get(emntPME));
1974 gmx_pmeonly(pmedata,
1978 walltime_accounting,
1981 deviceStreamManager.get());
1984 wallcycle_stop(wcycle, ewcRUN);
1986 /* Finish up, write some stuff
1987 * if rerunMD, don't write last frame again
1995 walltime_accounting,
1996 fr ? fr->nbv.get() : nullptr,
1998 EI_DYNAMICS(inputrec->eI) && !isMultiSim(ms));
2000 // clean up cycle counter
2001 wallcycle_destroy(wcycle);
2003 deviceStreamManager.reset(nullptr);
2007 gmx_pme_destroy(pmedata);
2011 // FIXME: this is only here to manually unpin mdAtoms->chargeA_ and state->x,
2012 // before we destroy the GPU context(s)
2013 // Pinned buffers are associated with contexts in CUDA.
2014 // As soon as we destroy GPU contexts after mdrunner() exits, these lines should go.
2015 mdAtoms.reset(nullptr);
2016 globalState.reset(nullptr);
2017 mdModules_.reset(nullptr); // destruct force providers here as they might also use the GPU
2018 gpuBonded.reset(nullptr);
2019 /* Free pinned buffers in *fr */
2022 // TODO convert to C++ so we can get rid of these frees
2026 if (!hwinfo_->deviceInfoList.empty())
2028 /* stop the GPU profiler (only CUDA) */
2032 /* With tMPI we need to wait for all ranks to finish deallocation before
2033 * destroying the CUDA context as some tMPI ranks may be sharing
2036 * This is not a concern in OpenCL where we use one context per rank.
2038 * Note: it is safe to not call the barrier on the ranks which do not use GPU,
2039 * but it is easier and more futureproof to call it on the whole node.
2041 * Note that this function needs to be called even if GPUs are not used
2042 * in this run because the PME ranks have no knowledge of whether GPUs
2043 * are used or not, but all ranks need to enter the barrier below.
2044 * \todo Remove this physical node barrier after making sure
2045 * that it's not needed anymore (with a shared GPU run).
2049 physicalNodeComm.barrier();
2051 releaseDevice(deviceInfo);
2053 /* Does what it says */
2054 print_date_and_time(fplog, cr->nodeid, "Finished mdrun", gmx_gettime());
2055 walltime_accounting_destroy(walltime_accounting);
2057 // Ensure log file content is written
2060 gmx_fio_flush(logFileHandle);
2063 /* Reset FPEs (important for unit tests) by disabling them. Assumes no
2064 * exceptions were enabled before function was called. */
2067 gmx_fedisableexcept();
2070 auto rc = static_cast<int>(gmx_get_stop_condition());
2073 /* we need to join all threads. The sub-threads join when they
2074 exit this function, but the master thread needs to be told to
2084 Mdrunner::~Mdrunner()
2086 // Clean up of the Manager.
2087 // This will end up getting called on every thread-MPI rank, which is unnecessary,
2088 // but okay as long as threads synchronize some time before adding or accessing
2089 // a new set of restraints.
2090 if (restraintManager_)
2092 restraintManager_->clear();
2093 GMX_ASSERT(restraintManager_->countRestraints() == 0,
2094 "restraints added during runner life time should be cleared at runner "
2099 void Mdrunner::addPotential(std::shared_ptr<gmx::IRestraintPotential> puller, const std::string& name)
2101 GMX_ASSERT(restraintManager_, "Mdrunner must have a restraint manager.");
2102 // Not sure if this should be logged through the md logger or something else,
2103 // but it is helpful to have some sort of INFO level message sent somewhere.
2104 // std::cout << "Registering restraint named " << name << std::endl;
2106 // When multiple restraints are used, it may be wasteful to register them separately.
2107 // Maybe instead register an entire Restraint Manager as a force provider.
2108 restraintManager_->addToSpec(std::move(puller), name);
2111 Mdrunner::Mdrunner(std::unique_ptr<MDModules> mdModules) : mdModules_(std::move(mdModules)) {}
2113 Mdrunner::Mdrunner(Mdrunner&&) noexcept = default;
2115 //NOLINTNEXTLINE(performance-noexcept-move-constructor) working around GCC bug 58265 in CentOS 7
2116 Mdrunner& Mdrunner::operator=(Mdrunner&& /*handle*/) noexcept(BUGFREE_NOEXCEPT_STRING) = default;
2118 class Mdrunner::BuilderImplementation
2121 BuilderImplementation() = delete;
2122 BuilderImplementation(std::unique_ptr<MDModules> mdModules, compat::not_null<SimulationContext*> context);
2123 ~BuilderImplementation();
2125 BuilderImplementation& setExtraMdrunOptions(const MdrunOptions& options,
2126 real forceWarningThreshold,
2127 StartingBehavior startingBehavior);
2129 void addHardwareDetectionResult(const gmx_hw_info_t* hwinfo);
2131 void addDomdec(const DomdecOptions& options);
2133 void addInput(SimulationInputHandle inputHolder);
2135 void addVerletList(int nstlist);
2137 void addReplicaExchange(const ReplicaExchangeParameters& params);
2139 void addNonBonded(const char* nbpu_opt);
2141 void addPME(const char* pme_opt_, const char* pme_fft_opt_);
2143 void addBondedTaskAssignment(const char* bonded_opt);
2145 void addUpdateTaskAssignment(const char* update_opt);
2147 void addHardwareOptions(const gmx_hw_opt_t& hardwareOptions);
2149 void addFilenames(ArrayRef<const t_filenm> filenames);
2151 void addOutputEnvironment(gmx_output_env_t* outputEnvironment);
2153 void addLogFile(t_fileio* logFileHandle);
2155 void addStopHandlerBuilder(std::unique_ptr<StopHandlerBuilder> builder);
2160 // Default parameters copied from runner.h
2161 // \todo Clarify source(s) of default parameters.
2163 const char* nbpu_opt_ = nullptr;
2164 const char* pme_opt_ = nullptr;
2165 const char* pme_fft_opt_ = nullptr;
2166 const char* bonded_opt_ = nullptr;
2167 const char* update_opt_ = nullptr;
2169 MdrunOptions mdrunOptions_;
2171 DomdecOptions domdecOptions_;
2173 ReplicaExchangeParameters replicaExchangeParameters_;
2175 //! Command-line override for the duration of a neighbor list with the Verlet scheme.
2178 //! World communicator, used for hardware detection and task assignment
2179 MPI_Comm libraryWorldCommunicator_ = MPI_COMM_NULL;
2181 //! Multisim communicator handle.
2182 gmx_multisim_t* multiSimulation_;
2184 //! mdrun communicator
2185 MPI_Comm simulationCommunicator_ = MPI_COMM_NULL;
2187 //! Print a warning if any force is larger than this (in kJ/mol nm).
2188 real forceWarningThreshold_ = -1;
2190 //! Whether the simulation will start afresh, or restart with/without appending.
2191 StartingBehavior startingBehavior_ = StartingBehavior::NewSimulation;
2193 //! The modules that comprise the functionality of mdrun.
2194 std::unique_ptr<MDModules> mdModules_;
2196 //! Detected hardware.
2197 const gmx_hw_info_t* hwinfo_ = nullptr;
2199 //! \brief Parallelism information.
2200 gmx_hw_opt_t hardwareOptions_;
2202 //! filename options for simulation.
2203 ArrayRef<const t_filenm> filenames_;
2205 /*! \brief Handle to output environment.
2207 * \todo gmx_output_env_t needs lifetime management.
2209 gmx_output_env_t* outputEnvironment_ = nullptr;
2211 /*! \brief Non-owning handle to MD log file.
2213 * \todo Context should own output facilities for client.
2214 * \todo Improve log file handle management.
2216 * Code managing the FILE* relies on the ability to set it to
2217 * nullptr to check whether the filehandle is valid.
2219 t_fileio* logFileHandle_ = nullptr;
2222 * \brief Builder for simulation stop signal handler.
2224 std::unique_ptr<StopHandlerBuilder> stopHandlerBuilder_ = nullptr;
2227 * \brief Sources for initial simulation state.
2229 * See issue #3652 for near-term refinements to the SimulationInput interface.
2231 * See issue #3379 for broader discussion on API aspects of simulation inputs and outputs.
2233 SimulationInputHandle inputHolder_;
2236 Mdrunner::BuilderImplementation::BuilderImplementation(std::unique_ptr<MDModules> mdModules,
2237 compat::not_null<SimulationContext*> context) :
2238 mdModules_(std::move(mdModules))
2240 libraryWorldCommunicator_ = context->libraryWorldCommunicator_;
2241 simulationCommunicator_ = context->simulationCommunicator_;
2242 multiSimulation_ = context->multiSimulation_.get();
2245 Mdrunner::BuilderImplementation::~BuilderImplementation() = default;
2247 Mdrunner::BuilderImplementation&
2248 Mdrunner::BuilderImplementation::setExtraMdrunOptions(const MdrunOptions& options,
2249 const real forceWarningThreshold,
2250 const StartingBehavior startingBehavior)
2252 mdrunOptions_ = options;
2253 forceWarningThreshold_ = forceWarningThreshold;
2254 startingBehavior_ = startingBehavior;
2258 void Mdrunner::BuilderImplementation::addDomdec(const DomdecOptions& options)
2260 domdecOptions_ = options;
2263 void Mdrunner::BuilderImplementation::addVerletList(int nstlist)
2268 void Mdrunner::BuilderImplementation::addReplicaExchange(const ReplicaExchangeParameters& params)
2270 replicaExchangeParameters_ = params;
2273 Mdrunner Mdrunner::BuilderImplementation::build()
2275 auto newRunner = Mdrunner(std::move(mdModules_));
2277 newRunner.mdrunOptions = mdrunOptions_;
2278 newRunner.pforce = forceWarningThreshold_;
2279 newRunner.startingBehavior = startingBehavior_;
2280 newRunner.domdecOptions = domdecOptions_;
2282 // \todo determine an invariant to check or confirm that all gmx_hw_opt_t objects are valid
2283 newRunner.hw_opt = hardwareOptions_;
2285 // No invariant to check. This parameter exists to optionally override other behavior.
2286 newRunner.nstlist_cmdline = nstlist_;
2288 newRunner.replExParams = replicaExchangeParameters_;
2290 newRunner.filenames = filenames_;
2292 newRunner.libraryWorldCommunicator = libraryWorldCommunicator_;
2294 newRunner.simulationCommunicator = simulationCommunicator_;
2296 // nullptr is a valid value for the multisim handle
2297 newRunner.ms = multiSimulation_;
2301 newRunner.hwinfo_ = hwinfo_;
2305 GMX_THROW(gmx::APIError(
2306 "MdrunnerBuilder::addHardwareDetectionResult() is required before build()"));
2311 newRunner.inputHolder_ = std::move(inputHolder_);
2315 GMX_THROW(gmx::APIError("MdrunnerBuilder::addInput() is required before build()."));
2318 // \todo Clarify ownership and lifetime management for gmx_output_env_t
2319 // \todo Update sanity checking when output environment has clearly specified invariants.
2320 // Initialization and default values for oenv are not well specified in the current version.
2321 if (outputEnvironment_)
2323 newRunner.oenv = outputEnvironment_;
2327 GMX_THROW(gmx::APIError(
2328 "MdrunnerBuilder::addOutputEnvironment() is required before build()"));
2331 newRunner.logFileHandle = logFileHandle_;
2335 newRunner.nbpu_opt = nbpu_opt_;
2339 GMX_THROW(gmx::APIError("MdrunnerBuilder::addNonBonded() is required before build()"));
2342 if (pme_opt_ && pme_fft_opt_)
2344 newRunner.pme_opt = pme_opt_;
2345 newRunner.pme_fft_opt = pme_fft_opt_;
2349 GMX_THROW(gmx::APIError("MdrunnerBuilder::addElectrostatics() is required before build()"));
2354 newRunner.bonded_opt = bonded_opt_;
2358 GMX_THROW(gmx::APIError(
2359 "MdrunnerBuilder::addBondedTaskAssignment() is required before build()"));
2364 newRunner.update_opt = update_opt_;
2368 GMX_THROW(gmx::APIError(
2369 "MdrunnerBuilder::addUpdateTaskAssignment() is required before build() "));
2373 newRunner.restraintManager_ = std::make_unique<gmx::RestraintManager>();
2375 if (stopHandlerBuilder_)
2377 newRunner.stopHandlerBuilder_ = std::move(stopHandlerBuilder_);
2381 newRunner.stopHandlerBuilder_ = std::make_unique<StopHandlerBuilder>();
2387 void Mdrunner::BuilderImplementation::addHardwareDetectionResult(const gmx_hw_info_t* hwinfo)
2392 void Mdrunner::BuilderImplementation::addNonBonded(const char* nbpu_opt)
2394 nbpu_opt_ = nbpu_opt;
2397 void Mdrunner::BuilderImplementation::addPME(const char* pme_opt, const char* pme_fft_opt)
2400 pme_fft_opt_ = pme_fft_opt;
2403 void Mdrunner::BuilderImplementation::addBondedTaskAssignment(const char* bonded_opt)
2405 bonded_opt_ = bonded_opt;
2408 void Mdrunner::BuilderImplementation::addUpdateTaskAssignment(const char* update_opt)
2410 update_opt_ = update_opt;
2413 void Mdrunner::BuilderImplementation::addHardwareOptions(const gmx_hw_opt_t& hardwareOptions)
2415 hardwareOptions_ = hardwareOptions;
2418 void Mdrunner::BuilderImplementation::addFilenames(ArrayRef<const t_filenm> filenames)
2420 filenames_ = filenames;
2423 void Mdrunner::BuilderImplementation::addOutputEnvironment(gmx_output_env_t* outputEnvironment)
2425 outputEnvironment_ = outputEnvironment;
2428 void Mdrunner::BuilderImplementation::addLogFile(t_fileio* logFileHandle)
2430 logFileHandle_ = logFileHandle;
2433 void Mdrunner::BuilderImplementation::addStopHandlerBuilder(std::unique_ptr<StopHandlerBuilder> builder)
2435 stopHandlerBuilder_ = std::move(builder);
2438 void Mdrunner::BuilderImplementation::addInput(SimulationInputHandle inputHolder)
2440 inputHolder_ = std::move(inputHolder);
2443 MdrunnerBuilder::MdrunnerBuilder(std::unique_ptr<MDModules> mdModules,
2444 compat::not_null<SimulationContext*> context) :
2445 impl_{ std::make_unique<Mdrunner::BuilderImplementation>(std::move(mdModules), context) }
2449 MdrunnerBuilder::~MdrunnerBuilder() = default;
2451 MdrunnerBuilder& MdrunnerBuilder::addHardwareDetectionResult(const gmx_hw_info_t* hwinfo)
2453 impl_->addHardwareDetectionResult(hwinfo);
2457 MdrunnerBuilder& MdrunnerBuilder::addSimulationMethod(const MdrunOptions& options,
2458 real forceWarningThreshold,
2459 const StartingBehavior startingBehavior)
2461 impl_->setExtraMdrunOptions(options, forceWarningThreshold, startingBehavior);
2465 MdrunnerBuilder& MdrunnerBuilder::addDomainDecomposition(const DomdecOptions& options)
2467 impl_->addDomdec(options);
2471 MdrunnerBuilder& MdrunnerBuilder::addNeighborList(int nstlist)
2473 impl_->addVerletList(nstlist);
2477 MdrunnerBuilder& MdrunnerBuilder::addReplicaExchange(const ReplicaExchangeParameters& params)
2479 impl_->addReplicaExchange(params);
2483 MdrunnerBuilder& MdrunnerBuilder::addNonBonded(const char* nbpu_opt)
2485 impl_->addNonBonded(nbpu_opt);
2489 MdrunnerBuilder& MdrunnerBuilder::addElectrostatics(const char* pme_opt, const char* pme_fft_opt)
2491 // The builder method may become more general in the future, but in this version,
2492 // parameters for PME electrostatics are both required and the only parameters
2494 if (pme_opt && pme_fft_opt)
2496 impl_->addPME(pme_opt, pme_fft_opt);
2501 gmx::InvalidInputError("addElectrostatics() arguments must be non-null pointers."));
2506 MdrunnerBuilder& MdrunnerBuilder::addBondedTaskAssignment(const char* bonded_opt)
2508 impl_->addBondedTaskAssignment(bonded_opt);
2512 MdrunnerBuilder& MdrunnerBuilder::addUpdateTaskAssignment(const char* update_opt)
2514 impl_->addUpdateTaskAssignment(update_opt);
2518 Mdrunner MdrunnerBuilder::build()
2520 return impl_->build();
2523 MdrunnerBuilder& MdrunnerBuilder::addHardwareOptions(const gmx_hw_opt_t& hardwareOptions)
2525 impl_->addHardwareOptions(hardwareOptions);
2529 MdrunnerBuilder& MdrunnerBuilder::addFilenames(ArrayRef<const t_filenm> filenames)
2531 impl_->addFilenames(filenames);
2535 MdrunnerBuilder& MdrunnerBuilder::addOutputEnvironment(gmx_output_env_t* outputEnvironment)
2537 impl_->addOutputEnvironment(outputEnvironment);
2541 MdrunnerBuilder& MdrunnerBuilder::addLogFile(t_fileio* logFileHandle)
2543 impl_->addLogFile(logFileHandle);
2547 MdrunnerBuilder& MdrunnerBuilder::addStopHandlerBuilder(std::unique_ptr<StopHandlerBuilder> builder)
2549 impl_->addStopHandlerBuilder(std::move(builder));
2553 MdrunnerBuilder& MdrunnerBuilder::addInput(SimulationInputHandle input)
2555 impl_->addInput(std::move(input));
2559 MdrunnerBuilder::MdrunnerBuilder(MdrunnerBuilder&&) noexcept = default;
2561 MdrunnerBuilder& MdrunnerBuilder::operator=(MdrunnerBuilder&&) noexcept = default;