2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2014,2015,2016,2017,2018,2019, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
37 * \brief This file defines high-level functions for mdrun to compute
38 * energies and forces for listed interactions.
40 * \author Mark Abraham <mark.j.abraham@gmail.com>
42 * \ingroup module_listed_forces
46 #include "listed_forces.h"
53 #include "gromacs/gmxlib/network.h"
54 #include "gromacs/gmxlib/nrnb.h"
55 #include "gromacs/listed_forces/bonded.h"
56 #include "gromacs/listed_forces/disre.h"
57 #include "gromacs/listed_forces/orires.h"
58 #include "gromacs/listed_forces/pairs.h"
59 #include "gromacs/listed_forces/position_restraints.h"
60 #include "gromacs/math/vec.h"
61 #include "gromacs/mdlib/enerdata_utils.h"
62 #include "gromacs/mdlib/force.h"
63 #include "gromacs/mdtypes/commrec.h"
64 #include "gromacs/mdtypes/fcdata.h"
65 #include "gromacs/mdtypes/forcerec.h"
66 #include "gromacs/mdtypes/inputrec.h"
67 #include "gromacs/mdtypes/md_enums.h"
68 #include "gromacs/mdtypes/simulation_workload.h"
69 #include "gromacs/pbcutil/ishift.h"
70 #include "gromacs/pbcutil/pbc.h"
71 #include "gromacs/timing/wallcycle.h"
72 #include "gromacs/topology/topology.h"
73 #include "gromacs/utility/exceptions.h"
74 #include "gromacs/utility/fatalerror.h"
75 #include "gromacs/utility/smalloc.h"
77 #include "listed_internal.h"
78 #include "utilities.h"
85 /*! \brief Return true if ftype is an explicit pair-listed LJ or
86 * COULOMB interaction type: bonded LJ (usually 1-4), or special
87 * listed non-bonded for FEP. */
88 bool isPairInteraction(int ftype)
90 return ((ftype) >= F_LJ14 && (ftype) <= F_LJC_PAIRS_NB);
93 /*! \brief Zero thread-local output buffers */
94 void zero_thread_output(f_thread_t* f_t)
96 constexpr int nelem_fa = sizeof(f_t->f[0]) / sizeof(real);
98 for (int i = 0; i < f_t->nblock_used; i++)
100 int a0 = f_t->block_index[i] * reduction_block_size;
101 int a1 = a0 + reduction_block_size;
102 for (int a = a0; a < a1; a++)
104 for (int d = 0; d < nelem_fa; d++)
111 for (int i = 0; i < SHIFTS; i++)
113 clear_rvec(f_t->fshift[i]);
115 for (int i = 0; i < F_NRE; i++)
119 for (int i = 0; i < egNR; i++)
121 for (int j = 0; j < f_t->grpp.nener; j++)
123 f_t->grpp.ener[i][j] = 0;
126 for (int i = 0; i < efptNR; i++)
132 /*! \brief The max thread number is arbitrary, we used a fixed number
133 * to avoid memory management. Using more than 16 threads is probably
134 * never useful performance wise. */
135 #define MAX_BONDED_THREADS 256
137 /*! \brief Reduce thread-local force buffers */
138 void reduce_thread_forces(int n, gmx::ArrayRef<gmx::RVec> force, const bonded_threading_t* bt, int nthreads)
140 if (nthreads > MAX_BONDED_THREADS)
142 gmx_fatal(FARGS, "Can not reduce bonded forces on more than %d threads", MAX_BONDED_THREADS);
145 rvec* gmx_restrict f = as_rvec_array(force.data());
147 /* This reduction can run on any number of threads,
148 * independently of bt->nthreads.
149 * But if nthreads matches bt->nthreads (which it currently does)
150 * the uniform distribution of the touched blocks over nthreads will
151 * match the distribution of bonded over threads well in most cases,
152 * which means that threads mostly reduce their own data which increases
153 * the number of cache hits.
155 #pragma omp parallel for num_threads(nthreads) schedule(static)
156 for (int b = 0; b < bt->nblock_used; b++)
160 int ind = bt->block_index[b];
161 rvec4* fp[MAX_BONDED_THREADS];
163 /* Determine which threads contribute to this block */
165 for (int ft = 0; ft < bt->nthreads; ft++)
167 if (bitmask_is_set(bt->mask[ind], ft))
169 fp[nfb++] = bt->f_t[ft]->f;
174 /* Reduce force buffers for threads that contribute */
175 int a0 = ind * reduction_block_size;
176 int a1 = (ind + 1) * reduction_block_size;
177 /* It would be nice if we could pad f to avoid this min */
178 a1 = std::min(a1, n);
179 for (int a = a0; a < a1; a++)
181 for (int fb = 0; fb < nfb; fb++)
183 rvec_inc(f[a], fp[fb][a]);
188 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
192 /*! \brief Reduce thread-local forces, shift forces and energies */
193 void reduce_thread_output(int n,
194 gmx::ForceWithShiftForces* forceWithShiftForces,
196 gmx_grppairener_t* grpp,
198 const bonded_threading_t* bt,
199 const gmx::StepWorkload& stepWork)
201 assert(bt->haveBondeds);
203 if (bt->nblock_used > 0)
205 /* Reduce the bonded force buffer */
206 reduce_thread_forces(n, forceWithShiftForces->force(), bt, bt->nthreads);
209 rvec* gmx_restrict fshift = as_rvec_array(forceWithShiftForces->shiftForces().data());
211 /* When necessary, reduce energy and virial using one thread only */
212 if ((stepWork.computeEnergy || stepWork.computeVirial || stepWork.computeDhdl) && bt->nthreads > 1)
214 gmx::ArrayRef<const std::unique_ptr<f_thread_t>> f_t = bt->f_t;
216 if (stepWork.computeVirial)
218 for (int i = 0; i < SHIFTS; i++)
220 for (int t = 1; t < bt->nthreads; t++)
222 rvec_inc(fshift[i], f_t[t]->fshift[i]);
226 if (stepWork.computeEnergy)
228 for (int i = 0; i < F_NRE; i++)
230 for (int t = 1; t < bt->nthreads; t++)
232 ener[i] += f_t[t]->ener[i];
235 for (int i = 0; i < egNR; i++)
237 for (int j = 0; j < f_t[1]->grpp.nener; j++)
239 for (int t = 1; t < bt->nthreads; t++)
241 grpp->ener[i][j] += f_t[t]->grpp.ener[i][j];
246 if (stepWork.computeDhdl)
248 for (int i = 0; i < efptNR; i++)
251 for (int t = 1; t < bt->nthreads; t++)
253 dvdl[i] += f_t[t]->dvdl[i];
260 /*! \brief Returns the bonded kernel flavor
262 * Note that energies are always requested when the virial
263 * is requested (performance gain would be small).
264 * Note that currently we do not have bonded kernels that
265 * do not compute forces.
267 BondedKernelFlavor selectBondedKernelFlavor(const gmx::StepWorkload& stepWork,
268 const bool useSimdKernels,
269 const bool havePerturbedInteractions)
271 BondedKernelFlavor flavor;
272 if (stepWork.computeEnergy || stepWork.computeVirial)
274 if (stepWork.computeVirial)
276 flavor = BondedKernelFlavor::ForcesAndVirialAndEnergy;
280 flavor = BondedKernelFlavor::ForcesAndEnergy;
285 if (useSimdKernels && !havePerturbedInteractions)
287 flavor = BondedKernelFlavor::ForcesSimdWhenAvailable;
291 flavor = BondedKernelFlavor::ForcesNoSimd;
298 /*! \brief Calculate one element of the list of bonded interactions
300 real calc_one_bond(int thread,
303 ArrayRef<const int> iatoms,
304 const int numNonperturbedInteractions,
305 const WorkDivision& workDivision,
309 const t_forcerec* fr,
312 gmx_grppairener_t* grpp,
318 const gmx::StepWorkload& stepWork,
319 int* global_atom_index)
321 GMX_ASSERT(idef->ilsort == ilsortNO_FE || idef->ilsort == ilsortFE_SORTED,
322 "The topology should be marked either as no FE or sorted on FE");
324 const bool havePerturbedInteractions =
325 (idef->ilsort == ilsortFE_SORTED && numNonperturbedInteractions < iatoms.ssize());
326 BondedKernelFlavor flavor =
327 selectBondedKernelFlavor(stepWork, fr->use_simd_kernels, havePerturbedInteractions);
329 if (IS_RESTRAINT_TYPE(ftype))
331 efptFTYPE = efptRESTRAINT;
335 efptFTYPE = efptBONDED;
338 const int nat1 = interaction_function[ftype].nratoms + 1;
339 const int nbonds = iatoms.ssize() / nat1;
341 GMX_ASSERT(fr->gpuBonded != nullptr || workDivision.end(ftype) == iatoms.ssize(),
342 "The thread division should match the topology");
344 const int nb0 = workDivision.bound(ftype, thread);
345 const int nbn = workDivision.bound(ftype, thread + 1) - nb0;
348 if (!isPairInteraction(ftype))
352 /* TODO The execution time for CMAP dihedrals might be
353 nice to account to its own subtimer, but first
354 wallcycle needs to be extended to support calling from
356 v = cmap_dihs(nbn, iatoms.data() + nb0, idef->iparams, idef->cmap_grid, x, f, fshift,
357 pbc, g, lambda[efptFTYPE], &(dvdl[efptFTYPE]), md, fcd, global_atom_index);
361 v = calculateSimpleBond(ftype, nbn, iatoms.data() + nb0, idef->iparams, x, f, fshift,
362 pbc, g, lambda[efptFTYPE], &(dvdl[efptFTYPE]), md, fcd,
363 global_atom_index, flavor);
368 /* TODO The execution time for pairs might be nice to account
369 to its own subtimer, but first wallcycle needs to be
370 extended to support calling from multiple threads. */
371 do_pairs(ftype, nbn, iatoms.data() + nb0, idef->iparams, x, f, fshift, pbc, g, lambda, dvdl,
372 md, fr, havePerturbedInteractions, stepWork, grpp, global_atom_index);
377 inc_nrnb(nrnb, nrnbIndex(ftype), nbonds);
385 /*! \brief Compute the bonded part of the listed forces, parallelized over threads
387 static void calcBondedForces(const t_idef* idef,
389 const t_forcerec* fr,
390 const t_pbc* pbc_null,
392 rvec* fshiftMasterBuffer,
393 gmx_enerdata_t* enerd,
399 const gmx::StepWorkload& stepWork,
400 int* global_atom_index)
402 bonded_threading_t* bt = fr->bondedThreading;
404 #pragma omp parallel for num_threads(bt->nthreads) schedule(static)
405 for (int thread = 0; thread < bt->nthreads; thread++)
409 f_thread_t& threadBuffers = *bt->f_t[thread];
415 gmx_grppairener_t* grpp;
417 zero_thread_output(&threadBuffers);
419 rvec4* ft = threadBuffers.f;
421 /* Thread 0 writes directly to the main output buffers.
422 * We might want to reconsider this.
426 fshift = fshiftMasterBuffer;
433 fshift = threadBuffers.fshift;
434 epot = threadBuffers.ener;
435 grpp = &threadBuffers.grpp;
436 dvdlt = threadBuffers.dvdl;
438 /* Loop over all bonded force types to calculate the bonded forces */
439 for (ftype = 0; (ftype < F_NRE); ftype++)
441 const t_ilist& ilist = idef->il[ftype];
442 if (ilist.nr > 0 && ftype_is_bonded_potential(ftype))
444 ArrayRef<const int> iatoms = gmx::constArrayRefFromArray(ilist.iatoms, ilist.nr);
446 thread, ftype, idef, iatoms, idef->numNonperturbedInteractions[ftype],
447 fr->bondedThreading->workDivision, x, ft, fshift, fr, pbc_null, g, grpp,
448 nrnb, lambda, dvdlt, md, fcd, stepWork, global_atom_index);
453 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
457 bool haveRestraints(const t_idef& idef, const t_fcdata& fcd)
459 return ((idef.il[F_POSRES].nr > 0) || (idef.il[F_FBPOSRES].nr > 0) || fcd.orires.nr > 0
460 || fcd.disres.nres > 0);
463 bool haveCpuBondeds(const t_forcerec& fr)
465 return fr.bondedThreading->haveBondeds;
468 bool haveCpuListedForces(const t_forcerec& fr, const t_idef& idef, const t_fcdata& fcd)
470 return haveCpuBondeds(fr) || haveRestraints(idef, fcd);
473 void calc_listed(const t_commrec* cr,
474 const gmx_multisim_t* ms,
475 struct gmx_wallcycle* wcycle,
479 gmx::ForceOutputs* forceOutputs,
480 const t_forcerec* fr,
481 const struct t_pbc* pbc,
482 const struct t_pbc* pbc_full,
483 const struct t_graph* g,
484 gmx_enerdata_t* enerd,
489 int* global_atom_index,
490 const gmx::StepWorkload& stepWork)
492 const t_pbc* pbc_null;
493 bonded_threading_t* bt = fr->bondedThreading;
504 if (haveRestraints(*idef, *fcd))
506 /* TODO Use of restraints triggers further function calls
507 inside the loop over calc_one_bond(), but those are too
508 awkward to account to this subtimer properly in the present
509 code. We don't test / care much about performance with
510 restraints, anyway. */
511 wallcycle_sub_start(wcycle, ewcsRESTRAINTS);
513 if (idef->il[F_POSRES].nr > 0)
515 posres_wrapper(nrnb, idef, pbc_full, x, enerd, lambda, fr, &forceOutputs->forceWithVirial());
518 if (idef->il[F_FBPOSRES].nr > 0)
520 fbposres_wrapper(nrnb, idef, pbc_full, x, enerd, fr, &forceOutputs->forceWithVirial());
523 /* Do pre force calculation stuff which might require communication */
524 if (fcd->orires.nr > 0)
526 /* This assertion is to ensure we have whole molecules.
527 * Unfortunately we do not have an mdrun state variable that tells
528 * us if molecules in x are not broken over PBC, so we have to make
529 * do with checking graph!=nullptr, which should tell us if we made
530 * molecules whole before calling the current function.
532 GMX_RELEASE_ASSERT(fr->ePBC == epbcNONE || g != nullptr,
533 "With orientation restraints molecules should be whole");
534 enerd->term[F_ORIRESDEV] = calc_orires_dev(ms, idef->il[F_ORIRES].nr, idef->il[F_ORIRES].iatoms,
535 idef->iparams, md, x, pbc_null, fcd, hist);
537 if (fcd->disres.nres > 0)
539 calc_disres_R_6(cr, ms, idef->il[F_DISRES].nr, idef->il[F_DISRES].iatoms, x, pbc_null,
543 wallcycle_sub_stop(wcycle, ewcsRESTRAINTS);
546 if (haveCpuBondeds(*fr))
548 gmx::ForceWithShiftForces& forceWithShiftForces = forceOutputs->forceWithShiftForces();
550 wallcycle_sub_start(wcycle, ewcsLISTED);
551 /* The dummy array is to have a place to store the dhdl at other values
552 of lambda, which will be thrown away in the end */
553 real dvdl[efptNR] = { 0 };
554 calcBondedForces(idef, x, fr, pbc_null, g,
555 as_rvec_array(forceWithShiftForces.shiftForces().data()), enerd, nrnb,
556 lambda, dvdl, md, fcd, stepWork, global_atom_index);
557 wallcycle_sub_stop(wcycle, ewcsLISTED);
559 wallcycle_sub_start(wcycle, ewcsLISTED_BUF_OPS);
560 reduce_thread_output(fr->natoms_force, &forceWithShiftForces, enerd->term, &enerd->grpp,
563 if (stepWork.computeDhdl)
565 for (int i = 0; i < efptNR; i++)
567 enerd->dvdl_nonlin[i] += dvdl[i];
570 wallcycle_sub_stop(wcycle, ewcsLISTED_BUF_OPS);
573 /* Copy the sum of violations for the distance restraints from fcd */
576 enerd->term[F_DISRESVIOL] = fcd->disres.sumviol;
580 void calc_listed_lambda(const t_idef* idef,
582 const t_forcerec* fr,
583 const struct t_pbc* pbc,
584 const struct t_graph* g,
585 gmx_grppairener_t* grpp,
591 int* global_atom_index)
594 real dvdl_dum[efptNR] = { 0 };
597 const t_pbc* pbc_null;
599 WorkDivision& workDivision = fr->bondedThreading->foreignLambdaWorkDivision;
610 /* Copy the whole idef, so we can modify the contents locally */
613 /* We already have the forces, so we use temp buffers here */
614 // TODO: Get rid of these allocations by using permanent force buffers
615 snew(f, fr->natoms_force);
616 snew(fshift, SHIFTS);
618 /* Loop over all bonded force types to calculate the bonded energies */
619 for (int ftype = 0; (ftype < F_NRE); ftype++)
621 if (ftype_is_bonded_potential(ftype))
623 const t_ilist& ilist = idef->il[ftype];
624 /* Create a temporary iatom list with only perturbed interactions */
625 const int numNonperturbed = idef->numNonperturbedInteractions[ftype];
626 ArrayRef<const int> iatoms = gmx::constArrayRefFromArray(ilist.iatoms + numNonperturbed,
627 ilist.nr - numNonperturbed);
628 t_ilist& ilist_fe = idef_fe.il[ftype];
629 /* Set the work range of thread 0 to the perturbed bondeds */
630 workDivision.setBound(ftype, 0, 0);
631 workDivision.setBound(ftype, 1, iatoms.ssize());
635 gmx::StepWorkload tempFlags;
636 tempFlags.computeEnergy = true;
637 v = calc_one_bond(0, ftype, idef, iatoms, iatoms.ssize(), workDivision, x, f,
638 fshift, fr, pbc_null, g, grpp, nrnb, lambda, dvdl_dum, md, fcd,
639 tempFlags, global_atom_index);
649 void do_force_listed(struct gmx_wallcycle* wcycle,
651 const t_lambda* fepvals,
653 const gmx_multisim_t* ms,
657 gmx::ForceOutputs* forceOutputs,
658 const t_forcerec* fr,
659 const struct t_pbc* pbc,
660 const struct t_graph* graph,
661 gmx_enerdata_t* enerd,
666 int* global_atom_index,
667 const gmx::StepWorkload& stepWork)
669 t_pbc pbc_full; /* Full PBC is needed for position restraints */
671 if (!stepWork.computeListedForces)
676 if ((idef->il[F_POSRES].nr > 0) || (idef->il[F_FBPOSRES].nr > 0))
678 /* Not enough flops to bother counting */
679 set_pbc(&pbc_full, fr->ePBC, box);
681 calc_listed(cr, ms, wcycle, idef, x, hist, forceOutputs, fr, pbc, &pbc_full, graph, enerd, nrnb,
682 lambda, md, fcd, global_atom_index, stepWork);
684 /* Check if we have to determine energy differences
685 * at foreign lambda's.
687 if (fepvals->n_lambda > 0 && stepWork.computeDhdl)
689 posres_wrapper_lambda(wcycle, fepvals, idef, &pbc_full, x, enerd, lambda, fr);
691 if (idef->ilsort != ilsortNO_FE)
693 wallcycle_sub_start(wcycle, ewcsLISTED_FEP);
694 if (idef->ilsort != ilsortFE_SORTED)
696 gmx_incons("The bonded interactions are not sorted for free energy");
698 for (size_t i = 0; i < enerd->enerpart_lambda.size(); i++)
702 reset_foreign_enerdata(enerd);
703 for (int j = 0; j < efptNR; j++)
705 lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i - 1]);
707 calc_listed_lambda(idef, x, fr, pbc, graph, &(enerd->foreign_grpp),
708 enerd->foreign_term, nrnb, lam_i, md, fcd, global_atom_index);
709 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
710 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
712 wallcycle_sub_stop(wcycle, ewcsLISTED_FEP);