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/mdlib/force_flags.h"
64 #include "gromacs/mdtypes/commrec.h"
65 #include "gromacs/mdtypes/fcdata.h"
66 #include "gromacs/mdtypes/forcerec.h"
67 #include "gromacs/mdtypes/inputrec.h"
68 #include "gromacs/mdtypes/md_enums.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"
83 /*! \brief Return true if ftype is an explicit pair-listed LJ or
84 * COULOMB interaction type: bonded LJ (usually 1-4), or special
85 * listed non-bonded for FEP. */
87 isPairInteraction(int ftype)
89 return ((ftype) >= F_LJ14 && (ftype) <= F_LJC_PAIRS_NB);
92 /*! \brief Zero thread-local output buffers */
94 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 */
139 reduce_thread_forces(int n, rvec *f,
140 const bonded_threading_t *bt,
143 if (nthreads > MAX_BONDED_THREADS)
145 gmx_fatal(FARGS, "Can not reduce bonded forces on more than %d threads",
149 /* This reduction can run on any number of threads,
150 * independently of bt->nthreads.
151 * But if nthreads matches bt->nthreads (which it currently does)
152 * the uniform distribution of the touched blocks over nthreads will
153 * match the distribution of bonded over threads well in most cases,
154 * which means that threads mostly reduce their own data which increases
155 * the number of cache hits.
157 #pragma omp parallel for num_threads(nthreads) schedule(static)
158 for (int b = 0; b < bt->nblock_used; b++)
162 int ind = bt->block_index[b];
163 rvec4 *fp[MAX_BONDED_THREADS];
165 /* Determine which threads contribute to this block */
167 for (int ft = 0; ft < bt->nthreads; ft++)
169 if (bitmask_is_set(bt->mask[ind], ft))
171 fp[nfb++] = bt->f_t[ft]->f;
176 /* Reduce force buffers for threads that contribute */
177 int a0 = ind *reduction_block_size;
178 int a1 = (ind + 1)*reduction_block_size;
179 /* It would be nice if we could pad f to avoid this min */
180 a1 = std::min(a1, n);
181 for (int a = a0; a < a1; a++)
183 for (int fb = 0; fb < nfb; fb++)
185 rvec_inc(f[a], fp[fb][a]);
190 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
194 /*! \brief Reduce thread-local forces, shift forces and energies */
196 reduce_thread_output(int n, rvec *f, rvec *fshift,
197 real *ener, gmx_grppairener_t *grpp, real *dvdl,
198 const bonded_threading_t *bt,
199 gmx_bool bCalcEnerVir,
202 assert(bt->haveBondeds);
204 if (bt->nblock_used > 0)
206 /* Reduce the bonded force buffer */
207 reduce_thread_forces(n, f, bt, bt->nthreads);
210 /* When necessary, reduce energy and virial using one thread only */
211 if (bCalcEnerVir && bt->nthreads > 1)
213 gmx::ArrayRef < const std::unique_ptr < f_thread_t>> f_t = bt->f_t;
215 for (int i = 0; i < SHIFTS; i++)
217 for (int t = 1; t < bt->nthreads; t++)
219 rvec_inc(fshift[i], f_t[t]->fshift[i]);
222 for (int i = 0; i < F_NRE; i++)
224 for (int t = 1; t < bt->nthreads; t++)
226 ener[i] += f_t[t]->ener[i];
229 for (int i = 0; i < egNR; i++)
231 for (int j = 0; j < f_t[1]->grpp.nener; j++)
233 for (int t = 1; t < bt->nthreads; t++)
235 grpp->ener[i][j] += f_t[t]->grpp.ener[i][j];
241 for (int i = 0; i < efptNR; i++)
244 for (int t = 1; t < bt->nthreads; t++)
246 dvdl[i] += f_t[t]->dvdl[i];
253 /*! \brief Calculate one element of the list of bonded interactions
256 calc_one_bond(int thread,
257 int ftype, const t_idef *idef,
258 const WorkDivision &workDivision,
259 const rvec x[], rvec4 f[], rvec fshift[],
260 const t_forcerec *fr,
261 const t_pbc *pbc, const t_graph *g,
262 gmx_grppairener_t *grpp,
264 const real *lambda, real *dvdl,
265 const t_mdatoms *md, t_fcdata *fcd,
266 gmx_bool bCalcEnerVir,
267 int *global_atom_index)
269 BondedKernelFlavor bondedKernelFlavor;
272 bondedKernelFlavor = BondedKernelFlavor::ForcesAndVirialAndEnergy;
274 else if (fr->use_simd_kernels)
276 bondedKernelFlavor = BondedKernelFlavor::ForcesSimdWhenAvailable;
280 bondedKernelFlavor = BondedKernelFlavor::ForcesNoSimd;
283 int nat1, nbonds, efptFTYPE;
288 if (IS_RESTRAINT_TYPE(ftype))
290 efptFTYPE = efptRESTRAINT;
294 efptFTYPE = efptBONDED;
297 GMX_ASSERT(fr->efep == efepNO || idef->ilsort == ilsortNO_FE || idef->ilsort == ilsortFE_SORTED, "With free-energy calculations, we should either have no perturbed bondeds or sorted perturbed bondeds");
298 const bool useFreeEnergy = (idef->ilsort == ilsortFE_SORTED && idef->il[ftype].nr_nonperturbed < idef->il[ftype].nr);
299 const bool computeForcesOnly = (!bCalcEnerVir && !useFreeEnergy);
301 nat1 = interaction_function[ftype].nratoms + 1;
302 nbonds = idef->il[ftype].nr/nat1;
303 iatoms = idef->il[ftype].iatoms;
305 GMX_ASSERT(fr->gpuBonded != nullptr || workDivision.end(ftype) == idef->il[ftype].nr,
306 "The thread division should match the topology");
308 nb0 = workDivision.bound(ftype, thread);
309 nbn = workDivision.bound(ftype, thread + 1) - nb0;
311 if (!isPairInteraction(ftype))
315 /* TODO The execution time for CMAP dihedrals might be
316 nice to account to its own subtimer, but first
317 wallcycle needs to be extended to support calling from
319 v = cmap_dihs(nbn, iatoms+nb0,
320 idef->iparams, idef->cmap_grid,
322 pbc, g, lambda[efptFTYPE], &(dvdl[efptFTYPE]),
323 md, fcd, global_atom_index);
327 v = calculateSimpleBond(ftype, nbn, iatoms + nb0,
330 pbc, g, lambda[efptFTYPE], &(dvdl[efptFTYPE]),
331 md, fcd, global_atom_index,
337 /* TODO The execution time for pairs might be nice to account
338 to its own subtimer, but first wallcycle needs to be
339 extended to support calling from multiple threads. */
340 do_pairs(ftype, nbn, iatoms+nb0, idef->iparams, x, f, fshift,
341 pbc, g, lambda, dvdl, md, fr,
342 computeForcesOnly, grpp, global_atom_index);
348 inc_nrnb(nrnb, nrnbIndex(ftype), nbonds);
356 /*! \brief Compute the bonded part of the listed forces, parallelized over threads
359 calcBondedForces(const t_idef *idef,
361 const t_forcerec *fr,
362 const t_pbc *pbc_null,
364 gmx_enerdata_t *enerd,
370 gmx_bool bCalcEnerVir,
371 int *global_atom_index)
373 bonded_threading_t *bt = fr->bondedThreading;
375 #pragma omp parallel for num_threads(bt->nthreads) schedule(static)
376 for (int thread = 0; thread < bt->nthreads; thread++)
380 f_thread_t &threadBuffers = *bt->f_t[thread];
386 gmx_grppairener_t *grpp;
388 zero_thread_output(&threadBuffers);
390 rvec4 *ft = threadBuffers.f;
392 /* Thread 0 writes directly to the main output buffers.
393 * We might want to reconsider this.
404 fshift = threadBuffers.fshift;
405 epot = threadBuffers.ener;
406 grpp = &threadBuffers.grpp;
407 dvdlt = threadBuffers.dvdl;
409 /* Loop over all bonded force types to calculate the bonded forces */
410 for (ftype = 0; (ftype < F_NRE); ftype++)
412 if (idef->il[ftype].nr > 0 && ftype_is_bonded_potential(ftype))
414 v = calc_one_bond(thread, ftype, idef,
415 fr->bondedThreading->workDivision, x,
416 ft, fshift, fr, pbc_null, g, grpp,
418 md, fcd, bCalcEnerVir,
424 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
428 bool havePositionRestraints(const t_idef &idef,
432 ((idef.il[F_POSRES].nr > 0) ||
433 (idef.il[F_FBPOSRES].nr > 0) ||
435 fcd.disres.nres > 0);
438 bool haveCpuBondeds(const t_forcerec &fr)
440 return fr.bondedThreading->haveBondeds;
443 bool haveCpuListedForces(const t_forcerec &fr,
447 return haveCpuBondeds(fr) || havePositionRestraints(idef, fcd);
450 void calc_listed(const t_commrec *cr,
451 const gmx_multisim_t *ms,
452 struct gmx_wallcycle *wcycle,
454 const rvec x[], history_t *hist,
456 gmx::ForceWithVirial *forceWithVirial,
457 const t_forcerec *fr,
458 const struct t_pbc *pbc,
459 const struct t_pbc *pbc_full,
460 const struct t_graph *g,
461 gmx_enerdata_t *enerd, t_nrnb *nrnb,
464 t_fcdata *fcd, int *global_atom_index,
467 gmx_bool bCalcEnerVir;
468 const t_pbc *pbc_null;
469 bonded_threading_t *bt = fr->bondedThreading;
471 bCalcEnerVir = ((force_flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)) != 0);
482 if (havePositionRestraints(*idef, *fcd))
484 /* TODO Use of restraints triggers further function calls
485 inside the loop over calc_one_bond(), but those are too
486 awkward to account to this subtimer properly in the present
487 code. We don't test / care much about performance with
488 restraints, anyway. */
489 wallcycle_sub_start(wcycle, ewcsRESTRAINTS);
491 if (idef->il[F_POSRES].nr > 0)
493 posres_wrapper(nrnb, idef, pbc_full, x, enerd, lambda, fr,
497 if (idef->il[F_FBPOSRES].nr > 0)
499 fbposres_wrapper(nrnb, idef, pbc_full, x, enerd, fr,
503 /* Do pre force calculation stuff which might require communication */
504 if (fcd->orires.nr > 0)
506 /* This assertion is to ensure we have whole molecules.
507 * Unfortunately we do not have an mdrun state variable that tells
508 * us if molecules in x are not broken over PBC, so we have to make
509 * do with checking graph!=nullptr, which should tell us if we made
510 * molecules whole before calling the current function.
512 GMX_RELEASE_ASSERT(fr->ePBC == epbcNONE || g != nullptr, "With orientation restraints molecules should be whole");
513 enerd->term[F_ORIRESDEV] =
514 calc_orires_dev(ms, idef->il[F_ORIRES].nr,
515 idef->il[F_ORIRES].iatoms,
516 idef->iparams, md, x,
517 pbc_null, fcd, hist);
519 if (fcd->disres.nres > 0)
521 calc_disres_R_6(cr, ms,
522 idef->il[F_DISRES].nr,
523 idef->il[F_DISRES].iatoms,
528 wallcycle_sub_stop(wcycle, ewcsRESTRAINTS);
531 if (haveCpuBondeds(*fr))
533 wallcycle_sub_start(wcycle, ewcsLISTED);
534 /* The dummy array is to have a place to store the dhdl at other values
535 of lambda, which will be thrown away in the end */
536 real dvdl[efptNR] = {0};
537 calcBondedForces(idef, x, fr, pbc_null, g, enerd, nrnb, lambda, dvdl, md,
538 fcd, bCalcEnerVir, global_atom_index);
539 wallcycle_sub_stop(wcycle, ewcsLISTED);
541 wallcycle_sub_start(wcycle, ewcsLISTED_BUF_OPS);
542 reduce_thread_output(fr->natoms_force, f, fr->fshift,
543 enerd->term, &enerd->grpp, dvdl,
546 (force_flags & GMX_FORCE_DHDL) != 0);
548 if (force_flags & GMX_FORCE_DHDL)
550 for (int i = 0; i < efptNR; i++)
552 enerd->dvdl_nonlin[i] += dvdl[i];
555 wallcycle_sub_stop(wcycle, ewcsLISTED_BUF_OPS);
558 /* Copy the sum of violations for the distance restraints from fcd */
561 enerd->term[F_DISRESVIOL] = fcd->disres.sumviol;
565 void calc_listed_lambda(const t_idef *idef,
567 const t_forcerec *fr,
568 const struct t_pbc *pbc, const struct t_graph *g,
569 gmx_grppairener_t *grpp, real *epot, t_nrnb *nrnb,
573 int *global_atom_index)
576 real dvdl_dum[efptNR] = {0};
579 const t_pbc *pbc_null;
581 WorkDivision &workDivision = fr->bondedThreading->foreignLambdaWorkDivision;
592 /* Copy the whole idef, so we can modify the contents locally */
595 /* We already have the forces, so we use temp buffers here */
596 // TODO: Get rid of these allocations by using permanent force buffers
597 snew(f, fr->natoms_force);
598 snew(fshift, SHIFTS);
600 /* Loop over all bonded force types to calculate the bonded energies */
601 for (int ftype = 0; (ftype < F_NRE); ftype++)
603 if (ftype_is_bonded_potential(ftype))
605 const t_ilist &ilist = idef->il[ftype];
606 /* Create a temporary t_ilist with only perturbed interactions */
607 t_ilist &ilist_fe = idef_fe.il[ftype];
608 ilist_fe.iatoms = ilist.iatoms + ilist.nr_nonperturbed;
609 ilist_fe.nr_nonperturbed = 0;
610 ilist_fe.nr = ilist.nr - ilist.nr_nonperturbed;
611 /* Set the work range of thread 0 to the perturbed bondeds */
612 workDivision.setBound(ftype, 0, 0);
613 workDivision.setBound(ftype, 1, ilist_fe.nr);
617 v = calc_one_bond(0, ftype, &idef_fe, workDivision,
618 x, f, fshift, fr, pbc_null, g,
619 grpp, nrnb, lambda, dvdl_dum,
632 do_force_listed(struct gmx_wallcycle *wcycle,
634 const t_lambda *fepvals,
636 const gmx_multisim_t *ms,
640 rvec *forceForUseWithShiftForces,
641 gmx::ForceWithVirial *forceWithVirial,
642 const t_forcerec *fr,
643 const struct t_pbc *pbc,
644 const struct t_graph *graph,
645 gmx_enerdata_t *enerd,
650 int *global_atom_index,
653 t_pbc pbc_full; /* Full PBC is needed for position restraints */
655 if (!(flags & GMX_FORCE_LISTED))
660 if ((idef->il[F_POSRES].nr > 0) ||
661 (idef->il[F_FBPOSRES].nr > 0))
663 /* Not enough flops to bother counting */
664 set_pbc(&pbc_full, fr->ePBC, box);
666 calc_listed(cr, ms, wcycle, idef, x, hist,
667 forceForUseWithShiftForces, forceWithVirial,
669 graph, enerd, nrnb, lambda, md, fcd,
670 global_atom_index, flags);
672 /* Check if we have to determine energy differences
673 * at foreign lambda's.
675 if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL))
677 posres_wrapper_lambda(wcycle, fepvals, idef, &pbc_full, x, enerd, lambda, fr);
679 if (idef->ilsort != ilsortNO_FE)
681 wallcycle_sub_start(wcycle, ewcsLISTED_FEP);
682 if (idef->ilsort != ilsortFE_SORTED)
684 gmx_incons("The bonded interactions are not sorted for free energy");
686 for (size_t i = 0; i < enerd->enerpart_lambda.size(); i++)
690 reset_foreign_enerdata(enerd);
691 for (int j = 0; j < efptNR; j++)
693 lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
695 calc_listed_lambda(idef, x, fr, pbc, graph, &(enerd->foreign_grpp), enerd->foreign_term, nrnb, lam_i, md,
696 fcd, global_atom_index);
697 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
698 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
700 wallcycle_sub_stop(wcycle, ewcsLISTED_FEP);