2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2014,2015,2016,2017, 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"
52 #include "gromacs/gmxlib/network.h"
53 #include "gromacs/gmxlib/nrnb.h"
54 #include "gromacs/listed-forces/bonded.h"
55 #include "gromacs/listed-forces/disre.h"
56 #include "gromacs/listed-forces/orires.h"
57 #include "gromacs/listed-forces/pairs.h"
58 #include "gromacs/listed-forces/position-restraints.h"
59 #include "gromacs/math/vec.h"
60 #include "gromacs/mdlib/force.h"
61 #include "gromacs/mdlib/force_flags.h"
62 #include "gromacs/mdtypes/commrec.h"
63 #include "gromacs/mdtypes/fcdata.h"
64 #include "gromacs/mdtypes/forcerec.h"
65 #include "gromacs/mdtypes/inputrec.h"
66 #include "gromacs/mdtypes/md_enums.h"
67 #include "gromacs/pbcutil/ishift.h"
68 #include "gromacs/pbcutil/pbc.h"
69 #include "gromacs/simd/simd.h"
70 #include "gromacs/timing/wallcycle.h"
71 #include "gromacs/topology/topology.h"
72 #include "gromacs/utility/exceptions.h"
73 #include "gromacs/utility/fatalerror.h"
74 #include "gromacs/utility/smalloc.h"
76 #include "listed-internal.h"
81 /*! \brief Return true if ftype is an explicit pair-listed LJ or
82 * COULOMB interaction type: bonded LJ (usually 1-4), or special
83 * listed non-bonded for FEP. */
85 isPairInteraction(int ftype)
87 return ((ftype) >= F_LJ14 && (ftype) <= F_LJC_PAIRS_NB);
90 /*! \brief Zero thread-local output buffers */
92 zero_thread_output(struct bonded_threading_t *bt, int thread)
99 f_thread_t *f_t = &bt->f_t[thread];
100 const int nelem_fa = sizeof(*f_t->f)/sizeof(real);
102 for (int i = 0; i < f_t->nblock_used; i++)
104 int a0 = f_t->block_index[i]*reduction_block_size;
105 int a1 = a0 + reduction_block_size;
106 for (int a = a0; a < a1; a++)
108 for (int d = 0; d < nelem_fa; d++)
115 for (int i = 0; i < SHIFTS; i++)
117 clear_rvec(f_t->fshift[i]);
119 for (int i = 0; i < F_NRE; i++)
123 for (int i = 0; i < egNR; i++)
125 for (int j = 0; j < f_t->grpp.nener; j++)
127 f_t->grpp.ener[i][j] = 0;
130 for (int i = 0; i < efptNR; i++)
136 /*! \brief The max thread number is arbitrary, we used a fixed number
137 * to avoid memory management. Using more than 16 threads is probably
138 * never useful performance wise. */
139 #define MAX_BONDED_THREADS 256
141 /*! \brief Reduce thread-local force buffers */
143 reduce_thread_forces(int n, rvec *f,
144 struct bonded_threading_t *bt,
147 if (nthreads > MAX_BONDED_THREADS)
149 gmx_fatal(FARGS, "Can not reduce bonded forces on more than %d threads",
153 /* This reduction can run on any number of threads,
154 * independently of bt->nthreads.
155 * But if nthreads matches bt->nthreads (which it currently does)
156 * the uniform distribution of the touched blocks over nthreads will
157 * match the distribution of bonded over threads well in most cases,
158 * which means that threads mostly reduce their own data which increases
159 * the number of cache hits.
161 #pragma omp parallel for num_threads(nthreads) schedule(static)
162 for (int b = 0; b < bt->nblock_used; b++)
166 int ind = bt->block_index[b];
167 rvec4 *fp[MAX_BONDED_THREADS];
169 /* Determine which threads contribute to this block */
171 for (int ft = 0; ft < bt->nthreads; ft++)
173 if (bitmask_is_set(bt->mask[ind], ft))
175 fp[nfb++] = bt->f_t[ft].f;
180 /* Reduce force buffers for threads that contribute */
181 int a0 = ind *reduction_block_size;
182 int a1 = (ind + 1)*reduction_block_size;
183 /* It would be nice if we could pad f to avoid this min */
184 a1 = std::min(a1, n);
185 for (int a = a0; a < a1; a++)
187 for (int fb = 0; fb < nfb; fb++)
189 rvec_inc(f[a], fp[fb][a]);
194 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
198 /*! \brief Reduce thread-local forces, shift forces and energies */
200 reduce_thread_output(int n, rvec *f, rvec *fshift,
201 real *ener, gmx_grppairener_t *grpp, real *dvdl,
202 struct bonded_threading_t *bt,
203 gmx_bool bCalcEnerVir,
206 if (!bt->haveBondeds)
211 if (bt->nblock_used > 0)
213 /* Reduce the bonded force buffer */
214 reduce_thread_forces(n, f, bt, bt->nthreads);
217 /* When necessary, reduce energy and virial using one thread only */
218 if (bCalcEnerVir && bt->nthreads > 1)
220 f_thread_t *f_t = bt->f_t;
222 for (int i = 0; i < SHIFTS; i++)
224 for (int t = 1; t < bt->nthreads; t++)
226 rvec_inc(fshift[i], f_t[t].fshift[i]);
229 for (int i = 0; i < F_NRE; i++)
231 for (int t = 1; t < bt->nthreads; t++)
233 ener[i] += f_t[t].ener[i];
236 for (int i = 0; i < egNR; i++)
238 for (int j = 0; j < f_t[1].grpp.nener; j++)
240 for (int t = 1; t < bt->nthreads; t++)
242 grpp->ener[i][j] += f_t[t].grpp.ener[i][j];
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 Calculate one element of the list of bonded interactions
263 calc_one_bond(int thread,
264 int ftype, const t_idef *idef,
265 const rvec x[], rvec4 f[], rvec fshift[],
267 const t_pbc *pbc, const t_graph *g,
268 gmx_grppairener_t *grpp,
270 real *lambda, real *dvdl,
271 const t_mdatoms *md, t_fcdata *fcd,
272 gmx_bool bCalcEnerVir,
273 int *global_atom_index)
275 #if GMX_SIMD_HAVE_REAL
276 bool bUseSIMD = fr->use_simd_kernels;
279 int nat1, nbonds, efptFTYPE;
284 if (IS_RESTRAINT_TYPE(ftype))
286 efptFTYPE = efptRESTRAINT;
290 efptFTYPE = efptBONDED;
293 nat1 = interaction_function[ftype].nratoms + 1;
294 nbonds = idef->il[ftype].nr/nat1;
295 iatoms = idef->il[ftype].iatoms;
297 nb0 = idef->il_thread_division[ftype*(idef->nthreads+1)+thread];
298 nbn = idef->il_thread_division[ftype*(idef->nthreads+1)+thread+1] - nb0;
300 if (!isPairInteraction(ftype))
304 /* TODO The execution time for CMAP dihedrals might be
305 nice to account to its own subtimer, but first
306 wallcycle needs to be extended to support calling from
308 v = cmap_dihs(nbn, iatoms+nb0,
309 idef->iparams, &idef->cmap_grid,
311 pbc, g, lambda[efptFTYPE], &(dvdl[efptFTYPE]),
312 md, fcd, global_atom_index);
314 #if GMX_SIMD_HAVE_REAL
315 else if (ftype == F_ANGLES && bUseSIMD &&
316 !bCalcEnerVir && fr->efep == efepNO)
318 /* No energies, shift forces, dvdl */
319 angles_noener_simd(nbn, idef->il[ftype].iatoms+nb0,
322 pbc, g, lambda[efptFTYPE], md, fcd,
327 else if (ftype == F_PDIHS &&
328 !bCalcEnerVir && fr->efep == efepNO)
330 /* No energies, shift forces, dvdl */
331 #if GMX_SIMD_HAVE_REAL
334 pdihs_noener_simd(nbn, idef->il[ftype].iatoms+nb0,
337 pbc, g, lambda[efptFTYPE], md, fcd,
343 pdihs_noener(nbn, idef->il[ftype].iatoms+nb0,
346 pbc, g, lambda[efptFTYPE], md, fcd,
351 #if GMX_SIMD_HAVE_REAL
352 else if (ftype == F_RBDIHS && bUseSIMD &&
353 !bCalcEnerVir && fr->efep == efepNO)
355 /* No energies, shift forces, dvdl */
356 rbdihs_noener_simd(nbn, idef->il[ftype].iatoms+nb0,
359 pbc, g, lambda[efptFTYPE], md, fcd,
366 v = interaction_function[ftype].ifunc(nbn, iatoms+nb0,
369 pbc, g, lambda[efptFTYPE], &(dvdl[efptFTYPE]),
370 md, fcd, global_atom_index);
375 /* TODO The execution time for pairs might be nice to account
376 to its own subtimer, but first wallcycle needs to be
377 extended to support calling from multiple threads. */
378 do_pairs(ftype, nbn, iatoms+nb0, idef->iparams, x, f, fshift,
379 pbc, g, lambda, dvdl, md, fr,
380 bCalcEnerVir, grpp, global_atom_index);
386 inc_nrnb(nrnb, interaction_function[ftype].nrnb_ind, nbonds);
395 ftype_is_bonded_potential(int ftype)
398 (interaction_function[ftype].flags & IF_BOND) &&
399 !(ftype == F_CONNBONDS || ftype == F_POSRES || ftype == F_FBPOSRES) &&
400 (ftype < F_GB12 || ftype > F_GB14);
403 void calc_listed(const t_commrec *cr,
404 struct gmx_wallcycle *wcycle,
406 const rvec x[], history_t *hist,
407 rvec f[], t_forcerec *fr,
408 const struct t_pbc *pbc,
409 const struct t_pbc *pbc_full,
410 const struct t_graph *g,
411 gmx_enerdata_t *enerd, t_nrnb *nrnb,
414 t_fcdata *fcd, int *global_atom_index,
417 struct bonded_threading_t *bt;
418 gmx_bool bCalcEnerVir;
420 /* The dummy array is to have a place to store the dhdl at other values
421 of lambda, which will be thrown away in the end */
423 const t_pbc *pbc_null;
426 bt = fr->bonded_threading;
428 assert(bt->nthreads == idef->nthreads);
430 bCalcEnerVir = (force_flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY));
432 for (i = 0; i < efptNR; i++)
448 p_graph(debug, "Bondage is fun", g);
452 if ((idef->il[F_POSRES].nr > 0) ||
453 (idef->il[F_FBPOSRES].nr > 0) ||
454 fcd->orires.nr > 0 ||
455 fcd->disres.nres > 0)
457 /* TODO Use of restraints triggers further function calls
458 inside the loop over calc_one_bond(), but those are too
459 awkward to account to this subtimer properly in the present
460 code. We don't test / care much about performance with
461 restraints, anyway. */
462 wallcycle_sub_start(wcycle, ewcsRESTRAINTS);
464 if (idef->il[F_POSRES].nr > 0)
466 posres_wrapper(nrnb, idef, pbc_full, x, enerd, lambda, fr);
469 if (idef->il[F_FBPOSRES].nr > 0)
471 fbposres_wrapper(nrnb, idef, pbc_full, x, enerd, fr);
474 /* Do pre force calculation stuff which might require communication */
475 if (fcd->orires.nr > 0)
477 enerd->term[F_ORIRESDEV] =
478 calc_orires_dev(cr->ms, idef->il[F_ORIRES].nr,
479 idef->il[F_ORIRES].iatoms,
480 idef->iparams, md, x,
481 pbc_null, fcd, hist);
483 if (fcd->disres.nres > 0)
486 idef->il[F_DISRES].nr,
487 idef->il[F_DISRES].iatoms,
492 wallcycle_sub_stop(wcycle, ewcsRESTRAINTS);
495 /* TODO: Skip this whole loop with a system/domain without listeds */
496 wallcycle_sub_start(wcycle, ewcsLISTED);
497 #pragma omp parallel for num_threads(bt->nthreads) schedule(static)
498 for (thread = 0; thread < bt->nthreads; thread++)
508 gmx_grppairener_t *grpp;
510 zero_thread_output(bt, thread);
512 ft = bt->f_t[thread].f;
523 fshift = bt->f_t[thread].fshift;
524 epot = bt->f_t[thread].ener;
525 grpp = &bt->f_t[thread].grpp;
526 dvdlt = bt->f_t[thread].dvdl;
528 /* Loop over all bonded force types to calculate the bonded forces */
529 for (ftype = 0; (ftype < F_NRE); ftype++)
531 if (idef->il[ftype].nr > 0 && ftype_is_bonded_potential(ftype))
533 v = calc_one_bond(thread, ftype, idef, x,
534 ft, fshift, fr, pbc_null, g, grpp,
536 md, fcd, bCalcEnerVir,
542 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
544 wallcycle_sub_stop(wcycle, ewcsLISTED);
546 wallcycle_sub_start(wcycle, ewcsLISTED_BUF_OPS);
547 reduce_thread_output(fr->natoms_force, f, fr->fshift,
548 enerd->term, &enerd->grpp, dvdl,
551 force_flags & GMX_FORCE_DHDL);
552 wallcycle_sub_stop(wcycle, ewcsLISTED_BUF_OPS);
554 /* Remaining code does not have enough flops to bother counting */
555 if (force_flags & GMX_FORCE_DHDL)
557 for (i = 0; i < efptNR; i++)
559 enerd->dvdl_nonlin[i] += dvdl[i];
563 /* Copy the sum of violations for the distance restraints from fcd */
566 enerd->term[F_DISRESVIOL] = fcd->disres.sumviol;
571 void calc_listed_lambda(const t_idef *idef,
574 const struct t_pbc *pbc, const struct t_graph *g,
575 gmx_grppairener_t *grpp, real *epot, t_nrnb *nrnb,
579 int *global_atom_index)
581 int ftype, nr_nonperturbed, nr;
583 real dvdl_dum[efptNR] = {0};
586 const t_pbc *pbc_null;
598 /* Copy the whole idef, so we can modify the contents locally */
600 idef_fe.nthreads = 1;
601 snew(idef_fe.il_thread_division, F_NRE*(idef_fe.nthreads+1));
603 /* We already have the forces, so we use temp buffers here */
604 snew(f, fr->natoms_force);
605 snew(fshift, SHIFTS);
607 /* Loop over all bonded force types to calculate the bonded energies */
608 for (ftype = 0; (ftype < F_NRE); ftype++)
610 if (ftype_is_bonded_potential(ftype))
612 /* Set the work range of thread 0 to the perturbed bondeds only */
613 nr_nonperturbed = idef->il[ftype].nr_nonperturbed;
614 nr = idef->il[ftype].nr;
615 idef_fe.il_thread_division[ftype*2+0] = nr_nonperturbed;
616 idef_fe.il_thread_division[ftype*2+1] = nr;
618 /* This is only to get the flop count correct */
619 idef_fe.il[ftype].nr = nr - nr_nonperturbed;
621 if (nr - nr_nonperturbed > 0)
623 v = calc_one_bond(0, ftype, &idef_fe,
624 x, f, fshift, fr, pbc_null, g,
625 grpp, nrnb, lambda, dvdl_dum,
636 sfree(idef_fe.il_thread_division);
640 do_force_listed(struct gmx_wallcycle *wcycle,
642 const t_lambda *fepvals,
649 const struct t_pbc *pbc,
650 const struct t_graph *graph,
651 gmx_enerdata_t *enerd,
656 int *global_atom_index,
659 t_pbc pbc_full; /* Full PBC is needed for position restraints */
661 if (!(flags & GMX_FORCE_LISTED))
666 if ((idef->il[F_POSRES].nr > 0) ||
667 (idef->il[F_FBPOSRES].nr > 0))
669 /* Not enough flops to bother counting */
670 set_pbc(&pbc_full, fr->ePBC, box);
672 calc_listed(cr, wcycle, idef, x, hist, f, fr, pbc, &pbc_full,
673 graph, enerd, nrnb, lambda, md, fcd,
674 global_atom_index, flags);
676 /* Check if we have to determine energy differences
677 * at foreign lambda's.
679 if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL))
681 posres_wrapper_lambda(wcycle, fepvals, idef, &pbc_full, x, enerd, lambda, fr);
683 if (idef->ilsort != ilsortNO_FE)
685 wallcycle_sub_start(wcycle, ewcsLISTED_FEP);
686 if (idef->ilsort != ilsortFE_SORTED)
688 gmx_incons("The bonded interactions are not sorted for free energy");
690 for (int i = 0; i < enerd->n_lambda; i++)
694 reset_foreign_enerdata(enerd);
695 for (int j = 0; j < efptNR; j++)
697 lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
699 calc_listed_lambda(idef, x, fr, pbc, graph, &(enerd->foreign_grpp), enerd->foreign_term, nrnb, lam_i, md,
700 fcd, global_atom_index);
701 sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
702 enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
704 wallcycle_sub_stop(wcycle, ewcsLISTED_FEP);