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) 2013,2014,2015,2016,2017,2018,2019, 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.
50 #include "gromacs/commandline/filenm.h"
51 #include "gromacs/compat/make_unique.h"
52 #include "gromacs/domdec/domdec.h"
53 #include "gromacs/domdec/domdec_struct.h"
54 #include "gromacs/ewald/ewald.h"
55 #include "gromacs/ewald/ewald-utils.h"
56 #include "gromacs/fileio/filetypes.h"
57 #include "gromacs/gmxlib/network.h"
58 #include "gromacs/gmxlib/nonbonded/nonbonded.h"
59 #include "gromacs/gpu_utils/gpu_utils.h"
60 #include "gromacs/hardware/hw_info.h"
61 #include "gromacs/listed-forces/gpubonded.h"
62 #include "gromacs/listed-forces/manage-threading.h"
63 #include "gromacs/listed-forces/pairs.h"
64 #include "gromacs/math/functions.h"
65 #include "gromacs/math/units.h"
66 #include "gromacs/math/utilities.h"
67 #include "gromacs/math/vec.h"
68 #include "gromacs/mdlib/force.h"
69 #include "gromacs/mdlib/forcerec-threading.h"
70 #include "gromacs/mdlib/gmx_omp_nthreads.h"
71 #include "gromacs/mdlib/md_support.h"
72 #include "gromacs/mdlib/nb_verlet.h"
73 #include "gromacs/mdlib/nbnxn_atomdata.h"
74 #include "gromacs/mdlib/nbnxn_gpu_data_mgmt.h"
75 #include "gromacs/mdlib/nbnxn_grid.h"
76 #include "gromacs/mdlib/nbnxn_internal.h"
77 #include "gromacs/mdlib/nbnxn_search.h"
78 #include "gromacs/mdlib/nbnxn_simd.h"
79 #include "gromacs/mdlib/nbnxn_tuning.h"
80 #include "gromacs/mdlib/nbnxn_util.h"
81 #include "gromacs/mdlib/ns.h"
82 #include "gromacs/mdlib/qmmm.h"
83 #include "gromacs/mdlib/rf_util.h"
84 #include "gromacs/mdlib/sim_util.h"
85 #include "gromacs/mdlib/wall.h"
86 #include "gromacs/mdtypes/commrec.h"
87 #include "gromacs/mdtypes/fcdata.h"
88 #include "gromacs/mdtypes/group.h"
89 #include "gromacs/mdtypes/iforceprovider.h"
90 #include "gromacs/mdtypes/inputrec.h"
91 #include "gromacs/mdtypes/md_enums.h"
92 #include "gromacs/pbcutil/ishift.h"
93 #include "gromacs/pbcutil/pbc.h"
94 #include "gromacs/simd/simd.h"
95 #include "gromacs/tables/forcetable.h"
96 #include "gromacs/topology/mtop_util.h"
97 #include "gromacs/trajectory/trajectoryframe.h"
98 #include "gromacs/utility/cstringutil.h"
99 #include "gromacs/utility/exceptions.h"
100 #include "gromacs/utility/fatalerror.h"
101 #include "gromacs/utility/gmxassert.h"
102 #include "gromacs/utility/logger.h"
103 #include "gromacs/utility/physicalnodecommunicator.h"
104 #include "gromacs/utility/pleasecite.h"
105 #include "gromacs/utility/smalloc.h"
106 #include "gromacs/utility/strconvert.h"
108 #include "nbnxn_gpu_jit_support.h"
110 t_forcerec *mk_forcerec()
119 static real *mk_nbfp(const gmx_ffparams_t *idef, gmx_bool bBHAM)
127 snew(nbfp, 3*atnr*atnr);
128 for (i = k = 0; (i < atnr); i++)
130 for (j = 0; (j < atnr); j++, k++)
132 BHAMA(nbfp, atnr, i, j) = idef->iparams[k].bham.a;
133 BHAMB(nbfp, atnr, i, j) = idef->iparams[k].bham.b;
134 /* nbfp now includes the 6.0 derivative prefactor */
135 BHAMC(nbfp, atnr, i, j) = idef->iparams[k].bham.c*6.0;
141 snew(nbfp, 2*atnr*atnr);
142 for (i = k = 0; (i < atnr); i++)
144 for (j = 0; (j < atnr); j++, k++)
146 /* nbfp now includes the 6.0/12.0 derivative prefactors */
147 C6(nbfp, atnr, i, j) = idef->iparams[k].lj.c6*6.0;
148 C12(nbfp, atnr, i, j) = idef->iparams[k].lj.c12*12.0;
156 static real *make_ljpme_c6grid(const gmx_ffparams_t *idef, t_forcerec *fr)
159 real c6, c6i, c6j, c12i, c12j, epsi, epsj, sigmai, sigmaj;
162 /* For LJ-PME simulations, we correct the energies with the reciprocal space
163 * inside of the cut-off. To do this the non-bonded kernels needs to have
164 * access to the C6-values used on the reciprocal grid in pme.c
168 snew(grid, 2*atnr*atnr);
169 for (i = k = 0; (i < atnr); i++)
171 for (j = 0; (j < atnr); j++, k++)
173 c6i = idef->iparams[i*(atnr+1)].lj.c6;
174 c12i = idef->iparams[i*(atnr+1)].lj.c12;
175 c6j = idef->iparams[j*(atnr+1)].lj.c6;
176 c12j = idef->iparams[j*(atnr+1)].lj.c12;
177 c6 = std::sqrt(c6i * c6j);
178 if (fr->ljpme_combination_rule == eljpmeLB
179 && !gmx_numzero(c6) && !gmx_numzero(c12i) && !gmx_numzero(c12j))
181 sigmai = gmx::sixthroot(c12i / c6i);
182 sigmaj = gmx::sixthroot(c12j / c6j);
183 epsi = c6i * c6i / c12i;
184 epsj = c6j * c6j / c12j;
185 c6 = std::sqrt(epsi * epsj) * gmx::power6(0.5*(sigmai+sigmaj));
187 /* Store the elements at the same relative positions as C6 in nbfp in order
188 * to simplify access in the kernels
190 grid[2*(atnr*i+j)] = c6*6.0;
196 static real *mk_nbfp_combination_rule(const gmx_ffparams_t *idef, int comb_rule)
200 real c6i, c6j, c12i, c12j, epsi, epsj, sigmai, sigmaj;
204 snew(nbfp, 2*atnr*atnr);
205 for (i = 0; i < atnr; ++i)
207 for (j = 0; j < atnr; ++j)
209 c6i = idef->iparams[i*(atnr+1)].lj.c6;
210 c12i = idef->iparams[i*(atnr+1)].lj.c12;
211 c6j = idef->iparams[j*(atnr+1)].lj.c6;
212 c12j = idef->iparams[j*(atnr+1)].lj.c12;
213 c6 = std::sqrt(c6i * c6j);
214 c12 = std::sqrt(c12i * c12j);
215 if (comb_rule == eCOMB_ARITHMETIC
216 && !gmx_numzero(c6) && !gmx_numzero(c12))
218 sigmai = gmx::sixthroot(c12i / c6i);
219 sigmaj = gmx::sixthroot(c12j / c6j);
220 epsi = c6i * c6i / c12i;
221 epsj = c6j * c6j / c12j;
222 c6 = std::sqrt(epsi * epsj) * gmx::power6(0.5*(sigmai+sigmaj));
223 c12 = std::sqrt(epsi * epsj) * gmx::power12(0.5*(sigmai+sigmaj));
225 C6(nbfp, atnr, i, j) = c6*6.0;
226 C12(nbfp, atnr, i, j) = c12*12.0;
232 /* This routine sets fr->solvent_opt to the most common solvent in the
233 * system, e.g. esolSPC or esolTIP4P. It will also mark each charge group in
234 * the fr->solvent_type array with the correct type (or esolNO).
236 * Charge groups that fulfill the conditions but are not identical to the
237 * most common one will be marked as esolNO in the solvent_type array.
239 * TIP3p is identical to SPC for these purposes, so we call it
240 * SPC in the arrays (Apologies to Bill Jorgensen ;-)
242 * NOTE: QM particle should not
243 * become an optimized solvent. Not even if there is only one charge
253 } solvent_parameters_t;
256 check_solvent_cg(const gmx_moltype_t *molt,
259 const unsigned char *qm_grpnr,
260 const t_grps *qm_grps,
262 int *n_solvent_parameters,
263 solvent_parameters_t **solvent_parameters_p,
273 real tmp_charge[4] = { 0.0 }; /* init to zero to make gcc4.8 happy */
274 int tmp_vdwtype[4] = { 0 }; /* init to zero to make gcc4.8 happy */
277 solvent_parameters_t *solvent_parameters;
279 /* We use a list with parameters for each solvent type.
280 * Every time we discover a new molecule that fulfills the basic
281 * conditions for a solvent we compare with the previous entries
282 * in these lists. If the parameters are the same we just increment
283 * the counter for that type, and otherwise we create a new type
284 * based on the current molecule.
286 * Once we've finished going through all molecules we check which
287 * solvent is most common, and mark all those molecules while we
288 * clear the flag on all others.
291 solvent_parameters = *solvent_parameters_p;
293 /* Mark the cg first as non optimized */
296 /* Check if this cg has no exclusions with atoms in other charge groups
297 * and all atoms inside the charge group excluded.
298 * We only have 3 or 4 atom solvent loops.
300 if (GET_CGINFO_EXCL_INTER(cginfo) ||
301 !GET_CGINFO_EXCL_INTRA(cginfo))
306 /* Get the indices of the first atom in this charge group */
307 j0 = molt->cgs.index[cg0];
308 j1 = molt->cgs.index[cg0+1];
310 /* Number of atoms in our molecule */
316 "Moltype '%s': there are %d atoms in this charge group\n",
320 /* Check if it could be an SPC (3 atoms) or TIP4p (4) water,
323 if (nj < 3 || nj > 4)
328 /* Check if we are doing QM on this group */
330 if (qm_grpnr != nullptr)
332 for (j = j0; j < j1 && !qm; j++)
334 qm = (qm_grpnr[j] < qm_grps->nr - 1);
337 /* Cannot use solvent optimization with QM */
343 atom = molt->atoms.atom;
345 /* Still looks like a solvent, time to check parameters */
347 /* If it is perturbed (free energy) we can't use the solvent loops,
348 * so then we just skip to the next molecule.
352 for (j = j0; j < j1 && !perturbed; j++)
354 perturbed = PERTURBED(atom[j]);
362 /* Now it's only a question if the VdW and charge parameters
363 * are OK. Before doing the check we compare and see if they are
364 * identical to a possible previous solvent type.
365 * First we assign the current types and charges.
367 for (j = 0; j < nj; j++)
369 tmp_vdwtype[j] = atom[j0+j].type;
370 tmp_charge[j] = atom[j0+j].q;
373 /* Does it match any previous solvent type? */
374 for (k = 0; k < *n_solvent_parameters; k++)
379 /* We can only match SPC with 3 atoms and TIP4p with 4 atoms */
380 if ( (solvent_parameters[k].model == esolSPC && nj != 3) ||
381 (solvent_parameters[k].model == esolTIP4P && nj != 4) )
386 /* Check that types & charges match for all atoms in molecule */
387 for (j = 0; j < nj && match; j++)
389 if (tmp_vdwtype[j] != solvent_parameters[k].vdwtype[j])
393 if (tmp_charge[j] != solvent_parameters[k].charge[j])
400 /* Congratulations! We have a matched solvent.
401 * Flag it with this type for later processing.
404 solvent_parameters[k].count += nmol;
406 /* We are done with this charge group */
411 /* If we get here, we have a tentative new solvent type.
412 * Before we add it we must check that it fulfills the requirements
413 * of the solvent optimized loops. First determine which atoms have
416 for (j = 0; j < nj; j++)
419 tjA = tmp_vdwtype[j];
421 /* Go through all other tpes and see if any have non-zero
422 * VdW parameters when combined with this one.
424 for (k = 0; k < fr->ntype && (!has_vdw[j]); k++)
426 /* We already checked that the atoms weren't perturbed,
427 * so we only need to check state A now.
431 has_vdw[j] = (has_vdw[j] ||
432 (BHAMA(fr->nbfp, fr->ntype, tjA, k) != 0.0) ||
433 (BHAMB(fr->nbfp, fr->ntype, tjA, k) != 0.0) ||
434 (BHAMC(fr->nbfp, fr->ntype, tjA, k) != 0.0));
439 has_vdw[j] = (has_vdw[j] ||
440 (C6(fr->nbfp, fr->ntype, tjA, k) != 0.0) ||
441 (C12(fr->nbfp, fr->ntype, tjA, k) != 0.0));
446 /* Now we know all we need to make the final check and assignment. */
450 * For this we require thatn all atoms have charge,
451 * the charges on atom 2 & 3 should be the same, and only
452 * atom 1 might have VdW.
456 tmp_charge[0] != 0 &&
457 tmp_charge[1] != 0 &&
458 tmp_charge[2] == tmp_charge[1])
460 srenew(solvent_parameters, *n_solvent_parameters+1);
461 solvent_parameters[*n_solvent_parameters].model = esolSPC;
462 solvent_parameters[*n_solvent_parameters].count = nmol;
463 for (k = 0; k < 3; k++)
465 solvent_parameters[*n_solvent_parameters].vdwtype[k] = tmp_vdwtype[k];
466 solvent_parameters[*n_solvent_parameters].charge[k] = tmp_charge[k];
469 *cg_sp = *n_solvent_parameters;
470 (*n_solvent_parameters)++;
475 /* Or could it be a TIP4P?
476 * For this we require thatn atoms 2,3,4 have charge, but not atom 1.
477 * Only atom 1 mght have VdW.
482 tmp_charge[0] == 0 &&
483 tmp_charge[1] != 0 &&
484 tmp_charge[2] == tmp_charge[1] &&
487 srenew(solvent_parameters, *n_solvent_parameters+1);
488 solvent_parameters[*n_solvent_parameters].model = esolTIP4P;
489 solvent_parameters[*n_solvent_parameters].count = nmol;
490 for (k = 0; k < 4; k++)
492 solvent_parameters[*n_solvent_parameters].vdwtype[k] = tmp_vdwtype[k];
493 solvent_parameters[*n_solvent_parameters].charge[k] = tmp_charge[k];
496 *cg_sp = *n_solvent_parameters;
497 (*n_solvent_parameters)++;
501 *solvent_parameters_p = solvent_parameters;
505 check_solvent(FILE * fp,
506 const gmx_mtop_t * mtop,
508 cginfo_mb_t *cginfo_mb)
511 const gmx_moltype_t *molt;
512 int mol, cg_mol, at_offset, am, cgm, i, nmol_ch, nmol;
513 int n_solvent_parameters;
514 solvent_parameters_t *solvent_parameters;
520 fprintf(debug, "Going to determine what solvent types we have.\n");
523 n_solvent_parameters = 0;
524 solvent_parameters = nullptr;
525 /* Allocate temporary array for solvent type */
526 snew(cg_sp, mtop->molblock.size());
529 for (size_t mb = 0; mb < mtop->molblock.size(); mb++)
531 molt = &mtop->moltype[mtop->molblock[mb].type];
533 /* Here we have to loop over all individual molecules
534 * because we need to check for QMMM particles.
536 snew(cg_sp[mb], cginfo_mb[mb].cg_mod);
537 nmol_ch = cginfo_mb[mb].cg_mod/cgs->nr;
538 nmol = mtop->molblock[mb].nmol/nmol_ch;
539 for (mol = 0; mol < nmol_ch; mol++)
542 am = mol*cgs->index[cgs->nr];
543 for (cg_mol = 0; cg_mol < cgs->nr; cg_mol++)
545 check_solvent_cg(molt, cg_mol, nmol,
546 mtop->groups.grpnr[egcQMMM] ?
547 mtop->groups.grpnr[egcQMMM]+at_offset+am : nullptr,
548 &mtop->groups.grps[egcQMMM],
550 &n_solvent_parameters, &solvent_parameters,
551 cginfo_mb[mb].cginfo[cgm+cg_mol],
552 &cg_sp[mb][cgm+cg_mol]);
555 at_offset += cgs->index[cgs->nr];
558 /* Puh! We finished going through all charge groups.
559 * Now find the most common solvent model.
562 /* Most common solvent this far */
564 for (i = 0; i < n_solvent_parameters; i++)
567 solvent_parameters[i].count > solvent_parameters[bestsp].count)
575 bestsol = solvent_parameters[bestsp].model;
583 for (size_t mb = 0; mb < mtop->molblock.size(); mb++)
585 cgs = &mtop->moltype[mtop->molblock[mb].type].cgs;
586 nmol = (mtop->molblock[mb].nmol*cgs->nr)/cginfo_mb[mb].cg_mod;
587 for (i = 0; i < cginfo_mb[mb].cg_mod; i++)
589 if (cg_sp[mb][i] == bestsp)
591 SET_CGINFO_SOLOPT(cginfo_mb[mb].cginfo[i], bestsol);
596 SET_CGINFO_SOLOPT(cginfo_mb[mb].cginfo[i], esolNO);
603 if (bestsol != esolNO && fp != nullptr)
605 fprintf(fp, "\nEnabling %s-like water optimization for %d molecules.\n\n",
607 solvent_parameters[bestsp].count);
610 sfree(solvent_parameters);
611 fr->solvent_opt = bestsol;
615 acNONE = 0, acCONSTRAINT, acSETTLE
618 static cginfo_mb_t *init_cginfo_mb(FILE *fplog, const gmx_mtop_t *mtop,
619 t_forcerec *fr, gmx_bool bNoSolvOpt,
620 gmx_bool *bFEP_NonBonded,
621 gmx_bool *bExcl_IntraCGAll_InterCGNone)
624 const t_blocka *excl;
625 const gmx_moltype_t *molt;
626 const gmx_molblock_t *molb;
627 cginfo_mb_t *cginfo_mb;
630 int cg_offset, a_offset;
631 int m, cg, a0, a1, gid, ai, j, aj, excl_nalloc;
635 gmx_bool bId, *bExcl, bExclIntraAll, bExclInter, bHaveVDW, bHaveQ, bHavePerturbedAtoms;
637 snew(cginfo_mb, mtop->molblock.size());
639 snew(type_VDW, fr->ntype);
640 for (ai = 0; ai < fr->ntype; ai++)
642 type_VDW[ai] = FALSE;
643 for (j = 0; j < fr->ntype; j++)
645 type_VDW[ai] = type_VDW[ai] ||
647 C6(fr->nbfp, fr->ntype, ai, j) != 0 ||
648 C12(fr->nbfp, fr->ntype, ai, j) != 0;
652 *bFEP_NonBonded = FALSE;
653 *bExcl_IntraCGAll_InterCGNone = TRUE;
656 snew(bExcl, excl_nalloc);
659 for (size_t mb = 0; mb < mtop->molblock.size(); mb++)
661 molb = &mtop->molblock[mb];
662 molt = &mtop->moltype[molb->type];
666 /* Check if the cginfo is identical for all molecules in this block.
667 * If so, we only need an array of the size of one molecule.
668 * Otherwise we make an array of #mol times #cgs per molecule.
671 for (m = 0; m < molb->nmol; m++)
673 int am = m*cgs->index[cgs->nr];
674 for (cg = 0; cg < cgs->nr; cg++)
677 a1 = cgs->index[cg+1];
678 if (getGroupType(mtop->groups, egcENER, a_offset+am+a0) !=
679 getGroupType(mtop->groups, egcENER, a_offset +a0))
683 if (mtop->groups.grpnr[egcQMMM] != nullptr)
685 for (ai = a0; ai < a1; ai++)
687 if (mtop->groups.grpnr[egcQMMM][a_offset+am+ai] !=
688 mtop->groups.grpnr[egcQMMM][a_offset +ai])
697 cginfo_mb[mb].cg_start = cg_offset;
698 cginfo_mb[mb].cg_end = cg_offset + molb->nmol*cgs->nr;
699 cginfo_mb[mb].cg_mod = (bId ? 1 : molb->nmol)*cgs->nr;
700 snew(cginfo_mb[mb].cginfo, cginfo_mb[mb].cg_mod);
701 cginfo = cginfo_mb[mb].cginfo;
703 /* Set constraints flags for constrained atoms */
704 snew(a_con, molt->atoms.nr);
705 for (ftype = 0; ftype < F_NRE; ftype++)
707 if (interaction_function[ftype].flags & IF_CONSTRAINT)
712 for (ia = 0; ia < molt->ilist[ftype].size(); ia += 1+nral)
716 for (a = 0; a < nral; a++)
718 a_con[molt->ilist[ftype].iatoms[ia+1+a]] =
719 (ftype == F_SETTLE ? acSETTLE : acCONSTRAINT);
725 for (m = 0; m < (bId ? 1 : molb->nmol); m++)
728 int am = m*cgs->index[cgs->nr];
729 for (cg = 0; cg < cgs->nr; cg++)
732 a1 = cgs->index[cg+1];
734 /* Store the energy group in cginfo */
735 gid = getGroupType(mtop->groups, egcENER, a_offset+am+a0);
736 SET_CGINFO_GID(cginfo[cgm+cg], gid);
738 /* Check the intra/inter charge group exclusions */
739 if (a1-a0 > excl_nalloc)
741 excl_nalloc = a1 - a0;
742 srenew(bExcl, excl_nalloc);
744 /* bExclIntraAll: all intra cg interactions excluded
745 * bExclInter: any inter cg interactions excluded
747 bExclIntraAll = TRUE;
751 bHavePerturbedAtoms = FALSE;
752 for (ai = a0; ai < a1; ai++)
754 /* Check VDW and electrostatic interactions */
755 bHaveVDW = bHaveVDW || (type_VDW[molt->atoms.atom[ai].type] ||
756 type_VDW[molt->atoms.atom[ai].typeB]);
757 bHaveQ = bHaveQ || (molt->atoms.atom[ai].q != 0 ||
758 molt->atoms.atom[ai].qB != 0);
760 bHavePerturbedAtoms = bHavePerturbedAtoms || (PERTURBED(molt->atoms.atom[ai]) != 0);
762 /* Clear the exclusion list for atom ai */
763 for (aj = a0; aj < a1; aj++)
765 bExcl[aj-a0] = FALSE;
767 /* Loop over all the exclusions of atom ai */
768 for (j = excl->index[ai]; j < excl->index[ai+1]; j++)
771 if (aj < a0 || aj >= a1)
780 /* Check if ai excludes a0 to a1 */
781 for (aj = a0; aj < a1; aj++)
785 bExclIntraAll = FALSE;
792 SET_CGINFO_CONSTR(cginfo[cgm+cg]);
795 SET_CGINFO_SETTLE(cginfo[cgm+cg]);
803 SET_CGINFO_EXCL_INTRA(cginfo[cgm+cg]);
807 SET_CGINFO_EXCL_INTER(cginfo[cgm+cg]);
809 if (a1 - a0 > MAX_CHARGEGROUP_SIZE)
811 /* The size in cginfo is currently only read with DD */
812 gmx_fatal(FARGS, "A charge group has size %d which is larger than the limit of %d atoms", a1-a0, MAX_CHARGEGROUP_SIZE);
816 SET_CGINFO_HAS_VDW(cginfo[cgm+cg]);
820 SET_CGINFO_HAS_Q(cginfo[cgm+cg]);
822 if (bHavePerturbedAtoms && fr->efep != efepNO)
824 SET_CGINFO_FEP(cginfo[cgm+cg]);
825 *bFEP_NonBonded = TRUE;
827 /* Store the charge group size */
828 SET_CGINFO_NATOMS(cginfo[cgm+cg], a1-a0);
830 if (!bExclIntraAll || bExclInter)
832 *bExcl_IntraCGAll_InterCGNone = FALSE;
839 cg_offset += molb->nmol*cgs->nr;
840 a_offset += molb->nmol*cgs->index[cgs->nr];
845 /* the solvent optimizer is called after the QM is initialized,
846 * because we don't want to have the QM subsystemto become an
850 check_solvent(fplog, mtop, fr, cginfo_mb);
852 if (getenv("GMX_NO_SOLV_OPT"))
856 fprintf(fplog, "Found environment variable GMX_NO_SOLV_OPT.\n"
857 "Disabling all solvent optimization\n");
859 fr->solvent_opt = esolNO;
863 fr->solvent_opt = esolNO;
865 if (!fr->solvent_opt)
867 for (size_t mb = 0; mb < mtop->molblock.size(); mb++)
869 for (cg = 0; cg < cginfo_mb[mb].cg_mod; cg++)
871 SET_CGINFO_SOLOPT(cginfo_mb[mb].cginfo[cg], esolNO);
879 static int *cginfo_expand(int nmb, cginfo_mb_t *cgi_mb)
884 ncg = cgi_mb[nmb-1].cg_end;
887 for (cg = 0; cg < ncg; cg++)
889 while (cg >= cgi_mb[mb].cg_end)
894 cgi_mb[mb].cginfo[(cg - cgi_mb[mb].cg_start) % cgi_mb[mb].cg_mod];
900 static void done_cginfo_mb(cginfo_mb_t *cginfo_mb, int numMolBlocks)
902 if (cginfo_mb == nullptr)
906 for (int mb = 0; mb < numMolBlocks; ++mb)
908 sfree(cginfo_mb[mb].cginfo);
913 /* Sets the sum of charges (squared) and C6 in the system in fr.
914 * Returns whether the system has a net charge.
916 static bool set_chargesum(FILE *log, t_forcerec *fr, const gmx_mtop_t *mtop)
918 /*This now calculates sum for q and c6*/
919 double qsum, q2sum, q, c6sum, c6;
924 for (const gmx_molblock_t &molb : mtop->molblock)
926 int nmol = molb.nmol;
927 const t_atoms *atoms = &mtop->moltype[molb.type].atoms;
928 for (int i = 0; i < atoms->nr; i++)
930 q = atoms->atom[i].q;
933 c6 = mtop->ffparams.iparams[atoms->atom[i].type*(mtop->ffparams.atnr+1)].lj.c6;
938 fr->q2sum[0] = q2sum;
939 fr->c6sum[0] = c6sum;
941 if (fr->efep != efepNO)
946 for (const gmx_molblock_t &molb : mtop->molblock)
948 int nmol = molb.nmol;
949 const t_atoms *atoms = &mtop->moltype[molb.type].atoms;
950 for (int i = 0; i < atoms->nr; i++)
952 q = atoms->atom[i].qB;
955 c6 = mtop->ffparams.iparams[atoms->atom[i].typeB*(mtop->ffparams.atnr+1)].lj.c6;
959 fr->q2sum[1] = q2sum;
960 fr->c6sum[1] = c6sum;
965 fr->qsum[1] = fr->qsum[0];
966 fr->q2sum[1] = fr->q2sum[0];
967 fr->c6sum[1] = fr->c6sum[0];
971 if (fr->efep == efepNO)
973 fprintf(log, "System total charge: %.3f\n", fr->qsum[0]);
977 fprintf(log, "System total charge, top. A: %.3f top. B: %.3f\n",
978 fr->qsum[0], fr->qsum[1]);
982 /* A cut-off of 1e-4 is used to catch rounding errors due to ascii input */
983 return (std::abs(fr->qsum[0]) > 1e-4 ||
984 std::abs(fr->qsum[1]) > 1e-4);
987 void update_forcerec(t_forcerec *fr, matrix box)
989 if (fr->ic->eeltype == eelGRF)
991 calc_rffac(nullptr, fr->ic->eeltype, fr->ic->epsilon_r, fr->ic->epsilon_rf,
992 fr->ic->rcoulomb, fr->temp, fr->zsquare, box,
993 &fr->ic->k_rf, &fr->ic->c_rf);
997 void set_avcsixtwelve(FILE *fplog, t_forcerec *fr, const gmx_mtop_t *mtop)
999 const t_atoms *atoms, *atoms_tpi;
1000 const t_blocka *excl;
1001 int nmolc, i, j, tpi, tpj, j1, j2, k, nexcl, q;
1002 int64_t npair, npair_ij, tmpi, tmpj;
1003 double csix, ctwelve;
1004 int ntp, *typecount;
1007 real *nbfp_comb = nullptr;
1013 /* For LJ-PME, we want to correct for the difference between the
1014 * actual C6 values and the C6 values used by the LJ-PME based on
1015 * combination rules. */
1017 if (EVDW_PME(fr->ic->vdwtype))
1019 nbfp_comb = mk_nbfp_combination_rule(&mtop->ffparams,
1020 (fr->ljpme_combination_rule == eljpmeLB) ? eCOMB_ARITHMETIC : eCOMB_GEOMETRIC);
1021 for (tpi = 0; tpi < ntp; ++tpi)
1023 for (tpj = 0; tpj < ntp; ++tpj)
1025 C6(nbfp_comb, ntp, tpi, tpj) =
1026 C6(nbfp, ntp, tpi, tpj) - C6(nbfp_comb, ntp, tpi, tpj);
1027 C12(nbfp_comb, ntp, tpi, tpj) = C12(nbfp, ntp, tpi, tpj);
1032 for (q = 0; q < (fr->efep == efepNO ? 1 : 2); q++)
1040 /* Count the types so we avoid natoms^2 operations */
1041 snew(typecount, ntp);
1042 gmx_mtop_count_atomtypes(mtop, q, typecount);
1044 for (tpi = 0; tpi < ntp; tpi++)
1046 for (tpj = tpi; tpj < ntp; tpj++)
1048 tmpi = typecount[tpi];
1049 tmpj = typecount[tpj];
1052 npair_ij = tmpi*tmpj;
1056 npair_ij = tmpi*(tmpi - 1)/2;
1060 /* nbfp now includes the 6.0 derivative prefactor */
1061 csix += npair_ij*BHAMC(nbfp, ntp, tpi, tpj)/6.0;
1065 /* nbfp now includes the 6.0/12.0 derivative prefactors */
1066 csix += npair_ij* C6(nbfp, ntp, tpi, tpj)/6.0;
1067 ctwelve += npair_ij* C12(nbfp, ntp, tpi, tpj)/12.0;
1073 /* Subtract the excluded pairs.
1074 * The main reason for substracting exclusions is that in some cases
1075 * some combinations might never occur and the parameters could have
1076 * any value. These unused values should not influence the dispersion
1079 for (const gmx_molblock_t &molb : mtop->molblock)
1081 int nmol = molb.nmol;
1082 atoms = &mtop->moltype[molb.type].atoms;
1083 excl = &mtop->moltype[molb.type].excls;
1084 for (int i = 0; (i < atoms->nr); i++)
1088 tpi = atoms->atom[i].type;
1092 tpi = atoms->atom[i].typeB;
1094 j1 = excl->index[i];
1095 j2 = excl->index[i+1];
1096 for (j = j1; j < j2; j++)
1103 tpj = atoms->atom[k].type;
1107 tpj = atoms->atom[k].typeB;
1111 /* nbfp now includes the 6.0 derivative prefactor */
1112 csix -= nmol*BHAMC(nbfp, ntp, tpi, tpj)/6.0;
1116 /* nbfp now includes the 6.0/12.0 derivative prefactors */
1117 csix -= nmol*C6 (nbfp, ntp, tpi, tpj)/6.0;
1118 ctwelve -= nmol*C12(nbfp, ntp, tpi, tpj)/12.0;
1128 /* Only correct for the interaction of the test particle
1129 * with the rest of the system.
1132 &mtop->moltype[mtop->molblock.back().type].atoms;
1135 for (size_t mb = 0; mb < mtop->molblock.size(); mb++)
1137 const gmx_molblock_t &molb = mtop->molblock[mb];
1138 atoms = &mtop->moltype[molb.type].atoms;
1139 for (j = 0; j < atoms->nr; j++)
1142 /* Remove the interaction of the test charge group
1145 if (mb == mtop->molblock.size() - 1)
1149 if (mb == 0 && molb.nmol == 1)
1151 gmx_fatal(FARGS, "Old format tpr with TPI, please generate a new tpr file");
1156 tpj = atoms->atom[j].type;
1160 tpj = atoms->atom[j].typeB;
1162 for (i = 0; i < fr->n_tpi; i++)
1166 tpi = atoms_tpi->atom[i].type;
1170 tpi = atoms_tpi->atom[i].typeB;
1174 /* nbfp now includes the 6.0 derivative prefactor */
1175 csix += nmolc*BHAMC(nbfp, ntp, tpi, tpj)/6.0;
1179 /* nbfp now includes the 6.0/12.0 derivative prefactors */
1180 csix += nmolc*C6 (nbfp, ntp, tpi, tpj)/6.0;
1181 ctwelve += nmolc*C12(nbfp, ntp, tpi, tpj)/12.0;
1188 if (npair - nexcl <= 0 && fplog)
1190 fprintf(fplog, "\nWARNING: There are no atom pairs for dispersion correction\n\n");
1196 csix /= npair - nexcl;
1197 ctwelve /= npair - nexcl;
1201 fprintf(debug, "Counted %d exclusions\n", nexcl);
1202 fprintf(debug, "Average C6 parameter is: %10g\n", csix);
1203 fprintf(debug, "Average C12 parameter is: %10g\n", ctwelve);
1205 fr->avcsix[q] = csix;
1206 fr->avctwelve[q] = ctwelve;
1209 if (EVDW_PME(fr->ic->vdwtype))
1214 if (fplog != nullptr)
1216 if (fr->eDispCorr == edispcAllEner ||
1217 fr->eDispCorr == edispcAllEnerPres)
1219 fprintf(fplog, "Long Range LJ corr.: <C6> %10.4e, <C12> %10.4e\n",
1220 fr->avcsix[0], fr->avctwelve[0]);
1224 fprintf(fplog, "Long Range LJ corr.: <C6> %10.4e\n", fr->avcsix[0]);
1230 static real calcBuckinghamBMax(FILE *fplog, const gmx_mtop_t *mtop)
1232 const t_atoms *at1, *at2;
1233 int i, j, tpi, tpj, ntypes;
1238 fprintf(fplog, "Determining largest Buckingham b parameter for table\n");
1240 ntypes = mtop->ffparams.atnr;
1243 real bham_b_max = 0;
1244 for (size_t mt1 = 0; mt1 < mtop->moltype.size(); mt1++)
1246 at1 = &mtop->moltype[mt1].atoms;
1247 for (i = 0; (i < at1->nr); i++)
1249 tpi = at1->atom[i].type;
1252 gmx_fatal(FARGS, "Atomtype[%d] = %d, maximum = %d", i, tpi, ntypes);
1255 for (size_t mt2 = mt1; mt2 < mtop->moltype.size(); mt2++)
1257 at2 = &mtop->moltype[mt2].atoms;
1258 for (j = 0; (j < at2->nr); j++)
1260 tpj = at2->atom[j].type;
1263 gmx_fatal(FARGS, "Atomtype[%d] = %d, maximum = %d", j, tpj, ntypes);
1265 b = mtop->ffparams.iparams[tpi*ntypes + tpj].bham.b;
1270 if ((b < bmin) || (bmin == -1))
1280 fprintf(fplog, "Buckingham b parameters, min: %g, max: %g\n",
1287 static void make_nbf_tables(FILE *fp,
1288 const interaction_const_t *ic, real rtab,
1289 const char *tabfn, char *eg1, char *eg2,
1295 if (tabfn == nullptr)
1299 fprintf(debug, "No table file name passed, can not read table, can not do non-bonded interactions\n");
1304 sprintf(buf, "%s", tabfn);
1307 /* Append the two energy group names */
1308 sprintf(buf + strlen(tabfn) - strlen(ftp2ext(efXVG)) - 1, "_%s_%s.%s",
1309 eg1, eg2, ftp2ext(efXVG));
1311 nbl->table_elec_vdw = make_tables(fp, ic, buf, rtab, 0);
1312 /* Copy the contents of the table to separate coulomb and LJ tables too,
1313 * to improve cache performance.
1315 /* For performance reasons we want
1316 * the table data to be aligned to 16-byte. The pointers could be freed
1317 * but currently aren't.
1319 snew(nbl->table_elec, 1);
1320 nbl->table_elec->interaction = GMX_TABLE_INTERACTION_ELEC;
1321 nbl->table_elec->format = nbl->table_elec_vdw->format;
1322 nbl->table_elec->r = nbl->table_elec_vdw->r;
1323 nbl->table_elec->n = nbl->table_elec_vdw->n;
1324 nbl->table_elec->scale = nbl->table_elec_vdw->scale;
1325 nbl->table_elec->formatsize = nbl->table_elec_vdw->formatsize;
1326 nbl->table_elec->ninteractions = 1;
1327 nbl->table_elec->stride = nbl->table_elec->formatsize * nbl->table_elec->ninteractions;
1328 snew_aligned(nbl->table_elec->data, nbl->table_elec->stride*(nbl->table_elec->n+1), 32);
1330 snew(nbl->table_vdw, 1);
1331 nbl->table_vdw->interaction = GMX_TABLE_INTERACTION_VDWREP_VDWDISP;
1332 nbl->table_vdw->format = nbl->table_elec_vdw->format;
1333 nbl->table_vdw->r = nbl->table_elec_vdw->r;
1334 nbl->table_vdw->n = nbl->table_elec_vdw->n;
1335 nbl->table_vdw->scale = nbl->table_elec_vdw->scale;
1336 nbl->table_vdw->formatsize = nbl->table_elec_vdw->formatsize;
1337 nbl->table_vdw->ninteractions = 2;
1338 nbl->table_vdw->stride = nbl->table_vdw->formatsize * nbl->table_vdw->ninteractions;
1339 snew_aligned(nbl->table_vdw->data, nbl->table_vdw->stride*(nbl->table_vdw->n+1), 32);
1341 /* NOTE: Using a single i-loop here leads to mix-up of data in table_vdw
1342 * with (at least) gcc 6.2, 6.3 and 6.4 when compiled with -O3 and AVX
1344 for (i = 0; i <= nbl->table_elec_vdw->n; i++)
1346 for (j = 0; j < 4; j++)
1348 nbl->table_elec->data[4*i+j] = nbl->table_elec_vdw->data[12*i+j];
1351 for (i = 0; i <= nbl->table_elec_vdw->n; i++)
1353 for (j = 0; j < 8; j++)
1355 nbl->table_vdw->data[8*i+j] = nbl->table_elec_vdw->data[12*i+4+j];
1360 /*!\brief If there's bonded interactions of type \c ftype1 or \c
1361 * ftype2 present in the topology, build an array of the number of
1362 * interactions present for each bonded interaction index found in the
1365 * \c ftype1 or \c ftype2 may be set to -1 to disable seeking for a
1366 * valid type with that parameter.
1368 * \c count will be reallocated as necessary to fit the largest bonded
1369 * interaction index found, and its current size will be returned in
1370 * \c ncount. It will contain zero for every bonded interaction index
1371 * for which no interactions are present in the topology.
1373 static void count_tables(int ftype1, int ftype2, const gmx_mtop_t *mtop,
1374 int *ncount, int **count)
1376 int ftype, i, j, tabnr;
1378 // Loop over all moleculetypes
1379 for (const gmx_moltype_t &molt : mtop->moltype)
1381 // Loop over all interaction types
1382 for (ftype = 0; ftype < F_NRE; ftype++)
1384 // If the current interaction type is one of the types whose tables we're trying to count...
1385 if (ftype == ftype1 || ftype == ftype2)
1387 const InteractionList &il = molt.ilist[ftype];
1388 const int stride = 1 + NRAL(ftype);
1389 // ... and there are actually some interactions for this type
1390 for (i = 0; i < il.size(); i += stride)
1392 // Find out which table index the user wanted
1393 tabnr = mtop->ffparams.iparams[il.iatoms[i]].tab.table;
1396 gmx_fatal(FARGS, "A bonded table number is smaller than 0: %d\n", tabnr);
1398 // Make room for this index in the data structure
1399 if (tabnr >= *ncount)
1401 srenew(*count, tabnr+1);
1402 for (j = *ncount; j < tabnr+1; j++)
1408 // Record that this table index is used and must have a valid file
1416 /*!\brief If there's bonded interactions of flavour \c tabext and type
1417 * \c ftype1 or \c ftype2 present in the topology, seek them in the
1418 * list of filenames passed to mdrun, and make bonded tables from
1421 * \c ftype1 or \c ftype2 may be set to -1 to disable seeking for a
1422 * valid type with that parameter.
1424 * A fatal error occurs if no matching filename is found.
1426 static bondedtable_t *make_bonded_tables(FILE *fplog,
1427 int ftype1, int ftype2,
1428 const gmx_mtop_t *mtop,
1429 gmx::ArrayRef<const std::string> tabbfnm,
1439 count_tables(ftype1, ftype2, mtop, &ncount, &count);
1441 // Are there any relevant tabulated bond interactions?
1445 for (int i = 0; i < ncount; i++)
1447 // Do any interactions exist that requires this table?
1450 // This pattern enforces the current requirement that
1451 // table filenames end in a characteristic sequence
1452 // before the file type extension, and avoids table 13
1453 // being recognized and used for table 1.
1454 std::string patternToFind = gmx::formatString("_%s%d.%s", tabext, i, ftp2ext(efXVG));
1455 bool madeTable = false;
1456 for (gmx::index j = 0; j < tabbfnm.size() && !madeTable; ++j)
1458 if (gmx::endsWith(tabbfnm[j], patternToFind))
1460 // Finally read the table from the file found
1461 tab[i] = make_bonded_table(fplog, tabbfnm[j].c_str(), NRAL(ftype1)-2);
1467 bool isPlural = (ftype2 != -1);
1468 gmx_fatal(FARGS, "Tabulated interaction of type '%s%s%s' with index %d cannot be used because no table file whose name matched '%s' was passed via the gmx mdrun -tableb command-line option.",
1469 interaction_function[ftype1].longname,
1470 isPlural ? "' or '" : "",
1471 isPlural ? interaction_function[ftype2].longname : "",
1473 patternToFind.c_str());
1483 void forcerec_set_ranges(t_forcerec *fr,
1484 int ncg_home, int ncg_force,
1486 int natoms_force_constr, int natoms_f_novirsum)
1491 /* fr->ncg_force is unused in the standard code,
1492 * but it can be useful for modified code dealing with charge groups.
1494 fr->ncg_force = ncg_force;
1495 fr->natoms_force = natoms_force;
1496 fr->natoms_force_constr = natoms_force_constr;
1498 if (fr->natoms_force_constr > fr->nalloc_force)
1500 fr->nalloc_force = over_alloc_dd(fr->natoms_force_constr);
1503 if (fr->haveDirectVirialContributions)
1505 fr->forceBufferForDirectVirialContributions->resize(natoms_f_novirsum);
1509 static real cutoff_inf(real cutoff)
1513 cutoff = GMX_CUTOFF_INF;
1519 gmx_bool nbnxn_simd_supported(const gmx::MDLogger &mdlog,
1520 const t_inputrec *ir)
1522 if (ir->vdwtype == evdwPME && ir->ljpme_combination_rule == eljpmeLB)
1524 /* LJ PME with LB combination rule does 7 mesh operations.
1525 * This so slow that we don't compile SIMD non-bonded kernels
1527 GMX_LOG(mdlog.warning).asParagraph().appendText("LJ-PME with Lorentz-Berthelot is not supported with SIMD kernels, falling back to plain C kernels");
1535 static void pick_nbnxn_kernel_cpu(const t_inputrec gmx_unused *ir,
1538 const gmx_hw_info_t gmx_unused &hardwareInfo)
1540 *kernel_type = nbnxnk4x4_PlainC;
1541 *ewald_excl = ewaldexclTable;
1545 #ifdef GMX_NBNXN_SIMD_4XN
1546 *kernel_type = nbnxnk4xN_SIMD_4xN;
1548 #ifdef GMX_NBNXN_SIMD_2XNN
1549 *kernel_type = nbnxnk4xN_SIMD_2xNN;
1552 #if defined GMX_NBNXN_SIMD_2XNN && defined GMX_NBNXN_SIMD_4XN
1553 /* We need to choose if we want 2x(N+N) or 4xN kernels.
1554 * This is based on the SIMD acceleration choice and CPU information
1555 * detected at runtime.
1557 * 4xN calculates more (zero) interactions, but has less pair-search
1558 * work and much better kernel instruction scheduling.
1560 * Up till now we have only seen that on Intel Sandy/Ivy Bridge,
1561 * which doesn't have FMA, both the analytical and tabulated Ewald
1562 * kernels have similar pair rates for 4x8 and 2x(4+4), so we choose
1563 * 2x(4+4) because it results in significantly fewer pairs.
1564 * For RF, the raw pair rate of the 4x8 kernel is higher than 2x(4+4),
1565 * 10% with HT, 50% without HT. As we currently don't detect the actual
1566 * use of HT, use 4x8 to avoid a potential performance hit.
1567 * On Intel Haswell 4x8 is always faster.
1569 *kernel_type = nbnxnk4xN_SIMD_4xN;
1571 #if !GMX_SIMD_HAVE_FMA
1572 if (EEL_PME_EWALD(ir->coulombtype) ||
1573 EVDW_PME(ir->vdwtype))
1575 /* We have Ewald kernels without FMA (Intel Sandy/Ivy Bridge).
1576 * There are enough instructions to make 2x(4+4) efficient.
1578 *kernel_type = nbnxnk4xN_SIMD_2xNN;
1581 if (hardwareInfo.haveAmdZenCpu)
1583 /* One 256-bit FMA per cycle makes 2xNN faster */
1584 *kernel_type = nbnxnk4xN_SIMD_2xNN;
1586 #endif /* GMX_NBNXN_SIMD_2XNN && GMX_NBNXN_SIMD_4XN */
1589 if (getenv("GMX_NBNXN_SIMD_4XN") != nullptr)
1591 #ifdef GMX_NBNXN_SIMD_4XN
1592 *kernel_type = nbnxnk4xN_SIMD_4xN;
1594 gmx_fatal(FARGS, "SIMD 4xN kernels requested, but GROMACS has been compiled without support for these kernels");
1597 if (getenv("GMX_NBNXN_SIMD_2XNN") != nullptr)
1599 #ifdef GMX_NBNXN_SIMD_2XNN
1600 *kernel_type = nbnxnk4xN_SIMD_2xNN;
1602 gmx_fatal(FARGS, "SIMD 2x(N+N) kernels requested, but GROMACS has been compiled without support for these kernels");
1606 /* Analytical Ewald exclusion correction is only an option in
1608 * Since table lookup's don't parallelize with SIMD, analytical
1609 * will probably always be faster for a SIMD width of 8 or more.
1610 * With FMA analytical is sometimes faster for a width if 4 as well.
1611 * In single precision, this is faster on Bulldozer.
1613 #if GMX_SIMD_REAL_WIDTH >= 8 || \
1614 (GMX_SIMD_REAL_WIDTH >= 4 && GMX_SIMD_HAVE_FMA && !GMX_DOUBLE)
1615 /* On AMD Zen, tabulated Ewald kernels are faster on all 4 combinations
1616 * of single or double precision and 128 or 256-bit AVX2.
1618 if (!hardwareInfo.haveAmdZenCpu)
1620 *ewald_excl = ewaldexclAnalytical;
1623 if (getenv("GMX_NBNXN_EWALD_TABLE") != nullptr)
1625 *ewald_excl = ewaldexclTable;
1627 if (getenv("GMX_NBNXN_EWALD_ANALYTICAL") != nullptr)
1629 *ewald_excl = ewaldexclAnalytical;
1637 const char *lookup_nbnxn_kernel_name(int kernel_type)
1639 const char *returnvalue = nullptr;
1640 switch (kernel_type)
1643 returnvalue = "not set";
1645 case nbnxnk4x4_PlainC:
1646 returnvalue = "plain C";
1648 case nbnxnk4xN_SIMD_4xN:
1649 case nbnxnk4xN_SIMD_2xNN:
1651 returnvalue = "SIMD";
1653 returnvalue = "not available";
1656 case nbnxnk8x8x8_GPU: returnvalue = "GPU"; break;
1657 case nbnxnk8x8x8_PlainC: returnvalue = "plain C"; break;
1661 gmx_fatal(FARGS, "Illegal kernel type selected");
1666 static void pick_nbnxn_kernel(const gmx::MDLogger &mdlog,
1667 gmx_bool use_simd_kernels,
1668 const gmx_hw_info_t &hardwareInfo,
1670 EmulateGpuNonbonded emulateGpu,
1671 const t_inputrec *ir,
1674 gmx_bool bDoNonbonded)
1676 assert(kernel_type);
1678 *kernel_type = nbnxnkNotSet;
1679 *ewald_excl = ewaldexclTable;
1681 if (emulateGpu == EmulateGpuNonbonded::Yes)
1683 *kernel_type = nbnxnk8x8x8_PlainC;
1687 GMX_LOG(mdlog.warning).asParagraph().appendText("Emulating a GPU run on the CPU (slow)");
1692 *kernel_type = nbnxnk8x8x8_GPU;
1695 if (*kernel_type == nbnxnkNotSet)
1697 if (use_simd_kernels &&
1698 nbnxn_simd_supported(mdlog, ir))
1700 pick_nbnxn_kernel_cpu(ir, kernel_type, ewald_excl, hardwareInfo);
1704 *kernel_type = nbnxnk4x4_PlainC;
1710 GMX_LOG(mdlog.info).asParagraph().appendTextFormatted(
1711 "Using %s %dx%d nonbonded short-range kernels",
1712 lookup_nbnxn_kernel_name(*kernel_type),
1713 nbnxn_kernel_to_cluster_i_size(*kernel_type),
1714 nbnxn_kernel_to_cluster_j_size(*kernel_type));
1716 if (nbnxnk4x4_PlainC == *kernel_type ||
1717 nbnxnk8x8x8_PlainC == *kernel_type)
1719 GMX_LOG(mdlog.warning).asParagraph().appendTextFormatted(
1720 "WARNING: Using the slow %s kernels. This should\n"
1721 "not happen during routine usage on supported platforms.",
1722 lookup_nbnxn_kernel_name(*kernel_type));
1727 /*! \brief Print Coulomb Ewald citations and set ewald coefficients */
1728 static void initCoulombEwaldParameters(FILE *fp, const t_inputrec *ir,
1729 bool systemHasNetCharge,
1730 interaction_const_t *ic)
1732 if (!EEL_PME_EWALD(ir->coulombtype))
1739 fprintf(fp, "Will do PME sum in reciprocal space for electrostatic interactions.\n");
1741 if (ir->coulombtype == eelP3M_AD)
1743 please_cite(fp, "Hockney1988");
1744 please_cite(fp, "Ballenegger2012");
1748 please_cite(fp, "Essmann95a");
1751 if (ir->ewald_geometry == eewg3DC)
1755 fprintf(fp, "Using the Ewald3DC correction for systems with a slab geometry%s.\n",
1756 systemHasNetCharge ? " and net charge" : "");
1758 please_cite(fp, "In-Chul99a");
1759 if (systemHasNetCharge)
1761 please_cite(fp, "Ballenegger2009");
1766 ic->ewaldcoeff_q = calc_ewaldcoeff_q(ir->rcoulomb, ir->ewald_rtol);
1769 fprintf(fp, "Using a Gaussian width (1/beta) of %g nm for Ewald\n",
1770 1/ic->ewaldcoeff_q);
1773 if (ic->coulomb_modifier == eintmodPOTSHIFT)
1775 GMX_RELEASE_ASSERT(ic->rcoulomb != 0, "Cutoff radius cannot be zero");
1776 ic->sh_ewald = std::erfc(ic->ewaldcoeff_q*ic->rcoulomb) / ic->rcoulomb;
1784 /*! \brief Print Van der Waals Ewald citations and set ewald coefficients */
1785 static void initVdwEwaldParameters(FILE *fp, const t_inputrec *ir,
1786 interaction_const_t *ic)
1788 if (!EVDW_PME(ir->vdwtype))
1795 fprintf(fp, "Will do PME sum in reciprocal space for LJ dispersion interactions.\n");
1796 please_cite(fp, "Essmann95a");
1798 ic->ewaldcoeff_lj = calc_ewaldcoeff_lj(ir->rvdw, ir->ewald_rtol_lj);
1801 fprintf(fp, "Using a Gaussian width (1/beta) of %g nm for LJ Ewald\n",
1802 1/ic->ewaldcoeff_lj);
1805 if (ic->vdw_modifier == eintmodPOTSHIFT)
1807 real crc2 = gmx::square(ic->ewaldcoeff_lj*ic->rvdw);
1808 ic->sh_lj_ewald = (std::exp(-crc2)*(1 + crc2 + 0.5*crc2*crc2) - 1)/gmx::power6(ic->rvdw);
1812 ic->sh_lj_ewald = 0;
1816 gmx_bool uses_simple_tables(int cutoff_scheme,
1817 nonbonded_verlet_t *nbv,
1820 gmx_bool bUsesSimpleTables = TRUE;
1823 switch (cutoff_scheme)
1826 bUsesSimpleTables = TRUE;
1829 assert(nullptr != nbv);
1830 grp_index = (group < 0) ? 0 : (nbv->ngrp - 1);
1831 bUsesSimpleTables = nbnxn_kernel_pairlist_simple(nbv->grp[grp_index].kernel_type);
1834 gmx_incons("unimplemented");
1836 return bUsesSimpleTables;
1839 static void init_ewald_f_table(interaction_const_t *ic,
1844 /* Get the Ewald table spacing based on Coulomb and/or LJ
1845 * Ewald coefficients and rtol.
1847 ic->tabq_scale = ewald_spline3_table_scale(ic);
1849 if (ic->cutoff_scheme == ecutsVERLET)
1851 maxr = ic->rcoulomb;
1855 maxr = std::max(ic->rcoulomb, rtab);
1857 ic->tabq_size = static_cast<int>(maxr*ic->tabq_scale) + 2;
1859 sfree_aligned(ic->tabq_coul_FDV0);
1860 sfree_aligned(ic->tabq_coul_F);
1861 sfree_aligned(ic->tabq_coul_V);
1863 sfree_aligned(ic->tabq_vdw_FDV0);
1864 sfree_aligned(ic->tabq_vdw_F);
1865 sfree_aligned(ic->tabq_vdw_V);
1867 if (EEL_PME_EWALD(ic->eeltype))
1869 /* Create the original table data in FDV0 */
1870 snew_aligned(ic->tabq_coul_FDV0, ic->tabq_size*4, 32);
1871 snew_aligned(ic->tabq_coul_F, ic->tabq_size, 32);
1872 snew_aligned(ic->tabq_coul_V, ic->tabq_size, 32);
1873 table_spline3_fill_ewald_lr(ic->tabq_coul_F, ic->tabq_coul_V, ic->tabq_coul_FDV0,
1874 ic->tabq_size, 1/ic->tabq_scale, ic->ewaldcoeff_q, v_q_ewald_lr);
1877 if (EVDW_PME(ic->vdwtype))
1879 snew_aligned(ic->tabq_vdw_FDV0, ic->tabq_size*4, 32);
1880 snew_aligned(ic->tabq_vdw_F, ic->tabq_size, 32);
1881 snew_aligned(ic->tabq_vdw_V, ic->tabq_size, 32);
1882 table_spline3_fill_ewald_lr(ic->tabq_vdw_F, ic->tabq_vdw_V, ic->tabq_vdw_FDV0,
1883 ic->tabq_size, 1/ic->tabq_scale, ic->ewaldcoeff_lj, v_lj_ewald_lr);
1887 void init_interaction_const_tables(FILE *fp,
1888 interaction_const_t *ic,
1891 if (EEL_PME_EWALD(ic->eeltype) || EVDW_PME(ic->vdwtype))
1893 init_ewald_f_table(ic, rtab);
1897 fprintf(fp, "Initialized non-bonded Ewald correction tables, spacing: %.2e size: %d\n\n",
1898 1/ic->tabq_scale, ic->tabq_size);
1903 static void clear_force_switch_constants(shift_consts_t *sc)
1910 static void force_switch_constants(real p,
1914 /* Here we determine the coefficient for shifting the force to zero
1915 * between distance rsw and the cut-off rc.
1916 * For a potential of r^-p, we have force p*r^-(p+1).
1917 * But to save flops we absorb p in the coefficient.
1919 * force/p = r^-(p+1) + c2*r^2 + c3*r^3
1920 * potential = r^-p + c2/3*r^3 + c3/4*r^4 + cpot
1922 sc->c2 = ((p + 1)*rsw - (p + 4)*rc)/(pow(rc, p + 2)*gmx::square(rc - rsw));
1923 sc->c3 = -((p + 1)*rsw - (p + 3)*rc)/(pow(rc, p + 2)*gmx::power3(rc - rsw));
1924 sc->cpot = -pow(rc, -p) + p*sc->c2/3*gmx::power3(rc - rsw) + p*sc->c3/4*gmx::power4(rc - rsw);
1927 static void potential_switch_constants(real rsw, real rc,
1928 switch_consts_t *sc)
1930 /* The switch function is 1 at rsw and 0 at rc.
1931 * The derivative and second derivate are zero at both ends.
1932 * rsw = max(r - r_switch, 0)
1933 * sw = 1 + c3*rsw^3 + c4*rsw^4 + c5*rsw^5
1934 * dsw = 3*c3*rsw^2 + 4*c4*rsw^3 + 5*c5*rsw^4
1935 * force = force*dsw - potential*sw
1938 sc->c3 = -10/gmx::power3(rc - rsw);
1939 sc->c4 = 15/gmx::power4(rc - rsw);
1940 sc->c5 = -6/gmx::power5(rc - rsw);
1943 /*! \brief Construct interaction constants
1945 * This data is used (particularly) by search and force code for
1946 * short-range interactions. Many of these are constant for the whole
1947 * simulation; some are constant only after PME tuning completes.
1950 init_interaction_const(FILE *fp,
1951 interaction_const_t **interaction_const,
1952 const t_inputrec *ir,
1953 const gmx_mtop_t *mtop,
1954 bool systemHasNetCharge)
1956 interaction_const_t *ic;
1960 ic->cutoff_scheme = ir->cutoff_scheme;
1962 /* Just allocate something so we can free it */
1963 snew_aligned(ic->tabq_coul_FDV0, 16, 32);
1964 snew_aligned(ic->tabq_coul_F, 16, 32);
1965 snew_aligned(ic->tabq_coul_V, 16, 32);
1968 ic->vdwtype = ir->vdwtype;
1969 ic->vdw_modifier = ir->vdw_modifier;
1970 ic->reppow = mtop->ffparams.reppow;
1971 ic->rvdw = cutoff_inf(ir->rvdw);
1972 ic->rvdw_switch = ir->rvdw_switch;
1973 ic->ljpme_comb_rule = ir->ljpme_combination_rule;
1974 ic->useBuckingham = (mtop->ffparams.functype[0] == F_BHAM);
1975 if (ic->useBuckingham)
1977 ic->buckinghamBMax = calcBuckinghamBMax(fp, mtop);
1980 initVdwEwaldParameters(fp, ir, ic);
1982 clear_force_switch_constants(&ic->dispersion_shift);
1983 clear_force_switch_constants(&ic->repulsion_shift);
1985 switch (ic->vdw_modifier)
1987 case eintmodPOTSHIFT:
1988 /* Only shift the potential, don't touch the force */
1989 ic->dispersion_shift.cpot = -1.0/gmx::power6(ic->rvdw);
1990 ic->repulsion_shift.cpot = -1.0/gmx::power12(ic->rvdw);
1992 case eintmodFORCESWITCH:
1993 /* Switch the force, switch and shift the potential */
1994 force_switch_constants(6.0, ic->rvdw_switch, ic->rvdw,
1995 &ic->dispersion_shift);
1996 force_switch_constants(12.0, ic->rvdw_switch, ic->rvdw,
1997 &ic->repulsion_shift);
1999 case eintmodPOTSWITCH:
2000 /* Switch the potential and force */
2001 potential_switch_constants(ic->rvdw_switch, ic->rvdw,
2005 case eintmodEXACTCUTOFF:
2006 /* Nothing to do here */
2009 gmx_incons("unimplemented potential modifier");
2012 ic->sh_invrc6 = -ic->dispersion_shift.cpot;
2014 /* Electrostatics */
2015 ic->eeltype = ir->coulombtype;
2016 ic->coulomb_modifier = ir->coulomb_modifier;
2017 ic->rcoulomb = cutoff_inf(ir->rcoulomb);
2018 ic->rcoulomb_switch = ir->rcoulomb_switch;
2019 ic->epsilon_r = ir->epsilon_r;
2021 /* Set the Coulomb energy conversion factor */
2022 if (ic->epsilon_r != 0)
2024 ic->epsfac = ONE_4PI_EPS0/ic->epsilon_r;
2028 /* eps = 0 is infinite dieletric: no Coulomb interactions */
2032 /* Reaction-field */
2033 if (EEL_RF(ic->eeltype))
2035 ic->epsilon_rf = ir->epsilon_rf;
2036 /* Generalized reaction field parameters are updated every step */
2037 if (ic->eeltype != eelGRF)
2039 calc_rffac(fp, ic->eeltype, ic->epsilon_r, ic->epsilon_rf,
2040 ic->rcoulomb, 0, 0, nullptr,
2041 &ic->k_rf, &ic->c_rf);
2044 if (ir->cutoff_scheme == ecutsGROUP && ic->eeltype == eelRF_ZERO)
2046 /* grompp should have done this, but this scheme is obsolete */
2047 ic->coulomb_modifier = eintmodEXACTCUTOFF;
2052 /* For plain cut-off we might use the reaction-field kernels */
2053 ic->epsilon_rf = ic->epsilon_r;
2055 if (ir->coulomb_modifier == eintmodPOTSHIFT)
2057 ic->c_rf = 1/ic->rcoulomb;
2065 initCoulombEwaldParameters(fp, ir, systemHasNetCharge, ic);
2069 real dispersion_shift;
2071 dispersion_shift = ic->dispersion_shift.cpot;
2072 if (EVDW_PME(ic->vdwtype))
2074 dispersion_shift -= ic->sh_lj_ewald;
2076 fprintf(fp, "Potential shift: LJ r^-12: %.3e r^-6: %.3e",
2077 ic->repulsion_shift.cpot, dispersion_shift);
2079 if (ic->eeltype == eelCUT)
2081 fprintf(fp, ", Coulomb %.e", -ic->c_rf);
2083 else if (EEL_PME(ic->eeltype))
2085 fprintf(fp, ", Ewald %.3e", -ic->sh_ewald);
2090 *interaction_const = ic;
2094 done_interaction_const(interaction_const_t *interaction_const)
2096 sfree_aligned(interaction_const->tabq_coul_FDV0);
2097 sfree_aligned(interaction_const->tabq_coul_F);
2098 sfree_aligned(interaction_const->tabq_coul_V);
2099 sfree(interaction_const);
2102 static void init_nb_verlet(const gmx::MDLogger &mdlog,
2103 nonbonded_verlet_t **nb_verlet,
2104 gmx_bool bFEP_NonBonded,
2105 const t_inputrec *ir,
2106 const t_forcerec *fr,
2107 const t_commrec *cr,
2108 const gmx_hw_info_t &hardwareInfo,
2109 const gmx_device_info_t *deviceInfo,
2110 const gmx_mtop_t *mtop,
2113 nonbonded_verlet_t *nbv;
2116 nbnxn_alloc_t *nb_alloc;
2117 nbnxn_free_t *nb_free;
2119 nbv = new nonbonded_verlet_t();
2121 nbv->emulateGpu = ((getenv("GMX_EMULATE_GPU") != nullptr) ? EmulateGpuNonbonded::Yes : EmulateGpuNonbonded::No);
2122 nbv->bUseGPU = deviceInfo != nullptr;
2124 GMX_RELEASE_ASSERT(!(nbv->emulateGpu == EmulateGpuNonbonded::Yes && nbv->bUseGPU), "When GPU emulation is active, there cannot be a GPU assignment");
2127 nbv->min_ci_balanced = 0;
2129 nbv->ngrp = (DOMAINDECOMP(cr) ? 2 : 1);
2130 for (int i = 0; i < nbv->ngrp; i++)
2132 nbv->grp[i].nbl_lists.nnbl = 0;
2133 nbv->grp[i].kernel_type = nbnxnkNotSet;
2135 if (i == 0) /* local */
2137 pick_nbnxn_kernel(mdlog, fr->use_simd_kernels, hardwareInfo,
2138 nbv->bUseGPU, nbv->emulateGpu, ir,
2139 &nbv->grp[i].kernel_type,
2140 &nbv->grp[i].ewald_excl,
2143 else /* non-local */
2145 /* Use the same kernel for local and non-local interactions */
2146 nbv->grp[i].kernel_type = nbv->grp[0].kernel_type;
2147 nbv->grp[i].ewald_excl = nbv->grp[0].ewald_excl;
2151 nbv->listParams = gmx::compat::make_unique<NbnxnListParameters>(ir->rlist);
2152 setupDynamicPairlistPruning(mdlog, ir, mtop, box, nbv->grp[0].kernel_type, fr->ic,
2153 nbv->listParams.get());
2155 nbv->nbs = gmx::compat::make_unique<nbnxn_search>(DOMAINDECOMP(cr) ? &cr->dd->nc : nullptr,
2156 DOMAINDECOMP(cr) ? domdec_zones(cr->dd) : nullptr,
2158 gmx_omp_nthreads_get(emntPairsearch));
2160 gpu_set_host_malloc_and_free(nbv->grp[0].kernel_type == nbnxnk8x8x8_GPU,
2161 &nb_alloc, &nb_free);
2163 for (int i = 0; i < nbv->ngrp; i++)
2165 nbnxn_init_pairlist_set(&nbv->grp[i].nbl_lists,
2166 nbnxn_kernel_pairlist_simple(nbv->grp[i].kernel_type),
2167 /* 8x8x8 "non-simple" lists are ATM always combined */
2168 !nbnxn_kernel_pairlist_simple(nbv->grp[i].kernel_type),
2172 int enbnxninitcombrule;
2173 if (fr->ic->vdwtype == evdwCUT &&
2174 (fr->ic->vdw_modifier == eintmodNONE ||
2175 fr->ic->vdw_modifier == eintmodPOTSHIFT) &&
2176 getenv("GMX_NO_LJ_COMB_RULE") == nullptr)
2178 /* Plain LJ cut-off: we can optimize with combination rules */
2179 enbnxninitcombrule = enbnxninitcombruleDETECT;
2181 else if (fr->ic->vdwtype == evdwPME)
2183 /* LJ-PME: we need to use a combination rule for the grid */
2184 if (fr->ljpme_combination_rule == eljpmeGEOM)
2186 enbnxninitcombrule = enbnxninitcombruleGEOM;
2190 enbnxninitcombrule = enbnxninitcombruleLB;
2195 /* We use a full combination matrix: no rule required */
2196 enbnxninitcombrule = enbnxninitcombruleNONE;
2199 nbv->nbat = new nbnxn_atomdata_t(nbv->bUseGPU ? gmx::PinningPolicy::PinnedIfSupported : gmx::PinningPolicy::CannotBePinned);
2200 int mimimumNumEnergyGroupNonbonded = ir->opts.ngener;
2201 if (ir->opts.ngener - ir->nwall == 1)
2203 /* We have only one non-wall energy group, we do not need energy group
2204 * support in the non-bondeds kernels, since all non-bonded energy
2205 * contributions go to the first element of the energy group matrix.
2207 mimimumNumEnergyGroupNonbonded = 1;
2209 bool bSimpleList = nbnxn_kernel_pairlist_simple(nbv->grp[0].kernel_type);
2210 nbnxn_atomdata_init(mdlog,
2212 nbv->grp[0].kernel_type,
2214 fr->ntype, fr->nbfp,
2215 mimimumNumEnergyGroupNonbonded,
2216 bSimpleList ? gmx_omp_nthreads_get(emntNonbonded) : 1);
2220 /* init the NxN GPU data; the last argument tells whether we'll have
2221 * both local and non-local NB calculation on GPU */
2222 nbnxn_gpu_init(&nbv->gpu_nbv,
2225 nbv->listParams.get(),
2230 if ((env = getenv("GMX_NB_MIN_CI")) != nullptr)
2234 nbv->min_ci_balanced = strtol(env, &end, 10);
2235 if (!end || (*end != 0) || nbv->min_ci_balanced < 0)
2237 gmx_fatal(FARGS, "Invalid value passed in GMX_NB_MIN_CI=%s, non-negative integer required", env);
2242 fprintf(debug, "Neighbor-list balancing parameter: %d (passed as env. var.)\n",
2243 nbv->min_ci_balanced);
2248 nbv->min_ci_balanced = nbnxn_gpu_min_ci_balanced(nbv->gpu_nbv);
2251 fprintf(debug, "Neighbor-list balancing parameter: %d (auto-adjusted to the number of GPU multi-processors)\n",
2252 nbv->min_ci_balanced);
2261 gmx_bool usingGpu(nonbonded_verlet_t *nbv)
2263 return nbv != nullptr && nbv->bUseGPU;
2266 void init_forcerec(FILE *fp,
2267 const gmx::MDLogger &mdlog,
2270 const t_inputrec *ir,
2271 const gmx_mtop_t *mtop,
2272 const t_commrec *cr,
2276 gmx::ArrayRef<const std::string> tabbfnm,
2277 const gmx_hw_info_t &hardwareInfo,
2278 const gmx_device_info_t *deviceInfo,
2279 const bool useGpuForBonded,
2280 gmx_bool bNoSolvOpt,
2283 int m, negp_pp, negptable, egi, egj;
2288 gmx_bool bGenericKernelOnly;
2289 gmx_bool needGroupSchemeTables, bSomeNormalNbListsAreInUse;
2290 gmx_bool bFEP_NonBonded;
2291 int *nm_ind, egp_flags;
2293 /* By default we turn SIMD kernels on, but it might be turned off further down... */
2294 fr->use_simd_kernels = TRUE;
2296 if (check_box(ir->ePBC, box))
2298 gmx_fatal(FARGS, "%s", check_box(ir->ePBC, box));
2301 /* Test particle insertion ? */
2304 /* Set to the size of the molecule to be inserted (the last one) */
2305 /* Because of old style topologies, we have to use the last cg
2306 * instead of the last molecule type.
2308 cgs = &mtop->moltype[mtop->molblock.back().type].cgs;
2309 fr->n_tpi = cgs->index[cgs->nr] - cgs->index[cgs->nr-1];
2310 gmx::RangePartitioning molecules = gmx_mtop_molecules(*mtop);
2311 if (fr->n_tpi != molecules.block(molecules.numBlocks() - 1).size())
2313 gmx_fatal(FARGS, "The molecule to insert can not consist of multiple charge groups.\nMake it a single charge group.");
2321 if (ir->coulombtype == eelRF_NEC_UNSUPPORTED)
2323 gmx_fatal(FARGS, "%s electrostatics is no longer supported",
2324 eel_names[ir->coulombtype]);
2329 gmx_fatal(FARGS, "AdResS simulations are no longer supported");
2331 if (ir->useTwinRange)
2333 gmx_fatal(FARGS, "Twin-range simulations are no longer supported");
2335 /* Copy the user determined parameters */
2336 fr->userint1 = ir->userint1;
2337 fr->userint2 = ir->userint2;
2338 fr->userint3 = ir->userint3;
2339 fr->userint4 = ir->userint4;
2340 fr->userreal1 = ir->userreal1;
2341 fr->userreal2 = ir->userreal2;
2342 fr->userreal3 = ir->userreal3;
2343 fr->userreal4 = ir->userreal4;
2346 fr->fc_stepsize = ir->fc_stepsize;
2349 fr->efep = ir->efep;
2350 fr->sc_alphavdw = ir->fepvals->sc_alpha;
2351 if (ir->fepvals->bScCoul)
2353 fr->sc_alphacoul = ir->fepvals->sc_alpha;
2354 fr->sc_sigma6_min = gmx::power6(ir->fepvals->sc_sigma_min);
2358 fr->sc_alphacoul = 0;
2359 fr->sc_sigma6_min = 0; /* only needed when bScCoul is on */
2361 fr->sc_power = ir->fepvals->sc_power;
2362 fr->sc_r_power = ir->fepvals->sc_r_power;
2363 fr->sc_sigma6_def = gmx::power6(ir->fepvals->sc_sigma);
2365 env = getenv("GMX_SCSIGMA_MIN");
2369 sscanf(env, "%20lf", &dbl);
2370 fr->sc_sigma6_min = gmx::power6(dbl);
2373 fprintf(fp, "Setting the minimum soft core sigma to %g nm\n", dbl);
2377 fr->bNonbonded = TRUE;
2378 if (getenv("GMX_NO_NONBONDED") != nullptr)
2380 /* turn off non-bonded calculations */
2381 fr->bNonbonded = FALSE;
2382 GMX_LOG(mdlog.warning).asParagraph().appendText(
2383 "Found environment variable GMX_NO_NONBONDED.\n"
2384 "Disabling nonbonded calculations.");
2387 bGenericKernelOnly = FALSE;
2389 /* We now check in the NS code whether a particular combination of interactions
2390 * can be used with water optimization, and disable it if that is not the case.
2393 if (getenv("GMX_NB_GENERIC") != nullptr)
2398 "Found environment variable GMX_NB_GENERIC.\n"
2399 "Disabling all interaction-specific nonbonded kernels, will only\n"
2400 "use the slow generic ones in src/gmxlib/nonbonded/nb_generic.c\n\n");
2402 bGenericKernelOnly = TRUE;
2405 if (bGenericKernelOnly)
2410 if ( (getenv("GMX_DISABLE_SIMD_KERNELS") != nullptr) || (getenv("GMX_NOOPTIMIZEDKERNELS") != nullptr) )
2412 fr->use_simd_kernels = FALSE;
2416 "\nFound environment variable GMX_DISABLE_SIMD_KERNELS.\n"
2417 "Disabling the usage of any SIMD-specific non-bonded & bonded kernel routines\n"
2418 "(e.g. SSE2/SSE4.1/AVX).\n\n");
2422 fr->bBHAM = (mtop->ffparams.functype[0] == F_BHAM);
2424 /* Neighbour searching stuff */
2425 fr->cutoff_scheme = ir->cutoff_scheme;
2426 fr->bGrid = (ir->ns_type == ensGRID);
2427 fr->ePBC = ir->ePBC;
2429 if (fr->cutoff_scheme == ecutsGROUP)
2431 const char *note = "NOTE: This file uses the deprecated 'group' cutoff_scheme. This will be\n"
2432 "removed in a future release when 'verlet' supports all interaction forms.\n";
2436 fprintf(stderr, "\n%s\n", note);
2440 fprintf(fp, "\n%s\n", note);
2444 /* Determine if we will do PBC for distances in bonded interactions */
2445 if (fr->ePBC == epbcNONE)
2447 fr->bMolPBC = FALSE;
2451 if (!DOMAINDECOMP(cr))
2455 bSHAKE = (ir->eConstrAlg == econtSHAKE &&
2456 (gmx_mtop_ftype_count(mtop, F_CONSTR) > 0 ||
2457 gmx_mtop_ftype_count(mtop, F_CONSTRNC) > 0));
2459 /* The group cut-off scheme and SHAKE assume charge groups
2460 * are whole, but not using molpbc is faster in most cases.
2461 * With intermolecular interactions we need PBC for calculating
2462 * distances between atoms in different molecules.
2464 if ((fr->cutoff_scheme == ecutsGROUP || bSHAKE) &&
2465 !mtop->bIntermolecularInteractions)
2467 fr->bMolPBC = ir->bPeriodicMols;
2469 if (bSHAKE && fr->bMolPBC)
2471 gmx_fatal(FARGS, "SHAKE is not supported with periodic molecules");
2476 /* Not making molecules whole is faster in most cases,
2477 * but With orientation restraints we need whole molecules.
2479 fr->bMolPBC = (fcd->orires.nr == 0);
2481 if (getenv("GMX_USE_GRAPH") != nullptr)
2483 fr->bMolPBC = FALSE;
2486 GMX_LOG(mdlog.warning).asParagraph().appendText("GMX_USE_GRAPH is set, using the graph for bonded interactions");
2489 if (mtop->bIntermolecularInteractions)
2491 GMX_LOG(mdlog.warning).asParagraph().appendText("WARNING: Molecules linked by intermolecular interactions have to reside in the same periodic image, otherwise artifacts will occur!");
2495 GMX_RELEASE_ASSERT(fr->bMolPBC || !mtop->bIntermolecularInteractions, "We need to use PBC within molecules with inter-molecular interactions");
2497 if (bSHAKE && fr->bMolPBC)
2499 gmx_fatal(FARGS, "SHAKE is not properly supported with intermolecular interactions. For short simulations where linked molecules remain in the same periodic image, the environment variable GMX_USE_GRAPH can be used to override this check.\n");
2505 fr->bMolPBC = dd_bonded_molpbc(cr->dd, fr->ePBC);
2509 fr->rc_scaling = ir->refcoord_scaling;
2510 copy_rvec(ir->posres_com, fr->posres_com);
2511 copy_rvec(ir->posres_comB, fr->posres_comB);
2512 fr->rlist = cutoff_inf(ir->rlist);
2513 fr->ljpme_combination_rule = ir->ljpme_combination_rule;
2515 /* This now calculates sum for q and c6*/
2516 bool systemHasNetCharge = set_chargesum(fp, fr, mtop);
2518 /* fr->ic is used both by verlet and group kernels (to some extent) now */
2519 init_interaction_const(fp, &fr->ic, ir, mtop, systemHasNetCharge);
2520 init_interaction_const_tables(fp, fr->ic, ir->rlist + ir->tabext);
2522 const interaction_const_t *ic = fr->ic;
2524 /* TODO: Replace this Ewald table or move it into interaction_const_t */
2525 if (ir->coulombtype == eelEWALD)
2527 init_ewald_tab(&(fr->ewald_table), ir, fp);
2530 /* Electrostatics: Translate from interaction-setting-in-mdp-file to kernel interaction format */
2531 switch (ic->eeltype)
2534 fr->nbkernel_elec_interaction = GMX_NBKERNEL_ELEC_COULOMB;
2539 fr->nbkernel_elec_interaction = GMX_NBKERNEL_ELEC_REACTIONFIELD;
2543 fr->nbkernel_elec_interaction = GMX_NBKERNEL_ELEC_REACTIONFIELD;
2544 GMX_RELEASE_ASSERT(ic->coulomb_modifier == eintmodEXACTCUTOFF, "With the group scheme RF-zero needs the exact cut-off modifier");
2553 case eelPMEUSERSWITCH:
2554 fr->nbkernel_elec_interaction = GMX_NBKERNEL_ELEC_CUBICSPLINETABLE;
2560 fr->nbkernel_elec_interaction = GMX_NBKERNEL_ELEC_EWALD;
2564 gmx_fatal(FARGS, "Unsupported electrostatic interaction: %s", eel_names[ic->eeltype]);
2566 fr->nbkernel_elec_modifier = ic->coulomb_modifier;
2568 /* Vdw: Translate from mdp settings to kernel format */
2569 switch (ic->vdwtype)
2574 fr->nbkernel_vdw_interaction = GMX_NBKERNEL_VDW_BUCKINGHAM;
2578 fr->nbkernel_vdw_interaction = GMX_NBKERNEL_VDW_LENNARDJONES;
2582 fr->nbkernel_vdw_interaction = GMX_NBKERNEL_VDW_LJEWALD;
2588 case evdwENCADSHIFT:
2589 fr->nbkernel_vdw_interaction = GMX_NBKERNEL_VDW_CUBICSPLINETABLE;
2593 gmx_fatal(FARGS, "Unsupported vdw interaction: %s", evdw_names[ic->vdwtype]);
2595 fr->nbkernel_vdw_modifier = ic->vdw_modifier;
2597 if (ir->cutoff_scheme == ecutsGROUP)
2599 fr->bvdwtab = ((ic->vdwtype != evdwCUT || !gmx_within_tol(ic->reppow, 12.0, 10*GMX_DOUBLE_EPS))
2600 && !EVDW_PME(ic->vdwtype));
2601 /* We have special kernels for standard Ewald and PME, but the pme-switch ones are tabulated above */
2602 fr->bcoultab = !(ic->eeltype == eelCUT ||
2603 ic->eeltype == eelEWALD ||
2604 ic->eeltype == eelPME ||
2605 ic->eeltype == eelP3M_AD ||
2606 ic->eeltype == eelRF ||
2607 ic->eeltype == eelRF_ZERO);
2609 /* If the user absolutely wants different switch/shift settings for coul/vdw, it is likely
2610 * going to be faster to tabulate the interaction than calling the generic kernel.
2611 * However, if generic kernels have been requested we keep things analytically.
2613 if (fr->nbkernel_elec_modifier == eintmodPOTSWITCH &&
2614 fr->nbkernel_vdw_modifier == eintmodPOTSWITCH &&
2615 !bGenericKernelOnly)
2617 if ((ic->rcoulomb_switch != ic->rvdw_switch) || (ic->rcoulomb != ic->rvdw))
2619 fr->bcoultab = TRUE;
2620 /* Once we tabulate electrostatics, we can use the switch function for LJ,
2621 * which would otherwise need two tables.
2625 else if ((fr->nbkernel_elec_modifier == eintmodPOTSHIFT && fr->nbkernel_vdw_modifier == eintmodPOTSHIFT) ||
2626 ((fr->nbkernel_elec_interaction == GMX_NBKERNEL_ELEC_REACTIONFIELD &&
2627 fr->nbkernel_elec_modifier == eintmodEXACTCUTOFF &&
2628 (fr->nbkernel_vdw_modifier == eintmodPOTSWITCH || fr->nbkernel_vdw_modifier == eintmodPOTSHIFT))))
2630 if ((ic->rcoulomb != ic->rvdw) && (!bGenericKernelOnly))
2632 fr->bcoultab = TRUE;
2636 if (fr->nbkernel_elec_modifier == eintmodFORCESWITCH)
2638 fr->bcoultab = TRUE;
2640 if (fr->nbkernel_vdw_modifier == eintmodFORCESWITCH)
2645 if (getenv("GMX_REQUIRE_TABLES"))
2648 fr->bcoultab = TRUE;
2653 fprintf(fp, "Table routines are used for coulomb: %s\n",
2654 gmx::boolToString(fr->bcoultab));
2655 fprintf(fp, "Table routines are used for vdw: %s\n",
2656 gmx::boolToString(fr->bvdwtab));
2661 fr->nbkernel_vdw_interaction = GMX_NBKERNEL_VDW_CUBICSPLINETABLE;
2662 fr->nbkernel_vdw_modifier = eintmodNONE;
2666 fr->nbkernel_elec_interaction = GMX_NBKERNEL_ELEC_CUBICSPLINETABLE;
2667 fr->nbkernel_elec_modifier = eintmodNONE;
2671 if (ir->cutoff_scheme == ecutsVERLET)
2673 if (!gmx_within_tol(ic->reppow, 12.0, 10*GMX_DOUBLE_EPS))
2675 gmx_fatal(FARGS, "Cut-off scheme %s only supports LJ repulsion power 12", ecutscheme_names[ir->cutoff_scheme]);
2677 /* Older tpr files can contain Coulomb user tables with the Verlet cutoff-scheme,
2678 * while mdrun does not (and never did) support this.
2680 if (EEL_USER(fr->ic->eeltype))
2682 gmx_fatal(FARGS, "Combination of %s and cutoff scheme %s is not supported",
2683 eel_names[ir->coulombtype], ecutscheme_names[ir->cutoff_scheme]);
2686 fr->bvdwtab = FALSE;
2687 fr->bcoultab = FALSE;
2690 /* 1-4 interaction electrostatics */
2691 fr->fudgeQQ = mtop->ffparams.fudgeQQ;
2693 /* Parameters for generalized RF */
2697 if (ic->eeltype == eelGRF)
2699 init_generalized_rf(fp, mtop, ir, fr);
2702 fr->haveDirectVirialContributions =
2703 (EEL_FULL(ic->eeltype) || EVDW_PME(ic->vdwtype) ||
2704 fr->forceProviders->hasForceProvider() ||
2705 gmx_mtop_ftype_count(mtop, F_POSRES) > 0 ||
2706 gmx_mtop_ftype_count(mtop, F_FBPOSRES) > 0 ||
2712 if (fr->haveDirectVirialContributions)
2714 fr->forceBufferForDirectVirialContributions = new std::vector<gmx::RVec>;
2717 if (fr->cutoff_scheme == ecutsGROUP &&
2718 ncg_mtop(mtop) > fr->cg_nalloc && !DOMAINDECOMP(cr))
2720 /* Count the total number of charge groups */
2721 fr->cg_nalloc = ncg_mtop(mtop);
2722 srenew(fr->cg_cm, fr->cg_nalloc);
2724 if (fr->shift_vec == nullptr)
2726 snew(fr->shift_vec, SHIFTS);
2729 if (fr->fshift == nullptr)
2731 snew(fr->fshift, SHIFTS);
2734 if (fr->nbfp == nullptr)
2736 fr->ntype = mtop->ffparams.atnr;
2737 fr->nbfp = mk_nbfp(&mtop->ffparams, fr->bBHAM);
2738 if (EVDW_PME(ic->vdwtype))
2740 fr->ljpme_c6grid = make_ljpme_c6grid(&mtop->ffparams, fr);
2744 /* Copy the energy group exclusions */
2745 fr->egp_flags = ir->opts.egp_flags;
2747 /* Van der Waals stuff */
2748 if ((ic->vdwtype != evdwCUT) && (ic->vdwtype != evdwUSER) && !fr->bBHAM)
2750 if (ic->rvdw_switch >= ic->rvdw)
2752 gmx_fatal(FARGS, "rvdw_switch (%f) must be < rvdw (%f)",
2753 ic->rvdw_switch, ic->rvdw);
2757 fprintf(fp, "Using %s Lennard-Jones, switch between %g and %g nm\n",
2758 (ic->eeltype == eelSWITCH) ? "switched" : "shifted",
2759 ic->rvdw_switch, ic->rvdw);
2763 if (fr->bBHAM && EVDW_PME(ic->vdwtype))
2765 gmx_fatal(FARGS, "LJ PME not supported with Buckingham");
2768 if (fr->bBHAM && (ic->vdwtype == evdwSHIFT || ic->vdwtype == evdwSWITCH))
2770 gmx_fatal(FARGS, "Switch/shift interaction not supported with Buckingham");
2773 if (fr->bBHAM && fr->cutoff_scheme == ecutsVERLET)
2775 gmx_fatal(FARGS, "Verlet cutoff-scheme is not supported with Buckingham");
2778 if (fp && fr->cutoff_scheme == ecutsGROUP)
2780 fprintf(fp, "Cut-off's: NS: %g Coulomb: %g %s: %g\n",
2781 fr->rlist, ic->rcoulomb, fr->bBHAM ? "BHAM" : "LJ", ic->rvdw);
2784 fr->eDispCorr = ir->eDispCorr;
2785 fr->numAtomsForDispersionCorrection = mtop->natoms;
2786 if (ir->eDispCorr != edispcNO)
2788 set_avcsixtwelve(fp, fr, mtop);
2791 if (ir->implicit_solvent)
2793 gmx_fatal(FARGS, "Implict solvation is no longer supported.");
2796 /* Construct tables for the group scheme. A little unnecessary to
2797 * make both vdw and coul tables sometimes, but what the
2798 * heck. Note that both cutoff schemes construct Ewald tables in
2799 * init_interaction_const_tables. */
2800 needGroupSchemeTables = (ir->cutoff_scheme == ecutsGROUP &&
2801 (fr->bcoultab || fr->bvdwtab));
2803 negp_pp = ir->opts.ngener - ir->nwall;
2805 if (!needGroupSchemeTables)
2807 bSomeNormalNbListsAreInUse = TRUE;
2812 bSomeNormalNbListsAreInUse = FALSE;
2813 for (egi = 0; egi < negp_pp; egi++)
2815 for (egj = egi; egj < negp_pp; egj++)
2817 egp_flags = ir->opts.egp_flags[GID(egi, egj, ir->opts.ngener)];
2818 if (!(egp_flags & EGP_EXCL))
2820 if (egp_flags & EGP_TABLE)
2826 bSomeNormalNbListsAreInUse = TRUE;
2831 if (bSomeNormalNbListsAreInUse)
2833 fr->nnblists = negptable + 1;
2837 fr->nnblists = negptable;
2839 if (fr->nnblists > 1)
2841 snew(fr->gid2nblists, ir->opts.ngener*ir->opts.ngener);
2845 snew(fr->nblists, fr->nnblists);
2847 /* This code automatically gives table length tabext without cut-off's,
2848 * in that case grompp should already have checked that we do not need
2849 * normal tables and we only generate tables for 1-4 interactions.
2851 rtab = ir->rlist + ir->tabext;
2853 if (needGroupSchemeTables)
2855 /* make tables for ordinary interactions */
2856 if (bSomeNormalNbListsAreInUse)
2858 make_nbf_tables(fp, ic, rtab, tabfn, nullptr, nullptr, &fr->nblists[0]);
2867 /* Read the special tables for certain energy group pairs */
2868 nm_ind = mtop->groups.grps[egcENER].nm_ind;
2869 for (egi = 0; egi < negp_pp; egi++)
2871 for (egj = egi; egj < negp_pp; egj++)
2873 egp_flags = ir->opts.egp_flags[GID(egi, egj, ir->opts.ngener)];
2874 if ((egp_flags & EGP_TABLE) && !(egp_flags & EGP_EXCL))
2876 if (fr->nnblists > 1)
2878 fr->gid2nblists[GID(egi, egj, ir->opts.ngener)] = m;
2880 /* Read the table file with the two energy groups names appended */
2881 make_nbf_tables(fp, ic, rtab, tabfn,
2882 *mtop->groups.grpname[nm_ind[egi]],
2883 *mtop->groups.grpname[nm_ind[egj]],
2887 else if (fr->nnblists > 1)
2889 fr->gid2nblists[GID(egi, egj, ir->opts.ngener)] = 0;
2896 /* Tables might not be used for the potential modifier
2897 * interactions per se, but we still need them to evaluate
2898 * switch/shift dispersion corrections in this case. */
2899 if (fr->eDispCorr != edispcNO)
2901 fr->dispersionCorrectionTable = makeDispersionCorrectionTable(fp, ic, rtab, tabfn);
2904 /* We want to use unmodified tables for 1-4 coulombic
2905 * interactions, so we must in general have an extra set of
2907 if (gmx_mtop_ftype_count(mtop, F_LJ14) > 0 ||
2908 gmx_mtop_ftype_count(mtop, F_LJC14_Q) > 0 ||
2909 gmx_mtop_ftype_count(mtop, F_LJC_PAIRS_NB) > 0)
2911 fr->pairsTable = make_tables(fp, ic, tabpfn, rtab,
2912 GMX_MAKETABLES_14ONLY);
2916 fr->nwall = ir->nwall;
2917 if (ir->nwall && ir->wall_type == ewtTABLE)
2919 make_wall_tables(fp, ir, tabfn, &mtop->groups, fr);
2922 if (fcd && !tabbfnm.empty())
2924 // Need to catch std::bad_alloc
2925 // TODO Don't need to catch this here, when merging with master branch
2928 fcd->bondtab = make_bonded_tables(fp,
2929 F_TABBONDS, F_TABBONDSNC,
2930 mtop, tabbfnm, "b");
2931 fcd->angletab = make_bonded_tables(fp,
2933 mtop, tabbfnm, "a");
2934 fcd->dihtab = make_bonded_tables(fp,
2936 mtop, tabbfnm, "d");
2938 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
2944 fprintf(debug, "No fcdata or table file name passed, can not read table, can not do bonded interactions\n");
2948 // QM/MM initialization if requested
2949 fr->bQMMM = ir->bQMMM;
2952 // Initialize QM/MM if supported
2955 GMX_LOG(mdlog.info).asParagraph().
2956 appendText("Large parts of the QM/MM support is deprecated, and may be removed in a future "
2957 "version. Please get in touch with the developers if you find the support useful, "
2958 "as help is needed if the functionality is to continue to be available.");
2959 fr->qr = mk_QMMMrec();
2960 init_QMMMrec(cr, mtop, ir, fr);
2964 gmx_incons("QM/MM was requested, but is only available when GROMACS "
2965 "is configured with QM/MM support");
2969 /* Set all the static charge group info */
2970 fr->cginfo_mb = init_cginfo_mb(fp, mtop, fr, bNoSolvOpt,
2972 &fr->bExcl_IntraCGAll_InterCGNone);
2973 if (DOMAINDECOMP(cr))
2975 fr->cginfo = nullptr;
2979 fr->cginfo = cginfo_expand(mtop->molblock.size(), fr->cginfo_mb);
2982 if (!DOMAINDECOMP(cr))
2984 forcerec_set_ranges(fr, ncg_mtop(mtop), ncg_mtop(mtop),
2985 mtop->natoms, mtop->natoms, mtop->natoms);
2988 fr->print_force = print_force;
2991 /* coarse load balancing vars */
2996 /* Initialize neighbor search */
2998 init_ns(fp, cr, fr->ns, fr, mtop);
3000 if (thisRankHasDuty(cr, DUTY_PP))
3002 gmx_nonbonded_setup(fr, bGenericKernelOnly);
3005 /* Initialize the thread working data for bonded interactions */
3006 init_bonded_threading(fp, mtop->groups.grps[egcENER].nr,
3007 &fr->bondedThreading);
3009 fr->nthread_ewc = gmx_omp_nthreads_get(emntBonded);
3010 snew(fr->ewc_t, fr->nthread_ewc);
3012 if (fr->cutoff_scheme == ecutsVERLET)
3014 // We checked the cut-offs in grompp, but double-check here.
3015 // We have PME+LJcutoff kernels for rcoulomb>rvdw.
3016 if (EEL_PME_EWALD(ir->coulombtype) && ir->vdwtype == eelCUT)
3018 GMX_RELEASE_ASSERT(ir->rcoulomb >= ir->rvdw, "With Verlet lists and PME we should have rcoulomb>=rvdw");
3022 GMX_RELEASE_ASSERT(ir->rcoulomb == ir->rvdw, "With Verlet lists and no PME rcoulomb and rvdw should be identical");
3025 init_nb_verlet(mdlog, &fr->nbv, bFEP_NonBonded, ir, fr,
3026 cr, hardwareInfo, deviceInfo,
3029 if (useGpuForBonded)
3031 auto stream = DOMAINDECOMP(cr) ?
3032 nbnxn_gpu_get_command_stream(fr->nbv->gpu_nbv, eintNonlocal) :
3033 nbnxn_gpu_get_command_stream(fr->nbv->gpu_nbv, eintLocal);
3034 // TODO the heap allocation is only needed while
3035 // t_forcerec lacks a constructor.
3036 fr->gpuBonded = new gmx::GpuBonded(mtop->ffparams,
3043 /* Here we switch from using mdlog, which prints the newline before
3044 * the paragraph, to our old fprintf logging, which prints the newline
3045 * after the paragraph, so we should add a newline here.
3050 if (ir->eDispCorr != edispcNO)
3052 calc_enervirdiff(fp, ir->eDispCorr, fr);
3056 /* Frees GPU memory and sets a tMPI node barrier.
3058 * Note that this function needs to be called even if GPUs are not used
3059 * in this run because the PME ranks have no knowledge of whether GPUs
3060 * are used or not, but all ranks need to enter the barrier below.
3061 * \todo Remove physical node barrier from this function after making sure
3062 * that it's not needed anymore (with a shared GPU run).
3064 void free_gpu_resources(t_forcerec *fr,
3065 const gmx::PhysicalNodeCommunicator &physicalNodeCommunicator)
3067 bool isPPrankUsingGPU = (fr != nullptr) && (fr->nbv != nullptr) && fr->nbv->bUseGPU;
3069 /* stop the GPU profiler (only CUDA) */
3072 if (isPPrankUsingGPU)
3074 /* free nbnxn data in GPU memory */
3075 nbnxn_gpu_free(fr->nbv->gpu_nbv);
3076 delete fr->gpuBonded;
3077 fr->gpuBonded = nullptr;
3080 /* With tMPI we need to wait for all ranks to finish deallocation before
3081 * destroying the CUDA context in free_gpu() as some tMPI ranks may be sharing
3084 * This is not a concern in OpenCL where we use one context per rank which
3085 * is freed in nbnxn_gpu_free().
3087 * Note: it is safe to not call the barrier on the ranks which do not use GPU,
3088 * but it is easier and more futureproof to call it on the whole node.
3092 physicalNodeCommunicator.barrier();
3096 void done_forcerec(t_forcerec *fr, int numMolBlocks, int numEnergyGroups)
3100 // PME-only ranks don't have a forcerec
3103 // cginfo is dynamically allocated if no domain decomposition
3104 if (fr->cginfo != nullptr)
3108 done_cginfo_mb(fr->cginfo_mb, numMolBlocks);
3110 done_interaction_const(fr->ic);
3111 sfree(fr->shift_vec);
3114 done_ns(fr->ns, numEnergyGroups);
3116 tear_down_bonded_threading(fr->bondedThreading);
3117 GMX_RELEASE_ASSERT(fr->gpuBonded == nullptr, "Should have been deleted earlier, when used");
3118 fr->bondedThreading = nullptr;