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 by the GROMACS development team.
7 * Copyright (c) 2018,2019,2020, by the GROMACS development team, led by
8 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
9 * and including many others, as listed in the AUTHORS file in the
10 * top-level source directory and at http://www.gromacs.org.
12 * GROMACS is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public License
14 * as published by the Free Software Foundation; either version 2.1
15 * of the License, or (at your option) any later version.
17 * GROMACS is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with GROMACS; if not, see
24 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
25 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
27 * If you want to redistribute modifications to GROMACS, please
28 * consider that scientific software is very special. Version
29 * control is crucial - bugs must be traceable. We will be happy to
30 * consider code for inclusion in the official distribution, but
31 * derived work must not be called official GROMACS. Details are found
32 * in the README & COPYING files - if they are missing, get the
33 * official version at http://www.gromacs.org.
35 * To help us fund GROMACS development, we humbly ask that you cite
36 * the research papers on the package. Check out http://www.gromacs.org.
40 #include "nb_free_energy.h"
48 #include "gromacs/gmxlib/nrnb.h"
49 #include "gromacs/gmxlib/nonbonded/nb_kernel.h"
50 #include "gromacs/gmxlib/nonbonded/nonbonded.h"
51 #include "gromacs/math/functions.h"
52 #include "gromacs/math/vec.h"
53 #include "gromacs/mdtypes/forceoutput.h"
54 #include "gromacs/mdtypes/forcerec.h"
55 #include "gromacs/mdtypes/interaction_const.h"
56 #include "gromacs/mdtypes/md_enums.h"
57 #include "gromacs/mdtypes/mdatom.h"
58 #include "gromacs/simd/simd.h"
59 #include "gromacs/utility/fatalerror.h"
62 //! Scalar (non-SIMD) data types.
63 struct ScalarDataTypes
65 using RealType = real; //!< The data type to use as real.
66 using IntType = int; //!< The data type to use as int.
67 static constexpr int simdRealWidth = 1; //!< The width of the RealType.
68 static constexpr int simdIntWidth = 1; //!< The width of the IntType.
71 #if GMX_SIMD_HAVE_REAL && GMX_SIMD_HAVE_INT32_ARITHMETICS
75 using RealType = gmx::SimdReal; //!< The data type to use as real.
76 using IntType = gmx::SimdInt32; //!< The data type to use as int.
77 static constexpr int simdRealWidth = GMX_SIMD_REAL_WIDTH; //!< The width of the RealType.
78 static constexpr int simdIntWidth = GMX_SIMD_FINT32_WIDTH; //!< The width of the IntType.
82 //! Computes r^(1/p) and 1/r^(1/p) for the standard p=6
83 template<class RealType>
84 static inline void pthRoot(const RealType r, RealType* pthRoot, RealType* invPthRoot)
86 *invPthRoot = gmx::invsqrt(std::cbrt(r));
87 *pthRoot = 1 / (*invPthRoot);
90 template<class RealType>
91 static inline RealType calculateRinv6(const RealType rinvV)
93 RealType rinv6 = rinvV * rinvV;
94 return (rinv6 * rinv6 * rinv6);
97 template<class RealType>
98 static inline RealType calculateVdw6(const RealType c6, const RealType rinv6)
103 template<class RealType>
104 static inline RealType calculateVdw12(const RealType c12, const RealType rinv6)
106 return (c12 * rinv6 * rinv6);
109 /* reaction-field electrostatics */
110 template<class RealType>
111 static inline RealType reactionFieldScalarForce(const RealType qq,
117 return (qq * (rinv - two * krf * r * r));
119 template<class RealType>
120 static inline RealType reactionFieldPotential(const RealType qq,
124 const real potentialShift)
126 return (qq * (rinv + krf * r * r - potentialShift));
129 /* Ewald electrostatics */
130 template<class RealType>
131 static inline RealType ewaldScalarForce(const RealType coulomb, const RealType rinv)
133 return (coulomb * rinv);
135 template<class RealType>
136 static inline RealType ewaldPotential(const RealType coulomb, const RealType rinv, const real potentialShift)
138 return (coulomb * (rinv - potentialShift));
142 template<class RealType>
143 static inline RealType lennardJonesScalarForce(const RealType v6, const RealType v12)
147 template<class RealType>
148 static inline RealType lennardJonesPotential(const RealType v6,
152 const real repulsionShift,
153 const real dispersionShift,
155 const real onetwelfth)
157 return ((v12 + c12 * repulsionShift) * onetwelfth - (v6 + c6 * dispersionShift) * onesixth);
161 static inline real ewaldLennardJonesGridSubtract(const real c6grid, const real potentialShift, const real onesixth)
163 return (c6grid * potentialShift * onesixth);
166 /* LJ Potential switch */
167 template<class RealType>
168 static inline RealType potSwitchScalarForceMod(const RealType fScalarInp,
169 const RealType potential,
178 real fScalar = fScalarInp * sw - r * potential * dsw;
183 template<class RealType>
184 static inline RealType potSwitchPotentialMod(const RealType potentialInp,
192 real potential = potentialInp * sw;
199 //! Templated free-energy non-bonded kernel
200 template<typename DataTypes, bool useSoftCore, bool scLambdasOrAlphasDiffer, bool vdwInteractionTypeIsEwald, bool elecInteractionTypeIsEwald, bool vdwModifierIsPotSwitch>
201 static void nb_free_energy_kernel(const t_nblist* gmx_restrict nlist,
202 rvec* gmx_restrict xx,
203 gmx::ForceWithShiftForces* forceWithShiftForces,
204 const t_forcerec* gmx_restrict fr,
205 const t_mdatoms* gmx_restrict mdatoms,
206 nb_kernel_data_t* gmx_restrict kernel_data,
207 t_nrnb* gmx_restrict nrnb)
213 using RealType = typename DataTypes::RealType;
214 using IntType = typename DataTypes::IntType;
216 /* FIXME: How should these be handled with SIMD? */
217 constexpr real onetwelfth = 1.0 / 12.0;
218 constexpr real onesixth = 1.0 / 6.0;
219 constexpr real zero = 0.0;
220 constexpr real half = 0.5;
221 constexpr real one = 1.0;
222 constexpr real two = 2.0;
223 constexpr real six = 6.0;
225 /* Extract pointer to non-bonded interaction constants */
226 const interaction_const_t* ic = fr->ic;
228 // Extract pair list data
229 const int nri = nlist->nri;
230 const int* iinr = nlist->iinr;
231 const int* jindex = nlist->jindex;
232 const int* jjnr = nlist->jjnr;
233 const int* shift = nlist->shift;
234 const int* gid = nlist->gid;
236 const real* shiftvec = fr->shift_vec[0];
237 const real* chargeA = mdatoms->chargeA;
238 const real* chargeB = mdatoms->chargeB;
239 real* Vc = kernel_data->energygrp_elec;
240 const int* typeA = mdatoms->typeA;
241 const int* typeB = mdatoms->typeB;
242 const int ntype = fr->ntype;
243 const real* nbfp = fr->nbfp.data();
244 const real* nbfp_grid = fr->ljpme_c6grid;
245 real* Vv = kernel_data->energygrp_vdw;
246 const real lambda_coul = kernel_data->lambda[efptCOUL];
247 const real lambda_vdw = kernel_data->lambda[efptVDW];
248 real* dvdl = kernel_data->dvdl;
249 const auto& scParams = *ic->softCoreParameters;
250 const real alpha_coul = scParams.alphaCoulomb;
251 const real alpha_vdw = scParams.alphaVdw;
252 const real lam_power = scParams.lambdaPower;
253 const real sigma6_def = scParams.sigma6WithInvalidSigma;
254 const real sigma6_min = scParams.sigma6Minimum;
255 const bool doForces = ((kernel_data->flags & GMX_NONBONDED_DO_FORCE) != 0);
256 const bool doShiftForces = ((kernel_data->flags & GMX_NONBONDED_DO_SHIFTFORCE) != 0);
257 const bool doPotential = ((kernel_data->flags & GMX_NONBONDED_DO_POTENTIAL) != 0);
259 // Extract data from interaction_const_t
260 const real facel = ic->epsfac;
261 const real rcoulomb = ic->rcoulomb;
262 const real krf = ic->k_rf;
263 const real crf = ic->c_rf;
264 const real sh_lj_ewald = ic->sh_lj_ewald;
265 const real rvdw = ic->rvdw;
266 const real dispersionShift = ic->dispersion_shift.cpot;
267 const real repulsionShift = ic->repulsion_shift.cpot;
269 // Note that the nbnxm kernels do not support Coulomb potential switching at all
270 GMX_ASSERT(ic->coulomb_modifier != eintmodPOTSWITCH,
271 "Potential switching is not supported for Coulomb with FEP");
273 real vdw_swV3, vdw_swV4, vdw_swV5, vdw_swF2, vdw_swF3, vdw_swF4;
274 if (vdwModifierIsPotSwitch)
276 const real d = ic->rvdw - ic->rvdw_switch;
277 vdw_swV3 = -10.0 / (d * d * d);
278 vdw_swV4 = 15.0 / (d * d * d * d);
279 vdw_swV5 = -6.0 / (d * d * d * d * d);
280 vdw_swF2 = -30.0 / (d * d * d);
281 vdw_swF3 = 60.0 / (d * d * d * d);
282 vdw_swF4 = -30.0 / (d * d * d * d * d);
286 /* Avoid warnings from stupid compilers (looking at you, Clang!) */
287 vdw_swV3 = vdw_swV4 = vdw_swV5 = vdw_swF2 = vdw_swF3 = vdw_swF4 = 0.0;
291 if (ic->eeltype == eelCUT || EEL_RF(ic->eeltype))
293 icoul = GMX_NBKERNEL_ELEC_REACTIONFIELD;
297 icoul = GMX_NBKERNEL_ELEC_NONE;
300 real rcutoff_max2 = std::max(ic->rcoulomb, ic->rvdw);
301 rcutoff_max2 = rcutoff_max2 * rcutoff_max2;
303 const real* tab_ewald_F_lj = nullptr;
304 const real* tab_ewald_V_lj = nullptr;
305 const real* ewtab = nullptr;
306 real coulombTableScale = 0;
307 real coulombTableScaleInvHalf = 0;
308 real vdwTableScale = 0;
309 real vdwTableScaleInvHalf = 0;
311 if (elecInteractionTypeIsEwald || vdwInteractionTypeIsEwald)
313 sh_ewald = ic->sh_ewald;
315 if (elecInteractionTypeIsEwald)
317 const auto& coulombTables = *ic->coulombEwaldTables;
318 ewtab = coulombTables.tableFDV0.data();
319 coulombTableScale = coulombTables.scale;
320 coulombTableScaleInvHalf = half / coulombTableScale;
322 if (vdwInteractionTypeIsEwald)
324 const auto& vdwTables = *ic->vdwEwaldTables;
325 tab_ewald_F_lj = vdwTables.tableF.data();
326 tab_ewald_V_lj = vdwTables.tableV.data();
327 vdwTableScale = vdwTables.scale;
328 vdwTableScaleInvHalf = half / vdwTableScale;
331 /* For Ewald/PME interactions we cannot easily apply the soft-core component to
332 * reciprocal space. When we use non-switched Ewald interactions, we
333 * assume the soft-coring does not significantly affect the grid contribution
334 * and apply the soft-core only to the full 1/r (- shift) pair contribution.
336 * However, we cannot use this approach for switch-modified since we would then
337 * effectively end up evaluating a significantly different interaction here compared to the
338 * normal (non-free-energy) kernels, either by applying a cutoff at a different
339 * position than what the user requested, or by switching different
340 * things (1/r rather than short-range Ewald). For these settings, we just
341 * use the traditional short-range Ewald interaction in that case.
343 GMX_RELEASE_ASSERT(!(vdwInteractionTypeIsEwald && vdwModifierIsPotSwitch),
344 "Can not apply soft-core to switched Ewald potentials");
349 /* Lambda factor for state A, 1-lambda*/
350 real LFC[NSTATES], LFV[NSTATES];
351 LFC[STATE_A] = one - lambda_coul;
352 LFV[STATE_A] = one - lambda_vdw;
354 /* Lambda factor for state B, lambda*/
355 LFC[STATE_B] = lambda_coul;
356 LFV[STATE_B] = lambda_vdw;
358 /*derivative of the lambda factor for state A and B */
363 real lfac_coul[NSTATES], dlfac_coul[NSTATES], lfac_vdw[NSTATES], dlfac_vdw[NSTATES];
364 constexpr real sc_r_power = 6.0_real;
365 for (int i = 0; i < NSTATES; i++)
367 lfac_coul[i] = (lam_power == 2 ? (1 - LFC[i]) * (1 - LFC[i]) : (1 - LFC[i]));
368 dlfac_coul[i] = DLF[i] * lam_power / sc_r_power * (lam_power == 2 ? (1 - LFC[i]) : 1);
369 lfac_vdw[i] = (lam_power == 2 ? (1 - LFV[i]) * (1 - LFV[i]) : (1 - LFV[i]));
370 dlfac_vdw[i] = DLF[i] * lam_power / sc_r_power * (lam_power == 2 ? (1 - LFV[i]) : 1);
373 // TODO: We should get rid of using pointers to real
374 const real* x = xx[0];
375 real* gmx_restrict f = &(forceWithShiftForces->force()[0][0]);
376 real* gmx_restrict fshift = &(forceWithShiftForces->shiftForces()[0][0]);
378 const real rlistSquared = gmx::square(fr->rlist);
380 int numExcludedPairsBeyondRlist = 0;
382 for (int n = 0; n < nri; n++)
384 int npair_within_cutoff = 0;
386 const int is3 = 3 * shift[n];
387 const real shX = shiftvec[is3];
388 const real shY = shiftvec[is3 + 1];
389 const real shZ = shiftvec[is3 + 2];
390 const int nj0 = jindex[n];
391 const int nj1 = jindex[n + 1];
392 const int ii = iinr[n];
393 const int ii3 = 3 * ii;
394 const real ix = shX + x[ii3 + 0];
395 const real iy = shY + x[ii3 + 1];
396 const real iz = shZ + x[ii3 + 2];
397 const real iqA = facel * chargeA[ii];
398 const real iqB = facel * chargeB[ii];
399 const int ntiA = 2 * ntype * typeA[ii];
400 const int ntiB = 2 * ntype * typeB[ii];
407 for (int k = nj0; k < nj1; k++)
410 const int jnr = jjnr[k];
411 const int j3 = 3 * jnr;
412 RealType c6[NSTATES], c12[NSTATES], qq[NSTATES], Vcoul[NSTATES], Vvdw[NSTATES];
413 RealType r, rinv, rp, rpm2;
414 RealType alpha_vdw_eff, alpha_coul_eff, sigma6[NSTATES];
415 const RealType dx = ix - x[j3];
416 const RealType dy = iy - x[j3 + 1];
417 const RealType dz = iz - x[j3 + 2];
418 const RealType rsq = dx * dx + dy * dy + dz * dz;
419 RealType FscalC[NSTATES], FscalV[NSTATES];
420 /* Check if this pair on the exlusions list.*/
421 const bool bPairIncluded = nlist->excl_fep == nullptr || nlist->excl_fep[k];
423 if (rsq >= rcutoff_max2 && bPairIncluded)
425 /* We save significant time by skipping all code below.
426 * Note that with soft-core interactions, the actual cut-off
427 * check might be different. But since the soft-core distance
428 * is always larger than r, checking on r here is safe.
429 * Exclusions outside the cutoff can not be skipped as
430 * when using Ewald: the reciprocal-space
431 * Ewald component still needs to be subtracted.
436 npair_within_cutoff++;
438 if (rsq > rlistSquared)
440 numExcludedPairsBeyondRlist++;
445 /* Note that unlike in the nbnxn kernels, we do not need
446 * to clamp the value of rsq before taking the invsqrt
447 * to avoid NaN in the LJ calculation, since here we do
448 * not calculate LJ interactions when C6 and C12 are zero.
451 rinv = gmx::invsqrt(rsq);
456 /* The force at r=0 is zero, because of symmetry.
457 * But note that the potential is in general non-zero,
458 * since the soft-cored r will be non-zero.
466 rpm2 = rsq * rsq; /* r4 */
467 rp = rpm2 * rsq; /* r6 */
471 /* The soft-core power p will not affect the results
472 * with not using soft-core, so we use power of 0 which gives
473 * the simplest math and cheapest code.
481 qq[STATE_A] = iqA * chargeA[jnr];
482 qq[STATE_B] = iqB * chargeB[jnr];
484 tj[STATE_A] = ntiA + 2 * typeA[jnr];
485 tj[STATE_B] = ntiB + 2 * typeB[jnr];
489 c6[STATE_A] = nbfp[tj[STATE_A]];
490 c6[STATE_B] = nbfp[tj[STATE_B]];
492 for (int i = 0; i < NSTATES; i++)
494 c12[i] = nbfp[tj[i] + 1];
497 if ((c6[i] > 0) && (c12[i] > 0))
499 /* c12 is stored scaled with 12.0 and c6 is scaled with 6.0 - correct for this */
500 sigma6[i] = half * c12[i] / c6[i];
501 if (sigma6[i] < sigma6_min) /* for disappearing coul and vdw with soft core at the same time */
503 sigma6[i] = sigma6_min;
508 sigma6[i] = sigma6_def;
515 /* only use softcore if one of the states has a zero endstate - softcore is for avoiding infinities!*/
516 if ((c12[STATE_A] > 0) && (c12[STATE_B] > 0))
523 alpha_vdw_eff = alpha_vdw;
524 alpha_coul_eff = alpha_coul;
528 for (int i = 0; i < NSTATES; i++)
535 RealType rinvC, rinvV, rC, rV, rpinvC, rpinvV;
537 /* Only spend time on A or B state if it is non-zero */
538 if ((qq[i] != 0) || (c6[i] != 0) || (c12[i] != 0))
540 /* this section has to be inside the loop because of the dependence on sigma6 */
543 rpinvC = one / (alpha_coul_eff * lfac_coul[i] * sigma6[i] + rp);
544 pthRoot(rpinvC, &rinvC, &rC);
545 if (scLambdasOrAlphasDiffer)
547 rpinvV = one / (alpha_vdw_eff * lfac_vdw[i] * sigma6[i] + rp);
548 pthRoot(rpinvV, &rinvV, &rV);
552 /* We can avoid one expensive pow and one / operation */
569 /* Only process the coulomb interactions if we have charges,
570 * and if we either include all entries in the list (no cutoff
571 * used in the kernel), or if we are within the cutoff.
573 bool computeElecInteraction = (elecInteractionTypeIsEwald && r < rcoulomb)
574 || (!elecInteractionTypeIsEwald && rC < rcoulomb);
576 if ((qq[i] != 0) && computeElecInteraction)
578 if (elecInteractionTypeIsEwald)
580 Vcoul[i] = ewaldPotential(qq[i], rinvC, sh_ewald);
581 FscalC[i] = ewaldScalarForce(qq[i], rinvC);
585 Vcoul[i] = reactionFieldPotential(qq[i], rinvC, rC, krf, crf);
586 FscalC[i] = reactionFieldScalarForce(qq[i], rinvC, rC, krf, two);
590 /* Only process the VDW interactions if we have
591 * some non-zero parameters, and if we either
592 * include all entries in the list (no cutoff used
593 * in the kernel), or if we are within the cutoff.
595 bool computeVdwInteraction = (vdwInteractionTypeIsEwald && r < rvdw)
596 || (!vdwInteractionTypeIsEwald && rV < rvdw);
597 if ((c6[i] != 0 || c12[i] != 0) && computeVdwInteraction)
606 rinv6 = calculateRinv6(rinvV);
608 RealType Vvdw6 = calculateVdw6(c6[i], rinv6);
609 RealType Vvdw12 = calculateVdw12(c12[i], rinv6);
611 Vvdw[i] = lennardJonesPotential(Vvdw6, Vvdw12, c6[i], c12[i], repulsionShift,
612 dispersionShift, onesixth, onetwelfth);
613 FscalV[i] = lennardJonesScalarForce(Vvdw6, Vvdw12);
615 if (vdwInteractionTypeIsEwald)
617 /* Subtract the grid potential at the cut-off */
618 Vvdw[i] += ewaldLennardJonesGridSubtract(nbfp_grid[tj[i]],
619 sh_lj_ewald, onesixth);
622 if (vdwModifierIsPotSwitch)
624 RealType d = rV - ic->rvdw_switch;
625 d = (d > zero) ? d : zero;
626 const RealType d2 = d * d;
628 one + d2 * d * (vdw_swV3 + d * (vdw_swV4 + d * vdw_swV5));
629 const RealType dsw = d2 * (vdw_swF2 + d * (vdw_swF3 + d * vdw_swF4));
631 FscalV[i] = potSwitchScalarForceMod(FscalV[i], Vvdw[i], sw, rV,
633 Vvdw[i] = potSwitchPotentialMod(Vvdw[i], sw, rV, rvdw, zero);
637 /* FscalC (and FscalV) now contain: dV/drC * rC
638 * Now we multiply by rC^-p, so it will be: dV/drC * rC^1-p
639 * Further down we first multiply by r^p-2 and then by
640 * the vector r, which in total gives: dV/drC * (r/rC)^1-p
645 } // end for (int i = 0; i < NSTATES; i++)
647 /* Assemble A and B states */
648 for (int i = 0; i < NSTATES; i++)
650 vctot += LFC[i] * Vcoul[i];
651 vvtot += LFV[i] * Vvdw[i];
653 Fscal += LFC[i] * FscalC[i] * rpm2;
654 Fscal += LFV[i] * FscalV[i] * rpm2;
658 dvdl_coul += Vcoul[i] * DLF[i]
659 + LFC[i] * alpha_coul_eff * dlfac_coul[i] * FscalC[i] * sigma6[i];
660 dvdl_vdw += Vvdw[i] * DLF[i]
661 + LFV[i] * alpha_vdw_eff * dlfac_vdw[i] * FscalV[i] * sigma6[i];
665 dvdl_coul += Vcoul[i] * DLF[i];
666 dvdl_vdw += Vvdw[i] * DLF[i];
669 } // end if (bPairIncluded)
670 else if (icoul == GMX_NBKERNEL_ELEC_REACTIONFIELD)
672 /* For excluded pairs, which are only in this pair list when
673 * using the Verlet scheme, we don't use soft-core.
674 * As there is no singularity, there is no need for soft-core.
676 const real FF = -two * krf;
677 RealType VV = krf * rsq - crf;
684 for (int i = 0; i < NSTATES; i++)
686 vctot += LFC[i] * qq[i] * VV;
687 Fscal += LFC[i] * qq[i] * FF;
688 dvdl_coul += DLF[i] * qq[i] * VV;
692 if (elecInteractionTypeIsEwald && (r < rcoulomb || !bPairIncluded))
694 /* See comment in the preamble. When using Ewald interactions
695 * (unless we use a switch modifier) we subtract the reciprocal-space
696 * Ewald component here which made it possible to apply the free
697 * energy interaction to 1/r (vanilla coulomb short-range part)
698 * above. This gets us closer to the ideal case of applying
699 * the softcore to the entire electrostatic interaction,
700 * including the reciprocal-space component.
704 const RealType ewrt = r * coulombTableScale;
705 IntType ewitab = static_cast<IntType>(ewrt);
706 const RealType eweps = ewrt - ewitab;
708 f_lr = ewtab[ewitab] + eweps * ewtab[ewitab + 1];
709 v_lr = (ewtab[ewitab + 2] - coulombTableScaleInvHalf * eweps * (ewtab[ewitab] + f_lr));
712 /* Note that any possible Ewald shift has already been applied in
713 * the normal interaction part above.
718 /* If we get here, the i particle (ii) has itself (jnr)
719 * in its neighborlist. This can only happen with the Verlet
720 * scheme, and corresponds to a self-interaction that will
721 * occur twice. Scale it down by 50% to only include it once.
726 for (int i = 0; i < NSTATES; i++)
728 vctot -= LFC[i] * qq[i] * v_lr;
729 Fscal -= LFC[i] * qq[i] * f_lr;
730 dvdl_coul -= (DLF[i] * qq[i]) * v_lr;
734 if (vdwInteractionTypeIsEwald && (r < rvdw || !bPairIncluded))
736 /* See comment in the preamble. When using LJ-Ewald interactions
737 * (unless we use a switch modifier) we subtract the reciprocal-space
738 * Ewald component here which made it possible to apply the free
739 * energy interaction to r^-6 (vanilla LJ6 short-range part)
740 * above. This gets us closer to the ideal case of applying
741 * the softcore to the entire VdW interaction,
742 * including the reciprocal-space component.
744 /* We could also use the analytical form here
745 * iso a table, but that can cause issues for
746 * r close to 0 for non-interacting pairs.
749 const RealType rs = rsq * rinv * vdwTableScale;
750 const IntType ri = static_cast<IntType>(rs);
751 const RealType frac = rs - ri;
752 const RealType f_lr = (1 - frac) * tab_ewald_F_lj[ri] + frac * tab_ewald_F_lj[ri + 1];
753 /* TODO: Currently the Ewald LJ table does not contain
754 * the factor 1/6, we should add this.
756 const RealType FF = f_lr * rinv / six;
758 (tab_ewald_V_lj[ri] - vdwTableScaleInvHalf * frac * (tab_ewald_F_lj[ri] + f_lr))
763 /* If we get here, the i particle (ii) has itself (jnr)
764 * in its neighborlist. This can only happen with the Verlet
765 * scheme, and corresponds to a self-interaction that will
766 * occur twice. Scale it down by 50% to only include it once.
771 for (int i = 0; i < NSTATES; i++)
773 const real c6grid = nbfp_grid[tj[i]];
774 vvtot += LFV[i] * c6grid * VV;
775 Fscal += LFV[i] * c6grid * FF;
776 dvdl_vdw += (DLF[i] * c6grid) * VV;
782 const real tx = Fscal * dx;
783 const real ty = Fscal * dy;
784 const real tz = Fscal * dz;
788 /* OpenMP atomics are expensive, but this kernels is also
789 * expensive, so we can take this hit, instead of using
790 * thread-local output buffers and extra reduction.
792 * All the OpenMP regions in this file are trivial and should
793 * not throw, so no need for try/catch.
802 } // end for (int k = nj0; k < nj1; k++)
804 /* The atomics below are expensive with many OpenMP threads.
805 * Here unperturbed i-particles will usually only have a few
806 * (perturbed) j-particles in the list. Thus with a buffered list
807 * we can skip a significant number of i-reductions with a check.
809 if (npair_within_cutoff > 0)
825 fshift[is3 + 1] += fiy;
827 fshift[is3 + 2] += fiz;
838 } // end for (int n = 0; n < nri; n++)
841 dvdl[efptCOUL] += dvdl_coul;
843 dvdl[efptVDW] += dvdl_vdw;
845 /* Estimate flops, average for free energy stuff:
846 * 12 flops per outer iteration
847 * 150 flops per inner iteration
850 inc_nrnb(nrnb, eNR_NBKERNEL_FREE_ENERGY, nlist->nri * 12 + nlist->jindex[nri] * 150);
852 if (numExcludedPairsBeyondRlist > 0)
855 "There are %d perturbed non-bonded pair interactions beyond the pair-list cutoff "
856 "of %g nm, which is not supported. This can happen because the system is "
857 "unstable or because intra-molecular interactions at long distances are "
859 "latter is the case, you can try to increase nstlist or rlist to avoid this."
860 "The error is likely triggered by the use of couple-intramol=no "
861 "and the maximal distance in the decoupled molecule exceeding rlist.",
862 numExcludedPairsBeyondRlist, fr->rlist);
866 typedef void (*KernelFunction)(const t_nblist* gmx_restrict nlist,
867 rvec* gmx_restrict xx,
868 gmx::ForceWithShiftForces* forceWithShiftForces,
869 const t_forcerec* gmx_restrict fr,
870 const t_mdatoms* gmx_restrict mdatoms,
871 nb_kernel_data_t* gmx_restrict kernel_data,
872 t_nrnb* gmx_restrict nrnb);
874 template<bool useSoftCore, bool scLambdasOrAlphasDiffer, bool vdwInteractionTypeIsEwald, bool elecInteractionTypeIsEwald, bool vdwModifierIsPotSwitch>
875 static KernelFunction dispatchKernelOnUseSimd(const bool useSimd)
879 #if GMX_SIMD_HAVE_REAL && GMX_SIMD_HAVE_INT32_ARITHMETICS && GMX_USE_SIMD_KERNELS
880 /* FIXME: Here SimdDataTypes should be used to enable SIMD. So far, the code in nb_free_energy_kernel is not adapted to SIMD */
881 return (nb_free_energy_kernel<ScalarDataTypes, useSoftCore, scLambdasOrAlphasDiffer, vdwInteractionTypeIsEwald,
882 elecInteractionTypeIsEwald, vdwModifierIsPotSwitch>);
884 return (nb_free_energy_kernel<ScalarDataTypes, useSoftCore, scLambdasOrAlphasDiffer, vdwInteractionTypeIsEwald,
885 elecInteractionTypeIsEwald, vdwModifierIsPotSwitch>);
890 return (nb_free_energy_kernel<ScalarDataTypes, useSoftCore, scLambdasOrAlphasDiffer, vdwInteractionTypeIsEwald,
891 elecInteractionTypeIsEwald, vdwModifierIsPotSwitch>);
895 template<bool useSoftCore, bool scLambdasOrAlphasDiffer, bool vdwInteractionTypeIsEwald, bool elecInteractionTypeIsEwald>
896 static KernelFunction dispatchKernelOnVdwModifier(const bool vdwModifierIsPotSwitch, const bool useSimd)
898 if (vdwModifierIsPotSwitch)
900 return (dispatchKernelOnUseSimd<useSoftCore, scLambdasOrAlphasDiffer, vdwInteractionTypeIsEwald,
901 elecInteractionTypeIsEwald, true>(useSimd));
905 return (dispatchKernelOnUseSimd<useSoftCore, scLambdasOrAlphasDiffer, vdwInteractionTypeIsEwald,
906 elecInteractionTypeIsEwald, false>(useSimd));
910 template<bool useSoftCore, bool scLambdasOrAlphasDiffer, bool vdwInteractionTypeIsEwald>
911 static KernelFunction dispatchKernelOnElecInteractionType(const bool elecInteractionTypeIsEwald,
912 const bool vdwModifierIsPotSwitch,
915 if (elecInteractionTypeIsEwald)
917 return (dispatchKernelOnVdwModifier<useSoftCore, scLambdasOrAlphasDiffer, vdwInteractionTypeIsEwald, true>(
918 vdwModifierIsPotSwitch, useSimd));
922 return (dispatchKernelOnVdwModifier<useSoftCore, scLambdasOrAlphasDiffer, vdwInteractionTypeIsEwald, false>(
923 vdwModifierIsPotSwitch, useSimd));
927 template<bool useSoftCore, bool scLambdasOrAlphasDiffer>
928 static KernelFunction dispatchKernelOnVdwInteractionType(const bool vdwInteractionTypeIsEwald,
929 const bool elecInteractionTypeIsEwald,
930 const bool vdwModifierIsPotSwitch,
933 if (vdwInteractionTypeIsEwald)
935 return (dispatchKernelOnElecInteractionType<useSoftCore, scLambdasOrAlphasDiffer, true>(
936 elecInteractionTypeIsEwald, vdwModifierIsPotSwitch, useSimd));
940 return (dispatchKernelOnElecInteractionType<useSoftCore, scLambdasOrAlphasDiffer, false>(
941 elecInteractionTypeIsEwald, vdwModifierIsPotSwitch, useSimd));
945 template<bool useSoftCore>
946 static KernelFunction dispatchKernelOnScLambdasOrAlphasDifference(const bool scLambdasOrAlphasDiffer,
947 const bool vdwInteractionTypeIsEwald,
948 const bool elecInteractionTypeIsEwald,
949 const bool vdwModifierIsPotSwitch,
952 if (scLambdasOrAlphasDiffer)
954 return (dispatchKernelOnVdwInteractionType<useSoftCore, true>(
955 vdwInteractionTypeIsEwald, elecInteractionTypeIsEwald, vdwModifierIsPotSwitch, useSimd));
959 return (dispatchKernelOnVdwInteractionType<useSoftCore, false>(
960 vdwInteractionTypeIsEwald, elecInteractionTypeIsEwald, vdwModifierIsPotSwitch, useSimd));
964 static KernelFunction dispatchKernel(const bool scLambdasOrAlphasDiffer,
965 const bool vdwInteractionTypeIsEwald,
966 const bool elecInteractionTypeIsEwald,
967 const bool vdwModifierIsPotSwitch,
969 const interaction_const_t& ic)
971 if (ic.softCoreParameters->alphaCoulomb == 0 && ic.softCoreParameters->alphaVdw == 0)
973 return (dispatchKernelOnScLambdasOrAlphasDifference<false>(
974 scLambdasOrAlphasDiffer, vdwInteractionTypeIsEwald, elecInteractionTypeIsEwald,
975 vdwModifierIsPotSwitch, useSimd));
979 return (dispatchKernelOnScLambdasOrAlphasDifference<true>(
980 scLambdasOrAlphasDiffer, vdwInteractionTypeIsEwald, elecInteractionTypeIsEwald,
981 vdwModifierIsPotSwitch, useSimd));
986 void gmx_nb_free_energy_kernel(const t_nblist* nlist,
988 gmx::ForceWithShiftForces* ff,
989 const t_forcerec* fr,
990 const t_mdatoms* mdatoms,
991 nb_kernel_data_t* kernel_data,
994 const interaction_const_t& ic = *fr->ic;
995 GMX_ASSERT(EEL_PME_EWALD(ic.eeltype) || ic.eeltype == eelCUT || EEL_RF(ic.eeltype),
996 "Unsupported eeltype with free energy");
997 GMX_ASSERT(ic.softCoreParameters, "We need soft-core parameters");
999 const auto& scParams = *ic.softCoreParameters;
1000 const bool vdwInteractionTypeIsEwald = (EVDW_PME(ic.vdwtype));
1001 const bool elecInteractionTypeIsEwald = (EEL_PME_EWALD(ic.eeltype));
1002 const bool vdwModifierIsPotSwitch = (ic.vdw_modifier == eintmodPOTSWITCH);
1003 bool scLambdasOrAlphasDiffer = true;
1004 const bool useSimd = fr->use_simd_kernels;
1006 if (scParams.alphaCoulomb == 0 && scParams.alphaVdw == 0)
1008 scLambdasOrAlphasDiffer = false;
1012 if (kernel_data->lambda[efptCOUL] == kernel_data->lambda[efptVDW]
1013 && scParams.alphaCoulomb == scParams.alphaVdw)
1015 scLambdasOrAlphasDiffer = false;
1019 KernelFunction kernelFunc;
1020 kernelFunc = dispatchKernel(scLambdasOrAlphasDiffer, vdwInteractionTypeIsEwald,
1021 elecInteractionTypeIsEwald, vdwModifierIsPotSwitch, useSimd, ic);
1022 kernelFunc(nlist, xx, ff, fr, mdatoms, kernel_data, nrnb);