2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2016 by the GROMACS development team.
5 * Copyright (c) 2017,2018,2019,2020,2021, by the GROMACS development team, led by
6 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
7 * and including many others, as listed in the AUTHORS file in the
8 * top-level source directory and at http://www.gromacs.org.
10 * GROMACS is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public License
12 * as published by the Free Software Foundation; either version 2.1
13 * of the License, or (at your option) any later version.
15 * GROMACS is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with GROMACS; if not, see
22 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
23 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
25 * If you want to redistribute modifications to GROMACS, please
26 * consider that scientific software is very special. Version
27 * control is crucial - bugs must be traceable. We will be happy to
28 * consider code for inclusion in the official distribution, but
29 * derived work must not be called official GROMACS. Details are found
30 * in the README & COPYING files - if they are missing, get the
31 * official version at http://www.gromacs.org.
33 * To help us fund GROMACS development, we humbly ask that you cite
34 * the research papers on the package. Check out http://www.gromacs.org.
39 #include "gromacs/gmxlib/nrnb.h"
40 #include "gromacs/gmxlib/nonbonded/nb_free_energy.h"
41 #include "gromacs/gmxlib/nonbonded/nonbonded.h"
42 #include "gromacs/math/vectypes.h"
43 #include "gromacs/mdlib/enerdata_utils.h"
44 #include "gromacs/mdlib/force.h"
45 #include "gromacs/mdlib/gmx_omp_nthreads.h"
46 #include "gromacs/mdtypes/enerdata.h"
47 #include "gromacs/mdtypes/forceoutput.h"
48 #include "gromacs/mdtypes/forcerec.h"
49 #include "gromacs/mdtypes/inputrec.h"
50 #include "gromacs/mdtypes/interaction_const.h"
51 #include "gromacs/mdtypes/md_enums.h"
52 #include "gromacs/mdtypes/mdatom.h"
53 #include "gromacs/mdtypes/nblist.h"
54 #include "gromacs/mdtypes/simulation_workload.h"
55 #include "gromacs/nbnxm/gpu_data_mgmt.h"
56 #include "gromacs/nbnxm/nbnxm.h"
57 #include "gromacs/simd/simd.h"
58 #include "gromacs/timing/wallcycle.h"
59 #include "gromacs/utility/enumerationhelpers.h"
60 #include "gromacs/utility/fatalerror.h"
61 #include "gromacs/utility/gmxassert.h"
62 #include "gromacs/utility/real.h"
64 #include "kernel_common.h"
65 #include "nbnxm_gpu.h"
66 #include "nbnxm_simd.h"
67 #include "pairlistset.h"
68 #include "pairlistsets.h"
69 #include "kernels_reference/kernel_gpu_ref.h"
70 #define INCLUDE_KERNELFUNCTION_TABLES
71 #include "kernels_reference/kernel_ref.h"
72 #ifdef GMX_NBNXN_SIMD_2XNN
73 # include "kernels_simd_2xmm/kernels.h"
75 #ifdef GMX_NBNXN_SIMD_4XN
76 # include "kernels_simd_4xm/kernels.h"
78 #undef INCLUDE_FUNCTION_TABLES
80 /*! \brief Clears the energy group output buffers
82 * \param[in,out] out nbnxn kernel output struct
84 static void clearGroupEnergies(nbnxn_atomdata_output_t* out)
86 std::fill(out->Vvdw.begin(), out->Vvdw.end(), 0.0_real);
87 std::fill(out->Vc.begin(), out->Vc.end(), 0.0_real);
88 std::fill(out->VSvdw.begin(), out->VSvdw.end(), 0.0_real);
89 std::fill(out->VSc.begin(), out->VSc.end(), 0.0_real);
92 /*! \brief Reduce the group-pair energy buffers produced by a SIMD kernel
93 * to single terms in the output buffers.
95 * The SIMD kernels produce a large number of energy buffer in SIMD registers
96 * to avoid scattered reads and writes.
98 * \tparam unrollj The unroll size for j-particles in the SIMD kernel
99 * \param[in] numGroups The number of energy groups
100 * \param[in] numGroups_2log Log2 of numGroups, rounded up
101 * \param[in,out] out Struct with energy buffers
103 template<int unrollj>
104 static void reduceGroupEnergySimdBuffers(int numGroups, int numGroups_2log, nbnxn_atomdata_output_t* out)
106 const int unrollj_half = unrollj / 2;
107 /* Energies are stored in SIMD registers with size 2^numGroups_2log */
108 const int numGroupsStorage = (1 << numGroups_2log);
110 const real* gmx_restrict vVdwSimd = out->VSvdw.data();
111 const real* gmx_restrict vCoulombSimd = out->VSc.data();
112 real* gmx_restrict vVdw = out->Vvdw.data();
113 real* gmx_restrict vCoulomb = out->Vc.data();
115 /* The size of the SIMD energy group buffer array is:
116 * numGroups*numGroups*numGroupsStorage*unrollj_half*simd_width
118 for (int i = 0; i < numGroups; i++)
120 for (int j1 = 0; j1 < numGroups; j1++)
122 for (int j0 = 0; j0 < numGroups; j0++)
124 int c = ((i * numGroups + j1) * numGroupsStorage + j0) * unrollj_half * unrollj;
125 for (int s = 0; s < unrollj_half; s++)
127 vVdw[i * numGroups + j0] += vVdwSimd[c + 0];
128 vVdw[i * numGroups + j1] += vVdwSimd[c + 1];
129 vCoulomb[i * numGroups + j0] += vCoulombSimd[c + 0];
130 vCoulomb[i * numGroups + j1] += vCoulombSimd[c + 1];
138 CoulombKernelType getCoulombKernelType(const Nbnxm::EwaldExclusionType ewaldExclusionType,
139 const CoulombInteractionType coulombInteractionType,
140 const bool haveEqualCoulombVwdRadii)
143 if (EEL_RF(coulombInteractionType) || coulombInteractionType == CoulombInteractionType::Cut)
145 return CoulombKernelType::ReactionField;
149 if (ewaldExclusionType == Nbnxm::EwaldExclusionType::Table)
151 if (haveEqualCoulombVwdRadii)
153 return CoulombKernelType::Table;
157 return CoulombKernelType::TableTwin;
162 if (haveEqualCoulombVwdRadii)
164 return CoulombKernelType::Ewald;
168 return CoulombKernelType::EwaldTwin;
174 static int getVdwKernelType(const Nbnxm::KernelSetup& kernelSetup,
175 const nbnxn_atomdata_t::Params& nbatParams,
176 const interaction_const_t& ic)
178 if (ic.vdwtype == VanDerWaalsType::Cut)
180 switch (ic.vdw_modifier)
182 case InteractionModifiers::None:
183 case InteractionModifiers::PotShift:
184 switch (nbatParams.ljCombinationRule)
186 case LJCombinationRule::Geometric: return vdwktLJCUT_COMBGEOM;
187 case LJCombinationRule::LorentzBerthelot: return vdwktLJCUT_COMBLB;
188 case LJCombinationRule::None: return vdwktLJCUT_COMBNONE;
189 default: gmx_incons("Unknown combination rule");
191 case InteractionModifiers::ForceSwitch: return vdwktLJFORCESWITCH;
192 case InteractionModifiers::PotSwitch: return vdwktLJPOTSWITCH;
194 std::string errorMsg =
195 gmx::formatString("Unsupported VdW interaction modifier %s (%d)",
196 enumValueToString(ic.vdw_modifier),
197 static_cast<int>(ic.vdw_modifier));
198 gmx_incons(errorMsg);
201 else if (ic.vdwtype == VanDerWaalsType::Pme)
203 if (ic.ljpme_comb_rule == LongRangeVdW::Geom)
205 return vdwktLJEWALDCOMBGEOM;
209 /* At setup we (should have) selected the C reference kernel */
210 GMX_RELEASE_ASSERT(kernelSetup.kernelType == Nbnxm::KernelType::Cpu4x4_PlainC,
211 "Only the C reference nbnxn SIMD kernel supports LJ-PME with LB "
212 "combination rules");
213 return vdwktLJEWALDCOMBLB;
218 std::string errorMsg = gmx::formatString("Unsupported VdW interaction type %s (%d)",
219 enumValueToString(ic.vdwtype),
220 static_cast<int>(ic.vdwtype));
221 gmx_incons(errorMsg);
225 /*! \brief Dispatches the non-bonded N versus M atom cluster CPU kernels.
227 * OpenMP parallelization is performed within this function.
228 * Energy reduction, but not force and shift force reduction, is performed
229 * within this function.
231 * \param[in] pairlistSet Pairlists with local or non-local interactions to compute
232 * \param[in] kernelSetup The non-bonded kernel setup
233 * \param[in,out] nbat The atomdata for the interactions
234 * \param[in] ic Non-bonded interaction constants
235 * \param[in] shiftVectors The PBC shift vectors
236 * \param[in] stepWork Flags that tell what to compute
237 * \param[in] clearF Enum that tells if to clear the force output buffer
238 * \param[out] vCoulomb Output buffer for Coulomb energies
239 * \param[out] vVdw Output buffer for Van der Waals energies
240 * \param[in] wcycle Pointer to cycle counting data structure.
242 static void nbnxn_kernel_cpu(const PairlistSet& pairlistSet,
243 const Nbnxm::KernelSetup& kernelSetup,
244 nbnxn_atomdata_t* nbat,
245 const interaction_const_t& ic,
246 gmx::ArrayRef<const gmx::RVec> shiftVectors,
247 const gmx::StepWorkload& stepWork,
251 gmx_wallcycle* wcycle)
254 const nbnxn_atomdata_t::Params& nbatParams = nbat->params();
256 const int coulkt = static_cast<int>(getCoulombKernelType(
257 kernelSetup.ewaldExclusionType, ic.eeltype, (ic.rcoulomb == ic.rvdw)));
258 const int vdwkt = getVdwKernelType(kernelSetup, nbatParams, ic);
260 gmx::ArrayRef<const NbnxnPairlistCpu> pairlists = pairlistSet.cpuLists();
262 const auto* shiftVecPointer = as_rvec_array(shiftVectors.data());
264 int gmx_unused nthreads = gmx_omp_nthreads_get(ModuleMultiThread::Nonbonded);
265 wallcycle_sub_start(wcycle, WallCycleSubCounter::NonbondedClear);
266 #pragma omp parallel for schedule(static) num_threads(nthreads)
267 for (gmx::index nb = 0; nb < pairlists.ssize(); nb++)
269 // Presently, the kernels do not call C++ code that can throw,
270 // so no need for a try/catch pair in this OpenMP region.
271 nbnxn_atomdata_output_t* out = &nbat->out[nb];
273 if (clearF == enbvClearFYes)
275 clearForceBuffer(nbat, nb);
277 clear_fshift(out->fshift.data());
282 wallcycle_sub_stop(wcycle, WallCycleSubCounter::NonbondedClear);
283 wallcycle_sub_start(wcycle, WallCycleSubCounter::NonbondedKernel);
286 // TODO: Change to reference
287 const NbnxnPairlistCpu* pairlist = &pairlists[nb];
289 if (!stepWork.computeEnergy)
291 /* Don't calculate energies */
292 switch (kernelSetup.kernelType)
294 case Nbnxm::KernelType::Cpu4x4_PlainC:
295 nbnxn_kernel_noener_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVecPointer, out);
297 #ifdef GMX_NBNXN_SIMD_2XNN
298 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
299 nbnxm_kernel_noener_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVecPointer, out);
302 #ifdef GMX_NBNXN_SIMD_4XN
303 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
304 nbnxm_kernel_noener_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVecPointer, out);
307 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
310 else if (out->Vvdw.size() == 1)
312 /* A single energy group (pair) */
316 switch (kernelSetup.kernelType)
318 case Nbnxm::KernelType::Cpu4x4_PlainC:
319 nbnxn_kernel_ener_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVecPointer, out);
321 #ifdef GMX_NBNXN_SIMD_2XNN
322 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
323 nbnxm_kernel_ener_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVecPointer, out);
326 #ifdef GMX_NBNXN_SIMD_4XN
327 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
328 nbnxm_kernel_ener_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVecPointer, out);
331 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
336 /* Calculate energy group contributions */
337 clearGroupEnergies(out);
341 switch (kernelSetup.kernelType)
343 case Nbnxm::KernelType::Cpu4x4_PlainC:
344 unrollj = c_nbnxnCpuIClusterSize;
345 nbnxn_kernel_energrp_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVecPointer, out);
347 #ifdef GMX_NBNXN_SIMD_2XNN
348 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
349 unrollj = GMX_SIMD_REAL_WIDTH / 2;
350 nbnxm_kernel_energrp_simd_2xmm[coulkt][vdwkt](
351 pairlist, nbat, &ic, shiftVecPointer, out);
354 #ifdef GMX_NBNXN_SIMD_4XN
355 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
356 unrollj = GMX_SIMD_REAL_WIDTH;
357 nbnxm_kernel_energrp_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVecPointer, out);
360 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
363 if (kernelSetup.kernelType != Nbnxm::KernelType::Cpu4x4_PlainC)
368 reduceGroupEnergySimdBuffers<2>(nbatParams.nenergrp, nbatParams.neg_2log, out);
371 reduceGroupEnergySimdBuffers<4>(nbatParams.nenergrp, nbatParams.neg_2log, out);
374 reduceGroupEnergySimdBuffers<8>(nbatParams.nenergrp, nbatParams.neg_2log, out);
376 default: GMX_RELEASE_ASSERT(false, "Unsupported j-unroll size");
381 wallcycle_sub_stop(wcycle, WallCycleSubCounter::NonbondedKernel);
383 if (stepWork.computeEnergy)
385 reduce_energies_over_lists(nbat, pairlists.ssize(), vVdw, vCoulomb);
389 static void accountFlops(t_nrnb* nrnb,
390 const PairlistSet& pairlistSet,
391 const nonbonded_verlet_t& nbv,
392 const interaction_const_t& ic,
393 const gmx::StepWorkload& stepWork)
395 const bool usingGpuKernels = nbv.useGpu();
397 int enr_nbnxn_kernel_ljc = eNRNB;
398 if (EEL_RF(ic.eeltype) || ic.eeltype == CoulombInteractionType::Cut)
400 enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
402 else if ((!usingGpuKernels && nbv.kernelSetup().ewaldExclusionType == Nbnxm::EwaldExclusionType::Analytical)
403 || (usingGpuKernels && Nbnxm::gpu_is_kernel_ewald_analytical(nbv.gpu_nbv)))
405 enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
409 enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
411 int enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
412 if (stepWork.computeEnergy)
414 /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
415 enr_nbnxn_kernel_ljc += 1;
416 enr_nbnxn_kernel_lj += 1;
419 inc_nrnb(nrnb, enr_nbnxn_kernel_ljc, pairlistSet.natpair_ljq_);
420 inc_nrnb(nrnb, enr_nbnxn_kernel_lj, pairlistSet.natpair_lj_);
421 /* The Coulomb-only kernels are offset -eNR_NBNXN_LJ_RF+eNR_NBNXN_RF */
422 inc_nrnb(nrnb, enr_nbnxn_kernel_ljc - eNR_NBNXN_LJ_RF + eNR_NBNXN_RF, pairlistSet.natpair_q_);
424 if (ic.vdw_modifier == InteractionModifiers::ForceSwitch)
426 /* We add up the switch cost separately */
428 eNR_NBNXN_ADD_LJ_FSW + (stepWork.computeEnergy ? 1 : 0),
429 pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
431 if (ic.vdw_modifier == InteractionModifiers::PotSwitch)
433 /* We add up the switch cost separately */
435 eNR_NBNXN_ADD_LJ_PSW + (stepWork.computeEnergy ? 1 : 0),
436 pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
438 if (ic.vdwtype == VanDerWaalsType::Pme)
440 /* We add up the LJ Ewald cost separately */
442 eNR_NBNXN_ADD_LJ_EWALD + (stepWork.computeEnergy ? 1 : 0),
443 pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
447 void nonbonded_verlet_t::dispatchNonbondedKernel(gmx::InteractionLocality iLocality,
448 const interaction_const_t& ic,
449 const gmx::StepWorkload& stepWork,
451 gmx::ArrayRef<const gmx::RVec> shiftvec,
452 gmx::ArrayRef<real> repulsionDispersionSR,
453 gmx::ArrayRef<real> CoulombSR,
456 const PairlistSet& pairlistSet = pairlistSets().pairlistSet(iLocality);
458 switch (kernelSetup().kernelType)
460 case Nbnxm::KernelType::Cpu4x4_PlainC:
461 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
462 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
463 nbnxn_kernel_cpu(pairlistSet,
471 repulsionDispersionSR.data(),
475 case Nbnxm::KernelType::Gpu8x8x8:
476 Nbnxm::gpu_launch_kernel(gpu_nbv, stepWork, iLocality);
479 case Nbnxm::KernelType::Cpu8x8x8_PlainC:
480 nbnxn_kernel_gpu_ref(pairlistSet.gpuList(),
487 nbat->out[0].fshift.data(),
489 repulsionDispersionSR.data());
492 default: GMX_RELEASE_ASSERT(false, "Invalid nonbonded kernel type passed!");
495 accountFlops(nrnb, pairlistSet, *this, ic, stepWork);
498 void nonbonded_verlet_t::dispatchFreeEnergyKernel(gmx::InteractionLocality iLocality,
499 gmx::ArrayRef<const gmx::RVec> coords,
500 gmx::ForceWithShiftForces* forceWithShiftForces,
504 const interaction_const_t& ic,
505 gmx::ArrayRef<const gmx::RVec> shiftvec,
506 gmx::ArrayRef<const real> nbfp,
507 gmx::ArrayRef<const real> nbfp_grid,
508 gmx::ArrayRef<const real> chargeA,
509 gmx::ArrayRef<const real> chargeB,
510 gmx::ArrayRef<const int> typeA,
511 gmx::ArrayRef<const int> typeB,
513 gmx::ArrayRef<const real> lambda,
514 gmx_enerdata_t* enerd,
515 const gmx::StepWorkload& stepWork,
518 const auto nbl_fep = pairlistSets().pairlistSet(iLocality).fepLists();
520 /* When the first list is empty, all are empty and there is nothing to do */
521 if (!pairlistSets().params().haveFep || nbl_fep[0]->nrj == 0)
527 /* Add short-range interactions */
528 donb_flags |= GMX_NONBONDED_DO_SR;
530 if (stepWork.computeForces)
532 donb_flags |= GMX_NONBONDED_DO_FORCE;
534 if (stepWork.computeVirial)
536 donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
538 if (stepWork.computeEnergy)
540 donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
543 gmx::EnumerationArray<FreeEnergyPerturbationCouplingType, real> dvdl_nb = { 0 };
544 int kernelFlags = donb_flags;
545 gmx::ArrayRef<const real> kernelLambda = lambda;
546 gmx::ArrayRef<real> kernelDvdl = dvdl_nb;
548 gmx::ArrayRef<real> energygrp_elec = enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR];
549 gmx::ArrayRef<real> energygrp_vdw = enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR];
551 GMX_ASSERT(gmx_omp_nthreads_get(ModuleMultiThread::Nonbonded) == nbl_fep.ssize(),
552 "Number of lists should be same as number of NB threads");
554 wallcycle_sub_start(wcycle_, WallCycleSubCounter::NonbondedFep);
555 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
556 for (gmx::index th = 0; th < nbl_fep.ssize(); th++)
560 gmx_nb_free_energy_kernel(*nbl_fep[th],
562 forceWithShiftForces,
581 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
584 if (fepvals->sc_alpha != 0)
586 enerd->dvdl_nonlin[FreeEnergyPerturbationCouplingType::Vdw] +=
587 dvdl_nb[FreeEnergyPerturbationCouplingType::Vdw];
588 enerd->dvdl_nonlin[FreeEnergyPerturbationCouplingType::Coul] +=
589 dvdl_nb[FreeEnergyPerturbationCouplingType::Coul];
593 enerd->dvdl_lin[FreeEnergyPerturbationCouplingType::Vdw] +=
594 dvdl_nb[FreeEnergyPerturbationCouplingType::Vdw];
595 enerd->dvdl_lin[FreeEnergyPerturbationCouplingType::Coul] +=
596 dvdl_nb[FreeEnergyPerturbationCouplingType::Coul];
599 /* If we do foreign lambda and we have soft-core interactions
600 * we have to recalculate the (non-linear) energies contributions.
602 if (fepvals->n_lambda > 0 && stepWork.computeDhdl && fepvals->sc_alpha != 0)
604 gmx::EnumerationArray<FreeEnergyPerturbationCouplingType, real> lam_i;
605 kernelFlags = (donb_flags & ~(GMX_NONBONDED_DO_FORCE | GMX_NONBONDED_DO_SHIFTFORCE))
606 | GMX_NONBONDED_DO_FOREIGNLAMBDA;
607 kernelLambda = lam_i;
608 kernelDvdl = dvdl_nb;
609 gmx::ArrayRef<real> energygrp_elec =
610 foreignEnergyGroups_->energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR];
611 gmx::ArrayRef<real> energygrp_vdw =
612 foreignEnergyGroups_->energyGroupPairTerms[NonBondedEnergyTerms::LJSR];
614 for (gmx::index i = 0; i < 1 + enerd->foreignLambdaTerms.numLambdas(); i++)
616 std::fill(std::begin(dvdl_nb), std::end(dvdl_nb), 0);
617 for (int j = 0; j < static_cast<int>(FreeEnergyPerturbationCouplingType::Count); j++)
619 lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i - 1]);
621 foreignEnergyGroups_->clear();
622 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
623 for (gmx::index th = 0; th < nbl_fep.ssize(); th++)
627 gmx_nb_free_energy_kernel(*nbl_fep[th],
629 forceWithShiftForces,
648 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
650 std::array<real, F_NRE> foreign_term = { 0 };
651 sum_epot(*foreignEnergyGroups_, foreign_term.data());
652 enerd->foreignLambdaTerms.accumulate(
654 foreign_term[F_EPOT],
655 dvdl_nb[FreeEnergyPerturbationCouplingType::Vdw]
656 + dvdl_nb[FreeEnergyPerturbationCouplingType::Coul]);
659 wallcycle_sub_stop(wcycle_, WallCycleSubCounter::NonbondedFep);