src/gromacs/nbnxm/kerneldispatch.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   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.
   9  *
  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.
  14  *
  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.
  19  *
  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.
  24  *
  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.
  32  *
  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.
  35  */
  36
  37 #include "gmxpre.h"
  38
  39 #include "gromacs/gmxlib/nrnb.h"
  40 #include "gromacs/gmxlib/nonbonded/nb_free_energy.h"
  41 #include "gromacs/gmxlib/nonbonded/nb_kernel.h"
  42 #include "gromacs/gmxlib/nonbonded/nonbonded.h"
  43 #include "gromacs/math/vectypes.h"
  44 #include "gromacs/mdlib/enerdata_utils.h"
  45 #include "gromacs/mdlib/force.h"
  46 #include "gromacs/mdlib/gmx_omp_nthreads.h"
  47 #include "gromacs/mdtypes/enerdata.h"
  48 #include "gromacs/mdtypes/forceoutput.h"
  49 #include "gromacs/mdtypes/forcerec.h"
  50 #include "gromacs/mdtypes/inputrec.h"
  51 #include "gromacs/mdtypes/interaction_const.h"
  52 #include "gromacs/mdtypes/md_enums.h"
  53 #include "gromacs/mdtypes/mdatom.h"
  54 #include "gromacs/mdtypes/nblist.h"
  55 #include "gromacs/mdtypes/simulation_workload.h"
  56 #include "gromacs/nbnxm/gpu_data_mgmt.h"
  57 #include "gromacs/nbnxm/nbnxm.h"
  58 #include "gromacs/simd/simd.h"
  59 #include "gromacs/timing/wallcycle.h"
  60 #include "gromacs/utility/enumerationhelpers.h"
  61 #include "gromacs/utility/fatalerror.h"
  62 #include "gromacs/utility/gmxassert.h"
  63 #include "gromacs/utility/real.h"
  64
  65 #include "kernel_common.h"
  66 #include "nbnxm_gpu.h"
  67 #include "nbnxm_gpu_data_mgmt.h"
  68 #include "nbnxm_simd.h"
  69 #include "pairlistset.h"
  70 #include "pairlistsets.h"
  71 #include "kernels_reference/kernel_gpu_ref.h"
  72 #define INCLUDE_KERNELFUNCTION_TABLES
  73 #include "kernels_reference/kernel_ref.h"
  74 #ifdef GMX_NBNXN_SIMD_2XNN
  75 #    include "kernels_simd_2xmm/kernels.h"
  76 #endif
  77 #ifdef GMX_NBNXN_SIMD_4XN
  78 #    include "kernels_simd_4xm/kernels.h"
  79 #endif
  80 #undef INCLUDE_FUNCTION_TABLES
  81
  82 /*! \brief Clears the energy group output buffers
  83  *
  84  * \param[in,out] out  nbnxn kernel output struct
  85  */
  86 static void clearGroupEnergies(nbnxn_atomdata_output_t* out)
  87 {
  88     std::fill(out->Vvdw.begin(), out->Vvdw.end(), 0.0_real);
  89     std::fill(out->Vc.begin(), out->Vc.end(), 0.0_real);
  90     std::fill(out->VSvdw.begin(), out->VSvdw.end(), 0.0_real);
  91     std::fill(out->VSc.begin(), out->VSc.end(), 0.0_real);
  92 }
  93
  94 /*! \brief Reduce the group-pair energy buffers produced by a SIMD kernel
  95  * to single terms in the output buffers.
  96  *
  97  * The SIMD kernels produce a large number of energy buffer in SIMD registers
  98  * to avoid scattered reads and writes.
  99  *
 100  * \tparam        unrollj         The unroll size for j-particles in the SIMD kernel
 101  * \param[in]     numGroups       The number of energy groups
 102  * \param[in]     numGroups_2log  Log2 of numGroups, rounded up
 103  * \param[in,out] out             Struct with energy buffers
 104  */
 105 template<int unrollj>
 106 static void reduceGroupEnergySimdBuffers(int numGroups, int numGroups_2log, nbnxn_atomdata_output_t* out)
 107 {
 108     const int unrollj_half = unrollj / 2;
 109     /* Energies are stored in SIMD registers with size 2^numGroups_2log */
 110     const int numGroupsStorage = (1 << numGroups_2log);
 111
 112     const real* gmx_restrict vVdwSimd     = out->VSvdw.data();
 113     const real* gmx_restrict vCoulombSimd = out->VSc.data();
 114     real* gmx_restrict vVdw               = out->Vvdw.data();
 115     real* gmx_restrict vCoulomb           = out->Vc.data();
 116
 117     /* The size of the SIMD energy group buffer array is:
 118      * numGroups*numGroups*numGroupsStorage*unrollj_half*simd_width
 119      */
 120     for (int i = 0; i < numGroups; i++)
 121     {
 122         for (int j1 = 0; j1 < numGroups; j1++)
 123         {
 124             for (int j0 = 0; j0 < numGroups; j0++)
 125             {
 126                 int c = ((i * numGroups + j1) * numGroupsStorage + j0) * unrollj_half * unrollj;
 127                 for (int s = 0; s < unrollj_half; s++)
 128                 {
 129                     vVdw[i * numGroups + j0] += vVdwSimd[c + 0];
 130                     vVdw[i * numGroups + j1] += vVdwSimd[c + 1];
 131                     vCoulomb[i * numGroups + j0] += vCoulombSimd[c + 0];
 132                     vCoulomb[i * numGroups + j1] += vCoulombSimd[c + 1];
 133                     c += unrollj + 2;
 134                 }
 135             }
 136         }
 137     }
 138 }
 139
 140 static int getCoulombKernelType(const Nbnxm::KernelSetup& kernelSetup, const interaction_const_t& ic)
 141 {
 142
 143     if (EEL_RF(ic.eeltype) || ic.eeltype == CoulombInteractionType::Cut)
 144     {
 145         return coulktRF;
 146     }
 147     else
 148     {
 149         if (kernelSetup.ewaldExclusionType == Nbnxm::EwaldExclusionType::Table)
 150         {
 151             if (ic.rcoulomb == ic.rvdw)
 152             {
 153                 return coulktTAB;
 154             }
 155             else
 156             {
 157                 return coulktTAB_TWIN;
 158             }
 159         }
 160         else
 161         {
 162             if (ic.rcoulomb == ic.rvdw)
 163             {
 164                 return coulktEWALD;
 165             }
 166             else
 167             {
 168                 return coulktEWALD_TWIN;
 169             }
 170         }
 171     }
 172 }
 173
 174 static int getVdwKernelType(const Nbnxm::KernelSetup&       kernelSetup,
 175                             const nbnxn_atomdata_t::Params& nbatParams,
 176                             const interaction_const_t&      ic)
 177 {
 178     if (ic.vdwtype == VanDerWaalsType::Cut)
 179     {
 180         switch (ic.vdw_modifier)
 181         {
 182             case InteractionModifiers::None:
 183             case InteractionModifiers::PotShift:
 184                 switch (nbatParams.ljCombinationRule)
 185                 {
 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");
 190                 }
 191             case InteractionModifiers::ForceSwitch: return vdwktLJFORCESWITCH;
 192             case InteractionModifiers::PotSwitch: return vdwktLJPOTSWITCH;
 193             default:
 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);
 199         }
 200     }
 201     else if (ic.vdwtype == VanDerWaalsType::Pme)
 202     {
 203         if (ic.ljpme_comb_rule == LongRangeVdW::Geom)
 204         {
 205             return vdwktLJEWALDCOMBGEOM;
 206         }
 207         else
 208         {
 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;
 214         }
 215     }
 216     else
 217     {
 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);
 222     }
 223 }
 224
 225 /*! \brief Dispatches the non-bonded N versus M atom cluster CPU kernels.
 226  *
 227  * OpenMP parallelization is performed within this function.
 228  * Energy reduction, but not force and shift force reduction, is performed
 229  * within this function.
 230  *
 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.
 241  */
 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                              rvec*                      shiftVectors,
 247                              const gmx::StepWorkload&   stepWork,
 248                              int                        clearF,
 249                              real*                      vCoulomb,
 250                              real*                      vVdw,
 251                              gmx_wallcycle*             wcycle)
 252 {
 253
 254     const nbnxn_atomdata_t::Params& nbatParams = nbat->params();
 255
 256     const int coulkt = getCoulombKernelType(kernelSetup, ic);
 257     const int vdwkt  = getVdwKernelType(kernelSetup, nbatParams, ic);
 258
 259     gmx::ArrayRef<const NbnxnPairlistCpu> pairlists = pairlistSet.cpuLists();
 260
 261     int gmx_unused nthreads = gmx_omp_nthreads_get(emntNonbonded);
 262     wallcycle_sub_start(wcycle, ewcsNONBONDED_CLEAR);
 263 #pragma omp parallel for schedule(static) num_threads(nthreads)
 264     for (gmx::index nb = 0; nb < pairlists.ssize(); nb++)
 265     {
 266         // Presently, the kernels do not call C++ code that can throw,
 267         // so no need for a try/catch pair in this OpenMP region.
 268         nbnxn_atomdata_output_t* out = &nbat->out[nb];
 269
 270         if (clearF == enbvClearFYes)
 271         {
 272             clearForceBuffer(nbat, nb);
 273
 274             clear_fshift(out->fshift.data());
 275         }
 276
 277         if (nb == 0)
 278         {
 279             wallcycle_sub_stop(wcycle, ewcsNONBONDED_CLEAR);
 280             wallcycle_sub_start(wcycle, ewcsNONBONDED_KERNEL);
 281         }
 282
 283         // TODO: Change to reference
 284         const NbnxnPairlistCpu* pairlist = &pairlists[nb];
 285
 286         if (!stepWork.computeEnergy)
 287         {
 288             /* Don't calculate energies */
 289             switch (kernelSetup.kernelType)
 290             {
 291                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 292                     nbnxn_kernel_noener_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 293                     break;
 294 #ifdef GMX_NBNXN_SIMD_2XNN
 295                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 296                     nbnxm_kernel_noener_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 297                     break;
 298 #endif
 299 #ifdef GMX_NBNXN_SIMD_4XN
 300                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 301                     nbnxm_kernel_noener_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 302                     break;
 303 #endif
 304                 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 305             }
 306         }
 307         else if (out->Vvdw.size() == 1)
 308         {
 309             /* A single energy group (pair) */
 310             out->Vvdw[0] = 0;
 311             out->Vc[0]   = 0;
 312
 313             switch (kernelSetup.kernelType)
 314             {
 315                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 316                     nbnxn_kernel_ener_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 317                     break;
 318 #ifdef GMX_NBNXN_SIMD_2XNN
 319                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 320                     nbnxm_kernel_ener_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 321                     break;
 322 #endif
 323 #ifdef GMX_NBNXN_SIMD_4XN
 324                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 325                     nbnxm_kernel_ener_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 326                     break;
 327 #endif
 328                 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 329             }
 330         }
 331         else
 332         {
 333             /* Calculate energy group contributions */
 334             clearGroupEnergies(out);
 335
 336             int unrollj = 0;
 337
 338             switch (kernelSetup.kernelType)
 339             {
 340                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 341                     unrollj = c_nbnxnCpuIClusterSize;
 342                     nbnxn_kernel_energrp_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 343                     break;
 344 #ifdef GMX_NBNXN_SIMD_2XNN
 345                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 346                     unrollj = GMX_SIMD_REAL_WIDTH / 2;
 347                     nbnxm_kernel_energrp_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 348                     break;
 349 #endif
 350 #ifdef GMX_NBNXN_SIMD_4XN
 351                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 352                     unrollj = GMX_SIMD_REAL_WIDTH;
 353                     nbnxm_kernel_energrp_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 354                     break;
 355 #endif
 356                 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 357             }
 358
 359             if (kernelSetup.kernelType != Nbnxm::KernelType::Cpu4x4_PlainC)
 360             {
 361                 switch (unrollj)
 362                 {
 363                     case 2:
 364                         reduceGroupEnergySimdBuffers<2>(nbatParams.nenergrp, nbatParams.neg_2log, out);
 365                         break;
 366                     case 4:
 367                         reduceGroupEnergySimdBuffers<4>(nbatParams.nenergrp, nbatParams.neg_2log, out);
 368                         break;
 369                     case 8:
 370                         reduceGroupEnergySimdBuffers<8>(nbatParams.nenergrp, nbatParams.neg_2log, out);
 371                         break;
 372                     default: GMX_RELEASE_ASSERT(false, "Unsupported j-unroll size");
 373                 }
 374             }
 375         }
 376     }
 377     wallcycle_sub_stop(wcycle, ewcsNONBONDED_KERNEL);
 378
 379     if (stepWork.computeEnergy)
 380     {
 381         reduce_energies_over_lists(nbat, pairlists.ssize(), vVdw, vCoulomb);
 382     }
 383 }
 384
 385 static void accountFlops(t_nrnb*                    nrnb,
 386                          const PairlistSet&         pairlistSet,
 387                          const nonbonded_verlet_t&  nbv,
 388                          const interaction_const_t& ic,
 389                          const gmx::StepWorkload&   stepWork)
 390 {
 391     const bool usingGpuKernels = nbv.useGpu();
 392
 393     int enr_nbnxn_kernel_ljc = eNRNB;
 394     if (EEL_RF(ic.eeltype) || ic.eeltype == CoulombInteractionType::Cut)
 395     {
 396         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
 397     }
 398     else if ((!usingGpuKernels && nbv.kernelSetup().ewaldExclusionType == Nbnxm::EwaldExclusionType::Analytical)
 399              || (usingGpuKernels && Nbnxm::gpu_is_kernel_ewald_analytical(nbv.gpu_nbv)))
 400     {
 401         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
 402     }
 403     else
 404     {
 405         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
 406     }
 407     int enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
 408     if (stepWork.computeEnergy)
 409     {
 410         /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
 411         enr_nbnxn_kernel_ljc += 1;
 412         enr_nbnxn_kernel_lj += 1;
 413     }
 414
 415     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc, pairlistSet.natpair_ljq_);
 416     inc_nrnb(nrnb, enr_nbnxn_kernel_lj, pairlistSet.natpair_lj_);
 417     /* The Coulomb-only kernels are offset -eNR_NBNXN_LJ_RF+eNR_NBNXN_RF */
 418     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc - eNR_NBNXN_LJ_RF + eNR_NBNXN_RF, pairlistSet.natpair_q_);
 419
 420     if (ic.vdw_modifier == InteractionModifiers::ForceSwitch)
 421     {
 422         /* We add up the switch cost separately */
 423         inc_nrnb(nrnb,
 424                  eNR_NBNXN_ADD_LJ_FSW + (stepWork.computeEnergy ? 1 : 0),
 425                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 426     }
 427     if (ic.vdw_modifier == InteractionModifiers::PotSwitch)
 428     {
 429         /* We add up the switch cost separately */
 430         inc_nrnb(nrnb,
 431                  eNR_NBNXN_ADD_LJ_PSW + (stepWork.computeEnergy ? 1 : 0),
 432                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 433     }
 434     if (ic.vdwtype == VanDerWaalsType::Pme)
 435     {
 436         /* We add up the LJ Ewald cost separately */
 437         inc_nrnb(nrnb,
 438                  eNR_NBNXN_ADD_LJ_EWALD + (stepWork.computeEnergy ? 1 : 0),
 439                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 440     }
 441 }
 442
 443 void nonbonded_verlet_t::dispatchNonbondedKernel(gmx::InteractionLocality   iLocality,
 444                                                  const interaction_const_t& ic,
 445                                                  const gmx::StepWorkload&   stepWork,
 446                                                  int                        clearF,
 447                                                  const t_forcerec&          fr,
 448                                                  gmx_enerdata_t*            enerd,
 449                                                  t_nrnb*                    nrnb)
 450 {
 451     const PairlistSet& pairlistSet = pairlistSets().pairlistSet(iLocality);
 452
 453     switch (kernelSetup().kernelType)
 454     {
 455         case Nbnxm::KernelType::Cpu4x4_PlainC:
 456         case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 457         case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 458             nbnxn_kernel_cpu(
 459                     pairlistSet,
 460                     kernelSetup(),
 461                     nbat.get(),
 462                     ic,
 463                     fr.shift_vec,
 464                     stepWork,
 465                     clearF,
 466                     enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR].data(),
 467                     fr.bBHAM ? enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::BuckinghamSR].data()
 468                              : enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR].data(),
 469                     wcycle_);
 470             break;
 471
 472         case Nbnxm::KernelType::Gpu8x8x8:
 473             Nbnxm::gpu_launch_kernel(gpu_nbv, stepWork, iLocality);
 474             break;
 475
 476         case Nbnxm::KernelType::Cpu8x8x8_PlainC:
 477             nbnxn_kernel_gpu_ref(
 478                     pairlistSet.gpuList(),
 479                     nbat.get(),
 480                     &ic,
 481                     fr.shift_vec,
 482                     stepWork,
 483                     clearF,
 484                     nbat->out[0].f,
 485                     nbat->out[0].fshift.data(),
 486                     enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR].data(),
 487                     fr.bBHAM ? enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::BuckinghamSR].data()
 488                              : enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR].data());
 489             break;
 490
 491         default: GMX_RELEASE_ASSERT(false, "Invalid nonbonded kernel type passed!");
 492     }
 493
 494     accountFlops(nrnb, pairlistSet, *this, ic, stepWork);
 495 }
 496
 497 void nonbonded_verlet_t::dispatchFreeEnergyKernel(gmx::InteractionLocality   iLocality,
 498                                                   const t_forcerec*          fr,
 499                                                   rvec                       x[],
 500                                                   gmx::ForceWithShiftForces* forceWithShiftForces,
 501                                                   const t_mdatoms&           mdatoms,
 502                                                   t_lambda*                  fepvals,
 503                                                   gmx::ArrayRef<const real>  lambda,
 504                                                   gmx_enerdata_t*            enerd,
 505                                                   const gmx::StepWorkload&   stepWork,
 506                                                   t_nrnb*                    nrnb)
 507 {
 508     const auto nbl_fep = pairlistSets().pairlistSet(iLocality).fepLists();
 509
 510     /* When the first list is empty, all are empty and there is nothing to do */
 511     if (!pairlistSets().params().haveFep || nbl_fep[0]->nrj == 0)
 512     {
 513         return;
 514     }
 515
 516     int donb_flags = 0;
 517     /* Add short-range interactions */
 518     donb_flags |= GMX_NONBONDED_DO_SR;
 519
 520     if (stepWork.computeForces)
 521     {
 522         donb_flags |= GMX_NONBONDED_DO_FORCE;
 523     }
 524     if (stepWork.computeVirial)
 525     {
 526         donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 527     }
 528     if (stepWork.computeEnergy)
 529     {
 530         donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 531     }
 532
 533     nb_kernel_data_t                                                kernel_data;
 534     gmx::EnumerationArray<FreeEnergyPerturbationCouplingType, real> dvdl_nb = { 0 };
 535     kernel_data.flags                                                       = donb_flags;
 536     kernel_data.lambda                                                      = lambda;
 537     kernel_data.dvdl                                                        = dvdl_nb;
 538
 539     gmx::ArrayRef<real> energygrp_elec = enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR];
 540     gmx::ArrayRef<real> energygrp_vdw = enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR];
 541
 542     GMX_ASSERT(gmx_omp_nthreads_get(emntNonbonded) == nbl_fep.ssize(),
 543                "Number of lists should be same as number of NB threads");
 544
 545     wallcycle_sub_start(wcycle_, ewcsNONBONDED_FEP);
 546 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
 547     for (gmx::index th = 0; th < nbl_fep.ssize(); th++)
 548     {
 549         try
 550         {
 551             gmx_nb_free_energy_kernel(
 552                     nbl_fep[th].get(), x, forceWithShiftForces, fr, &mdatoms, &kernel_data, energygrp_elec, energygrp_vdw, nrnb);
 553         }
 554         GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
 555     }
 556
 557     if (fepvals->sc_alpha != 0)
 558     {
 559         enerd->dvdl_nonlin[FreeEnergyPerturbationCouplingType::Vdw] +=
 560                 dvdl_nb[FreeEnergyPerturbationCouplingType::Vdw];
 561         enerd->dvdl_nonlin[FreeEnergyPerturbationCouplingType::Coul] +=
 562                 dvdl_nb[FreeEnergyPerturbationCouplingType::Coul];
 563     }
 564     else
 565     {
 566         enerd->dvdl_lin[FreeEnergyPerturbationCouplingType::Vdw] +=
 567                 dvdl_nb[FreeEnergyPerturbationCouplingType::Vdw];
 568         enerd->dvdl_lin[FreeEnergyPerturbationCouplingType::Coul] +=
 569                 dvdl_nb[FreeEnergyPerturbationCouplingType::Coul];
 570     }
 571
 572     /* If we do foreign lambda and we have soft-core interactions
 573      * we have to recalculate the (non-linear) energies contributions.
 574      */
 575     if (fepvals->n_lambda > 0 && stepWork.computeDhdl && fepvals->sc_alpha != 0)
 576     {
 577         gmx::EnumerationArray<FreeEnergyPerturbationCouplingType, real> lam_i;
 578         kernel_data.flags = (donb_flags & ~(GMX_NONBONDED_DO_FORCE | GMX_NONBONDED_DO_SHIFTFORCE))
 579                             | GMX_NONBONDED_DO_FOREIGNLAMBDA;
 580         kernel_data.lambda = lam_i;
 581         kernel_data.dvdl   = dvdl_nb;
 582         gmx::ArrayRef<real> energygrp_elec =
 583                 enerd->foreign_grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR];
 584         gmx::ArrayRef<real> energygrp_vdw =
 585                 enerd->foreign_grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR];
 586
 587         for (gmx::index i = 0; i < 1 + enerd->foreignLambdaTerms.numLambdas(); i++)
 588         {
 589             std::fill(std::begin(dvdl_nb), std::end(dvdl_nb), 0);
 590             for (int j = 0; j < static_cast<int>(FreeEnergyPerturbationCouplingType::Count); j++)
 591             {
 592                 lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i - 1]);
 593             }
 594             reset_foreign_enerdata(enerd);
 595 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
 596             for (gmx::index th = 0; th < nbl_fep.ssize(); th++)
 597             {
 598                 try
 599                 {
 600                     gmx_nb_free_energy_kernel(nbl_fep[th].get(),
 601                                               x,
 602                                               forceWithShiftForces,
 603                                               fr,
 604                                               &mdatoms,
 605                                               &kernel_data,
 606                                               energygrp_elec,
 607                                               energygrp_vdw,
 608                                               nrnb);
 609                 }
 610                 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
 611             }
 612
 613             sum_epot(enerd->foreign_grpp, enerd->foreign_term);
 614             enerd->foreignLambdaTerms.accumulate(
 615                     i,
 616                     enerd->foreign_term[F_EPOT],
 617                     dvdl_nb[FreeEnergyPerturbationCouplingType::Vdw]
 618                             + dvdl_nb[FreeEnergyPerturbationCouplingType::Coul]);
 619         }
 620     }
 621     wallcycle_sub_stop(wcycle_, ewcsNONBONDED_FEP);
 622 }