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/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"
  63
  64 #include "kernel_common.h"
  65 #include "nbnxm_gpu.h"
  66 #include "nbnxm_gpu_data_mgmt.h"
  67 #include "nbnxm_simd.h"
  68 #include "pairlistset.h"
  69 #include "pairlistsets.h"
  70 #include "kernels_reference/kernel_gpu_ref.h"
  71 #define INCLUDE_KERNELFUNCTION_TABLES
  72 #include "kernels_reference/kernel_ref.h"
  73 #ifdef GMX_NBNXN_SIMD_2XNN
  74 #    include "kernels_simd_2xmm/kernels.h"
  75 #endif
  76 #ifdef GMX_NBNXN_SIMD_4XN
  77 #    include "kernels_simd_4xm/kernels.h"
  78 #endif
  79 #undef INCLUDE_FUNCTION_TABLES
  80
  81 /*! \brief Clears the energy group output buffers
  82  *
  83  * \param[in,out] out  nbnxn kernel output struct
  84  */
  85 static void clearGroupEnergies(nbnxn_atomdata_output_t* out)
  86 {
  87     std::fill(out->Vvdw.begin(), out->Vvdw.end(), 0.0_real);
  88     std::fill(out->Vc.begin(), out->Vc.end(), 0.0_real);
  89     std::fill(out->VSvdw.begin(), out->VSvdw.end(), 0.0_real);
  90     std::fill(out->VSc.begin(), out->VSc.end(), 0.0_real);
  91 }
  92
  93 /*! \brief Reduce the group-pair energy buffers produced by a SIMD kernel
  94  * to single terms in the output buffers.
  95  *
  96  * The SIMD kernels produce a large number of energy buffer in SIMD registers
  97  * to avoid scattered reads and writes.
  98  *
  99  * \tparam        unrollj         The unroll size for j-particles in the SIMD kernel
 100  * \param[in]     numGroups       The number of energy groups
 101  * \param[in]     numGroups_2log  Log2 of numGroups, rounded up
 102  * \param[in,out] out             Struct with energy buffers
 103  */
 104 template<int unrollj>
 105 static void reduceGroupEnergySimdBuffers(int numGroups, int numGroups_2log, nbnxn_atomdata_output_t* out)
 106 {
 107     const int unrollj_half = unrollj / 2;
 108     /* Energies are stored in SIMD registers with size 2^numGroups_2log */
 109     const int numGroupsStorage = (1 << numGroups_2log);
 110
 111     const real* gmx_restrict vVdwSimd     = out->VSvdw.data();
 112     const real* gmx_restrict vCoulombSimd = out->VSc.data();
 113     real* gmx_restrict vVdw               = out->Vvdw.data();
 114     real* gmx_restrict vCoulomb           = out->Vc.data();
 115
 116     /* The size of the SIMD energy group buffer array is:
 117      * numGroups*numGroups*numGroupsStorage*unrollj_half*simd_width
 118      */
 119     for (int i = 0; i < numGroups; i++)
 120     {
 121         for (int j1 = 0; j1 < numGroups; j1++)
 122         {
 123             for (int j0 = 0; j0 < numGroups; j0++)
 124             {
 125                 int c = ((i * numGroups + j1) * numGroupsStorage + j0) * unrollj_half * unrollj;
 126                 for (int s = 0; s < unrollj_half; s++)
 127                 {
 128                     vVdw[i * numGroups + j0] += vVdwSimd[c + 0];
 129                     vVdw[i * numGroups + j1] += vVdwSimd[c + 1];
 130                     vCoulomb[i * numGroups + j0] += vCoulombSimd[c + 0];
 131                     vCoulomb[i * numGroups + j1] += vCoulombSimd[c + 1];
 132                     c += unrollj + 2;
 133                 }
 134             }
 135         }
 136     }
 137 }
 138
 139 static int getCoulombKernelType(const Nbnxm::KernelSetup& kernelSetup, const interaction_const_t& ic)
 140 {
 141
 142     if (EEL_RF(ic.eeltype) || ic.eeltype == CoulombInteractionType::Cut)
 143     {
 144         return coulktRF;
 145     }
 146     else
 147     {
 148         if (kernelSetup.ewaldExclusionType == Nbnxm::EwaldExclusionType::Table)
 149         {
 150             if (ic.rcoulomb == ic.rvdw)
 151             {
 152                 return coulktTAB;
 153             }
 154             else
 155             {
 156                 return coulktTAB_TWIN;
 157             }
 158         }
 159         else
 160         {
 161             if (ic.rcoulomb == ic.rvdw)
 162             {
 163                 return coulktEWALD;
 164             }
 165             else
 166             {
 167                 return coulktEWALD_TWIN;
 168             }
 169         }
 170     }
 171 }
 172
 173 static int getVdwKernelType(const Nbnxm::KernelSetup&       kernelSetup,
 174                             const nbnxn_atomdata_t::Params& nbatParams,
 175                             const interaction_const_t&      ic)
 176 {
 177     if (ic.vdwtype == VanDerWaalsType::Cut)
 178     {
 179         switch (ic.vdw_modifier)
 180         {
 181             case InteractionModifiers::None:
 182             case InteractionModifiers::PotShift:
 183                 switch (nbatParams.ljCombinationRule)
 184                 {
 185                     case LJCombinationRule::Geometric: return vdwktLJCUT_COMBGEOM;
 186                     case LJCombinationRule::LorentzBerthelot: return vdwktLJCUT_COMBLB;
 187                     case LJCombinationRule::None: return vdwktLJCUT_COMBNONE;
 188                     default: gmx_incons("Unknown combination rule");
 189                 }
 190             case InteractionModifiers::ForceSwitch: return vdwktLJFORCESWITCH;
 191             case InteractionModifiers::PotSwitch: return vdwktLJPOTSWITCH;
 192             default:
 193                 std::string errorMsg =
 194                         gmx::formatString("Unsupported VdW interaction modifier %s (%d)",
 195                                           enumValueToString(ic.vdw_modifier),
 196                                           static_cast<int>(ic.vdw_modifier));
 197                 gmx_incons(errorMsg);
 198         }
 199     }
 200     else if (ic.vdwtype == VanDerWaalsType::Pme)
 201     {
 202         if (ic.ljpme_comb_rule == LongRangeVdW::Geom)
 203         {
 204             return vdwktLJEWALDCOMBGEOM;
 205         }
 206         else
 207         {
 208             /* At setup we (should have) selected the C reference kernel */
 209             GMX_RELEASE_ASSERT(kernelSetup.kernelType == Nbnxm::KernelType::Cpu4x4_PlainC,
 210                                "Only the C reference nbnxn SIMD kernel supports LJ-PME with LB "
 211                                "combination rules");
 212             return vdwktLJEWALDCOMBLB;
 213         }
 214     }
 215     else
 216     {
 217         std::string errorMsg = gmx::formatString("Unsupported VdW interaction type %s (%d)",
 218                                                  enumValueToString(ic.vdwtype),
 219                                                  static_cast<int>(ic.vdwtype));
 220         gmx_incons(errorMsg);
 221     }
 222 }
 223
 224 /*! \brief Dispatches the non-bonded N versus M atom cluster CPU kernels.
 225  *
 226  * OpenMP parallelization is performed within this function.
 227  * Energy reduction, but not force and shift force reduction, is performed
 228  * within this function.
 229  *
 230  * \param[in]     pairlistSet   Pairlists with local or non-local interactions to compute
 231  * \param[in]     kernelSetup   The non-bonded kernel setup
 232  * \param[in,out] nbat          The atomdata for the interactions
 233  * \param[in]     ic            Non-bonded interaction constants
 234  * \param[in]     shiftVectors  The PBC shift vectors
 235  * \param[in]     stepWork      Flags that tell what to compute
 236  * \param[in]     clearF        Enum that tells if to clear the force output buffer
 237  * \param[out]    vCoulomb      Output buffer for Coulomb energies
 238  * \param[out]    vVdw          Output buffer for Van der Waals energies
 239  * \param[in]     wcycle        Pointer to cycle counting data structure.
 240  */
 241 static void nbnxn_kernel_cpu(const PairlistSet&         pairlistSet,
 242                              const Nbnxm::KernelSetup&  kernelSetup,
 243                              nbnxn_atomdata_t*          nbat,
 244                              const interaction_const_t& ic,
 245                              rvec*                      shiftVectors,
 246                              const gmx::StepWorkload&   stepWork,
 247                              int                        clearF,
 248                              real*                      vCoulomb,
 249                              real*                      vVdw,
 250                              gmx_wallcycle*             wcycle)
 251 {
 252
 253     const nbnxn_atomdata_t::Params& nbatParams = nbat->params();
 254
 255     const int coulkt = getCoulombKernelType(kernelSetup, ic);
 256     const int vdwkt  = getVdwKernelType(kernelSetup, nbatParams, ic);
 257
 258     gmx::ArrayRef<const NbnxnPairlistCpu> pairlists = pairlistSet.cpuLists();
 259
 260     int gmx_unused nthreads = gmx_omp_nthreads_get(emntNonbonded);
 261     wallcycle_sub_start(wcycle, ewcsNONBONDED_CLEAR);
 262 #pragma omp parallel for schedule(static) num_threads(nthreads)
 263     for (gmx::index nb = 0; nb < pairlists.ssize(); nb++)
 264     {
 265         // Presently, the kernels do not call C++ code that can throw,
 266         // so no need for a try/catch pair in this OpenMP region.
 267         nbnxn_atomdata_output_t* out = &nbat->out[nb];
 268
 269         if (clearF == enbvClearFYes)
 270         {
 271             clearForceBuffer(nbat, nb);
 272
 273             clear_fshift(out->fshift.data());
 274         }
 275
 276         if (nb == 0)
 277         {
 278             wallcycle_sub_stop(wcycle, ewcsNONBONDED_CLEAR);
 279             wallcycle_sub_start(wcycle, ewcsNONBONDED_KERNEL);
 280         }
 281
 282         // TODO: Change to reference
 283         const NbnxnPairlistCpu* pairlist = &pairlists[nb];
 284
 285         if (!stepWork.computeEnergy)
 286         {
 287             /* Don't calculate energies */
 288             switch (kernelSetup.kernelType)
 289             {
 290                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 291                     nbnxn_kernel_noener_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 292                     break;
 293 #ifdef GMX_NBNXN_SIMD_2XNN
 294                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 295                     nbnxm_kernel_noener_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 296                     break;
 297 #endif
 298 #ifdef GMX_NBNXN_SIMD_4XN
 299                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 300                     nbnxm_kernel_noener_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 301                     break;
 302 #endif
 303                 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 304             }
 305         }
 306         else if (out->Vvdw.size() == 1)
 307         {
 308             /* A single energy group (pair) */
 309             out->Vvdw[0] = 0;
 310             out->Vc[0]   = 0;
 311
 312             switch (kernelSetup.kernelType)
 313             {
 314                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 315                     nbnxn_kernel_ener_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 316                     break;
 317 #ifdef GMX_NBNXN_SIMD_2XNN
 318                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 319                     nbnxm_kernel_ener_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 320                     break;
 321 #endif
 322 #ifdef GMX_NBNXN_SIMD_4XN
 323                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 324                     nbnxm_kernel_ener_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 325                     break;
 326 #endif
 327                 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 328             }
 329         }
 330         else
 331         {
 332             /* Calculate energy group contributions */
 333             clearGroupEnergies(out);
 334
 335             int unrollj = 0;
 336
 337             switch (kernelSetup.kernelType)
 338             {
 339                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 340                     unrollj = c_nbnxnCpuIClusterSize;
 341                     nbnxn_kernel_energrp_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 342                     break;
 343 #ifdef GMX_NBNXN_SIMD_2XNN
 344                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 345                     unrollj = GMX_SIMD_REAL_WIDTH / 2;
 346                     nbnxm_kernel_energrp_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 347                     break;
 348 #endif
 349 #ifdef GMX_NBNXN_SIMD_4XN
 350                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 351                     unrollj = GMX_SIMD_REAL_WIDTH;
 352                     nbnxm_kernel_energrp_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 353                     break;
 354 #endif
 355                 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 356             }
 357
 358             if (kernelSetup.kernelType != Nbnxm::KernelType::Cpu4x4_PlainC)
 359             {
 360                 switch (unrollj)
 361                 {
 362                     case 2:
 363                         reduceGroupEnergySimdBuffers<2>(nbatParams.nenergrp, nbatParams.neg_2log, out);
 364                         break;
 365                     case 4:
 366                         reduceGroupEnergySimdBuffers<4>(nbatParams.nenergrp, nbatParams.neg_2log, out);
 367                         break;
 368                     case 8:
 369                         reduceGroupEnergySimdBuffers<8>(nbatParams.nenergrp, nbatParams.neg_2log, out);
 370                         break;
 371                     default: GMX_RELEASE_ASSERT(false, "Unsupported j-unroll size");
 372                 }
 373             }
 374         }
 375     }
 376     wallcycle_sub_stop(wcycle, ewcsNONBONDED_KERNEL);
 377
 378     if (stepWork.computeEnergy)
 379     {
 380         reduce_energies_over_lists(nbat, pairlists.ssize(), vVdw, vCoulomb);
 381     }
 382 }
 383
 384 static void accountFlops(t_nrnb*                    nrnb,
 385                          const PairlistSet&         pairlistSet,
 386                          const nonbonded_verlet_t&  nbv,
 387                          const interaction_const_t& ic,
 388                          const gmx::StepWorkload&   stepWork)
 389 {
 390     const bool usingGpuKernels = nbv.useGpu();
 391
 392     int enr_nbnxn_kernel_ljc = eNRNB;
 393     if (EEL_RF(ic.eeltype) || ic.eeltype == CoulombInteractionType::Cut)
 394     {
 395         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
 396     }
 397     else if ((!usingGpuKernels && nbv.kernelSetup().ewaldExclusionType == Nbnxm::EwaldExclusionType::Analytical)
 398              || (usingGpuKernels && Nbnxm::gpu_is_kernel_ewald_analytical(nbv.gpu_nbv)))
 399     {
 400         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
 401     }
 402     else
 403     {
 404         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
 405     }
 406     int enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
 407     if (stepWork.computeEnergy)
 408     {
 409         /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
 410         enr_nbnxn_kernel_ljc += 1;
 411         enr_nbnxn_kernel_lj += 1;
 412     }
 413
 414     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc, pairlistSet.natpair_ljq_);
 415     inc_nrnb(nrnb, enr_nbnxn_kernel_lj, pairlistSet.natpair_lj_);
 416     /* The Coulomb-only kernels are offset -eNR_NBNXN_LJ_RF+eNR_NBNXN_RF */
 417     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc - eNR_NBNXN_LJ_RF + eNR_NBNXN_RF, pairlistSet.natpair_q_);
 418
 419     if (ic.vdw_modifier == InteractionModifiers::ForceSwitch)
 420     {
 421         /* We add up the switch cost separately */
 422         inc_nrnb(nrnb,
 423                  eNR_NBNXN_ADD_LJ_FSW + (stepWork.computeEnergy ? 1 : 0),
 424                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 425     }
 426     if (ic.vdw_modifier == InteractionModifiers::PotSwitch)
 427     {
 428         /* We add up the switch cost separately */
 429         inc_nrnb(nrnb,
 430                  eNR_NBNXN_ADD_LJ_PSW + (stepWork.computeEnergy ? 1 : 0),
 431                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 432     }
 433     if (ic.vdwtype == VanDerWaalsType::Pme)
 434     {
 435         /* We add up the LJ Ewald cost separately */
 436         inc_nrnb(nrnb,
 437                  eNR_NBNXN_ADD_LJ_EWALD + (stepWork.computeEnergy ? 1 : 0),
 438                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 439     }
 440 }
 441
 442 void nonbonded_verlet_t::dispatchNonbondedKernel(gmx::InteractionLocality   iLocality,
 443                                                  const interaction_const_t& ic,
 444                                                  const gmx::StepWorkload&   stepWork,
 445                                                  int                        clearF,
 446                                                  const t_forcerec&          fr,
 447                                                  gmx_enerdata_t*            enerd,
 448                                                  t_nrnb*                    nrnb)
 449 {
 450     const PairlistSet& pairlistSet = pairlistSets().pairlistSet(iLocality);
 451
 452     switch (kernelSetup().kernelType)
 453     {
 454         case Nbnxm::KernelType::Cpu4x4_PlainC:
 455         case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 456         case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 457             nbnxn_kernel_cpu(
 458                     pairlistSet,
 459                     kernelSetup(),
 460                     nbat.get(),
 461                     ic,
 462                     fr.shift_vec,
 463                     stepWork,
 464                     clearF,
 465                     enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR].data(),
 466                     fr.bBHAM ? enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::BuckinghamSR].data()
 467                              : enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR].data(),
 468                     wcycle_);
 469             break;
 470
 471         case Nbnxm::KernelType::Gpu8x8x8:
 472             Nbnxm::gpu_launch_kernel(gpu_nbv, stepWork, iLocality);
 473             break;
 474
 475         case Nbnxm::KernelType::Cpu8x8x8_PlainC:
 476             nbnxn_kernel_gpu_ref(
 477                     pairlistSet.gpuList(),
 478                     nbat.get(),
 479                     &ic,
 480                     fr.shift_vec,
 481                     stepWork,
 482                     clearF,
 483                     nbat->out[0].f,
 484                     nbat->out[0].fshift.data(),
 485                     enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR].data(),
 486                     fr.bBHAM ? enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::BuckinghamSR].data()
 487                              : enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR].data());
 488             break;
 489
 490         default: GMX_RELEASE_ASSERT(false, "Invalid nonbonded kernel type passed!");
 491     }
 492
 493     accountFlops(nrnb, pairlistSet, *this, ic, stepWork);
 494 }
 495
 496 void nonbonded_verlet_t::dispatchFreeEnergyKernel(gmx::InteractionLocality   iLocality,
 497                                                   const t_forcerec*          fr,
 498                                                   rvec                       x[],
 499                                                   gmx::ForceWithShiftForces* forceWithShiftForces,
 500                                                   const t_mdatoms&           mdatoms,
 501                                                   t_lambda*                  fepvals,
 502                                                   gmx::ArrayRef<const real>  lambda,
 503                                                   gmx_enerdata_t*            enerd,
 504                                                   const gmx::StepWorkload&   stepWork,
 505                                                   t_nrnb*                    nrnb)
 506 {
 507     const auto nbl_fep = pairlistSets().pairlistSet(iLocality).fepLists();
 508
 509     /* When the first list is empty, all are empty and there is nothing to do */
 510     if (!pairlistSets().params().haveFep || nbl_fep[0]->nrj == 0)
 511     {
 512         return;
 513     }
 514
 515     int donb_flags = 0;
 516     /* Add short-range interactions */
 517     donb_flags |= GMX_NONBONDED_DO_SR;
 518
 519     if (stepWork.computeForces)
 520     {
 521         donb_flags |= GMX_NONBONDED_DO_FORCE;
 522     }
 523     if (stepWork.computeVirial)
 524     {
 525         donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 526     }
 527     if (stepWork.computeEnergy)
 528     {
 529         donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 530     }
 531
 532     nb_kernel_data_t                                                kernel_data;
 533     gmx::EnumerationArray<FreeEnergyPerturbationCouplingType, real> dvdl_nb = { 0 };
 534     kernel_data.flags                                                       = donb_flags;
 535     kernel_data.lambda                                                      = lambda;
 536     kernel_data.dvdl                                                        = dvdl_nb;
 537
 538     kernel_data.energygrp_elec = enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR].data();
 539     kernel_data.energygrp_vdw = enerd->grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR].data();
 540
 541     GMX_ASSERT(gmx_omp_nthreads_get(emntNonbonded) == nbl_fep.ssize(),
 542                "Number of lists should be same as number of NB threads");
 543
 544     wallcycle_sub_start(wcycle_, ewcsNONBONDED_FEP);
 545 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
 546     for (gmx::index th = 0; th < nbl_fep.ssize(); th++)
 547     {
 548         try
 549         {
 550             gmx_nb_free_energy_kernel(
 551                     nbl_fep[th].get(), x, forceWithShiftForces, fr, &mdatoms, &kernel_data, nrnb);
 552         }
 553         GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
 554     }
 555
 556     if (fepvals->sc_alpha != 0)
 557     {
 558         enerd->dvdl_nonlin[FreeEnergyPerturbationCouplingType::Vdw] +=
 559                 dvdl_nb[FreeEnergyPerturbationCouplingType::Vdw];
 560         enerd->dvdl_nonlin[FreeEnergyPerturbationCouplingType::Coul] +=
 561                 dvdl_nb[FreeEnergyPerturbationCouplingType::Coul];
 562     }
 563     else
 564     {
 565         enerd->dvdl_lin[FreeEnergyPerturbationCouplingType::Vdw] +=
 566                 dvdl_nb[FreeEnergyPerturbationCouplingType::Vdw];
 567         enerd->dvdl_lin[FreeEnergyPerturbationCouplingType::Coul] +=
 568                 dvdl_nb[FreeEnergyPerturbationCouplingType::Coul];
 569     }
 570
 571     /* If we do foreign lambda and we have soft-core interactions
 572      * we have to recalculate the (non-linear) energies contributions.
 573      */
 574     if (fepvals->n_lambda > 0 && stepWork.computeDhdl && fepvals->sc_alpha != 0)
 575     {
 576         gmx::EnumerationArray<FreeEnergyPerturbationCouplingType, real> lam_i;
 577         kernel_data.flags = (donb_flags & ~(GMX_NONBONDED_DO_FORCE | GMX_NONBONDED_DO_SHIFTFORCE))
 578                             | GMX_NONBONDED_DO_FOREIGNLAMBDA;
 579         kernel_data.lambda = lam_i;
 580         kernel_data.dvdl   = dvdl_nb;
 581         kernel_data.energygrp_elec =
 582                 enerd->foreign_grpp.energyGroupPairTerms[NonBondedEnergyTerms::CoulombSR].data();
 583         kernel_data.energygrp_vdw =
 584                 enerd->foreign_grpp.energyGroupPairTerms[NonBondedEnergyTerms::LJSR].data();
 585
 586         for (gmx::index i = 0; i < 1 + enerd->foreignLambdaTerms.numLambdas(); i++)
 587         {
 588             std::fill(std::begin(dvdl_nb), std::end(dvdl_nb), 0);
 589             for (int j = 0; j < static_cast<int>(FreeEnergyPerturbationCouplingType::Count); j++)
 590             {
 591                 lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i - 1]);
 592             }
 593             reset_foreign_enerdata(enerd);
 594 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
 595             for (gmx::index th = 0; th < nbl_fep.ssize(); th++)
 596             {
 597                 try
 598                 {
 599                     gmx_nb_free_energy_kernel(
 600                             nbl_fep[th].get(), x, forceWithShiftForces, fr, &mdatoms, &kernel_data, nrnb);
 601                 }
 602                 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
 603             }
 604
 605             sum_epot(enerd->foreign_grpp, enerd->foreign_term);
 606             enerd->foreignLambdaTerms.accumulate(
 607                     i,
 608                     enerd->foreign_term[F_EPOT],
 609                     dvdl_nb[FreeEnergyPerturbationCouplingType::Vdw]
 610                             + dvdl_nb[FreeEnergyPerturbationCouplingType::Coul]);
 611         }
 612     }
 613     wallcycle_sub_stop(wcycle_, ewcsNONBONDED_FEP);
 614 }