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, 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/gmxassert.h"
  60 #include "gromacs/utility/real.h"
  61
  62 #include "kernel_common.h"
  63 #include "nbnxm_gpu.h"
  64 #include "nbnxm_simd.h"
  65 #include "pairlistset.h"
  66 #include "pairlistsets.h"
  67 #include "kernels_reference/kernel_gpu_ref.h"
  68 #define INCLUDE_KERNELFUNCTION_TABLES
  69 #include "kernels_reference/kernel_ref.h"
  70 #ifdef GMX_NBNXN_SIMD_2XNN
  71 #    include "kernels_simd_2xmm/kernels.h"
  72 #endif
  73 #ifdef GMX_NBNXN_SIMD_4XN
  74 #    include "kernels_simd_4xm/kernels.h"
  75 #endif
  76 #undef INCLUDE_FUNCTION_TABLES
  77
  78 /*! \brief Clears the energy group output buffers
  79  *
  80  * \param[in,out] out  nbnxn kernel output struct
  81  */
  82 static void clearGroupEnergies(nbnxn_atomdata_output_t* out)
  83 {
  84     std::fill(out->Vvdw.begin(), out->Vvdw.end(), 0.0_real);
  85     std::fill(out->Vc.begin(), out->Vc.end(), 0.0_real);
  86     std::fill(out->VSvdw.begin(), out->VSvdw.end(), 0.0_real);
  87     std::fill(out->VSc.begin(), out->VSc.end(), 0.0_real);
  88 }
  89
  90 /*! \brief Reduce the group-pair energy buffers produced by a SIMD kernel
  91  * to single terms in the output buffers.
  92  *
  93  * The SIMD kernels produce a large number of energy buffer in SIMD registers
  94  * to avoid scattered reads and writes.
  95  *
  96  * \tparam        unrollj         The unroll size for j-particles in the SIMD kernel
  97  * \param[in]     numGroups       The number of energy groups
  98  * \param[in]     numGroups_2log  Log2 of numGroups, rounded up
  99  * \param[in,out] out             Struct with energy buffers
 100  */
 101 template<int unrollj>
 102 static void reduceGroupEnergySimdBuffers(int numGroups, int numGroups_2log, nbnxn_atomdata_output_t* out)
 103 {
 104     const int unrollj_half = unrollj / 2;
 105     /* Energies are stored in SIMD registers with size 2^numGroups_2log */
 106     const int numGroupsStorage = (1 << numGroups_2log);
 107
 108     const real* gmx_restrict vVdwSimd     = out->VSvdw.data();
 109     const real* gmx_restrict vCoulombSimd = out->VSc.data();
 110     real* gmx_restrict vVdw               = out->Vvdw.data();
 111     real* gmx_restrict vCoulomb           = out->Vc.data();
 112
 113     /* The size of the SIMD energy group buffer array is:
 114      * numGroups*numGroups*numGroupsStorage*unrollj_half*simd_width
 115      */
 116     for (int i = 0; i < numGroups; i++)
 117     {
 118         for (int j1 = 0; j1 < numGroups; j1++)
 119         {
 120             for (int j0 = 0; j0 < numGroups; j0++)
 121             {
 122                 int c = ((i * numGroups + j1) * numGroupsStorage + j0) * unrollj_half * unrollj;
 123                 for (int s = 0; s < unrollj_half; s++)
 124                 {
 125                     vVdw[i * numGroups + j0] += vVdwSimd[c + 0];
 126                     vVdw[i * numGroups + j1] += vVdwSimd[c + 1];
 127                     vCoulomb[i * numGroups + j0] += vCoulombSimd[c + 0];
 128                     vCoulomb[i * numGroups + j1] += vCoulombSimd[c + 1];
 129                     c += unrollj + 2;
 130                 }
 131             }
 132         }
 133     }
 134 }
 135
 136 /*! \brief Dispatches the non-bonded N versus M atom cluster CPU kernels.
 137  *
 138  * OpenMP parallelization is performed within this function.
 139  * Energy reduction, but not force and shift force reduction, is performed
 140  * within this function.
 141  *
 142  * \param[in]     pairlistSet   Pairlists with local or non-local interactions to compute
 143  * \param[in]     kernelSetup   The non-bonded kernel setup
 144  * \param[in,out] nbat          The atomdata for the interactions
 145  * \param[in]     ic            Non-bonded interaction constants
 146  * \param[in]     shiftVectors  The PBC shift vectors
 147  * \param[in]     stepWork      Flags that tell what to compute
 148  * \param[in]     clearF        Enum that tells if to clear the force output buffer
 149  * \param[out]    vCoulomb      Output buffer for Coulomb energies
 150  * \param[out]    vVdw          Output buffer for Van der Waals energies
 151  * \param[in]     wcycle        Pointer to cycle counting data structure.
 152  */
 153 static void nbnxn_kernel_cpu(const PairlistSet&         pairlistSet,
 154                              const Nbnxm::KernelSetup&  kernelSetup,
 155                              nbnxn_atomdata_t*          nbat,
 156                              const interaction_const_t& ic,
 157                              rvec*                      shiftVectors,
 158                              const gmx::StepWorkload&   stepWork,
 159                              int                        clearF,
 160                              real*                      vCoulomb,
 161                              real*                      vVdw,
 162                              gmx_wallcycle*             wcycle)
 163 {
 164
 165     int coulkt;
 166     if (EEL_RF(ic.eeltype) || ic.eeltype == eelCUT)
 167     {
 168         coulkt = coulktRF;
 169     }
 170     else
 171     {
 172         if (kernelSetup.ewaldExclusionType == Nbnxm::EwaldExclusionType::Table)
 173         {
 174             if (ic.rcoulomb == ic.rvdw)
 175             {
 176                 coulkt = coulktTAB;
 177             }
 178             else
 179             {
 180                 coulkt = coulktTAB_TWIN;
 181             }
 182         }
 183         else
 184         {
 185             if (ic.rcoulomb == ic.rvdw)
 186             {
 187                 coulkt = coulktEWALD;
 188             }
 189             else
 190             {
 191                 coulkt = coulktEWALD_TWIN;
 192             }
 193         }
 194     }
 195
 196     const nbnxn_atomdata_t::Params& nbatParams = nbat->params();
 197
 198     int vdwkt = 0;
 199     if (ic.vdwtype == evdwCUT)
 200     {
 201         switch (ic.vdw_modifier)
 202         {
 203             case eintmodNONE:
 204             case eintmodPOTSHIFT:
 205                 switch (nbatParams.comb_rule)
 206                 {
 207                     case ljcrGEOM: vdwkt = vdwktLJCUT_COMBGEOM; break;
 208                     case ljcrLB: vdwkt = vdwktLJCUT_COMBLB; break;
 209                     case ljcrNONE: vdwkt = vdwktLJCUT_COMBNONE; break;
 210                     default: GMX_RELEASE_ASSERT(false, "Unknown combination rule");
 211                 }
 212                 break;
 213             case eintmodFORCESWITCH: vdwkt = vdwktLJFORCESWITCH; break;
 214             case eintmodPOTSWITCH: vdwkt = vdwktLJPOTSWITCH; break;
 215             default: GMX_RELEASE_ASSERT(false, "Unsupported VdW interaction modifier");
 216         }
 217     }
 218     else if (ic.vdwtype == evdwPME)
 219     {
 220         if (ic.ljpme_comb_rule == eljpmeGEOM)
 221         {
 222             vdwkt = vdwktLJEWALDCOMBGEOM;
 223         }
 224         else
 225         {
 226             vdwkt = vdwktLJEWALDCOMBLB;
 227             /* At setup we (should have) selected the C reference kernel */
 228             GMX_RELEASE_ASSERT(kernelSetup.kernelType == Nbnxm::KernelType::Cpu4x4_PlainC,
 229                                "Only the C reference nbnxn SIMD kernel supports LJ-PME with LB "
 230                                "combination rules");
 231         }
 232     }
 233     else
 234     {
 235         GMX_RELEASE_ASSERT(false, "Unsupported VdW interaction type");
 236     }
 237
 238     gmx::ArrayRef<const NbnxnPairlistCpu> pairlists = pairlistSet.cpuLists();
 239
 240     int gmx_unused nthreads = gmx_omp_nthreads_get(emntNonbonded);
 241     wallcycle_sub_start(wcycle, ewcsNONBONDED_CLEAR);
 242 #pragma omp parallel for schedule(static) num_threads(nthreads)
 243     for (gmx::index nb = 0; nb < pairlists.ssize(); nb++)
 244     {
 245         // Presently, the kernels do not call C++ code that can throw,
 246         // so no need for a try/catch pair in this OpenMP region.
 247         nbnxn_atomdata_output_t* out = &nbat->out[nb];
 248
 249         if (clearF == enbvClearFYes)
 250         {
 251             clearForceBuffer(nbat, nb);
 252
 253             clear_fshift(out->fshift.data());
 254         }
 255
 256         if (nb == 0)
 257         {
 258             wallcycle_sub_stop(wcycle, ewcsNONBONDED_CLEAR);
 259             wallcycle_sub_start(wcycle, ewcsNONBONDED_KERNEL);
 260         }
 261
 262         // TODO: Change to reference
 263         const NbnxnPairlistCpu* pairlist = &pairlists[nb];
 264
 265         if (!stepWork.computeEnergy)
 266         {
 267             /* Don't calculate energies */
 268             switch (kernelSetup.kernelType)
 269             {
 270                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 271                     nbnxn_kernel_noener_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 272                     break;
 273 #ifdef GMX_NBNXN_SIMD_2XNN
 274                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 275                     nbnxm_kernel_noener_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 276                     break;
 277 #endif
 278 #ifdef GMX_NBNXN_SIMD_4XN
 279                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 280                     nbnxm_kernel_noener_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 281                     break;
 282 #endif
 283                 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 284             }
 285         }
 286         else if (out->Vvdw.size() == 1)
 287         {
 288             /* A single energy group (pair) */
 289             out->Vvdw[0] = 0;
 290             out->Vc[0]   = 0;
 291
 292             switch (kernelSetup.kernelType)
 293             {
 294                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 295                     nbnxn_kernel_ener_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 296                     break;
 297 #ifdef GMX_NBNXN_SIMD_2XNN
 298                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 299                     nbnxm_kernel_ener_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 300                     break;
 301 #endif
 302 #ifdef GMX_NBNXN_SIMD_4XN
 303                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 304                     nbnxm_kernel_ener_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 305                     break;
 306 #endif
 307                 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 308             }
 309         }
 310         else
 311         {
 312             /* Calculate energy group contributions */
 313             clearGroupEnergies(out);
 314
 315             int unrollj = 0;
 316
 317             switch (kernelSetup.kernelType)
 318             {
 319                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 320                     unrollj = c_nbnxnCpuIClusterSize;
 321                     nbnxn_kernel_energrp_ref[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 322                     break;
 323 #ifdef GMX_NBNXN_SIMD_2XNN
 324                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 325                     unrollj = GMX_SIMD_REAL_WIDTH / 2;
 326                     nbnxm_kernel_energrp_simd_2xmm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 327                     break;
 328 #endif
 329 #ifdef GMX_NBNXN_SIMD_4XN
 330                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 331                     unrollj = GMX_SIMD_REAL_WIDTH;
 332                     nbnxm_kernel_energrp_simd_4xm[coulkt][vdwkt](pairlist, nbat, &ic, shiftVectors, out);
 333                     break;
 334 #endif
 335                 default: GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 336             }
 337
 338             if (kernelSetup.kernelType != Nbnxm::KernelType::Cpu4x4_PlainC)
 339             {
 340                 switch (unrollj)
 341                 {
 342                     case 2:
 343                         reduceGroupEnergySimdBuffers<2>(nbatParams.nenergrp, nbatParams.neg_2log, out);
 344                         break;
 345                     case 4:
 346                         reduceGroupEnergySimdBuffers<4>(nbatParams.nenergrp, nbatParams.neg_2log, out);
 347                         break;
 348                     case 8:
 349                         reduceGroupEnergySimdBuffers<8>(nbatParams.nenergrp, nbatParams.neg_2log, out);
 350                         break;
 351                     default: GMX_RELEASE_ASSERT(false, "Unsupported j-unroll size");
 352                 }
 353             }
 354         }
 355     }
 356     wallcycle_sub_stop(wcycle, ewcsNONBONDED_KERNEL);
 357
 358     if (stepWork.computeEnergy)
 359     {
 360         reduce_energies_over_lists(nbat, pairlists.ssize(), vVdw, vCoulomb);
 361     }
 362 }
 363
 364 static void accountFlops(t_nrnb*                    nrnb,
 365                          const PairlistSet&         pairlistSet,
 366                          const nonbonded_verlet_t&  nbv,
 367                          const interaction_const_t& ic,
 368                          const gmx::StepWorkload&   stepWork)
 369 {
 370     const bool usingGpuKernels = nbv.useGpu();
 371
 372     int enr_nbnxn_kernel_ljc;
 373     if (EEL_RF(ic.eeltype) || ic.eeltype == eelCUT)
 374     {
 375         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
 376     }
 377     else if ((!usingGpuKernels && nbv.kernelSetup().ewaldExclusionType == Nbnxm::EwaldExclusionType::Analytical)
 378              || (usingGpuKernels && Nbnxm::gpu_is_kernel_ewald_analytical(nbv.gpu_nbv)))
 379     {
 380         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
 381     }
 382     else
 383     {
 384         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
 385     }
 386     int enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
 387     if (stepWork.computeEnergy)
 388     {
 389         /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
 390         enr_nbnxn_kernel_ljc += 1;
 391         enr_nbnxn_kernel_lj += 1;
 392     }
 393
 394     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc, pairlistSet.natpair_ljq_);
 395     inc_nrnb(nrnb, enr_nbnxn_kernel_lj, pairlistSet.natpair_lj_);
 396     /* The Coulomb-only kernels are offset -eNR_NBNXN_LJ_RF+eNR_NBNXN_RF */
 397     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc - eNR_NBNXN_LJ_RF + eNR_NBNXN_RF, pairlistSet.natpair_q_);
 398
 399     if (ic.vdw_modifier == eintmodFORCESWITCH)
 400     {
 401         /* We add up the switch cost separately */
 402         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_FSW + (stepWork.computeEnergy ? 1 : 0),
 403                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 404     }
 405     if (ic.vdw_modifier == eintmodPOTSWITCH)
 406     {
 407         /* We add up the switch cost separately */
 408         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_PSW + (stepWork.computeEnergy ? 1 : 0),
 409                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 410     }
 411     if (ic.vdwtype == evdwPME)
 412     {
 413         /* We add up the LJ Ewald cost separately */
 414         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_EWALD + (stepWork.computeEnergy ? 1 : 0),
 415                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 416     }
 417 }
 418
 419 void nonbonded_verlet_t::dispatchNonbondedKernel(gmx::InteractionLocality   iLocality,
 420                                                  const interaction_const_t& ic,
 421                                                  const gmx::StepWorkload&   stepWork,
 422                                                  int                        clearF,
 423                                                  const t_forcerec&          fr,
 424                                                  gmx_enerdata_t*            enerd,
 425                                                  t_nrnb*                    nrnb)
 426 {
 427     const PairlistSet& pairlistSet = pairlistSets().pairlistSet(iLocality);
 428
 429     switch (kernelSetup().kernelType)
 430     {
 431         case Nbnxm::KernelType::Cpu4x4_PlainC:
 432         case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 433         case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 434             nbnxn_kernel_cpu(pairlistSet, kernelSetup(), nbat.get(), ic, fr.shift_vec, stepWork,
 435                              clearF, enerd->grpp.ener[egCOULSR].data(),
 436                              fr.bBHAM ? enerd->grpp.ener[egBHAMSR].data() : enerd->grpp.ener[egLJSR].data(),
 437                              wcycle_);
 438             break;
 439
 440         case Nbnxm::KernelType::Gpu8x8x8:
 441             Nbnxm::gpu_launch_kernel(gpu_nbv, stepWork, iLocality);
 442             break;
 443
 444         case Nbnxm::KernelType::Cpu8x8x8_PlainC:
 445             nbnxn_kernel_gpu_ref(
 446                     pairlistSet.gpuList(), nbat.get(), &ic, fr.shift_vec, stepWork, clearF,
 447                     nbat->out[0].f, nbat->out[0].fshift.data(), enerd->grpp.ener[egCOULSR].data(),
 448                     fr.bBHAM ? enerd->grpp.ener[egBHAMSR].data() : enerd->grpp.ener[egLJSR].data());
 449             break;
 450
 451         default: GMX_RELEASE_ASSERT(false, "Invalid nonbonded kernel type passed!");
 452     }
 453
 454     accountFlops(nrnb, pairlistSet, *this, ic, stepWork);
 455 }
 456
 457 void nonbonded_verlet_t::dispatchFreeEnergyKernel(gmx::InteractionLocality   iLocality,
 458                                                   const t_forcerec*          fr,
 459                                                   rvec                       x[],
 460                                                   gmx::ForceWithShiftForces* forceWithShiftForces,
 461                                                   const t_mdatoms&           mdatoms,
 462                                                   t_lambda*                  fepvals,
 463                                                   real*                      lambda,
 464                                                   gmx_enerdata_t*            enerd,
 465                                                   const gmx::StepWorkload&   stepWork,
 466                                                   t_nrnb*                    nrnb)
 467 {
 468     const auto nbl_fep = pairlistSets().pairlistSet(iLocality).fepLists();
 469
 470     /* When the first list is empty, all are empty and there is nothing to do */
 471     if (!pairlistSets().params().haveFep || nbl_fep[0]->nrj == 0)
 472     {
 473         return;
 474     }
 475
 476     int donb_flags = 0;
 477     /* Add short-range interactions */
 478     donb_flags |= GMX_NONBONDED_DO_SR;
 479
 480     if (stepWork.computeForces)
 481     {
 482         donb_flags |= GMX_NONBONDED_DO_FORCE;
 483     }
 484     if (stepWork.computeVirial)
 485     {
 486         donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 487     }
 488     if (stepWork.computeEnergy)
 489     {
 490         donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 491     }
 492
 493     nb_kernel_data_t kernel_data;
 494     real             dvdl_nb[efptNR] = { 0 };
 495     kernel_data.flags                = donb_flags;
 496     kernel_data.lambda               = lambda;
 497     kernel_data.dvdl                 = dvdl_nb;
 498
 499     kernel_data.energygrp_elec = enerd->grpp.ener[egCOULSR].data();
 500     kernel_data.energygrp_vdw  = enerd->grpp.ener[egLJSR].data();
 501
 502     GMX_ASSERT(gmx_omp_nthreads_get(emntNonbonded) == nbl_fep.ssize(),
 503                "Number of lists should be same as number of NB threads");
 504
 505     wallcycle_sub_start(wcycle_, ewcsNONBONDED_FEP);
 506 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
 507     for (gmx::index th = 0; th < nbl_fep.ssize(); th++)
 508     {
 509         try
 510         {
 511             gmx_nb_free_energy_kernel(nbl_fep[th].get(), x, forceWithShiftForces, fr, &mdatoms,
 512                                       &kernel_data, nrnb);
 513         }
 514         GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
 515     }
 516
 517     if (fepvals->sc_alpha != 0)
 518     {
 519         enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
 520         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 521     }
 522     else
 523     {
 524         enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
 525         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 526     }
 527
 528     /* If we do foreign lambda and we have soft-core interactions
 529      * we have to recalculate the (non-linear) energies contributions.
 530      */
 531     if (fepvals->n_lambda > 0 && stepWork.computeDhdl && fepvals->sc_alpha != 0)
 532     {
 533         real lam_i[efptNR];
 534         kernel_data.flags = (donb_flags & ~(GMX_NONBONDED_DO_FORCE | GMX_NONBONDED_DO_SHIFTFORCE))
 535                             | GMX_NONBONDED_DO_FOREIGNLAMBDA;
 536         kernel_data.lambda         = lam_i;
 537         kernel_data.dvdl           = dvdl_nb;
 538         kernel_data.energygrp_elec = enerd->foreign_grpp.ener[egCOULSR].data();
 539         kernel_data.energygrp_vdw  = enerd->foreign_grpp.ener[egLJSR].data();
 540
 541         for (size_t i = 0; i < enerd->enerpart_lambda.size(); i++)
 542         {
 543             std::fill(std::begin(dvdl_nb), std::end(dvdl_nb), 0);
 544             for (int j = 0; j < efptNR; j++)
 545             {
 546                 lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i - 1]);
 547             }
 548             reset_foreign_enerdata(enerd);
 549 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
 550             for (gmx::index th = 0; th < nbl_fep.ssize(); th++)
 551             {
 552                 try
 553                 {
 554                     gmx_nb_free_energy_kernel(nbl_fep[th].get(), x, forceWithShiftForces, fr,
 555                                               &mdatoms, &kernel_data, nrnb);
 556                 }
 557                 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR
 558             }
 559
 560             sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 561             enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 562             enerd->dhdlLambda[i] += dvdl_nb[efptVDW] + dvdl_nb[efptCOUL];
 563         }
 564     }
 565     else
 566     {
 567         for (size_t i = 0; i < enerd->enerpart_lambda.size(); i++)
 568         {
 569             enerd->dhdlLambda[i] += dvdl_nb[efptVDW] + dvdl_nb[efptCOUL];
 570         }
 571     }
 572     wallcycle_sub_stop(wcycle_, ewcsNONBONDED_FEP);
 573 }