src/gromacs/nbnxm/kerneldispatch.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2016,2017,2018,2019, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35
  36 #include "gmxpre.h"
  37
  38 #include "gromacs/gmxlib/nrnb.h"
  39 #include "gromacs/gmxlib/nonbonded/nb_free_energy.h"
  40 #include "gromacs/gmxlib/nonbonded/nb_kernel.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/force_flags.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/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/nbnxm/gpu_data_mgmt.h"
  54 #include "gromacs/nbnxm/nbnxm.h"
  55 #include "gromacs/nbnxm/nbnxm_simd.h"
  56 #include "gromacs/nbnxm/kernels_reference/kernel_gpu_ref.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 "pairlistset.h"
  64 #include "pairlistsets.h"
  65 #define INCLUDE_KERNELFUNCTION_TABLES
  66 #include "gromacs/nbnxm/kernels_reference/kernel_ref.h"
  67 #ifdef GMX_NBNXN_SIMD_2XNN
  68 #include "gromacs/nbnxm/kernels_simd_2xmm/kernels.h"
  69 #endif
  70 #ifdef GMX_NBNXN_SIMD_4XN
  71 #include "gromacs/nbnxm/kernels_simd_4xm/kernels.h"
  72 #endif
  73 #undef INCLUDE_FUNCTION_TABLES
  74
  75 /*! \brief Clears the energy group output buffers
  76  *
  77  * \param[in,out] out  nbnxn kernel output struct
  78  */
  79 static void clearGroupEnergies(nbnxn_atomdata_output_t *out)
  80 {
  81     std::fill(out->Vvdw.begin(), out->Vvdw.end(), 0.0_real);
  82     std::fill(out->Vc.begin(), out->Vc.end(), 0.0_real);
  83     std::fill(out->VSvdw.begin(), out->VSvdw.end(), 0.0_real);
  84     std::fill(out->VSc.begin(), out->VSc.end(), 0.0_real);
  85 }
  86
  87 /*! \brief Reduce the group-pair energy buffers produced by a SIMD kernel
  88  * to single terms in the output buffers.
  89  *
  90  * The SIMD kernels produce a large number of energy buffer in SIMD registers
  91  * to avoid scattered reads and writes.
  92  *
  93  * \tparam        unrollj         The unroll size for j-particles in the SIMD kernel
  94  * \param[in]     numGroups       The number of energy groups
  95  * \param[in]     numGroups_2log  Log2 of numGroups, rounded up
  96  * \param[in,out] out             Struct with energy buffers
  97  */
  98 template <int unrollj> static void
  99 reduceGroupEnergySimdBuffers(int                       numGroups,
 100                              int                       numGroups_2log,
 101                              nbnxn_atomdata_output_t  *out)
 102 {
 103     const int                 unrollj_half     = unrollj/2;
 104     /* Energies are stored in SIMD registers with size 2^numGroups_2log */
 105     const int                 numGroupsStorage = (1 << numGroups_2log);
 106
 107     const real * gmx_restrict vVdwSimd     = out->VSvdw.data();
 108     const real * gmx_restrict vCoulombSimd = out->VSc.data();
 109     real * gmx_restrict       vVdw         = out->Vvdw.data();
 110     real * gmx_restrict       vCoulomb     = out->Vc.data();
 111
 112     /* The size of the SIMD energy group buffer array is:
 113      * numGroups*numGroups*numGroupsStorage*unrollj_half*simd_width
 114      */
 115     for (int i = 0; i < numGroups; i++)
 116     {
 117         for (int j1 = 0; j1 < numGroups; j1++)
 118         {
 119             for (int j0 = 0; j0 < numGroups; j0++)
 120             {
 121                 int c = ((i*numGroups + j1)*numGroupsStorage + j0)*unrollj_half*unrollj;
 122                 for (int s = 0; s < unrollj_half; s++)
 123                 {
 124                     vVdw    [i*numGroups + j0] += vVdwSimd    [c + 0];
 125                     vVdw    [i*numGroups + j1] += vVdwSimd    [c + 1];
 126                     vCoulomb[i*numGroups + j0] += vCoulombSimd[c + 0];
 127                     vCoulomb[i*numGroups + j1] += vCoulombSimd[c + 1];
 128                     c                          += unrollj + 2;
 129                 }
 130             }
 131         }
 132     }
 133 }
 134
 135 /*! \brief Dispatches the non-bonded N versus M atom cluster CPU kernels.
 136  *
 137  * OpenMP parallelization is performed within this function.
 138  * Energy reduction, but not force and shift force reduction, is performed
 139  * within this function.
 140  *
 141  * \param[in]     pairlistSet   Pairlists with local or non-local interactions to compute
 142  * \param[in]     kernelSetup   The non-bonded kernel setup
 143  * \param[in,out] nbat          The atomdata for the interactions
 144  * \param[in]     ic            Non-bonded interaction constants
 145  * \param[in]     shiftVectors  The PBC shift vectors
 146  * \param[in]     forceFlags    Flags that tell what to compute
 147  * \param[in]     clearF        Enum that tells if to clear the force output buffer
 148  * \param[out]    vCoulomb      Output buffer for Coulomb energies
 149  * \param[out]    vVdw          Output buffer for Van der Waals energies
 150  * \param[in]     wcycle        Pointer to cycle counting data structure.
 151  */
 152 static void
 153 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                  int                             forceFlags,
 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:
 211                         GMX_RELEASE_ASSERT(false, "Unknown combination rule");
 212                 }
 213                 break;
 214             case eintmodFORCESWITCH:
 215                 vdwkt = vdwktLJFORCESWITCH;
 216                 break;
 217             case eintmodPOTSWITCH:
 218                 vdwkt = vdwktLJPOTSWITCH;
 219                 break;
 220             default:
 221                 GMX_RELEASE_ASSERT(false, "Unsupported VdW interaction modifier");
 222         }
 223     }
 224     else if (ic.vdwtype == evdwPME)
 225     {
 226         if (ic.ljpme_comb_rule == eljpmeGEOM)
 227         {
 228             vdwkt = vdwktLJEWALDCOMBGEOM;
 229         }
 230         else
 231         {
 232             vdwkt = vdwktLJEWALDCOMBLB;
 233             /* At setup we (should have) selected the C reference kernel */
 234             GMX_RELEASE_ASSERT(kernelSetup.kernelType == Nbnxm::KernelType::Cpu4x4_PlainC, "Only the C reference nbnxn SIMD kernel supports LJ-PME with LB combination rules");
 235         }
 236     }
 237     else
 238     {
 239         GMX_RELEASE_ASSERT(false, "Unsupported VdW interaction type");
 240     }
 241
 242     gmx::ArrayRef<const NbnxnPairlistCpu> pairlists = pairlistSet.cpuLists();
 243
 244     int gmx_unused                        nthreads = gmx_omp_nthreads_get(emntNonbonded);
 245     wallcycle_sub_start(wcycle, ewcsNONBONDED_CLEAR);
 246 #pragma omp parallel for schedule(static) num_threads(nthreads)
 247     for (gmx::index nb = 0; nb < pairlists.ssize(); nb++)
 248     {
 249         // Presently, the kernels do not call C++ code that can throw,
 250         // so no need for a try/catch pair in this OpenMP region.
 251         nbnxn_atomdata_output_t *out = &nbat->out[nb];
 252
 253         if (clearF == enbvClearFYes)
 254         {
 255             clearForceBuffer(nbat, nb);
 256
 257             clear_fshift(out->fshift.data());
 258         }
 259
 260         if (nb == 0)
 261         {
 262             wallcycle_sub_stop(wcycle, ewcsNONBONDED_CLEAR);
 263             wallcycle_sub_start(wcycle, ewcsNONBONDED_KERNEL);
 264         }
 265
 266         // TODO: Change to reference
 267         const NbnxnPairlistCpu *pairlist = &pairlists[nb];
 268
 269         if (!(forceFlags & GMX_FORCE_ENERGY))
 270         {
 271             /* Don't calculate energies */
 272             switch (kernelSetup.kernelType)
 273             {
 274                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 275                     nbnxn_kernel_noener_ref[coulkt][vdwkt](pairlist, nbat,
 276                                                            &ic,
 277                                                            shiftVectors,
 278                                                            out);
 279                     break;
 280 #ifdef GMX_NBNXN_SIMD_2XNN
 281                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 282                     nbnxm_kernel_noener_simd_2xmm[coulkt][vdwkt](pairlist, nbat,
 283                                                                  &ic,
 284                                                                  shiftVectors,
 285                                                                  out);
 286                     break;
 287 #endif
 288 #ifdef GMX_NBNXN_SIMD_4XN
 289                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 290                     nbnxm_kernel_noener_simd_4xm[coulkt][vdwkt](pairlist, nbat,
 291                                                                 &ic,
 292                                                                 shiftVectors,
 293                                                                 out);
 294                     break;
 295 #endif
 296                 default:
 297                     GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 298             }
 299         }
 300         else if (out->Vvdw.size() == 1)
 301         {
 302             /* A single energy group (pair) */
 303             out->Vvdw[0] = 0;
 304             out->Vc[0]   = 0;
 305
 306             switch (kernelSetup.kernelType)
 307             {
 308                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 309                     nbnxn_kernel_ener_ref[coulkt][vdwkt](pairlist, nbat,
 310                                                          &ic,
 311                                                          shiftVectors,
 312                                                          out);
 313                     break;
 314 #ifdef GMX_NBNXN_SIMD_2XNN
 315                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 316                     nbnxm_kernel_ener_simd_2xmm[coulkt][vdwkt](pairlist, nbat,
 317                                                                &ic,
 318                                                                shiftVectors,
 319                                                                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,
 325                                                               &ic,
 326                                                               shiftVectors,
 327                                                               out);
 328                     break;
 329 #endif
 330                 default:
 331                     GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 332             }
 333         }
 334         else
 335         {
 336             /* Calculate energy group contributions */
 337             clearGroupEnergies(out);
 338
 339             int unrollj = 0;
 340
 341             switch (kernelSetup.kernelType)
 342             {
 343                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 344                     unrollj = c_nbnxnCpuIClusterSize;
 345                     nbnxn_kernel_energrp_ref[coulkt][vdwkt](pairlist, nbat,
 346                                                             &ic,
 347                                                             shiftVectors,
 348                                                             out);
 349                     break;
 350 #ifdef GMX_NBNXN_SIMD_2XNN
 351                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 352                     unrollj = GMX_SIMD_REAL_WIDTH/2;
 353                     nbnxm_kernel_energrp_simd_2xmm[coulkt][vdwkt](pairlist, nbat,
 354                                                                   &ic,
 355                                                                   shiftVectors,
 356                                                                   out);
 357                     break;
 358 #endif
 359 #ifdef GMX_NBNXN_SIMD_4XN
 360                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 361                     unrollj = GMX_SIMD_REAL_WIDTH;
 362                     nbnxm_kernel_energrp_simd_4xm[coulkt][vdwkt](pairlist, nbat,
 363                                                                  &ic,
 364                                                                  shiftVectors,
 365                                                                  out);
 366                     break;
 367 #endif
 368                 default:
 369                     GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 370             }
 371
 372             if (kernelSetup.kernelType != Nbnxm::KernelType::Cpu4x4_PlainC)
 373             {
 374                 switch (unrollj)
 375                 {
 376                     case 2:
 377                         reduceGroupEnergySimdBuffers<2>(nbatParams.nenergrp,
 378                                                         nbatParams.neg_2log,
 379                                                         out);
 380                         break;
 381                     case 4:
 382                         reduceGroupEnergySimdBuffers<4>(nbatParams.nenergrp,
 383                                                         nbatParams.neg_2log,
 384                                                         out);
 385                         break;
 386                     case 8:
 387                         reduceGroupEnergySimdBuffers<8>(nbatParams.nenergrp,
 388                                                         nbatParams.neg_2log,
 389                                                         out);
 390                         break;
 391                     default:
 392                         GMX_RELEASE_ASSERT(false, "Unsupported j-unroll size");
 393                 }
 394             }
 395         }
 396     }
 397     wallcycle_sub_stop(wcycle, ewcsNONBONDED_KERNEL);
 398
 399     if (forceFlags & GMX_FORCE_ENERGY)
 400     {
 401         reduce_energies_over_lists(nbat, pairlists.ssize(), vVdw, vCoulomb);
 402     }
 403 }
 404
 405 static void accountFlops(t_nrnb                           *nrnb,
 406                          const PairlistSet                &pairlistSet,
 407                          const nonbonded_verlet_t         &nbv,
 408                          const interaction_const_t        &ic,
 409                          const int                         forceFlags)
 410 {
 411     const bool usingGpuKernels = nbv.useGpu();
 412
 413     int        enr_nbnxn_kernel_ljc;
 414     if (EEL_RF(ic.eeltype) || ic.eeltype == eelCUT)
 415     {
 416         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
 417     }
 418     else if ((!usingGpuKernels && nbv.kernelSetup().ewaldExclusionType == Nbnxm::EwaldExclusionType::Analytical) ||
 419              (usingGpuKernels && Nbnxm::gpu_is_kernel_ewald_analytical(nbv.gpu_nbv)))
 420     {
 421         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
 422     }
 423     else
 424     {
 425         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
 426     }
 427     int enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
 428     if (forceFlags & GMX_FORCE_ENERGY)
 429     {
 430         /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
 431         enr_nbnxn_kernel_ljc += 1;
 432         enr_nbnxn_kernel_lj  += 1;
 433     }
 434
 435     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc,
 436              pairlistSet.natpair_ljq_);
 437     inc_nrnb(nrnb, enr_nbnxn_kernel_lj,
 438              pairlistSet.natpair_lj_);
 439     /* The Coulomb-only kernels are offset -eNR_NBNXN_LJ_RF+eNR_NBNXN_RF */
 440     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc-eNR_NBNXN_LJ_RF+eNR_NBNXN_RF,
 441              pairlistSet.natpair_q_);
 442
 443     const bool calcEnergy = ((forceFlags & GMX_FORCE_ENERGY) != 0);
 444     if (ic.vdw_modifier == eintmodFORCESWITCH)
 445     {
 446         /* We add up the switch cost separately */
 447         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_FSW + (calcEnergy ? 1 : 0),
 448                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 449     }
 450     if (ic.vdw_modifier == eintmodPOTSWITCH)
 451     {
 452         /* We add up the switch cost separately */
 453         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_PSW + (calcEnergy ? 1 : 0),
 454                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 455     }
 456     if (ic.vdwtype == evdwPME)
 457     {
 458         /* We add up the LJ Ewald cost separately */
 459         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_EWALD + (calcEnergy ? 1 : 0),
 460                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 461     }
 462 }
 463
 464 void
 465 nonbonded_verlet_t::dispatchNonbondedKernel(Nbnxm::InteractionLocality iLocality,
 466                                             const interaction_const_t &ic,
 467                                             int                        forceFlags,
 468                                             int                        clearF,
 469                                             const t_forcerec          &fr,
 470                                             gmx_enerdata_t            *enerd,
 471                                             t_nrnb                    *nrnb)
 472 {
 473     const PairlistSet &pairlistSet = pairlistSets().pairlistSet(iLocality);
 474
 475     switch (kernelSetup().kernelType)
 476     {
 477         case Nbnxm::KernelType::Cpu4x4_PlainC:
 478         case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 479         case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 480             nbnxn_kernel_cpu(pairlistSet,
 481                              kernelSetup(),
 482                              nbat.get(),
 483                              ic,
 484                              fr.shift_vec,
 485                              forceFlags,
 486                              clearF,
 487                              enerd->grpp.ener[egCOULSR].data(),
 488                              fr.bBHAM ?
 489                              enerd->grpp.ener[egBHAMSR].data() :
 490                              enerd->grpp.ener[egLJSR].data(),
 491                              wcycle_);
 492             break;
 493
 494         case Nbnxm::KernelType::Gpu8x8x8:
 495             Nbnxm::gpu_launch_kernel(gpu_nbv, forceFlags, iLocality);
 496             break;
 497
 498         case Nbnxm::KernelType::Cpu8x8x8_PlainC:
 499             nbnxn_kernel_gpu_ref(pairlistSet.gpuList(),
 500                                  nbat.get(), &ic,
 501                                  fr.shift_vec,
 502                                  forceFlags,
 503                                  clearF,
 504                                  nbat->out[0].f,
 505                                  nbat->out[0].fshift.data(),
 506                                  enerd->grpp.ener[egCOULSR].data(),
 507                                  fr.bBHAM ?
 508                                  enerd->grpp.ener[egBHAMSR].data() :
 509                                  enerd->grpp.ener[egLJSR].data());
 510             break;
 511
 512         default:
 513             GMX_RELEASE_ASSERT(false, "Invalid nonbonded kernel type passed!");
 514
 515     }
 516
 517     accountFlops(nrnb, pairlistSet, *this, ic, forceFlags);
 518 }
 519
 520 void
 521 nonbonded_verlet_t::dispatchFreeEnergyKernel(Nbnxm::InteractionLocality  iLocality,
 522                                              const t_forcerec           *fr,
 523                                              rvec                        x[],
 524                                              gmx::ForceWithShiftForces  *forceWithShiftForces,
 525                                              const t_mdatoms            &mdatoms,
 526                                              t_lambda                   *fepvals,
 527                                              real                       *lambda,
 528                                              gmx_enerdata_t             *enerd,
 529                                              const int                   forceFlags,
 530                                              t_nrnb                     *nrnb)
 531 {
 532     const auto nbl_fep = pairlistSets().pairlistSet(iLocality).fepLists();
 533
 534     /* When the first list is empty, all are empty and there is nothing to do */
 535     if (!pairlistSets().params().haveFep || nbl_fep[0]->nrj == 0)
 536     {
 537         return;
 538     }
 539
 540     int donb_flags = 0;
 541     /* Add short-range interactions */
 542     donb_flags |= GMX_NONBONDED_DO_SR;
 543
 544     /* Currently all group scheme kernels always calculate (shift-)forces */
 545     if (forceFlags & GMX_FORCE_FORCES)
 546     {
 547         donb_flags |= GMX_NONBONDED_DO_FORCE;
 548     }
 549     if (forceFlags & GMX_FORCE_VIRIAL)
 550     {
 551         donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 552     }
 553     if (forceFlags & GMX_FORCE_ENERGY)
 554     {
 555         donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 556     }
 557
 558     nb_kernel_data_t kernel_data;
 559     real             dvdl_nb[efptNR] = { 0 };
 560     kernel_data.flags  = donb_flags;
 561     kernel_data.lambda = lambda;
 562     kernel_data.dvdl   = dvdl_nb;
 563
 564     kernel_data.energygrp_elec = enerd->grpp.ener[egCOULSR].data();
 565     kernel_data.energygrp_vdw  = enerd->grpp.ener[egLJSR].data();
 566
 567     GMX_ASSERT(gmx_omp_nthreads_get(emntNonbonded) == nbl_fep.ssize(), "Number of lists should be same as number of NB threads");
 568
 569     wallcycle_sub_start(wcycle_, ewcsNONBONDED_FEP);
 570 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
 571     for (gmx::index th = 0; th < nbl_fep.ssize(); th++)
 572     {
 573         try
 574         {
 575             gmx_nb_free_energy_kernel(nbl_fep[th].get(),
 576                                       x, forceWithShiftForces,
 577                                       fr, &mdatoms, &kernel_data, nrnb);
 578         }
 579         GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
 580     }
 581
 582     if (fepvals->sc_alpha != 0)
 583     {
 584         enerd->dvdl_nonlin[efptVDW]  += dvdl_nb[efptVDW];
 585         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 586     }
 587     else
 588     {
 589         enerd->dvdl_lin[efptVDW]  += dvdl_nb[efptVDW];
 590         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 591     }
 592
 593     /* If we do foreign lambda and we have soft-core interactions
 594      * we have to recalculate the (non-linear) energies contributions.
 595      */
 596     if (fepvals->n_lambda > 0 && (forceFlags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 597     {
 598         real lam_i[efptNR];
 599         kernel_data.flags          = (donb_flags & ~(GMX_NONBONDED_DO_FORCE | GMX_NONBONDED_DO_SHIFTFORCE)) | GMX_NONBONDED_DO_FOREIGNLAMBDA;
 600         kernel_data.lambda         = lam_i;
 601         kernel_data.energygrp_elec = enerd->foreign_grpp.ener[egCOULSR].data();
 602         kernel_data.energygrp_vdw  = enerd->foreign_grpp.ener[egLJSR].data();
 603         /* Note that we add to kernel_data.dvdl, but ignore the result */
 604
 605         for (size_t i = 0; i < enerd->enerpart_lambda.size(); i++)
 606         {
 607             for (int j = 0; j < efptNR; j++)
 608             {
 609                 lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 610             }
 611             reset_foreign_enerdata(enerd);
 612 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
 613             for (gmx::index th = 0; th < nbl_fep.ssize(); th++)
 614             {
 615                 try
 616                 {
 617                     gmx_nb_free_energy_kernel(nbl_fep[th].get(),
 618                                               x, forceWithShiftForces,
 619                                               fr, &mdatoms, &kernel_data, nrnb);
 620                 }
 621                 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
 622             }
 623
 624             sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 625             enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 626         }
 627     }
 628     wallcycle_sub_stop(wcycle_, ewcsNONBONDED_FEP);
 629 }