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/inputrec.h"
  49 #include "gromacs/mdtypes/interaction_const.h"
  50 #include "gromacs/mdtypes/md_enums.h"
  51 #include "gromacs/mdtypes/mdatom.h"
  52 #include "gromacs/nbnxm/gpu_data_mgmt.h"
  53 #include "gromacs/nbnxm/nbnxm.h"
  54 #include "gromacs/nbnxm/nbnxm_simd.h"
  55 #include "gromacs/nbnxm/kernels_reference/kernel_gpu_ref.h"
  56 #include "gromacs/simd/simd.h"
  57 #include "gromacs/utility/gmxassert.h"
  58 #include "gromacs/utility/real.h"
  59
  60 #include "kernel_common.h"
  61 #include "pairlistset.h"
  62 #include "pairlistsets.h"
  63 #define INCLUDE_KERNELFUNCTION_TABLES
  64 #include "gromacs/nbnxm/kernels_reference/kernel_ref.h"
  65 #ifdef GMX_NBNXN_SIMD_2XNN
  66 #include "gromacs/nbnxm/kernels_simd_2xmm/kernels.h"
  67 #endif
  68 #ifdef GMX_NBNXN_SIMD_4XN
  69 #include "gromacs/nbnxm/kernels_simd_4xm/kernels.h"
  70 #endif
  71 #undef INCLUDE_FUNCTION_TABLES
  72
  73 /*! \brief Clears the energy group output buffers
  74  *
  75  * \param[in,out] out  nbnxn kernel output struct
  76  */
  77 static void clearGroupEnergies(nbnxn_atomdata_output_t *out)
  78 {
  79     std::fill(out->Vvdw.begin(), out->Vvdw.end(), 0.0_real);
  80     std::fill(out->Vc.begin(), out->Vc.end(), 0.0_real);
  81     std::fill(out->VSvdw.begin(), out->VSvdw.end(), 0.0_real);
  82     std::fill(out->VSc.begin(), out->VSc.end(), 0.0_real);
  83 }
  84
  85 /*! \brief Reduce the group-pair energy buffers produced by a SIMD kernel
  86  * to single terms in the output buffers.
  87  *
  88  * The SIMD kernels produce a large number of energy buffer in SIMD registers
  89  * to avoid scattered reads and writes.
  90  *
  91  * \tparam        unrollj         The unroll size for j-particles in the SIMD kernel
  92  * \param[in]     numGroups       The number of energy groups
  93  * \param[in]     numGroups_2log  Log2 of numGroups, rounded up
  94  * \param[in,out] out             Struct with energy buffers
  95  */
  96 template <int unrollj> static void
  97 reduceGroupEnergySimdBuffers(int                       numGroups,
  98                              int                       numGroups_2log,
  99                              nbnxn_atomdata_output_t  *out)
 100 {
 101     const int                 unrollj_half     = unrollj/2;
 102     /* Energies are stored in SIMD registers with size 2^numGroups_2log */
 103     const int                 numGroupsStorage = (1 << numGroups_2log);
 104
 105     const real * gmx_restrict vVdwSimd     = out->VSvdw.data();
 106     const real * gmx_restrict vCoulombSimd = out->VSc.data();
 107     real * gmx_restrict       vVdw         = out->Vvdw.data();
 108     real * gmx_restrict       vCoulomb     = out->Vc.data();
 109
 110     /* The size of the SIMD energy group buffer array is:
 111      * numGroups*numGroups*numGroupsStorage*unrollj_half*simd_width
 112      */
 113     for (int i = 0; i < numGroups; i++)
 114     {
 115         for (int j1 = 0; j1 < numGroups; j1++)
 116         {
 117             for (int j0 = 0; j0 < numGroups; j0++)
 118             {
 119                 int c = ((i*numGroups + j1)*numGroupsStorage + j0)*unrollj_half*unrollj;
 120                 for (int s = 0; s < unrollj_half; s++)
 121                 {
 122                     vVdw    [i*numGroups + j0] += vVdwSimd    [c + 0];
 123                     vVdw    [i*numGroups + j1] += vVdwSimd    [c + 1];
 124                     vCoulomb[i*numGroups + j0] += vCoulombSimd[c + 0];
 125                     vCoulomb[i*numGroups + j1] += vCoulombSimd[c + 1];
 126                     c                          += unrollj + 2;
 127                 }
 128             }
 129         }
 130     }
 131 }
 132
 133 /*! \brief Dispatches the non-bonded N versus M atom cluster CPU kernels.
 134  *
 135  * OpenMP parallelization is performed within this function.
 136  * Energy reduction, but not force and shift force reduction, is performed
 137  * within this function.
 138  *
 139  * \param[in]     pairlistSet   Pairlists with local or non-local interactions to compute
 140  * \param[in]     kernelSetup   The non-bonded kernel setup
 141  * \param[in,out] nbat          The atomdata for the interactions
 142  * \param[in]     ic            Non-bonded interaction constants
 143  * \param[in]     shiftVectors  The PBC shift vectors
 144  * \param[in]     forceFlags    Flags that tell what to compute
 145  * \param[in]     clearF        Enum that tells if to clear the force output buffer
 146  * \param[out]    vCoulomb      Output buffer for Coulomb energies
 147  * \param[out]    vVdw          Output buffer for Van der Waals energies
 148  */
 149 static void
 150 nbnxn_kernel_cpu(const PairlistSet              &pairlistSet,
 151                  const Nbnxm::KernelSetup       &kernelSetup,
 152                  nbnxn_atomdata_t               *nbat,
 153                  const interaction_const_t      &ic,
 154                  rvec                           *shiftVectors,
 155                  int                             forceFlags,
 156                  int                             clearF,
 157                  real                           *vCoulomb,
 158                  real                           *vVdw)
 159 {
 160
 161     int                      coulkt;
 162     if (EEL_RF(ic.eeltype) || ic.eeltype == eelCUT)
 163     {
 164         coulkt = coulktRF;
 165     }
 166     else
 167     {
 168         if (kernelSetup.ewaldExclusionType == Nbnxm::EwaldExclusionType::Table)
 169         {
 170             if (ic.rcoulomb == ic.rvdw)
 171             {
 172                 coulkt = coulktTAB;
 173             }
 174             else
 175             {
 176                 coulkt = coulktTAB_TWIN;
 177             }
 178         }
 179         else
 180         {
 181             if (ic.rcoulomb == ic.rvdw)
 182             {
 183                 coulkt = coulktEWALD;
 184             }
 185             else
 186             {
 187                 coulkt = coulktEWALD_TWIN;
 188             }
 189         }
 190     }
 191
 192     const nbnxn_atomdata_t::Params &nbatParams = nbat->params();
 193
 194     int vdwkt = 0;
 195     if (ic.vdwtype == evdwCUT)
 196     {
 197         switch (ic.vdw_modifier)
 198         {
 199             case eintmodNONE:
 200             case eintmodPOTSHIFT:
 201                 switch (nbatParams.comb_rule)
 202                 {
 203                     case ljcrGEOM: vdwkt = vdwktLJCUT_COMBGEOM; break;
 204                     case ljcrLB:   vdwkt = vdwktLJCUT_COMBLB;   break;
 205                     case ljcrNONE: vdwkt = vdwktLJCUT_COMBNONE; break;
 206                     default:
 207                         GMX_RELEASE_ASSERT(false, "Unknown combination rule");
 208                 }
 209                 break;
 210             case eintmodFORCESWITCH:
 211                 vdwkt = vdwktLJFORCESWITCH;
 212                 break;
 213             case eintmodPOTSWITCH:
 214                 vdwkt = vdwktLJPOTSWITCH;
 215                 break;
 216             default:
 217                 GMX_RELEASE_ASSERT(false, "Unsupported VdW interaction modifier");
 218         }
 219     }
 220     else if (ic.vdwtype == evdwPME)
 221     {
 222         if (ic.ljpme_comb_rule == eljpmeGEOM)
 223         {
 224             vdwkt = vdwktLJEWALDCOMBGEOM;
 225         }
 226         else
 227         {
 228             vdwkt = vdwktLJEWALDCOMBLB;
 229             /* At setup we (should have) selected the C reference kernel */
 230             GMX_RELEASE_ASSERT(kernelSetup.kernelType == Nbnxm::KernelType::Cpu4x4_PlainC, "Only the C reference nbnxn SIMD kernel supports LJ-PME with LB 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 #pragma omp parallel for schedule(static) num_threads(nthreads)
 242     for (int nb = 0; nb < pairlists.ssize(); nb++)
 243     {
 244         // Presently, the kernels do not call C++ code that can throw,
 245         // so no need for a try/catch pair in this OpenMP region.
 246         nbnxn_atomdata_output_t *out = &nbat->out[nb];
 247
 248         if (clearF == enbvClearFYes)
 249         {
 250             clearForceBuffer(nbat, nb);
 251
 252             clear_fshift(out->fshift.data());
 253         }
 254
 255         // TODO: Change to reference
 256         const NbnxnPairlistCpu *pairlist = &pairlists[nb];
 257
 258         if (!(forceFlags & GMX_FORCE_ENERGY))
 259         {
 260             /* Don't calculate energies */
 261             switch (kernelSetup.kernelType)
 262             {
 263                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 264                     nbnxn_kernel_noener_ref[coulkt][vdwkt](pairlist, nbat,
 265                                                            &ic,
 266                                                            shiftVectors,
 267                                                            out);
 268                     break;
 269 #ifdef GMX_NBNXN_SIMD_2XNN
 270                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 271                     nbnxm_kernel_noener_simd_2xmm[coulkt][vdwkt](pairlist, nbat,
 272                                                                  &ic,
 273                                                                  shiftVectors,
 274                                                                  out);
 275                     break;
 276 #endif
 277 #ifdef GMX_NBNXN_SIMD_4XN
 278                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 279                     nbnxm_kernel_noener_simd_4xm[coulkt][vdwkt](pairlist, nbat,
 280                                                                 &ic,
 281                                                                 shiftVectors,
 282                                                                 out);
 283                     break;
 284 #endif
 285                 default:
 286                     GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 287             }
 288         }
 289         else if (out->Vvdw.size() == 1)
 290         {
 291             /* A single energy group (pair) */
 292             out->Vvdw[0] = 0;
 293             out->Vc[0]   = 0;
 294
 295             switch (kernelSetup.kernelType)
 296             {
 297                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 298                     nbnxn_kernel_ener_ref[coulkt][vdwkt](pairlist, nbat,
 299                                                          &ic,
 300                                                          shiftVectors,
 301                                                          out);
 302                     break;
 303 #ifdef GMX_NBNXN_SIMD_2XNN
 304                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 305                     nbnxm_kernel_ener_simd_2xmm[coulkt][vdwkt](pairlist, nbat,
 306                                                                &ic,
 307                                                                shiftVectors,
 308                                                                out);
 309                     break;
 310 #endif
 311 #ifdef GMX_NBNXN_SIMD_4XN
 312                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 313                     nbnxm_kernel_ener_simd_4xm[coulkt][vdwkt](pairlist, nbat,
 314                                                               &ic,
 315                                                               shiftVectors,
 316                                                               out);
 317                     break;
 318 #endif
 319                 default:
 320                     GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 321             }
 322         }
 323         else
 324         {
 325             /* Calculate energy group contributions */
 326             clearGroupEnergies(out);
 327
 328             int unrollj = 0;
 329
 330             switch (kernelSetup.kernelType)
 331             {
 332                 case Nbnxm::KernelType::Cpu4x4_PlainC:
 333                     unrollj = c_nbnxnCpuIClusterSize;
 334                     nbnxn_kernel_energrp_ref[coulkt][vdwkt](pairlist, nbat,
 335                                                             &ic,
 336                                                             shiftVectors,
 337                                                             out);
 338                     break;
 339 #ifdef GMX_NBNXN_SIMD_2XNN
 340                 case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 341                     unrollj = GMX_SIMD_REAL_WIDTH/2;
 342                     nbnxm_kernel_energrp_simd_2xmm[coulkt][vdwkt](pairlist, nbat,
 343                                                                   &ic,
 344                                                                   shiftVectors,
 345                                                                   out);
 346                     break;
 347 #endif
 348 #ifdef GMX_NBNXN_SIMD_4XN
 349                 case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 350                     unrollj = GMX_SIMD_REAL_WIDTH;
 351                     nbnxm_kernel_energrp_simd_4xm[coulkt][vdwkt](pairlist, nbat,
 352                                                                  &ic,
 353                                                                  shiftVectors,
 354                                                                  out);
 355                     break;
 356 #endif
 357                 default:
 358                     GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 359             }
 360
 361             if (kernelSetup.kernelType != Nbnxm::KernelType::Cpu4x4_PlainC)
 362             {
 363                 switch (unrollj)
 364                 {
 365                     case 2:
 366                         reduceGroupEnergySimdBuffers<2>(nbatParams.nenergrp,
 367                                                         nbatParams.neg_2log,
 368                                                         out);
 369                         break;
 370                     case 4:
 371                         reduceGroupEnergySimdBuffers<4>(nbatParams.nenergrp,
 372                                                         nbatParams.neg_2log,
 373                                                         out);
 374                         break;
 375                     case 8:
 376                         reduceGroupEnergySimdBuffers<8>(nbatParams.nenergrp,
 377                                                         nbatParams.neg_2log,
 378                                                         out);
 379                         break;
 380                     default:
 381                         GMX_RELEASE_ASSERT(false, "Unsupported j-unroll size");
 382                 }
 383             }
 384         }
 385     }
 386
 387     if (forceFlags & GMX_FORCE_ENERGY)
 388     {
 389         reduce_energies_over_lists(nbat, pairlists.ssize(), vVdw, vCoulomb);
 390     }
 391 }
 392
 393 static void accountFlops(t_nrnb                           *nrnb,
 394                          const PairlistSet                &pairlistSet,
 395                          const nonbonded_verlet_t         &nbv,
 396                          const interaction_const_t        &ic,
 397                          const int                         forceFlags)
 398 {
 399     const bool usingGpuKernels = nbv.useGpu();
 400
 401     int        enr_nbnxn_kernel_ljc;
 402     if (EEL_RF(ic.eeltype) || ic.eeltype == eelCUT)
 403     {
 404         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
 405     }
 406     else if ((!usingGpuKernels && nbv.kernelSetup().ewaldExclusionType == Nbnxm::EwaldExclusionType::Analytical) ||
 407              (usingGpuKernels && Nbnxm::gpu_is_kernel_ewald_analytical(nbv.gpu_nbv)))
 408     {
 409         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
 410     }
 411     else
 412     {
 413         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
 414     }
 415     int enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
 416     if (forceFlags & GMX_FORCE_ENERGY)
 417     {
 418         /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
 419         enr_nbnxn_kernel_ljc += 1;
 420         enr_nbnxn_kernel_lj  += 1;
 421     }
 422
 423     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc,
 424              pairlistSet.natpair_ljq_);
 425     inc_nrnb(nrnb, enr_nbnxn_kernel_lj,
 426              pairlistSet.natpair_lj_);
 427     /* The Coulomb-only kernels are offset -eNR_NBNXN_LJ_RF+eNR_NBNXN_RF */
 428     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc-eNR_NBNXN_LJ_RF+eNR_NBNXN_RF,
 429              pairlistSet.natpair_q_);
 430
 431     const bool calcEnergy = ((forceFlags & GMX_FORCE_ENERGY) != 0);
 432     if (ic.vdw_modifier == eintmodFORCESWITCH)
 433     {
 434         /* We add up the switch cost separately */
 435         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_FSW + (calcEnergy ? 1 : 0),
 436                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 437     }
 438     if (ic.vdw_modifier == eintmodPOTSWITCH)
 439     {
 440         /* We add up the switch cost separately */
 441         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_PSW + (calcEnergy ? 1 : 0),
 442                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 443     }
 444     if (ic.vdwtype == evdwPME)
 445     {
 446         /* We add up the LJ Ewald cost separately */
 447         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_EWALD + (calcEnergy ? 1 : 0),
 448                  pairlistSet.natpair_ljq_ + pairlistSet.natpair_lj_);
 449     }
 450 }
 451
 452 void
 453 nonbonded_verlet_t::dispatchNonbondedKernel(Nbnxm::InteractionLocality iLocality,
 454                                             const interaction_const_t &ic,
 455                                             int                        forceFlags,
 456                                             int                        clearF,
 457                                             const t_forcerec          &fr,
 458                                             gmx_enerdata_t            *enerd,
 459                                             t_nrnb                    *nrnb)
 460 {
 461     const PairlistSet &pairlistSet = pairlistSets().pairlistSet(iLocality);
 462
 463     switch (kernelSetup().kernelType)
 464     {
 465         case Nbnxm::KernelType::Cpu4x4_PlainC:
 466         case Nbnxm::KernelType::Cpu4xN_Simd_4xN:
 467         case Nbnxm::KernelType::Cpu4xN_Simd_2xNN:
 468             nbnxn_kernel_cpu(pairlistSet,
 469                              kernelSetup(),
 470                              nbat.get(),
 471                              ic,
 472                              fr.shift_vec,
 473                              forceFlags,
 474                              clearF,
 475                              enerd->grpp.ener[egCOULSR].data(),
 476                              fr.bBHAM ?
 477                              enerd->grpp.ener[egBHAMSR].data() :
 478                              enerd->grpp.ener[egLJSR].data());
 479             break;
 480
 481         case Nbnxm::KernelType::Gpu8x8x8:
 482             Nbnxm::gpu_launch_kernel(gpu_nbv, forceFlags, iLocality);
 483             break;
 484
 485         case Nbnxm::KernelType::Cpu8x8x8_PlainC:
 486             nbnxn_kernel_gpu_ref(pairlistSet.gpuList(),
 487                                  nbat.get(), &ic,
 488                                  fr.shift_vec,
 489                                  forceFlags,
 490                                  clearF,
 491                                  nbat->out[0].f,
 492                                  nbat->out[0].fshift.data(),
 493                                  enerd->grpp.ener[egCOULSR].data(),
 494                                  fr.bBHAM ?
 495                                  enerd->grpp.ener[egBHAMSR].data() :
 496                                  enerd->grpp.ener[egLJSR].data());
 497             break;
 498
 499         default:
 500             GMX_RELEASE_ASSERT(false, "Invalid nonbonded kernel type passed!");
 501
 502     }
 503
 504     accountFlops(nrnb, pairlistSet, *this, ic, forceFlags);
 505 }
 506
 507 void
 508 nonbonded_verlet_t::dispatchFreeEnergyKernel(Nbnxm::InteractionLocality  iLocality,
 509                                              t_forcerec                 *fr,
 510                                              rvec                        x[],
 511                                              rvec                        f[],
 512                                              const t_mdatoms            &mdatoms,
 513                                              t_lambda                   *fepvals,
 514                                              real                       *lambda,
 515                                              gmx_enerdata_t             *enerd,
 516                                              const int                   forceFlags,
 517                                              t_nrnb                     *nrnb)
 518 {
 519     const auto nbl_fep = pairlistSets().pairlistSet(iLocality).fepLists();
 520
 521     /* When the first list is empty, all are empty and there is nothing to do */
 522     if (!pairlistSets().params().haveFep || nbl_fep[0]->nrj == 0)
 523     {
 524         return;
 525     }
 526
 527     int donb_flags = 0;
 528     /* Add short-range interactions */
 529     donb_flags |= GMX_NONBONDED_DO_SR;
 530
 531     /* Currently all group scheme kernels always calculate (shift-)forces */
 532     if (forceFlags & GMX_FORCE_FORCES)
 533     {
 534         donb_flags |= GMX_NONBONDED_DO_FORCE;
 535     }
 536     if (forceFlags & GMX_FORCE_VIRIAL)
 537     {
 538         donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
 539     }
 540     if (forceFlags & GMX_FORCE_ENERGY)
 541     {
 542         donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
 543     }
 544
 545     nb_kernel_data_t kernel_data;
 546     real             dvdl_nb[efptNR] = { 0 };
 547     kernel_data.flags  = donb_flags;
 548     kernel_data.lambda = lambda;
 549     kernel_data.dvdl   = dvdl_nb;
 550
 551     kernel_data.energygrp_elec = enerd->grpp.ener[egCOULSR].data();
 552     kernel_data.energygrp_vdw  = enerd->grpp.ener[egLJSR].data();
 553
 554     GMX_ASSERT(gmx_omp_nthreads_get(emntNonbonded) == nbl_fep.ssize(), "Number of lists should be same as number of NB threads");
 555
 556 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
 557     for (int th = 0; th < nbl_fep.ssize(); th++)
 558     {
 559         try
 560         {
 561             gmx_nb_free_energy_kernel(nbl_fep[th].get(),
 562                                       x, f, fr, &mdatoms, &kernel_data, nrnb);
 563         }
 564         GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
 565     }
 566
 567     if (fepvals->sc_alpha != 0)
 568     {
 569         enerd->dvdl_nonlin[efptVDW]  += dvdl_nb[efptVDW];
 570         enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
 571     }
 572     else
 573     {
 574         enerd->dvdl_lin[efptVDW]  += dvdl_nb[efptVDW];
 575         enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
 576     }
 577
 578     /* If we do foreign lambda and we have soft-core interactions
 579      * we have to recalculate the (non-linear) energies contributions.
 580      */
 581     if (fepvals->n_lambda > 0 && (forceFlags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
 582     {
 583         real lam_i[efptNR];
 584         kernel_data.flags          = (donb_flags & ~(GMX_NONBONDED_DO_FORCE | GMX_NONBONDED_DO_SHIFTFORCE)) | GMX_NONBONDED_DO_FOREIGNLAMBDA;
 585         kernel_data.lambda         = lam_i;
 586         kernel_data.energygrp_elec = enerd->foreign_grpp.ener[egCOULSR].data();
 587         kernel_data.energygrp_vdw  = enerd->foreign_grpp.ener[egLJSR].data();
 588         /* Note that we add to kernel_data.dvdl, but ignore the result */
 589
 590         for (size_t i = 0; i < enerd->enerpart_lambda.size(); i++)
 591         {
 592             for (int j = 0; j < efptNR; j++)
 593             {
 594                 lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
 595             }
 596             reset_foreign_enerdata(enerd);
 597 #pragma omp parallel for schedule(static) num_threads(nbl_fep.ssize())
 598             for (int th = 0; th < nbl_fep.ssize(); th++)
 599             {
 600                 try
 601                 {
 602                     gmx_nb_free_energy_kernel(nbl_fep[th].get(),
 603                                               x, f, fr, &mdatoms, &kernel_data, nrnb);
 604                 }
 605                 GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
 606             }
 607
 608             sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
 609             enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
 610         }
 611     }
 612 }