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/math/vectypes.h"
  40 #include "gromacs/mdlib/force_flags.h"
  41 #include "gromacs/mdlib/gmx_omp_nthreads.h"
  42 #include "gromacs/mdtypes/enerdata.h"
  43 #include "gromacs/mdtypes/interaction_const.h"
  44 #include "gromacs/mdtypes/md_enums.h"
  45 #include "gromacs/nbnxm/gpu_data_mgmt.h"
  46 #include "gromacs/nbnxm/nbnxm.h"
  47 #include "gromacs/nbnxm/nbnxm_simd.h"
  48 #include "gromacs/nbnxm/kernels_reference/kernel_gpu_ref.h"
  49 #include "gromacs/simd/simd.h"
  50 #include "gromacs/utility/gmxassert.h"
  51 #include "gromacs/utility/real.h"
  52
  53 #include "kernel_common.h"
  54 #define INCLUDE_KERNELFUNCTION_TABLES
  55 #include "gromacs/nbnxm/kernels_reference/kernel_ref.h"
  56 #ifdef GMX_NBNXN_SIMD_2XNN
  57 #include "gromacs/nbnxm/kernels_simd_2xmm/kernels.h"
  58 #endif
  59 #ifdef GMX_NBNXN_SIMD_4XN
  60 #include "gromacs/nbnxm/kernels_simd_4xm/kernels.h"
  61 #endif
  62 #undef INCLUDE_FUNCTION_TABLES
  63
  64 /*! \brief Clears the energy group output buffers
  65  *
  66  * \param[in,out] out  nbnxn kernel output struct
  67  */
  68 static void clearGroupEnergies(nbnxn_atomdata_output_t *out)
  69 {
  70     std::fill(out->Vvdw.begin(), out->Vvdw.end(), 0.0_real);
  71     std::fill(out->Vc.begin(), out->Vc.end(), 0.0_real);
  72     std::fill(out->VSvdw.begin(), out->VSvdw.end(), 0.0_real);
  73     std::fill(out->VSc.begin(), out->VSc.end(), 0.0_real);
  74 }
  75
  76 /*! \brief Reduce the group-pair energy buffers produced by a SIMD kernel
  77  * to single terms in the output buffers.
  78  *
  79  * The SIMD kernels produce a large number of energy buffer in SIMD registers
  80  * to avoid scattered reads and writes.
  81  *
  82  * \tparam        unrollj         The unroll size for j-particles in the SIMD kernel
  83  * \param[in]     numGroups       The number of energy groups
  84  * \param[in]     numGroups_2log  Log2 of numGroups, rounded up
  85  * \param[in,out] out             Struct with energy buffers
  86  */
  87 template <int unrollj> static void
  88 reduceGroupEnergySimdBuffers(int                       numGroups,
  89                              int                       numGroups_2log,
  90                              nbnxn_atomdata_output_t  *out)
  91 {
  92     const int                 unrollj_half     = unrollj/2;
  93     /* Energies are stored in SIMD registers with size 2^numGroups_2log */
  94     const int                 numGroupsStorage = (1 << numGroups_2log);
  95
  96     const real * gmx_restrict vVdwSimd     = out->VSvdw.data();
  97     const real * gmx_restrict vCoulombSimd = out->VSc.data();
  98     real * gmx_restrict       vVdw         = out->Vvdw.data();
  99     real * gmx_restrict       vCoulomb     = out->Vc.data();
 100
 101     /* The size of the SIMD energy group buffer array is:
 102      * numGroups*numGroups*numGroupsStorage*unrollj_half*simd_width
 103      */
 104     for (int i = 0; i < numGroups; i++)
 105     {
 106         for (int j1 = 0; j1 < numGroups; j1++)
 107         {
 108             for (int j0 = 0; j0 < numGroups; j0++)
 109             {
 110                 int c = ((i*numGroups + j1)*numGroupsStorage + j0)*unrollj_half*unrollj;
 111                 for (int s = 0; s < unrollj_half; s++)
 112                 {
 113                     vVdw    [i*numGroups + j0] += vVdwSimd    [c + 0];
 114                     vVdw    [i*numGroups + j1] += vVdwSimd    [c + 1];
 115                     vCoulomb[i*numGroups + j0] += vCoulombSimd[c + 0];
 116                     vCoulomb[i*numGroups + j1] += vCoulombSimd[c + 1];
 117                     c                          += unrollj + 2;
 118                 }
 119             }
 120         }
 121     }
 122 }
 123
 124 /*! \brief Dispatches the non-bonded N versus M atom cluster CPU kernels.
 125  *
 126  * OpenMP parallelization is performed within this function.
 127  * Energy reduction, but not force and shift force reduction, is performed
 128  * within this function.
 129  *
 130  * \param[in]     pairlistSet   Pairlists with local or non-local interactions to compute
 131  * \param[in]     kernel_type   The non-bonded kernel type
 132  * \param[in]     ewald_excl    The Ewald exclusion treatment
 133  * \param[in,out] nbat          The atomdata for the interactions
 134  * \param[in]     ic            Non-bonded interaction constants
 135  * \param[in]     shiftVectors  The PBC shift vectors
 136  * \param[in]     forceFlags    Flags that tell what to compute
 137  * \param[in]     clearF        Enum that tells if to clear the force output buffer
 138  * \param[out]    fshift        Shift force output buffer
 139  * \param[out]    vCoulomb      Output buffer for Coulomb energies
 140  * \param[out]    vVdw          Output buffer for Van der Waals energies
 141  */
 142 static void
 143 nbnxn_kernel_cpu(const nbnxn_pairlist_set_t     &pairlistSet,
 144                  const int                       kernel_type,
 145                  const int                       ewald_excl,
 146                  nbnxn_atomdata_t               *nbat,
 147                  const interaction_const_t      &ic,
 148                  rvec                           *shiftVectors,
 149                  int                             forceFlags,
 150                  int                             clearF,
 151                  real                           *fshift,
 152                  real                           *vCoulomb,
 153                  real                           *vVdw)
 154 {
 155
 156     int                      coulkt;
 157     if (EEL_RF(ic.eeltype) || ic.eeltype == eelCUT)
 158     {
 159         coulkt = coulktRF;
 160     }
 161     else
 162     {
 163         if (ewald_excl == ewaldexclTable)
 164         {
 165             if (ic.rcoulomb == ic.rvdw)
 166             {
 167                 coulkt = coulktTAB;
 168             }
 169             else
 170             {
 171                 coulkt = coulktTAB_TWIN;
 172             }
 173         }
 174         else
 175         {
 176             if (ic.rcoulomb == ic.rvdw)
 177             {
 178                 coulkt = coulktEWALD;
 179             }
 180             else
 181             {
 182                 coulkt = coulktEWALD_TWIN;
 183             }
 184         }
 185     }
 186
 187     const nbnxn_atomdata_t::Params &nbatParams = nbat->params();
 188
 189     int vdwkt = 0;
 190     if (ic.vdwtype == evdwCUT)
 191     {
 192         switch (ic.vdw_modifier)
 193         {
 194             case eintmodNONE:
 195             case eintmodPOTSHIFT:
 196                 switch (nbatParams.comb_rule)
 197                 {
 198                     case ljcrGEOM: vdwkt = vdwktLJCUT_COMBGEOM; break;
 199                     case ljcrLB:   vdwkt = vdwktLJCUT_COMBLB;   break;
 200                     case ljcrNONE: vdwkt = vdwktLJCUT_COMBNONE; break;
 201                     default:
 202                         GMX_RELEASE_ASSERT(false, "Unknown combination rule");
 203                 }
 204                 break;
 205             case eintmodFORCESWITCH:
 206                 vdwkt = vdwktLJFORCESWITCH;
 207                 break;
 208             case eintmodPOTSWITCH:
 209                 vdwkt = vdwktLJPOTSWITCH;
 210                 break;
 211             default:
 212                 GMX_RELEASE_ASSERT(false, "Unsupported VdW interaction modifier");
 213         }
 214     }
 215     else if (ic.vdwtype == evdwPME)
 216     {
 217         if (ic.ljpme_comb_rule == eljpmeGEOM)
 218         {
 219             vdwkt = vdwktLJEWALDCOMBGEOM;
 220         }
 221         else
 222         {
 223             vdwkt = vdwktLJEWALDCOMBLB;
 224             /* At setup we (should have) selected the C reference kernel */
 225             GMX_RELEASE_ASSERT(kernel_type == nbnxnk4x4_PlainC, "Only the C reference nbnxn SIMD kernel supports LJ-PME with LB combination rules");
 226         }
 227     }
 228     else
 229     {
 230         GMX_RELEASE_ASSERT(false, "Unsupported VdW interaction type");
 231     }
 232
 233     int                        nnbl = pairlistSet.nnbl;
 234     NbnxnPairlistCpu * const * nbl  = pairlistSet.nbl;
 235
 236     int gmx_unused             nthreads = gmx_omp_nthreads_get(emntNonbonded);
 237 #pragma omp parallel for schedule(static) num_threads(nthreads)
 238     for (int nb = 0; nb < nnbl; nb++)
 239     {
 240         // Presently, the kernels do not call C++ code that can throw,
 241         // so no need for a try/catch pair in this OpenMP region.
 242         nbnxn_atomdata_output_t *out = &nbat->out[nb];
 243
 244         if (clearF == enbvClearFYes)
 245         {
 246             clear_f(nbat, nb, out->f.data());
 247         }
 248
 249         real *fshift_p;
 250         if ((forceFlags & GMX_FORCE_VIRIAL) && nnbl == 1)
 251         {
 252             fshift_p = fshift;
 253         }
 254         else
 255         {
 256             fshift_p = out->fshift.data();
 257
 258             if (clearF == enbvClearFYes)
 259             {
 260                 clear_fshift(fshift_p);
 261             }
 262         }
 263
 264         if (!(forceFlags & GMX_FORCE_ENERGY))
 265         {
 266             /* Don't calculate energies */
 267             switch (kernel_type)
 268             {
 269                 case nbnxnk4x4_PlainC:
 270                     nbnxn_kernel_noener_ref[coulkt][vdwkt](nbl[nb], nbat,
 271                                                            &ic,
 272                                                            shiftVectors,
 273                                                            out->f.data(),
 274                                                            fshift_p);
 275                     break;
 276 #ifdef GMX_NBNXN_SIMD_2XNN
 277                 case nbnxnk4xN_SIMD_2xNN:
 278                     nbnxm_kernel_noener_simd_2xmm[coulkt][vdwkt](nbl[nb], nbat,
 279                                                                  &ic,
 280                                                                  shiftVectors,
 281                                                                  out->f.data(),
 282                                                                  fshift_p);
 283                     break;
 284 #endif
 285 #ifdef GMX_NBNXN_SIMD_4XN
 286                 case nbnxnk4xN_SIMD_4xN:
 287                     nbnxm_kernel_noener_simd_4xm[coulkt][vdwkt](nbl[nb], nbat,
 288                                                                 &ic,
 289                                                                 shiftVectors,
 290                                                                 out->f.data(),
 291                                                                 fshift_p);
 292                     break;
 293 #endif
 294                 default:
 295                     GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 296             }
 297         }
 298         else if (out->Vvdw.size() == 1)
 299         {
 300             /* A single energy group (pair) */
 301             out->Vvdw[0] = 0;
 302             out->Vc[0]   = 0;
 303
 304             switch (kernel_type)
 305             {
 306                 case nbnxnk4x4_PlainC:
 307                     nbnxn_kernel_ener_ref[coulkt][vdwkt](nbl[nb], nbat,
 308                                                          &ic,
 309                                                          shiftVectors,
 310                                                          out->f.data(),
 311                                                          fshift_p,
 312                                                          out->Vvdw.data(),
 313                                                          out->Vc.data());
 314                     break;
 315 #ifdef GMX_NBNXN_SIMD_2XNN
 316                 case nbnxnk4xN_SIMD_2xNN:
 317                     nbnxm_kernel_ener_simd_2xmm[coulkt][vdwkt](nbl[nb], nbat,
 318                                                                &ic,
 319                                                                shiftVectors,
 320                                                                out->f.data(),
 321                                                                fshift_p,
 322                                                                out->Vvdw.data(),
 323                                                                out->Vc.data());
 324                     break;
 325 #endif
 326 #ifdef GMX_NBNXN_SIMD_4XN
 327                 case nbnxnk4xN_SIMD_4xN:
 328                     nbnxm_kernel_ener_simd_4xm[coulkt][vdwkt](nbl[nb], nbat,
 329                                                               &ic,
 330                                                               shiftVectors,
 331                                                               out->f.data(),
 332                                                               fshift_p,
 333                                                               out->Vvdw.data(),
 334                                                               out->Vc.data());
 335                     break;
 336 #endif
 337                 default:
 338                     GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 339             }
 340         }
 341         else
 342         {
 343             /* Calculate energy group contributions */
 344             clearGroupEnergies(out);
 345
 346             int unrollj = 0;
 347
 348             switch (kernel_type)
 349             {
 350                 case nbnxnk4x4_PlainC:
 351                     unrollj = c_nbnxnCpuIClusterSize;
 352                     nbnxn_kernel_energrp_ref[coulkt][vdwkt](nbl[nb], nbat,
 353                                                             &ic,
 354                                                             shiftVectors,
 355                                                             out->f.data(),
 356                                                             fshift_p,
 357                                                             out->Vvdw.data(),
 358                                                             out->Vc.data());
 359                     break;
 360 #ifdef GMX_NBNXN_SIMD_2XNN
 361                 case nbnxnk4xN_SIMD_2xNN:
 362                     unrollj = GMX_SIMD_REAL_WIDTH/2;
 363                     nbnxm_kernel_energrp_simd_2xmm[coulkt][vdwkt](nbl[nb], nbat,
 364                                                                   &ic,
 365                                                                   shiftVectors,
 366                                                                   out->f.data(),
 367                                                                   fshift_p,
 368                                                                   out->VSvdw.data(),
 369                                                                   out->VSc.data());
 370                     break;
 371 #endif
 372 #ifdef GMX_NBNXN_SIMD_4XN
 373                 case nbnxnk4xN_SIMD_4xN:
 374                     unrollj = GMX_SIMD_REAL_WIDTH;
 375                     nbnxm_kernel_energrp_simd_4xm[coulkt][vdwkt](nbl[nb], nbat,
 376                                                                  &ic,
 377                                                                  shiftVectors,
 378                                                                  out->f.data(),
 379                                                                  fshift_p,
 380                                                                  out->VSvdw.data(),
 381                                                                  out->VSc.data());
 382                     break;
 383 #endif
 384                 default:
 385                     GMX_RELEASE_ASSERT(false, "Unsupported kernel architecture");
 386             }
 387
 388             if (kernel_type != nbnxnk4x4_PlainC)
 389             {
 390                 switch (unrollj)
 391                 {
 392                     case 2:
 393                         reduceGroupEnergySimdBuffers<2>(nbatParams.nenergrp,
 394                                                         nbatParams.neg_2log,
 395                                                         out);
 396                         break;
 397                     case 4:
 398                         reduceGroupEnergySimdBuffers<4>(nbatParams.nenergrp,
 399                                                         nbatParams.neg_2log,
 400                                                         out);
 401                         break;
 402                     case 8:
 403                         reduceGroupEnergySimdBuffers<8>(nbatParams.nenergrp,
 404                                                         nbatParams.neg_2log,
 405                                                         out);
 406                         break;
 407                     default:
 408                         GMX_RELEASE_ASSERT(false, "Unsupported j-unroll size");
 409                 }
 410             }
 411         }
 412     }
 413
 414     if (forceFlags & GMX_FORCE_ENERGY)
 415     {
 416         reduce_energies_over_lists(nbat, nnbl, vVdw, vCoulomb);
 417     }
 418 }
 419
 420 static void accountFlops(t_nrnb                           *nrnb,
 421                          const nonbonded_verlet_t         &nbv,
 422                          const Nbnxm::InteractionLocality  iLocality,
 423                          const interaction_const_t        &ic,
 424                          const int                         forceFlags)
 425 {
 426     const nbnxn_pairlist_set_t &pairlistSet     = nbv.pairlistSets[iLocality];
 427     const bool                  usingGpuKernels = nbv.useGpu();
 428
 429     int enr_nbnxn_kernel_ljc;
 430     if (EEL_RF(ic.eeltype) || ic.eeltype == eelCUT)
 431     {
 432         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
 433     }
 434     else if ((!usingGpuKernels && nbv.ewaldExclusionType_ == ewaldexclAnalytical) ||
 435              (usingGpuKernels && Nbnxm::gpu_is_kernel_ewald_analytical(nbv.gpu_nbv)))
 436     {
 437         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
 438     }
 439     else
 440     {
 441         enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
 442     }
 443     int enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
 444     if (forceFlags & GMX_FORCE_ENERGY)
 445     {
 446         /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
 447         enr_nbnxn_kernel_ljc += 1;
 448         enr_nbnxn_kernel_lj  += 1;
 449     }
 450
 451     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc,
 452              pairlistSet.natpair_ljq);
 453     inc_nrnb(nrnb, enr_nbnxn_kernel_lj,
 454              pairlistSet.natpair_lj);
 455     /* The Coulomb-only kernels are offset -eNR_NBNXN_LJ_RF+eNR_NBNXN_RF */
 456     inc_nrnb(nrnb, enr_nbnxn_kernel_ljc-eNR_NBNXN_LJ_RF+eNR_NBNXN_RF,
 457              pairlistSet.natpair_q);
 458
 459     const bool calcEnergy = ((forceFlags & GMX_FORCE_ENERGY) != 0);
 460     if (ic.vdw_modifier == eintmodFORCESWITCH)
 461     {
 462         /* We add up the switch cost separately */
 463         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_FSW + (calcEnergy ? 1 : 0),
 464                  pairlistSet.natpair_ljq + pairlistSet.natpair_lj);
 465     }
 466     if (ic.vdw_modifier == eintmodPOTSWITCH)
 467     {
 468         /* We add up the switch cost separately */
 469         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_PSW + (calcEnergy ? 1 : 0),
 470                  pairlistSet.natpair_ljq + pairlistSet.natpair_lj);
 471     }
 472     if (ic.vdwtype == evdwPME)
 473     {
 474         /* We add up the LJ Ewald cost separately */
 475         inc_nrnb(nrnb, eNR_NBNXN_ADD_LJ_EWALD + (calcEnergy ? 1 : 0),
 476                  pairlistSet.natpair_ljq + pairlistSet.natpair_lj);
 477     }
 478 }
 479
 480 void NbnxnDispatchKernel(nonbonded_verlet_t        *nbv,
 481                          Nbnxm::InteractionLocality iLocality,
 482                          const interaction_const_t &ic,
 483                          int                        forceFlags,
 484                          int                        clearF,
 485                          t_forcerec                *fr,
 486                          gmx_enerdata_t            *enerd,
 487                          t_nrnb                    *nrnb)
 488 {
 489     const nbnxn_pairlist_set_t &pairlistSet = nbv->pairlistSets[iLocality];
 490
 491     switch (nbv->kernelType_)
 492     {
 493         case nbnxnk4x4_PlainC:
 494         case nbnxnk4xN_SIMD_4xN:
 495         case nbnxnk4xN_SIMD_2xNN:
 496             nbnxn_kernel_cpu(pairlistSet,
 497                              nbv->kernelType_,
 498                              nbv->ewaldExclusionType_,
 499                              nbv->nbat,
 500                              ic,
 501                              fr->shift_vec,
 502                              forceFlags,
 503                              clearF,
 504                              fr->fshift[0],
 505                              enerd->grpp.ener[egCOULSR],
 506                              fr->bBHAM ?
 507                              enerd->grpp.ener[egBHAMSR] :
 508                              enerd->grpp.ener[egLJSR]);
 509             break;
 510
 511         case nbnxnk8x8x8_GPU:
 512             Nbnxm::gpu_launch_kernel(nbv->gpu_nbv, forceFlags, iLocality);
 513             break;
 514
 515         case nbnxnk8x8x8_PlainC:
 516             nbnxn_kernel_gpu_ref(pairlistSet.nblGpu[0],
 517                                  nbv->nbat, &ic,
 518                                  fr->shift_vec,
 519                                  forceFlags,
 520                                  clearF,
 521                                  nbv->nbat->out[0].f,
 522                                  fr->fshift[0],
 523                                  enerd->grpp.ener[egCOULSR],
 524                                  fr->bBHAM ?
 525                                  enerd->grpp.ener[egBHAMSR] :
 526                                  enerd->grpp.ener[egLJSR]);
 527             break;
 528
 529         default:
 530             GMX_RELEASE_ASSERT(false, "Invalid nonbonded kernel type passed!");
 531
 532     }
 533
 534     accountFlops(nrnb, *nbv, iLocality, ic, forceFlags);
 535 }