src/gromacs/mdlib/nbnxn_kernels/simd_4xn/nbnxn_kernel_simd_4xn.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, 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  * Note: this file was generated by the Verlet kernel generator for
  37  * kernel type 4xn.
  38  */
  39
  40 #include "gmxpre.h"
  41
  42 #include "config.h"
  43
  44 #include "gromacs/math/vectypes.h"
  45 #include "gromacs/mdlib/nb_verlet.h"
  46 #include "gromacs/mdlib/nbnxn_simd.h"
  47 #include "gromacs/mdtypes/interaction_const.h"
  48 #include "gromacs/mdtypes/md_enums.h"
  49
  50 #ifdef GMX_NBNXN_SIMD_4XN
  51
  52 #include "gromacs/simd/vector_operations.h"
  53
  54 #if !(GMX_SIMD_REAL_WIDTH == 2 || GMX_SIMD_REAL_WIDTH == 4 || GMX_SIMD_REAL_WIDTH == 8)
  55 #error "unsupported SIMD width"
  56 #endif
  57
  58 #define GMX_SIMD_J_UNROLL_SIZE 1
  59 #include "nbnxn_kernel_simd_4xn.h"
  60
  61 #include "gromacs/mdlib/force_flags.h"
  62 #include "gromacs/mdlib/gmx_omp_nthreads.h"
  63 #include "gromacs/mdlib/nbnxn_kernels/nbnxn_kernel_common.h"
  64 #include "gromacs/simd/simd.h"
  65 #include "gromacs/utility/fatalerror.h"
  66 #include "gromacs/utility/real.h"
  67
  68 /*! \brief Kinds of electrostatic treatments in SIMD Verlet kernels
  69  */
  70 enum {
  71     coulktRF, coulktTAB, coulktTAB_TWIN, coulktEWALD, coulktEWALD_TWIN, coulktNR
  72 };
  73
  74 /*! \brief Kinds of Van der Waals treatments in SIMD Verlet kernels
  75  */
  76 enum {
  77     vdwktLJCUT_COMBGEOM, vdwktLJCUT_COMBLB, vdwktLJCUT_COMBNONE, vdwktLJFORCESWITCH, vdwktLJPOTSWITCH, vdwktLJEWALDCOMBGEOM, vdwktNR
  78 };
  79
  80 /* Declare and define the kernel function pointer lookup tables.
  81  * The minor index of the array goes over both the LJ combination rules,
  82  * which is only supported by plain cut-off, and the LJ switch/PME functions.
  83  */
  84 static p_nbk_func_noener p_nbk_noener[coulktNR][vdwktNR] =
  85 {
  86     {
  87         nbnxn_kernel_ElecRF_VdwLJCombGeom_F_4xn,
  88         nbnxn_kernel_ElecRF_VdwLJCombLB_F_4xn,
  89         nbnxn_kernel_ElecRF_VdwLJ_F_4xn,
  90         nbnxn_kernel_ElecRF_VdwLJFSw_F_4xn,
  91         nbnxn_kernel_ElecRF_VdwLJPSw_F_4xn,
  92         nbnxn_kernel_ElecRF_VdwLJEwCombGeom_F_4xn,
  93     },
  94     {
  95         nbnxn_kernel_ElecQSTab_VdwLJCombGeom_F_4xn,
  96         nbnxn_kernel_ElecQSTab_VdwLJCombLB_F_4xn,
  97         nbnxn_kernel_ElecQSTab_VdwLJ_F_4xn,
  98         nbnxn_kernel_ElecQSTab_VdwLJFSw_F_4xn,
  99         nbnxn_kernel_ElecQSTab_VdwLJPSw_F_4xn,
 100         nbnxn_kernel_ElecQSTab_VdwLJEwCombGeom_F_4xn,
 101     },
 102     {
 103         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombGeom_F_4xn,
 104         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombLB_F_4xn,
 105         nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_F_4xn,
 106         nbnxn_kernel_ElecQSTabTwinCut_VdwLJFSw_F_4xn,
 107         nbnxn_kernel_ElecQSTabTwinCut_VdwLJPSw_F_4xn,
 108         nbnxn_kernel_ElecQSTabTwinCut_VdwLJEwCombGeom_F_4xn,
 109     },
 110     {
 111         nbnxn_kernel_ElecEw_VdwLJCombGeom_F_4xn,
 112         nbnxn_kernel_ElecEw_VdwLJCombLB_F_4xn,
 113         nbnxn_kernel_ElecEw_VdwLJ_F_4xn,
 114         nbnxn_kernel_ElecEw_VdwLJFSw_F_4xn,
 115         nbnxn_kernel_ElecEw_VdwLJPSw_F_4xn,
 116         nbnxn_kernel_ElecEw_VdwLJEwCombGeom_F_4xn,
 117     },
 118     {
 119         nbnxn_kernel_ElecEwTwinCut_VdwLJCombGeom_F_4xn,
 120         nbnxn_kernel_ElecEwTwinCut_VdwLJCombLB_F_4xn,
 121         nbnxn_kernel_ElecEwTwinCut_VdwLJ_F_4xn,
 122         nbnxn_kernel_ElecEwTwinCut_VdwLJFSw_F_4xn,
 123         nbnxn_kernel_ElecEwTwinCut_VdwLJPSw_F_4xn,
 124         nbnxn_kernel_ElecEwTwinCut_VdwLJEwCombGeom_F_4xn,
 125     },
 126 };
 127
 128 static p_nbk_func_ener p_nbk_ener[coulktNR][vdwktNR] =
 129 {
 130     {
 131         nbnxn_kernel_ElecRF_VdwLJCombGeom_VF_4xn,
 132         nbnxn_kernel_ElecRF_VdwLJCombLB_VF_4xn,
 133         nbnxn_kernel_ElecRF_VdwLJ_VF_4xn,
 134         nbnxn_kernel_ElecRF_VdwLJFSw_VF_4xn,
 135         nbnxn_kernel_ElecRF_VdwLJPSw_VF_4xn,
 136         nbnxn_kernel_ElecRF_VdwLJEwCombGeom_VF_4xn,
 137     },
 138     {
 139         nbnxn_kernel_ElecQSTab_VdwLJCombGeom_VF_4xn,
 140         nbnxn_kernel_ElecQSTab_VdwLJCombLB_VF_4xn,
 141         nbnxn_kernel_ElecQSTab_VdwLJ_VF_4xn,
 142         nbnxn_kernel_ElecQSTab_VdwLJFSw_VF_4xn,
 143         nbnxn_kernel_ElecQSTab_VdwLJPSw_VF_4xn,
 144         nbnxn_kernel_ElecQSTab_VdwLJEwCombGeom_VF_4xn,
 145     },
 146     {
 147         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombGeom_VF_4xn,
 148         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombLB_VF_4xn,
 149         nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_VF_4xn,
 150         nbnxn_kernel_ElecQSTabTwinCut_VdwLJFSw_VF_4xn,
 151         nbnxn_kernel_ElecQSTabTwinCut_VdwLJPSw_VF_4xn,
 152         nbnxn_kernel_ElecQSTabTwinCut_VdwLJEwCombGeom_VF_4xn,
 153     },
 154     {
 155         nbnxn_kernel_ElecEw_VdwLJCombGeom_VF_4xn,
 156         nbnxn_kernel_ElecEw_VdwLJCombLB_VF_4xn,
 157         nbnxn_kernel_ElecEw_VdwLJ_VF_4xn,
 158         nbnxn_kernel_ElecEw_VdwLJFSw_VF_4xn,
 159         nbnxn_kernel_ElecEw_VdwLJPSw_VF_4xn,
 160         nbnxn_kernel_ElecEw_VdwLJEwCombGeom_VF_4xn,
 161     },
 162     {
 163         nbnxn_kernel_ElecEwTwinCut_VdwLJCombGeom_VF_4xn,
 164         nbnxn_kernel_ElecEwTwinCut_VdwLJCombLB_VF_4xn,
 165         nbnxn_kernel_ElecEwTwinCut_VdwLJ_VF_4xn,
 166         nbnxn_kernel_ElecEwTwinCut_VdwLJFSw_VF_4xn,
 167         nbnxn_kernel_ElecEwTwinCut_VdwLJPSw_VF_4xn,
 168         nbnxn_kernel_ElecEwTwinCut_VdwLJEwCombGeom_VF_4xn,
 169     },
 170 };
 171
 172 static p_nbk_func_ener p_nbk_energrp[coulktNR][vdwktNR] =
 173 {
 174     {
 175         nbnxn_kernel_ElecRF_VdwLJCombGeom_VgrpF_4xn,
 176         nbnxn_kernel_ElecRF_VdwLJCombLB_VgrpF_4xn,
 177         nbnxn_kernel_ElecRF_VdwLJ_VgrpF_4xn,
 178         nbnxn_kernel_ElecRF_VdwLJFSw_VgrpF_4xn,
 179         nbnxn_kernel_ElecRF_VdwLJPSw_VgrpF_4xn,
 180         nbnxn_kernel_ElecRF_VdwLJEwCombGeom_VgrpF_4xn,
 181     },
 182     {
 183         nbnxn_kernel_ElecQSTab_VdwLJCombGeom_VgrpF_4xn,
 184         nbnxn_kernel_ElecQSTab_VdwLJCombLB_VgrpF_4xn,
 185         nbnxn_kernel_ElecQSTab_VdwLJ_VgrpF_4xn,
 186         nbnxn_kernel_ElecQSTab_VdwLJFSw_VgrpF_4xn,
 187         nbnxn_kernel_ElecQSTab_VdwLJPSw_VgrpF_4xn,
 188         nbnxn_kernel_ElecQSTab_VdwLJEwCombGeom_VgrpF_4xn,
 189     },
 190     {
 191         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombGeom_VgrpF_4xn,
 192         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombLB_VgrpF_4xn,
 193         nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_VgrpF_4xn,
 194         nbnxn_kernel_ElecQSTabTwinCut_VdwLJFSw_VgrpF_4xn,
 195         nbnxn_kernel_ElecQSTabTwinCut_VdwLJPSw_VgrpF_4xn,
 196         nbnxn_kernel_ElecQSTabTwinCut_VdwLJEwCombGeom_VgrpF_4xn,
 197     },
 198     {
 199         nbnxn_kernel_ElecEw_VdwLJCombGeom_VgrpF_4xn,
 200         nbnxn_kernel_ElecEw_VdwLJCombLB_VgrpF_4xn,
 201         nbnxn_kernel_ElecEw_VdwLJ_VgrpF_4xn,
 202         nbnxn_kernel_ElecEw_VdwLJFSw_VgrpF_4xn,
 203         nbnxn_kernel_ElecEw_VdwLJPSw_VgrpF_4xn,
 204         nbnxn_kernel_ElecEw_VdwLJEwCombGeom_VgrpF_4xn,
 205     },
 206     {
 207         nbnxn_kernel_ElecEwTwinCut_VdwLJCombGeom_VgrpF_4xn,
 208         nbnxn_kernel_ElecEwTwinCut_VdwLJCombLB_VgrpF_4xn,
 209         nbnxn_kernel_ElecEwTwinCut_VdwLJ_VgrpF_4xn,
 210         nbnxn_kernel_ElecEwTwinCut_VdwLJFSw_VgrpF_4xn,
 211         nbnxn_kernel_ElecEwTwinCut_VdwLJPSw_VgrpF_4xn,
 212         nbnxn_kernel_ElecEwTwinCut_VdwLJEwCombGeom_VgrpF_4xn,
 213     },
 214 };
 215
 216
 217 static void
 218 reduce_group_energies(int ng, int ng_2log,
 219                       const real *VSvdw, const real *VSc,
 220                       real *Vvdw, real *Vc)
 221 {
 222     const int unrollj      = GMX_SIMD_REAL_WIDTH/GMX_SIMD_J_UNROLL_SIZE;
 223     const int unrollj_half = unrollj/2;
 224     int       ng_p2, i, j, j0, j1, c, s;
 225
 226     ng_p2 = (1<<ng_2log);
 227
 228     /* The size of the x86 SIMD energy group buffer array is:
 229      * ng*ng*ng_p2*unrollj_half*simd_width
 230      */
 231     for (i = 0; i < ng; i++)
 232     {
 233         for (j = 0; j < ng; j++)
 234         {
 235             Vvdw[i*ng+j] = 0;
 236             Vc[i*ng+j]   = 0;
 237         }
 238
 239         for (j1 = 0; j1 < ng; j1++)
 240         {
 241             for (j0 = 0; j0 < ng; j0++)
 242             {
 243                 c = ((i*ng + j1)*ng_p2 + j0)*unrollj_half*unrollj;
 244                 for (s = 0; s < unrollj_half; s++)
 245                 {
 246                     Vvdw[i*ng+j0] += VSvdw[c+0];
 247                     Vvdw[i*ng+j1] += VSvdw[c+1];
 248                     Vc  [i*ng+j0] += VSc  [c+0];
 249                     Vc  [i*ng+j1] += VSc  [c+1];
 250                     c             += unrollj + 2;
 251                 }
 252             }
 253         }
 254     }
 255 }
 256
 257 #else /* GMX_NBNXN_SIMD_4XN */
 258
 259 #include "gromacs/utility/fatalerror.h"
 260
 261 #endif /* GMX_NBNXN_SIMD_4XN */
 262
 263 void
 264 nbnxn_kernel_simd_4xn(nbnxn_pairlist_set_t      gmx_unused *nbl_list,
 265                       const nbnxn_atomdata_t    gmx_unused *nbat,
 266                       const interaction_const_t gmx_unused *ic,
 267                       int                       gmx_unused  ewald_excl,
 268                       rvec                      gmx_unused *shift_vec,
 269                       int                       gmx_unused  force_flags,
 270                       int                       gmx_unused  clearF,
 271                       real                      gmx_unused *fshift,
 272                       real                      gmx_unused *Vc,
 273                       real                      gmx_unused *Vvdw)
 274 #ifdef GMX_NBNXN_SIMD_4XN
 275 {
 276     int                nnbl;
 277     nbnxn_pairlist_t **nbl;
 278     int                coulkt, vdwkt = 0;
 279     int                nb, nthreads;
 280
 281     nnbl = nbl_list->nnbl;
 282     nbl  = nbl_list->nbl;
 283
 284     if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
 285     {
 286         coulkt = coulktRF;
 287     }
 288     else
 289     {
 290         if (ewald_excl == ewaldexclTable)
 291         {
 292             if (ic->rcoulomb == ic->rvdw)
 293             {
 294                 coulkt = coulktTAB;
 295             }
 296             else
 297             {
 298                 coulkt = coulktTAB_TWIN;
 299             }
 300         }
 301         else
 302         {
 303             if (ic->rcoulomb == ic->rvdw)
 304             {
 305                 coulkt = coulktEWALD;
 306             }
 307             else
 308             {
 309                 coulkt = coulktEWALD_TWIN;
 310             }
 311         }
 312     }
 313
 314     if (ic->vdwtype == evdwCUT)
 315     {
 316         switch (ic->vdw_modifier)
 317         {
 318             case eintmodNONE:
 319             case eintmodPOTSHIFT:
 320                 switch (nbat->comb_rule)
 321                 {
 322                     case ljcrGEOM: vdwkt = vdwktLJCUT_COMBGEOM; break;
 323                     case ljcrLB:   vdwkt = vdwktLJCUT_COMBLB;   break;
 324                     case ljcrNONE: vdwkt = vdwktLJCUT_COMBNONE; break;
 325                     default:       gmx_incons("Unknown combination rule");
 326                 }
 327                 break;
 328             case eintmodFORCESWITCH:
 329                 vdwkt = vdwktLJFORCESWITCH;
 330                 break;
 331             case eintmodPOTSWITCH:
 332                 vdwkt = vdwktLJPOTSWITCH;
 333                 break;
 334             default:
 335                 gmx_incons("Unsupported VdW interaction modifier");
 336         }
 337     }
 338     else if (ic->vdwtype == evdwPME)
 339     {
 340         if (ic->ljpme_comb_rule == eljpmeLB)
 341         {
 342             gmx_incons("The nbnxn SIMD kernels don't support LJ-PME with LB");
 343         }
 344         vdwkt = vdwktLJEWALDCOMBGEOM;
 345     }
 346     else
 347     {
 348         gmx_incons("Unsupported VdW interaction type");
 349     }
 350     // cppcheck-suppress unreadVariable
 351     nthreads = gmx_omp_nthreads_get(emntNonbonded);
 352 #pragma omp parallel for schedule(static) num_threads(nthreads)
 353     for (nb = 0; nb < nnbl; nb++)
 354     {
 355         // Presently, the kernels do not call C++ code that can throw, so
 356         // no need for a try/catch pair in this OpenMP region.
 357         nbnxn_atomdata_output_t *out;
 358         real                    *fshift_p;
 359
 360         out = &nbat->out[nb];
 361
 362         if (clearF == enbvClearFYes)
 363         {
 364             clear_f(nbat, nb, out->f);
 365         }
 366
 367         if ((force_flags & GMX_FORCE_VIRIAL) && nnbl == 1)
 368         {
 369             fshift_p = fshift;
 370         }
 371         else
 372         {
 373             fshift_p = out->fshift;
 374
 375             if (clearF == enbvClearFYes)
 376             {
 377                 clear_fshift(fshift_p);
 378             }
 379         }
 380
 381         if (!(force_flags & GMX_FORCE_ENERGY))
 382         {
 383             /* Don't calculate energies */
 384             p_nbk_noener[coulkt][vdwkt](nbl[nb], nbat,
 385                                         ic,
 386                                         shift_vec,
 387                                         out->f,
 388                                         fshift_p);
 389         }
 390         else if (out->nV == 1)
 391         {
 392             /* No energy groups */
 393             out->Vvdw[0] = 0;
 394             out->Vc[0]   = 0;
 395
 396             p_nbk_ener[coulkt][vdwkt](nbl[nb], nbat,
 397                                       ic,
 398                                       shift_vec,
 399                                       out->f,
 400                                       fshift_p,
 401                                       out->Vvdw,
 402                                       out->Vc);
 403         }
 404         else
 405         {
 406             /* Calculate energy group contributions */
 407             int i;
 408
 409             for (i = 0; i < out->nVS; i++)
 410             {
 411                 out->VSvdw[i] = 0;
 412             }
 413             for (i = 0; i < out->nVS; i++)
 414             {
 415                 out->VSc[i] = 0;
 416             }
 417
 418             p_nbk_energrp[coulkt][vdwkt](nbl[nb], nbat,
 419                                          ic,
 420                                          shift_vec,
 421                                          out->f,
 422                                          fshift_p,
 423                                          out->VSvdw,
 424                                          out->VSc);
 425
 426             reduce_group_energies(nbat->nenergrp, nbat->neg_2log,
 427                                   out->VSvdw, out->VSc,
 428                                   out->Vvdw, out->Vc);
 429         }
 430     }
 431
 432     if (force_flags & GMX_FORCE_ENERGY)
 433     {
 434         reduce_energies_over_lists(nbat, nnbl, Vvdw, Vc);
 435     }
 436 }
 437 #else
 438 {
 439     gmx_incons("nbnxn_kernel_simd_4xn called when such kernels "
 440                " are not enabled.");
 441 }
 442 #endif
 443 #undef GMX_SIMD_J_UNROLL_SIZE