src/gromacs/mdlib/nbnxn_kernels/simd_2xnn/nbnxn_kernel_simd_2xnn.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 2xnn.
  38  */
  39
  40 #ifdef HAVE_CONFIG_H
  41 #include <config.h>
  42 #endif
  43
  44 #include "typedefs.h"
  45
  46 #ifdef GMX_NBNXN_SIMD_2XNN
  47
  48 /* Include the full-width SIMD macros */
  49
  50 #include "gromacs/simd/macros.h"
  51 #include "gromacs/simd/vector_operations.h"
  52
  53 #if !(GMX_SIMD_REAL_WIDTH == 8 || GMX_SIMD_REAL_WIDTH == 16)
  54 #error "unsupported SIMD width"
  55 #endif
  56
  57 #define GMX_SIMD_J_UNROLL_SIZE 2
  58 #include "nbnxn_kernel_simd_2xnn.h"
  59 #include "../nbnxn_kernel_common.h"
  60 #include "gmx_omp_nthreads.h"
  61 #include "types/force_flags.h"
  62 #include "gmx_fatal.h"
  63
  64 /*! \brief Kinds of electrostatic treatments in SIMD Verlet kernels
  65  */
  66 enum {
  67     coultRF, coultTAB, coultTAB_TWIN, coultEWALD, coultEWALD_TWIN, coultNR
  68 };
  69
  70 /* Declare and define the kernel function pointer lookup tables.
  71  * The minor index of the array goes over both the LJ combination rules,
  72  * which is only supported by plain cut-off, and the LJ switch functions.
  73  */
  74 static p_nbk_func_noener p_nbk_noener[coultNR][ljcrNR+2] =
  75 {
  76     {
  77         nbnxn_kernel_ElecRF_VdwLJCombGeom_F_2xnn,
  78         nbnxn_kernel_ElecRF_VdwLJCombLB_F_2xnn,
  79         nbnxn_kernel_ElecRF_VdwLJ_F_2xnn,
  80         nbnxn_kernel_ElecRF_VdwLJFSw_F_2xnn,
  81         nbnxn_kernel_ElecRF_VdwLJPSw_F_2xnn,
  82     },
  83     {
  84         nbnxn_kernel_ElecQSTab_VdwLJCombGeom_F_2xnn,
  85         nbnxn_kernel_ElecQSTab_VdwLJCombLB_F_2xnn,
  86         nbnxn_kernel_ElecQSTab_VdwLJ_F_2xnn,
  87         nbnxn_kernel_ElecQSTab_VdwLJFSw_F_2xnn,
  88         nbnxn_kernel_ElecQSTab_VdwLJPSw_F_2xnn,
  89     },
  90     {
  91         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombGeom_F_2xnn,
  92         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombLB_F_2xnn,
  93         nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_F_2xnn,
  94         nbnxn_kernel_ElecQSTabTwinCut_VdwLJFSw_F_2xnn,
  95         nbnxn_kernel_ElecQSTabTwinCut_VdwLJPSw_F_2xnn,
  96     },
  97     {
  98         nbnxn_kernel_ElecEw_VdwLJCombGeom_F_2xnn,
  99         nbnxn_kernel_ElecEw_VdwLJCombLB_F_2xnn,
 100         nbnxn_kernel_ElecEw_VdwLJ_F_2xnn,
 101         nbnxn_kernel_ElecEw_VdwLJFSw_F_2xnn,
 102         nbnxn_kernel_ElecEw_VdwLJPSw_F_2xnn,
 103     },
 104     {
 105         nbnxn_kernel_ElecEwTwinCut_VdwLJCombGeom_F_2xnn,
 106         nbnxn_kernel_ElecEwTwinCut_VdwLJCombLB_F_2xnn,
 107         nbnxn_kernel_ElecEwTwinCut_VdwLJ_F_2xnn,
 108         nbnxn_kernel_ElecEwTwinCut_VdwLJFSw_F_2xnn,
 109         nbnxn_kernel_ElecEwTwinCut_VdwLJPSw_F_2xnn,
 110     },
 111 };
 112
 113 static p_nbk_func_ener p_nbk_ener[coultNR][ljcrNR+2] =
 114 {
 115     {
 116         nbnxn_kernel_ElecRF_VdwLJCombGeom_VF_2xnn,
 117         nbnxn_kernel_ElecRF_VdwLJCombLB_VF_2xnn,
 118         nbnxn_kernel_ElecRF_VdwLJ_VF_2xnn,
 119         nbnxn_kernel_ElecRF_VdwLJFSw_VF_2xnn,
 120         nbnxn_kernel_ElecRF_VdwLJPSw_VF_2xnn,
 121     },
 122     {
 123         nbnxn_kernel_ElecQSTab_VdwLJCombGeom_VF_2xnn,
 124         nbnxn_kernel_ElecQSTab_VdwLJCombLB_VF_2xnn,
 125         nbnxn_kernel_ElecQSTab_VdwLJ_VF_2xnn,
 126         nbnxn_kernel_ElecQSTab_VdwLJFSw_VF_2xnn,
 127         nbnxn_kernel_ElecQSTab_VdwLJPSw_VF_2xnn,
 128     },
 129     {
 130         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombGeom_VF_2xnn,
 131         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombLB_VF_2xnn,
 132         nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_VF_2xnn,
 133         nbnxn_kernel_ElecQSTabTwinCut_VdwLJFSw_VF_2xnn,
 134         nbnxn_kernel_ElecQSTabTwinCut_VdwLJPSw_VF_2xnn,
 135     },
 136     {
 137         nbnxn_kernel_ElecEw_VdwLJCombGeom_VF_2xnn,
 138         nbnxn_kernel_ElecEw_VdwLJCombLB_VF_2xnn,
 139         nbnxn_kernel_ElecEw_VdwLJ_VF_2xnn,
 140         nbnxn_kernel_ElecEw_VdwLJFSw_VF_2xnn,
 141         nbnxn_kernel_ElecEw_VdwLJPSw_VF_2xnn,
 142     },
 143     {
 144         nbnxn_kernel_ElecEwTwinCut_VdwLJCombGeom_VF_2xnn,
 145         nbnxn_kernel_ElecEwTwinCut_VdwLJCombLB_VF_2xnn,
 146         nbnxn_kernel_ElecEwTwinCut_VdwLJ_VF_2xnn,
 147         nbnxn_kernel_ElecEwTwinCut_VdwLJFSw_VF_2xnn,
 148         nbnxn_kernel_ElecEwTwinCut_VdwLJPSw_VF_2xnn,
 149     },
 150 };
 151
 152 static p_nbk_func_ener p_nbk_energrp[coultNR][ljcrNR+2] =
 153 {
 154     {
 155         nbnxn_kernel_ElecRF_VdwLJCombGeom_VgrpF_2xnn,
 156         nbnxn_kernel_ElecRF_VdwLJCombLB_VgrpF_2xnn,
 157         nbnxn_kernel_ElecRF_VdwLJ_VgrpF_2xnn,
 158         nbnxn_kernel_ElecRF_VdwLJFSw_VgrpF_2xnn,
 159         nbnxn_kernel_ElecRF_VdwLJPSw_VgrpF_2xnn,
 160     },
 161     {
 162         nbnxn_kernel_ElecQSTab_VdwLJCombGeom_VgrpF_2xnn,
 163         nbnxn_kernel_ElecQSTab_VdwLJCombLB_VgrpF_2xnn,
 164         nbnxn_kernel_ElecQSTab_VdwLJ_VgrpF_2xnn,
 165         nbnxn_kernel_ElecQSTab_VdwLJFSw_VgrpF_2xnn,
 166         nbnxn_kernel_ElecQSTab_VdwLJPSw_VgrpF_2xnn,
 167     },
 168     {
 169         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombGeom_VgrpF_2xnn,
 170         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombLB_VgrpF_2xnn,
 171         nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_VgrpF_2xnn,
 172         nbnxn_kernel_ElecQSTabTwinCut_VdwLJFSw_VgrpF_2xnn,
 173         nbnxn_kernel_ElecQSTabTwinCut_VdwLJPSw_VgrpF_2xnn,
 174     },
 175     {
 176         nbnxn_kernel_ElecEw_VdwLJCombGeom_VgrpF_2xnn,
 177         nbnxn_kernel_ElecEw_VdwLJCombLB_VgrpF_2xnn,
 178         nbnxn_kernel_ElecEw_VdwLJ_VgrpF_2xnn,
 179         nbnxn_kernel_ElecEw_VdwLJFSw_VgrpF_2xnn,
 180         nbnxn_kernel_ElecEw_VdwLJPSw_VgrpF_2xnn,
 181     },
 182     {
 183         nbnxn_kernel_ElecEwTwinCut_VdwLJCombGeom_VgrpF_2xnn,
 184         nbnxn_kernel_ElecEwTwinCut_VdwLJCombLB_VgrpF_2xnn,
 185         nbnxn_kernel_ElecEwTwinCut_VdwLJ_VgrpF_2xnn,
 186         nbnxn_kernel_ElecEwTwinCut_VdwLJFSw_VgrpF_2xnn,
 187         nbnxn_kernel_ElecEwTwinCut_VdwLJPSw_VgrpF_2xnn,
 188     },
 189 };
 190
 191
 192 static void
 193 reduce_group_energies(int ng, int ng_2log,
 194                       const real *VSvdw, const real *VSc,
 195                       real *Vvdw, real *Vc)
 196 {
 197     const int unrollj      = GMX_SIMD_REAL_WIDTH/GMX_SIMD_J_UNROLL_SIZE;
 198     const int unrollj_half = unrollj/2;
 199     int       ng_p2, i, j, j0, j1, c, s;
 200
 201     ng_p2 = (1<<ng_2log);
 202
 203     /* The size of the x86 SIMD energy group buffer array is:
 204      * ng*ng*ng_p2*unrollj_half*simd_width
 205      */
 206     for (i = 0; i < ng; i++)
 207     {
 208         for (j = 0; j < ng; j++)
 209         {
 210             Vvdw[i*ng+j] = 0;
 211             Vc[i*ng+j]   = 0;
 212         }
 213
 214         for (j1 = 0; j1 < ng; j1++)
 215         {
 216             for (j0 = 0; j0 < ng; j0++)
 217             {
 218                 c = ((i*ng + j1)*ng_p2 + j0)*unrollj_half*unrollj;
 219                 for (s = 0; s < unrollj_half; s++)
 220                 {
 221                     Vvdw[i*ng+j0] += VSvdw[c+0];
 222                     Vvdw[i*ng+j1] += VSvdw[c+1];
 223                     Vc  [i*ng+j0] += VSc  [c+0];
 224                     Vc  [i*ng+j1] += VSc  [c+1];
 225                     c             += unrollj + 2;
 226                 }
 227             }
 228         }
 229     }
 230 }
 231
 232 #else /* GMX_NBNXN_SIMD_2XNN */
 233
 234 #include "gmx_fatal.h"
 235
 236 #endif /* GMX_NBNXN_SIMD_2XNN */
 237
 238 void
 239 nbnxn_kernel_simd_2xnn(nbnxn_pairlist_set_t      gmx_unused *nbl_list,
 240                        const nbnxn_atomdata_t    gmx_unused *nbat,
 241                        const interaction_const_t gmx_unused *ic,
 242                        int                       gmx_unused  ewald_excl,
 243                        rvec                      gmx_unused *shift_vec,
 244                        int                       gmx_unused  force_flags,
 245                        int                       gmx_unused  clearF,
 246                        real                      gmx_unused *fshift,
 247                        real                      gmx_unused *Vc,
 248                        real                      gmx_unused *Vvdw)
 249 #ifdef GMX_NBNXN_SIMD_2XNN
 250 {
 251     int                nnbl;
 252     nbnxn_pairlist_t **nbl;
 253     int                coult, ljtreatment = 0;
 254     int                nb;
 255
 256     nnbl = nbl_list->nnbl;
 257     nbl  = nbl_list->nbl;
 258
 259     if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
 260     {
 261         coult = coultRF;
 262     }
 263     else
 264     {
 265         if (ewald_excl == ewaldexclTable)
 266         {
 267             if (ic->rcoulomb == ic->rvdw)
 268             {
 269                 coult = coultTAB;
 270             }
 271             else
 272             {
 273                 coult = coultTAB_TWIN;
 274             }
 275         }
 276         else
 277         {
 278             if (ic->rcoulomb == ic->rvdw)
 279             {
 280                 coult = coultEWALD;
 281             }
 282             else
 283             {
 284                 coult = coultEWALD_TWIN;
 285             }
 286         }
 287     }
 288
 289     switch (ic->vdw_modifier)
 290     {
 291         case eintmodNONE:
 292         case eintmodPOTSHIFT:
 293             ljtreatment = nbat->comb_rule;
 294             break;
 295         /* Switch functions follow after cut-off combination rule kernels */
 296         case eintmodFORCESWITCH:
 297             ljtreatment = ljcrNR;
 298             break;
 299         case eintmodPOTSWITCH:
 300             ljtreatment = ljcrNR + 1;
 301             break;
 302         default:
 303             gmx_incons("Unsupported VdW interaction modifier");
 304     }
 305
 306 #pragma omp parallel for schedule(static) num_threads(gmx_omp_nthreads_get(emntNonbonded))
 307     for (nb = 0; nb < nnbl; nb++)
 308     {
 309         nbnxn_atomdata_output_t *out;
 310         real                    *fshift_p;
 311
 312         out = &nbat->out[nb];
 313
 314         if (clearF == enbvClearFYes)
 315         {
 316             clear_f(nbat, nb, out->f);
 317         }
 318
 319         if ((force_flags & GMX_FORCE_VIRIAL) && nnbl == 1)
 320         {
 321             fshift_p = fshift;
 322         }
 323         else
 324         {
 325             fshift_p = out->fshift;
 326
 327             if (clearF == enbvClearFYes)
 328             {
 329                 clear_fshift(fshift_p);
 330             }
 331         }
 332
 333         if (!(force_flags & GMX_FORCE_ENERGY))
 334         {
 335             /* Don't calculate energies */
 336             p_nbk_noener[coult][ljtreatment](nbl[nb], nbat,
 337                                              ic,
 338                                              shift_vec,
 339                                              out->f,
 340                                              fshift_p);
 341         }
 342         else if (out->nV == 1)
 343         {
 344             /* No energy groups */
 345             out->Vvdw[0] = 0;
 346             out->Vc[0]   = 0;
 347
 348             p_nbk_ener[coult][ljtreatment](nbl[nb], nbat,
 349                                            ic,
 350                                            shift_vec,
 351                                            out->f,
 352                                            fshift_p,
 353                                            out->Vvdw,
 354                                            out->Vc);
 355         }
 356         else
 357         {
 358             /* Calculate energy group contributions */
 359             int i;
 360
 361             for (i = 0; i < out->nVS; i++)
 362             {
 363                 out->VSvdw[i] = 0;
 364             }
 365             for (i = 0; i < out->nVS; i++)
 366             {
 367                 out->VSc[i] = 0;
 368             }
 369
 370             p_nbk_energrp[coult][ljtreatment](nbl[nb], nbat,
 371                                               ic,
 372                                               shift_vec,
 373                                               out->f,
 374                                               fshift_p,
 375                                               out->VSvdw,
 376                                               out->VSc);
 377
 378             reduce_group_energies(nbat->nenergrp, nbat->neg_2log,
 379                                   out->VSvdw, out->VSc,
 380                                   out->Vvdw, out->Vc);
 381         }
 382     }
 383
 384     if (force_flags & GMX_FORCE_ENERGY)
 385     {
 386         reduce_energies_over_lists(nbat, nnbl, Vvdw, Vc);
 387     }
 388 }
 389 #else
 390 {
 391     gmx_incons("nbnxn_kernel_simd_2xnn called when such kernels "
 392                " are not enabled.");
 393 }
 394 #endif
 395 #undef GMX_SIMD_J_UNROLL_SIZE