src/gromacs/mdlib/nbnxn_kernels/simd_4xn/nbnxn_kernel_simd_4xn.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 4xn.
  38  */
  39
  40 #ifdef HAVE_CONFIG_H
  41 #include <config.h>
  42 #endif
  43
  44 #include "typedefs.h"
  45
  46 #ifdef GMX_NBNXN_SIMD_4XN
  47
  48 #ifdef GMX_NBNXN_HALF_WIDTH_SIMD
  49 #define GMX_USE_HALF_WIDTH_SIMD_HERE
  50 #endif
  51
  52 #include "gromacs/simd/macros.h"
  53 #include "gromacs/simd/vector_operations.h"
  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 #include "../nbnxn_kernel_common.h"
  61 #include "gmx_omp_nthreads.h"
  62 #include "types/force_flags.h"
  63 #include "gmx_fatal.h"
  64
  65 /*! \brief Kinds of electrostatic treatments in SIMD Verlet kernels
  66  */
  67 enum {
  68     coultRF, coultTAB, coultTAB_TWIN, coultEWALD, coultEWALD_TWIN, coultNR
  69 };
  70
  71 /* Declare and define the kernel function pointer lookup tables.
  72  * The minor index of the array goes over both the LJ combination rules,
  73  * which is only supported by plain cut-off, and the LJ switch functions.
  74  */
  75 static p_nbk_func_noener p_nbk_noener[coultNR][ljcrNR+2] =
  76 {
  77     {
  78         nbnxn_kernel_ElecRF_VdwLJCombGeom_F_4xn,
  79         nbnxn_kernel_ElecRF_VdwLJCombLB_F_4xn,
  80         nbnxn_kernel_ElecRF_VdwLJ_F_4xn,
  81         nbnxn_kernel_ElecRF_VdwLJFSw_F_4xn,
  82         nbnxn_kernel_ElecRF_VdwLJPSw_F_4xn,
  83     },
  84     {
  85         nbnxn_kernel_ElecQSTab_VdwLJCombGeom_F_4xn,
  86         nbnxn_kernel_ElecQSTab_VdwLJCombLB_F_4xn,
  87         nbnxn_kernel_ElecQSTab_VdwLJ_F_4xn,
  88         nbnxn_kernel_ElecQSTab_VdwLJFSw_F_4xn,
  89         nbnxn_kernel_ElecQSTab_VdwLJPSw_F_4xn,
  90     },
  91     {
  92         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombGeom_F_4xn,
  93         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombLB_F_4xn,
  94         nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_F_4xn,
  95         nbnxn_kernel_ElecQSTabTwinCut_VdwLJFSw_F_4xn,
  96         nbnxn_kernel_ElecQSTabTwinCut_VdwLJPSw_F_4xn,
  97     },
  98     {
  99         nbnxn_kernel_ElecEw_VdwLJCombGeom_F_4xn,
 100         nbnxn_kernel_ElecEw_VdwLJCombLB_F_4xn,
 101         nbnxn_kernel_ElecEw_VdwLJ_F_4xn,
 102         nbnxn_kernel_ElecEw_VdwLJFSw_F_4xn,
 103         nbnxn_kernel_ElecEw_VdwLJPSw_F_4xn,
 104     },
 105     {
 106         nbnxn_kernel_ElecEwTwinCut_VdwLJCombGeom_F_4xn,
 107         nbnxn_kernel_ElecEwTwinCut_VdwLJCombLB_F_4xn,
 108         nbnxn_kernel_ElecEwTwinCut_VdwLJ_F_4xn,
 109         nbnxn_kernel_ElecEwTwinCut_VdwLJFSw_F_4xn,
 110         nbnxn_kernel_ElecEwTwinCut_VdwLJPSw_F_4xn,
 111     },
 112 };
 113
 114 static p_nbk_func_ener p_nbk_ener[coultNR][ljcrNR+2] =
 115 {
 116     {
 117         nbnxn_kernel_ElecRF_VdwLJCombGeom_VF_4xn,
 118         nbnxn_kernel_ElecRF_VdwLJCombLB_VF_4xn,
 119         nbnxn_kernel_ElecRF_VdwLJ_VF_4xn,
 120         nbnxn_kernel_ElecRF_VdwLJFSw_VF_4xn,
 121         nbnxn_kernel_ElecRF_VdwLJPSw_VF_4xn,
 122     },
 123     {
 124         nbnxn_kernel_ElecQSTab_VdwLJCombGeom_VF_4xn,
 125         nbnxn_kernel_ElecQSTab_VdwLJCombLB_VF_4xn,
 126         nbnxn_kernel_ElecQSTab_VdwLJ_VF_4xn,
 127         nbnxn_kernel_ElecQSTab_VdwLJFSw_VF_4xn,
 128         nbnxn_kernel_ElecQSTab_VdwLJPSw_VF_4xn,
 129     },
 130     {
 131         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombGeom_VF_4xn,
 132         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombLB_VF_4xn,
 133         nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_VF_4xn,
 134         nbnxn_kernel_ElecQSTabTwinCut_VdwLJFSw_VF_4xn,
 135         nbnxn_kernel_ElecQSTabTwinCut_VdwLJPSw_VF_4xn,
 136     },
 137     {
 138         nbnxn_kernel_ElecEw_VdwLJCombGeom_VF_4xn,
 139         nbnxn_kernel_ElecEw_VdwLJCombLB_VF_4xn,
 140         nbnxn_kernel_ElecEw_VdwLJ_VF_4xn,
 141         nbnxn_kernel_ElecEw_VdwLJFSw_VF_4xn,
 142         nbnxn_kernel_ElecEw_VdwLJPSw_VF_4xn,
 143     },
 144     {
 145         nbnxn_kernel_ElecEwTwinCut_VdwLJCombGeom_VF_4xn,
 146         nbnxn_kernel_ElecEwTwinCut_VdwLJCombLB_VF_4xn,
 147         nbnxn_kernel_ElecEwTwinCut_VdwLJ_VF_4xn,
 148         nbnxn_kernel_ElecEwTwinCut_VdwLJFSw_VF_4xn,
 149         nbnxn_kernel_ElecEwTwinCut_VdwLJPSw_VF_4xn,
 150     },
 151 };
 152
 153 static p_nbk_func_ener p_nbk_energrp[coultNR][ljcrNR+2] =
 154 {
 155     {
 156         nbnxn_kernel_ElecRF_VdwLJCombGeom_VgrpF_4xn,
 157         nbnxn_kernel_ElecRF_VdwLJCombLB_VgrpF_4xn,
 158         nbnxn_kernel_ElecRF_VdwLJ_VgrpF_4xn,
 159         nbnxn_kernel_ElecRF_VdwLJFSw_VgrpF_4xn,
 160         nbnxn_kernel_ElecRF_VdwLJPSw_VgrpF_4xn,
 161     },
 162     {
 163         nbnxn_kernel_ElecQSTab_VdwLJCombGeom_VgrpF_4xn,
 164         nbnxn_kernel_ElecQSTab_VdwLJCombLB_VgrpF_4xn,
 165         nbnxn_kernel_ElecQSTab_VdwLJ_VgrpF_4xn,
 166         nbnxn_kernel_ElecQSTab_VdwLJFSw_VgrpF_4xn,
 167         nbnxn_kernel_ElecQSTab_VdwLJPSw_VgrpF_4xn,
 168     },
 169     {
 170         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombGeom_VgrpF_4xn,
 171         nbnxn_kernel_ElecQSTabTwinCut_VdwLJCombLB_VgrpF_4xn,
 172         nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_VgrpF_4xn,
 173         nbnxn_kernel_ElecQSTabTwinCut_VdwLJFSw_VgrpF_4xn,
 174         nbnxn_kernel_ElecQSTabTwinCut_VdwLJPSw_VgrpF_4xn,
 175     },
 176     {
 177         nbnxn_kernel_ElecEw_VdwLJCombGeom_VgrpF_4xn,
 178         nbnxn_kernel_ElecEw_VdwLJCombLB_VgrpF_4xn,
 179         nbnxn_kernel_ElecEw_VdwLJ_VgrpF_4xn,
 180         nbnxn_kernel_ElecEw_VdwLJFSw_VgrpF_4xn,
 181         nbnxn_kernel_ElecEw_VdwLJPSw_VgrpF_4xn,
 182     },
 183     {
 184         nbnxn_kernel_ElecEwTwinCut_VdwLJCombGeom_VgrpF_4xn,
 185         nbnxn_kernel_ElecEwTwinCut_VdwLJCombLB_VgrpF_4xn,
 186         nbnxn_kernel_ElecEwTwinCut_VdwLJ_VgrpF_4xn,
 187         nbnxn_kernel_ElecEwTwinCut_VdwLJFSw_VgrpF_4xn,
 188         nbnxn_kernel_ElecEwTwinCut_VdwLJPSw_VgrpF_4xn,
 189     },
 190 };
 191
 192
 193 static void
 194 reduce_group_energies(int ng, int ng_2log,
 195                       const real *VSvdw, const real *VSc,
 196                       real *Vvdw, real *Vc)
 197 {
 198     const int unrollj      = GMX_SIMD_REAL_WIDTH/GMX_SIMD_J_UNROLL_SIZE;
 199     const int unrollj_half = unrollj/2;
 200     int       ng_p2, i, j, j0, j1, c, s;
 201
 202     ng_p2 = (1<<ng_2log);
 203
 204     /* The size of the x86 SIMD energy group buffer array is:
 205      * ng*ng*ng_p2*unrollj_half*simd_width
 206      */
 207     for (i = 0; i < ng; i++)
 208     {
 209         for (j = 0; j < ng; j++)
 210         {
 211             Vvdw[i*ng+j] = 0;
 212             Vc[i*ng+j]   = 0;
 213         }
 214
 215         for (j1 = 0; j1 < ng; j1++)
 216         {
 217             for (j0 = 0; j0 < ng; j0++)
 218             {
 219                 c = ((i*ng + j1)*ng_p2 + j0)*unrollj_half*unrollj;
 220                 for (s = 0; s < unrollj_half; s++)
 221                 {
 222                     Vvdw[i*ng+j0] += VSvdw[c+0];
 223                     Vvdw[i*ng+j1] += VSvdw[c+1];
 224                     Vc  [i*ng+j0] += VSc  [c+0];
 225                     Vc  [i*ng+j1] += VSc  [c+1];
 226                     c             += unrollj + 2;
 227                 }
 228             }
 229         }
 230     }
 231 }
 232
 233 #else /* GMX_NBNXN_SIMD_4XN */
 234
 235 #include "gmx_fatal.h"
 236
 237 #endif /* GMX_NBNXN_SIMD_4XN */
 238
 239 void
 240 nbnxn_kernel_simd_4xn(nbnxn_pairlist_set_t      gmx_unused *nbl_list,
 241                       const nbnxn_atomdata_t    gmx_unused *nbat,
 242                       const interaction_const_t gmx_unused *ic,
 243                       int                       gmx_unused  ewald_excl,
 244                       rvec                      gmx_unused *shift_vec,
 245                       int                       gmx_unused  force_flags,
 246                       int                       gmx_unused  clearF,
 247                       real                      gmx_unused *fshift,
 248                       real                      gmx_unused *Vc,
 249                       real                      gmx_unused *Vvdw)
 250 #ifdef GMX_NBNXN_SIMD_4XN
 251 {
 252     int                nnbl;
 253     nbnxn_pairlist_t **nbl;
 254     int                coult, ljtreatment = 0;
 255     int                nb;
 256
 257     nnbl = nbl_list->nnbl;
 258     nbl  = nbl_list->nbl;
 259
 260     if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
 261     {
 262         coult = coultRF;
 263     }
 264     else
 265     {
 266         if (ewald_excl == ewaldexclTable)
 267         {
 268             if (ic->rcoulomb == ic->rvdw)
 269             {
 270                 coult = coultTAB;
 271             }
 272             else
 273             {
 274                 coult = coultTAB_TWIN;
 275             }
 276         }
 277         else
 278         {
 279             if (ic->rcoulomb == ic->rvdw)
 280             {
 281                 coult = coultEWALD;
 282             }
 283             else
 284             {
 285                 coult = coultEWALD_TWIN;
 286             }
 287         }
 288     }
 289
 290     switch (ic->vdw_modifier)
 291     {
 292         case eintmodNONE:
 293         case eintmodPOTSHIFT:
 294             ljtreatment = nbat->comb_rule;
 295             break;
 296         /* Switch functions follow after cut-off combination rule kernels */
 297         case eintmodFORCESWITCH:
 298             ljtreatment = ljcrNR;
 299             break;
 300         case eintmodPOTSWITCH:
 301             ljtreatment = ljcrNR + 1;
 302             break;
 303         default:
 304             gmx_incons("Unsupported VdW interaction modifier");
 305     }
 306
 307 #pragma omp parallel for schedule(static) num_threads(gmx_omp_nthreads_get(emntNonbonded))
 308     for (nb = 0; nb < nnbl; nb++)
 309     {
 310         nbnxn_atomdata_output_t *out;
 311         real                    *fshift_p;
 312
 313         out = &nbat->out[nb];
 314
 315         if (clearF == enbvClearFYes)
 316         {
 317             clear_f(nbat, nb, out->f);
 318         }
 319
 320         if ((force_flags & GMX_FORCE_VIRIAL) && nnbl == 1)
 321         {
 322             fshift_p = fshift;
 323         }
 324         else
 325         {
 326             fshift_p = out->fshift;
 327
 328             if (clearF == enbvClearFYes)
 329             {
 330                 clear_fshift(fshift_p);
 331             }
 332         }
 333
 334         if (!(force_flags & GMX_FORCE_ENERGY))
 335         {
 336             /* Don't calculate energies */
 337             p_nbk_noener[coult][ljtreatment](nbl[nb], nbat,
 338                                              ic,
 339                                              shift_vec,
 340                                              out->f,
 341                                              fshift_p);
 342         }
 343         else if (out->nV == 1)
 344         {
 345             /* No energy groups */
 346             out->Vvdw[0] = 0;
 347             out->Vc[0]   = 0;
 348
 349             p_nbk_ener[coult][ljtreatment](nbl[nb], nbat,
 350                                            ic,
 351                                            shift_vec,
 352                                            out->f,
 353                                            fshift_p,
 354                                            out->Vvdw,
 355                                            out->Vc);
 356         }
 357         else
 358         {
 359             /* Calculate energy group contributions */
 360             int i;
 361
 362             for (i = 0; i < out->nVS; i++)
 363             {
 364                 out->VSvdw[i] = 0;
 365             }
 366             for (i = 0; i < out->nVS; i++)
 367             {
 368                 out->VSc[i] = 0;
 369             }
 370
 371             p_nbk_energrp[coult][ljtreatment](nbl[nb], nbat,
 372                                               ic,
 373                                               shift_vec,
 374                                               out->f,
 375                                               fshift_p,
 376                                               out->VSvdw,
 377                                               out->VSc);
 378
 379             reduce_group_energies(nbat->nenergrp, nbat->neg_2log,
 380                                   out->VSvdw, out->VSc,
 381                                   out->Vvdw, out->Vc);
 382         }
 383     }
 384
 385     if (force_flags & GMX_FORCE_ENERGY)
 386     {
 387         reduce_energies_over_lists(nbat, nnbl, Vvdw, Vc);
 388     }
 389 }
 390 #else
 391 {
 392     gmx_incons("nbnxn_kernel_simd_4xn called when such kernels "
 393                " are not enabled.");
 394 }
 395 #endif
 396 #undef GMX_SIMD_J_UNROLL_SIZE