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