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, 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 "gmx_simd_macros.h"
  51 #include "gmx_simd_vec.h"
  52 #if !(GMX_SIMD_WIDTH_HERE == 8 || GMX_SIMD_WIDTH_HERE == 16)
  53 #error "unsupported SIMD width"
  54 #endif
  55
  56 #define GMX_SIMD_J_UNROLL_SIZE 2
  57 #include "nbnxn_kernel_simd_2xnn.h"
  58 #include "../nbnxn_kernel_common.h"
  59 #include "gmx_omp_nthreads.h"
  60 #include "types/force_flags.h"
  61
  62 /*! \brief Kinds of electrostatic treatments in SIMD Verlet kernels
  63  */
  64 enum {
  65     coultRF, coultTAB, coultTAB_TWIN, coultEWALD, coultEWALD_TWIN, coultNR
  66 };
  67
  68 /* Declare and define the kernel function pointer lookup tables. */
  69 static p_nbk_func_ener p_nbk_ener[coultNR][ljcrNR] =
  70 {
  71     {
  72         nbnxn_kernel_simd_2xnn_rf_comb_geom_ener,
  73         nbnxn_kernel_simd_2xnn_rf_comb_lb_ener,
  74         nbnxn_kernel_simd_2xnn_rf_comb_none_ener,
  75     },
  76     {
  77         nbnxn_kernel_simd_2xnn_tab_comb_geom_ener,
  78         nbnxn_kernel_simd_2xnn_tab_comb_lb_ener,
  79         nbnxn_kernel_simd_2xnn_tab_comb_none_ener,
  80     },
  81     {
  82         nbnxn_kernel_simd_2xnn_tab_twin_comb_geom_ener,
  83         nbnxn_kernel_simd_2xnn_tab_twin_comb_lb_ener,
  84         nbnxn_kernel_simd_2xnn_tab_twin_comb_none_ener,
  85     },
  86     {
  87         nbnxn_kernel_simd_2xnn_ewald_comb_geom_ener,
  88         nbnxn_kernel_simd_2xnn_ewald_comb_lb_ener,
  89         nbnxn_kernel_simd_2xnn_ewald_comb_none_ener,
  90     },
  91     {
  92         nbnxn_kernel_simd_2xnn_ewald_twin_comb_geom_ener,
  93         nbnxn_kernel_simd_2xnn_ewald_twin_comb_lb_ener,
  94         nbnxn_kernel_simd_2xnn_ewald_twin_comb_none_ener,
  95     },
  96 };
  97
  98 static p_nbk_func_ener p_nbk_energrp[coultNR][ljcrNR] =
  99 {
 100     {
 101         nbnxn_kernel_simd_2xnn_rf_comb_geom_energrp,
 102         nbnxn_kernel_simd_2xnn_rf_comb_lb_energrp,
 103         nbnxn_kernel_simd_2xnn_rf_comb_none_energrp,
 104     },
 105     {
 106         nbnxn_kernel_simd_2xnn_tab_comb_geom_energrp,
 107         nbnxn_kernel_simd_2xnn_tab_comb_lb_energrp,
 108         nbnxn_kernel_simd_2xnn_tab_comb_none_energrp,
 109     },
 110     {
 111         nbnxn_kernel_simd_2xnn_tab_twin_comb_geom_energrp,
 112         nbnxn_kernel_simd_2xnn_tab_twin_comb_lb_energrp,
 113         nbnxn_kernel_simd_2xnn_tab_twin_comb_none_energrp,
 114     },
 115     {
 116         nbnxn_kernel_simd_2xnn_ewald_comb_geom_energrp,
 117         nbnxn_kernel_simd_2xnn_ewald_comb_lb_energrp,
 118         nbnxn_kernel_simd_2xnn_ewald_comb_none_energrp,
 119     },
 120     {
 121         nbnxn_kernel_simd_2xnn_ewald_twin_comb_geom_energrp,
 122         nbnxn_kernel_simd_2xnn_ewald_twin_comb_lb_energrp,
 123         nbnxn_kernel_simd_2xnn_ewald_twin_comb_none_energrp,
 124     },
 125 };
 126
 127 static p_nbk_func_noener p_nbk_noener[coultNR][ljcrNR] =
 128 {
 129     {
 130         nbnxn_kernel_simd_2xnn_rf_comb_geom_noener,
 131         nbnxn_kernel_simd_2xnn_rf_comb_lb_noener,
 132         nbnxn_kernel_simd_2xnn_rf_comb_none_noener,
 133     },
 134     {
 135         nbnxn_kernel_simd_2xnn_tab_comb_geom_noener,
 136         nbnxn_kernel_simd_2xnn_tab_comb_lb_noener,
 137         nbnxn_kernel_simd_2xnn_tab_comb_none_noener,
 138     },
 139     {
 140         nbnxn_kernel_simd_2xnn_tab_twin_comb_geom_noener,
 141         nbnxn_kernel_simd_2xnn_tab_twin_comb_lb_noener,
 142         nbnxn_kernel_simd_2xnn_tab_twin_comb_none_noener,
 143     },
 144     {
 145         nbnxn_kernel_simd_2xnn_ewald_comb_geom_noener,
 146         nbnxn_kernel_simd_2xnn_ewald_comb_lb_noener,
 147         nbnxn_kernel_simd_2xnn_ewald_comb_none_noener,
 148     },
 149     {
 150         nbnxn_kernel_simd_2xnn_ewald_twin_comb_geom_noener,
 151         nbnxn_kernel_simd_2xnn_ewald_twin_comb_lb_noener,
 152         nbnxn_kernel_simd_2xnn_ewald_twin_comb_none_noener,
 153     },
 154 };
 155
 156
 157 static void
 158 reduce_group_energies(int ng, int ng_2log,
 159                       const real *VSvdw, const real *VSc,
 160                       real *Vvdw, real *Vc)
 161 {
 162     const int unrollj      = GMX_SIMD_WIDTH_HERE/GMX_SIMD_J_UNROLL_SIZE;
 163     const int unrollj_half = unrollj/2;
 164     int       ng_p2, i, j, j0, j1, c, s;
 165
 166     ng_p2 = (1<<ng_2log);
 167
 168     /* The size of the x86 SIMD energy group buffer array is:
 169      * ng*ng*ng_p2*unrollj_half*simd_width
 170      */
 171     for (i = 0; i < ng; i++)
 172     {
 173         for (j = 0; j < ng; j++)
 174         {
 175             Vvdw[i*ng+j] = 0;
 176             Vc[i*ng+j]   = 0;
 177         }
 178
 179         for (j1 = 0; j1 < ng; j1++)
 180         {
 181             for (j0 = 0; j0 < ng; j0++)
 182             {
 183                 c = ((i*ng + j1)*ng_p2 + j0)*unrollj_half*unrollj;
 184                 for (s = 0; s < unrollj_half; s++)
 185                 {
 186                     Vvdw[i*ng+j0] += VSvdw[c+0];
 187                     Vvdw[i*ng+j1] += VSvdw[c+1];
 188                     Vc  [i*ng+j0] += VSc  [c+0];
 189                     Vc  [i*ng+j1] += VSc  [c+1];
 190                     c             += unrollj + 2;
 191                 }
 192             }
 193         }
 194     }
 195 }
 196
 197 #else /* GMX_NBNXN_SIMD_2XNN */
 198
 199 #include "gmx_fatal.h"
 200
 201 #endif /* GMX_NBNXN_SIMD_2XNN */
 202
 203 void
 204 nbnxn_kernel_simd_2xnn(nbnxn_pairlist_set_t      gmx_unused *nbl_list,
 205                        const nbnxn_atomdata_t    gmx_unused *nbat,
 206                        const interaction_const_t gmx_unused *ic,
 207                        int                       gmx_unused  ewald_excl,
 208                        rvec                      gmx_unused *shift_vec,
 209                        int                       gmx_unused  force_flags,
 210                        int                       gmx_unused  clearF,
 211                        real                      gmx_unused *fshift,
 212                        real                      gmx_unused *Vc,
 213                        real                      gmx_unused *Vvdw)
 214 #ifdef GMX_NBNXN_SIMD_2XNN
 215 {
 216     int                nnbl;
 217     nbnxn_pairlist_t **nbl;
 218     int                coult;
 219     int                nb;
 220
 221     nnbl = nbl_list->nnbl;
 222     nbl  = nbl_list->nbl;
 223
 224     if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
 225     {
 226         coult = coultRF;
 227     }
 228     else
 229     {
 230         if (ewald_excl == ewaldexclTable)
 231         {
 232             if (ic->rcoulomb == ic->rvdw)
 233             {
 234                 coult = coultTAB;
 235             }
 236             else
 237             {
 238                 coult = coultTAB_TWIN;
 239             }
 240         }
 241         else
 242         {
 243             if (ic->rcoulomb == ic->rvdw)
 244             {
 245                 coult = coultEWALD;
 246             }
 247             else
 248             {
 249                 coult = coultEWALD_TWIN;
 250             }
 251         }
 252     }
 253
 254 #pragma omp parallel for schedule(static) num_threads(gmx_omp_nthreads_get(emntNonbonded))
 255     for (nb = 0; nb < nnbl; nb++)
 256     {
 257         nbnxn_atomdata_output_t *out;
 258         real                    *fshift_p;
 259
 260         out = &nbat->out[nb];
 261
 262         if (clearF == enbvClearFYes)
 263         {
 264             clear_f(nbat, nb, out->f);
 265         }
 266
 267         if ((force_flags & GMX_FORCE_VIRIAL) && nnbl == 1)
 268         {
 269             fshift_p = fshift;
 270         }
 271         else
 272         {
 273             fshift_p = out->fshift;
 274
 275             if (clearF == enbvClearFYes)
 276             {
 277                 clear_fshift(fshift_p);
 278             }
 279         }
 280
 281         if (!(force_flags & GMX_FORCE_ENERGY))
 282         {
 283             /* Don't calculate energies */
 284             p_nbk_noener[coult][nbat->comb_rule](nbl[nb], nbat,
 285                                                  ic,
 286                                                  shift_vec,
 287                                                  out->f,
 288                                                  fshift_p);
 289         }
 290         else if (out->nV == 1)
 291         {
 292             /* No energy groups */
 293             out->Vvdw[0] = 0;
 294             out->Vc[0]   = 0;
 295
 296             p_nbk_ener[coult][nbat->comb_rule](nbl[nb], nbat,
 297                                                ic,
 298                                                shift_vec,
 299                                                out->f,
 300                                                fshift_p,
 301                                                out->Vvdw,
 302                                                out->Vc);
 303         }
 304         else
 305         {
 306             /* Calculate energy group contributions */
 307             int i;
 308
 309             for (i = 0; i < out->nVS; i++)
 310             {
 311                 out->VSvdw[i] = 0;
 312             }
 313             for (i = 0; i < out->nVS; i++)
 314             {
 315                 out->VSc[i] = 0;
 316             }
 317
 318             p_nbk_energrp[coult][nbat->comb_rule](nbl[nb], nbat,
 319                                                   ic,
 320                                                   shift_vec,
 321                                                   out->f,
 322                                                   fshift_p,
 323                                                   out->VSvdw,
 324                                                   out->VSc);
 325
 326             reduce_group_energies(nbat->nenergrp, nbat->neg_2log,
 327                                   out->VSvdw, out->VSc,
 328                                   out->Vvdw, out->Vc);
 329         }
 330     }
 331
 332     if (force_flags & GMX_FORCE_ENERGY)
 333     {
 334         reduce_energies_over_lists(nbat, nnbl, Vvdw, Vc);
 335     }
 336 }
 337 #else
 338 {
 339     gmx_incons("nbnxn_kernel_simd_2xnn called when such kernels "
 340                " are not enabled.");
 341 }
 342 #endif
 343 #undef GMX_SIMD_J_UNROLL_SIZE