src/gromacs/mdlib/nbnxn_kernels/nbnxn_kernel_file_generator/nbnxn_kernel_simd_template.c.pre

   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 #ifdef HAVE_CONFIG_H
  36 #include <config.h>
  37 #endif
  38
  39 #include "typedefs.h"
  40
  41 #ifdef {0}
  42
  43 {1}
  44 #include "gromacs/simd/macros.h"
  45 #include "gromacs/simd/vector_operations.h"
  46 {2}
  47 #define GMX_SIMD_J_UNROLL_SIZE {3}
  48 #include "{4}"
  49 #include "../nbnxn_kernel_common.h"
  50 #include "gmx_omp_nthreads.h"
  51 #include "types/force_flags.h"
  52 #include "gmx_fatal.h"
  53
  54 /*! \brief Kinds of electrostatic treatments in SIMD Verlet kernels
  55  */
  56 enum {{
  57     coultRF, coultTAB, coultTAB_TWIN, coultEWALD, coultEWALD_TWIN, coultNR
  58 }};
  59
  60 /* Declare and define the kernel function pointer lookup tables.
  61  * The minor index of the array goes over both the LJ combination rules,
  62  * which is only supported by plain cut-off, and the LJ switch functions.
  63  */
  64 static p_nbk_func_noener p_nbk_noener[coultNR][ljcrNR+2] =
  65 {7}
  66 static p_nbk_func_ener p_nbk_ener[coultNR][ljcrNR+2] =
  67 {8}
  68 static p_nbk_func_ener p_nbk_energrp[coultNR][ljcrNR+2] =
  69 {9}
  70
  71 static void
  72 reduce_group_energies(int ng, int ng_2log,
  73                       const real *VSvdw, const real *VSc,
  74                       real *Vvdw, real *Vc)
  75 {{
  76     const int unrollj      = GMX_SIMD_REAL_WIDTH/GMX_SIMD_J_UNROLL_SIZE;
  77     const int unrollj_half = unrollj/2;
  78     int       ng_p2, i, j, j0, j1, c, s;
  79
  80     ng_p2 = (1<<ng_2log);
  81
  82     /* The size of the x86 SIMD energy group buffer array is:
  83      * ng*ng*ng_p2*unrollj_half*simd_width
  84      */
  85     for (i = 0; i < ng; i++)
  86     {{
  87         for (j = 0; j < ng; j++)
  88         {{
  89             Vvdw[i*ng+j] = 0;
  90             Vc[i*ng+j]   = 0;
  91         }}
  92
  93         for (j1 = 0; j1 < ng; j1++)
  94         {{
  95             for (j0 = 0; j0 < ng; j0++)
  96             {{
  97                 c = ((i*ng + j1)*ng_p2 + j0)*unrollj_half*unrollj;
  98                 for (s = 0; s < unrollj_half; s++)
  99                 {{
 100                     Vvdw[i*ng+j0] += VSvdw[c+0];
 101                     Vvdw[i*ng+j1] += VSvdw[c+1];
 102                     Vc  [i*ng+j0] += VSc  [c+0];
 103                     Vc  [i*ng+j1] += VSc  [c+1];
 104                     c             += unrollj + 2;
 105                 }}
 106             }}
 107         }}
 108     }}
 109 }}
 110
 111 #else /* {0} */
 112
 113 #include "gmx_fatal.h"
 114
 115 #endif /* {0} */
 116
 117 void
 118 {5}(nbnxn_pairlist_set_t      gmx_unused *nbl_list,
 119 {6}const nbnxn_atomdata_t    gmx_unused *nbat,
 120 {6}const interaction_const_t gmx_unused *ic,
 121 {6}int                       gmx_unused  ewald_excl,
 122 {6}rvec                      gmx_unused *shift_vec,
 123 {6}int                       gmx_unused  force_flags,
 124 {6}int                       gmx_unused  clearF,
 125 {6}real                      gmx_unused *fshift,
 126 {6}real                      gmx_unused *Vc,
 127 {6}real                      gmx_unused *Vvdw)
 128 #ifdef {0}
 129 {{
 130     int                nnbl;
 131     nbnxn_pairlist_t **nbl;
 132     int                coult, ljtreatment = 0;
 133     int                nb;
 134
 135     nnbl = nbl_list->nnbl;
 136     nbl  = nbl_list->nbl;
 137
 138     if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
 139     {{
 140         coult = coultRF;
 141     }}
 142     else
 143     {{
 144         if (ewald_excl == ewaldexclTable)
 145         {{
 146             if (ic->rcoulomb == ic->rvdw)
 147             {{
 148                 coult = coultTAB;
 149             }}
 150             else
 151             {{
 152                 coult = coultTAB_TWIN;
 153             }}
 154         }}
 155         else
 156         {{
 157             if (ic->rcoulomb == ic->rvdw)
 158             {{
 159                 coult = coultEWALD;
 160             }}
 161             else
 162             {{
 163                 coult = coultEWALD_TWIN;
 164             }}
 165         }}
 166     }}
 167
 168     switch (ic->vdw_modifier)
 169     {{
 170         case eintmodNONE:
 171         case eintmodPOTSHIFT:
 172             ljtreatment = nbat->comb_rule;
 173             break;
 174         /* Switch functions follow after cut-off combination rule kernels */
 175         case eintmodFORCESWITCH:
 176             ljtreatment = ljcrNR;
 177             break;
 178         case eintmodPOTSWITCH:
 179             ljtreatment = ljcrNR + 1;
 180             break;
 181         default:
 182             gmx_incons("Unsupported VdW interaction modifier");
 183     }}
 184
 185 #pragma omp parallel for schedule(static) num_threads(gmx_omp_nthreads_get(emntNonbonded))
 186     for (nb = 0; nb < nnbl; nb++)
 187     {{
 188         nbnxn_atomdata_output_t *out;
 189         real                    *fshift_p;
 190
 191         out = &nbat->out[nb];
 192
 193         if (clearF == enbvClearFYes)
 194         {{
 195             clear_f(nbat, nb, out->f);
 196         }}
 197
 198         if ((force_flags & GMX_FORCE_VIRIAL) && nnbl == 1)
 199         {{
 200             fshift_p = fshift;
 201         }}
 202         else
 203         {{
 204             fshift_p = out->fshift;
 205
 206             if (clearF == enbvClearFYes)
 207             {{
 208                 clear_fshift(fshift_p);
 209             }}
 210         }}
 211
 212         if (!(force_flags & GMX_FORCE_ENERGY))
 213         {{
 214             /* Don't calculate energies */
 215             p_nbk_noener[coult][ljtreatment](nbl[nb], nbat,
 216                                              ic,
 217                                              shift_vec,
 218                                              out->f,
 219                                              fshift_p);
 220         }}
 221         else if (out->nV == 1)
 222         {{
 223             /* No energy groups */
 224             out->Vvdw[0] = 0;
 225             out->Vc[0]   = 0;
 226
 227             p_nbk_ener[coult][ljtreatment](nbl[nb], nbat,
 228                                            ic,
 229                                            shift_vec,
 230                                            out->f,
 231                                            fshift_p,
 232                                            out->Vvdw,
 233                                            out->Vc);
 234         }}
 235         else
 236         {{
 237             /* Calculate energy group contributions */
 238             int i;
 239
 240             for (i = 0; i < out->nVS; i++)
 241             {{
 242                 out->VSvdw[i] = 0;
 243             }}
 244             for (i = 0; i < out->nVS; i++)
 245             {{
 246                 out->VSc[i] = 0;
 247             }}
 248
 249             p_nbk_energrp[coult][ljtreatment](nbl[nb], nbat,
 250                                               ic,
 251                                               shift_vec,
 252                                               out->f,
 253                                               fshift_p,
 254                                               out->VSvdw,
 255                                               out->VSc);
 256
 257             reduce_group_energies(nbat->nenergrp, nbat->neg_2log,
 258                                   out->VSvdw, out->VSc,
 259                                   out->Vvdw, out->Vc);
 260         }}
 261     }}
 262
 263     if (force_flags & GMX_FORCE_ENERGY)
 264     {{
 265         reduce_energies_over_lists(nbat, nnbl, Vvdw, Vc);
 266     }}
 267 }}
 268 #else
 269 {{
 270     gmx_incons("{5} called when such kernels "
 271                " are not enabled.");
 272 }}
 273 #endif
 274 #undef GMX_SIMD_J_UNROLL_SIZE