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

   1 #ifdef HAVE_CONFIG_H
   2 #include <config.h>
   3 #endif
   4
   5 #include "typedefs.h"
   6
   7 #ifdef {0}
   8
   9 {1}
  10 #include "gmx_simd_macros.h"
  11 #include "gmx_simd_vec.h"
  12 {2}
  13 #define GMX_SIMD_J_UNROLL_SIZE {3}
  14 #include "{4}"
  15 #include "../nbnxn_kernel_common.h"
  16 #include "gmx_omp_nthreads.h"
  17 #include "types/force_flags.h"
  18
  19 /*! \brief Kinds of electrostatic treatments in SIMD Verlet kernels
  20  */
  21 enum {{
  22     coultRF, coultTAB, coultTAB_TWIN, coultEWALD, coultEWALD_TWIN, coultNR
  23 }};
  24
  25 /* Declare and define the kernel function pointer lookup tables. */
  26 static p_nbk_func_ener p_nbk_ener[coultNR][ljcrNR] =
  27 {7}
  28 static p_nbk_func_ener p_nbk_energrp[coultNR][ljcrNR] =
  29 {8}
  30 static p_nbk_func_noener p_nbk_noener[coultNR][ljcrNR] =
  31 {9}
  32
  33 static void
  34 reduce_group_energies(int ng, int ng_2log,
  35                       const real *VSvdw, const real *VSc,
  36                       real *Vvdw, real *Vc)
  37 {{
  38     const int unrollj      = GMX_SIMD_WIDTH_HERE/GMX_SIMD_J_UNROLL_SIZE;
  39     const int unrollj_half = unrollj/2;
  40     int       ng_p2, i, j, j0, j1, c, s;
  41
  42     ng_p2 = (1<<ng_2log);
  43
  44     /* The size of the x86 SIMD energy group buffer array is:
  45      * ng*ng*ng_p2*unrollj_half*simd_width
  46      */
  47     for (i = 0; i < ng; i++)
  48     {{
  49         for (j = 0; j < ng; j++)
  50         {{
  51             Vvdw[i*ng+j] = 0;
  52             Vc[i*ng+j]   = 0;
  53         }}
  54
  55         for (j1 = 0; j1 < ng; j1++)
  56         {{
  57             for (j0 = 0; j0 < ng; j0++)
  58             {{
  59                 c = ((i*ng + j1)*ng_p2 + j0)*unrollj_half*unrollj;
  60                 for (s = 0; s < unrollj_half; s++)
  61                 {{
  62                     Vvdw[i*ng+j0] += VSvdw[c+0];
  63                     Vvdw[i*ng+j1] += VSvdw[c+1];
  64                     Vc  [i*ng+j0] += VSc  [c+0];
  65                     Vc  [i*ng+j1] += VSc  [c+1];
  66                     c             += unrollj + 2;
  67                 }}
  68             }}
  69         }}
  70     }}
  71 }}
  72
  73 #else /* {0} */
  74
  75 #include "gmx_fatal.h"
  76
  77 #endif /* {0} */
  78
  79 void
  80 {5}(nbnxn_pairlist_set_t       *nbl_list,
  81 {6}const nbnxn_atomdata_t     *nbat,
  82 {6}const interaction_const_t  *ic,
  83 {6}int                         ewald_excl,
  84 {6}rvec                       *shift_vec,
  85 {6}int                         force_flags,
  86 {6}int                         clearF,
  87 {6}real                       *fshift,
  88 {6}real                       *Vc,
  89 {6}real                       *Vvdw)
  90 #ifdef {0}
  91 {{
  92     int                nnbl;
  93     nbnxn_pairlist_t **nbl;
  94     int                coult;
  95     int                nb;
  96
  97     nnbl = nbl_list->nnbl;
  98     nbl  = nbl_list->nbl;
  99
 100     if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
 101     {{
 102         coult = coultRF;
 103     }}
 104     else
 105     {{
 106         if (ewald_excl == ewaldexclTable)
 107         {{
 108             if (ic->rcoulomb == ic->rvdw)
 109             {{
 110                 coult = coultTAB;
 111             }}
 112             else
 113             {{
 114                 coult = coultTAB_TWIN;
 115             }}
 116         }}
 117         else
 118         {{
 119             if (ic->rcoulomb == ic->rvdw)
 120             {{
 121                 coult = coultEWALD;
 122             }}
 123             else
 124             {{
 125                 coult = coultEWALD_TWIN;
 126             }}
 127         }}
 128     }}
 129
 130 #pragma omp parallel for schedule(static) num_threads(gmx_omp_nthreads_get(emntNonbonded))
 131     for (nb = 0; nb < nnbl; nb++)
 132     {{
 133         nbnxn_atomdata_output_t *out;
 134         real                    *fshift_p;
 135
 136         out = &nbat->out[nb];
 137
 138         if (clearF == enbvClearFYes)
 139         {{
 140             clear_f(nbat, nb, out->f);
 141         }}
 142
 143         if ((force_flags & GMX_FORCE_VIRIAL) && nnbl == 1)
 144         {{
 145             fshift_p = fshift;
 146         }}
 147         else
 148         {{
 149             fshift_p = out->fshift;
 150
 151             if (clearF == enbvClearFYes)
 152             {{
 153                 clear_fshift(fshift_p);
 154             }}
 155         }}
 156
 157         if (!(force_flags & GMX_FORCE_ENERGY))
 158         {{
 159             /* Don't calculate energies */
 160             p_nbk_noener[coult][nbat->comb_rule](nbl[nb], nbat,
 161                                                  ic,
 162                                                  shift_vec,
 163                                                  out->f,
 164                                                  fshift_p);
 165         }}
 166         else if (out->nV == 1)
 167         {{
 168             /* No energy groups */
 169             out->Vvdw[0] = 0;
 170             out->Vc[0]   = 0;
 171
 172             p_nbk_ener[coult][nbat->comb_rule](nbl[nb], nbat,
 173                                                ic,
 174                                                shift_vec,
 175                                                out->f,
 176                                                fshift_p,
 177                                                out->Vvdw,
 178                                                out->Vc);
 179         }}
 180         else
 181         {{
 182             /* Calculate energy group contributions */
 183             int i;
 184
 185             for (i = 0; i < out->nVS; i++)
 186             {{
 187                 out->VSvdw[i] = 0;
 188             }}
 189             for (i = 0; i < out->nVS; i++)
 190             {{
 191                 out->VSc[i] = 0;
 192             }}
 193
 194             p_nbk_energrp[coult][nbat->comb_rule](nbl[nb], nbat,
 195                                                   ic,
 196                                                   shift_vec,
 197                                                   out->f,
 198                                                   fshift_p,
 199                                                   out->VSvdw,
 200                                                   out->VSc);
 201
 202             reduce_group_energies(nbat->nenergrp, nbat->neg_2log,
 203                                   out->VSvdw, out->VSc,
 204                                   out->Vvdw, out->Vc);
 205         }}
 206     }}
 207
 208     if (force_flags & GMX_FORCE_ENERGY)
 209     {{
 210         reduce_energies_over_lists(nbat, nnbl, Vvdw, Vc);
 211     }}
 212 }}
 213 #else
 214 {{
 215     gmx_incons("{5} called when such kernels "
 216                " are not enabled.");
 217 }}
 218 #endif
 219 #undef GMX_SIMD_J_UNROLL_SIZE