src/gromacs/mdlib/nbnxn_kernels/simd_4xn/nbnxn_kernel_simd_4xn_outer.h

   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
  37 {
  38     const nbnxn_ci_t   *nbln;
  39     const nbnxn_cj_t   *l_cj;
  40     const int *         type;
  41     const real *        q;
  42     const real         *shiftvec;
  43     const real         *x;
  44     const real         *nbfp0, *nbfp1, *nbfp2 = NULL, *nbfp3 = NULL;
  45     real                facel;
  46     real               *nbfp_ptr;
  47     int                 n, ci, ci_sh;
  48     int                 ish, ish3;
  49     gmx_bool            do_LJ, half_LJ, do_coul, do_self;
  50     int                 sci, scix, sciy, sciz, sci2;
  51     int                 cjind0, cjind1, cjind;
  52     int                 ip, jp;
  53
  54 #ifdef ENERGY_GROUPS
  55     int         Vstride_i;
  56     int         egps_ishift, egps_imask;
  57     int         egps_jshift, egps_jmask, egps_jstride;
  58     int         egps_i;
  59     real       *vvdwtp[UNROLLI];
  60     real       *vctp[UNROLLI];
  61 #endif
  62
  63     gmx_simd_real_t  shX_S;
  64     gmx_simd_real_t  shY_S;
  65     gmx_simd_real_t  shZ_S;
  66     gmx_simd_real_t  ix_S0, iy_S0, iz_S0;
  67     gmx_simd_real_t  ix_S1, iy_S1, iz_S1;
  68     gmx_simd_real_t  ix_S2, iy_S2, iz_S2;
  69     gmx_simd_real_t  ix_S3, iy_S3, iz_S3;
  70     gmx_simd_real_t  fix_S0, fiy_S0, fiz_S0;
  71     gmx_simd_real_t  fix_S1, fiy_S1, fiz_S1;
  72     gmx_simd_real_t  fix_S2, fiy_S2, fiz_S2;
  73     gmx_simd_real_t  fix_S3, fiy_S3, fiz_S3;
  74 #if UNROLLJ >= 4
  75     /* We use an i-force SIMD register width of 4 */
  76     gmx_simd4_real_t fix_S, fiy_S, fiz_S;
  77 #else
  78     /* We use an i-force SIMD register width of 2 */
  79     gmx_simd_real_t  fix0_S, fiy0_S, fiz0_S;
  80     gmx_simd_real_t  fix2_S, fiy2_S, fiz2_S;
  81 #endif
  82
  83     gmx_simd_real_t  diagonal_jmi_S;
  84 #if UNROLLI == UNROLLJ
  85     gmx_simd_bool_t  diagonal_mask_S0, diagonal_mask_S1, diagonal_mask_S2, diagonal_mask_S3;
  86 #else
  87     gmx_simd_bool_t  diagonal_mask0_S0, diagonal_mask0_S1, diagonal_mask0_S2, diagonal_mask0_S3;
  88     gmx_simd_bool_t  diagonal_mask1_S0, diagonal_mask1_S1, diagonal_mask1_S2, diagonal_mask1_S3;
  89 #endif
  90
  91     unsigned            *exclusion_filter;
  92     gmx_exclfilter       filter_S0, filter_S1, filter_S2, filter_S3;
  93
  94     gmx_simd_real_t      zero_S = gmx_simd_set1_r(0.0);
  95
  96     gmx_simd_real_t      one_S  = gmx_simd_set1_r(1.0);
  97     gmx_simd_real_t      iq_S0  = gmx_simd_setzero_r();
  98     gmx_simd_real_t      iq_S1  = gmx_simd_setzero_r();
  99     gmx_simd_real_t      iq_S2  = gmx_simd_setzero_r();
 100     gmx_simd_real_t      iq_S3  = gmx_simd_setzero_r();
 101
 102 #ifdef CALC_COUL_RF
 103     gmx_simd_real_t      mrc_3_S;
 104 #ifdef CALC_ENERGIES
 105     gmx_simd_real_t      hrc_3_S, moh_rc_S;
 106 #endif
 107 #endif
 108
 109 #ifdef CALC_COUL_TAB
 110     /* Coulomb table variables */
 111     gmx_simd_real_t   invtsp_S;
 112     const real *      tab_coul_F;
 113 #ifndef TAB_FDV0
 114     const real *      tab_coul_V;
 115 #endif
 116     /* Thread-local working buffers for force and potential lookups */
 117     int               ti0_array[2*GMX_SIMD_REAL_WIDTH], *ti0 = NULL;
 118     int               ti1_array[2*GMX_SIMD_REAL_WIDTH], *ti1 = NULL;
 119     int               ti2_array[2*GMX_SIMD_REAL_WIDTH], *ti2 = NULL;
 120     int               ti3_array[2*GMX_SIMD_REAL_WIDTH], *ti3 = NULL;
 121 #ifdef CALC_ENERGIES
 122     gmx_simd_real_t   mhalfsp_S;
 123 #endif
 124 #endif
 125
 126 #ifdef CALC_COUL_EWALD
 127     gmx_simd_real_t beta2_S, beta_S;
 128 #endif
 129
 130 #if defined CALC_ENERGIES && (defined CALC_COUL_EWALD || defined CALC_COUL_TAB)
 131     gmx_simd_real_t  sh_ewald_S;
 132 #endif
 133
 134 #if defined LJ_CUT && defined CALC_ENERGIES
 135     gmx_simd_real_t   p6_cpot_S, p12_cpot_S;
 136 #endif
 137 #ifdef LJ_POT_SWITCH
 138     gmx_simd_real_t   rswitch_S;
 139     gmx_simd_real_t   swV3_S, swV4_S, swV5_S;
 140     gmx_simd_real_t   swF2_S, swF3_S, swF4_S;
 141 #endif
 142 #ifdef LJ_FORCE_SWITCH
 143     gmx_simd_real_t   rswitch_S;
 144     gmx_simd_real_t   p6_fc2_S, p6_fc3_S;
 145     gmx_simd_real_t   p12_fc2_S, p12_fc3_S;
 146 #ifdef CALC_ENERGIES
 147     gmx_simd_real_t   p6_vc3_S, p6_vc4_S;
 148     gmx_simd_real_t   p12_vc3_S, p12_vc4_S;
 149     gmx_simd_real_t   p6_6cpot_S, p12_12cpot_S;
 150 #endif
 151 #endif
 152 #ifdef LJ_EWALD_GEOM
 153     real              lj_ewaldcoeff2, lj_ewaldcoeff6_6;
 154     gmx_simd_real_t   mone_S, half_S, lje_c2_S, lje_c6_6_S, lje_vc_S;
 155 #endif
 156
 157 #ifdef LJ_COMB_LB
 158     const real       *ljc;
 159
 160     gmx_simd_real_t   hsig_i_S0, seps_i_S0;
 161     gmx_simd_real_t   hsig_i_S1, seps_i_S1;
 162     gmx_simd_real_t   hsig_i_S2, seps_i_S2;
 163     gmx_simd_real_t   hsig_i_S3, seps_i_S3;
 164 #else
 165 #ifdef FIX_LJ_C
 166     real              pvdw_array[2*UNROLLI*UNROLLJ+3];
 167     real             *pvdw_c6, *pvdw_c12;
 168     gmx_simd_real_t   c6_S0, c12_S0;
 169     gmx_simd_real_t   c6_S1, c12_S1;
 170     gmx_simd_real_t   c6_S2, c12_S2;
 171     gmx_simd_real_t   c6_S3, c12_S3;
 172 #endif
 173
 174 #if defined LJ_COMB_GEOM || defined LJ_EWALD_GEOM
 175     const real       *ljc;
 176
 177     gmx_simd_real_t   c6s_S0, c12s_S0;
 178     gmx_simd_real_t   c6s_S1, c12s_S1;
 179     gmx_simd_real_t   c6s_S2  = gmx_simd_setzero_r();
 180     gmx_simd_real_t   c12s_S2 = gmx_simd_setzero_r();
 181     gmx_simd_real_t   c6s_S3  = gmx_simd_setzero_r();
 182     gmx_simd_real_t   c12s_S3 = gmx_simd_setzero_r();
 183 #endif
 184 #endif /* LJ_COMB_LB */
 185
 186     gmx_simd_real_t  vctot_S, Vvdwtot_S;
 187     gmx_simd_real_t  sixth_S, twelveth_S;
 188
 189     gmx_simd_real_t  avoid_sing_S;
 190     gmx_simd_real_t  rc2_S;
 191 #ifdef VDW_CUTOFF_CHECK
 192     gmx_simd_real_t  rcvdw2_S;
 193 #endif
 194
 195     int ninner;
 196
 197 #ifdef COUNT_PAIRS
 198     int npair = 0;
 199 #endif
 200
 201 #if defined LJ_COMB_GEOM || defined LJ_COMB_LB || defined LJ_EWALD_GEOM
 202     ljc = nbat->lj_comb;
 203 #endif
 204 #if !(defined LJ_COMB_GEOM || defined LJ_COMB_LB)
 205     /* No combination rule used */
 206     nbfp_ptr    = (4 == nbfp_stride) ? nbat->nbfp_s4 : nbat->nbfp;
 207 #endif
 208
 209     /* Load j-i for the first i */
 210     diagonal_jmi_S    = gmx_simd_load_r(nbat->simd_4xn_diagonal_j_minus_i);
 211     /* Generate all the diagonal masks as comparison results */
 212 #if UNROLLI == UNROLLJ
 213     diagonal_mask_S0  = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 214     diagonal_jmi_S    = gmx_simd_sub_r(diagonal_jmi_S, one_S);
 215     diagonal_mask_S1  = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 216     diagonal_jmi_S    = gmx_simd_sub_r(diagonal_jmi_S, one_S);
 217     diagonal_mask_S2  = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 218     diagonal_jmi_S    = gmx_simd_sub_r(diagonal_jmi_S, one_S);
 219     diagonal_mask_S3  = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 220 #else
 221 #if UNROLLI == 2*UNROLLJ || 2*UNROLLI == UNROLLJ
 222     diagonal_mask0_S0 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 223     diagonal_jmi_S    = gmx_simd_sub_r(diagonal_jmi_S, one_S);
 224     diagonal_mask0_S1 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 225     diagonal_jmi_S    = gmx_simd_sub_r(diagonal_jmi_S, one_S);
 226     diagonal_mask0_S2 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 227     diagonal_jmi_S    = gmx_simd_sub_r(diagonal_jmi_S, one_S);
 228     diagonal_mask0_S3 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 229     diagonal_jmi_S    = gmx_simd_sub_r(diagonal_jmi_S, one_S);
 230
 231 #if UNROLLI == 2*UNROLLJ
 232     /* Load j-i for the second half of the j-cluster */
 233     diagonal_jmi_S    = gmx_simd_load_r(nbat->simd_4xn_diagonal_j_minus_i + UNROLLJ);
 234 #endif
 235
 236     diagonal_mask1_S0 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 237     diagonal_jmi_S    = gmx_simd_sub_r(diagonal_jmi_S, one_S);
 238     diagonal_mask1_S1 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 239     diagonal_jmi_S    = gmx_simd_sub_r(diagonal_jmi_S, one_S);
 240     diagonal_mask1_S2 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 241     diagonal_jmi_S    = gmx_simd_sub_r(diagonal_jmi_S, one_S);
 242     diagonal_mask1_S3 = gmx_simd_cmplt_r(zero_S, diagonal_jmi_S);
 243 #endif
 244 #endif
 245
 246     /* Load masks for topology exclusion masking. filter_stride is
 247        static const, so the conditional will be optimized away. */
 248     if (1 == filter_stride)
 249     {
 250         exclusion_filter = nbat->simd_exclusion_filter1;
 251     }
 252     else /* (2 == filter_stride) */
 253     {
 254         exclusion_filter = nbat->simd_exclusion_filter2;
 255     }
 256
 257     /* Here we cast the exclusion filters from unsigned * to int * or real *.
 258      * Since we only check bits, the actual value they represent does not
 259      * matter, as long as both filter and mask data are treated the same way.
 260      */
 261     filter_S0 = gmx_load_exclusion_filter(exclusion_filter + 0*UNROLLJ*filter_stride);
 262     filter_S1 = gmx_load_exclusion_filter(exclusion_filter + 1*UNROLLJ*filter_stride);
 263     filter_S2 = gmx_load_exclusion_filter(exclusion_filter + 2*UNROLLJ*filter_stride);
 264     filter_S3 = gmx_load_exclusion_filter(exclusion_filter + 3*UNROLLJ*filter_stride);
 265
 266 #ifdef CALC_COUL_RF
 267     /* Reaction-field constants */
 268     mrc_3_S  = gmx_simd_set1_r(-2*ic->k_rf);
 269 #ifdef CALC_ENERGIES
 270     hrc_3_S  = gmx_simd_set1_r(ic->k_rf);
 271     moh_rc_S = gmx_simd_set1_r(-ic->c_rf);
 272 #endif
 273 #endif
 274
 275 #ifdef CALC_COUL_TAB
 276     /* Generate aligned table index pointers */
 277     ti0 = prepare_table_load_buffer(ti0_array);
 278     ti1 = prepare_table_load_buffer(ti1_array);
 279     ti2 = prepare_table_load_buffer(ti2_array);
 280     ti3 = prepare_table_load_buffer(ti3_array);
 281
 282     invtsp_S  = gmx_simd_set1_r(ic->tabq_scale);
 283 #ifdef CALC_ENERGIES
 284     mhalfsp_S = gmx_simd_set1_r(-0.5/ic->tabq_scale);
 285 #endif
 286
 287 #ifdef TAB_FDV0
 288     tab_coul_F = ic->tabq_coul_FDV0;
 289 #else
 290     tab_coul_F = ic->tabq_coul_F;
 291     tab_coul_V = ic->tabq_coul_V;
 292 #endif
 293 #endif /* CALC_COUL_TAB */
 294
 295 #ifdef CALC_COUL_EWALD
 296     beta2_S = gmx_simd_set1_r(ic->ewaldcoeff_q*ic->ewaldcoeff_q);
 297     beta_S  = gmx_simd_set1_r(ic->ewaldcoeff_q);
 298 #endif
 299
 300 #if (defined CALC_COUL_TAB || defined CALC_COUL_EWALD) && defined CALC_ENERGIES
 301     sh_ewald_S = gmx_simd_set1_r(ic->sh_ewald);
 302 #endif
 303
 304     /* LJ function constants */
 305 #if defined CALC_ENERGIES || defined LJ_POT_SWITCH
 306     sixth_S      = gmx_simd_set1_r(1.0/6.0);
 307     twelveth_S   = gmx_simd_set1_r(1.0/12.0);
 308 #endif
 309
 310 #if defined LJ_CUT && defined CALC_ENERGIES
 311     /* We shift the potential by cpot, which can be zero */
 312     p6_cpot_S    = gmx_simd_set1_r(ic->dispersion_shift.cpot);
 313     p12_cpot_S   = gmx_simd_set1_r(ic->repulsion_shift.cpot);
 314 #endif
 315 #ifdef LJ_POT_SWITCH
 316     rswitch_S = gmx_simd_set1_r(ic->rvdw_switch);
 317     swV3_S    = gmx_simd_set1_r(ic->vdw_switch.c3);
 318     swV4_S    = gmx_simd_set1_r(ic->vdw_switch.c4);
 319     swV5_S    = gmx_simd_set1_r(ic->vdw_switch.c5);
 320     swF2_S    = gmx_simd_set1_r(3*ic->vdw_switch.c3);
 321     swF3_S    = gmx_simd_set1_r(4*ic->vdw_switch.c4);
 322     swF4_S    = gmx_simd_set1_r(5*ic->vdw_switch.c5);
 323 #endif
 324 #ifdef LJ_FORCE_SWITCH
 325     rswitch_S = gmx_simd_set1_r(ic->rvdw_switch);
 326     p6_fc2_S  = gmx_simd_set1_r(ic->dispersion_shift.c2);
 327     p6_fc3_S  = gmx_simd_set1_r(ic->dispersion_shift.c3);
 328     p12_fc2_S = gmx_simd_set1_r(ic->repulsion_shift.c2);
 329     p12_fc3_S = gmx_simd_set1_r(ic->repulsion_shift.c3);
 330 #ifdef CALC_ENERGIES
 331     {
 332         gmx_simd_real_t mthird_S  = gmx_simd_set1_r(-1.0/3.0);
 333         gmx_simd_real_t mfourth_S = gmx_simd_set1_r(-1.0/4.0);
 334
 335         p6_vc3_S     = gmx_simd_mul_r(mthird_S,  p6_fc2_S);
 336         p6_vc4_S     = gmx_simd_mul_r(mfourth_S, p6_fc3_S);
 337         p6_6cpot_S   = gmx_simd_set1_r(ic->dispersion_shift.cpot/6);
 338         p12_vc3_S    = gmx_simd_mul_r(mthird_S,  p12_fc2_S);
 339         p12_vc4_S    = gmx_simd_mul_r(mfourth_S, p12_fc3_S);
 340         p12_12cpot_S = gmx_simd_set1_r(ic->repulsion_shift.cpot/12);
 341     }
 342 #endif
 343 #endif
 344 #ifdef LJ_EWALD_GEOM
 345     mone_S           = gmx_simd_set1_r(-1.0);
 346     half_S           = gmx_simd_set1_r(0.5);
 347     lj_ewaldcoeff2   = ic->ewaldcoeff_lj*ic->ewaldcoeff_lj;
 348     lj_ewaldcoeff6_6 = lj_ewaldcoeff2*lj_ewaldcoeff2*lj_ewaldcoeff2/6;
 349     lje_c2_S         = gmx_simd_set1_r(lj_ewaldcoeff2);
 350     lje_c6_6_S       = gmx_simd_set1_r(lj_ewaldcoeff6_6);
 351     /* Determine the grid potential at the cut-off */
 352     lje_vc_S         = gmx_simd_set1_r(ic->sh_lj_ewald);
 353 #endif
 354
 355     /* The kernel either supports rcoulomb = rvdw or rcoulomb >= rvdw */
 356     rc2_S    = gmx_simd_set1_r(ic->rcoulomb*ic->rcoulomb);
 357 #ifdef VDW_CUTOFF_CHECK
 358     rcvdw2_S = gmx_simd_set1_r(ic->rvdw*ic->rvdw);
 359 #endif
 360
 361     avoid_sing_S = gmx_simd_set1_r(NBNXN_AVOID_SING_R2_INC);
 362
 363     q                   = nbat->q;
 364     type                = nbat->type;
 365     facel               = ic->epsfac;
 366     shiftvec            = shift_vec[0];
 367     x                   = nbat->x;
 368
 369 #ifdef FIX_LJ_C
 370     pvdw_c6  = gmx_simd_align_real(pvdw_array);
 371     pvdw_c12 = pvdw_c6 + UNROLLI*UNROLLJ;
 372
 373     for (jp = 0; jp < UNROLLJ; jp++)
 374     {
 375         pvdw_c6 [0*UNROLLJ+jp] = nbat->nbfp[0*2];
 376         pvdw_c6 [1*UNROLLJ+jp] = nbat->nbfp[0*2];
 377         pvdw_c6 [2*UNROLLJ+jp] = nbat->nbfp[0*2];
 378         pvdw_c6 [3*UNROLLJ+jp] = nbat->nbfp[0*2];
 379
 380         pvdw_c12[0*UNROLLJ+jp] = nbat->nbfp[0*2+1];
 381         pvdw_c12[1*UNROLLJ+jp] = nbat->nbfp[0*2+1];
 382         pvdw_c12[2*UNROLLJ+jp] = nbat->nbfp[0*2+1];
 383         pvdw_c12[3*UNROLLJ+jp] = nbat->nbfp[0*2+1];
 384     }
 385     c6_S0            = gmx_simd_load_r(pvdw_c6 +0*UNROLLJ);
 386     c6_S1            = gmx_simd_load_r(pvdw_c6 +1*UNROLLJ);
 387     c6_S2            = gmx_simd_load_r(pvdw_c6 +2*UNROLLJ);
 388     c6_S3            = gmx_simd_load_r(pvdw_c6 +3*UNROLLJ);
 389
 390     c12_S0           = gmx_simd_load_r(pvdw_c12+0*UNROLLJ);
 391     c12_S1           = gmx_simd_load_r(pvdw_c12+1*UNROLLJ);
 392     c12_S2           = gmx_simd_load_r(pvdw_c12+2*UNROLLJ);
 393     c12_S3           = gmx_simd_load_r(pvdw_c12+3*UNROLLJ);
 394 #endif /* FIX_LJ_C */
 395
 396 #ifdef ENERGY_GROUPS
 397     egps_ishift  = nbat->neg_2log;
 398     egps_imask   = (1<<egps_ishift) - 1;
 399     egps_jshift  = 2*nbat->neg_2log;
 400     egps_jmask   = (1<<egps_jshift) - 1;
 401     egps_jstride = (UNROLLJ>>1)*UNROLLJ;
 402     /* Major division is over i-particle energy groups, determine the stride */
 403     Vstride_i    = nbat->nenergrp*(1<<nbat->neg_2log)*egps_jstride;
 404 #endif
 405
 406     l_cj = nbl->cj;
 407
 408     ninner = 0;
 409     for (n = 0; n < nbl->nci; n++)
 410     {
 411         nbln = &nbl->ci[n];
 412
 413         ish              = (nbln->shift & NBNXN_CI_SHIFT);
 414         ish3             = ish*3;
 415         cjind0           = nbln->cj_ind_start;
 416         cjind1           = nbln->cj_ind_end;
 417         ci               = nbln->ci;
 418         ci_sh            = (ish == CENTRAL ? ci : -1);
 419
 420         shX_S = gmx_simd_load1_r(shiftvec+ish3);
 421         shY_S = gmx_simd_load1_r(shiftvec+ish3+1);
 422         shZ_S = gmx_simd_load1_r(shiftvec+ish3+2);
 423
 424 #if UNROLLJ <= 4
 425         sci              = ci*STRIDE;
 426         scix             = sci*DIM;
 427         sci2             = sci*2;
 428 #else
 429         sci              = (ci>>1)*STRIDE;
 430         scix             = sci*DIM + (ci & 1)*(STRIDE>>1);
 431         sci2             = sci*2 + (ci & 1)*(STRIDE>>1);
 432         sci             += (ci & 1)*(STRIDE>>1);
 433 #endif
 434
 435         /* We have 5 LJ/C combinations, but use only three inner loops,
 436          * as the other combinations are unlikely and/or not much faster:
 437          * inner half-LJ + C for half-LJ + C / no-LJ + C
 438          * inner LJ + C      for full-LJ + C
 439          * inner LJ          for full-LJ + no-C / half-LJ + no-C
 440          */
 441         do_LJ   = (nbln->shift & NBNXN_CI_DO_LJ(0));
 442         do_coul = (nbln->shift & NBNXN_CI_DO_COUL(0));
 443         half_LJ = ((nbln->shift & NBNXN_CI_HALF_LJ(0)) || !do_LJ) && do_coul;
 444 #ifdef LJ_EWALD_GEOM
 445         do_self = TRUE;
 446 #else
 447         do_self = do_coul;
 448 #endif
 449
 450 #ifdef ENERGY_GROUPS
 451         egps_i = nbat->energrp[ci];
 452         {
 453             int ia, egp_ia;
 454
 455             for (ia = 0; ia < UNROLLI; ia++)
 456             {
 457                 egp_ia     = (egps_i >> (ia*egps_ishift)) & egps_imask;
 458                 vvdwtp[ia] = Vvdw + egp_ia*Vstride_i;
 459                 vctp[ia]   = Vc   + egp_ia*Vstride_i;
 460             }
 461         }
 462 #endif
 463
 464 #ifdef CALC_ENERGIES
 465 #if UNROLLJ == 4
 466         if (do_self && l_cj[nbln->cj_ind_start].cj == ci_sh)
 467 #endif
 468 #if UNROLLJ == 2
 469         if (do_self && l_cj[nbln->cj_ind_start].cj == (ci_sh<<1))
 470 #endif
 471 #if UNROLLJ == 8
 472         if (do_self && l_cj[nbln->cj_ind_start].cj == (ci_sh>>1))
 473 #endif
 474         {
 475             if (do_coul)
 476             {
 477                 real Vc_sub_self;
 478                 int  ia;
 479
 480 #ifdef CALC_COUL_RF
 481                 Vc_sub_self = 0.5*ic->c_rf;
 482 #endif
 483 #ifdef CALC_COUL_TAB
 484 #ifdef TAB_FDV0
 485                 Vc_sub_self = 0.5*tab_coul_F[2];
 486 #else
 487                 Vc_sub_self = 0.5*tab_coul_V[0];
 488 #endif
 489 #endif
 490 #ifdef CALC_COUL_EWALD
 491                 /* beta/sqrt(pi) */
 492                 Vc_sub_self = 0.5*ic->ewaldcoeff_q*M_2_SQRTPI;
 493 #endif
 494
 495                 for (ia = 0; ia < UNROLLI; ia++)
 496                 {
 497                     real qi;
 498
 499                     qi = q[sci+ia];
 500 #ifdef ENERGY_GROUPS
 501                     vctp[ia][((egps_i>>(ia*egps_ishift)) & egps_imask)*egps_jstride]
 502 #else
 503                     Vc[0]
 504 #endif
 505                         -= facel*qi*qi*Vc_sub_self;
 506                 }
 507             }
 508
 509 #ifdef LJ_EWALD_GEOM
 510             {
 511                 int  ia;
 512
 513                 for (ia = 0; ia < UNROLLI; ia++)
 514                 {
 515                     real c6_i;
 516
 517                     c6_i = nbat->nbfp[nbat->type[sci+ia]*(nbat->ntype + 1)*2]/6;
 518 #ifdef ENERGY_GROUPS
 519                     vvdwtp[ia][((egps_i>>(ia*egps_ishift)) & egps_imask)*egps_jstride]
 520 #else
 521                     Vvdw[0]
 522 #endif
 523                         += 0.5*c6_i*lj_ewaldcoeff6_6;
 524                 }
 525             }
 526 #endif      /* LJ_EWALD_GEOM */
 527         }
 528 #endif
 529
 530         /* Load i atom data */
 531         sciy             = scix + STRIDE;
 532         sciz             = sciy + STRIDE;
 533         ix_S0            = gmx_simd_add_r(gmx_simd_load1_r(x+scix), shX_S);
 534         ix_S1            = gmx_simd_add_r(gmx_simd_load1_r(x+scix+1), shX_S);
 535         ix_S2            = gmx_simd_add_r(gmx_simd_load1_r(x+scix+2), shX_S);
 536         ix_S3            = gmx_simd_add_r(gmx_simd_load1_r(x+scix+3), shX_S);
 537         iy_S0            = gmx_simd_add_r(gmx_simd_load1_r(x+sciy), shY_S);
 538         iy_S1            = gmx_simd_add_r(gmx_simd_load1_r(x+sciy+1), shY_S);
 539         iy_S2            = gmx_simd_add_r(gmx_simd_load1_r(x+sciy+2), shY_S);
 540         iy_S3            = gmx_simd_add_r(gmx_simd_load1_r(x+sciy+3), shY_S);
 541         iz_S0            = gmx_simd_add_r(gmx_simd_load1_r(x+sciz), shZ_S);
 542         iz_S1            = gmx_simd_add_r(gmx_simd_load1_r(x+sciz+1), shZ_S);
 543         iz_S2            = gmx_simd_add_r(gmx_simd_load1_r(x+sciz+2), shZ_S);
 544         iz_S3            = gmx_simd_add_r(gmx_simd_load1_r(x+sciz+3), shZ_S);
 545
 546         if (do_coul)
 547         {
 548             iq_S0      = gmx_simd_set1_r(facel*q[sci]);
 549             iq_S1      = gmx_simd_set1_r(facel*q[sci+1]);
 550             iq_S2      = gmx_simd_set1_r(facel*q[sci+2]);
 551             iq_S3      = gmx_simd_set1_r(facel*q[sci+3]);
 552         }
 553
 554 #ifdef LJ_COMB_LB
 555         hsig_i_S0      = gmx_simd_load1_r(ljc+sci2+0);
 556         hsig_i_S1      = gmx_simd_load1_r(ljc+sci2+1);
 557         hsig_i_S2      = gmx_simd_load1_r(ljc+sci2+2);
 558         hsig_i_S3      = gmx_simd_load1_r(ljc+sci2+3);
 559         seps_i_S0      = gmx_simd_load1_r(ljc+sci2+STRIDE+0);
 560         seps_i_S1      = gmx_simd_load1_r(ljc+sci2+STRIDE+1);
 561         seps_i_S2      = gmx_simd_load1_r(ljc+sci2+STRIDE+2);
 562         seps_i_S3      = gmx_simd_load1_r(ljc+sci2+STRIDE+3);
 563 #else
 564 #ifdef LJ_COMB_GEOM
 565         c6s_S0         = gmx_simd_load1_r(ljc+sci2+0);
 566         c6s_S1         = gmx_simd_load1_r(ljc+sci2+1);
 567         if (!half_LJ)
 568         {
 569             c6s_S2     = gmx_simd_load1_r(ljc+sci2+2);
 570             c6s_S3     = gmx_simd_load1_r(ljc+sci2+3);
 571         }
 572         c12s_S0        = gmx_simd_load1_r(ljc+sci2+STRIDE+0);
 573         c12s_S1        = gmx_simd_load1_r(ljc+sci2+STRIDE+1);
 574         if (!half_LJ)
 575         {
 576             c12s_S2    = gmx_simd_load1_r(ljc+sci2+STRIDE+2);
 577             c12s_S3    = gmx_simd_load1_r(ljc+sci2+STRIDE+3);
 578         }
 579 #else
 580         nbfp0     = nbfp_ptr + type[sci  ]*nbat->ntype*nbfp_stride;
 581         nbfp1     = nbfp_ptr + type[sci+1]*nbat->ntype*nbfp_stride;
 582         if (!half_LJ)
 583         {
 584             nbfp2 = nbfp_ptr + type[sci+2]*nbat->ntype*nbfp_stride;
 585             nbfp3 = nbfp_ptr + type[sci+3]*nbat->ntype*nbfp_stride;
 586         }
 587 #endif
 588 #endif
 589 #ifdef LJ_EWALD_GEOM
 590         /* We need the geometrically combined C6 for the PME grid correction */
 591         c6s_S0 = gmx_simd_load1_r(ljc+sci2+0);
 592         c6s_S1 = gmx_simd_load1_r(ljc+sci2+1);
 593         if (!half_LJ)
 594         {
 595             c6s_S2 = gmx_simd_load1_r(ljc+sci2+2);
 596             c6s_S3 = gmx_simd_load1_r(ljc+sci2+3);
 597         }
 598 #endif
 599
 600         /* Zero the potential energy for this list */
 601         Vvdwtot_S        = gmx_simd_setzero_r();
 602         vctot_S          = gmx_simd_setzero_r();
 603
 604         /* Clear i atom forces */
 605         fix_S0           = gmx_simd_setzero_r();
 606         fix_S1           = gmx_simd_setzero_r();
 607         fix_S2           = gmx_simd_setzero_r();
 608         fix_S3           = gmx_simd_setzero_r();
 609         fiy_S0           = gmx_simd_setzero_r();
 610         fiy_S1           = gmx_simd_setzero_r();
 611         fiy_S2           = gmx_simd_setzero_r();
 612         fiy_S3           = gmx_simd_setzero_r();
 613         fiz_S0           = gmx_simd_setzero_r();
 614         fiz_S1           = gmx_simd_setzero_r();
 615         fiz_S2           = gmx_simd_setzero_r();
 616         fiz_S3           = gmx_simd_setzero_r();
 617
 618         cjind = cjind0;
 619
 620         /* Currently all kernels use (at least half) LJ */
 621 #define CALC_LJ
 622         if (half_LJ)
 623         {
 624             /* Coulomb: all i-atoms, LJ: first half i-atoms */
 625 #define CALC_COULOMB
 626 #define HALF_LJ
 627 #define CHECK_EXCLS
 628             while (cjind < cjind1 && nbl->cj[cjind].excl != NBNXN_INTERACTION_MASK_ALL)
 629             {
 630 #include "nbnxn_kernel_simd_4xn_inner.h"
 631                 cjind++;
 632             }
 633 #undef CHECK_EXCLS
 634             for (; (cjind < cjind1); cjind++)
 635             {
 636 #include "nbnxn_kernel_simd_4xn_inner.h"
 637             }
 638 #undef HALF_LJ
 639 #undef CALC_COULOMB
 640         }
 641         else if (do_coul)
 642         {
 643             /* Coulomb: all i-atoms, LJ: all i-atoms */
 644 #define CALC_COULOMB
 645 #define CHECK_EXCLS
 646             while (cjind < cjind1 && nbl->cj[cjind].excl != NBNXN_INTERACTION_MASK_ALL)
 647             {
 648 #include "nbnxn_kernel_simd_4xn_inner.h"
 649                 cjind++;
 650             }
 651 #undef CHECK_EXCLS
 652             for (; (cjind < cjind1); cjind++)
 653             {
 654 #include "nbnxn_kernel_simd_4xn_inner.h"
 655             }
 656 #undef CALC_COULOMB
 657         }
 658         else
 659         {
 660             /* Coulomb: none, LJ: all i-atoms */
 661 #define CHECK_EXCLS
 662             while (cjind < cjind1 && nbl->cj[cjind].excl != NBNXN_INTERACTION_MASK_ALL)
 663             {
 664 #include "nbnxn_kernel_simd_4xn_inner.h"
 665                 cjind++;
 666             }
 667 #undef CHECK_EXCLS
 668             for (; (cjind < cjind1); cjind++)
 669             {
 670 #include "nbnxn_kernel_simd_4xn_inner.h"
 671             }
 672         }
 673 #undef CALC_LJ
 674         ninner += cjind1 - cjind0;
 675
 676         /* Add accumulated i-forces to the force array */
 677 #if UNROLLJ >= 4
 678         fix_S = gmx_mm_transpose_sum4_pr(fix_S0, fix_S1, fix_S2, fix_S3);
 679         gmx_simd4_store_r(f+scix, gmx_simd4_add_r(fix_S, gmx_simd4_load_r(f+scix)));
 680
 681         fiy_S = gmx_mm_transpose_sum4_pr(fiy_S0, fiy_S1, fiy_S2, fiy_S3);
 682         gmx_simd4_store_r(f+sciy, gmx_simd4_add_r(fiy_S, gmx_simd4_load_r(f+sciy)));
 683
 684         fiz_S = gmx_mm_transpose_sum4_pr(fiz_S0, fiz_S1, fiz_S2, fiz_S3);
 685         gmx_simd4_store_r(f+sciz, gmx_simd4_add_r(fiz_S, gmx_simd4_load_r(f+sciz)));
 686
 687 #ifdef CALC_SHIFTFORCES
 688         fshift[ish3+0] += gmx_simd4_reduce_r(fix_S);
 689         fshift[ish3+1] += gmx_simd4_reduce_r(fiy_S);
 690         fshift[ish3+2] += gmx_simd4_reduce_r(fiz_S);
 691 #endif
 692 #else
 693         fix0_S = gmx_mm_transpose_sum2_pr(fix_S0, fix_S1);
 694         gmx_simd_store_r(f+scix, gmx_simd_add_r(fix0_S, gmx_simd_load_r(f+scix)));
 695         fix2_S = gmx_mm_transpose_sum2_pr(fix_S2, fix_S3);
 696         gmx_simd_store_r(f+scix+2, gmx_simd_add_r(fix2_S, gmx_simd_load_r(f+scix+2)));
 697
 698         fiy0_S = gmx_mm_transpose_sum2_pr(fiy_S0, fiy_S1);
 699         gmx_simd_store_r(f+sciy, gmx_simd_add_r(fiy0_S, gmx_simd_load_r(f+sciy)));
 700         fiy2_S = gmx_mm_transpose_sum2_pr(fiy_S2, fiy_S3);
 701         gmx_simd_store_r(f+sciy+2, gmx_simd_add_r(fiy2_S, gmx_simd_load_r(f+sciy+2)));
 702
 703         fiz0_S = gmx_mm_transpose_sum2_pr(fiz_S0, fiz_S1);
 704         gmx_simd_store_r(f+sciz, gmx_simd_add_r(fiz0_S, gmx_simd_load_r(f+sciz)));
 705         fiz2_S = gmx_mm_transpose_sum2_pr(fiz_S2, fiz_S3);
 706         gmx_simd_store_r(f+sciz+2, gmx_simd_add_r(fiz2_S, gmx_simd_load_r(f+sciz+2)));
 707
 708 #ifdef CALC_SHIFTFORCES
 709         fshift[ish3+0] += gmx_simd_reduce_r(gmx_simd_add_r(fix0_S, fix2_S));
 710         fshift[ish3+1] += gmx_simd_reduce_r(gmx_simd_add_r(fiy0_S, fiy2_S));
 711         fshift[ish3+2] += gmx_simd_reduce_r(gmx_simd_add_r(fiz0_S, fiz2_S));
 712 #endif
 713 #endif
 714
 715 #ifdef CALC_ENERGIES
 716         if (do_coul)
 717         {
 718             *Vc += gmx_simd_reduce_r(vctot_S);
 719         }
 720
 721         *Vvdw += gmx_simd_reduce_r(Vvdwtot_S);
 722 #endif
 723
 724         /* Outer loop uses 6 flops/iteration */
 725     }
 726
 727 #ifdef COUNT_PAIRS
 728     printf("atom pairs %d\n", npair);
 729 #endif
 730 }