src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_256_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     real *           vdwioffsetptr1;
  86     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 101     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 102     __m256d          dummy_mask,cutoff_mask;
 103     __m128           tmpmask0,tmpmask1;
 104     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 105     __m256d          one     = _mm256_set1_pd(1.0);
 106     __m256d          two     = _mm256_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm256_set1_pd(fr->ic->epsfac);
 119     charge           = mdatoms->chargeA;
 120     krf              = _mm256_set1_pd(fr->ic->k_rf);
 121     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 122     crf              = _mm256_set1_pd(fr->ic->c_rf);
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 130     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 131     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 132     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 133
 134     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 135     rcutoff_scalar   = fr->ic->rcoulomb;
 136     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 137     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 138
 139     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 140     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 141
 142     /* Avoid stupid compiler warnings */
 143     jnrA = jnrB = jnrC = jnrD = 0;
 144     j_coord_offsetA = 0;
 145     j_coord_offsetB = 0;
 146     j_coord_offsetC = 0;
 147     j_coord_offsetD = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     for(iidx=0;iidx<4*DIM;iidx++)
 153     {
 154         scratch[iidx] = 0.0;
 155     }
 156
 157     /* Start outer loop over neighborlists */
 158     for(iidx=0; iidx<nri; iidx++)
 159     {
 160         /* Load shift vector for this list */
 161         i_shift_offset   = DIM*shiftidx[iidx];
 162
 163         /* Load limits for loop over neighbors */
 164         j_index_start    = jindex[iidx];
 165         j_index_end      = jindex[iidx+1];
 166
 167         /* Get outer coordinate index */
 168         inr              = iinr[iidx];
 169         i_coord_offset   = DIM*inr;
 170
 171         /* Load i particle coords and add shift vector */
 172         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 173                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 174
 175         fix0             = _mm256_setzero_pd();
 176         fiy0             = _mm256_setzero_pd();
 177         fiz0             = _mm256_setzero_pd();
 178         fix1             = _mm256_setzero_pd();
 179         fiy1             = _mm256_setzero_pd();
 180         fiz1             = _mm256_setzero_pd();
 181         fix2             = _mm256_setzero_pd();
 182         fiy2             = _mm256_setzero_pd();
 183         fiz2             = _mm256_setzero_pd();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm256_setzero_pd();
 187         vvdwsum          = _mm256_setzero_pd();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 206                                                  &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm256_sub_pd(ix0,jx0);
 210             dy00             = _mm256_sub_pd(iy0,jy0);
 211             dz00             = _mm256_sub_pd(iz0,jz0);
 212             dx10             = _mm256_sub_pd(ix1,jx0);
 213             dy10             = _mm256_sub_pd(iy1,jy0);
 214             dz10             = _mm256_sub_pd(iz1,jz0);
 215             dx20             = _mm256_sub_pd(ix2,jx0);
 216             dy20             = _mm256_sub_pd(iy2,jy0);
 217             dz20             = _mm256_sub_pd(iz2,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 221             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 222             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 223
 224             rinv00           = avx256_invsqrt_d(rsq00);
 225             rinv10           = avx256_invsqrt_d(rsq10);
 226             rinv20           = avx256_invsqrt_d(rsq20);
 227
 228             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 229             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 230             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 234                                                                  charge+jnrC+0,charge+jnrD+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239
 240             fjx0             = _mm256_setzero_pd();
 241             fjy0             = _mm256_setzero_pd();
 242             fjz0             = _mm256_setzero_pd();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 249             {
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq00             = _mm256_mul_pd(iq0,jq0);
 253             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 254                                             vdwioffsetptr0+vdwjidx0B,
 255                                             vdwioffsetptr0+vdwjidx0C,
 256                                             vdwioffsetptr0+vdwjidx0D,
 257                                             &c6_00,&c12_00);
 258
 259             /* REACTION-FIELD ELECTROSTATICS */
 260             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 261             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 262
 263             /* LENNARD-JONES DISPERSION/REPULSION */
 264
 265             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 266             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 267             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 268             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 269                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 270             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 271
 272             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velec            = _mm256_and_pd(velec,cutoff_mask);
 276             velecsum         = _mm256_add_pd(velecsum,velec);
 277             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 278             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 279
 280             fscal            = _mm256_add_pd(felec,fvdw);
 281
 282             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm256_mul_pd(fscal,dx00);
 286             ty               = _mm256_mul_pd(fscal,dy00);
 287             tz               = _mm256_mul_pd(fscal,dz00);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm256_add_pd(fix0,tx);
 291             fiy0             = _mm256_add_pd(fiy0,ty);
 292             fiz0             = _mm256_add_pd(fiz0,tz);
 293
 294             fjx0             = _mm256_add_pd(fjx0,tx);
 295             fjy0             = _mm256_add_pd(fjy0,ty);
 296             fjz0             = _mm256_add_pd(fjz0,tz);
 297
 298             }
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 305             {
 306
 307             /* Compute parameters for interactions between i and j atoms */
 308             qq10             = _mm256_mul_pd(iq1,jq0);
 309
 310             /* REACTION-FIELD ELECTROSTATICS */
 311             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 312             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 313
 314             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velec            = _mm256_and_pd(velec,cutoff_mask);
 318             velecsum         = _mm256_add_pd(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm256_mul_pd(fscal,dx10);
 326             ty               = _mm256_mul_pd(fscal,dy10);
 327             tz               = _mm256_mul_pd(fscal,dz10);
 328
 329             /* Update vectorial force */
 330             fix1             = _mm256_add_pd(fix1,tx);
 331             fiy1             = _mm256_add_pd(fiy1,ty);
 332             fiz1             = _mm256_add_pd(fiz1,tz);
 333
 334             fjx0             = _mm256_add_pd(fjx0,tx);
 335             fjy0             = _mm256_add_pd(fjy0,ty);
 336             fjz0             = _mm256_add_pd(fjz0,tz);
 337
 338             }
 339
 340             /**************************
 341              * CALCULATE INTERACTIONS *
 342              **************************/
 343
 344             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 345             {
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq20             = _mm256_mul_pd(iq2,jq0);
 349
 350             /* REACTION-FIELD ELECTROSTATICS */
 351             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 352             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 353
 354             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velec            = _mm256_and_pd(velec,cutoff_mask);
 358             velecsum         = _mm256_add_pd(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm256_mul_pd(fscal,dx20);
 366             ty               = _mm256_mul_pd(fscal,dy20);
 367             tz               = _mm256_mul_pd(fscal,dz20);
 368
 369             /* Update vectorial force */
 370             fix2             = _mm256_add_pd(fix2,tx);
 371             fiy2             = _mm256_add_pd(fiy2,ty);
 372             fiz2             = _mm256_add_pd(fiz2,tz);
 373
 374             fjx0             = _mm256_add_pd(fjx0,tx);
 375             fjy0             = _mm256_add_pd(fjy0,ty);
 376             fjz0             = _mm256_add_pd(fjz0,tz);
 377
 378             }
 379
 380             fjptrA             = f+j_coord_offsetA;
 381             fjptrB             = f+j_coord_offsetB;
 382             fjptrC             = f+j_coord_offsetC;
 383             fjptrD             = f+j_coord_offsetD;
 384
 385             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 386
 387             /* Inner loop uses 129 flops */
 388         }
 389
 390         if(jidx<j_index_end)
 391         {
 392
 393             /* Get j neighbor index, and coordinate index */
 394             jnrlistA         = jjnr[jidx];
 395             jnrlistB         = jjnr[jidx+1];
 396             jnrlistC         = jjnr[jidx+2];
 397             jnrlistD         = jjnr[jidx+3];
 398             /* Sign of each element will be negative for non-real atoms.
 399              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 400              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 401              */
 402             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 403
 404             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 405             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 406             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 407
 408             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 409             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 410             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 411             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 412             j_coord_offsetA  = DIM*jnrA;
 413             j_coord_offsetB  = DIM*jnrB;
 414             j_coord_offsetC  = DIM*jnrC;
 415             j_coord_offsetD  = DIM*jnrD;
 416
 417             /* load j atom coordinates */
 418             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 419                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 420                                                  &jx0,&jy0,&jz0);
 421
 422             /* Calculate displacement vector */
 423             dx00             = _mm256_sub_pd(ix0,jx0);
 424             dy00             = _mm256_sub_pd(iy0,jy0);
 425             dz00             = _mm256_sub_pd(iz0,jz0);
 426             dx10             = _mm256_sub_pd(ix1,jx0);
 427             dy10             = _mm256_sub_pd(iy1,jy0);
 428             dz10             = _mm256_sub_pd(iz1,jz0);
 429             dx20             = _mm256_sub_pd(ix2,jx0);
 430             dy20             = _mm256_sub_pd(iy2,jy0);
 431             dz20             = _mm256_sub_pd(iz2,jz0);
 432
 433             /* Calculate squared distance and things based on it */
 434             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 435             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 436             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 437
 438             rinv00           = avx256_invsqrt_d(rsq00);
 439             rinv10           = avx256_invsqrt_d(rsq10);
 440             rinv20           = avx256_invsqrt_d(rsq20);
 441
 442             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 443             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 444             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 445
 446             /* Load parameters for j particles */
 447             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 448                                                                  charge+jnrC+0,charge+jnrD+0);
 449             vdwjidx0A        = 2*vdwtype[jnrA+0];
 450             vdwjidx0B        = 2*vdwtype[jnrB+0];
 451             vdwjidx0C        = 2*vdwtype[jnrC+0];
 452             vdwjidx0D        = 2*vdwtype[jnrD+0];
 453
 454             fjx0             = _mm256_setzero_pd();
 455             fjy0             = _mm256_setzero_pd();
 456             fjz0             = _mm256_setzero_pd();
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 463             {
 464
 465             /* Compute parameters for interactions between i and j atoms */
 466             qq00             = _mm256_mul_pd(iq0,jq0);
 467             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 468                                             vdwioffsetptr0+vdwjidx0B,
 469                                             vdwioffsetptr0+vdwjidx0C,
 470                                             vdwioffsetptr0+vdwjidx0D,
 471                                             &c6_00,&c12_00);
 472
 473             /* REACTION-FIELD ELECTROSTATICS */
 474             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 475             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 476
 477             /* LENNARD-JONES DISPERSION/REPULSION */
 478
 479             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 480             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 481             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 482             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 483                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 484             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 485
 486             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velec            = _mm256_and_pd(velec,cutoff_mask);
 490             velec            = _mm256_andnot_pd(dummy_mask,velec);
 491             velecsum         = _mm256_add_pd(velecsum,velec);
 492             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 493             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 494             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 495
 496             fscal            = _mm256_add_pd(felec,fvdw);
 497
 498             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 499
 500             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 501
 502             /* Calculate temporary vectorial force */
 503             tx               = _mm256_mul_pd(fscal,dx00);
 504             ty               = _mm256_mul_pd(fscal,dy00);
 505             tz               = _mm256_mul_pd(fscal,dz00);
 506
 507             /* Update vectorial force */
 508             fix0             = _mm256_add_pd(fix0,tx);
 509             fiy0             = _mm256_add_pd(fiy0,ty);
 510             fiz0             = _mm256_add_pd(fiz0,tz);
 511
 512             fjx0             = _mm256_add_pd(fjx0,tx);
 513             fjy0             = _mm256_add_pd(fjy0,ty);
 514             fjz0             = _mm256_add_pd(fjz0,tz);
 515
 516             }
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 523             {
 524
 525             /* Compute parameters for interactions between i and j atoms */
 526             qq10             = _mm256_mul_pd(iq1,jq0);
 527
 528             /* REACTION-FIELD ELECTROSTATICS */
 529             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 530             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 531
 532             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm256_and_pd(velec,cutoff_mask);
 536             velec            = _mm256_andnot_pd(dummy_mask,velec);
 537             velecsum         = _mm256_add_pd(velecsum,velec);
 538
 539             fscal            = felec;
 540
 541             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 542
 543             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm256_mul_pd(fscal,dx10);
 547             ty               = _mm256_mul_pd(fscal,dy10);
 548             tz               = _mm256_mul_pd(fscal,dz10);
 549
 550             /* Update vectorial force */
 551             fix1             = _mm256_add_pd(fix1,tx);
 552             fiy1             = _mm256_add_pd(fiy1,ty);
 553             fiz1             = _mm256_add_pd(fiz1,tz);
 554
 555             fjx0             = _mm256_add_pd(fjx0,tx);
 556             fjy0             = _mm256_add_pd(fjy0,ty);
 557             fjz0             = _mm256_add_pd(fjz0,tz);
 558
 559             }
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 566             {
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq20             = _mm256_mul_pd(iq2,jq0);
 570
 571             /* REACTION-FIELD ELECTROSTATICS */
 572             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 573             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 574
 575             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_and_pd(velec,cutoff_mask);
 579             velec            = _mm256_andnot_pd(dummy_mask,velec);
 580             velecsum         = _mm256_add_pd(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 585
 586             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 587
 588             /* Calculate temporary vectorial force */
 589             tx               = _mm256_mul_pd(fscal,dx20);
 590             ty               = _mm256_mul_pd(fscal,dy20);
 591             tz               = _mm256_mul_pd(fscal,dz20);
 592
 593             /* Update vectorial force */
 594             fix2             = _mm256_add_pd(fix2,tx);
 595             fiy2             = _mm256_add_pd(fiy2,ty);
 596             fiz2             = _mm256_add_pd(fiz2,tz);
 597
 598             fjx0             = _mm256_add_pd(fjx0,tx);
 599             fjy0             = _mm256_add_pd(fjy0,ty);
 600             fjz0             = _mm256_add_pd(fjz0,tz);
 601
 602             }
 603
 604             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 605             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 606             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 607             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 608
 609             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 610
 611             /* Inner loop uses 129 flops */
 612         }
 613
 614         /* End of innermost loop */
 615
 616         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 617                                                  f+i_coord_offset,fshift+i_shift_offset);
 618
 619         ggid                        = gid[iidx];
 620         /* Update potential energies */
 621         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 622         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 623
 624         /* Increment number of inner iterations */
 625         inneriter                  += j_index_end - j_index_start;
 626
 627         /* Outer loop uses 20 flops */
 628     }
 629
 630     /* Increment number of outer iterations */
 631     outeriter        += nri;
 632
 633     /* Update outer/inner flops */
 634
 635     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*129);
 636 }
 637 /*
 638  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_double
 639  * Electrostatics interaction: ReactionField
 640  * VdW interaction:            LennardJones
 641  * Geometry:                   Water3-Particle
 642  * Calculate force/pot:        Force
 643  */
 644 void
 645 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_256_double
 646                     (t_nblist                    * gmx_restrict       nlist,
 647                      rvec                        * gmx_restrict          xx,
 648                      rvec                        * gmx_restrict          ff,
 649                      struct t_forcerec           * gmx_restrict          fr,
 650                      t_mdatoms                   * gmx_restrict     mdatoms,
 651                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 652                      t_nrnb                      * gmx_restrict        nrnb)
 653 {
 654     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 655      * just 0 for non-waters.
 656      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 657      * jnr indices corresponding to data put in the four positions in the SIMD register.
 658      */
 659     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 660     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 661     int              jnrA,jnrB,jnrC,jnrD;
 662     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 663     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 664     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 665     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 666     real             rcutoff_scalar;
 667     real             *shiftvec,*fshift,*x,*f;
 668     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 669     real             scratch[4*DIM];
 670     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 671     real *           vdwioffsetptr0;
 672     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 673     real *           vdwioffsetptr1;
 674     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 675     real *           vdwioffsetptr2;
 676     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 677     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 678     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 679     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 680     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 681     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 682     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 683     real             *charge;
 684     int              nvdwtype;
 685     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 686     int              *vdwtype;
 687     real             *vdwparam;
 688     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 689     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 690     __m256d          dummy_mask,cutoff_mask;
 691     __m128           tmpmask0,tmpmask1;
 692     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 693     __m256d          one     = _mm256_set1_pd(1.0);
 694     __m256d          two     = _mm256_set1_pd(2.0);
 695     x                = xx[0];
 696     f                = ff[0];
 697
 698     nri              = nlist->nri;
 699     iinr             = nlist->iinr;
 700     jindex           = nlist->jindex;
 701     jjnr             = nlist->jjnr;
 702     shiftidx         = nlist->shift;
 703     gid              = nlist->gid;
 704     shiftvec         = fr->shift_vec[0];
 705     fshift           = fr->fshift[0];
 706     facel            = _mm256_set1_pd(fr->ic->epsfac);
 707     charge           = mdatoms->chargeA;
 708     krf              = _mm256_set1_pd(fr->ic->k_rf);
 709     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 710     crf              = _mm256_set1_pd(fr->ic->c_rf);
 711     nvdwtype         = fr->ntype;
 712     vdwparam         = fr->nbfp;
 713     vdwtype          = mdatoms->typeA;
 714
 715     /* Setup water-specific parameters */
 716     inr              = nlist->iinr[0];
 717     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 718     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 719     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 720     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 721
 722     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 723     rcutoff_scalar   = fr->ic->rcoulomb;
 724     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 725     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 726
 727     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 728     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 729
 730     /* Avoid stupid compiler warnings */
 731     jnrA = jnrB = jnrC = jnrD = 0;
 732     j_coord_offsetA = 0;
 733     j_coord_offsetB = 0;
 734     j_coord_offsetC = 0;
 735     j_coord_offsetD = 0;
 736
 737     outeriter        = 0;
 738     inneriter        = 0;
 739
 740     for(iidx=0;iidx<4*DIM;iidx++)
 741     {
 742         scratch[iidx] = 0.0;
 743     }
 744
 745     /* Start outer loop over neighborlists */
 746     for(iidx=0; iidx<nri; iidx++)
 747     {
 748         /* Load shift vector for this list */
 749         i_shift_offset   = DIM*shiftidx[iidx];
 750
 751         /* Load limits for loop over neighbors */
 752         j_index_start    = jindex[iidx];
 753         j_index_end      = jindex[iidx+1];
 754
 755         /* Get outer coordinate index */
 756         inr              = iinr[iidx];
 757         i_coord_offset   = DIM*inr;
 758
 759         /* Load i particle coords and add shift vector */
 760         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 761                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 762
 763         fix0             = _mm256_setzero_pd();
 764         fiy0             = _mm256_setzero_pd();
 765         fiz0             = _mm256_setzero_pd();
 766         fix1             = _mm256_setzero_pd();
 767         fiy1             = _mm256_setzero_pd();
 768         fiz1             = _mm256_setzero_pd();
 769         fix2             = _mm256_setzero_pd();
 770         fiy2             = _mm256_setzero_pd();
 771         fiz2             = _mm256_setzero_pd();
 772
 773         /* Start inner kernel loop */
 774         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 775         {
 776
 777             /* Get j neighbor index, and coordinate index */
 778             jnrA             = jjnr[jidx];
 779             jnrB             = jjnr[jidx+1];
 780             jnrC             = jjnr[jidx+2];
 781             jnrD             = jjnr[jidx+3];
 782             j_coord_offsetA  = DIM*jnrA;
 783             j_coord_offsetB  = DIM*jnrB;
 784             j_coord_offsetC  = DIM*jnrC;
 785             j_coord_offsetD  = DIM*jnrD;
 786
 787             /* load j atom coordinates */
 788             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 789                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 790                                                  &jx0,&jy0,&jz0);
 791
 792             /* Calculate displacement vector */
 793             dx00             = _mm256_sub_pd(ix0,jx0);
 794             dy00             = _mm256_sub_pd(iy0,jy0);
 795             dz00             = _mm256_sub_pd(iz0,jz0);
 796             dx10             = _mm256_sub_pd(ix1,jx0);
 797             dy10             = _mm256_sub_pd(iy1,jy0);
 798             dz10             = _mm256_sub_pd(iz1,jz0);
 799             dx20             = _mm256_sub_pd(ix2,jx0);
 800             dy20             = _mm256_sub_pd(iy2,jy0);
 801             dz20             = _mm256_sub_pd(iz2,jz0);
 802
 803             /* Calculate squared distance and things based on it */
 804             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 805             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 806             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 807
 808             rinv00           = avx256_invsqrt_d(rsq00);
 809             rinv10           = avx256_invsqrt_d(rsq10);
 810             rinv20           = avx256_invsqrt_d(rsq20);
 811
 812             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 813             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 814             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 815
 816             /* Load parameters for j particles */
 817             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 818                                                                  charge+jnrC+0,charge+jnrD+0);
 819             vdwjidx0A        = 2*vdwtype[jnrA+0];
 820             vdwjidx0B        = 2*vdwtype[jnrB+0];
 821             vdwjidx0C        = 2*vdwtype[jnrC+0];
 822             vdwjidx0D        = 2*vdwtype[jnrD+0];
 823
 824             fjx0             = _mm256_setzero_pd();
 825             fjy0             = _mm256_setzero_pd();
 826             fjz0             = _mm256_setzero_pd();
 827
 828             /**************************
 829              * CALCULATE INTERACTIONS *
 830              **************************/
 831
 832             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 833             {
 834
 835             /* Compute parameters for interactions between i and j atoms */
 836             qq00             = _mm256_mul_pd(iq0,jq0);
 837             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 838                                             vdwioffsetptr0+vdwjidx0B,
 839                                             vdwioffsetptr0+vdwjidx0C,
 840                                             vdwioffsetptr0+vdwjidx0D,
 841                                             &c6_00,&c12_00);
 842
 843             /* REACTION-FIELD ELECTROSTATICS */
 844             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 845
 846             /* LENNARD-JONES DISPERSION/REPULSION */
 847
 848             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 849             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 850
 851             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 852
 853             fscal            = _mm256_add_pd(felec,fvdw);
 854
 855             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 856
 857             /* Calculate temporary vectorial force */
 858             tx               = _mm256_mul_pd(fscal,dx00);
 859             ty               = _mm256_mul_pd(fscal,dy00);
 860             tz               = _mm256_mul_pd(fscal,dz00);
 861
 862             /* Update vectorial force */
 863             fix0             = _mm256_add_pd(fix0,tx);
 864             fiy0             = _mm256_add_pd(fiy0,ty);
 865             fiz0             = _mm256_add_pd(fiz0,tz);
 866
 867             fjx0             = _mm256_add_pd(fjx0,tx);
 868             fjy0             = _mm256_add_pd(fjy0,ty);
 869             fjz0             = _mm256_add_pd(fjz0,tz);
 870
 871             }
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 878             {
 879
 880             /* Compute parameters for interactions between i and j atoms */
 881             qq10             = _mm256_mul_pd(iq1,jq0);
 882
 883             /* REACTION-FIELD ELECTROSTATICS */
 884             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 885
 886             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 887
 888             fscal            = felec;
 889
 890             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 891
 892             /* Calculate temporary vectorial force */
 893             tx               = _mm256_mul_pd(fscal,dx10);
 894             ty               = _mm256_mul_pd(fscal,dy10);
 895             tz               = _mm256_mul_pd(fscal,dz10);
 896
 897             /* Update vectorial force */
 898             fix1             = _mm256_add_pd(fix1,tx);
 899             fiy1             = _mm256_add_pd(fiy1,ty);
 900             fiz1             = _mm256_add_pd(fiz1,tz);
 901
 902             fjx0             = _mm256_add_pd(fjx0,tx);
 903             fjy0             = _mm256_add_pd(fjy0,ty);
 904             fjz0             = _mm256_add_pd(fjz0,tz);
 905
 906             }
 907
 908             /**************************
 909              * CALCULATE INTERACTIONS *
 910              **************************/
 911
 912             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 913             {
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq20             = _mm256_mul_pd(iq2,jq0);
 917
 918             /* REACTION-FIELD ELECTROSTATICS */
 919             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 920
 921             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 922
 923             fscal            = felec;
 924
 925             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 926
 927             /* Calculate temporary vectorial force */
 928             tx               = _mm256_mul_pd(fscal,dx20);
 929             ty               = _mm256_mul_pd(fscal,dy20);
 930             tz               = _mm256_mul_pd(fscal,dz20);
 931
 932             /* Update vectorial force */
 933             fix2             = _mm256_add_pd(fix2,tx);
 934             fiy2             = _mm256_add_pd(fiy2,ty);
 935             fiz2             = _mm256_add_pd(fiz2,tz);
 936
 937             fjx0             = _mm256_add_pd(fjx0,tx);
 938             fjy0             = _mm256_add_pd(fjy0,ty);
 939             fjz0             = _mm256_add_pd(fjz0,tz);
 940
 941             }
 942
 943             fjptrA             = f+j_coord_offsetA;
 944             fjptrB             = f+j_coord_offsetB;
 945             fjptrC             = f+j_coord_offsetC;
 946             fjptrD             = f+j_coord_offsetD;
 947
 948             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 949
 950             /* Inner loop uses 100 flops */
 951         }
 952
 953         if(jidx<j_index_end)
 954         {
 955
 956             /* Get j neighbor index, and coordinate index */
 957             jnrlistA         = jjnr[jidx];
 958             jnrlistB         = jjnr[jidx+1];
 959             jnrlistC         = jjnr[jidx+2];
 960             jnrlistD         = jjnr[jidx+3];
 961             /* Sign of each element will be negative for non-real atoms.
 962              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 963              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 964              */
 965             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 966
 967             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 968             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 969             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 970
 971             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 972             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 973             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 974             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 975             j_coord_offsetA  = DIM*jnrA;
 976             j_coord_offsetB  = DIM*jnrB;
 977             j_coord_offsetC  = DIM*jnrC;
 978             j_coord_offsetD  = DIM*jnrD;
 979
 980             /* load j atom coordinates */
 981             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 982                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 983                                                  &jx0,&jy0,&jz0);
 984
 985             /* Calculate displacement vector */
 986             dx00             = _mm256_sub_pd(ix0,jx0);
 987             dy00             = _mm256_sub_pd(iy0,jy0);
 988             dz00             = _mm256_sub_pd(iz0,jz0);
 989             dx10             = _mm256_sub_pd(ix1,jx0);
 990             dy10             = _mm256_sub_pd(iy1,jy0);
 991             dz10             = _mm256_sub_pd(iz1,jz0);
 992             dx20             = _mm256_sub_pd(ix2,jx0);
 993             dy20             = _mm256_sub_pd(iy2,jy0);
 994             dz20             = _mm256_sub_pd(iz2,jz0);
 995
 996             /* Calculate squared distance and things based on it */
 997             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 998             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 999             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1000
1001             rinv00           = avx256_invsqrt_d(rsq00);
1002             rinv10           = avx256_invsqrt_d(rsq10);
1003             rinv20           = avx256_invsqrt_d(rsq20);
1004
1005             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1006             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1007             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1008
1009             /* Load parameters for j particles */
1010             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1011                                                                  charge+jnrC+0,charge+jnrD+0);
1012             vdwjidx0A        = 2*vdwtype[jnrA+0];
1013             vdwjidx0B        = 2*vdwtype[jnrB+0];
1014             vdwjidx0C        = 2*vdwtype[jnrC+0];
1015             vdwjidx0D        = 2*vdwtype[jnrD+0];
1016
1017             fjx0             = _mm256_setzero_pd();
1018             fjy0             = _mm256_setzero_pd();
1019             fjz0             = _mm256_setzero_pd();
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1026             {
1027
1028             /* Compute parameters for interactions between i and j atoms */
1029             qq00             = _mm256_mul_pd(iq0,jq0);
1030             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1031                                             vdwioffsetptr0+vdwjidx0B,
1032                                             vdwioffsetptr0+vdwjidx0C,
1033                                             vdwioffsetptr0+vdwjidx0D,
1034                                             &c6_00,&c12_00);
1035
1036             /* REACTION-FIELD ELECTROSTATICS */
1037             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
1038
1039             /* LENNARD-JONES DISPERSION/REPULSION */
1040
1041             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1042             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1043
1044             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1045
1046             fscal            = _mm256_add_pd(felec,fvdw);
1047
1048             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1049
1050             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1051
1052             /* Calculate temporary vectorial force */
1053             tx               = _mm256_mul_pd(fscal,dx00);
1054             ty               = _mm256_mul_pd(fscal,dy00);
1055             tz               = _mm256_mul_pd(fscal,dz00);
1056
1057             /* Update vectorial force */
1058             fix0             = _mm256_add_pd(fix0,tx);
1059             fiy0             = _mm256_add_pd(fiy0,ty);
1060             fiz0             = _mm256_add_pd(fiz0,tz);
1061
1062             fjx0             = _mm256_add_pd(fjx0,tx);
1063             fjy0             = _mm256_add_pd(fjy0,ty);
1064             fjz0             = _mm256_add_pd(fjz0,tz);
1065
1066             }
1067
1068             /**************************
1069              * CALCULATE INTERACTIONS *
1070              **************************/
1071
1072             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1073             {
1074
1075             /* Compute parameters for interactions between i and j atoms */
1076             qq10             = _mm256_mul_pd(iq1,jq0);
1077
1078             /* REACTION-FIELD ELECTROSTATICS */
1079             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1080
1081             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1082
1083             fscal            = felec;
1084
1085             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1086
1087             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm256_mul_pd(fscal,dx10);
1091             ty               = _mm256_mul_pd(fscal,dy10);
1092             tz               = _mm256_mul_pd(fscal,dz10);
1093
1094             /* Update vectorial force */
1095             fix1             = _mm256_add_pd(fix1,tx);
1096             fiy1             = _mm256_add_pd(fiy1,ty);
1097             fiz1             = _mm256_add_pd(fiz1,tz);
1098
1099             fjx0             = _mm256_add_pd(fjx0,tx);
1100             fjy0             = _mm256_add_pd(fjy0,ty);
1101             fjz0             = _mm256_add_pd(fjz0,tz);
1102
1103             }
1104
1105             /**************************
1106              * CALCULATE INTERACTIONS *
1107              **************************/
1108
1109             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1110             {
1111
1112             /* Compute parameters for interactions between i and j atoms */
1113             qq20             = _mm256_mul_pd(iq2,jq0);
1114
1115             /* REACTION-FIELD ELECTROSTATICS */
1116             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1117
1118             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1119
1120             fscal            = felec;
1121
1122             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1123
1124             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1125
1126             /* Calculate temporary vectorial force */
1127             tx               = _mm256_mul_pd(fscal,dx20);
1128             ty               = _mm256_mul_pd(fscal,dy20);
1129             tz               = _mm256_mul_pd(fscal,dz20);
1130
1131             /* Update vectorial force */
1132             fix2             = _mm256_add_pd(fix2,tx);
1133             fiy2             = _mm256_add_pd(fiy2,ty);
1134             fiz2             = _mm256_add_pd(fiz2,tz);
1135
1136             fjx0             = _mm256_add_pd(fjx0,tx);
1137             fjy0             = _mm256_add_pd(fjy0,ty);
1138             fjz0             = _mm256_add_pd(fjz0,tz);
1139
1140             }
1141
1142             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1143             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1144             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1145             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1146
1147             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1148
1149             /* Inner loop uses 100 flops */
1150         }
1151
1152         /* End of innermost loop */
1153
1154         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1155                                                  f+i_coord_offset,fshift+i_shift_offset);
1156
1157         /* Increment number of inner iterations */
1158         inneriter                  += j_index_end - j_index_start;
1159
1160         /* Outer loop uses 18 flops */
1161     }
1162
1163     /* Increment number of outer iterations */
1164     outeriter        += nri;
1165
1166     /* Update outer/inner flops */
1167
1168     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
1169 }