src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_avx_256_single.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_single 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_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 103     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 104     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 105     real             rswitch_scalar,d_scalar;
 106     __m256           dummy_mask,cutoff_mask;
 107     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 108     __m256           one     = _mm256_set1_ps(1.0);
 109     __m256           two     = _mm256_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_ps(fr->ic->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm256_set1_ps(fr->ic->k_rf);
 124     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 125     crf              = _mm256_set1_ps(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 133     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 134     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 135     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->ic->rcoulomb;
 139     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 140     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 141
 142     rswitch_scalar   = fr->ic->rvdw_switch;
 143     rswitch          = _mm256_set1_ps(rswitch_scalar);
 144     /* Setup switch parameters */
 145     d_scalar         = rcutoff_scalar-rswitch_scalar;
 146     d                = _mm256_set1_ps(d_scalar);
 147     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 148     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 149     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 150     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 151     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 152     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158     j_coord_offsetC = 0;
 159     j_coord_offsetD = 0;
 160     j_coord_offsetE = 0;
 161     j_coord_offsetF = 0;
 162     j_coord_offsetG = 0;
 163     j_coord_offsetH = 0;
 164
 165     outeriter        = 0;
 166     inneriter        = 0;
 167
 168     for(iidx=0;iidx<4*DIM;iidx++)
 169     {
 170         scratch[iidx] = 0.0;
 171     }
 172
 173     /* Start outer loop over neighborlists */
 174     for(iidx=0; iidx<nri; iidx++)
 175     {
 176         /* Load shift vector for this list */
 177         i_shift_offset   = DIM*shiftidx[iidx];
 178
 179         /* Load limits for loop over neighbors */
 180         j_index_start    = jindex[iidx];
 181         j_index_end      = jindex[iidx+1];
 182
 183         /* Get outer coordinate index */
 184         inr              = iinr[iidx];
 185         i_coord_offset   = DIM*inr;
 186
 187         /* Load i particle coords and add shift vector */
 188         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 189                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 190
 191         fix0             = _mm256_setzero_ps();
 192         fiy0             = _mm256_setzero_ps();
 193         fiz0             = _mm256_setzero_ps();
 194         fix1             = _mm256_setzero_ps();
 195         fiy1             = _mm256_setzero_ps();
 196         fiz1             = _mm256_setzero_ps();
 197         fix2             = _mm256_setzero_ps();
 198         fiy2             = _mm256_setzero_ps();
 199         fiz2             = _mm256_setzero_ps();
 200
 201         /* Reset potential sums */
 202         velecsum         = _mm256_setzero_ps();
 203         vvdwsum          = _mm256_setzero_ps();
 204
 205         /* Start inner kernel loop */
 206         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 207         {
 208
 209             /* Get j neighbor index, and coordinate index */
 210             jnrA             = jjnr[jidx];
 211             jnrB             = jjnr[jidx+1];
 212             jnrC             = jjnr[jidx+2];
 213             jnrD             = jjnr[jidx+3];
 214             jnrE             = jjnr[jidx+4];
 215             jnrF             = jjnr[jidx+5];
 216             jnrG             = jjnr[jidx+6];
 217             jnrH             = jjnr[jidx+7];
 218             j_coord_offsetA  = DIM*jnrA;
 219             j_coord_offsetB  = DIM*jnrB;
 220             j_coord_offsetC  = DIM*jnrC;
 221             j_coord_offsetD  = DIM*jnrD;
 222             j_coord_offsetE  = DIM*jnrE;
 223             j_coord_offsetF  = DIM*jnrF;
 224             j_coord_offsetG  = DIM*jnrG;
 225             j_coord_offsetH  = DIM*jnrH;
 226
 227             /* load j atom coordinates */
 228             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 229                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 230                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 231                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 232                                                  &jx0,&jy0,&jz0);
 233
 234             /* Calculate displacement vector */
 235             dx00             = _mm256_sub_ps(ix0,jx0);
 236             dy00             = _mm256_sub_ps(iy0,jy0);
 237             dz00             = _mm256_sub_ps(iz0,jz0);
 238             dx10             = _mm256_sub_ps(ix1,jx0);
 239             dy10             = _mm256_sub_ps(iy1,jy0);
 240             dz10             = _mm256_sub_ps(iz1,jz0);
 241             dx20             = _mm256_sub_ps(ix2,jx0);
 242             dy20             = _mm256_sub_ps(iy2,jy0);
 243             dz20             = _mm256_sub_ps(iz2,jz0);
 244
 245             /* Calculate squared distance and things based on it */
 246             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 247             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 248             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 249
 250             rinv00           = avx256_invsqrt_f(rsq00);
 251             rinv10           = avx256_invsqrt_f(rsq10);
 252             rinv20           = avx256_invsqrt_f(rsq20);
 253
 254             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 255             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 256             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 257
 258             /* Load parameters for j particles */
 259             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 260                                                                  charge+jnrC+0,charge+jnrD+0,
 261                                                                  charge+jnrE+0,charge+jnrF+0,
 262                                                                  charge+jnrG+0,charge+jnrH+0);
 263             vdwjidx0A        = 2*vdwtype[jnrA+0];
 264             vdwjidx0B        = 2*vdwtype[jnrB+0];
 265             vdwjidx0C        = 2*vdwtype[jnrC+0];
 266             vdwjidx0D        = 2*vdwtype[jnrD+0];
 267             vdwjidx0E        = 2*vdwtype[jnrE+0];
 268             vdwjidx0F        = 2*vdwtype[jnrF+0];
 269             vdwjidx0G        = 2*vdwtype[jnrG+0];
 270             vdwjidx0H        = 2*vdwtype[jnrH+0];
 271
 272             fjx0             = _mm256_setzero_ps();
 273             fjy0             = _mm256_setzero_ps();
 274             fjz0             = _mm256_setzero_ps();
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 281             {
 282
 283             r00              = _mm256_mul_ps(rsq00,rinv00);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq00             = _mm256_mul_ps(iq0,jq0);
 287             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 288                                             vdwioffsetptr0+vdwjidx0B,
 289                                             vdwioffsetptr0+vdwjidx0C,
 290                                             vdwioffsetptr0+vdwjidx0D,
 291                                             vdwioffsetptr0+vdwjidx0E,
 292                                             vdwioffsetptr0+vdwjidx0F,
 293                                             vdwioffsetptr0+vdwjidx0G,
 294                                             vdwioffsetptr0+vdwjidx0H,
 295                                             &c6_00,&c12_00);
 296
 297             /* REACTION-FIELD ELECTROSTATICS */
 298             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 299             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 300
 301             /* LENNARD-JONES DISPERSION/REPULSION */
 302
 303             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 304             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 305             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 306             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 307             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 308
 309             d                = _mm256_sub_ps(r00,rswitch);
 310             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 311             d2               = _mm256_mul_ps(d,d);
 312             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 313
 314             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 315
 316             /* Evaluate switch function */
 317             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 318             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 319             vvdw             = _mm256_mul_ps(vvdw,sw);
 320             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velec            = _mm256_and_ps(velec,cutoff_mask);
 324             velecsum         = _mm256_add_ps(velecsum,velec);
 325             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 326             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 327
 328             fscal            = _mm256_add_ps(felec,fvdw);
 329
 330             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm256_mul_ps(fscal,dx00);
 334             ty               = _mm256_mul_ps(fscal,dy00);
 335             tz               = _mm256_mul_ps(fscal,dz00);
 336
 337             /* Update vectorial force */
 338             fix0             = _mm256_add_ps(fix0,tx);
 339             fiy0             = _mm256_add_ps(fiy0,ty);
 340             fiz0             = _mm256_add_ps(fiz0,tz);
 341
 342             fjx0             = _mm256_add_ps(fjx0,tx);
 343             fjy0             = _mm256_add_ps(fjy0,ty);
 344             fjz0             = _mm256_add_ps(fjz0,tz);
 345
 346             }
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 353             {
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq10             = _mm256_mul_ps(iq1,jq0);
 357
 358             /* REACTION-FIELD ELECTROSTATICS */
 359             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 360             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 361
 362             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velec            = _mm256_and_ps(velec,cutoff_mask);
 366             velecsum         = _mm256_add_ps(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm256_mul_ps(fscal,dx10);
 374             ty               = _mm256_mul_ps(fscal,dy10);
 375             tz               = _mm256_mul_ps(fscal,dz10);
 376
 377             /* Update vectorial force */
 378             fix1             = _mm256_add_ps(fix1,tx);
 379             fiy1             = _mm256_add_ps(fiy1,ty);
 380             fiz1             = _mm256_add_ps(fiz1,tz);
 381
 382             fjx0             = _mm256_add_ps(fjx0,tx);
 383             fjy0             = _mm256_add_ps(fjy0,ty);
 384             fjz0             = _mm256_add_ps(fjz0,tz);
 385
 386             }
 387
 388             /**************************
 389              * CALCULATE INTERACTIONS *
 390              **************************/
 391
 392             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 393             {
 394
 395             /* Compute parameters for interactions between i and j atoms */
 396             qq20             = _mm256_mul_ps(iq2,jq0);
 397
 398             /* REACTION-FIELD ELECTROSTATICS */
 399             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 400             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 401
 402             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm256_and_ps(velec,cutoff_mask);
 406             velecsum         = _mm256_add_ps(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 411
 412             /* Calculate temporary vectorial force */
 413             tx               = _mm256_mul_ps(fscal,dx20);
 414             ty               = _mm256_mul_ps(fscal,dy20);
 415             tz               = _mm256_mul_ps(fscal,dz20);
 416
 417             /* Update vectorial force */
 418             fix2             = _mm256_add_ps(fix2,tx);
 419             fiy2             = _mm256_add_ps(fiy2,ty);
 420             fiz2             = _mm256_add_ps(fiz2,tz);
 421
 422             fjx0             = _mm256_add_ps(fjx0,tx);
 423             fjy0             = _mm256_add_ps(fjy0,ty);
 424             fjz0             = _mm256_add_ps(fjz0,tz);
 425
 426             }
 427
 428             fjptrA             = f+j_coord_offsetA;
 429             fjptrB             = f+j_coord_offsetB;
 430             fjptrC             = f+j_coord_offsetC;
 431             fjptrD             = f+j_coord_offsetD;
 432             fjptrE             = f+j_coord_offsetE;
 433             fjptrF             = f+j_coord_offsetF;
 434             fjptrG             = f+j_coord_offsetG;
 435             fjptrH             = f+j_coord_offsetH;
 436
 437             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 438
 439             /* Inner loop uses 145 flops */
 440         }
 441
 442         if(jidx<j_index_end)
 443         {
 444
 445             /* Get j neighbor index, and coordinate index */
 446             jnrlistA         = jjnr[jidx];
 447             jnrlistB         = jjnr[jidx+1];
 448             jnrlistC         = jjnr[jidx+2];
 449             jnrlistD         = jjnr[jidx+3];
 450             jnrlistE         = jjnr[jidx+4];
 451             jnrlistF         = jjnr[jidx+5];
 452             jnrlistG         = jjnr[jidx+6];
 453             jnrlistH         = jjnr[jidx+7];
 454             /* Sign of each element will be negative for non-real atoms.
 455              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 456              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 457              */
 458             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 459                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 460
 461             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 462             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 463             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 464             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 465             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 466             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 467             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 468             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 469             j_coord_offsetA  = DIM*jnrA;
 470             j_coord_offsetB  = DIM*jnrB;
 471             j_coord_offsetC  = DIM*jnrC;
 472             j_coord_offsetD  = DIM*jnrD;
 473             j_coord_offsetE  = DIM*jnrE;
 474             j_coord_offsetF  = DIM*jnrF;
 475             j_coord_offsetG  = DIM*jnrG;
 476             j_coord_offsetH  = DIM*jnrH;
 477
 478             /* load j atom coordinates */
 479             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 480                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 481                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 482                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 483                                                  &jx0,&jy0,&jz0);
 484
 485             /* Calculate displacement vector */
 486             dx00             = _mm256_sub_ps(ix0,jx0);
 487             dy00             = _mm256_sub_ps(iy0,jy0);
 488             dz00             = _mm256_sub_ps(iz0,jz0);
 489             dx10             = _mm256_sub_ps(ix1,jx0);
 490             dy10             = _mm256_sub_ps(iy1,jy0);
 491             dz10             = _mm256_sub_ps(iz1,jz0);
 492             dx20             = _mm256_sub_ps(ix2,jx0);
 493             dy20             = _mm256_sub_ps(iy2,jy0);
 494             dz20             = _mm256_sub_ps(iz2,jz0);
 495
 496             /* Calculate squared distance and things based on it */
 497             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 498             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 499             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 500
 501             rinv00           = avx256_invsqrt_f(rsq00);
 502             rinv10           = avx256_invsqrt_f(rsq10);
 503             rinv20           = avx256_invsqrt_f(rsq20);
 504
 505             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 506             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 507             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 508
 509             /* Load parameters for j particles */
 510             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 511                                                                  charge+jnrC+0,charge+jnrD+0,
 512                                                                  charge+jnrE+0,charge+jnrF+0,
 513                                                                  charge+jnrG+0,charge+jnrH+0);
 514             vdwjidx0A        = 2*vdwtype[jnrA+0];
 515             vdwjidx0B        = 2*vdwtype[jnrB+0];
 516             vdwjidx0C        = 2*vdwtype[jnrC+0];
 517             vdwjidx0D        = 2*vdwtype[jnrD+0];
 518             vdwjidx0E        = 2*vdwtype[jnrE+0];
 519             vdwjidx0F        = 2*vdwtype[jnrF+0];
 520             vdwjidx0G        = 2*vdwtype[jnrG+0];
 521             vdwjidx0H        = 2*vdwtype[jnrH+0];
 522
 523             fjx0             = _mm256_setzero_ps();
 524             fjy0             = _mm256_setzero_ps();
 525             fjz0             = _mm256_setzero_ps();
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 532             {
 533
 534             r00              = _mm256_mul_ps(rsq00,rinv00);
 535             r00              = _mm256_andnot_ps(dummy_mask,r00);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq00             = _mm256_mul_ps(iq0,jq0);
 539             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 540                                             vdwioffsetptr0+vdwjidx0B,
 541                                             vdwioffsetptr0+vdwjidx0C,
 542                                             vdwioffsetptr0+vdwjidx0D,
 543                                             vdwioffsetptr0+vdwjidx0E,
 544                                             vdwioffsetptr0+vdwjidx0F,
 545                                             vdwioffsetptr0+vdwjidx0G,
 546                                             vdwioffsetptr0+vdwjidx0H,
 547                                             &c6_00,&c12_00);
 548
 549             /* REACTION-FIELD ELECTROSTATICS */
 550             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 551             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 552
 553             /* LENNARD-JONES DISPERSION/REPULSION */
 554
 555             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 556             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 557             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 558             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 559             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 560
 561             d                = _mm256_sub_ps(r00,rswitch);
 562             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 563             d2               = _mm256_mul_ps(d,d);
 564             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 565
 566             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 567
 568             /* Evaluate switch function */
 569             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 570             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 571             vvdw             = _mm256_mul_ps(vvdw,sw);
 572             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm256_and_ps(velec,cutoff_mask);
 576             velec            = _mm256_andnot_ps(dummy_mask,velec);
 577             velecsum         = _mm256_add_ps(velecsum,velec);
 578             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 579             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 580             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 581
 582             fscal            = _mm256_add_ps(felec,fvdw);
 583
 584             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 585
 586             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 587
 588             /* Calculate temporary vectorial force */
 589             tx               = _mm256_mul_ps(fscal,dx00);
 590             ty               = _mm256_mul_ps(fscal,dy00);
 591             tz               = _mm256_mul_ps(fscal,dz00);
 592
 593             /* Update vectorial force */
 594             fix0             = _mm256_add_ps(fix0,tx);
 595             fiy0             = _mm256_add_ps(fiy0,ty);
 596             fiz0             = _mm256_add_ps(fiz0,tz);
 597
 598             fjx0             = _mm256_add_ps(fjx0,tx);
 599             fjy0             = _mm256_add_ps(fjy0,ty);
 600             fjz0             = _mm256_add_ps(fjz0,tz);
 601
 602             }
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 609             {
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq10             = _mm256_mul_ps(iq1,jq0);
 613
 614             /* REACTION-FIELD ELECTROSTATICS */
 615             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 616             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 617
 618             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 619
 620             /* Update potential sum for this i atom from the interaction with this j atom. */
 621             velec            = _mm256_and_ps(velec,cutoff_mask);
 622             velec            = _mm256_andnot_ps(dummy_mask,velec);
 623             velecsum         = _mm256_add_ps(velecsum,velec);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 628
 629             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 630
 631             /* Calculate temporary vectorial force */
 632             tx               = _mm256_mul_ps(fscal,dx10);
 633             ty               = _mm256_mul_ps(fscal,dy10);
 634             tz               = _mm256_mul_ps(fscal,dz10);
 635
 636             /* Update vectorial force */
 637             fix1             = _mm256_add_ps(fix1,tx);
 638             fiy1             = _mm256_add_ps(fiy1,ty);
 639             fiz1             = _mm256_add_ps(fiz1,tz);
 640
 641             fjx0             = _mm256_add_ps(fjx0,tx);
 642             fjy0             = _mm256_add_ps(fjy0,ty);
 643             fjz0             = _mm256_add_ps(fjz0,tz);
 644
 645             }
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 652             {
 653
 654             /* Compute parameters for interactions between i and j atoms */
 655             qq20             = _mm256_mul_ps(iq2,jq0);
 656
 657             /* REACTION-FIELD ELECTROSTATICS */
 658             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 659             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 660
 661             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 662
 663             /* Update potential sum for this i atom from the interaction with this j atom. */
 664             velec            = _mm256_and_ps(velec,cutoff_mask);
 665             velec            = _mm256_andnot_ps(dummy_mask,velec);
 666             velecsum         = _mm256_add_ps(velecsum,velec);
 667
 668             fscal            = felec;
 669
 670             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 671
 672             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 673
 674             /* Calculate temporary vectorial force */
 675             tx               = _mm256_mul_ps(fscal,dx20);
 676             ty               = _mm256_mul_ps(fscal,dy20);
 677             tz               = _mm256_mul_ps(fscal,dz20);
 678
 679             /* Update vectorial force */
 680             fix2             = _mm256_add_ps(fix2,tx);
 681             fiy2             = _mm256_add_ps(fiy2,ty);
 682             fiz2             = _mm256_add_ps(fiz2,tz);
 683
 684             fjx0             = _mm256_add_ps(fjx0,tx);
 685             fjy0             = _mm256_add_ps(fjy0,ty);
 686             fjz0             = _mm256_add_ps(fjz0,tz);
 687
 688             }
 689
 690             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 691             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 692             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 693             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 694             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 695             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 696             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 697             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 698
 699             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 700
 701             /* Inner loop uses 146 flops */
 702         }
 703
 704         /* End of innermost loop */
 705
 706         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 707                                                  f+i_coord_offset,fshift+i_shift_offset);
 708
 709         ggid                        = gid[iidx];
 710         /* Update potential energies */
 711         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 712         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 713
 714         /* Increment number of inner iterations */
 715         inneriter                  += j_index_end - j_index_start;
 716
 717         /* Outer loop uses 20 flops */
 718     }
 719
 720     /* Increment number of outer iterations */
 721     outeriter        += nri;
 722
 723     /* Update outer/inner flops */
 724
 725     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*146);
 726 }
 727 /*
 728  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_single
 729  * Electrostatics interaction: ReactionField
 730  * VdW interaction:            LennardJones
 731  * Geometry:                   Water3-Particle
 732  * Calculate force/pot:        Force
 733  */
 734 void
 735 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_single
 736                     (t_nblist                    * gmx_restrict       nlist,
 737                      rvec                        * gmx_restrict          xx,
 738                      rvec                        * gmx_restrict          ff,
 739                      struct t_forcerec           * gmx_restrict          fr,
 740                      t_mdatoms                   * gmx_restrict     mdatoms,
 741                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 742                      t_nrnb                      * gmx_restrict        nrnb)
 743 {
 744     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 745      * just 0 for non-waters.
 746      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 747      * jnr indices corresponding to data put in the four positions in the SIMD register.
 748      */
 749     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 750     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 751     int              jnrA,jnrB,jnrC,jnrD;
 752     int              jnrE,jnrF,jnrG,jnrH;
 753     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 754     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 755     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 756     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 757     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 758     real             rcutoff_scalar;
 759     real             *shiftvec,*fshift,*x,*f;
 760     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 761     real             scratch[4*DIM];
 762     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 763     real *           vdwioffsetptr0;
 764     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 765     real *           vdwioffsetptr1;
 766     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 767     real *           vdwioffsetptr2;
 768     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 769     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 770     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 771     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 772     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 773     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 774     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 775     real             *charge;
 776     int              nvdwtype;
 777     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 778     int              *vdwtype;
 779     real             *vdwparam;
 780     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 781     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 782     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 783     real             rswitch_scalar,d_scalar;
 784     __m256           dummy_mask,cutoff_mask;
 785     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 786     __m256           one     = _mm256_set1_ps(1.0);
 787     __m256           two     = _mm256_set1_ps(2.0);
 788     x                = xx[0];
 789     f                = ff[0];
 790
 791     nri              = nlist->nri;
 792     iinr             = nlist->iinr;
 793     jindex           = nlist->jindex;
 794     jjnr             = nlist->jjnr;
 795     shiftidx         = nlist->shift;
 796     gid              = nlist->gid;
 797     shiftvec         = fr->shift_vec[0];
 798     fshift           = fr->fshift[0];
 799     facel            = _mm256_set1_ps(fr->ic->epsfac);
 800     charge           = mdatoms->chargeA;
 801     krf              = _mm256_set1_ps(fr->ic->k_rf);
 802     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 803     crf              = _mm256_set1_ps(fr->ic->c_rf);
 804     nvdwtype         = fr->ntype;
 805     vdwparam         = fr->nbfp;
 806     vdwtype          = mdatoms->typeA;
 807
 808     /* Setup water-specific parameters */
 809     inr              = nlist->iinr[0];
 810     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 811     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 812     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 813     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 814
 815     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 816     rcutoff_scalar   = fr->ic->rcoulomb;
 817     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 818     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 819
 820     rswitch_scalar   = fr->ic->rvdw_switch;
 821     rswitch          = _mm256_set1_ps(rswitch_scalar);
 822     /* Setup switch parameters */
 823     d_scalar         = rcutoff_scalar-rswitch_scalar;
 824     d                = _mm256_set1_ps(d_scalar);
 825     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 826     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 827     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 828     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 829     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 830     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 831
 832     /* Avoid stupid compiler warnings */
 833     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 834     j_coord_offsetA = 0;
 835     j_coord_offsetB = 0;
 836     j_coord_offsetC = 0;
 837     j_coord_offsetD = 0;
 838     j_coord_offsetE = 0;
 839     j_coord_offsetF = 0;
 840     j_coord_offsetG = 0;
 841     j_coord_offsetH = 0;
 842
 843     outeriter        = 0;
 844     inneriter        = 0;
 845
 846     for(iidx=0;iidx<4*DIM;iidx++)
 847     {
 848         scratch[iidx] = 0.0;
 849     }
 850
 851     /* Start outer loop over neighborlists */
 852     for(iidx=0; iidx<nri; iidx++)
 853     {
 854         /* Load shift vector for this list */
 855         i_shift_offset   = DIM*shiftidx[iidx];
 856
 857         /* Load limits for loop over neighbors */
 858         j_index_start    = jindex[iidx];
 859         j_index_end      = jindex[iidx+1];
 860
 861         /* Get outer coordinate index */
 862         inr              = iinr[iidx];
 863         i_coord_offset   = DIM*inr;
 864
 865         /* Load i particle coords and add shift vector */
 866         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 867                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 868
 869         fix0             = _mm256_setzero_ps();
 870         fiy0             = _mm256_setzero_ps();
 871         fiz0             = _mm256_setzero_ps();
 872         fix1             = _mm256_setzero_ps();
 873         fiy1             = _mm256_setzero_ps();
 874         fiz1             = _mm256_setzero_ps();
 875         fix2             = _mm256_setzero_ps();
 876         fiy2             = _mm256_setzero_ps();
 877         fiz2             = _mm256_setzero_ps();
 878
 879         /* Start inner kernel loop */
 880         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 881         {
 882
 883             /* Get j neighbor index, and coordinate index */
 884             jnrA             = jjnr[jidx];
 885             jnrB             = jjnr[jidx+1];
 886             jnrC             = jjnr[jidx+2];
 887             jnrD             = jjnr[jidx+3];
 888             jnrE             = jjnr[jidx+4];
 889             jnrF             = jjnr[jidx+5];
 890             jnrG             = jjnr[jidx+6];
 891             jnrH             = jjnr[jidx+7];
 892             j_coord_offsetA  = DIM*jnrA;
 893             j_coord_offsetB  = DIM*jnrB;
 894             j_coord_offsetC  = DIM*jnrC;
 895             j_coord_offsetD  = DIM*jnrD;
 896             j_coord_offsetE  = DIM*jnrE;
 897             j_coord_offsetF  = DIM*jnrF;
 898             j_coord_offsetG  = DIM*jnrG;
 899             j_coord_offsetH  = DIM*jnrH;
 900
 901             /* load j atom coordinates */
 902             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 903                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 904                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 905                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 906                                                  &jx0,&jy0,&jz0);
 907
 908             /* Calculate displacement vector */
 909             dx00             = _mm256_sub_ps(ix0,jx0);
 910             dy00             = _mm256_sub_ps(iy0,jy0);
 911             dz00             = _mm256_sub_ps(iz0,jz0);
 912             dx10             = _mm256_sub_ps(ix1,jx0);
 913             dy10             = _mm256_sub_ps(iy1,jy0);
 914             dz10             = _mm256_sub_ps(iz1,jz0);
 915             dx20             = _mm256_sub_ps(ix2,jx0);
 916             dy20             = _mm256_sub_ps(iy2,jy0);
 917             dz20             = _mm256_sub_ps(iz2,jz0);
 918
 919             /* Calculate squared distance and things based on it */
 920             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 921             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 922             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 923
 924             rinv00           = avx256_invsqrt_f(rsq00);
 925             rinv10           = avx256_invsqrt_f(rsq10);
 926             rinv20           = avx256_invsqrt_f(rsq20);
 927
 928             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 929             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 930             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 931
 932             /* Load parameters for j particles */
 933             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 934                                                                  charge+jnrC+0,charge+jnrD+0,
 935                                                                  charge+jnrE+0,charge+jnrF+0,
 936                                                                  charge+jnrG+0,charge+jnrH+0);
 937             vdwjidx0A        = 2*vdwtype[jnrA+0];
 938             vdwjidx0B        = 2*vdwtype[jnrB+0];
 939             vdwjidx0C        = 2*vdwtype[jnrC+0];
 940             vdwjidx0D        = 2*vdwtype[jnrD+0];
 941             vdwjidx0E        = 2*vdwtype[jnrE+0];
 942             vdwjidx0F        = 2*vdwtype[jnrF+0];
 943             vdwjidx0G        = 2*vdwtype[jnrG+0];
 944             vdwjidx0H        = 2*vdwtype[jnrH+0];
 945
 946             fjx0             = _mm256_setzero_ps();
 947             fjy0             = _mm256_setzero_ps();
 948             fjz0             = _mm256_setzero_ps();
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 955             {
 956
 957             r00              = _mm256_mul_ps(rsq00,rinv00);
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             qq00             = _mm256_mul_ps(iq0,jq0);
 961             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 962                                             vdwioffsetptr0+vdwjidx0B,
 963                                             vdwioffsetptr0+vdwjidx0C,
 964                                             vdwioffsetptr0+vdwjidx0D,
 965                                             vdwioffsetptr0+vdwjidx0E,
 966                                             vdwioffsetptr0+vdwjidx0F,
 967                                             vdwioffsetptr0+vdwjidx0G,
 968                                             vdwioffsetptr0+vdwjidx0H,
 969                                             &c6_00,&c12_00);
 970
 971             /* REACTION-FIELD ELECTROSTATICS */
 972             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 973
 974             /* LENNARD-JONES DISPERSION/REPULSION */
 975
 976             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 977             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 978             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 979             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 980             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 981
 982             d                = _mm256_sub_ps(r00,rswitch);
 983             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 984             d2               = _mm256_mul_ps(d,d);
 985             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 986
 987             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 988
 989             /* Evaluate switch function */
 990             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 991             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 992             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 993
 994             fscal            = _mm256_add_ps(felec,fvdw);
 995
 996             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm256_mul_ps(fscal,dx00);
1000             ty               = _mm256_mul_ps(fscal,dy00);
1001             tz               = _mm256_mul_ps(fscal,dz00);
1002
1003             /* Update vectorial force */
1004             fix0             = _mm256_add_ps(fix0,tx);
1005             fiy0             = _mm256_add_ps(fiy0,ty);
1006             fiz0             = _mm256_add_ps(fiz0,tz);
1007
1008             fjx0             = _mm256_add_ps(fjx0,tx);
1009             fjy0             = _mm256_add_ps(fjy0,ty);
1010             fjz0             = _mm256_add_ps(fjz0,tz);
1011
1012             }
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1019             {
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq10             = _mm256_mul_ps(iq1,jq0);
1023
1024             /* REACTION-FIELD ELECTROSTATICS */
1025             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1026
1027             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1028
1029             fscal            = felec;
1030
1031             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1032
1033             /* Calculate temporary vectorial force */
1034             tx               = _mm256_mul_ps(fscal,dx10);
1035             ty               = _mm256_mul_ps(fscal,dy10);
1036             tz               = _mm256_mul_ps(fscal,dz10);
1037
1038             /* Update vectorial force */
1039             fix1             = _mm256_add_ps(fix1,tx);
1040             fiy1             = _mm256_add_ps(fiy1,ty);
1041             fiz1             = _mm256_add_ps(fiz1,tz);
1042
1043             fjx0             = _mm256_add_ps(fjx0,tx);
1044             fjy0             = _mm256_add_ps(fjy0,ty);
1045             fjz0             = _mm256_add_ps(fjz0,tz);
1046
1047             }
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1054             {
1055
1056             /* Compute parameters for interactions between i and j atoms */
1057             qq20             = _mm256_mul_ps(iq2,jq0);
1058
1059             /* REACTION-FIELD ELECTROSTATICS */
1060             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1061
1062             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1063
1064             fscal            = felec;
1065
1066             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1067
1068             /* Calculate temporary vectorial force */
1069             tx               = _mm256_mul_ps(fscal,dx20);
1070             ty               = _mm256_mul_ps(fscal,dy20);
1071             tz               = _mm256_mul_ps(fscal,dz20);
1072
1073             /* Update vectorial force */
1074             fix2             = _mm256_add_ps(fix2,tx);
1075             fiy2             = _mm256_add_ps(fiy2,ty);
1076             fiz2             = _mm256_add_ps(fiz2,tz);
1077
1078             fjx0             = _mm256_add_ps(fjx0,tx);
1079             fjy0             = _mm256_add_ps(fjy0,ty);
1080             fjz0             = _mm256_add_ps(fjz0,tz);
1081
1082             }
1083
1084             fjptrA             = f+j_coord_offsetA;
1085             fjptrB             = f+j_coord_offsetB;
1086             fjptrC             = f+j_coord_offsetC;
1087             fjptrD             = f+j_coord_offsetD;
1088             fjptrE             = f+j_coord_offsetE;
1089             fjptrF             = f+j_coord_offsetF;
1090             fjptrG             = f+j_coord_offsetG;
1091             fjptrH             = f+j_coord_offsetH;
1092
1093             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1094
1095             /* Inner loop uses 124 flops */
1096         }
1097
1098         if(jidx<j_index_end)
1099         {
1100
1101             /* Get j neighbor index, and coordinate index */
1102             jnrlistA         = jjnr[jidx];
1103             jnrlistB         = jjnr[jidx+1];
1104             jnrlistC         = jjnr[jidx+2];
1105             jnrlistD         = jjnr[jidx+3];
1106             jnrlistE         = jjnr[jidx+4];
1107             jnrlistF         = jjnr[jidx+5];
1108             jnrlistG         = jjnr[jidx+6];
1109             jnrlistH         = jjnr[jidx+7];
1110             /* Sign of each element will be negative for non-real atoms.
1111              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1112              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1113              */
1114             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1115                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1116
1117             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1118             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1119             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1120             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1121             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1122             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1123             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1124             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1125             j_coord_offsetA  = DIM*jnrA;
1126             j_coord_offsetB  = DIM*jnrB;
1127             j_coord_offsetC  = DIM*jnrC;
1128             j_coord_offsetD  = DIM*jnrD;
1129             j_coord_offsetE  = DIM*jnrE;
1130             j_coord_offsetF  = DIM*jnrF;
1131             j_coord_offsetG  = DIM*jnrG;
1132             j_coord_offsetH  = DIM*jnrH;
1133
1134             /* load j atom coordinates */
1135             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1136                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1137                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1138                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1139                                                  &jx0,&jy0,&jz0);
1140
1141             /* Calculate displacement vector */
1142             dx00             = _mm256_sub_ps(ix0,jx0);
1143             dy00             = _mm256_sub_ps(iy0,jy0);
1144             dz00             = _mm256_sub_ps(iz0,jz0);
1145             dx10             = _mm256_sub_ps(ix1,jx0);
1146             dy10             = _mm256_sub_ps(iy1,jy0);
1147             dz10             = _mm256_sub_ps(iz1,jz0);
1148             dx20             = _mm256_sub_ps(ix2,jx0);
1149             dy20             = _mm256_sub_ps(iy2,jy0);
1150             dz20             = _mm256_sub_ps(iz2,jz0);
1151
1152             /* Calculate squared distance and things based on it */
1153             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1154             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1155             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1156
1157             rinv00           = avx256_invsqrt_f(rsq00);
1158             rinv10           = avx256_invsqrt_f(rsq10);
1159             rinv20           = avx256_invsqrt_f(rsq20);
1160
1161             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1162             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1163             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1164
1165             /* Load parameters for j particles */
1166             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1167                                                                  charge+jnrC+0,charge+jnrD+0,
1168                                                                  charge+jnrE+0,charge+jnrF+0,
1169                                                                  charge+jnrG+0,charge+jnrH+0);
1170             vdwjidx0A        = 2*vdwtype[jnrA+0];
1171             vdwjidx0B        = 2*vdwtype[jnrB+0];
1172             vdwjidx0C        = 2*vdwtype[jnrC+0];
1173             vdwjidx0D        = 2*vdwtype[jnrD+0];
1174             vdwjidx0E        = 2*vdwtype[jnrE+0];
1175             vdwjidx0F        = 2*vdwtype[jnrF+0];
1176             vdwjidx0G        = 2*vdwtype[jnrG+0];
1177             vdwjidx0H        = 2*vdwtype[jnrH+0];
1178
1179             fjx0             = _mm256_setzero_ps();
1180             fjy0             = _mm256_setzero_ps();
1181             fjz0             = _mm256_setzero_ps();
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1188             {
1189
1190             r00              = _mm256_mul_ps(rsq00,rinv00);
1191             r00              = _mm256_andnot_ps(dummy_mask,r00);
1192
1193             /* Compute parameters for interactions between i and j atoms */
1194             qq00             = _mm256_mul_ps(iq0,jq0);
1195             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1196                                             vdwioffsetptr0+vdwjidx0B,
1197                                             vdwioffsetptr0+vdwjidx0C,
1198                                             vdwioffsetptr0+vdwjidx0D,
1199                                             vdwioffsetptr0+vdwjidx0E,
1200                                             vdwioffsetptr0+vdwjidx0F,
1201                                             vdwioffsetptr0+vdwjidx0G,
1202                                             vdwioffsetptr0+vdwjidx0H,
1203                                             &c6_00,&c12_00);
1204
1205             /* REACTION-FIELD ELECTROSTATICS */
1206             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1207
1208             /* LENNARD-JONES DISPERSION/REPULSION */
1209
1210             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1211             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1212             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1213             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1214             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1215
1216             d                = _mm256_sub_ps(r00,rswitch);
1217             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1218             d2               = _mm256_mul_ps(d,d);
1219             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1220
1221             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1222
1223             /* Evaluate switch function */
1224             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1225             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1226             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1227
1228             fscal            = _mm256_add_ps(felec,fvdw);
1229
1230             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1231
1232             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1233
1234             /* Calculate temporary vectorial force */
1235             tx               = _mm256_mul_ps(fscal,dx00);
1236             ty               = _mm256_mul_ps(fscal,dy00);
1237             tz               = _mm256_mul_ps(fscal,dz00);
1238
1239             /* Update vectorial force */
1240             fix0             = _mm256_add_ps(fix0,tx);
1241             fiy0             = _mm256_add_ps(fiy0,ty);
1242             fiz0             = _mm256_add_ps(fiz0,tz);
1243
1244             fjx0             = _mm256_add_ps(fjx0,tx);
1245             fjy0             = _mm256_add_ps(fjy0,ty);
1246             fjz0             = _mm256_add_ps(fjz0,tz);
1247
1248             }
1249
1250             /**************************
1251              * CALCULATE INTERACTIONS *
1252              **************************/
1253
1254             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1255             {
1256
1257             /* Compute parameters for interactions between i and j atoms */
1258             qq10             = _mm256_mul_ps(iq1,jq0);
1259
1260             /* REACTION-FIELD ELECTROSTATICS */
1261             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1262
1263             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1264
1265             fscal            = felec;
1266
1267             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1268
1269             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1270
1271             /* Calculate temporary vectorial force */
1272             tx               = _mm256_mul_ps(fscal,dx10);
1273             ty               = _mm256_mul_ps(fscal,dy10);
1274             tz               = _mm256_mul_ps(fscal,dz10);
1275
1276             /* Update vectorial force */
1277             fix1             = _mm256_add_ps(fix1,tx);
1278             fiy1             = _mm256_add_ps(fiy1,ty);
1279             fiz1             = _mm256_add_ps(fiz1,tz);
1280
1281             fjx0             = _mm256_add_ps(fjx0,tx);
1282             fjy0             = _mm256_add_ps(fjy0,ty);
1283             fjz0             = _mm256_add_ps(fjz0,tz);
1284
1285             }
1286
1287             /**************************
1288              * CALCULATE INTERACTIONS *
1289              **************************/
1290
1291             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1292             {
1293
1294             /* Compute parameters for interactions between i and j atoms */
1295             qq20             = _mm256_mul_ps(iq2,jq0);
1296
1297             /* REACTION-FIELD ELECTROSTATICS */
1298             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1299
1300             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1301
1302             fscal            = felec;
1303
1304             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1305
1306             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1307
1308             /* Calculate temporary vectorial force */
1309             tx               = _mm256_mul_ps(fscal,dx20);
1310             ty               = _mm256_mul_ps(fscal,dy20);
1311             tz               = _mm256_mul_ps(fscal,dz20);
1312
1313             /* Update vectorial force */
1314             fix2             = _mm256_add_ps(fix2,tx);
1315             fiy2             = _mm256_add_ps(fiy2,ty);
1316             fiz2             = _mm256_add_ps(fiz2,tz);
1317
1318             fjx0             = _mm256_add_ps(fjx0,tx);
1319             fjy0             = _mm256_add_ps(fjy0,ty);
1320             fjz0             = _mm256_add_ps(fjz0,tz);
1321
1322             }
1323
1324             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1325             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1326             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1327             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1328             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1329             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1330             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1331             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1332
1333             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1334
1335             /* Inner loop uses 125 flops */
1336         }
1337
1338         /* End of innermost loop */
1339
1340         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1341                                                  f+i_coord_offset,fshift+i_shift_offset);
1342
1343         /* Increment number of inner iterations */
1344         inneriter                  += j_index_end - j_index_start;
1345
1346         /* Outer loop uses 18 flops */
1347     }
1348
1349     /* Increment number of outer iterations */
1350     outeriter        += nri;
1351
1352     /* Update outer/inner flops */
1353
1354     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*125);
1355 }