src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_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_VdwLJSw_GeomW4P1_VF_avx_256_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_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     real *           vdwioffsetptr3;
  90     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 106     real             rswitch_scalar,d_scalar;
 107     __m256d          dummy_mask,cutoff_mask;
 108     __m128           tmpmask0,tmpmask1;
 109     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 110     __m256d          one     = _mm256_set1_pd(1.0);
 111     __m256d          two     = _mm256_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_pd(fr->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm256_set1_pd(fr->ic->k_rf);
 126     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 127     crf              = _mm256_set1_pd(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 135     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 136     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 137     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 138
 139     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 140     rcutoff_scalar   = fr->ic->rcoulomb;
 141     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 142     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 143
 144     rswitch_scalar   = fr->ic->rvdw_switch;
 145     rswitch          = _mm256_set1_pd(rswitch_scalar);
 146     /* Setup switch parameters */
 147     d_scalar         = rcutoff_scalar-rswitch_scalar;
 148     d                = _mm256_set1_pd(d_scalar);
 149     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 150     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 151     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 152     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 153     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 154     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = jnrC = jnrD = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160     j_coord_offsetC = 0;
 161     j_coord_offsetD = 0;
 162
 163     outeriter        = 0;
 164     inneriter        = 0;
 165
 166     for(iidx=0;iidx<4*DIM;iidx++)
 167     {
 168         scratch[iidx] = 0.0;
 169     }
 170
 171     /* Start outer loop over neighborlists */
 172     for(iidx=0; iidx<nri; iidx++)
 173     {
 174         /* Load shift vector for this list */
 175         i_shift_offset   = DIM*shiftidx[iidx];
 176
 177         /* Load limits for loop over neighbors */
 178         j_index_start    = jindex[iidx];
 179         j_index_end      = jindex[iidx+1];
 180
 181         /* Get outer coordinate index */
 182         inr              = iinr[iidx];
 183         i_coord_offset   = DIM*inr;
 184
 185         /* Load i particle coords and add shift vector */
 186         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 187                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 188
 189         fix0             = _mm256_setzero_pd();
 190         fiy0             = _mm256_setzero_pd();
 191         fiz0             = _mm256_setzero_pd();
 192         fix1             = _mm256_setzero_pd();
 193         fiy1             = _mm256_setzero_pd();
 194         fiz1             = _mm256_setzero_pd();
 195         fix2             = _mm256_setzero_pd();
 196         fiy2             = _mm256_setzero_pd();
 197         fiz2             = _mm256_setzero_pd();
 198         fix3             = _mm256_setzero_pd();
 199         fiy3             = _mm256_setzero_pd();
 200         fiz3             = _mm256_setzero_pd();
 201
 202         /* Reset potential sums */
 203         velecsum         = _mm256_setzero_pd();
 204         vvdwsum          = _mm256_setzero_pd();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             j_coord_offsetA  = DIM*jnrA;
 216             j_coord_offsetB  = DIM*jnrB;
 217             j_coord_offsetC  = DIM*jnrC;
 218             j_coord_offsetD  = DIM*jnrD;
 219
 220             /* load j atom coordinates */
 221             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 223                                                  &jx0,&jy0,&jz0);
 224
 225             /* Calculate displacement vector */
 226             dx00             = _mm256_sub_pd(ix0,jx0);
 227             dy00             = _mm256_sub_pd(iy0,jy0);
 228             dz00             = _mm256_sub_pd(iz0,jz0);
 229             dx10             = _mm256_sub_pd(ix1,jx0);
 230             dy10             = _mm256_sub_pd(iy1,jy0);
 231             dz10             = _mm256_sub_pd(iz1,jz0);
 232             dx20             = _mm256_sub_pd(ix2,jx0);
 233             dy20             = _mm256_sub_pd(iy2,jy0);
 234             dz20             = _mm256_sub_pd(iz2,jz0);
 235             dx30             = _mm256_sub_pd(ix3,jx0);
 236             dy30             = _mm256_sub_pd(iy3,jy0);
 237             dz30             = _mm256_sub_pd(iz3,jz0);
 238
 239             /* Calculate squared distance and things based on it */
 240             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 241             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 242             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 243             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 244
 245             rinv00           = avx256_invsqrt_d(rsq00);
 246             rinv10           = avx256_invsqrt_d(rsq10);
 247             rinv20           = avx256_invsqrt_d(rsq20);
 248             rinv30           = avx256_invsqrt_d(rsq30);
 249
 250             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 251             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 252             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 253             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 254
 255             /* Load parameters for j particles */
 256             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 257                                                                  charge+jnrC+0,charge+jnrD+0);
 258             vdwjidx0A        = 2*vdwtype[jnrA+0];
 259             vdwjidx0B        = 2*vdwtype[jnrB+0];
 260             vdwjidx0C        = 2*vdwtype[jnrC+0];
 261             vdwjidx0D        = 2*vdwtype[jnrD+0];
 262
 263             fjx0             = _mm256_setzero_pd();
 264             fjy0             = _mm256_setzero_pd();
 265             fjz0             = _mm256_setzero_pd();
 266
 267             /**************************
 268              * CALCULATE INTERACTIONS *
 269              **************************/
 270
 271             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 272             {
 273
 274             r00              = _mm256_mul_pd(rsq00,rinv00);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 278                                             vdwioffsetptr0+vdwjidx0B,
 279                                             vdwioffsetptr0+vdwjidx0C,
 280                                             vdwioffsetptr0+vdwjidx0D,
 281                                             &c6_00,&c12_00);
 282
 283             /* LENNARD-JONES DISPERSION/REPULSION */
 284
 285             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 286             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 287             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 288             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 289             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 290
 291             d                = _mm256_sub_pd(r00,rswitch);
 292             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 293             d2               = _mm256_mul_pd(d,d);
 294             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 295
 296             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 297
 298             /* Evaluate switch function */
 299             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 300             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 301             vvdw             = _mm256_mul_pd(vvdw,sw);
 302             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 306             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 307
 308             fscal            = fvdw;
 309
 310             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm256_mul_pd(fscal,dx00);
 314             ty               = _mm256_mul_pd(fscal,dy00);
 315             tz               = _mm256_mul_pd(fscal,dz00);
 316
 317             /* Update vectorial force */
 318             fix0             = _mm256_add_pd(fix0,tx);
 319             fiy0             = _mm256_add_pd(fiy0,ty);
 320             fiz0             = _mm256_add_pd(fiz0,tz);
 321
 322             fjx0             = _mm256_add_pd(fjx0,tx);
 323             fjy0             = _mm256_add_pd(fjy0,ty);
 324             fjz0             = _mm256_add_pd(fjz0,tz);
 325
 326             }
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 333             {
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq10             = _mm256_mul_pd(iq1,jq0);
 337
 338             /* REACTION-FIELD ELECTROSTATICS */
 339             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 340             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 341
 342             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velec            = _mm256_and_pd(velec,cutoff_mask);
 346             velecsum         = _mm256_add_pd(velecsum,velec);
 347
 348             fscal            = felec;
 349
 350             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm256_mul_pd(fscal,dx10);
 354             ty               = _mm256_mul_pd(fscal,dy10);
 355             tz               = _mm256_mul_pd(fscal,dz10);
 356
 357             /* Update vectorial force */
 358             fix1             = _mm256_add_pd(fix1,tx);
 359             fiy1             = _mm256_add_pd(fiy1,ty);
 360             fiz1             = _mm256_add_pd(fiz1,tz);
 361
 362             fjx0             = _mm256_add_pd(fjx0,tx);
 363             fjy0             = _mm256_add_pd(fjy0,ty);
 364             fjz0             = _mm256_add_pd(fjz0,tz);
 365
 366             }
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 373             {
 374
 375             /* Compute parameters for interactions between i and j atoms */
 376             qq20             = _mm256_mul_pd(iq2,jq0);
 377
 378             /* REACTION-FIELD ELECTROSTATICS */
 379             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 380             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 381
 382             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velec            = _mm256_and_pd(velec,cutoff_mask);
 386             velecsum         = _mm256_add_pd(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm256_mul_pd(fscal,dx20);
 394             ty               = _mm256_mul_pd(fscal,dy20);
 395             tz               = _mm256_mul_pd(fscal,dz20);
 396
 397             /* Update vectorial force */
 398             fix2             = _mm256_add_pd(fix2,tx);
 399             fiy2             = _mm256_add_pd(fiy2,ty);
 400             fiz2             = _mm256_add_pd(fiz2,tz);
 401
 402             fjx0             = _mm256_add_pd(fjx0,tx);
 403             fjy0             = _mm256_add_pd(fjy0,ty);
 404             fjz0             = _mm256_add_pd(fjz0,tz);
 405
 406             }
 407
 408             /**************************
 409              * CALCULATE INTERACTIONS *
 410              **************************/
 411
 412             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 413             {
 414
 415             /* Compute parameters for interactions between i and j atoms */
 416             qq30             = _mm256_mul_pd(iq3,jq0);
 417
 418             /* REACTION-FIELD ELECTROSTATICS */
 419             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 420             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 421
 422             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 423
 424             /* Update potential sum for this i atom from the interaction with this j atom. */
 425             velec            = _mm256_and_pd(velec,cutoff_mask);
 426             velecsum         = _mm256_add_pd(velecsum,velec);
 427
 428             fscal            = felec;
 429
 430             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 431
 432             /* Calculate temporary vectorial force */
 433             tx               = _mm256_mul_pd(fscal,dx30);
 434             ty               = _mm256_mul_pd(fscal,dy30);
 435             tz               = _mm256_mul_pd(fscal,dz30);
 436
 437             /* Update vectorial force */
 438             fix3             = _mm256_add_pd(fix3,tx);
 439             fiy3             = _mm256_add_pd(fiy3,ty);
 440             fiz3             = _mm256_add_pd(fiz3,tz);
 441
 442             fjx0             = _mm256_add_pd(fjx0,tx);
 443             fjy0             = _mm256_add_pd(fjy0,ty);
 444             fjz0             = _mm256_add_pd(fjz0,tz);
 445
 446             }
 447
 448             fjptrA             = f+j_coord_offsetA;
 449             fjptrB             = f+j_coord_offsetB;
 450             fjptrC             = f+j_coord_offsetC;
 451             fjptrD             = f+j_coord_offsetD;
 452
 453             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 454
 455             /* Inner loop uses 170 flops */
 456         }
 457
 458         if(jidx<j_index_end)
 459         {
 460
 461             /* Get j neighbor index, and coordinate index */
 462             jnrlistA         = jjnr[jidx];
 463             jnrlistB         = jjnr[jidx+1];
 464             jnrlistC         = jjnr[jidx+2];
 465             jnrlistD         = jjnr[jidx+3];
 466             /* Sign of each element will be negative for non-real atoms.
 467              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 468              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 469              */
 470             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 471
 472             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 473             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 474             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 475
 476             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 477             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 478             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 479             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 480             j_coord_offsetA  = DIM*jnrA;
 481             j_coord_offsetB  = DIM*jnrB;
 482             j_coord_offsetC  = DIM*jnrC;
 483             j_coord_offsetD  = DIM*jnrD;
 484
 485             /* load j atom coordinates */
 486             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 487                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 488                                                  &jx0,&jy0,&jz0);
 489
 490             /* Calculate displacement vector */
 491             dx00             = _mm256_sub_pd(ix0,jx0);
 492             dy00             = _mm256_sub_pd(iy0,jy0);
 493             dz00             = _mm256_sub_pd(iz0,jz0);
 494             dx10             = _mm256_sub_pd(ix1,jx0);
 495             dy10             = _mm256_sub_pd(iy1,jy0);
 496             dz10             = _mm256_sub_pd(iz1,jz0);
 497             dx20             = _mm256_sub_pd(ix2,jx0);
 498             dy20             = _mm256_sub_pd(iy2,jy0);
 499             dz20             = _mm256_sub_pd(iz2,jz0);
 500             dx30             = _mm256_sub_pd(ix3,jx0);
 501             dy30             = _mm256_sub_pd(iy3,jy0);
 502             dz30             = _mm256_sub_pd(iz3,jz0);
 503
 504             /* Calculate squared distance and things based on it */
 505             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 506             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 507             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 508             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 509
 510             rinv00           = avx256_invsqrt_d(rsq00);
 511             rinv10           = avx256_invsqrt_d(rsq10);
 512             rinv20           = avx256_invsqrt_d(rsq20);
 513             rinv30           = avx256_invsqrt_d(rsq30);
 514
 515             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 516             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 517             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 518             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 519
 520             /* Load parameters for j particles */
 521             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 522                                                                  charge+jnrC+0,charge+jnrD+0);
 523             vdwjidx0A        = 2*vdwtype[jnrA+0];
 524             vdwjidx0B        = 2*vdwtype[jnrB+0];
 525             vdwjidx0C        = 2*vdwtype[jnrC+0];
 526             vdwjidx0D        = 2*vdwtype[jnrD+0];
 527
 528             fjx0             = _mm256_setzero_pd();
 529             fjy0             = _mm256_setzero_pd();
 530             fjz0             = _mm256_setzero_pd();
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 537             {
 538
 539             r00              = _mm256_mul_pd(rsq00,rinv00);
 540             r00              = _mm256_andnot_pd(dummy_mask,r00);
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 544                                             vdwioffsetptr0+vdwjidx0B,
 545                                             vdwioffsetptr0+vdwjidx0C,
 546                                             vdwioffsetptr0+vdwjidx0D,
 547                                             &c6_00,&c12_00);
 548
 549             /* LENNARD-JONES DISPERSION/REPULSION */
 550
 551             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 552             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 553             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 554             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 555             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 556
 557             d                = _mm256_sub_pd(r00,rswitch);
 558             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 559             d2               = _mm256_mul_pd(d,d);
 560             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 561
 562             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 563
 564             /* Evaluate switch function */
 565             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 566             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 567             vvdw             = _mm256_mul_pd(vvdw,sw);
 568             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 572             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 573             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 574
 575             fscal            = fvdw;
 576
 577             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 578
 579             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm256_mul_pd(fscal,dx00);
 583             ty               = _mm256_mul_pd(fscal,dy00);
 584             tz               = _mm256_mul_pd(fscal,dz00);
 585
 586             /* Update vectorial force */
 587             fix0             = _mm256_add_pd(fix0,tx);
 588             fiy0             = _mm256_add_pd(fiy0,ty);
 589             fiz0             = _mm256_add_pd(fiz0,tz);
 590
 591             fjx0             = _mm256_add_pd(fjx0,tx);
 592             fjy0             = _mm256_add_pd(fjy0,ty);
 593             fjz0             = _mm256_add_pd(fjz0,tz);
 594
 595             }
 596
 597             /**************************
 598              * CALCULATE INTERACTIONS *
 599              **************************/
 600
 601             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 602             {
 603
 604             /* Compute parameters for interactions between i and j atoms */
 605             qq10             = _mm256_mul_pd(iq1,jq0);
 606
 607             /* REACTION-FIELD ELECTROSTATICS */
 608             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 609             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 610
 611             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 612
 613             /* Update potential sum for this i atom from the interaction with this j atom. */
 614             velec            = _mm256_and_pd(velec,cutoff_mask);
 615             velec            = _mm256_andnot_pd(dummy_mask,velec);
 616             velecsum         = _mm256_add_pd(velecsum,velec);
 617
 618             fscal            = felec;
 619
 620             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 621
 622             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 623
 624             /* Calculate temporary vectorial force */
 625             tx               = _mm256_mul_pd(fscal,dx10);
 626             ty               = _mm256_mul_pd(fscal,dy10);
 627             tz               = _mm256_mul_pd(fscal,dz10);
 628
 629             /* Update vectorial force */
 630             fix1             = _mm256_add_pd(fix1,tx);
 631             fiy1             = _mm256_add_pd(fiy1,ty);
 632             fiz1             = _mm256_add_pd(fiz1,tz);
 633
 634             fjx0             = _mm256_add_pd(fjx0,tx);
 635             fjy0             = _mm256_add_pd(fjy0,ty);
 636             fjz0             = _mm256_add_pd(fjz0,tz);
 637
 638             }
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 645             {
 646
 647             /* Compute parameters for interactions between i and j atoms */
 648             qq20             = _mm256_mul_pd(iq2,jq0);
 649
 650             /* REACTION-FIELD ELECTROSTATICS */
 651             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 652             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 653
 654             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 655
 656             /* Update potential sum for this i atom from the interaction with this j atom. */
 657             velec            = _mm256_and_pd(velec,cutoff_mask);
 658             velec            = _mm256_andnot_pd(dummy_mask,velec);
 659             velecsum         = _mm256_add_pd(velecsum,velec);
 660
 661             fscal            = felec;
 662
 663             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 664
 665             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 666
 667             /* Calculate temporary vectorial force */
 668             tx               = _mm256_mul_pd(fscal,dx20);
 669             ty               = _mm256_mul_pd(fscal,dy20);
 670             tz               = _mm256_mul_pd(fscal,dz20);
 671
 672             /* Update vectorial force */
 673             fix2             = _mm256_add_pd(fix2,tx);
 674             fiy2             = _mm256_add_pd(fiy2,ty);
 675             fiz2             = _mm256_add_pd(fiz2,tz);
 676
 677             fjx0             = _mm256_add_pd(fjx0,tx);
 678             fjy0             = _mm256_add_pd(fjy0,ty);
 679             fjz0             = _mm256_add_pd(fjz0,tz);
 680
 681             }
 682
 683             /**************************
 684              * CALCULATE INTERACTIONS *
 685              **************************/
 686
 687             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 688             {
 689
 690             /* Compute parameters for interactions between i and j atoms */
 691             qq30             = _mm256_mul_pd(iq3,jq0);
 692
 693             /* REACTION-FIELD ELECTROSTATICS */
 694             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 695             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 696
 697             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 698
 699             /* Update potential sum for this i atom from the interaction with this j atom. */
 700             velec            = _mm256_and_pd(velec,cutoff_mask);
 701             velec            = _mm256_andnot_pd(dummy_mask,velec);
 702             velecsum         = _mm256_add_pd(velecsum,velec);
 703
 704             fscal            = felec;
 705
 706             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 707
 708             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 709
 710             /* Calculate temporary vectorial force */
 711             tx               = _mm256_mul_pd(fscal,dx30);
 712             ty               = _mm256_mul_pd(fscal,dy30);
 713             tz               = _mm256_mul_pd(fscal,dz30);
 714
 715             /* Update vectorial force */
 716             fix3             = _mm256_add_pd(fix3,tx);
 717             fiy3             = _mm256_add_pd(fiy3,ty);
 718             fiz3             = _mm256_add_pd(fiz3,tz);
 719
 720             fjx0             = _mm256_add_pd(fjx0,tx);
 721             fjy0             = _mm256_add_pd(fjy0,ty);
 722             fjz0             = _mm256_add_pd(fjz0,tz);
 723
 724             }
 725
 726             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 727             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 728             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 729             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 730
 731             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 732
 733             /* Inner loop uses 171 flops */
 734         }
 735
 736         /* End of innermost loop */
 737
 738         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 739                                                  f+i_coord_offset,fshift+i_shift_offset);
 740
 741         ggid                        = gid[iidx];
 742         /* Update potential energies */
 743         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 744         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 745
 746         /* Increment number of inner iterations */
 747         inneriter                  += j_index_end - j_index_start;
 748
 749         /* Outer loop uses 26 flops */
 750     }
 751
 752     /* Increment number of outer iterations */
 753     outeriter        += nri;
 754
 755     /* Update outer/inner flops */
 756
 757     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*171);
 758 }
 759 /*
 760  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_double
 761  * Electrostatics interaction: ReactionField
 762  * VdW interaction:            LennardJones
 763  * Geometry:                   Water4-Particle
 764  * Calculate force/pot:        Force
 765  */
 766 void
 767 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_double
 768                     (t_nblist                    * gmx_restrict       nlist,
 769                      rvec                        * gmx_restrict          xx,
 770                      rvec                        * gmx_restrict          ff,
 771                      struct t_forcerec           * gmx_restrict          fr,
 772                      t_mdatoms                   * gmx_restrict     mdatoms,
 773                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 774                      t_nrnb                      * gmx_restrict        nrnb)
 775 {
 776     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 777      * just 0 for non-waters.
 778      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 779      * jnr indices corresponding to data put in the four positions in the SIMD register.
 780      */
 781     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 782     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 783     int              jnrA,jnrB,jnrC,jnrD;
 784     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 785     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 786     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 787     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 788     real             rcutoff_scalar;
 789     real             *shiftvec,*fshift,*x,*f;
 790     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 791     real             scratch[4*DIM];
 792     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 793     real *           vdwioffsetptr0;
 794     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 795     real *           vdwioffsetptr1;
 796     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 797     real *           vdwioffsetptr2;
 798     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 799     real *           vdwioffsetptr3;
 800     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 801     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 802     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 803     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 804     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 805     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 806     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 807     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 808     real             *charge;
 809     int              nvdwtype;
 810     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 811     int              *vdwtype;
 812     real             *vdwparam;
 813     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 814     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 815     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 816     real             rswitch_scalar,d_scalar;
 817     __m256d          dummy_mask,cutoff_mask;
 818     __m128           tmpmask0,tmpmask1;
 819     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 820     __m256d          one     = _mm256_set1_pd(1.0);
 821     __m256d          two     = _mm256_set1_pd(2.0);
 822     x                = xx[0];
 823     f                = ff[0];
 824
 825     nri              = nlist->nri;
 826     iinr             = nlist->iinr;
 827     jindex           = nlist->jindex;
 828     jjnr             = nlist->jjnr;
 829     shiftidx         = nlist->shift;
 830     gid              = nlist->gid;
 831     shiftvec         = fr->shift_vec[0];
 832     fshift           = fr->fshift[0];
 833     facel            = _mm256_set1_pd(fr->ic->epsfac);
 834     charge           = mdatoms->chargeA;
 835     krf              = _mm256_set1_pd(fr->ic->k_rf);
 836     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 837     crf              = _mm256_set1_pd(fr->ic->c_rf);
 838     nvdwtype         = fr->ntype;
 839     vdwparam         = fr->nbfp;
 840     vdwtype          = mdatoms->typeA;
 841
 842     /* Setup water-specific parameters */
 843     inr              = nlist->iinr[0];
 844     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 845     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 846     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 847     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 848
 849     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 850     rcutoff_scalar   = fr->ic->rcoulomb;
 851     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 852     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 853
 854     rswitch_scalar   = fr->ic->rvdw_switch;
 855     rswitch          = _mm256_set1_pd(rswitch_scalar);
 856     /* Setup switch parameters */
 857     d_scalar         = rcutoff_scalar-rswitch_scalar;
 858     d                = _mm256_set1_pd(d_scalar);
 859     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 860     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 861     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 862     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 863     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 864     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 865
 866     /* Avoid stupid compiler warnings */
 867     jnrA = jnrB = jnrC = jnrD = 0;
 868     j_coord_offsetA = 0;
 869     j_coord_offsetB = 0;
 870     j_coord_offsetC = 0;
 871     j_coord_offsetD = 0;
 872
 873     outeriter        = 0;
 874     inneriter        = 0;
 875
 876     for(iidx=0;iidx<4*DIM;iidx++)
 877     {
 878         scratch[iidx] = 0.0;
 879     }
 880
 881     /* Start outer loop over neighborlists */
 882     for(iidx=0; iidx<nri; iidx++)
 883     {
 884         /* Load shift vector for this list */
 885         i_shift_offset   = DIM*shiftidx[iidx];
 886
 887         /* Load limits for loop over neighbors */
 888         j_index_start    = jindex[iidx];
 889         j_index_end      = jindex[iidx+1];
 890
 891         /* Get outer coordinate index */
 892         inr              = iinr[iidx];
 893         i_coord_offset   = DIM*inr;
 894
 895         /* Load i particle coords and add shift vector */
 896         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 897                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 898
 899         fix0             = _mm256_setzero_pd();
 900         fiy0             = _mm256_setzero_pd();
 901         fiz0             = _mm256_setzero_pd();
 902         fix1             = _mm256_setzero_pd();
 903         fiy1             = _mm256_setzero_pd();
 904         fiz1             = _mm256_setzero_pd();
 905         fix2             = _mm256_setzero_pd();
 906         fiy2             = _mm256_setzero_pd();
 907         fiz2             = _mm256_setzero_pd();
 908         fix3             = _mm256_setzero_pd();
 909         fiy3             = _mm256_setzero_pd();
 910         fiz3             = _mm256_setzero_pd();
 911
 912         /* Start inner kernel loop */
 913         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 914         {
 915
 916             /* Get j neighbor index, and coordinate index */
 917             jnrA             = jjnr[jidx];
 918             jnrB             = jjnr[jidx+1];
 919             jnrC             = jjnr[jidx+2];
 920             jnrD             = jjnr[jidx+3];
 921             j_coord_offsetA  = DIM*jnrA;
 922             j_coord_offsetB  = DIM*jnrB;
 923             j_coord_offsetC  = DIM*jnrC;
 924             j_coord_offsetD  = DIM*jnrD;
 925
 926             /* load j atom coordinates */
 927             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 928                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 929                                                  &jx0,&jy0,&jz0);
 930
 931             /* Calculate displacement vector */
 932             dx00             = _mm256_sub_pd(ix0,jx0);
 933             dy00             = _mm256_sub_pd(iy0,jy0);
 934             dz00             = _mm256_sub_pd(iz0,jz0);
 935             dx10             = _mm256_sub_pd(ix1,jx0);
 936             dy10             = _mm256_sub_pd(iy1,jy0);
 937             dz10             = _mm256_sub_pd(iz1,jz0);
 938             dx20             = _mm256_sub_pd(ix2,jx0);
 939             dy20             = _mm256_sub_pd(iy2,jy0);
 940             dz20             = _mm256_sub_pd(iz2,jz0);
 941             dx30             = _mm256_sub_pd(ix3,jx0);
 942             dy30             = _mm256_sub_pd(iy3,jy0);
 943             dz30             = _mm256_sub_pd(iz3,jz0);
 944
 945             /* Calculate squared distance and things based on it */
 946             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 947             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 948             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 949             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 950
 951             rinv00           = avx256_invsqrt_d(rsq00);
 952             rinv10           = avx256_invsqrt_d(rsq10);
 953             rinv20           = avx256_invsqrt_d(rsq20);
 954             rinv30           = avx256_invsqrt_d(rsq30);
 955
 956             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 957             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 958             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 959             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 960
 961             /* Load parameters for j particles */
 962             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 963                                                                  charge+jnrC+0,charge+jnrD+0);
 964             vdwjidx0A        = 2*vdwtype[jnrA+0];
 965             vdwjidx0B        = 2*vdwtype[jnrB+0];
 966             vdwjidx0C        = 2*vdwtype[jnrC+0];
 967             vdwjidx0D        = 2*vdwtype[jnrD+0];
 968
 969             fjx0             = _mm256_setzero_pd();
 970             fjy0             = _mm256_setzero_pd();
 971             fjz0             = _mm256_setzero_pd();
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 978             {
 979
 980             r00              = _mm256_mul_pd(rsq00,rinv00);
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 984                                             vdwioffsetptr0+vdwjidx0B,
 985                                             vdwioffsetptr0+vdwjidx0C,
 986                                             vdwioffsetptr0+vdwjidx0D,
 987                                             &c6_00,&c12_00);
 988
 989             /* LENNARD-JONES DISPERSION/REPULSION */
 990
 991             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 992             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 993             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 994             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 995             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 996
 997             d                = _mm256_sub_pd(r00,rswitch);
 998             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 999             d2               = _mm256_mul_pd(d,d);
1000             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1001
1002             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1003
1004             /* Evaluate switch function */
1005             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1006             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1007             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1008
1009             fscal            = fvdw;
1010
1011             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1012
1013             /* Calculate temporary vectorial force */
1014             tx               = _mm256_mul_pd(fscal,dx00);
1015             ty               = _mm256_mul_pd(fscal,dy00);
1016             tz               = _mm256_mul_pd(fscal,dz00);
1017
1018             /* Update vectorial force */
1019             fix0             = _mm256_add_pd(fix0,tx);
1020             fiy0             = _mm256_add_pd(fiy0,ty);
1021             fiz0             = _mm256_add_pd(fiz0,tz);
1022
1023             fjx0             = _mm256_add_pd(fjx0,tx);
1024             fjy0             = _mm256_add_pd(fjy0,ty);
1025             fjz0             = _mm256_add_pd(fjz0,tz);
1026
1027             }
1028
1029             /**************************
1030              * CALCULATE INTERACTIONS *
1031              **************************/
1032
1033             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1034             {
1035
1036             /* Compute parameters for interactions between i and j atoms */
1037             qq10             = _mm256_mul_pd(iq1,jq0);
1038
1039             /* REACTION-FIELD ELECTROSTATICS */
1040             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1041
1042             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1043
1044             fscal            = felec;
1045
1046             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1047
1048             /* Calculate temporary vectorial force */
1049             tx               = _mm256_mul_pd(fscal,dx10);
1050             ty               = _mm256_mul_pd(fscal,dy10);
1051             tz               = _mm256_mul_pd(fscal,dz10);
1052
1053             /* Update vectorial force */
1054             fix1             = _mm256_add_pd(fix1,tx);
1055             fiy1             = _mm256_add_pd(fiy1,ty);
1056             fiz1             = _mm256_add_pd(fiz1,tz);
1057
1058             fjx0             = _mm256_add_pd(fjx0,tx);
1059             fjy0             = _mm256_add_pd(fjy0,ty);
1060             fjz0             = _mm256_add_pd(fjz0,tz);
1061
1062             }
1063
1064             /**************************
1065              * CALCULATE INTERACTIONS *
1066              **************************/
1067
1068             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1069             {
1070
1071             /* Compute parameters for interactions between i and j atoms */
1072             qq20             = _mm256_mul_pd(iq2,jq0);
1073
1074             /* REACTION-FIELD ELECTROSTATICS */
1075             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1076
1077             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1078
1079             fscal            = felec;
1080
1081             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1082
1083             /* Calculate temporary vectorial force */
1084             tx               = _mm256_mul_pd(fscal,dx20);
1085             ty               = _mm256_mul_pd(fscal,dy20);
1086             tz               = _mm256_mul_pd(fscal,dz20);
1087
1088             /* Update vectorial force */
1089             fix2             = _mm256_add_pd(fix2,tx);
1090             fiy2             = _mm256_add_pd(fiy2,ty);
1091             fiz2             = _mm256_add_pd(fiz2,tz);
1092
1093             fjx0             = _mm256_add_pd(fjx0,tx);
1094             fjy0             = _mm256_add_pd(fjy0,ty);
1095             fjz0             = _mm256_add_pd(fjz0,tz);
1096
1097             }
1098
1099             /**************************
1100              * CALCULATE INTERACTIONS *
1101              **************************/
1102
1103             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1104             {
1105
1106             /* Compute parameters for interactions between i and j atoms */
1107             qq30             = _mm256_mul_pd(iq3,jq0);
1108
1109             /* REACTION-FIELD ELECTROSTATICS */
1110             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1111
1112             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1113
1114             fscal            = felec;
1115
1116             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm256_mul_pd(fscal,dx30);
1120             ty               = _mm256_mul_pd(fscal,dy30);
1121             tz               = _mm256_mul_pd(fscal,dz30);
1122
1123             /* Update vectorial force */
1124             fix3             = _mm256_add_pd(fix3,tx);
1125             fiy3             = _mm256_add_pd(fiy3,ty);
1126             fiz3             = _mm256_add_pd(fiz3,tz);
1127
1128             fjx0             = _mm256_add_pd(fjx0,tx);
1129             fjy0             = _mm256_add_pd(fjy0,ty);
1130             fjz0             = _mm256_add_pd(fjz0,tz);
1131
1132             }
1133
1134             fjptrA             = f+j_coord_offsetA;
1135             fjptrB             = f+j_coord_offsetB;
1136             fjptrC             = f+j_coord_offsetC;
1137             fjptrD             = f+j_coord_offsetD;
1138
1139             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1140
1141             /* Inner loop uses 149 flops */
1142         }
1143
1144         if(jidx<j_index_end)
1145         {
1146
1147             /* Get j neighbor index, and coordinate index */
1148             jnrlistA         = jjnr[jidx];
1149             jnrlistB         = jjnr[jidx+1];
1150             jnrlistC         = jjnr[jidx+2];
1151             jnrlistD         = jjnr[jidx+3];
1152             /* Sign of each element will be negative for non-real atoms.
1153              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1154              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1155              */
1156             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1157
1158             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1159             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1160             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1161
1162             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1163             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1164             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1165             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1166             j_coord_offsetA  = DIM*jnrA;
1167             j_coord_offsetB  = DIM*jnrB;
1168             j_coord_offsetC  = DIM*jnrC;
1169             j_coord_offsetD  = DIM*jnrD;
1170
1171             /* load j atom coordinates */
1172             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1173                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1174                                                  &jx0,&jy0,&jz0);
1175
1176             /* Calculate displacement vector */
1177             dx00             = _mm256_sub_pd(ix0,jx0);
1178             dy00             = _mm256_sub_pd(iy0,jy0);
1179             dz00             = _mm256_sub_pd(iz0,jz0);
1180             dx10             = _mm256_sub_pd(ix1,jx0);
1181             dy10             = _mm256_sub_pd(iy1,jy0);
1182             dz10             = _mm256_sub_pd(iz1,jz0);
1183             dx20             = _mm256_sub_pd(ix2,jx0);
1184             dy20             = _mm256_sub_pd(iy2,jy0);
1185             dz20             = _mm256_sub_pd(iz2,jz0);
1186             dx30             = _mm256_sub_pd(ix3,jx0);
1187             dy30             = _mm256_sub_pd(iy3,jy0);
1188             dz30             = _mm256_sub_pd(iz3,jz0);
1189
1190             /* Calculate squared distance and things based on it */
1191             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1192             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1193             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1194             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1195
1196             rinv00           = avx256_invsqrt_d(rsq00);
1197             rinv10           = avx256_invsqrt_d(rsq10);
1198             rinv20           = avx256_invsqrt_d(rsq20);
1199             rinv30           = avx256_invsqrt_d(rsq30);
1200
1201             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1202             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1203             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1204             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1205
1206             /* Load parameters for j particles */
1207             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1208                                                                  charge+jnrC+0,charge+jnrD+0);
1209             vdwjidx0A        = 2*vdwtype[jnrA+0];
1210             vdwjidx0B        = 2*vdwtype[jnrB+0];
1211             vdwjidx0C        = 2*vdwtype[jnrC+0];
1212             vdwjidx0D        = 2*vdwtype[jnrD+0];
1213
1214             fjx0             = _mm256_setzero_pd();
1215             fjy0             = _mm256_setzero_pd();
1216             fjz0             = _mm256_setzero_pd();
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1223             {
1224
1225             r00              = _mm256_mul_pd(rsq00,rinv00);
1226             r00              = _mm256_andnot_pd(dummy_mask,r00);
1227
1228             /* Compute parameters for interactions between i and j atoms */
1229             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1230                                             vdwioffsetptr0+vdwjidx0B,
1231                                             vdwioffsetptr0+vdwjidx0C,
1232                                             vdwioffsetptr0+vdwjidx0D,
1233                                             &c6_00,&c12_00);
1234
1235             /* LENNARD-JONES DISPERSION/REPULSION */
1236
1237             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1238             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1239             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1240             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1241             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1242
1243             d                = _mm256_sub_pd(r00,rswitch);
1244             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1245             d2               = _mm256_mul_pd(d,d);
1246             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
1247
1248             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1249
1250             /* Evaluate switch function */
1251             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1252             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1253             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1254
1255             fscal            = fvdw;
1256
1257             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1258
1259             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1260
1261             /* Calculate temporary vectorial force */
1262             tx               = _mm256_mul_pd(fscal,dx00);
1263             ty               = _mm256_mul_pd(fscal,dy00);
1264             tz               = _mm256_mul_pd(fscal,dz00);
1265
1266             /* Update vectorial force */
1267             fix0             = _mm256_add_pd(fix0,tx);
1268             fiy0             = _mm256_add_pd(fiy0,ty);
1269             fiz0             = _mm256_add_pd(fiz0,tz);
1270
1271             fjx0             = _mm256_add_pd(fjx0,tx);
1272             fjy0             = _mm256_add_pd(fjy0,ty);
1273             fjz0             = _mm256_add_pd(fjz0,tz);
1274
1275             }
1276
1277             /**************************
1278              * CALCULATE INTERACTIONS *
1279              **************************/
1280
1281             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1282             {
1283
1284             /* Compute parameters for interactions between i and j atoms */
1285             qq10             = _mm256_mul_pd(iq1,jq0);
1286
1287             /* REACTION-FIELD ELECTROSTATICS */
1288             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1289
1290             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1291
1292             fscal            = felec;
1293
1294             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1295
1296             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1297
1298             /* Calculate temporary vectorial force */
1299             tx               = _mm256_mul_pd(fscal,dx10);
1300             ty               = _mm256_mul_pd(fscal,dy10);
1301             tz               = _mm256_mul_pd(fscal,dz10);
1302
1303             /* Update vectorial force */
1304             fix1             = _mm256_add_pd(fix1,tx);
1305             fiy1             = _mm256_add_pd(fiy1,ty);
1306             fiz1             = _mm256_add_pd(fiz1,tz);
1307
1308             fjx0             = _mm256_add_pd(fjx0,tx);
1309             fjy0             = _mm256_add_pd(fjy0,ty);
1310             fjz0             = _mm256_add_pd(fjz0,tz);
1311
1312             }
1313
1314             /**************************
1315              * CALCULATE INTERACTIONS *
1316              **************************/
1317
1318             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1319             {
1320
1321             /* Compute parameters for interactions between i and j atoms */
1322             qq20             = _mm256_mul_pd(iq2,jq0);
1323
1324             /* REACTION-FIELD ELECTROSTATICS */
1325             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1326
1327             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1328
1329             fscal            = felec;
1330
1331             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1332
1333             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1334
1335             /* Calculate temporary vectorial force */
1336             tx               = _mm256_mul_pd(fscal,dx20);
1337             ty               = _mm256_mul_pd(fscal,dy20);
1338             tz               = _mm256_mul_pd(fscal,dz20);
1339
1340             /* Update vectorial force */
1341             fix2             = _mm256_add_pd(fix2,tx);
1342             fiy2             = _mm256_add_pd(fiy2,ty);
1343             fiz2             = _mm256_add_pd(fiz2,tz);
1344
1345             fjx0             = _mm256_add_pd(fjx0,tx);
1346             fjy0             = _mm256_add_pd(fjy0,ty);
1347             fjz0             = _mm256_add_pd(fjz0,tz);
1348
1349             }
1350
1351             /**************************
1352              * CALCULATE INTERACTIONS *
1353              **************************/
1354
1355             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1356             {
1357
1358             /* Compute parameters for interactions between i and j atoms */
1359             qq30             = _mm256_mul_pd(iq3,jq0);
1360
1361             /* REACTION-FIELD ELECTROSTATICS */
1362             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1363
1364             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1365
1366             fscal            = felec;
1367
1368             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1369
1370             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1371
1372             /* Calculate temporary vectorial force */
1373             tx               = _mm256_mul_pd(fscal,dx30);
1374             ty               = _mm256_mul_pd(fscal,dy30);
1375             tz               = _mm256_mul_pd(fscal,dz30);
1376
1377             /* Update vectorial force */
1378             fix3             = _mm256_add_pd(fix3,tx);
1379             fiy3             = _mm256_add_pd(fiy3,ty);
1380             fiz3             = _mm256_add_pd(fiz3,tz);
1381
1382             fjx0             = _mm256_add_pd(fjx0,tx);
1383             fjy0             = _mm256_add_pd(fjy0,ty);
1384             fjz0             = _mm256_add_pd(fjz0,tz);
1385
1386             }
1387
1388             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1389             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1390             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1391             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1392
1393             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1394
1395             /* Inner loop uses 150 flops */
1396         }
1397
1398         /* End of innermost loop */
1399
1400         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1401                                                  f+i_coord_offset,fshift+i_shift_offset);
1402
1403         /* Increment number of inner iterations */
1404         inneriter                  += j_index_end - j_index_start;
1405
1406         /* Outer loop uses 24 flops */
1407     }
1408
1409     /* Increment number of outer iterations */
1410     outeriter        += nri;
1411
1412     /* Update outer/inner flops */
1413
1414     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1415 }