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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 107     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 108     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 109     real             rswitch_scalar,d_scalar;
 110     __m256d          dummy_mask,cutoff_mask;
 111     __m128           tmpmask0,tmpmask1;
 112     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 113     __m256d          one     = _mm256_set1_pd(1.0);
 114     __m256d          two     = _mm256_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     krf              = _mm256_set1_pd(fr->ic->k_rf);
 129     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 130     crf              = _mm256_set1_pd(fr->ic->c_rf);
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 138     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 139     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 140     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 141
 142     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 143     rcutoff_scalar   = fr->rcoulomb;
 144     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 145     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 146
 147     rswitch_scalar   = fr->rvdw_switch;
 148     rswitch          = _mm256_set1_pd(rswitch_scalar);
 149     /* Setup switch parameters */
 150     d_scalar         = rcutoff_scalar-rswitch_scalar;
 151     d                = _mm256_set1_pd(d_scalar);
 152     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 153     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 154     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 155     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 156     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 157     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 158
 159     /* Avoid stupid compiler warnings */
 160     jnrA = jnrB = jnrC = jnrD = 0;
 161     j_coord_offsetA = 0;
 162     j_coord_offsetB = 0;
 163     j_coord_offsetC = 0;
 164     j_coord_offsetD = 0;
 165
 166     outeriter        = 0;
 167     inneriter        = 0;
 168
 169     for(iidx=0;iidx<4*DIM;iidx++)
 170     {
 171         scratch[iidx] = 0.0;
 172     }
 173
 174     /* Start outer loop over neighborlists */
 175     for(iidx=0; iidx<nri; iidx++)
 176     {
 177         /* Load shift vector for this list */
 178         i_shift_offset   = DIM*shiftidx[iidx];
 179
 180         /* Load limits for loop over neighbors */
 181         j_index_start    = jindex[iidx];
 182         j_index_end      = jindex[iidx+1];
 183
 184         /* Get outer coordinate index */
 185         inr              = iinr[iidx];
 186         i_coord_offset   = DIM*inr;
 187
 188         /* Load i particle coords and add shift vector */
 189         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 190                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 191
 192         fix0             = _mm256_setzero_pd();
 193         fiy0             = _mm256_setzero_pd();
 194         fiz0             = _mm256_setzero_pd();
 195         fix1             = _mm256_setzero_pd();
 196         fiy1             = _mm256_setzero_pd();
 197         fiz1             = _mm256_setzero_pd();
 198         fix2             = _mm256_setzero_pd();
 199         fiy2             = _mm256_setzero_pd();
 200         fiz2             = _mm256_setzero_pd();
 201         fix3             = _mm256_setzero_pd();
 202         fiy3             = _mm256_setzero_pd();
 203         fiz3             = _mm256_setzero_pd();
 204
 205         /* Reset potential sums */
 206         velecsum         = _mm256_setzero_pd();
 207         vvdwsum          = _mm256_setzero_pd();
 208
 209         /* Start inner kernel loop */
 210         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 211         {
 212
 213             /* Get j neighbor index, and coordinate index */
 214             jnrA             = jjnr[jidx];
 215             jnrB             = jjnr[jidx+1];
 216             jnrC             = jjnr[jidx+2];
 217             jnrD             = jjnr[jidx+3];
 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
 223             /* load j atom coordinates */
 224             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 225                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 226                                                  &jx0,&jy0,&jz0);
 227
 228             /* Calculate displacement vector */
 229             dx00             = _mm256_sub_pd(ix0,jx0);
 230             dy00             = _mm256_sub_pd(iy0,jy0);
 231             dz00             = _mm256_sub_pd(iz0,jz0);
 232             dx10             = _mm256_sub_pd(ix1,jx0);
 233             dy10             = _mm256_sub_pd(iy1,jy0);
 234             dz10             = _mm256_sub_pd(iz1,jz0);
 235             dx20             = _mm256_sub_pd(ix2,jx0);
 236             dy20             = _mm256_sub_pd(iy2,jy0);
 237             dz20             = _mm256_sub_pd(iz2,jz0);
 238             dx30             = _mm256_sub_pd(ix3,jx0);
 239             dy30             = _mm256_sub_pd(iy3,jy0);
 240             dz30             = _mm256_sub_pd(iz3,jz0);
 241
 242             /* Calculate squared distance and things based on it */
 243             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 244             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 245             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 246             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 247
 248             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 249             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 250             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 251             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 252
 253             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 254             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 255             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 256             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 257
 258             /* Load parameters for j particles */
 259             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 260                                                                  charge+jnrC+0,charge+jnrD+0);
 261             vdwjidx0A        = 2*vdwtype[jnrA+0];
 262             vdwjidx0B        = 2*vdwtype[jnrB+0];
 263             vdwjidx0C        = 2*vdwtype[jnrC+0];
 264             vdwjidx0D        = 2*vdwtype[jnrD+0];
 265
 266             fjx0             = _mm256_setzero_pd();
 267             fjy0             = _mm256_setzero_pd();
 268             fjz0             = _mm256_setzero_pd();
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 275             {
 276
 277             r00              = _mm256_mul_pd(rsq00,rinv00);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 281                                             vdwioffsetptr0+vdwjidx0B,
 282                                             vdwioffsetptr0+vdwjidx0C,
 283                                             vdwioffsetptr0+vdwjidx0D,
 284                                             &c6_00,&c12_00);
 285
 286             /* LENNARD-JONES DISPERSION/REPULSION */
 287
 288             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 289             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 290             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 291             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 292             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 293
 294             d                = _mm256_sub_pd(r00,rswitch);
 295             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 296             d2               = _mm256_mul_pd(d,d);
 297             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)))))));
 298
 299             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 300
 301             /* Evaluate switch function */
 302             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 303             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 304             vvdw             = _mm256_mul_pd(vvdw,sw);
 305             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 309             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 310
 311             fscal            = fvdw;
 312
 313             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm256_mul_pd(fscal,dx00);
 317             ty               = _mm256_mul_pd(fscal,dy00);
 318             tz               = _mm256_mul_pd(fscal,dz00);
 319
 320             /* Update vectorial force */
 321             fix0             = _mm256_add_pd(fix0,tx);
 322             fiy0             = _mm256_add_pd(fiy0,ty);
 323             fiz0             = _mm256_add_pd(fiz0,tz);
 324
 325             fjx0             = _mm256_add_pd(fjx0,tx);
 326             fjy0             = _mm256_add_pd(fjy0,ty);
 327             fjz0             = _mm256_add_pd(fjz0,tz);
 328
 329             }
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 336             {
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq10             = _mm256_mul_pd(iq1,jq0);
 340
 341             /* REACTION-FIELD ELECTROSTATICS */
 342             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 343             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 344
 345             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velec            = _mm256_and_pd(velec,cutoff_mask);
 349             velecsum         = _mm256_add_pd(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm256_mul_pd(fscal,dx10);
 357             ty               = _mm256_mul_pd(fscal,dy10);
 358             tz               = _mm256_mul_pd(fscal,dz10);
 359
 360             /* Update vectorial force */
 361             fix1             = _mm256_add_pd(fix1,tx);
 362             fiy1             = _mm256_add_pd(fiy1,ty);
 363             fiz1             = _mm256_add_pd(fiz1,tz);
 364
 365             fjx0             = _mm256_add_pd(fjx0,tx);
 366             fjy0             = _mm256_add_pd(fjy0,ty);
 367             fjz0             = _mm256_add_pd(fjz0,tz);
 368
 369             }
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 376             {
 377
 378             /* Compute parameters for interactions between i and j atoms */
 379             qq20             = _mm256_mul_pd(iq2,jq0);
 380
 381             /* REACTION-FIELD ELECTROSTATICS */
 382             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 383             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 384
 385             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velec            = _mm256_and_pd(velec,cutoff_mask);
 389             velecsum         = _mm256_add_pd(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm256_mul_pd(fscal,dx20);
 397             ty               = _mm256_mul_pd(fscal,dy20);
 398             tz               = _mm256_mul_pd(fscal,dz20);
 399
 400             /* Update vectorial force */
 401             fix2             = _mm256_add_pd(fix2,tx);
 402             fiy2             = _mm256_add_pd(fiy2,ty);
 403             fiz2             = _mm256_add_pd(fiz2,tz);
 404
 405             fjx0             = _mm256_add_pd(fjx0,tx);
 406             fjy0             = _mm256_add_pd(fjy0,ty);
 407             fjz0             = _mm256_add_pd(fjz0,tz);
 408
 409             }
 410
 411             /**************************
 412              * CALCULATE INTERACTIONS *
 413              **************************/
 414
 415             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 416             {
 417
 418             /* Compute parameters for interactions between i and j atoms */
 419             qq30             = _mm256_mul_pd(iq3,jq0);
 420
 421             /* REACTION-FIELD ELECTROSTATICS */
 422             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 423             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 424
 425             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 426
 427             /* Update potential sum for this i atom from the interaction with this j atom. */
 428             velec            = _mm256_and_pd(velec,cutoff_mask);
 429             velecsum         = _mm256_add_pd(velecsum,velec);
 430
 431             fscal            = felec;
 432
 433             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 434
 435             /* Calculate temporary vectorial force */
 436             tx               = _mm256_mul_pd(fscal,dx30);
 437             ty               = _mm256_mul_pd(fscal,dy30);
 438             tz               = _mm256_mul_pd(fscal,dz30);
 439
 440             /* Update vectorial force */
 441             fix3             = _mm256_add_pd(fix3,tx);
 442             fiy3             = _mm256_add_pd(fiy3,ty);
 443             fiz3             = _mm256_add_pd(fiz3,tz);
 444
 445             fjx0             = _mm256_add_pd(fjx0,tx);
 446             fjy0             = _mm256_add_pd(fjy0,ty);
 447             fjz0             = _mm256_add_pd(fjz0,tz);
 448
 449             }
 450
 451             fjptrA             = f+j_coord_offsetA;
 452             fjptrB             = f+j_coord_offsetB;
 453             fjptrC             = f+j_coord_offsetC;
 454             fjptrD             = f+j_coord_offsetD;
 455
 456             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 457
 458             /* Inner loop uses 170 flops */
 459         }
 460
 461         if(jidx<j_index_end)
 462         {
 463
 464             /* Get j neighbor index, and coordinate index */
 465             jnrlistA         = jjnr[jidx];
 466             jnrlistB         = jjnr[jidx+1];
 467             jnrlistC         = jjnr[jidx+2];
 468             jnrlistD         = jjnr[jidx+3];
 469             /* Sign of each element will be negative for non-real atoms.
 470              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 471              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 472              */
 473             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 474
 475             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 476             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 477             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 478
 479             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 480             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 481             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 482             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 483             j_coord_offsetA  = DIM*jnrA;
 484             j_coord_offsetB  = DIM*jnrB;
 485             j_coord_offsetC  = DIM*jnrC;
 486             j_coord_offsetD  = DIM*jnrD;
 487
 488             /* load j atom coordinates */
 489             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 490                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 491                                                  &jx0,&jy0,&jz0);
 492
 493             /* Calculate displacement vector */
 494             dx00             = _mm256_sub_pd(ix0,jx0);
 495             dy00             = _mm256_sub_pd(iy0,jy0);
 496             dz00             = _mm256_sub_pd(iz0,jz0);
 497             dx10             = _mm256_sub_pd(ix1,jx0);
 498             dy10             = _mm256_sub_pd(iy1,jy0);
 499             dz10             = _mm256_sub_pd(iz1,jz0);
 500             dx20             = _mm256_sub_pd(ix2,jx0);
 501             dy20             = _mm256_sub_pd(iy2,jy0);
 502             dz20             = _mm256_sub_pd(iz2,jz0);
 503             dx30             = _mm256_sub_pd(ix3,jx0);
 504             dy30             = _mm256_sub_pd(iy3,jy0);
 505             dz30             = _mm256_sub_pd(iz3,jz0);
 506
 507             /* Calculate squared distance and things based on it */
 508             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 509             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 510             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 511             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 512
 513             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 514             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 515             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 516             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 517
 518             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 519             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 520             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 521             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 522
 523             /* Load parameters for j particles */
 524             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 525                                                                  charge+jnrC+0,charge+jnrD+0);
 526             vdwjidx0A        = 2*vdwtype[jnrA+0];
 527             vdwjidx0B        = 2*vdwtype[jnrB+0];
 528             vdwjidx0C        = 2*vdwtype[jnrC+0];
 529             vdwjidx0D        = 2*vdwtype[jnrD+0];
 530
 531             fjx0             = _mm256_setzero_pd();
 532             fjy0             = _mm256_setzero_pd();
 533             fjz0             = _mm256_setzero_pd();
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 540             {
 541
 542             r00              = _mm256_mul_pd(rsq00,rinv00);
 543             r00              = _mm256_andnot_pd(dummy_mask,r00);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 547                                             vdwioffsetptr0+vdwjidx0B,
 548                                             vdwioffsetptr0+vdwjidx0C,
 549                                             vdwioffsetptr0+vdwjidx0D,
 550                                             &c6_00,&c12_00);
 551
 552             /* LENNARD-JONES DISPERSION/REPULSION */
 553
 554             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 555             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 556             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 557             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 558             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 559
 560             d                = _mm256_sub_pd(r00,rswitch);
 561             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 562             d2               = _mm256_mul_pd(d,d);
 563             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)))))));
 564
 565             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 566
 567             /* Evaluate switch function */
 568             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 569             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
 570             vvdw             = _mm256_mul_pd(vvdw,sw);
 571             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 572
 573             /* Update potential sum for this i atom from the interaction with this j atom. */
 574             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 575             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 576             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 577
 578             fscal            = fvdw;
 579
 580             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 581
 582             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm256_mul_pd(fscal,dx00);
 586             ty               = _mm256_mul_pd(fscal,dy00);
 587             tz               = _mm256_mul_pd(fscal,dz00);
 588
 589             /* Update vectorial force */
 590             fix0             = _mm256_add_pd(fix0,tx);
 591             fiy0             = _mm256_add_pd(fiy0,ty);
 592             fiz0             = _mm256_add_pd(fiz0,tz);
 593
 594             fjx0             = _mm256_add_pd(fjx0,tx);
 595             fjy0             = _mm256_add_pd(fjy0,ty);
 596             fjz0             = _mm256_add_pd(fjz0,tz);
 597
 598             }
 599
 600             /**************************
 601              * CALCULATE INTERACTIONS *
 602              **************************/
 603
 604             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 605             {
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             qq10             = _mm256_mul_pd(iq1,jq0);
 609
 610             /* REACTION-FIELD ELECTROSTATICS */
 611             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 612             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 613
 614             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velec            = _mm256_and_pd(velec,cutoff_mask);
 618             velec            = _mm256_andnot_pd(dummy_mask,velec);
 619             velecsum         = _mm256_add_pd(velecsum,velec);
 620
 621             fscal            = felec;
 622
 623             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 624
 625             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 626
 627             /* Calculate temporary vectorial force */
 628             tx               = _mm256_mul_pd(fscal,dx10);
 629             ty               = _mm256_mul_pd(fscal,dy10);
 630             tz               = _mm256_mul_pd(fscal,dz10);
 631
 632             /* Update vectorial force */
 633             fix1             = _mm256_add_pd(fix1,tx);
 634             fiy1             = _mm256_add_pd(fiy1,ty);
 635             fiz1             = _mm256_add_pd(fiz1,tz);
 636
 637             fjx0             = _mm256_add_pd(fjx0,tx);
 638             fjy0             = _mm256_add_pd(fjy0,ty);
 639             fjz0             = _mm256_add_pd(fjz0,tz);
 640
 641             }
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 648             {
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq20             = _mm256_mul_pd(iq2,jq0);
 652
 653             /* REACTION-FIELD ELECTROSTATICS */
 654             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 655             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 656
 657             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 658
 659             /* Update potential sum for this i atom from the interaction with this j atom. */
 660             velec            = _mm256_and_pd(velec,cutoff_mask);
 661             velec            = _mm256_andnot_pd(dummy_mask,velec);
 662             velecsum         = _mm256_add_pd(velecsum,velec);
 663
 664             fscal            = felec;
 665
 666             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 667
 668             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 669
 670             /* Calculate temporary vectorial force */
 671             tx               = _mm256_mul_pd(fscal,dx20);
 672             ty               = _mm256_mul_pd(fscal,dy20);
 673             tz               = _mm256_mul_pd(fscal,dz20);
 674
 675             /* Update vectorial force */
 676             fix2             = _mm256_add_pd(fix2,tx);
 677             fiy2             = _mm256_add_pd(fiy2,ty);
 678             fiz2             = _mm256_add_pd(fiz2,tz);
 679
 680             fjx0             = _mm256_add_pd(fjx0,tx);
 681             fjy0             = _mm256_add_pd(fjy0,ty);
 682             fjz0             = _mm256_add_pd(fjz0,tz);
 683
 684             }
 685
 686             /**************************
 687              * CALCULATE INTERACTIONS *
 688              **************************/
 689
 690             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 691             {
 692
 693             /* Compute parameters for interactions between i and j atoms */
 694             qq30             = _mm256_mul_pd(iq3,jq0);
 695
 696             /* REACTION-FIELD ELECTROSTATICS */
 697             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 698             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 699
 700             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 701
 702             /* Update potential sum for this i atom from the interaction with this j atom. */
 703             velec            = _mm256_and_pd(velec,cutoff_mask);
 704             velec            = _mm256_andnot_pd(dummy_mask,velec);
 705             velecsum         = _mm256_add_pd(velecsum,velec);
 706
 707             fscal            = felec;
 708
 709             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 710
 711             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 712
 713             /* Calculate temporary vectorial force */
 714             tx               = _mm256_mul_pd(fscal,dx30);
 715             ty               = _mm256_mul_pd(fscal,dy30);
 716             tz               = _mm256_mul_pd(fscal,dz30);
 717
 718             /* Update vectorial force */
 719             fix3             = _mm256_add_pd(fix3,tx);
 720             fiy3             = _mm256_add_pd(fiy3,ty);
 721             fiz3             = _mm256_add_pd(fiz3,tz);
 722
 723             fjx0             = _mm256_add_pd(fjx0,tx);
 724             fjy0             = _mm256_add_pd(fjy0,ty);
 725             fjz0             = _mm256_add_pd(fjz0,tz);
 726
 727             }
 728
 729             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 730             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 731             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 732             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 733
 734             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 735
 736             /* Inner loop uses 171 flops */
 737         }
 738
 739         /* End of innermost loop */
 740
 741         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 742                                                  f+i_coord_offset,fshift+i_shift_offset);
 743
 744         ggid                        = gid[iidx];
 745         /* Update potential energies */
 746         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 747         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 748
 749         /* Increment number of inner iterations */
 750         inneriter                  += j_index_end - j_index_start;
 751
 752         /* Outer loop uses 26 flops */
 753     }
 754
 755     /* Increment number of outer iterations */
 756     outeriter        += nri;
 757
 758     /* Update outer/inner flops */
 759
 760     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*171);
 761 }
 762 /*
 763  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_double
 764  * Electrostatics interaction: ReactionField
 765  * VdW interaction:            LennardJones
 766  * Geometry:                   Water4-Particle
 767  * Calculate force/pot:        Force
 768  */
 769 void
 770 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_avx_256_double
 771                     (t_nblist                    * gmx_restrict       nlist,
 772                      rvec                        * gmx_restrict          xx,
 773                      rvec                        * gmx_restrict          ff,
 774                      t_forcerec                  * gmx_restrict          fr,
 775                      t_mdatoms                   * gmx_restrict     mdatoms,
 776                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 777                      t_nrnb                      * gmx_restrict        nrnb)
 778 {
 779     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 780      * just 0 for non-waters.
 781      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 782      * jnr indices corresponding to data put in the four positions in the SIMD register.
 783      */
 784     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 785     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 786     int              jnrA,jnrB,jnrC,jnrD;
 787     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 788     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 789     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 790     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 791     real             rcutoff_scalar;
 792     real             *shiftvec,*fshift,*x,*f;
 793     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 794     real             scratch[4*DIM];
 795     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 796     real *           vdwioffsetptr0;
 797     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 798     real *           vdwioffsetptr1;
 799     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 800     real *           vdwioffsetptr2;
 801     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 802     real *           vdwioffsetptr3;
 803     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 804     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 805     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 806     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 807     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 808     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 809     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 810     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 811     real             *charge;
 812     int              nvdwtype;
 813     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 814     int              *vdwtype;
 815     real             *vdwparam;
 816     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 817     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 818     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 819     real             rswitch_scalar,d_scalar;
 820     __m256d          dummy_mask,cutoff_mask;
 821     __m128           tmpmask0,tmpmask1;
 822     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 823     __m256d          one     = _mm256_set1_pd(1.0);
 824     __m256d          two     = _mm256_set1_pd(2.0);
 825     x                = xx[0];
 826     f                = ff[0];
 827
 828     nri              = nlist->nri;
 829     iinr             = nlist->iinr;
 830     jindex           = nlist->jindex;
 831     jjnr             = nlist->jjnr;
 832     shiftidx         = nlist->shift;
 833     gid              = nlist->gid;
 834     shiftvec         = fr->shift_vec[0];
 835     fshift           = fr->fshift[0];
 836     facel            = _mm256_set1_pd(fr->epsfac);
 837     charge           = mdatoms->chargeA;
 838     krf              = _mm256_set1_pd(fr->ic->k_rf);
 839     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 840     crf              = _mm256_set1_pd(fr->ic->c_rf);
 841     nvdwtype         = fr->ntype;
 842     vdwparam         = fr->nbfp;
 843     vdwtype          = mdatoms->typeA;
 844
 845     /* Setup water-specific parameters */
 846     inr              = nlist->iinr[0];
 847     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 848     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 849     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 850     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 851
 852     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 853     rcutoff_scalar   = fr->rcoulomb;
 854     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 855     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 856
 857     rswitch_scalar   = fr->rvdw_switch;
 858     rswitch          = _mm256_set1_pd(rswitch_scalar);
 859     /* Setup switch parameters */
 860     d_scalar         = rcutoff_scalar-rswitch_scalar;
 861     d                = _mm256_set1_pd(d_scalar);
 862     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 863     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 864     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 865     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 866     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 867     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 868
 869     /* Avoid stupid compiler warnings */
 870     jnrA = jnrB = jnrC = jnrD = 0;
 871     j_coord_offsetA = 0;
 872     j_coord_offsetB = 0;
 873     j_coord_offsetC = 0;
 874     j_coord_offsetD = 0;
 875
 876     outeriter        = 0;
 877     inneriter        = 0;
 878
 879     for(iidx=0;iidx<4*DIM;iidx++)
 880     {
 881         scratch[iidx] = 0.0;
 882     }
 883
 884     /* Start outer loop over neighborlists */
 885     for(iidx=0; iidx<nri; iidx++)
 886     {
 887         /* Load shift vector for this list */
 888         i_shift_offset   = DIM*shiftidx[iidx];
 889
 890         /* Load limits for loop over neighbors */
 891         j_index_start    = jindex[iidx];
 892         j_index_end      = jindex[iidx+1];
 893
 894         /* Get outer coordinate index */
 895         inr              = iinr[iidx];
 896         i_coord_offset   = DIM*inr;
 897
 898         /* Load i particle coords and add shift vector */
 899         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 900                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 901
 902         fix0             = _mm256_setzero_pd();
 903         fiy0             = _mm256_setzero_pd();
 904         fiz0             = _mm256_setzero_pd();
 905         fix1             = _mm256_setzero_pd();
 906         fiy1             = _mm256_setzero_pd();
 907         fiz1             = _mm256_setzero_pd();
 908         fix2             = _mm256_setzero_pd();
 909         fiy2             = _mm256_setzero_pd();
 910         fiz2             = _mm256_setzero_pd();
 911         fix3             = _mm256_setzero_pd();
 912         fiy3             = _mm256_setzero_pd();
 913         fiz3             = _mm256_setzero_pd();
 914
 915         /* Start inner kernel loop */
 916         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 917         {
 918
 919             /* Get j neighbor index, and coordinate index */
 920             jnrA             = jjnr[jidx];
 921             jnrB             = jjnr[jidx+1];
 922             jnrC             = jjnr[jidx+2];
 923             jnrD             = jjnr[jidx+3];
 924             j_coord_offsetA  = DIM*jnrA;
 925             j_coord_offsetB  = DIM*jnrB;
 926             j_coord_offsetC  = DIM*jnrC;
 927             j_coord_offsetD  = DIM*jnrD;
 928
 929             /* load j atom coordinates */
 930             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 931                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 932                                                  &jx0,&jy0,&jz0);
 933
 934             /* Calculate displacement vector */
 935             dx00             = _mm256_sub_pd(ix0,jx0);
 936             dy00             = _mm256_sub_pd(iy0,jy0);
 937             dz00             = _mm256_sub_pd(iz0,jz0);
 938             dx10             = _mm256_sub_pd(ix1,jx0);
 939             dy10             = _mm256_sub_pd(iy1,jy0);
 940             dz10             = _mm256_sub_pd(iz1,jz0);
 941             dx20             = _mm256_sub_pd(ix2,jx0);
 942             dy20             = _mm256_sub_pd(iy2,jy0);
 943             dz20             = _mm256_sub_pd(iz2,jz0);
 944             dx30             = _mm256_sub_pd(ix3,jx0);
 945             dy30             = _mm256_sub_pd(iy3,jy0);
 946             dz30             = _mm256_sub_pd(iz3,jz0);
 947
 948             /* Calculate squared distance and things based on it */
 949             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 950             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 951             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 952             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 953
 954             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 955             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 956             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 957             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 958
 959             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 960             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 961             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 962             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 963
 964             /* Load parameters for j particles */
 965             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 966                                                                  charge+jnrC+0,charge+jnrD+0);
 967             vdwjidx0A        = 2*vdwtype[jnrA+0];
 968             vdwjidx0B        = 2*vdwtype[jnrB+0];
 969             vdwjidx0C        = 2*vdwtype[jnrC+0];
 970             vdwjidx0D        = 2*vdwtype[jnrD+0];
 971
 972             fjx0             = _mm256_setzero_pd();
 973             fjy0             = _mm256_setzero_pd();
 974             fjz0             = _mm256_setzero_pd();
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 981             {
 982
 983             r00              = _mm256_mul_pd(rsq00,rinv00);
 984
 985             /* Compute parameters for interactions between i and j atoms */
 986             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 987                                             vdwioffsetptr0+vdwjidx0B,
 988                                             vdwioffsetptr0+vdwjidx0C,
 989                                             vdwioffsetptr0+vdwjidx0D,
 990                                             &c6_00,&c12_00);
 991
 992             /* LENNARD-JONES DISPERSION/REPULSION */
 993
 994             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 995             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 996             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 997             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 998             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 999
1000             d                = _mm256_sub_pd(r00,rswitch);
1001             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1002             d2               = _mm256_mul_pd(d,d);
1003             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)))))));
1004
1005             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1006
1007             /* Evaluate switch function */
1008             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1009             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1010             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1011
1012             fscal            = fvdw;
1013
1014             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1015
1016             /* Calculate temporary vectorial force */
1017             tx               = _mm256_mul_pd(fscal,dx00);
1018             ty               = _mm256_mul_pd(fscal,dy00);
1019             tz               = _mm256_mul_pd(fscal,dz00);
1020
1021             /* Update vectorial force */
1022             fix0             = _mm256_add_pd(fix0,tx);
1023             fiy0             = _mm256_add_pd(fiy0,ty);
1024             fiz0             = _mm256_add_pd(fiz0,tz);
1025
1026             fjx0             = _mm256_add_pd(fjx0,tx);
1027             fjy0             = _mm256_add_pd(fjy0,ty);
1028             fjz0             = _mm256_add_pd(fjz0,tz);
1029
1030             }
1031
1032             /**************************
1033              * CALCULATE INTERACTIONS *
1034              **************************/
1035
1036             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1037             {
1038
1039             /* Compute parameters for interactions between i and j atoms */
1040             qq10             = _mm256_mul_pd(iq1,jq0);
1041
1042             /* REACTION-FIELD ELECTROSTATICS */
1043             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1044
1045             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1046
1047             fscal            = felec;
1048
1049             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1050
1051             /* Calculate temporary vectorial force */
1052             tx               = _mm256_mul_pd(fscal,dx10);
1053             ty               = _mm256_mul_pd(fscal,dy10);
1054             tz               = _mm256_mul_pd(fscal,dz10);
1055
1056             /* Update vectorial force */
1057             fix1             = _mm256_add_pd(fix1,tx);
1058             fiy1             = _mm256_add_pd(fiy1,ty);
1059             fiz1             = _mm256_add_pd(fiz1,tz);
1060
1061             fjx0             = _mm256_add_pd(fjx0,tx);
1062             fjy0             = _mm256_add_pd(fjy0,ty);
1063             fjz0             = _mm256_add_pd(fjz0,tz);
1064
1065             }
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1072             {
1073
1074             /* Compute parameters for interactions between i and j atoms */
1075             qq20             = _mm256_mul_pd(iq2,jq0);
1076
1077             /* REACTION-FIELD ELECTROSTATICS */
1078             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1079
1080             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1081
1082             fscal            = felec;
1083
1084             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1085
1086             /* Calculate temporary vectorial force */
1087             tx               = _mm256_mul_pd(fscal,dx20);
1088             ty               = _mm256_mul_pd(fscal,dy20);
1089             tz               = _mm256_mul_pd(fscal,dz20);
1090
1091             /* Update vectorial force */
1092             fix2             = _mm256_add_pd(fix2,tx);
1093             fiy2             = _mm256_add_pd(fiy2,ty);
1094             fiz2             = _mm256_add_pd(fiz2,tz);
1095
1096             fjx0             = _mm256_add_pd(fjx0,tx);
1097             fjy0             = _mm256_add_pd(fjy0,ty);
1098             fjz0             = _mm256_add_pd(fjz0,tz);
1099
1100             }
1101
1102             /**************************
1103              * CALCULATE INTERACTIONS *
1104              **************************/
1105
1106             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1107             {
1108
1109             /* Compute parameters for interactions between i and j atoms */
1110             qq30             = _mm256_mul_pd(iq3,jq0);
1111
1112             /* REACTION-FIELD ELECTROSTATICS */
1113             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1114
1115             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1116
1117             fscal            = felec;
1118
1119             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1120
1121             /* Calculate temporary vectorial force */
1122             tx               = _mm256_mul_pd(fscal,dx30);
1123             ty               = _mm256_mul_pd(fscal,dy30);
1124             tz               = _mm256_mul_pd(fscal,dz30);
1125
1126             /* Update vectorial force */
1127             fix3             = _mm256_add_pd(fix3,tx);
1128             fiy3             = _mm256_add_pd(fiy3,ty);
1129             fiz3             = _mm256_add_pd(fiz3,tz);
1130
1131             fjx0             = _mm256_add_pd(fjx0,tx);
1132             fjy0             = _mm256_add_pd(fjy0,ty);
1133             fjz0             = _mm256_add_pd(fjz0,tz);
1134
1135             }
1136
1137             fjptrA             = f+j_coord_offsetA;
1138             fjptrB             = f+j_coord_offsetB;
1139             fjptrC             = f+j_coord_offsetC;
1140             fjptrD             = f+j_coord_offsetD;
1141
1142             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1143
1144             /* Inner loop uses 149 flops */
1145         }
1146
1147         if(jidx<j_index_end)
1148         {
1149
1150             /* Get j neighbor index, and coordinate index */
1151             jnrlistA         = jjnr[jidx];
1152             jnrlistB         = jjnr[jidx+1];
1153             jnrlistC         = jjnr[jidx+2];
1154             jnrlistD         = jjnr[jidx+3];
1155             /* Sign of each element will be negative for non-real atoms.
1156              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1157              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1158              */
1159             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1160
1161             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1162             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1163             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1164
1165             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1166             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1167             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1168             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1169             j_coord_offsetA  = DIM*jnrA;
1170             j_coord_offsetB  = DIM*jnrB;
1171             j_coord_offsetC  = DIM*jnrC;
1172             j_coord_offsetD  = DIM*jnrD;
1173
1174             /* load j atom coordinates */
1175             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1176                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1177                                                  &jx0,&jy0,&jz0);
1178
1179             /* Calculate displacement vector */
1180             dx00             = _mm256_sub_pd(ix0,jx0);
1181             dy00             = _mm256_sub_pd(iy0,jy0);
1182             dz00             = _mm256_sub_pd(iz0,jz0);
1183             dx10             = _mm256_sub_pd(ix1,jx0);
1184             dy10             = _mm256_sub_pd(iy1,jy0);
1185             dz10             = _mm256_sub_pd(iz1,jz0);
1186             dx20             = _mm256_sub_pd(ix2,jx0);
1187             dy20             = _mm256_sub_pd(iy2,jy0);
1188             dz20             = _mm256_sub_pd(iz2,jz0);
1189             dx30             = _mm256_sub_pd(ix3,jx0);
1190             dy30             = _mm256_sub_pd(iy3,jy0);
1191             dz30             = _mm256_sub_pd(iz3,jz0);
1192
1193             /* Calculate squared distance and things based on it */
1194             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1195             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1196             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1197             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1198
1199             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1200             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1201             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1202             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1203
1204             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1205             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1206             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1207             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1208
1209             /* Load parameters for j particles */
1210             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1211                                                                  charge+jnrC+0,charge+jnrD+0);
1212             vdwjidx0A        = 2*vdwtype[jnrA+0];
1213             vdwjidx0B        = 2*vdwtype[jnrB+0];
1214             vdwjidx0C        = 2*vdwtype[jnrC+0];
1215             vdwjidx0D        = 2*vdwtype[jnrD+0];
1216
1217             fjx0             = _mm256_setzero_pd();
1218             fjy0             = _mm256_setzero_pd();
1219             fjz0             = _mm256_setzero_pd();
1220
1221             /**************************
1222              * CALCULATE INTERACTIONS *
1223              **************************/
1224
1225             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1226             {
1227
1228             r00              = _mm256_mul_pd(rsq00,rinv00);
1229             r00              = _mm256_andnot_pd(dummy_mask,r00);
1230
1231             /* Compute parameters for interactions between i and j atoms */
1232             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1233                                             vdwioffsetptr0+vdwjidx0B,
1234                                             vdwioffsetptr0+vdwjidx0C,
1235                                             vdwioffsetptr0+vdwjidx0D,
1236                                             &c6_00,&c12_00);
1237
1238             /* LENNARD-JONES DISPERSION/REPULSION */
1239
1240             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1241             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
1242             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1243             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
1244             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
1245
1246             d                = _mm256_sub_pd(r00,rswitch);
1247             d                = _mm256_max_pd(d,_mm256_setzero_pd());
1248             d2               = _mm256_mul_pd(d,d);
1249             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)))))));
1250
1251             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
1252
1253             /* Evaluate switch function */
1254             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1255             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv00,_mm256_mul_pd(vvdw,dsw)) );
1256             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1257
1258             fscal            = fvdw;
1259
1260             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1261
1262             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1263
1264             /* Calculate temporary vectorial force */
1265             tx               = _mm256_mul_pd(fscal,dx00);
1266             ty               = _mm256_mul_pd(fscal,dy00);
1267             tz               = _mm256_mul_pd(fscal,dz00);
1268
1269             /* Update vectorial force */
1270             fix0             = _mm256_add_pd(fix0,tx);
1271             fiy0             = _mm256_add_pd(fiy0,ty);
1272             fiz0             = _mm256_add_pd(fiz0,tz);
1273
1274             fjx0             = _mm256_add_pd(fjx0,tx);
1275             fjy0             = _mm256_add_pd(fjy0,ty);
1276             fjz0             = _mm256_add_pd(fjz0,tz);
1277
1278             }
1279
1280             /**************************
1281              * CALCULATE INTERACTIONS *
1282              **************************/
1283
1284             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1285             {
1286
1287             /* Compute parameters for interactions between i and j atoms */
1288             qq10             = _mm256_mul_pd(iq1,jq0);
1289
1290             /* REACTION-FIELD ELECTROSTATICS */
1291             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1292
1293             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1294
1295             fscal            = felec;
1296
1297             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1298
1299             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1300
1301             /* Calculate temporary vectorial force */
1302             tx               = _mm256_mul_pd(fscal,dx10);
1303             ty               = _mm256_mul_pd(fscal,dy10);
1304             tz               = _mm256_mul_pd(fscal,dz10);
1305
1306             /* Update vectorial force */
1307             fix1             = _mm256_add_pd(fix1,tx);
1308             fiy1             = _mm256_add_pd(fiy1,ty);
1309             fiz1             = _mm256_add_pd(fiz1,tz);
1310
1311             fjx0             = _mm256_add_pd(fjx0,tx);
1312             fjy0             = _mm256_add_pd(fjy0,ty);
1313             fjz0             = _mm256_add_pd(fjz0,tz);
1314
1315             }
1316
1317             /**************************
1318              * CALCULATE INTERACTIONS *
1319              **************************/
1320
1321             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1322             {
1323
1324             /* Compute parameters for interactions between i and j atoms */
1325             qq20             = _mm256_mul_pd(iq2,jq0);
1326
1327             /* REACTION-FIELD ELECTROSTATICS */
1328             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1329
1330             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1331
1332             fscal            = felec;
1333
1334             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1335
1336             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1337
1338             /* Calculate temporary vectorial force */
1339             tx               = _mm256_mul_pd(fscal,dx20);
1340             ty               = _mm256_mul_pd(fscal,dy20);
1341             tz               = _mm256_mul_pd(fscal,dz20);
1342
1343             /* Update vectorial force */
1344             fix2             = _mm256_add_pd(fix2,tx);
1345             fiy2             = _mm256_add_pd(fiy2,ty);
1346             fiz2             = _mm256_add_pd(fiz2,tz);
1347
1348             fjx0             = _mm256_add_pd(fjx0,tx);
1349             fjy0             = _mm256_add_pd(fjy0,ty);
1350             fjz0             = _mm256_add_pd(fjz0,tz);
1351
1352             }
1353
1354             /**************************
1355              * CALCULATE INTERACTIONS *
1356              **************************/
1357
1358             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1359             {
1360
1361             /* Compute parameters for interactions between i and j atoms */
1362             qq30             = _mm256_mul_pd(iq3,jq0);
1363
1364             /* REACTION-FIELD ELECTROSTATICS */
1365             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1366
1367             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1368
1369             fscal            = felec;
1370
1371             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1372
1373             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1374
1375             /* Calculate temporary vectorial force */
1376             tx               = _mm256_mul_pd(fscal,dx30);
1377             ty               = _mm256_mul_pd(fscal,dy30);
1378             tz               = _mm256_mul_pd(fscal,dz30);
1379
1380             /* Update vectorial force */
1381             fix3             = _mm256_add_pd(fix3,tx);
1382             fiy3             = _mm256_add_pd(fiy3,ty);
1383             fiz3             = _mm256_add_pd(fiz3,tz);
1384
1385             fjx0             = _mm256_add_pd(fjx0,tx);
1386             fjy0             = _mm256_add_pd(fjy0,ty);
1387             fjz0             = _mm256_add_pd(fjz0,tz);
1388
1389             }
1390
1391             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1392             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1393             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1394             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1395
1396             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1397
1398             /* Inner loop uses 150 flops */
1399         }
1400
1401         /* End of innermost loop */
1402
1403         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1404                                                  f+i_coord_offset,fshift+i_shift_offset);
1405
1406         /* Increment number of inner iterations */
1407         inneriter                  += j_index_end - j_index_start;
1408
1409         /* Outer loop uses 24 flops */
1410     }
1411
1412     /* Increment number of outer iterations */
1413     outeriter        += nri;
1414
1415     /* Update outer/inner flops */
1416
1417     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1418 }