src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 107     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 108     __m256           dummy_mask,cutoff_mask;
 109     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 110     __m256           one     = _mm256_set1_ps(1.0);
 111     __m256           two     = _mm256_set1_ps(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_ps(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm256_set1_ps(fr->ic->k_rf);
 126     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 127     crf              = _mm256_set1_ps(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_ps(facel,_mm256_set1_ps(charge[inr+1]));
 135     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 136     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(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->rcoulomb;
 141     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 142     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 143
 144     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 145     rvdw             = _mm256_set1_ps(fr->rvdw);
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153     j_coord_offsetE = 0;
 154     j_coord_offsetF = 0;
 155     j_coord_offsetG = 0;
 156     j_coord_offsetH = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 183
 184         fix0             = _mm256_setzero_ps();
 185         fiy0             = _mm256_setzero_ps();
 186         fiz0             = _mm256_setzero_ps();
 187         fix1             = _mm256_setzero_ps();
 188         fiy1             = _mm256_setzero_ps();
 189         fiz1             = _mm256_setzero_ps();
 190         fix2             = _mm256_setzero_ps();
 191         fiy2             = _mm256_setzero_ps();
 192         fiz2             = _mm256_setzero_ps();
 193         fix3             = _mm256_setzero_ps();
 194         fiy3             = _mm256_setzero_ps();
 195         fiz3             = _mm256_setzero_ps();
 196
 197         /* Reset potential sums */
 198         velecsum         = _mm256_setzero_ps();
 199         vvdwsum          = _mm256_setzero_ps();
 200
 201         /* Start inner kernel loop */
 202         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 203         {
 204
 205             /* Get j neighbor index, and coordinate index */
 206             jnrA             = jjnr[jidx];
 207             jnrB             = jjnr[jidx+1];
 208             jnrC             = jjnr[jidx+2];
 209             jnrD             = jjnr[jidx+3];
 210             jnrE             = jjnr[jidx+4];
 211             jnrF             = jjnr[jidx+5];
 212             jnrG             = jjnr[jidx+6];
 213             jnrH             = jjnr[jidx+7];
 214             j_coord_offsetA  = DIM*jnrA;
 215             j_coord_offsetB  = DIM*jnrB;
 216             j_coord_offsetC  = DIM*jnrC;
 217             j_coord_offsetD  = DIM*jnrD;
 218             j_coord_offsetE  = DIM*jnrE;
 219             j_coord_offsetF  = DIM*jnrF;
 220             j_coord_offsetG  = DIM*jnrG;
 221             j_coord_offsetH  = DIM*jnrH;
 222
 223             /* load j atom coordinates */
 224             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 225                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 226                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 227                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 228                                                  &jx0,&jy0,&jz0);
 229
 230             /* Calculate displacement vector */
 231             dx00             = _mm256_sub_ps(ix0,jx0);
 232             dy00             = _mm256_sub_ps(iy0,jy0);
 233             dz00             = _mm256_sub_ps(iz0,jz0);
 234             dx10             = _mm256_sub_ps(ix1,jx0);
 235             dy10             = _mm256_sub_ps(iy1,jy0);
 236             dz10             = _mm256_sub_ps(iz1,jz0);
 237             dx20             = _mm256_sub_ps(ix2,jx0);
 238             dy20             = _mm256_sub_ps(iy2,jy0);
 239             dz20             = _mm256_sub_ps(iz2,jz0);
 240             dx30             = _mm256_sub_ps(ix3,jx0);
 241             dy30             = _mm256_sub_ps(iy3,jy0);
 242             dz30             = _mm256_sub_ps(iz3,jz0);
 243
 244             /* Calculate squared distance and things based on it */
 245             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 246             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 247             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 248             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 249
 250             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 251             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 252             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 253
 254             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 255             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 256             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 257             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 258
 259             /* Load parameters for j particles */
 260             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 261                                                                  charge+jnrC+0,charge+jnrD+0,
 262                                                                  charge+jnrE+0,charge+jnrF+0,
 263                                                                  charge+jnrG+0,charge+jnrH+0);
 264             vdwjidx0A        = 2*vdwtype[jnrA+0];
 265             vdwjidx0B        = 2*vdwtype[jnrB+0];
 266             vdwjidx0C        = 2*vdwtype[jnrC+0];
 267             vdwjidx0D        = 2*vdwtype[jnrD+0];
 268             vdwjidx0E        = 2*vdwtype[jnrE+0];
 269             vdwjidx0F        = 2*vdwtype[jnrF+0];
 270             vdwjidx0G        = 2*vdwtype[jnrG+0];
 271             vdwjidx0H        = 2*vdwtype[jnrH+0];
 272
 273             fjx0             = _mm256_setzero_ps();
 274             fjy0             = _mm256_setzero_ps();
 275             fjz0             = _mm256_setzero_ps();
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 282             {
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 286                                             vdwioffsetptr0+vdwjidx0B,
 287                                             vdwioffsetptr0+vdwjidx0C,
 288                                             vdwioffsetptr0+vdwjidx0D,
 289                                             vdwioffsetptr0+vdwjidx0E,
 290                                             vdwioffsetptr0+vdwjidx0F,
 291                                             vdwioffsetptr0+vdwjidx0G,
 292                                             vdwioffsetptr0+vdwjidx0H,
 293                                             &c6_00,&c12_00);
 294
 295             /* LENNARD-JONES DISPERSION/REPULSION */
 296
 297             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 298             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 299             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 300             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 301                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 302             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 303
 304             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 308             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 309
 310             fscal            = fvdw;
 311
 312             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 313
 314             /* Calculate temporary vectorial force */
 315             tx               = _mm256_mul_ps(fscal,dx00);
 316             ty               = _mm256_mul_ps(fscal,dy00);
 317             tz               = _mm256_mul_ps(fscal,dz00);
 318
 319             /* Update vectorial force */
 320             fix0             = _mm256_add_ps(fix0,tx);
 321             fiy0             = _mm256_add_ps(fiy0,ty);
 322             fiz0             = _mm256_add_ps(fiz0,tz);
 323
 324             fjx0             = _mm256_add_ps(fjx0,tx);
 325             fjy0             = _mm256_add_ps(fjy0,ty);
 326             fjz0             = _mm256_add_ps(fjz0,tz);
 327
 328             }
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 335             {
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq10             = _mm256_mul_ps(iq1,jq0);
 339
 340             /* REACTION-FIELD ELECTROSTATICS */
 341             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 342             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 343
 344             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velec            = _mm256_and_ps(velec,cutoff_mask);
 348             velecsum         = _mm256_add_ps(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm256_mul_ps(fscal,dx10);
 356             ty               = _mm256_mul_ps(fscal,dy10);
 357             tz               = _mm256_mul_ps(fscal,dz10);
 358
 359             /* Update vectorial force */
 360             fix1             = _mm256_add_ps(fix1,tx);
 361             fiy1             = _mm256_add_ps(fiy1,ty);
 362             fiz1             = _mm256_add_ps(fiz1,tz);
 363
 364             fjx0             = _mm256_add_ps(fjx0,tx);
 365             fjy0             = _mm256_add_ps(fjy0,ty);
 366             fjz0             = _mm256_add_ps(fjz0,tz);
 367
 368             }
 369
 370             /**************************
 371              * CALCULATE INTERACTIONS *
 372              **************************/
 373
 374             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 375             {
 376
 377             /* Compute parameters for interactions between i and j atoms */
 378             qq20             = _mm256_mul_ps(iq2,jq0);
 379
 380             /* REACTION-FIELD ELECTROSTATICS */
 381             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 382             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 383
 384             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velec            = _mm256_and_ps(velec,cutoff_mask);
 388             velecsum         = _mm256_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm256_mul_ps(fscal,dx20);
 396             ty               = _mm256_mul_ps(fscal,dy20);
 397             tz               = _mm256_mul_ps(fscal,dz20);
 398
 399             /* Update vectorial force */
 400             fix2             = _mm256_add_ps(fix2,tx);
 401             fiy2             = _mm256_add_ps(fiy2,ty);
 402             fiz2             = _mm256_add_ps(fiz2,tz);
 403
 404             fjx0             = _mm256_add_ps(fjx0,tx);
 405             fjy0             = _mm256_add_ps(fjy0,ty);
 406             fjz0             = _mm256_add_ps(fjz0,tz);
 407
 408             }
 409
 410             /**************************
 411              * CALCULATE INTERACTIONS *
 412              **************************/
 413
 414             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 415             {
 416
 417             /* Compute parameters for interactions between i and j atoms */
 418             qq30             = _mm256_mul_ps(iq3,jq0);
 419
 420             /* REACTION-FIELD ELECTROSTATICS */
 421             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 422             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 423
 424             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velec            = _mm256_and_ps(velec,cutoff_mask);
 428             velecsum         = _mm256_add_ps(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 433
 434             /* Calculate temporary vectorial force */
 435             tx               = _mm256_mul_ps(fscal,dx30);
 436             ty               = _mm256_mul_ps(fscal,dy30);
 437             tz               = _mm256_mul_ps(fscal,dz30);
 438
 439             /* Update vectorial force */
 440             fix3             = _mm256_add_ps(fix3,tx);
 441             fiy3             = _mm256_add_ps(fiy3,ty);
 442             fiz3             = _mm256_add_ps(fiz3,tz);
 443
 444             fjx0             = _mm256_add_ps(fjx0,tx);
 445             fjy0             = _mm256_add_ps(fjy0,ty);
 446             fjz0             = _mm256_add_ps(fjz0,tz);
 447
 448             }
 449
 450             fjptrA             = f+j_coord_offsetA;
 451             fjptrB             = f+j_coord_offsetB;
 452             fjptrC             = f+j_coord_offsetC;
 453             fjptrD             = f+j_coord_offsetD;
 454             fjptrE             = f+j_coord_offsetE;
 455             fjptrF             = f+j_coord_offsetF;
 456             fjptrG             = f+j_coord_offsetG;
 457             fjptrH             = f+j_coord_offsetH;
 458
 459             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 460
 461             /* Inner loop uses 152 flops */
 462         }
 463
 464         if(jidx<j_index_end)
 465         {
 466
 467             /* Get j neighbor index, and coordinate index */
 468             jnrlistA         = jjnr[jidx];
 469             jnrlistB         = jjnr[jidx+1];
 470             jnrlistC         = jjnr[jidx+2];
 471             jnrlistD         = jjnr[jidx+3];
 472             jnrlistE         = jjnr[jidx+4];
 473             jnrlistF         = jjnr[jidx+5];
 474             jnrlistG         = jjnr[jidx+6];
 475             jnrlistH         = jjnr[jidx+7];
 476             /* Sign of each element will be negative for non-real atoms.
 477              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 478              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 479              */
 480             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 481                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 482
 483             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 484             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 485             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 486             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 487             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 488             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 489             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 490             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 491             j_coord_offsetA  = DIM*jnrA;
 492             j_coord_offsetB  = DIM*jnrB;
 493             j_coord_offsetC  = DIM*jnrC;
 494             j_coord_offsetD  = DIM*jnrD;
 495             j_coord_offsetE  = DIM*jnrE;
 496             j_coord_offsetF  = DIM*jnrF;
 497             j_coord_offsetG  = DIM*jnrG;
 498             j_coord_offsetH  = DIM*jnrH;
 499
 500             /* load j atom coordinates */
 501             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 502                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 503                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 504                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 505                                                  &jx0,&jy0,&jz0);
 506
 507             /* Calculate displacement vector */
 508             dx00             = _mm256_sub_ps(ix0,jx0);
 509             dy00             = _mm256_sub_ps(iy0,jy0);
 510             dz00             = _mm256_sub_ps(iz0,jz0);
 511             dx10             = _mm256_sub_ps(ix1,jx0);
 512             dy10             = _mm256_sub_ps(iy1,jy0);
 513             dz10             = _mm256_sub_ps(iz1,jz0);
 514             dx20             = _mm256_sub_ps(ix2,jx0);
 515             dy20             = _mm256_sub_ps(iy2,jy0);
 516             dz20             = _mm256_sub_ps(iz2,jz0);
 517             dx30             = _mm256_sub_ps(ix3,jx0);
 518             dy30             = _mm256_sub_ps(iy3,jy0);
 519             dz30             = _mm256_sub_ps(iz3,jz0);
 520
 521             /* Calculate squared distance and things based on it */
 522             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 523             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 524             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 525             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 526
 527             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 528             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 529             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 530
 531             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 532             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 533             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 534             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 535
 536             /* Load parameters for j particles */
 537             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 538                                                                  charge+jnrC+0,charge+jnrD+0,
 539                                                                  charge+jnrE+0,charge+jnrF+0,
 540                                                                  charge+jnrG+0,charge+jnrH+0);
 541             vdwjidx0A        = 2*vdwtype[jnrA+0];
 542             vdwjidx0B        = 2*vdwtype[jnrB+0];
 543             vdwjidx0C        = 2*vdwtype[jnrC+0];
 544             vdwjidx0D        = 2*vdwtype[jnrD+0];
 545             vdwjidx0E        = 2*vdwtype[jnrE+0];
 546             vdwjidx0F        = 2*vdwtype[jnrF+0];
 547             vdwjidx0G        = 2*vdwtype[jnrG+0];
 548             vdwjidx0H        = 2*vdwtype[jnrH+0];
 549
 550             fjx0             = _mm256_setzero_ps();
 551             fjy0             = _mm256_setzero_ps();
 552             fjz0             = _mm256_setzero_ps();
 553
 554             /**************************
 555              * CALCULATE INTERACTIONS *
 556              **************************/
 557
 558             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 559             {
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 563                                             vdwioffsetptr0+vdwjidx0B,
 564                                             vdwioffsetptr0+vdwjidx0C,
 565                                             vdwioffsetptr0+vdwjidx0D,
 566                                             vdwioffsetptr0+vdwjidx0E,
 567                                             vdwioffsetptr0+vdwjidx0F,
 568                                             vdwioffsetptr0+vdwjidx0G,
 569                                             vdwioffsetptr0+vdwjidx0H,
 570                                             &c6_00,&c12_00);
 571
 572             /* LENNARD-JONES DISPERSION/REPULSION */
 573
 574             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 575             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 576             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 577             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 578                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 579             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 580
 581             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 585             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 586             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 587
 588             fscal            = fvdw;
 589
 590             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 591
 592             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm256_mul_ps(fscal,dx00);
 596             ty               = _mm256_mul_ps(fscal,dy00);
 597             tz               = _mm256_mul_ps(fscal,dz00);
 598
 599             /* Update vectorial force */
 600             fix0             = _mm256_add_ps(fix0,tx);
 601             fiy0             = _mm256_add_ps(fiy0,ty);
 602             fiz0             = _mm256_add_ps(fiz0,tz);
 603
 604             fjx0             = _mm256_add_ps(fjx0,tx);
 605             fjy0             = _mm256_add_ps(fjy0,ty);
 606             fjz0             = _mm256_add_ps(fjz0,tz);
 607
 608             }
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 615             {
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq10             = _mm256_mul_ps(iq1,jq0);
 619
 620             /* REACTION-FIELD ELECTROSTATICS */
 621             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 622             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 623
 624             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 625
 626             /* Update potential sum for this i atom from the interaction with this j atom. */
 627             velec            = _mm256_and_ps(velec,cutoff_mask);
 628             velec            = _mm256_andnot_ps(dummy_mask,velec);
 629             velecsum         = _mm256_add_ps(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 634
 635             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 636
 637             /* Calculate temporary vectorial force */
 638             tx               = _mm256_mul_ps(fscal,dx10);
 639             ty               = _mm256_mul_ps(fscal,dy10);
 640             tz               = _mm256_mul_ps(fscal,dz10);
 641
 642             /* Update vectorial force */
 643             fix1             = _mm256_add_ps(fix1,tx);
 644             fiy1             = _mm256_add_ps(fiy1,ty);
 645             fiz1             = _mm256_add_ps(fiz1,tz);
 646
 647             fjx0             = _mm256_add_ps(fjx0,tx);
 648             fjy0             = _mm256_add_ps(fjy0,ty);
 649             fjz0             = _mm256_add_ps(fjz0,tz);
 650
 651             }
 652
 653             /**************************
 654              * CALCULATE INTERACTIONS *
 655              **************************/
 656
 657             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 658             {
 659
 660             /* Compute parameters for interactions between i and j atoms */
 661             qq20             = _mm256_mul_ps(iq2,jq0);
 662
 663             /* REACTION-FIELD ELECTROSTATICS */
 664             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 665             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 666
 667             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 668
 669             /* Update potential sum for this i atom from the interaction with this j atom. */
 670             velec            = _mm256_and_ps(velec,cutoff_mask);
 671             velec            = _mm256_andnot_ps(dummy_mask,velec);
 672             velecsum         = _mm256_add_ps(velecsum,velec);
 673
 674             fscal            = felec;
 675
 676             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 677
 678             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 679
 680             /* Calculate temporary vectorial force */
 681             tx               = _mm256_mul_ps(fscal,dx20);
 682             ty               = _mm256_mul_ps(fscal,dy20);
 683             tz               = _mm256_mul_ps(fscal,dz20);
 684
 685             /* Update vectorial force */
 686             fix2             = _mm256_add_ps(fix2,tx);
 687             fiy2             = _mm256_add_ps(fiy2,ty);
 688             fiz2             = _mm256_add_ps(fiz2,tz);
 689
 690             fjx0             = _mm256_add_ps(fjx0,tx);
 691             fjy0             = _mm256_add_ps(fjy0,ty);
 692             fjz0             = _mm256_add_ps(fjz0,tz);
 693
 694             }
 695
 696             /**************************
 697              * CALCULATE INTERACTIONS *
 698              **************************/
 699
 700             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 701             {
 702
 703             /* Compute parameters for interactions between i and j atoms */
 704             qq30             = _mm256_mul_ps(iq3,jq0);
 705
 706             /* REACTION-FIELD ELECTROSTATICS */
 707             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 708             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 709
 710             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 711
 712             /* Update potential sum for this i atom from the interaction with this j atom. */
 713             velec            = _mm256_and_ps(velec,cutoff_mask);
 714             velec            = _mm256_andnot_ps(dummy_mask,velec);
 715             velecsum         = _mm256_add_ps(velecsum,velec);
 716
 717             fscal            = felec;
 718
 719             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 720
 721             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 722
 723             /* Calculate temporary vectorial force */
 724             tx               = _mm256_mul_ps(fscal,dx30);
 725             ty               = _mm256_mul_ps(fscal,dy30);
 726             tz               = _mm256_mul_ps(fscal,dz30);
 727
 728             /* Update vectorial force */
 729             fix3             = _mm256_add_ps(fix3,tx);
 730             fiy3             = _mm256_add_ps(fiy3,ty);
 731             fiz3             = _mm256_add_ps(fiz3,tz);
 732
 733             fjx0             = _mm256_add_ps(fjx0,tx);
 734             fjy0             = _mm256_add_ps(fjy0,ty);
 735             fjz0             = _mm256_add_ps(fjz0,tz);
 736
 737             }
 738
 739             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 740             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 741             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 742             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 743             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 744             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 745             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 746             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 747
 748             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 749
 750             /* Inner loop uses 152 flops */
 751         }
 752
 753         /* End of innermost loop */
 754
 755         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 756                                                  f+i_coord_offset,fshift+i_shift_offset);
 757
 758         ggid                        = gid[iidx];
 759         /* Update potential energies */
 760         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 761         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 762
 763         /* Increment number of inner iterations */
 764         inneriter                  += j_index_end - j_index_start;
 765
 766         /* Outer loop uses 26 flops */
 767     }
 768
 769     /* Increment number of outer iterations */
 770     outeriter        += nri;
 771
 772     /* Update outer/inner flops */
 773
 774     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*152);
 775 }
 776 /*
 777  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_single
 778  * Electrostatics interaction: ReactionField
 779  * VdW interaction:            LennardJones
 780  * Geometry:                   Water4-Particle
 781  * Calculate force/pot:        Force
 782  */
 783 void
 784 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_single
 785                     (t_nblist                    * gmx_restrict       nlist,
 786                      rvec                        * gmx_restrict          xx,
 787                      rvec                        * gmx_restrict          ff,
 788                      t_forcerec                  * gmx_restrict          fr,
 789                      t_mdatoms                   * gmx_restrict     mdatoms,
 790                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 791                      t_nrnb                      * gmx_restrict        nrnb)
 792 {
 793     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 794      * just 0 for non-waters.
 795      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 796      * jnr indices corresponding to data put in the four positions in the SIMD register.
 797      */
 798     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 799     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 800     int              jnrA,jnrB,jnrC,jnrD;
 801     int              jnrE,jnrF,jnrG,jnrH;
 802     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 803     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 804     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 805     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 806     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 807     real             rcutoff_scalar;
 808     real             *shiftvec,*fshift,*x,*f;
 809     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 810     real             scratch[4*DIM];
 811     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 812     real *           vdwioffsetptr0;
 813     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 814     real *           vdwioffsetptr1;
 815     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 816     real *           vdwioffsetptr2;
 817     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 818     real *           vdwioffsetptr3;
 819     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 820     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 821     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 822     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 823     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 824     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 825     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 826     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 827     real             *charge;
 828     int              nvdwtype;
 829     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 830     int              *vdwtype;
 831     real             *vdwparam;
 832     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 833     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 834     __m256           dummy_mask,cutoff_mask;
 835     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 836     __m256           one     = _mm256_set1_ps(1.0);
 837     __m256           two     = _mm256_set1_ps(2.0);
 838     x                = xx[0];
 839     f                = ff[0];
 840
 841     nri              = nlist->nri;
 842     iinr             = nlist->iinr;
 843     jindex           = nlist->jindex;
 844     jjnr             = nlist->jjnr;
 845     shiftidx         = nlist->shift;
 846     gid              = nlist->gid;
 847     shiftvec         = fr->shift_vec[0];
 848     fshift           = fr->fshift[0];
 849     facel            = _mm256_set1_ps(fr->epsfac);
 850     charge           = mdatoms->chargeA;
 851     krf              = _mm256_set1_ps(fr->ic->k_rf);
 852     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 853     crf              = _mm256_set1_ps(fr->ic->c_rf);
 854     nvdwtype         = fr->ntype;
 855     vdwparam         = fr->nbfp;
 856     vdwtype          = mdatoms->typeA;
 857
 858     /* Setup water-specific parameters */
 859     inr              = nlist->iinr[0];
 860     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 861     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 862     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 863     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 864
 865     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 866     rcutoff_scalar   = fr->rcoulomb;
 867     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 868     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 869
 870     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 871     rvdw             = _mm256_set1_ps(fr->rvdw);
 872
 873     /* Avoid stupid compiler warnings */
 874     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 875     j_coord_offsetA = 0;
 876     j_coord_offsetB = 0;
 877     j_coord_offsetC = 0;
 878     j_coord_offsetD = 0;
 879     j_coord_offsetE = 0;
 880     j_coord_offsetF = 0;
 881     j_coord_offsetG = 0;
 882     j_coord_offsetH = 0;
 883
 884     outeriter        = 0;
 885     inneriter        = 0;
 886
 887     for(iidx=0;iidx<4*DIM;iidx++)
 888     {
 889         scratch[iidx] = 0.0;
 890     }
 891
 892     /* Start outer loop over neighborlists */
 893     for(iidx=0; iidx<nri; iidx++)
 894     {
 895         /* Load shift vector for this list */
 896         i_shift_offset   = DIM*shiftidx[iidx];
 897
 898         /* Load limits for loop over neighbors */
 899         j_index_start    = jindex[iidx];
 900         j_index_end      = jindex[iidx+1];
 901
 902         /* Get outer coordinate index */
 903         inr              = iinr[iidx];
 904         i_coord_offset   = DIM*inr;
 905
 906         /* Load i particle coords and add shift vector */
 907         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 908                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 909
 910         fix0             = _mm256_setzero_ps();
 911         fiy0             = _mm256_setzero_ps();
 912         fiz0             = _mm256_setzero_ps();
 913         fix1             = _mm256_setzero_ps();
 914         fiy1             = _mm256_setzero_ps();
 915         fiz1             = _mm256_setzero_ps();
 916         fix2             = _mm256_setzero_ps();
 917         fiy2             = _mm256_setzero_ps();
 918         fiz2             = _mm256_setzero_ps();
 919         fix3             = _mm256_setzero_ps();
 920         fiy3             = _mm256_setzero_ps();
 921         fiz3             = _mm256_setzero_ps();
 922
 923         /* Start inner kernel loop */
 924         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 925         {
 926
 927             /* Get j neighbor index, and coordinate index */
 928             jnrA             = jjnr[jidx];
 929             jnrB             = jjnr[jidx+1];
 930             jnrC             = jjnr[jidx+2];
 931             jnrD             = jjnr[jidx+3];
 932             jnrE             = jjnr[jidx+4];
 933             jnrF             = jjnr[jidx+5];
 934             jnrG             = jjnr[jidx+6];
 935             jnrH             = jjnr[jidx+7];
 936             j_coord_offsetA  = DIM*jnrA;
 937             j_coord_offsetB  = DIM*jnrB;
 938             j_coord_offsetC  = DIM*jnrC;
 939             j_coord_offsetD  = DIM*jnrD;
 940             j_coord_offsetE  = DIM*jnrE;
 941             j_coord_offsetF  = DIM*jnrF;
 942             j_coord_offsetG  = DIM*jnrG;
 943             j_coord_offsetH  = DIM*jnrH;
 944
 945             /* load j atom coordinates */
 946             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 947                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 948                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 949                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 950                                                  &jx0,&jy0,&jz0);
 951
 952             /* Calculate displacement vector */
 953             dx00             = _mm256_sub_ps(ix0,jx0);
 954             dy00             = _mm256_sub_ps(iy0,jy0);
 955             dz00             = _mm256_sub_ps(iz0,jz0);
 956             dx10             = _mm256_sub_ps(ix1,jx0);
 957             dy10             = _mm256_sub_ps(iy1,jy0);
 958             dz10             = _mm256_sub_ps(iz1,jz0);
 959             dx20             = _mm256_sub_ps(ix2,jx0);
 960             dy20             = _mm256_sub_ps(iy2,jy0);
 961             dz20             = _mm256_sub_ps(iz2,jz0);
 962             dx30             = _mm256_sub_ps(ix3,jx0);
 963             dy30             = _mm256_sub_ps(iy3,jy0);
 964             dz30             = _mm256_sub_ps(iz3,jz0);
 965
 966             /* Calculate squared distance and things based on it */
 967             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 968             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 969             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 970             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 971
 972             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 973             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 974             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 975
 976             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 977             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 978             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 979             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 980
 981             /* Load parameters for j particles */
 982             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 983                                                                  charge+jnrC+0,charge+jnrD+0,
 984                                                                  charge+jnrE+0,charge+jnrF+0,
 985                                                                  charge+jnrG+0,charge+jnrH+0);
 986             vdwjidx0A        = 2*vdwtype[jnrA+0];
 987             vdwjidx0B        = 2*vdwtype[jnrB+0];
 988             vdwjidx0C        = 2*vdwtype[jnrC+0];
 989             vdwjidx0D        = 2*vdwtype[jnrD+0];
 990             vdwjidx0E        = 2*vdwtype[jnrE+0];
 991             vdwjidx0F        = 2*vdwtype[jnrF+0];
 992             vdwjidx0G        = 2*vdwtype[jnrG+0];
 993             vdwjidx0H        = 2*vdwtype[jnrH+0];
 994
 995             fjx0             = _mm256_setzero_ps();
 996             fjy0             = _mm256_setzero_ps();
 997             fjz0             = _mm256_setzero_ps();
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1004             {
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1008                                             vdwioffsetptr0+vdwjidx0B,
1009                                             vdwioffsetptr0+vdwjidx0C,
1010                                             vdwioffsetptr0+vdwjidx0D,
1011                                             vdwioffsetptr0+vdwjidx0E,
1012                                             vdwioffsetptr0+vdwjidx0F,
1013                                             vdwioffsetptr0+vdwjidx0G,
1014                                             vdwioffsetptr0+vdwjidx0H,
1015                                             &c6_00,&c12_00);
1016
1017             /* LENNARD-JONES DISPERSION/REPULSION */
1018
1019             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1020             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1021
1022             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1023
1024             fscal            = fvdw;
1025
1026             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1027
1028             /* Calculate temporary vectorial force */
1029             tx               = _mm256_mul_ps(fscal,dx00);
1030             ty               = _mm256_mul_ps(fscal,dy00);
1031             tz               = _mm256_mul_ps(fscal,dz00);
1032
1033             /* Update vectorial force */
1034             fix0             = _mm256_add_ps(fix0,tx);
1035             fiy0             = _mm256_add_ps(fiy0,ty);
1036             fiz0             = _mm256_add_ps(fiz0,tz);
1037
1038             fjx0             = _mm256_add_ps(fjx0,tx);
1039             fjy0             = _mm256_add_ps(fjy0,ty);
1040             fjz0             = _mm256_add_ps(fjz0,tz);
1041
1042             }
1043
1044             /**************************
1045              * CALCULATE INTERACTIONS *
1046              **************************/
1047
1048             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1049             {
1050
1051             /* Compute parameters for interactions between i and j atoms */
1052             qq10             = _mm256_mul_ps(iq1,jq0);
1053
1054             /* REACTION-FIELD ELECTROSTATICS */
1055             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1056
1057             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1058
1059             fscal            = felec;
1060
1061             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1062
1063             /* Calculate temporary vectorial force */
1064             tx               = _mm256_mul_ps(fscal,dx10);
1065             ty               = _mm256_mul_ps(fscal,dy10);
1066             tz               = _mm256_mul_ps(fscal,dz10);
1067
1068             /* Update vectorial force */
1069             fix1             = _mm256_add_ps(fix1,tx);
1070             fiy1             = _mm256_add_ps(fiy1,ty);
1071             fiz1             = _mm256_add_ps(fiz1,tz);
1072
1073             fjx0             = _mm256_add_ps(fjx0,tx);
1074             fjy0             = _mm256_add_ps(fjy0,ty);
1075             fjz0             = _mm256_add_ps(fjz0,tz);
1076
1077             }
1078
1079             /**************************
1080              * CALCULATE INTERACTIONS *
1081              **************************/
1082
1083             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1084             {
1085
1086             /* Compute parameters for interactions between i and j atoms */
1087             qq20             = _mm256_mul_ps(iq2,jq0);
1088
1089             /* REACTION-FIELD ELECTROSTATICS */
1090             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1091
1092             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1093
1094             fscal            = felec;
1095
1096             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1097
1098             /* Calculate temporary vectorial force */
1099             tx               = _mm256_mul_ps(fscal,dx20);
1100             ty               = _mm256_mul_ps(fscal,dy20);
1101             tz               = _mm256_mul_ps(fscal,dz20);
1102
1103             /* Update vectorial force */
1104             fix2             = _mm256_add_ps(fix2,tx);
1105             fiy2             = _mm256_add_ps(fiy2,ty);
1106             fiz2             = _mm256_add_ps(fiz2,tz);
1107
1108             fjx0             = _mm256_add_ps(fjx0,tx);
1109             fjy0             = _mm256_add_ps(fjy0,ty);
1110             fjz0             = _mm256_add_ps(fjz0,tz);
1111
1112             }
1113
1114             /**************************
1115              * CALCULATE INTERACTIONS *
1116              **************************/
1117
1118             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1119             {
1120
1121             /* Compute parameters for interactions between i and j atoms */
1122             qq30             = _mm256_mul_ps(iq3,jq0);
1123
1124             /* REACTION-FIELD ELECTROSTATICS */
1125             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1126
1127             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1128
1129             fscal            = felec;
1130
1131             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1132
1133             /* Calculate temporary vectorial force */
1134             tx               = _mm256_mul_ps(fscal,dx30);
1135             ty               = _mm256_mul_ps(fscal,dy30);
1136             tz               = _mm256_mul_ps(fscal,dz30);
1137
1138             /* Update vectorial force */
1139             fix3             = _mm256_add_ps(fix3,tx);
1140             fiy3             = _mm256_add_ps(fiy3,ty);
1141             fiz3             = _mm256_add_ps(fiz3,tz);
1142
1143             fjx0             = _mm256_add_ps(fjx0,tx);
1144             fjy0             = _mm256_add_ps(fjy0,ty);
1145             fjz0             = _mm256_add_ps(fjz0,tz);
1146
1147             }
1148
1149             fjptrA             = f+j_coord_offsetA;
1150             fjptrB             = f+j_coord_offsetB;
1151             fjptrC             = f+j_coord_offsetC;
1152             fjptrD             = f+j_coord_offsetD;
1153             fjptrE             = f+j_coord_offsetE;
1154             fjptrF             = f+j_coord_offsetF;
1155             fjptrG             = f+j_coord_offsetG;
1156             fjptrH             = f+j_coord_offsetH;
1157
1158             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1159
1160             /* Inner loop uses 123 flops */
1161         }
1162
1163         if(jidx<j_index_end)
1164         {
1165
1166             /* Get j neighbor index, and coordinate index */
1167             jnrlistA         = jjnr[jidx];
1168             jnrlistB         = jjnr[jidx+1];
1169             jnrlistC         = jjnr[jidx+2];
1170             jnrlistD         = jjnr[jidx+3];
1171             jnrlistE         = jjnr[jidx+4];
1172             jnrlistF         = jjnr[jidx+5];
1173             jnrlistG         = jjnr[jidx+6];
1174             jnrlistH         = jjnr[jidx+7];
1175             /* Sign of each element will be negative for non-real atoms.
1176              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1177              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1178              */
1179             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1180                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1181
1182             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1183             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1184             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1185             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1186             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1187             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1188             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1189             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1190             j_coord_offsetA  = DIM*jnrA;
1191             j_coord_offsetB  = DIM*jnrB;
1192             j_coord_offsetC  = DIM*jnrC;
1193             j_coord_offsetD  = DIM*jnrD;
1194             j_coord_offsetE  = DIM*jnrE;
1195             j_coord_offsetF  = DIM*jnrF;
1196             j_coord_offsetG  = DIM*jnrG;
1197             j_coord_offsetH  = DIM*jnrH;
1198
1199             /* load j atom coordinates */
1200             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1201                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1202                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1203                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1204                                                  &jx0,&jy0,&jz0);
1205
1206             /* Calculate displacement vector */
1207             dx00             = _mm256_sub_ps(ix0,jx0);
1208             dy00             = _mm256_sub_ps(iy0,jy0);
1209             dz00             = _mm256_sub_ps(iz0,jz0);
1210             dx10             = _mm256_sub_ps(ix1,jx0);
1211             dy10             = _mm256_sub_ps(iy1,jy0);
1212             dz10             = _mm256_sub_ps(iz1,jz0);
1213             dx20             = _mm256_sub_ps(ix2,jx0);
1214             dy20             = _mm256_sub_ps(iy2,jy0);
1215             dz20             = _mm256_sub_ps(iz2,jz0);
1216             dx30             = _mm256_sub_ps(ix3,jx0);
1217             dy30             = _mm256_sub_ps(iy3,jy0);
1218             dz30             = _mm256_sub_ps(iz3,jz0);
1219
1220             /* Calculate squared distance and things based on it */
1221             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1222             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1223             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1224             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1225
1226             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1227             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1228             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1229
1230             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1231             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1232             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1233             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1234
1235             /* Load parameters for j particles */
1236             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1237                                                                  charge+jnrC+0,charge+jnrD+0,
1238                                                                  charge+jnrE+0,charge+jnrF+0,
1239                                                                  charge+jnrG+0,charge+jnrH+0);
1240             vdwjidx0A        = 2*vdwtype[jnrA+0];
1241             vdwjidx0B        = 2*vdwtype[jnrB+0];
1242             vdwjidx0C        = 2*vdwtype[jnrC+0];
1243             vdwjidx0D        = 2*vdwtype[jnrD+0];
1244             vdwjidx0E        = 2*vdwtype[jnrE+0];
1245             vdwjidx0F        = 2*vdwtype[jnrF+0];
1246             vdwjidx0G        = 2*vdwtype[jnrG+0];
1247             vdwjidx0H        = 2*vdwtype[jnrH+0];
1248
1249             fjx0             = _mm256_setzero_ps();
1250             fjy0             = _mm256_setzero_ps();
1251             fjz0             = _mm256_setzero_ps();
1252
1253             /**************************
1254              * CALCULATE INTERACTIONS *
1255              **************************/
1256
1257             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1258             {
1259
1260             /* Compute parameters for interactions between i and j atoms */
1261             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1262                                             vdwioffsetptr0+vdwjidx0B,
1263                                             vdwioffsetptr0+vdwjidx0C,
1264                                             vdwioffsetptr0+vdwjidx0D,
1265                                             vdwioffsetptr0+vdwjidx0E,
1266                                             vdwioffsetptr0+vdwjidx0F,
1267                                             vdwioffsetptr0+vdwjidx0G,
1268                                             vdwioffsetptr0+vdwjidx0H,
1269                                             &c6_00,&c12_00);
1270
1271             /* LENNARD-JONES DISPERSION/REPULSION */
1272
1273             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1274             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1275
1276             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1277
1278             fscal            = fvdw;
1279
1280             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1281
1282             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1283
1284             /* Calculate temporary vectorial force */
1285             tx               = _mm256_mul_ps(fscal,dx00);
1286             ty               = _mm256_mul_ps(fscal,dy00);
1287             tz               = _mm256_mul_ps(fscal,dz00);
1288
1289             /* Update vectorial force */
1290             fix0             = _mm256_add_ps(fix0,tx);
1291             fiy0             = _mm256_add_ps(fiy0,ty);
1292             fiz0             = _mm256_add_ps(fiz0,tz);
1293
1294             fjx0             = _mm256_add_ps(fjx0,tx);
1295             fjy0             = _mm256_add_ps(fjy0,ty);
1296             fjz0             = _mm256_add_ps(fjz0,tz);
1297
1298             }
1299
1300             /**************************
1301              * CALCULATE INTERACTIONS *
1302              **************************/
1303
1304             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1305             {
1306
1307             /* Compute parameters for interactions between i and j atoms */
1308             qq10             = _mm256_mul_ps(iq1,jq0);
1309
1310             /* REACTION-FIELD ELECTROSTATICS */
1311             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1312
1313             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1314
1315             fscal            = felec;
1316
1317             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1318
1319             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1320
1321             /* Calculate temporary vectorial force */
1322             tx               = _mm256_mul_ps(fscal,dx10);
1323             ty               = _mm256_mul_ps(fscal,dy10);
1324             tz               = _mm256_mul_ps(fscal,dz10);
1325
1326             /* Update vectorial force */
1327             fix1             = _mm256_add_ps(fix1,tx);
1328             fiy1             = _mm256_add_ps(fiy1,ty);
1329             fiz1             = _mm256_add_ps(fiz1,tz);
1330
1331             fjx0             = _mm256_add_ps(fjx0,tx);
1332             fjy0             = _mm256_add_ps(fjy0,ty);
1333             fjz0             = _mm256_add_ps(fjz0,tz);
1334
1335             }
1336
1337             /**************************
1338              * CALCULATE INTERACTIONS *
1339              **************************/
1340
1341             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1342             {
1343
1344             /* Compute parameters for interactions between i and j atoms */
1345             qq20             = _mm256_mul_ps(iq2,jq0);
1346
1347             /* REACTION-FIELD ELECTROSTATICS */
1348             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1349
1350             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1351
1352             fscal            = felec;
1353
1354             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1355
1356             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1357
1358             /* Calculate temporary vectorial force */
1359             tx               = _mm256_mul_ps(fscal,dx20);
1360             ty               = _mm256_mul_ps(fscal,dy20);
1361             tz               = _mm256_mul_ps(fscal,dz20);
1362
1363             /* Update vectorial force */
1364             fix2             = _mm256_add_ps(fix2,tx);
1365             fiy2             = _mm256_add_ps(fiy2,ty);
1366             fiz2             = _mm256_add_ps(fiz2,tz);
1367
1368             fjx0             = _mm256_add_ps(fjx0,tx);
1369             fjy0             = _mm256_add_ps(fjy0,ty);
1370             fjz0             = _mm256_add_ps(fjz0,tz);
1371
1372             }
1373
1374             /**************************
1375              * CALCULATE INTERACTIONS *
1376              **************************/
1377
1378             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1379             {
1380
1381             /* Compute parameters for interactions between i and j atoms */
1382             qq30             = _mm256_mul_ps(iq3,jq0);
1383
1384             /* REACTION-FIELD ELECTROSTATICS */
1385             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1386
1387             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1388
1389             fscal            = felec;
1390
1391             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1392
1393             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1394
1395             /* Calculate temporary vectorial force */
1396             tx               = _mm256_mul_ps(fscal,dx30);
1397             ty               = _mm256_mul_ps(fscal,dy30);
1398             tz               = _mm256_mul_ps(fscal,dz30);
1399
1400             /* Update vectorial force */
1401             fix3             = _mm256_add_ps(fix3,tx);
1402             fiy3             = _mm256_add_ps(fiy3,ty);
1403             fiz3             = _mm256_add_ps(fiz3,tz);
1404
1405             fjx0             = _mm256_add_ps(fjx0,tx);
1406             fjy0             = _mm256_add_ps(fjy0,ty);
1407             fjz0             = _mm256_add_ps(fjz0,tz);
1408
1409             }
1410
1411             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1412             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1413             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1414             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1415             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1416             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1417             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1418             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1419
1420             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1421
1422             /* Inner loop uses 123 flops */
1423         }
1424
1425         /* End of innermost loop */
1426
1427         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1428                                                  f+i_coord_offset,fshift+i_shift_offset);
1429
1430         /* Increment number of inner iterations */
1431         inneriter                  += j_index_end - j_index_start;
1432
1433         /* Outer loop uses 24 flops */
1434     }
1435
1436     /* Increment number of outer iterations */
1437     outeriter        += nri;
1438
1439     /* Update outer/inner flops */
1440
1441     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
1442 }