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

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     real *           vdwioffsetptr3;
  79     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  80     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  81     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  82     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  84     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  85     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  86     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     int              nvdwtype;
  89     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  93     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  94     __m256           dummy_mask,cutoff_mask;
  95     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  96     __m256           one     = _mm256_set1_ps(1.0);
  97     __m256           two     = _mm256_set1_ps(2.0);
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     facel            = _mm256_set1_ps(fr->epsfac);
 110     charge           = mdatoms->chargeA;
 111     krf              = _mm256_set1_ps(fr->ic->k_rf);
 112     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 113     crf              = _mm256_set1_ps(fr->ic->c_rf);
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117
 118     /* Setup water-specific parameters */
 119     inr              = nlist->iinr[0];
 120     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 121     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 122     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 123     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 124
 125     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 126     rcutoff_scalar   = fr->rcoulomb;
 127     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 128     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 129
 130     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 131     rvdw             = _mm256_set1_ps(fr->rvdw);
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139     j_coord_offsetE = 0;
 140     j_coord_offsetF = 0;
 141     j_coord_offsetG = 0;
 142     j_coord_offsetH = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 169
 170         fix0             = _mm256_setzero_ps();
 171         fiy0             = _mm256_setzero_ps();
 172         fiz0             = _mm256_setzero_ps();
 173         fix1             = _mm256_setzero_ps();
 174         fiy1             = _mm256_setzero_ps();
 175         fiz1             = _mm256_setzero_ps();
 176         fix2             = _mm256_setzero_ps();
 177         fiy2             = _mm256_setzero_ps();
 178         fiz2             = _mm256_setzero_ps();
 179         fix3             = _mm256_setzero_ps();
 180         fiy3             = _mm256_setzero_ps();
 181         fiz3             = _mm256_setzero_ps();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm256_setzero_ps();
 185         vvdwsum          = _mm256_setzero_ps();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196             jnrE             = jjnr[jidx+4];
 197             jnrF             = jjnr[jidx+5];
 198             jnrG             = jjnr[jidx+6];
 199             jnrH             = jjnr[jidx+7];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204             j_coord_offsetE  = DIM*jnrE;
 205             j_coord_offsetF  = DIM*jnrF;
 206             j_coord_offsetG  = DIM*jnrG;
 207             j_coord_offsetH  = DIM*jnrH;
 208
 209             /* load j atom coordinates */
 210             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 212                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 213                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 214                                                  &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm256_sub_ps(ix0,jx0);
 218             dy00             = _mm256_sub_ps(iy0,jy0);
 219             dz00             = _mm256_sub_ps(iz0,jz0);
 220             dx10             = _mm256_sub_ps(ix1,jx0);
 221             dy10             = _mm256_sub_ps(iy1,jy0);
 222             dz10             = _mm256_sub_ps(iz1,jz0);
 223             dx20             = _mm256_sub_ps(ix2,jx0);
 224             dy20             = _mm256_sub_ps(iy2,jy0);
 225             dz20             = _mm256_sub_ps(iz2,jz0);
 226             dx30             = _mm256_sub_ps(ix3,jx0);
 227             dy30             = _mm256_sub_ps(iy3,jy0);
 228             dz30             = _mm256_sub_ps(iz3,jz0);
 229
 230             /* Calculate squared distance and things based on it */
 231             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 232             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 233             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 234             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 235
 236             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 237             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 238             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 239
 240             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 241             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 242             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 243             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 247                                                                  charge+jnrC+0,charge+jnrD+0,
 248                                                                  charge+jnrE+0,charge+jnrF+0,
 249                                                                  charge+jnrG+0,charge+jnrH+0);
 250             vdwjidx0A        = 2*vdwtype[jnrA+0];
 251             vdwjidx0B        = 2*vdwtype[jnrB+0];
 252             vdwjidx0C        = 2*vdwtype[jnrC+0];
 253             vdwjidx0D        = 2*vdwtype[jnrD+0];
 254             vdwjidx0E        = 2*vdwtype[jnrE+0];
 255             vdwjidx0F        = 2*vdwtype[jnrF+0];
 256             vdwjidx0G        = 2*vdwtype[jnrG+0];
 257             vdwjidx0H        = 2*vdwtype[jnrH+0];
 258
 259             fjx0             = _mm256_setzero_ps();
 260             fjy0             = _mm256_setzero_ps();
 261             fjz0             = _mm256_setzero_ps();
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 268             {
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 272                                             vdwioffsetptr0+vdwjidx0B,
 273                                             vdwioffsetptr0+vdwjidx0C,
 274                                             vdwioffsetptr0+vdwjidx0D,
 275                                             vdwioffsetptr0+vdwjidx0E,
 276                                             vdwioffsetptr0+vdwjidx0F,
 277                                             vdwioffsetptr0+vdwjidx0G,
 278                                             vdwioffsetptr0+vdwjidx0H,
 279                                             &c6_00,&c12_00);
 280
 281             /* LENNARD-JONES DISPERSION/REPULSION */
 282
 283             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 284             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 285             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 286             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) ,
 287                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 288             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 289
 290             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 294             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 295
 296             fscal            = fvdw;
 297
 298             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm256_mul_ps(fscal,dx00);
 302             ty               = _mm256_mul_ps(fscal,dy00);
 303             tz               = _mm256_mul_ps(fscal,dz00);
 304
 305             /* Update vectorial force */
 306             fix0             = _mm256_add_ps(fix0,tx);
 307             fiy0             = _mm256_add_ps(fiy0,ty);
 308             fiz0             = _mm256_add_ps(fiz0,tz);
 309
 310             fjx0             = _mm256_add_ps(fjx0,tx);
 311             fjy0             = _mm256_add_ps(fjy0,ty);
 312             fjz0             = _mm256_add_ps(fjz0,tz);
 313
 314             }
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 321             {
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm256_mul_ps(iq1,jq0);
 325
 326             /* REACTION-FIELD ELECTROSTATICS */
 327             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 328             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 329
 330             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm256_and_ps(velec,cutoff_mask);
 334             velecsum         = _mm256_add_ps(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 339
 340             /* Calculate temporary vectorial force */
 341             tx               = _mm256_mul_ps(fscal,dx10);
 342             ty               = _mm256_mul_ps(fscal,dy10);
 343             tz               = _mm256_mul_ps(fscal,dz10);
 344
 345             /* Update vectorial force */
 346             fix1             = _mm256_add_ps(fix1,tx);
 347             fiy1             = _mm256_add_ps(fiy1,ty);
 348             fiz1             = _mm256_add_ps(fiz1,tz);
 349
 350             fjx0             = _mm256_add_ps(fjx0,tx);
 351             fjy0             = _mm256_add_ps(fjy0,ty);
 352             fjz0             = _mm256_add_ps(fjz0,tz);
 353
 354             }
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 361             {
 362
 363             /* Compute parameters for interactions between i and j atoms */
 364             qq20             = _mm256_mul_ps(iq2,jq0);
 365
 366             /* REACTION-FIELD ELECTROSTATICS */
 367             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 368             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 369
 370             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velec            = _mm256_and_ps(velec,cutoff_mask);
 374             velecsum         = _mm256_add_ps(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm256_mul_ps(fscal,dx20);
 382             ty               = _mm256_mul_ps(fscal,dy20);
 383             tz               = _mm256_mul_ps(fscal,dz20);
 384
 385             /* Update vectorial force */
 386             fix2             = _mm256_add_ps(fix2,tx);
 387             fiy2             = _mm256_add_ps(fiy2,ty);
 388             fiz2             = _mm256_add_ps(fiz2,tz);
 389
 390             fjx0             = _mm256_add_ps(fjx0,tx);
 391             fjy0             = _mm256_add_ps(fjy0,ty);
 392             fjz0             = _mm256_add_ps(fjz0,tz);
 393
 394             }
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 401             {
 402
 403             /* Compute parameters for interactions between i and j atoms */
 404             qq30             = _mm256_mul_ps(iq3,jq0);
 405
 406             /* REACTION-FIELD ELECTROSTATICS */
 407             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 408             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 409
 410             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 411
 412             /* Update potential sum for this i atom from the interaction with this j atom. */
 413             velec            = _mm256_and_ps(velec,cutoff_mask);
 414             velecsum         = _mm256_add_ps(velecsum,velec);
 415
 416             fscal            = felec;
 417
 418             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 419
 420             /* Calculate temporary vectorial force */
 421             tx               = _mm256_mul_ps(fscal,dx30);
 422             ty               = _mm256_mul_ps(fscal,dy30);
 423             tz               = _mm256_mul_ps(fscal,dz30);
 424
 425             /* Update vectorial force */
 426             fix3             = _mm256_add_ps(fix3,tx);
 427             fiy3             = _mm256_add_ps(fiy3,ty);
 428             fiz3             = _mm256_add_ps(fiz3,tz);
 429
 430             fjx0             = _mm256_add_ps(fjx0,tx);
 431             fjy0             = _mm256_add_ps(fjy0,ty);
 432             fjz0             = _mm256_add_ps(fjz0,tz);
 433
 434             }
 435
 436             fjptrA             = f+j_coord_offsetA;
 437             fjptrB             = f+j_coord_offsetB;
 438             fjptrC             = f+j_coord_offsetC;
 439             fjptrD             = f+j_coord_offsetD;
 440             fjptrE             = f+j_coord_offsetE;
 441             fjptrF             = f+j_coord_offsetF;
 442             fjptrG             = f+j_coord_offsetG;
 443             fjptrH             = f+j_coord_offsetH;
 444
 445             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 446
 447             /* Inner loop uses 152 flops */
 448         }
 449
 450         if(jidx<j_index_end)
 451         {
 452
 453             /* Get j neighbor index, and coordinate index */
 454             jnrlistA         = jjnr[jidx];
 455             jnrlistB         = jjnr[jidx+1];
 456             jnrlistC         = jjnr[jidx+2];
 457             jnrlistD         = jjnr[jidx+3];
 458             jnrlistE         = jjnr[jidx+4];
 459             jnrlistF         = jjnr[jidx+5];
 460             jnrlistG         = jjnr[jidx+6];
 461             jnrlistH         = jjnr[jidx+7];
 462             /* Sign of each element will be negative for non-real atoms.
 463              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 464              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 465              */
 466             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 467                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 468
 469             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 470             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 471             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 472             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 473             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 474             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 475             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 476             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 477             j_coord_offsetA  = DIM*jnrA;
 478             j_coord_offsetB  = DIM*jnrB;
 479             j_coord_offsetC  = DIM*jnrC;
 480             j_coord_offsetD  = DIM*jnrD;
 481             j_coord_offsetE  = DIM*jnrE;
 482             j_coord_offsetF  = DIM*jnrF;
 483             j_coord_offsetG  = DIM*jnrG;
 484             j_coord_offsetH  = DIM*jnrH;
 485
 486             /* load j atom coordinates */
 487             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 488                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 489                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 490                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 491                                                  &jx0,&jy0,&jz0);
 492
 493             /* Calculate displacement vector */
 494             dx00             = _mm256_sub_ps(ix0,jx0);
 495             dy00             = _mm256_sub_ps(iy0,jy0);
 496             dz00             = _mm256_sub_ps(iz0,jz0);
 497             dx10             = _mm256_sub_ps(ix1,jx0);
 498             dy10             = _mm256_sub_ps(iy1,jy0);
 499             dz10             = _mm256_sub_ps(iz1,jz0);
 500             dx20             = _mm256_sub_ps(ix2,jx0);
 501             dy20             = _mm256_sub_ps(iy2,jy0);
 502             dz20             = _mm256_sub_ps(iz2,jz0);
 503             dx30             = _mm256_sub_ps(ix3,jx0);
 504             dy30             = _mm256_sub_ps(iy3,jy0);
 505             dz30             = _mm256_sub_ps(iz3,jz0);
 506
 507             /* Calculate squared distance and things based on it */
 508             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 509             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 510             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 511             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 512
 513             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 514             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 515             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 516
 517             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 518             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 519             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 520             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 521
 522             /* Load parameters for j particles */
 523             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 524                                                                  charge+jnrC+0,charge+jnrD+0,
 525                                                                  charge+jnrE+0,charge+jnrF+0,
 526                                                                  charge+jnrG+0,charge+jnrH+0);
 527             vdwjidx0A        = 2*vdwtype[jnrA+0];
 528             vdwjidx0B        = 2*vdwtype[jnrB+0];
 529             vdwjidx0C        = 2*vdwtype[jnrC+0];
 530             vdwjidx0D        = 2*vdwtype[jnrD+0];
 531             vdwjidx0E        = 2*vdwtype[jnrE+0];
 532             vdwjidx0F        = 2*vdwtype[jnrF+0];
 533             vdwjidx0G        = 2*vdwtype[jnrG+0];
 534             vdwjidx0H        = 2*vdwtype[jnrH+0];
 535
 536             fjx0             = _mm256_setzero_ps();
 537             fjy0             = _mm256_setzero_ps();
 538             fjz0             = _mm256_setzero_ps();
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 545             {
 546
 547             /* Compute parameters for interactions between i and j atoms */
 548             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 549                                             vdwioffsetptr0+vdwjidx0B,
 550                                             vdwioffsetptr0+vdwjidx0C,
 551                                             vdwioffsetptr0+vdwjidx0D,
 552                                             vdwioffsetptr0+vdwjidx0E,
 553                                             vdwioffsetptr0+vdwjidx0F,
 554                                             vdwioffsetptr0+vdwjidx0G,
 555                                             vdwioffsetptr0+vdwjidx0H,
 556                                             &c6_00,&c12_00);
 557
 558             /* LENNARD-JONES DISPERSION/REPULSION */
 559
 560             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 561             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 562             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 563             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) ,
 564                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 565             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 566
 567             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 568
 569             /* Update potential sum for this i atom from the interaction with this j atom. */
 570             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 571             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 572             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 573
 574             fscal            = fvdw;
 575
 576             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 577
 578             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm256_mul_ps(fscal,dx00);
 582             ty               = _mm256_mul_ps(fscal,dy00);
 583             tz               = _mm256_mul_ps(fscal,dz00);
 584
 585             /* Update vectorial force */
 586             fix0             = _mm256_add_ps(fix0,tx);
 587             fiy0             = _mm256_add_ps(fiy0,ty);
 588             fiz0             = _mm256_add_ps(fiz0,tz);
 589
 590             fjx0             = _mm256_add_ps(fjx0,tx);
 591             fjy0             = _mm256_add_ps(fjy0,ty);
 592             fjz0             = _mm256_add_ps(fjz0,tz);
 593
 594             }
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 601             {
 602
 603             /* Compute parameters for interactions between i and j atoms */
 604             qq10             = _mm256_mul_ps(iq1,jq0);
 605
 606             /* REACTION-FIELD ELECTROSTATICS */
 607             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 608             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 609
 610             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velec            = _mm256_and_ps(velec,cutoff_mask);
 614             velec            = _mm256_andnot_ps(dummy_mask,velec);
 615             velecsum         = _mm256_add_ps(velecsum,velec);
 616
 617             fscal            = felec;
 618
 619             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 620
 621             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm256_mul_ps(fscal,dx10);
 625             ty               = _mm256_mul_ps(fscal,dy10);
 626             tz               = _mm256_mul_ps(fscal,dz10);
 627
 628             /* Update vectorial force */
 629             fix1             = _mm256_add_ps(fix1,tx);
 630             fiy1             = _mm256_add_ps(fiy1,ty);
 631             fiz1             = _mm256_add_ps(fiz1,tz);
 632
 633             fjx0             = _mm256_add_ps(fjx0,tx);
 634             fjy0             = _mm256_add_ps(fjy0,ty);
 635             fjz0             = _mm256_add_ps(fjz0,tz);
 636
 637             }
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 644             {
 645
 646             /* Compute parameters for interactions between i and j atoms */
 647             qq20             = _mm256_mul_ps(iq2,jq0);
 648
 649             /* REACTION-FIELD ELECTROSTATICS */
 650             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 651             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 652
 653             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velec            = _mm256_and_ps(velec,cutoff_mask);
 657             velec            = _mm256_andnot_ps(dummy_mask,velec);
 658             velecsum         = _mm256_add_ps(velecsum,velec);
 659
 660             fscal            = felec;
 661
 662             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 663
 664             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 665
 666             /* Calculate temporary vectorial force */
 667             tx               = _mm256_mul_ps(fscal,dx20);
 668             ty               = _mm256_mul_ps(fscal,dy20);
 669             tz               = _mm256_mul_ps(fscal,dz20);
 670
 671             /* Update vectorial force */
 672             fix2             = _mm256_add_ps(fix2,tx);
 673             fiy2             = _mm256_add_ps(fiy2,ty);
 674             fiz2             = _mm256_add_ps(fiz2,tz);
 675
 676             fjx0             = _mm256_add_ps(fjx0,tx);
 677             fjy0             = _mm256_add_ps(fjy0,ty);
 678             fjz0             = _mm256_add_ps(fjz0,tz);
 679
 680             }
 681
 682             /**************************
 683              * CALCULATE INTERACTIONS *
 684              **************************/
 685
 686             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 687             {
 688
 689             /* Compute parameters for interactions between i and j atoms */
 690             qq30             = _mm256_mul_ps(iq3,jq0);
 691
 692             /* REACTION-FIELD ELECTROSTATICS */
 693             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 694             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 695
 696             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 697
 698             /* Update potential sum for this i atom from the interaction with this j atom. */
 699             velec            = _mm256_and_ps(velec,cutoff_mask);
 700             velec            = _mm256_andnot_ps(dummy_mask,velec);
 701             velecsum         = _mm256_add_ps(velecsum,velec);
 702
 703             fscal            = felec;
 704
 705             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 706
 707             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 708
 709             /* Calculate temporary vectorial force */
 710             tx               = _mm256_mul_ps(fscal,dx30);
 711             ty               = _mm256_mul_ps(fscal,dy30);
 712             tz               = _mm256_mul_ps(fscal,dz30);
 713
 714             /* Update vectorial force */
 715             fix3             = _mm256_add_ps(fix3,tx);
 716             fiy3             = _mm256_add_ps(fiy3,ty);
 717             fiz3             = _mm256_add_ps(fiz3,tz);
 718
 719             fjx0             = _mm256_add_ps(fjx0,tx);
 720             fjy0             = _mm256_add_ps(fjy0,ty);
 721             fjz0             = _mm256_add_ps(fjz0,tz);
 722
 723             }
 724
 725             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 726             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 727             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 728             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 729             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 730             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 731             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 732             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 733
 734             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 735
 736             /* Inner loop uses 152 flops */
 737         }
 738
 739         /* End of innermost loop */
 740
 741         gmx_mm256_update_iforce_4atom_swizzle_ps(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_ps(velecsum,kernel_data->energygrp_elec+ggid);
 747         gmx_mm256_update_1pot_ps(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*152);
 761 }
 762 /*
 763  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_single
 764  * Electrostatics interaction: ReactionField
 765  * VdW interaction:            LennardJones
 766  * Geometry:                   Water4-Particle
 767  * Calculate force/pot:        Force
 768  */
 769 void
 770 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_single
 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_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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
 788     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 789     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 790     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 791     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 792     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 793     real             rcutoff_scalar;
 794     real             *shiftvec,*fshift,*x,*f;
 795     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 796     real             scratch[4*DIM];
 797     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 798     real *           vdwioffsetptr0;
 799     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 800     real *           vdwioffsetptr1;
 801     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 802     real *           vdwioffsetptr2;
 803     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 804     real *           vdwioffsetptr3;
 805     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 806     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 807     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 808     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 809     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 810     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 811     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 812     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 813     real             *charge;
 814     int              nvdwtype;
 815     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 816     int              *vdwtype;
 817     real             *vdwparam;
 818     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 819     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 820     __m256           dummy_mask,cutoff_mask;
 821     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 822     __m256           one     = _mm256_set1_ps(1.0);
 823     __m256           two     = _mm256_set1_ps(2.0);
 824     x                = xx[0];
 825     f                = ff[0];
 826
 827     nri              = nlist->nri;
 828     iinr             = nlist->iinr;
 829     jindex           = nlist->jindex;
 830     jjnr             = nlist->jjnr;
 831     shiftidx         = nlist->shift;
 832     gid              = nlist->gid;
 833     shiftvec         = fr->shift_vec[0];
 834     fshift           = fr->fshift[0];
 835     facel            = _mm256_set1_ps(fr->epsfac);
 836     charge           = mdatoms->chargeA;
 837     krf              = _mm256_set1_ps(fr->ic->k_rf);
 838     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 839     crf              = _mm256_set1_ps(fr->ic->c_rf);
 840     nvdwtype         = fr->ntype;
 841     vdwparam         = fr->nbfp;
 842     vdwtype          = mdatoms->typeA;
 843
 844     /* Setup water-specific parameters */
 845     inr              = nlist->iinr[0];
 846     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 847     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 848     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 849     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 850
 851     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 852     rcutoff_scalar   = fr->rcoulomb;
 853     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 854     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 855
 856     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 857     rvdw             = _mm256_set1_ps(fr->rvdw);
 858
 859     /* Avoid stupid compiler warnings */
 860     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 861     j_coord_offsetA = 0;
 862     j_coord_offsetB = 0;
 863     j_coord_offsetC = 0;
 864     j_coord_offsetD = 0;
 865     j_coord_offsetE = 0;
 866     j_coord_offsetF = 0;
 867     j_coord_offsetG = 0;
 868     j_coord_offsetH = 0;
 869
 870     outeriter        = 0;
 871     inneriter        = 0;
 872
 873     for(iidx=0;iidx<4*DIM;iidx++)
 874     {
 875         scratch[iidx] = 0.0;
 876     }
 877
 878     /* Start outer loop over neighborlists */
 879     for(iidx=0; iidx<nri; iidx++)
 880     {
 881         /* Load shift vector for this list */
 882         i_shift_offset   = DIM*shiftidx[iidx];
 883
 884         /* Load limits for loop over neighbors */
 885         j_index_start    = jindex[iidx];
 886         j_index_end      = jindex[iidx+1];
 887
 888         /* Get outer coordinate index */
 889         inr              = iinr[iidx];
 890         i_coord_offset   = DIM*inr;
 891
 892         /* Load i particle coords and add shift vector */
 893         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 894                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 895
 896         fix0             = _mm256_setzero_ps();
 897         fiy0             = _mm256_setzero_ps();
 898         fiz0             = _mm256_setzero_ps();
 899         fix1             = _mm256_setzero_ps();
 900         fiy1             = _mm256_setzero_ps();
 901         fiz1             = _mm256_setzero_ps();
 902         fix2             = _mm256_setzero_ps();
 903         fiy2             = _mm256_setzero_ps();
 904         fiz2             = _mm256_setzero_ps();
 905         fix3             = _mm256_setzero_ps();
 906         fiy3             = _mm256_setzero_ps();
 907         fiz3             = _mm256_setzero_ps();
 908
 909         /* Start inner kernel loop */
 910         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 911         {
 912
 913             /* Get j neighbor index, and coordinate index */
 914             jnrA             = jjnr[jidx];
 915             jnrB             = jjnr[jidx+1];
 916             jnrC             = jjnr[jidx+2];
 917             jnrD             = jjnr[jidx+3];
 918             jnrE             = jjnr[jidx+4];
 919             jnrF             = jjnr[jidx+5];
 920             jnrG             = jjnr[jidx+6];
 921             jnrH             = jjnr[jidx+7];
 922             j_coord_offsetA  = DIM*jnrA;
 923             j_coord_offsetB  = DIM*jnrB;
 924             j_coord_offsetC  = DIM*jnrC;
 925             j_coord_offsetD  = DIM*jnrD;
 926             j_coord_offsetE  = DIM*jnrE;
 927             j_coord_offsetF  = DIM*jnrF;
 928             j_coord_offsetG  = DIM*jnrG;
 929             j_coord_offsetH  = DIM*jnrH;
 930
 931             /* load j atom coordinates */
 932             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 933                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 934                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 935                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 936                                                  &jx0,&jy0,&jz0);
 937
 938             /* Calculate displacement vector */
 939             dx00             = _mm256_sub_ps(ix0,jx0);
 940             dy00             = _mm256_sub_ps(iy0,jy0);
 941             dz00             = _mm256_sub_ps(iz0,jz0);
 942             dx10             = _mm256_sub_ps(ix1,jx0);
 943             dy10             = _mm256_sub_ps(iy1,jy0);
 944             dz10             = _mm256_sub_ps(iz1,jz0);
 945             dx20             = _mm256_sub_ps(ix2,jx0);
 946             dy20             = _mm256_sub_ps(iy2,jy0);
 947             dz20             = _mm256_sub_ps(iz2,jz0);
 948             dx30             = _mm256_sub_ps(ix3,jx0);
 949             dy30             = _mm256_sub_ps(iy3,jy0);
 950             dz30             = _mm256_sub_ps(iz3,jz0);
 951
 952             /* Calculate squared distance and things based on it */
 953             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 954             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 955             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 956             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 957
 958             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 959             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 960             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 961
 962             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 963             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 964             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 965             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 966
 967             /* Load parameters for j particles */
 968             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 969                                                                  charge+jnrC+0,charge+jnrD+0,
 970                                                                  charge+jnrE+0,charge+jnrF+0,
 971                                                                  charge+jnrG+0,charge+jnrH+0);
 972             vdwjidx0A        = 2*vdwtype[jnrA+0];
 973             vdwjidx0B        = 2*vdwtype[jnrB+0];
 974             vdwjidx0C        = 2*vdwtype[jnrC+0];
 975             vdwjidx0D        = 2*vdwtype[jnrD+0];
 976             vdwjidx0E        = 2*vdwtype[jnrE+0];
 977             vdwjidx0F        = 2*vdwtype[jnrF+0];
 978             vdwjidx0G        = 2*vdwtype[jnrG+0];
 979             vdwjidx0H        = 2*vdwtype[jnrH+0];
 980
 981             fjx0             = _mm256_setzero_ps();
 982             fjy0             = _mm256_setzero_ps();
 983             fjz0             = _mm256_setzero_ps();
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 990             {
 991
 992             /* Compute parameters for interactions between i and j atoms */
 993             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 994                                             vdwioffsetptr0+vdwjidx0B,
 995                                             vdwioffsetptr0+vdwjidx0C,
 996                                             vdwioffsetptr0+vdwjidx0D,
 997                                             vdwioffsetptr0+vdwjidx0E,
 998                                             vdwioffsetptr0+vdwjidx0F,
 999                                             vdwioffsetptr0+vdwjidx0G,
1000                                             vdwioffsetptr0+vdwjidx0H,
1001                                             &c6_00,&c12_00);
1002
1003             /* LENNARD-JONES DISPERSION/REPULSION */
1004
1005             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1006             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1007
1008             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1009
1010             fscal            = fvdw;
1011
1012             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm256_mul_ps(fscal,dx00);
1016             ty               = _mm256_mul_ps(fscal,dy00);
1017             tz               = _mm256_mul_ps(fscal,dz00);
1018
1019             /* Update vectorial force */
1020             fix0             = _mm256_add_ps(fix0,tx);
1021             fiy0             = _mm256_add_ps(fiy0,ty);
1022             fiz0             = _mm256_add_ps(fiz0,tz);
1023
1024             fjx0             = _mm256_add_ps(fjx0,tx);
1025             fjy0             = _mm256_add_ps(fjy0,ty);
1026             fjz0             = _mm256_add_ps(fjz0,tz);
1027
1028             }
1029
1030             /**************************
1031              * CALCULATE INTERACTIONS *
1032              **************************/
1033
1034             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1035             {
1036
1037             /* Compute parameters for interactions between i and j atoms */
1038             qq10             = _mm256_mul_ps(iq1,jq0);
1039
1040             /* REACTION-FIELD ELECTROSTATICS */
1041             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1042
1043             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1044
1045             fscal            = felec;
1046
1047             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1048
1049             /* Calculate temporary vectorial force */
1050             tx               = _mm256_mul_ps(fscal,dx10);
1051             ty               = _mm256_mul_ps(fscal,dy10);
1052             tz               = _mm256_mul_ps(fscal,dz10);
1053
1054             /* Update vectorial force */
1055             fix1             = _mm256_add_ps(fix1,tx);
1056             fiy1             = _mm256_add_ps(fiy1,ty);
1057             fiz1             = _mm256_add_ps(fiz1,tz);
1058
1059             fjx0             = _mm256_add_ps(fjx0,tx);
1060             fjy0             = _mm256_add_ps(fjy0,ty);
1061             fjz0             = _mm256_add_ps(fjz0,tz);
1062
1063             }
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1070             {
1071
1072             /* Compute parameters for interactions between i and j atoms */
1073             qq20             = _mm256_mul_ps(iq2,jq0);
1074
1075             /* REACTION-FIELD ELECTROSTATICS */
1076             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1077
1078             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1079
1080             fscal            = felec;
1081
1082             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm256_mul_ps(fscal,dx20);
1086             ty               = _mm256_mul_ps(fscal,dy20);
1087             tz               = _mm256_mul_ps(fscal,dz20);
1088
1089             /* Update vectorial force */
1090             fix2             = _mm256_add_ps(fix2,tx);
1091             fiy2             = _mm256_add_ps(fiy2,ty);
1092             fiz2             = _mm256_add_ps(fiz2,tz);
1093
1094             fjx0             = _mm256_add_ps(fjx0,tx);
1095             fjy0             = _mm256_add_ps(fjy0,ty);
1096             fjz0             = _mm256_add_ps(fjz0,tz);
1097
1098             }
1099
1100             /**************************
1101              * CALCULATE INTERACTIONS *
1102              **************************/
1103
1104             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1105             {
1106
1107             /* Compute parameters for interactions between i and j atoms */
1108             qq30             = _mm256_mul_ps(iq3,jq0);
1109
1110             /* REACTION-FIELD ELECTROSTATICS */
1111             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1112
1113             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1114
1115             fscal            = felec;
1116
1117             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1118
1119             /* Calculate temporary vectorial force */
1120             tx               = _mm256_mul_ps(fscal,dx30);
1121             ty               = _mm256_mul_ps(fscal,dy30);
1122             tz               = _mm256_mul_ps(fscal,dz30);
1123
1124             /* Update vectorial force */
1125             fix3             = _mm256_add_ps(fix3,tx);
1126             fiy3             = _mm256_add_ps(fiy3,ty);
1127             fiz3             = _mm256_add_ps(fiz3,tz);
1128
1129             fjx0             = _mm256_add_ps(fjx0,tx);
1130             fjy0             = _mm256_add_ps(fjy0,ty);
1131             fjz0             = _mm256_add_ps(fjz0,tz);
1132
1133             }
1134
1135             fjptrA             = f+j_coord_offsetA;
1136             fjptrB             = f+j_coord_offsetB;
1137             fjptrC             = f+j_coord_offsetC;
1138             fjptrD             = f+j_coord_offsetD;
1139             fjptrE             = f+j_coord_offsetE;
1140             fjptrF             = f+j_coord_offsetF;
1141             fjptrG             = f+j_coord_offsetG;
1142             fjptrH             = f+j_coord_offsetH;
1143
1144             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1145
1146             /* Inner loop uses 123 flops */
1147         }
1148
1149         if(jidx<j_index_end)
1150         {
1151
1152             /* Get j neighbor index, and coordinate index */
1153             jnrlistA         = jjnr[jidx];
1154             jnrlistB         = jjnr[jidx+1];
1155             jnrlistC         = jjnr[jidx+2];
1156             jnrlistD         = jjnr[jidx+3];
1157             jnrlistE         = jjnr[jidx+4];
1158             jnrlistF         = jjnr[jidx+5];
1159             jnrlistG         = jjnr[jidx+6];
1160             jnrlistH         = jjnr[jidx+7];
1161             /* Sign of each element will be negative for non-real atoms.
1162              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1163              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1164              */
1165             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1166                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1167
1168             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1169             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1170             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1171             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1172             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1173             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1174             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1175             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1176             j_coord_offsetA  = DIM*jnrA;
1177             j_coord_offsetB  = DIM*jnrB;
1178             j_coord_offsetC  = DIM*jnrC;
1179             j_coord_offsetD  = DIM*jnrD;
1180             j_coord_offsetE  = DIM*jnrE;
1181             j_coord_offsetF  = DIM*jnrF;
1182             j_coord_offsetG  = DIM*jnrG;
1183             j_coord_offsetH  = DIM*jnrH;
1184
1185             /* load j atom coordinates */
1186             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1187                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1188                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1189                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1190                                                  &jx0,&jy0,&jz0);
1191
1192             /* Calculate displacement vector */
1193             dx00             = _mm256_sub_ps(ix0,jx0);
1194             dy00             = _mm256_sub_ps(iy0,jy0);
1195             dz00             = _mm256_sub_ps(iz0,jz0);
1196             dx10             = _mm256_sub_ps(ix1,jx0);
1197             dy10             = _mm256_sub_ps(iy1,jy0);
1198             dz10             = _mm256_sub_ps(iz1,jz0);
1199             dx20             = _mm256_sub_ps(ix2,jx0);
1200             dy20             = _mm256_sub_ps(iy2,jy0);
1201             dz20             = _mm256_sub_ps(iz2,jz0);
1202             dx30             = _mm256_sub_ps(ix3,jx0);
1203             dy30             = _mm256_sub_ps(iy3,jy0);
1204             dz30             = _mm256_sub_ps(iz3,jz0);
1205
1206             /* Calculate squared distance and things based on it */
1207             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1208             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1209             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1210             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1211
1212             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1213             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1214             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1215
1216             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1217             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1218             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1219             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1220
1221             /* Load parameters for j particles */
1222             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1223                                                                  charge+jnrC+0,charge+jnrD+0,
1224                                                                  charge+jnrE+0,charge+jnrF+0,
1225                                                                  charge+jnrG+0,charge+jnrH+0);
1226             vdwjidx0A        = 2*vdwtype[jnrA+0];
1227             vdwjidx0B        = 2*vdwtype[jnrB+0];
1228             vdwjidx0C        = 2*vdwtype[jnrC+0];
1229             vdwjidx0D        = 2*vdwtype[jnrD+0];
1230             vdwjidx0E        = 2*vdwtype[jnrE+0];
1231             vdwjidx0F        = 2*vdwtype[jnrF+0];
1232             vdwjidx0G        = 2*vdwtype[jnrG+0];
1233             vdwjidx0H        = 2*vdwtype[jnrH+0];
1234
1235             fjx0             = _mm256_setzero_ps();
1236             fjy0             = _mm256_setzero_ps();
1237             fjz0             = _mm256_setzero_ps();
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1244             {
1245
1246             /* Compute parameters for interactions between i and j atoms */
1247             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1248                                             vdwioffsetptr0+vdwjidx0B,
1249                                             vdwioffsetptr0+vdwjidx0C,
1250                                             vdwioffsetptr0+vdwjidx0D,
1251                                             vdwioffsetptr0+vdwjidx0E,
1252                                             vdwioffsetptr0+vdwjidx0F,
1253                                             vdwioffsetptr0+vdwjidx0G,
1254                                             vdwioffsetptr0+vdwjidx0H,
1255                                             &c6_00,&c12_00);
1256
1257             /* LENNARD-JONES DISPERSION/REPULSION */
1258
1259             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1260             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1261
1262             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1263
1264             fscal            = fvdw;
1265
1266             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1267
1268             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1269
1270             /* Calculate temporary vectorial force */
1271             tx               = _mm256_mul_ps(fscal,dx00);
1272             ty               = _mm256_mul_ps(fscal,dy00);
1273             tz               = _mm256_mul_ps(fscal,dz00);
1274
1275             /* Update vectorial force */
1276             fix0             = _mm256_add_ps(fix0,tx);
1277             fiy0             = _mm256_add_ps(fiy0,ty);
1278             fiz0             = _mm256_add_ps(fiz0,tz);
1279
1280             fjx0             = _mm256_add_ps(fjx0,tx);
1281             fjy0             = _mm256_add_ps(fjy0,ty);
1282             fjz0             = _mm256_add_ps(fjz0,tz);
1283
1284             }
1285
1286             /**************************
1287              * CALCULATE INTERACTIONS *
1288              **************************/
1289
1290             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1291             {
1292
1293             /* Compute parameters for interactions between i and j atoms */
1294             qq10             = _mm256_mul_ps(iq1,jq0);
1295
1296             /* REACTION-FIELD ELECTROSTATICS */
1297             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1298
1299             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1300
1301             fscal            = felec;
1302
1303             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1304
1305             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1306
1307             /* Calculate temporary vectorial force */
1308             tx               = _mm256_mul_ps(fscal,dx10);
1309             ty               = _mm256_mul_ps(fscal,dy10);
1310             tz               = _mm256_mul_ps(fscal,dz10);
1311
1312             /* Update vectorial force */
1313             fix1             = _mm256_add_ps(fix1,tx);
1314             fiy1             = _mm256_add_ps(fiy1,ty);
1315             fiz1             = _mm256_add_ps(fiz1,tz);
1316
1317             fjx0             = _mm256_add_ps(fjx0,tx);
1318             fjy0             = _mm256_add_ps(fjy0,ty);
1319             fjz0             = _mm256_add_ps(fjz0,tz);
1320
1321             }
1322
1323             /**************************
1324              * CALCULATE INTERACTIONS *
1325              **************************/
1326
1327             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1328             {
1329
1330             /* Compute parameters for interactions between i and j atoms */
1331             qq20             = _mm256_mul_ps(iq2,jq0);
1332
1333             /* REACTION-FIELD ELECTROSTATICS */
1334             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1335
1336             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1337
1338             fscal            = felec;
1339
1340             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1341
1342             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1343
1344             /* Calculate temporary vectorial force */
1345             tx               = _mm256_mul_ps(fscal,dx20);
1346             ty               = _mm256_mul_ps(fscal,dy20);
1347             tz               = _mm256_mul_ps(fscal,dz20);
1348
1349             /* Update vectorial force */
1350             fix2             = _mm256_add_ps(fix2,tx);
1351             fiy2             = _mm256_add_ps(fiy2,ty);
1352             fiz2             = _mm256_add_ps(fiz2,tz);
1353
1354             fjx0             = _mm256_add_ps(fjx0,tx);
1355             fjy0             = _mm256_add_ps(fjy0,ty);
1356             fjz0             = _mm256_add_ps(fjz0,tz);
1357
1358             }
1359
1360             /**************************
1361              * CALCULATE INTERACTIONS *
1362              **************************/
1363
1364             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1365             {
1366
1367             /* Compute parameters for interactions between i and j atoms */
1368             qq30             = _mm256_mul_ps(iq3,jq0);
1369
1370             /* REACTION-FIELD ELECTROSTATICS */
1371             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1372
1373             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1374
1375             fscal            = felec;
1376
1377             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1378
1379             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1380
1381             /* Calculate temporary vectorial force */
1382             tx               = _mm256_mul_ps(fscal,dx30);
1383             ty               = _mm256_mul_ps(fscal,dy30);
1384             tz               = _mm256_mul_ps(fscal,dz30);
1385
1386             /* Update vectorial force */
1387             fix3             = _mm256_add_ps(fix3,tx);
1388             fiy3             = _mm256_add_ps(fiy3,ty);
1389             fiz3             = _mm256_add_ps(fiz3,tz);
1390
1391             fjx0             = _mm256_add_ps(fjx0,tx);
1392             fjy0             = _mm256_add_ps(fjy0,ty);
1393             fjz0             = _mm256_add_ps(fjz0,tz);
1394
1395             }
1396
1397             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1398             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1399             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1400             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1401             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1402             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1403             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1404             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1405
1406             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1407
1408             /* Inner loop uses 123 flops */
1409         }
1410
1411         /* End of innermost loop */
1412
1413         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1414                                                  f+i_coord_offset,fshift+i_shift_offset);
1415
1416         /* Increment number of inner iterations */
1417         inneriter                  += j_index_end - j_index_start;
1418
1419         /* Outer loop uses 24 flops */
1420     }
1421
1422     /* Increment number of outer iterations */
1423     outeriter        += nri;
1424
1425     /* Update outer/inner flops */
1426
1427     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
1428 }