src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwLJ_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_ElecEw_VdwLJ_GeomW4P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_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     __m256i          ewitab;
  95     __m128i          ewitab_lo,ewitab_hi;
  96     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  98     real             *ewtab;
  99     __m256           dummy_mask,cutoff_mask;
 100     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 101     __m256           one     = _mm256_set1_ps(1.0);
 102     __m256           two     = _mm256_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm256_set1_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 121     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 122     beta2            = _mm256_mul_ps(beta,beta);
 123     beta3            = _mm256_mul_ps(beta,beta2);
 124
 125     ewtab            = fr->ic->tabq_coul_FDV0;
 126     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 127     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 132     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 133     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 134     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142     j_coord_offsetE = 0;
 143     j_coord_offsetF = 0;
 144     j_coord_offsetG = 0;
 145     j_coord_offsetH = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 172
 173         fix0             = _mm256_setzero_ps();
 174         fiy0             = _mm256_setzero_ps();
 175         fiz0             = _mm256_setzero_ps();
 176         fix1             = _mm256_setzero_ps();
 177         fiy1             = _mm256_setzero_ps();
 178         fiz1             = _mm256_setzero_ps();
 179         fix2             = _mm256_setzero_ps();
 180         fiy2             = _mm256_setzero_ps();
 181         fiz2             = _mm256_setzero_ps();
 182         fix3             = _mm256_setzero_ps();
 183         fiy3             = _mm256_setzero_ps();
 184         fiz3             = _mm256_setzero_ps();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm256_setzero_ps();
 188         vvdwsum          = _mm256_setzero_ps();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             jnrE             = jjnr[jidx+4];
 200             jnrF             = jjnr[jidx+5];
 201             jnrG             = jjnr[jidx+6];
 202             jnrH             = jjnr[jidx+7];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207             j_coord_offsetE  = DIM*jnrE;
 208             j_coord_offsetF  = DIM*jnrF;
 209             j_coord_offsetG  = DIM*jnrG;
 210             j_coord_offsetH  = DIM*jnrH;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 216                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 217                                                  &jx0,&jy0,&jz0);
 218
 219             /* Calculate displacement vector */
 220             dx00             = _mm256_sub_ps(ix0,jx0);
 221             dy00             = _mm256_sub_ps(iy0,jy0);
 222             dz00             = _mm256_sub_ps(iz0,jz0);
 223             dx10             = _mm256_sub_ps(ix1,jx0);
 224             dy10             = _mm256_sub_ps(iy1,jy0);
 225             dz10             = _mm256_sub_ps(iz1,jz0);
 226             dx20             = _mm256_sub_ps(ix2,jx0);
 227             dy20             = _mm256_sub_ps(iy2,jy0);
 228             dz20             = _mm256_sub_ps(iz2,jz0);
 229             dx30             = _mm256_sub_ps(ix3,jx0);
 230             dy30             = _mm256_sub_ps(iy3,jy0);
 231             dz30             = _mm256_sub_ps(iz3,jz0);
 232
 233             /* Calculate squared distance and things based on it */
 234             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 235             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 236             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 237             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 238
 239             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 240             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 241             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 242
 243             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 244             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 245             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 246             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 247
 248             /* Load parameters for j particles */
 249             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 250                                                                  charge+jnrC+0,charge+jnrD+0,
 251                                                                  charge+jnrE+0,charge+jnrF+0,
 252                                                                  charge+jnrG+0,charge+jnrH+0);
 253             vdwjidx0A        = 2*vdwtype[jnrA+0];
 254             vdwjidx0B        = 2*vdwtype[jnrB+0];
 255             vdwjidx0C        = 2*vdwtype[jnrC+0];
 256             vdwjidx0D        = 2*vdwtype[jnrD+0];
 257             vdwjidx0E        = 2*vdwtype[jnrE+0];
 258             vdwjidx0F        = 2*vdwtype[jnrF+0];
 259             vdwjidx0G        = 2*vdwtype[jnrG+0];
 260             vdwjidx0H        = 2*vdwtype[jnrH+0];
 261
 262             fjx0             = _mm256_setzero_ps();
 263             fjy0             = _mm256_setzero_ps();
 264             fjz0             = _mm256_setzero_ps();
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 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(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 287             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 291
 292             fscal            = fvdw;
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm256_mul_ps(fscal,dx00);
 296             ty               = _mm256_mul_ps(fscal,dy00);
 297             tz               = _mm256_mul_ps(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm256_add_ps(fix0,tx);
 301             fiy0             = _mm256_add_ps(fiy0,ty);
 302             fiz0             = _mm256_add_ps(fiz0,tz);
 303
 304             fjx0             = _mm256_add_ps(fjx0,tx);
 305             fjy0             = _mm256_add_ps(fjy0,ty);
 306             fjz0             = _mm256_add_ps(fjz0,tz);
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             r10              = _mm256_mul_ps(rsq10,rinv10);
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq10             = _mm256_mul_ps(iq1,jq0);
 316
 317             /* EWALD ELECTROSTATICS */
 318
 319             /* Analytical PME correction */
 320             zeta2            = _mm256_mul_ps(beta2,rsq10);
 321             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 322             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 323             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 324             felec            = _mm256_mul_ps(qq10,felec);
 325             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 326             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 327             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 328             velec            = _mm256_mul_ps(qq10,velec);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm256_add_ps(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_ps(fscal,dx10);
 337             ty               = _mm256_mul_ps(fscal,dy10);
 338             tz               = _mm256_mul_ps(fscal,dz10);
 339
 340             /* Update vectorial force */
 341             fix1             = _mm256_add_ps(fix1,tx);
 342             fiy1             = _mm256_add_ps(fiy1,ty);
 343             fiz1             = _mm256_add_ps(fiz1,tz);
 344
 345             fjx0             = _mm256_add_ps(fjx0,tx);
 346             fjy0             = _mm256_add_ps(fjy0,ty);
 347             fjz0             = _mm256_add_ps(fjz0,tz);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             r20              = _mm256_mul_ps(rsq20,rinv20);
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq20             = _mm256_mul_ps(iq2,jq0);
 357
 358             /* EWALD ELECTROSTATICS */
 359
 360             /* Analytical PME correction */
 361             zeta2            = _mm256_mul_ps(beta2,rsq20);
 362             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 363             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 364             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 365             felec            = _mm256_mul_ps(qq20,felec);
 366             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 367             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 368             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 369             velec            = _mm256_mul_ps(qq20,velec);
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velecsum         = _mm256_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm256_mul_ps(fscal,dx20);
 378             ty               = _mm256_mul_ps(fscal,dy20);
 379             tz               = _mm256_mul_ps(fscal,dz20);
 380
 381             /* Update vectorial force */
 382             fix2             = _mm256_add_ps(fix2,tx);
 383             fiy2             = _mm256_add_ps(fiy2,ty);
 384             fiz2             = _mm256_add_ps(fiz2,tz);
 385
 386             fjx0             = _mm256_add_ps(fjx0,tx);
 387             fjy0             = _mm256_add_ps(fjy0,ty);
 388             fjz0             = _mm256_add_ps(fjz0,tz);
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             r30              = _mm256_mul_ps(rsq30,rinv30);
 395
 396             /* Compute parameters for interactions between i and j atoms */
 397             qq30             = _mm256_mul_ps(iq3,jq0);
 398
 399             /* EWALD ELECTROSTATICS */
 400
 401             /* Analytical PME correction */
 402             zeta2            = _mm256_mul_ps(beta2,rsq30);
 403             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 404             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 405             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 406             felec            = _mm256_mul_ps(qq30,felec);
 407             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 408             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 409             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 410             velec            = _mm256_mul_ps(qq30,velec);
 411
 412             /* Update potential sum for this i atom from the interaction with this j atom. */
 413             velecsum         = _mm256_add_ps(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm256_mul_ps(fscal,dx30);
 419             ty               = _mm256_mul_ps(fscal,dy30);
 420             tz               = _mm256_mul_ps(fscal,dz30);
 421
 422             /* Update vectorial force */
 423             fix3             = _mm256_add_ps(fix3,tx);
 424             fiy3             = _mm256_add_ps(fiy3,ty);
 425             fiz3             = _mm256_add_ps(fiz3,tz);
 426
 427             fjx0             = _mm256_add_ps(fjx0,tx);
 428             fjy0             = _mm256_add_ps(fjy0,ty);
 429             fjz0             = _mm256_add_ps(fjz0,tz);
 430
 431             fjptrA             = f+j_coord_offsetA;
 432             fjptrB             = f+j_coord_offsetB;
 433             fjptrC             = f+j_coord_offsetC;
 434             fjptrD             = f+j_coord_offsetD;
 435             fjptrE             = f+j_coord_offsetE;
 436             fjptrF             = f+j_coord_offsetF;
 437             fjptrG             = f+j_coord_offsetG;
 438             fjptrH             = f+j_coord_offsetH;
 439
 440             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 441
 442             /* Inner loop uses 287 flops */
 443         }
 444
 445         if(jidx<j_index_end)
 446         {
 447
 448             /* Get j neighbor index, and coordinate index */
 449             jnrlistA         = jjnr[jidx];
 450             jnrlistB         = jjnr[jidx+1];
 451             jnrlistC         = jjnr[jidx+2];
 452             jnrlistD         = jjnr[jidx+3];
 453             jnrlistE         = jjnr[jidx+4];
 454             jnrlistF         = jjnr[jidx+5];
 455             jnrlistG         = jjnr[jidx+6];
 456             jnrlistH         = jjnr[jidx+7];
 457             /* Sign of each element will be negative for non-real atoms.
 458              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 459              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 460              */
 461             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 462                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 463
 464             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 465             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 466             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 467             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 468             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 469             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 470             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 471             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 472             j_coord_offsetA  = DIM*jnrA;
 473             j_coord_offsetB  = DIM*jnrB;
 474             j_coord_offsetC  = DIM*jnrC;
 475             j_coord_offsetD  = DIM*jnrD;
 476             j_coord_offsetE  = DIM*jnrE;
 477             j_coord_offsetF  = DIM*jnrF;
 478             j_coord_offsetG  = DIM*jnrG;
 479             j_coord_offsetH  = DIM*jnrH;
 480
 481             /* load j atom coordinates */
 482             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 484                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 485                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 486                                                  &jx0,&jy0,&jz0);
 487
 488             /* Calculate displacement vector */
 489             dx00             = _mm256_sub_ps(ix0,jx0);
 490             dy00             = _mm256_sub_ps(iy0,jy0);
 491             dz00             = _mm256_sub_ps(iz0,jz0);
 492             dx10             = _mm256_sub_ps(ix1,jx0);
 493             dy10             = _mm256_sub_ps(iy1,jy0);
 494             dz10             = _mm256_sub_ps(iz1,jz0);
 495             dx20             = _mm256_sub_ps(ix2,jx0);
 496             dy20             = _mm256_sub_ps(iy2,jy0);
 497             dz20             = _mm256_sub_ps(iz2,jz0);
 498             dx30             = _mm256_sub_ps(ix3,jx0);
 499             dy30             = _mm256_sub_ps(iy3,jy0);
 500             dz30             = _mm256_sub_ps(iz3,jz0);
 501
 502             /* Calculate squared distance and things based on it */
 503             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 504             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 505             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 506             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 507
 508             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 509             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 510             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 511
 512             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 513             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 514             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 515             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 516
 517             /* Load parameters for j particles */
 518             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 519                                                                  charge+jnrC+0,charge+jnrD+0,
 520                                                                  charge+jnrE+0,charge+jnrF+0,
 521                                                                  charge+jnrG+0,charge+jnrH+0);
 522             vdwjidx0A        = 2*vdwtype[jnrA+0];
 523             vdwjidx0B        = 2*vdwtype[jnrB+0];
 524             vdwjidx0C        = 2*vdwtype[jnrC+0];
 525             vdwjidx0D        = 2*vdwtype[jnrD+0];
 526             vdwjidx0E        = 2*vdwtype[jnrE+0];
 527             vdwjidx0F        = 2*vdwtype[jnrF+0];
 528             vdwjidx0G        = 2*vdwtype[jnrG+0];
 529             vdwjidx0H        = 2*vdwtype[jnrH+0];
 530
 531             fjx0             = _mm256_setzero_ps();
 532             fjy0             = _mm256_setzero_ps();
 533             fjz0             = _mm256_setzero_ps();
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 541                                             vdwioffsetptr0+vdwjidx0B,
 542                                             vdwioffsetptr0+vdwjidx0C,
 543                                             vdwioffsetptr0+vdwjidx0D,
 544                                             vdwioffsetptr0+vdwjidx0E,
 545                                             vdwioffsetptr0+vdwjidx0F,
 546                                             vdwioffsetptr0+vdwjidx0G,
 547                                             vdwioffsetptr0+vdwjidx0H,
 548                                             &c6_00,&c12_00);
 549
 550             /* LENNARD-JONES DISPERSION/REPULSION */
 551
 552             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 553             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 554             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 555             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 556             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 560             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 561
 562             fscal            = fvdw;
 563
 564             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm256_mul_ps(fscal,dx00);
 568             ty               = _mm256_mul_ps(fscal,dy00);
 569             tz               = _mm256_mul_ps(fscal,dz00);
 570
 571             /* Update vectorial force */
 572             fix0             = _mm256_add_ps(fix0,tx);
 573             fiy0             = _mm256_add_ps(fiy0,ty);
 574             fiz0             = _mm256_add_ps(fiz0,tz);
 575
 576             fjx0             = _mm256_add_ps(fjx0,tx);
 577             fjy0             = _mm256_add_ps(fjy0,ty);
 578             fjz0             = _mm256_add_ps(fjz0,tz);
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             r10              = _mm256_mul_ps(rsq10,rinv10);
 585             r10              = _mm256_andnot_ps(dummy_mask,r10);
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             qq10             = _mm256_mul_ps(iq1,jq0);
 589
 590             /* EWALD ELECTROSTATICS */
 591
 592             /* Analytical PME correction */
 593             zeta2            = _mm256_mul_ps(beta2,rsq10);
 594             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 595             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 596             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 597             felec            = _mm256_mul_ps(qq10,felec);
 598             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 599             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 600             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 601             velec            = _mm256_mul_ps(qq10,velec);
 602
 603             /* Update potential sum for this i atom from the interaction with this j atom. */
 604             velec            = _mm256_andnot_ps(dummy_mask,velec);
 605             velecsum         = _mm256_add_ps(velecsum,velec);
 606
 607             fscal            = felec;
 608
 609             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm256_mul_ps(fscal,dx10);
 613             ty               = _mm256_mul_ps(fscal,dy10);
 614             tz               = _mm256_mul_ps(fscal,dz10);
 615
 616             /* Update vectorial force */
 617             fix1             = _mm256_add_ps(fix1,tx);
 618             fiy1             = _mm256_add_ps(fiy1,ty);
 619             fiz1             = _mm256_add_ps(fiz1,tz);
 620
 621             fjx0             = _mm256_add_ps(fjx0,tx);
 622             fjy0             = _mm256_add_ps(fjy0,ty);
 623             fjz0             = _mm256_add_ps(fjz0,tz);
 624
 625             /**************************
 626              * CALCULATE INTERACTIONS *
 627              **************************/
 628
 629             r20              = _mm256_mul_ps(rsq20,rinv20);
 630             r20              = _mm256_andnot_ps(dummy_mask,r20);
 631
 632             /* Compute parameters for interactions between i and j atoms */
 633             qq20             = _mm256_mul_ps(iq2,jq0);
 634
 635             /* EWALD ELECTROSTATICS */
 636
 637             /* Analytical PME correction */
 638             zeta2            = _mm256_mul_ps(beta2,rsq20);
 639             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 640             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 641             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 642             felec            = _mm256_mul_ps(qq20,felec);
 643             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 644             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 645             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 646             velec            = _mm256_mul_ps(qq20,velec);
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velec            = _mm256_andnot_ps(dummy_mask,velec);
 650             velecsum         = _mm256_add_ps(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 655
 656             /* Calculate temporary vectorial force */
 657             tx               = _mm256_mul_ps(fscal,dx20);
 658             ty               = _mm256_mul_ps(fscal,dy20);
 659             tz               = _mm256_mul_ps(fscal,dz20);
 660
 661             /* Update vectorial force */
 662             fix2             = _mm256_add_ps(fix2,tx);
 663             fiy2             = _mm256_add_ps(fiy2,ty);
 664             fiz2             = _mm256_add_ps(fiz2,tz);
 665
 666             fjx0             = _mm256_add_ps(fjx0,tx);
 667             fjy0             = _mm256_add_ps(fjy0,ty);
 668             fjz0             = _mm256_add_ps(fjz0,tz);
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             r30              = _mm256_mul_ps(rsq30,rinv30);
 675             r30              = _mm256_andnot_ps(dummy_mask,r30);
 676
 677             /* Compute parameters for interactions between i and j atoms */
 678             qq30             = _mm256_mul_ps(iq3,jq0);
 679
 680             /* EWALD ELECTROSTATICS */
 681
 682             /* Analytical PME correction */
 683             zeta2            = _mm256_mul_ps(beta2,rsq30);
 684             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 685             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 686             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 687             felec            = _mm256_mul_ps(qq30,felec);
 688             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 689             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 690             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 691             velec            = _mm256_mul_ps(qq30,velec);
 692
 693             /* Update potential sum for this i atom from the interaction with this j atom. */
 694             velec            = _mm256_andnot_ps(dummy_mask,velec);
 695             velecsum         = _mm256_add_ps(velecsum,velec);
 696
 697             fscal            = felec;
 698
 699             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 700
 701             /* Calculate temporary vectorial force */
 702             tx               = _mm256_mul_ps(fscal,dx30);
 703             ty               = _mm256_mul_ps(fscal,dy30);
 704             tz               = _mm256_mul_ps(fscal,dz30);
 705
 706             /* Update vectorial force */
 707             fix3             = _mm256_add_ps(fix3,tx);
 708             fiy3             = _mm256_add_ps(fiy3,ty);
 709             fiz3             = _mm256_add_ps(fiz3,tz);
 710
 711             fjx0             = _mm256_add_ps(fjx0,tx);
 712             fjy0             = _mm256_add_ps(fjy0,ty);
 713             fjz0             = _mm256_add_ps(fjz0,tz);
 714
 715             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 716             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 717             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 718             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 719             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 720             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 721             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 722             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 723
 724             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 725
 726             /* Inner loop uses 290 flops */
 727         }
 728
 729         /* End of innermost loop */
 730
 731         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 732                                                  f+i_coord_offset,fshift+i_shift_offset);
 733
 734         ggid                        = gid[iidx];
 735         /* Update potential energies */
 736         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 737         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 738
 739         /* Increment number of inner iterations */
 740         inneriter                  += j_index_end - j_index_start;
 741
 742         /* Outer loop uses 26 flops */
 743     }
 744
 745     /* Increment number of outer iterations */
 746     outeriter        += nri;
 747
 748     /* Update outer/inner flops */
 749
 750     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*290);
 751 }
 752 /*
 753  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_256_single
 754  * Electrostatics interaction: Ewald
 755  * VdW interaction:            LennardJones
 756  * Geometry:                   Water4-Particle
 757  * Calculate force/pot:        Force
 758  */
 759 void
 760 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_256_single
 761                     (t_nblist * gmx_restrict                nlist,
 762                      rvec * gmx_restrict                    xx,
 763                      rvec * gmx_restrict                    ff,
 764                      t_forcerec * gmx_restrict              fr,
 765                      t_mdatoms * gmx_restrict               mdatoms,
 766                      nb_kernel_data_t * gmx_restrict        kernel_data,
 767                      t_nrnb * gmx_restrict                  nrnb)
 768 {
 769     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 770      * just 0 for non-waters.
 771      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 772      * jnr indices corresponding to data put in the four positions in the SIMD register.
 773      */
 774     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 775     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 776     int              jnrA,jnrB,jnrC,jnrD;
 777     int              jnrE,jnrF,jnrG,jnrH;
 778     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 779     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 780     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 781     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 782     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 783     real             rcutoff_scalar;
 784     real             *shiftvec,*fshift,*x,*f;
 785     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 786     real             scratch[4*DIM];
 787     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 788     real *           vdwioffsetptr0;
 789     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 790     real *           vdwioffsetptr1;
 791     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 792     real *           vdwioffsetptr2;
 793     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 794     real *           vdwioffsetptr3;
 795     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 796     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 797     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 798     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 799     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 800     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 801     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 802     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 803     real             *charge;
 804     int              nvdwtype;
 805     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 806     int              *vdwtype;
 807     real             *vdwparam;
 808     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 809     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 810     __m256i          ewitab;
 811     __m128i          ewitab_lo,ewitab_hi;
 812     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 813     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 814     real             *ewtab;
 815     __m256           dummy_mask,cutoff_mask;
 816     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 817     __m256           one     = _mm256_set1_ps(1.0);
 818     __m256           two     = _mm256_set1_ps(2.0);
 819     x                = xx[0];
 820     f                = ff[0];
 821
 822     nri              = nlist->nri;
 823     iinr             = nlist->iinr;
 824     jindex           = nlist->jindex;
 825     jjnr             = nlist->jjnr;
 826     shiftidx         = nlist->shift;
 827     gid              = nlist->gid;
 828     shiftvec         = fr->shift_vec[0];
 829     fshift           = fr->fshift[0];
 830     facel            = _mm256_set1_ps(fr->epsfac);
 831     charge           = mdatoms->chargeA;
 832     nvdwtype         = fr->ntype;
 833     vdwparam         = fr->nbfp;
 834     vdwtype          = mdatoms->typeA;
 835
 836     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 837     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 838     beta2            = _mm256_mul_ps(beta,beta);
 839     beta3            = _mm256_mul_ps(beta,beta2);
 840
 841     ewtab            = fr->ic->tabq_coul_F;
 842     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 843     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 844
 845     /* Setup water-specific parameters */
 846     inr              = nlist->iinr[0];
 847     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 848     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 849     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 850     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 851
 852     /* Avoid stupid compiler warnings */
 853     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 854     j_coord_offsetA = 0;
 855     j_coord_offsetB = 0;
 856     j_coord_offsetC = 0;
 857     j_coord_offsetD = 0;
 858     j_coord_offsetE = 0;
 859     j_coord_offsetF = 0;
 860     j_coord_offsetG = 0;
 861     j_coord_offsetH = 0;
 862
 863     outeriter        = 0;
 864     inneriter        = 0;
 865
 866     for(iidx=0;iidx<4*DIM;iidx++)
 867     {
 868         scratch[iidx] = 0.0;
 869     }
 870
 871     /* Start outer loop over neighborlists */
 872     for(iidx=0; iidx<nri; iidx++)
 873     {
 874         /* Load shift vector for this list */
 875         i_shift_offset   = DIM*shiftidx[iidx];
 876
 877         /* Load limits for loop over neighbors */
 878         j_index_start    = jindex[iidx];
 879         j_index_end      = jindex[iidx+1];
 880
 881         /* Get outer coordinate index */
 882         inr              = iinr[iidx];
 883         i_coord_offset   = DIM*inr;
 884
 885         /* Load i particle coords and add shift vector */
 886         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 887                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 888
 889         fix0             = _mm256_setzero_ps();
 890         fiy0             = _mm256_setzero_ps();
 891         fiz0             = _mm256_setzero_ps();
 892         fix1             = _mm256_setzero_ps();
 893         fiy1             = _mm256_setzero_ps();
 894         fiz1             = _mm256_setzero_ps();
 895         fix2             = _mm256_setzero_ps();
 896         fiy2             = _mm256_setzero_ps();
 897         fiz2             = _mm256_setzero_ps();
 898         fix3             = _mm256_setzero_ps();
 899         fiy3             = _mm256_setzero_ps();
 900         fiz3             = _mm256_setzero_ps();
 901
 902         /* Start inner kernel loop */
 903         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 904         {
 905
 906             /* Get j neighbor index, and coordinate index */
 907             jnrA             = jjnr[jidx];
 908             jnrB             = jjnr[jidx+1];
 909             jnrC             = jjnr[jidx+2];
 910             jnrD             = jjnr[jidx+3];
 911             jnrE             = jjnr[jidx+4];
 912             jnrF             = jjnr[jidx+5];
 913             jnrG             = jjnr[jidx+6];
 914             jnrH             = jjnr[jidx+7];
 915             j_coord_offsetA  = DIM*jnrA;
 916             j_coord_offsetB  = DIM*jnrB;
 917             j_coord_offsetC  = DIM*jnrC;
 918             j_coord_offsetD  = DIM*jnrD;
 919             j_coord_offsetE  = DIM*jnrE;
 920             j_coord_offsetF  = DIM*jnrF;
 921             j_coord_offsetG  = DIM*jnrG;
 922             j_coord_offsetH  = DIM*jnrH;
 923
 924             /* load j atom coordinates */
 925             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 926                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 927                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 928                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 929                                                  &jx0,&jy0,&jz0);
 930
 931             /* Calculate displacement vector */
 932             dx00             = _mm256_sub_ps(ix0,jx0);
 933             dy00             = _mm256_sub_ps(iy0,jy0);
 934             dz00             = _mm256_sub_ps(iz0,jz0);
 935             dx10             = _mm256_sub_ps(ix1,jx0);
 936             dy10             = _mm256_sub_ps(iy1,jy0);
 937             dz10             = _mm256_sub_ps(iz1,jz0);
 938             dx20             = _mm256_sub_ps(ix2,jx0);
 939             dy20             = _mm256_sub_ps(iy2,jy0);
 940             dz20             = _mm256_sub_ps(iz2,jz0);
 941             dx30             = _mm256_sub_ps(ix3,jx0);
 942             dy30             = _mm256_sub_ps(iy3,jy0);
 943             dz30             = _mm256_sub_ps(iz3,jz0);
 944
 945             /* Calculate squared distance and things based on it */
 946             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 947             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 948             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 949             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 950
 951             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 952             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 953             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 954
 955             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 956             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 957             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 958             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 959
 960             /* Load parameters for j particles */
 961             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 962                                                                  charge+jnrC+0,charge+jnrD+0,
 963                                                                  charge+jnrE+0,charge+jnrF+0,
 964                                                                  charge+jnrG+0,charge+jnrH+0);
 965             vdwjidx0A        = 2*vdwtype[jnrA+0];
 966             vdwjidx0B        = 2*vdwtype[jnrB+0];
 967             vdwjidx0C        = 2*vdwtype[jnrC+0];
 968             vdwjidx0D        = 2*vdwtype[jnrD+0];
 969             vdwjidx0E        = 2*vdwtype[jnrE+0];
 970             vdwjidx0F        = 2*vdwtype[jnrF+0];
 971             vdwjidx0G        = 2*vdwtype[jnrG+0];
 972             vdwjidx0H        = 2*vdwtype[jnrH+0];
 973
 974             fjx0             = _mm256_setzero_ps();
 975             fjy0             = _mm256_setzero_ps();
 976             fjz0             = _mm256_setzero_ps();
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 984                                             vdwioffsetptr0+vdwjidx0B,
 985                                             vdwioffsetptr0+vdwjidx0C,
 986                                             vdwioffsetptr0+vdwjidx0D,
 987                                             vdwioffsetptr0+vdwjidx0E,
 988                                             vdwioffsetptr0+vdwjidx0F,
 989                                             vdwioffsetptr0+vdwjidx0G,
 990                                             vdwioffsetptr0+vdwjidx0H,
 991                                             &c6_00,&c12_00);
 992
 993             /* LENNARD-JONES DISPERSION/REPULSION */
 994
 995             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 996             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
 997
 998             fscal            = fvdw;
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm256_mul_ps(fscal,dx00);
1002             ty               = _mm256_mul_ps(fscal,dy00);
1003             tz               = _mm256_mul_ps(fscal,dz00);
1004
1005             /* Update vectorial force */
1006             fix0             = _mm256_add_ps(fix0,tx);
1007             fiy0             = _mm256_add_ps(fiy0,ty);
1008             fiz0             = _mm256_add_ps(fiz0,tz);
1009
1010             fjx0             = _mm256_add_ps(fjx0,tx);
1011             fjy0             = _mm256_add_ps(fjy0,ty);
1012             fjz0             = _mm256_add_ps(fjz0,tz);
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             r10              = _mm256_mul_ps(rsq10,rinv10);
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq10             = _mm256_mul_ps(iq1,jq0);
1022
1023             /* EWALD ELECTROSTATICS */
1024
1025             /* Analytical PME correction */
1026             zeta2            = _mm256_mul_ps(beta2,rsq10);
1027             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1028             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1029             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1030             felec            = _mm256_mul_ps(qq10,felec);
1031
1032             fscal            = felec;
1033
1034             /* Calculate temporary vectorial force */
1035             tx               = _mm256_mul_ps(fscal,dx10);
1036             ty               = _mm256_mul_ps(fscal,dy10);
1037             tz               = _mm256_mul_ps(fscal,dz10);
1038
1039             /* Update vectorial force */
1040             fix1             = _mm256_add_ps(fix1,tx);
1041             fiy1             = _mm256_add_ps(fiy1,ty);
1042             fiz1             = _mm256_add_ps(fiz1,tz);
1043
1044             fjx0             = _mm256_add_ps(fjx0,tx);
1045             fjy0             = _mm256_add_ps(fjy0,ty);
1046             fjz0             = _mm256_add_ps(fjz0,tz);
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             r20              = _mm256_mul_ps(rsq20,rinv20);
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq20             = _mm256_mul_ps(iq2,jq0);
1056
1057             /* EWALD ELECTROSTATICS */
1058
1059             /* Analytical PME correction */
1060             zeta2            = _mm256_mul_ps(beta2,rsq20);
1061             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1062             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1063             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1064             felec            = _mm256_mul_ps(qq20,felec);
1065
1066             fscal            = felec;
1067
1068             /* Calculate temporary vectorial force */
1069             tx               = _mm256_mul_ps(fscal,dx20);
1070             ty               = _mm256_mul_ps(fscal,dy20);
1071             tz               = _mm256_mul_ps(fscal,dz20);
1072
1073             /* Update vectorial force */
1074             fix2             = _mm256_add_ps(fix2,tx);
1075             fiy2             = _mm256_add_ps(fiy2,ty);
1076             fiz2             = _mm256_add_ps(fiz2,tz);
1077
1078             fjx0             = _mm256_add_ps(fjx0,tx);
1079             fjy0             = _mm256_add_ps(fjy0,ty);
1080             fjz0             = _mm256_add_ps(fjz0,tz);
1081
1082             /**************************
1083              * CALCULATE INTERACTIONS *
1084              **************************/
1085
1086             r30              = _mm256_mul_ps(rsq30,rinv30);
1087
1088             /* Compute parameters for interactions between i and j atoms */
1089             qq30             = _mm256_mul_ps(iq3,jq0);
1090
1091             /* EWALD ELECTROSTATICS */
1092
1093             /* Analytical PME correction */
1094             zeta2            = _mm256_mul_ps(beta2,rsq30);
1095             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1096             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1097             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1098             felec            = _mm256_mul_ps(qq30,felec);
1099
1100             fscal            = felec;
1101
1102             /* Calculate temporary vectorial force */
1103             tx               = _mm256_mul_ps(fscal,dx30);
1104             ty               = _mm256_mul_ps(fscal,dy30);
1105             tz               = _mm256_mul_ps(fscal,dz30);
1106
1107             /* Update vectorial force */
1108             fix3             = _mm256_add_ps(fix3,tx);
1109             fiy3             = _mm256_add_ps(fiy3,ty);
1110             fiz3             = _mm256_add_ps(fiz3,tz);
1111
1112             fjx0             = _mm256_add_ps(fjx0,tx);
1113             fjy0             = _mm256_add_ps(fjy0,ty);
1114             fjz0             = _mm256_add_ps(fjz0,tz);
1115
1116             fjptrA             = f+j_coord_offsetA;
1117             fjptrB             = f+j_coord_offsetB;
1118             fjptrC             = f+j_coord_offsetC;
1119             fjptrD             = f+j_coord_offsetD;
1120             fjptrE             = f+j_coord_offsetE;
1121             fjptrF             = f+j_coord_offsetF;
1122             fjptrG             = f+j_coord_offsetG;
1123             fjptrH             = f+j_coord_offsetH;
1124
1125             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1126
1127             /* Inner loop uses 198 flops */
1128         }
1129
1130         if(jidx<j_index_end)
1131         {
1132
1133             /* Get j neighbor index, and coordinate index */
1134             jnrlistA         = jjnr[jidx];
1135             jnrlistB         = jjnr[jidx+1];
1136             jnrlistC         = jjnr[jidx+2];
1137             jnrlistD         = jjnr[jidx+3];
1138             jnrlistE         = jjnr[jidx+4];
1139             jnrlistF         = jjnr[jidx+5];
1140             jnrlistG         = jjnr[jidx+6];
1141             jnrlistH         = jjnr[jidx+7];
1142             /* Sign of each element will be negative for non-real atoms.
1143              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1144              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1145              */
1146             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1147                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1148
1149             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1150             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1151             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1152             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1153             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1154             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1155             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1156             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1157             j_coord_offsetA  = DIM*jnrA;
1158             j_coord_offsetB  = DIM*jnrB;
1159             j_coord_offsetC  = DIM*jnrC;
1160             j_coord_offsetD  = DIM*jnrD;
1161             j_coord_offsetE  = DIM*jnrE;
1162             j_coord_offsetF  = DIM*jnrF;
1163             j_coord_offsetG  = DIM*jnrG;
1164             j_coord_offsetH  = DIM*jnrH;
1165
1166             /* load j atom coordinates */
1167             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1168                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1169                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1170                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1171                                                  &jx0,&jy0,&jz0);
1172
1173             /* Calculate displacement vector */
1174             dx00             = _mm256_sub_ps(ix0,jx0);
1175             dy00             = _mm256_sub_ps(iy0,jy0);
1176             dz00             = _mm256_sub_ps(iz0,jz0);
1177             dx10             = _mm256_sub_ps(ix1,jx0);
1178             dy10             = _mm256_sub_ps(iy1,jy0);
1179             dz10             = _mm256_sub_ps(iz1,jz0);
1180             dx20             = _mm256_sub_ps(ix2,jx0);
1181             dy20             = _mm256_sub_ps(iy2,jy0);
1182             dz20             = _mm256_sub_ps(iz2,jz0);
1183             dx30             = _mm256_sub_ps(ix3,jx0);
1184             dy30             = _mm256_sub_ps(iy3,jy0);
1185             dz30             = _mm256_sub_ps(iz3,jz0);
1186
1187             /* Calculate squared distance and things based on it */
1188             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1189             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1190             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1191             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1192
1193             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1194             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1195             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1196
1197             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1198             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1199             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1200             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1201
1202             /* Load parameters for j particles */
1203             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1204                                                                  charge+jnrC+0,charge+jnrD+0,
1205                                                                  charge+jnrE+0,charge+jnrF+0,
1206                                                                  charge+jnrG+0,charge+jnrH+0);
1207             vdwjidx0A        = 2*vdwtype[jnrA+0];
1208             vdwjidx0B        = 2*vdwtype[jnrB+0];
1209             vdwjidx0C        = 2*vdwtype[jnrC+0];
1210             vdwjidx0D        = 2*vdwtype[jnrD+0];
1211             vdwjidx0E        = 2*vdwtype[jnrE+0];
1212             vdwjidx0F        = 2*vdwtype[jnrF+0];
1213             vdwjidx0G        = 2*vdwtype[jnrG+0];
1214             vdwjidx0H        = 2*vdwtype[jnrH+0];
1215
1216             fjx0             = _mm256_setzero_ps();
1217             fjy0             = _mm256_setzero_ps();
1218             fjz0             = _mm256_setzero_ps();
1219
1220             /**************************
1221              * CALCULATE INTERACTIONS *
1222              **************************/
1223
1224             /* Compute parameters for interactions between i and j atoms */
1225             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1226                                             vdwioffsetptr0+vdwjidx0B,
1227                                             vdwioffsetptr0+vdwjidx0C,
1228                                             vdwioffsetptr0+vdwjidx0D,
1229                                             vdwioffsetptr0+vdwjidx0E,
1230                                             vdwioffsetptr0+vdwjidx0F,
1231                                             vdwioffsetptr0+vdwjidx0G,
1232                                             vdwioffsetptr0+vdwjidx0H,
1233                                             &c6_00,&c12_00);
1234
1235             /* LENNARD-JONES DISPERSION/REPULSION */
1236
1237             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1238             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1239
1240             fscal            = fvdw;
1241
1242             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1243
1244             /* Calculate temporary vectorial force */
1245             tx               = _mm256_mul_ps(fscal,dx00);
1246             ty               = _mm256_mul_ps(fscal,dy00);
1247             tz               = _mm256_mul_ps(fscal,dz00);
1248
1249             /* Update vectorial force */
1250             fix0             = _mm256_add_ps(fix0,tx);
1251             fiy0             = _mm256_add_ps(fiy0,ty);
1252             fiz0             = _mm256_add_ps(fiz0,tz);
1253
1254             fjx0             = _mm256_add_ps(fjx0,tx);
1255             fjy0             = _mm256_add_ps(fjy0,ty);
1256             fjz0             = _mm256_add_ps(fjz0,tz);
1257
1258             /**************************
1259              * CALCULATE INTERACTIONS *
1260              **************************/
1261
1262             r10              = _mm256_mul_ps(rsq10,rinv10);
1263             r10              = _mm256_andnot_ps(dummy_mask,r10);
1264
1265             /* Compute parameters for interactions between i and j atoms */
1266             qq10             = _mm256_mul_ps(iq1,jq0);
1267
1268             /* EWALD ELECTROSTATICS */
1269
1270             /* Analytical PME correction */
1271             zeta2            = _mm256_mul_ps(beta2,rsq10);
1272             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1273             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1274             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1275             felec            = _mm256_mul_ps(qq10,felec);
1276
1277             fscal            = felec;
1278
1279             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1280
1281             /* Calculate temporary vectorial force */
1282             tx               = _mm256_mul_ps(fscal,dx10);
1283             ty               = _mm256_mul_ps(fscal,dy10);
1284             tz               = _mm256_mul_ps(fscal,dz10);
1285
1286             /* Update vectorial force */
1287             fix1             = _mm256_add_ps(fix1,tx);
1288             fiy1             = _mm256_add_ps(fiy1,ty);
1289             fiz1             = _mm256_add_ps(fiz1,tz);
1290
1291             fjx0             = _mm256_add_ps(fjx0,tx);
1292             fjy0             = _mm256_add_ps(fjy0,ty);
1293             fjz0             = _mm256_add_ps(fjz0,tz);
1294
1295             /**************************
1296              * CALCULATE INTERACTIONS *
1297              **************************/
1298
1299             r20              = _mm256_mul_ps(rsq20,rinv20);
1300             r20              = _mm256_andnot_ps(dummy_mask,r20);
1301
1302             /* Compute parameters for interactions between i and j atoms */
1303             qq20             = _mm256_mul_ps(iq2,jq0);
1304
1305             /* EWALD ELECTROSTATICS */
1306
1307             /* Analytical PME correction */
1308             zeta2            = _mm256_mul_ps(beta2,rsq20);
1309             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1310             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1311             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1312             felec            = _mm256_mul_ps(qq20,felec);
1313
1314             fscal            = felec;
1315
1316             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1317
1318             /* Calculate temporary vectorial force */
1319             tx               = _mm256_mul_ps(fscal,dx20);
1320             ty               = _mm256_mul_ps(fscal,dy20);
1321             tz               = _mm256_mul_ps(fscal,dz20);
1322
1323             /* Update vectorial force */
1324             fix2             = _mm256_add_ps(fix2,tx);
1325             fiy2             = _mm256_add_ps(fiy2,ty);
1326             fiz2             = _mm256_add_ps(fiz2,tz);
1327
1328             fjx0             = _mm256_add_ps(fjx0,tx);
1329             fjy0             = _mm256_add_ps(fjy0,ty);
1330             fjz0             = _mm256_add_ps(fjz0,tz);
1331
1332             /**************************
1333              * CALCULATE INTERACTIONS *
1334              **************************/
1335
1336             r30              = _mm256_mul_ps(rsq30,rinv30);
1337             r30              = _mm256_andnot_ps(dummy_mask,r30);
1338
1339             /* Compute parameters for interactions between i and j atoms */
1340             qq30             = _mm256_mul_ps(iq3,jq0);
1341
1342             /* EWALD ELECTROSTATICS */
1343
1344             /* Analytical PME correction */
1345             zeta2            = _mm256_mul_ps(beta2,rsq30);
1346             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1347             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1348             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1349             felec            = _mm256_mul_ps(qq30,felec);
1350
1351             fscal            = felec;
1352
1353             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1354
1355             /* Calculate temporary vectorial force */
1356             tx               = _mm256_mul_ps(fscal,dx30);
1357             ty               = _mm256_mul_ps(fscal,dy30);
1358             tz               = _mm256_mul_ps(fscal,dz30);
1359
1360             /* Update vectorial force */
1361             fix3             = _mm256_add_ps(fix3,tx);
1362             fiy3             = _mm256_add_ps(fiy3,ty);
1363             fiz3             = _mm256_add_ps(fiz3,tz);
1364
1365             fjx0             = _mm256_add_ps(fjx0,tx);
1366             fjy0             = _mm256_add_ps(fjy0,ty);
1367             fjz0             = _mm256_add_ps(fjz0,tz);
1368
1369             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1370             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1371             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1372             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1373             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1374             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1375             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1376             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1377
1378             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1379
1380             /* Inner loop uses 201 flops */
1381         }
1382
1383         /* End of innermost loop */
1384
1385         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1386                                                  f+i_coord_offset,fshift+i_shift_offset);
1387
1388         /* Increment number of inner iterations */
1389         inneriter                  += j_index_end - j_index_start;
1390
1391         /* Outer loop uses 24 flops */
1392     }
1393
1394     /* Increment number of outer iterations */
1395     outeriter        += nri;
1396
1397     /* Update outer/inner flops */
1398
1399     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*201);
1400 }