src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_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_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  90     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  91     __m256i          ewitab;
  92     __m128i          ewitab_lo,ewitab_hi;
  93     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  95     real             *ewtab;
  96     __m256           dummy_mask,cutoff_mask;
  97     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  98     __m256           one     = _mm256_set1_ps(1.0);
  99     __m256           two     = _mm256_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm256_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 118     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 119     beta2            = _mm256_mul_ps(beta,beta);
 120     beta3            = _mm256_mul_ps(beta,beta2);
 121
 122     ewtab            = fr->ic->tabq_coul_FDV0;
 123     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 124     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 129     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 130     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 131     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 132
 133     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 134     rcutoff_scalar   = fr->rcoulomb;
 135     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 136     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 137
 138     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 139     rvdw             = _mm256_set1_ps(fr->rvdw);
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147     j_coord_offsetE = 0;
 148     j_coord_offsetF = 0;
 149     j_coord_offsetG = 0;
 150     j_coord_offsetH = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm256_setzero_ps();
 179         fiy0             = _mm256_setzero_ps();
 180         fiz0             = _mm256_setzero_ps();
 181         fix1             = _mm256_setzero_ps();
 182         fiy1             = _mm256_setzero_ps();
 183         fiz1             = _mm256_setzero_ps();
 184         fix2             = _mm256_setzero_ps();
 185         fiy2             = _mm256_setzero_ps();
 186         fiz2             = _mm256_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_ps();
 190         vvdwsum          = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225             dx10             = _mm256_sub_ps(ix1,jx0);
 226             dy10             = _mm256_sub_ps(iy1,jy0);
 227             dz10             = _mm256_sub_ps(iz1,jz0);
 228             dx20             = _mm256_sub_ps(ix2,jx0);
 229             dy20             = _mm256_sub_ps(iy2,jy0);
 230             dz20             = _mm256_sub_ps(iz2,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 236
 237             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 240
 241             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 242             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 243             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 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             r00              = _mm256_mul_ps(rsq00,rinv00);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq00             = _mm256_mul_ps(iq0,jq0);
 274             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 275                                             vdwioffsetptr0+vdwjidx0B,
 276                                             vdwioffsetptr0+vdwjidx0C,
 277                                             vdwioffsetptr0+vdwjidx0D,
 278                                             vdwioffsetptr0+vdwjidx0E,
 279                                             vdwioffsetptr0+vdwjidx0F,
 280                                             vdwioffsetptr0+vdwjidx0G,
 281                                             vdwioffsetptr0+vdwjidx0H,
 282                                             &c6_00,&c12_00);
 283
 284             /* EWALD ELECTROSTATICS */
 285
 286             /* Analytical PME correction */
 287             zeta2            = _mm256_mul_ps(beta2,rsq00);
 288             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 289             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 290             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 291             felec            = _mm256_mul_ps(qq00,felec);
 292             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 293             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 294             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 295             velec            = _mm256_mul_ps(qq00,velec);
 296
 297             /* LENNARD-JONES DISPERSION/REPULSION */
 298
 299             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 300             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 301             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 302             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) ,
 303                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 304             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 305
 306             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             velec            = _mm256_and_ps(velec,cutoff_mask);
 310             velecsum         = _mm256_add_ps(velecsum,velec);
 311             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 312             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 313
 314             fscal            = _mm256_add_ps(felec,fvdw);
 315
 316             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm256_mul_ps(fscal,dx00);
 320             ty               = _mm256_mul_ps(fscal,dy00);
 321             tz               = _mm256_mul_ps(fscal,dz00);
 322
 323             /* Update vectorial force */
 324             fix0             = _mm256_add_ps(fix0,tx);
 325             fiy0             = _mm256_add_ps(fiy0,ty);
 326             fiz0             = _mm256_add_ps(fiz0,tz);
 327
 328             fjx0             = _mm256_add_ps(fjx0,tx);
 329             fjy0             = _mm256_add_ps(fjy0,ty);
 330             fjz0             = _mm256_add_ps(fjz0,tz);
 331
 332             }
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 339             {
 340
 341             r10              = _mm256_mul_ps(rsq10,rinv10);
 342
 343             /* Compute parameters for interactions between i and j atoms */
 344             qq10             = _mm256_mul_ps(iq1,jq0);
 345
 346             /* EWALD ELECTROSTATICS */
 347
 348             /* Analytical PME correction */
 349             zeta2            = _mm256_mul_ps(beta2,rsq10);
 350             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 351             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 352             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 353             felec            = _mm256_mul_ps(qq10,felec);
 354             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 355             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 356             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 357             velec            = _mm256_mul_ps(qq10,velec);
 358
 359             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm256_and_ps(velec,cutoff_mask);
 363             velecsum         = _mm256_add_ps(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm256_mul_ps(fscal,dx10);
 371             ty               = _mm256_mul_ps(fscal,dy10);
 372             tz               = _mm256_mul_ps(fscal,dz10);
 373
 374             /* Update vectorial force */
 375             fix1             = _mm256_add_ps(fix1,tx);
 376             fiy1             = _mm256_add_ps(fiy1,ty);
 377             fiz1             = _mm256_add_ps(fiz1,tz);
 378
 379             fjx0             = _mm256_add_ps(fjx0,tx);
 380             fjy0             = _mm256_add_ps(fjy0,ty);
 381             fjz0             = _mm256_add_ps(fjz0,tz);
 382
 383             }
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 390             {
 391
 392             r20              = _mm256_mul_ps(rsq20,rinv20);
 393
 394             /* Compute parameters for interactions between i and j atoms */
 395             qq20             = _mm256_mul_ps(iq2,jq0);
 396
 397             /* EWALD ELECTROSTATICS */
 398
 399             /* Analytical PME correction */
 400             zeta2            = _mm256_mul_ps(beta2,rsq20);
 401             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 402             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 403             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 404             felec            = _mm256_mul_ps(qq20,felec);
 405             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 406             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 407             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 408             velec            = _mm256_mul_ps(qq20,velec);
 409
 410             cutoff_mask      = _mm256_cmp_ps(rsq20,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,dx20);
 422             ty               = _mm256_mul_ps(fscal,dy20);
 423             tz               = _mm256_mul_ps(fscal,dz20);
 424
 425             /* Update vectorial force */
 426             fix2             = _mm256_add_ps(fix2,tx);
 427             fiy2             = _mm256_add_ps(fiy2,ty);
 428             fiz2             = _mm256_add_ps(fiz2,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 348 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
 504             /* Calculate squared distance and things based on it */
 505             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 506             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 507             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 508
 509             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 510             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 511             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 512
 513             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 514             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 515             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 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             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 540             {
 541
 542             r00              = _mm256_mul_ps(rsq00,rinv00);
 543             r00              = _mm256_andnot_ps(dummy_mask,r00);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq00             = _mm256_mul_ps(iq0,jq0);
 547             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 548                                             vdwioffsetptr0+vdwjidx0B,
 549                                             vdwioffsetptr0+vdwjidx0C,
 550                                             vdwioffsetptr0+vdwjidx0D,
 551                                             vdwioffsetptr0+vdwjidx0E,
 552                                             vdwioffsetptr0+vdwjidx0F,
 553                                             vdwioffsetptr0+vdwjidx0G,
 554                                             vdwioffsetptr0+vdwjidx0H,
 555                                             &c6_00,&c12_00);
 556
 557             /* EWALD ELECTROSTATICS */
 558
 559             /* Analytical PME correction */
 560             zeta2            = _mm256_mul_ps(beta2,rsq00);
 561             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 562             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 563             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 564             felec            = _mm256_mul_ps(qq00,felec);
 565             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 566             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 567             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 568             velec            = _mm256_mul_ps(qq00,velec);
 569
 570             /* LENNARD-JONES DISPERSION/REPULSION */
 571
 572             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 573             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 574             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 575             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) ,
 576                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 577             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 578
 579             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             velec            = _mm256_and_ps(velec,cutoff_mask);
 583             velec            = _mm256_andnot_ps(dummy_mask,velec);
 584             velecsum         = _mm256_add_ps(velecsum,velec);
 585             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 586             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 587             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 588
 589             fscal            = _mm256_add_ps(felec,fvdw);
 590
 591             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 592
 593             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm256_mul_ps(fscal,dx00);
 597             ty               = _mm256_mul_ps(fscal,dy00);
 598             tz               = _mm256_mul_ps(fscal,dz00);
 599
 600             /* Update vectorial force */
 601             fix0             = _mm256_add_ps(fix0,tx);
 602             fiy0             = _mm256_add_ps(fiy0,ty);
 603             fiz0             = _mm256_add_ps(fiz0,tz);
 604
 605             fjx0             = _mm256_add_ps(fjx0,tx);
 606             fjy0             = _mm256_add_ps(fjy0,ty);
 607             fjz0             = _mm256_add_ps(fjz0,tz);
 608
 609             }
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 616             {
 617
 618             r10              = _mm256_mul_ps(rsq10,rinv10);
 619             r10              = _mm256_andnot_ps(dummy_mask,r10);
 620
 621             /* Compute parameters for interactions between i and j atoms */
 622             qq10             = _mm256_mul_ps(iq1,jq0);
 623
 624             /* EWALD ELECTROSTATICS */
 625
 626             /* Analytical PME correction */
 627             zeta2            = _mm256_mul_ps(beta2,rsq10);
 628             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 629             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 630             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 631             felec            = _mm256_mul_ps(qq10,felec);
 632             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 633             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 634             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 635             velec            = _mm256_mul_ps(qq10,velec);
 636
 637             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm256_and_ps(velec,cutoff_mask);
 641             velec            = _mm256_andnot_ps(dummy_mask,velec);
 642             velecsum         = _mm256_add_ps(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 647
 648             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm256_mul_ps(fscal,dx10);
 652             ty               = _mm256_mul_ps(fscal,dy10);
 653             tz               = _mm256_mul_ps(fscal,dz10);
 654
 655             /* Update vectorial force */
 656             fix1             = _mm256_add_ps(fix1,tx);
 657             fiy1             = _mm256_add_ps(fiy1,ty);
 658             fiz1             = _mm256_add_ps(fiz1,tz);
 659
 660             fjx0             = _mm256_add_ps(fjx0,tx);
 661             fjy0             = _mm256_add_ps(fjy0,ty);
 662             fjz0             = _mm256_add_ps(fjz0,tz);
 663
 664             }
 665
 666             /**************************
 667              * CALCULATE INTERACTIONS *
 668              **************************/
 669
 670             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 671             {
 672
 673             r20              = _mm256_mul_ps(rsq20,rinv20);
 674             r20              = _mm256_andnot_ps(dummy_mask,r20);
 675
 676             /* Compute parameters for interactions between i and j atoms */
 677             qq20             = _mm256_mul_ps(iq2,jq0);
 678
 679             /* EWALD ELECTROSTATICS */
 680
 681             /* Analytical PME correction */
 682             zeta2            = _mm256_mul_ps(beta2,rsq20);
 683             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 684             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 685             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 686             felec            = _mm256_mul_ps(qq20,felec);
 687             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 688             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 689             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 690             velec            = _mm256_mul_ps(qq20,velec);
 691
 692             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 693
 694             /* Update potential sum for this i atom from the interaction with this j atom. */
 695             velec            = _mm256_and_ps(velec,cutoff_mask);
 696             velec            = _mm256_andnot_ps(dummy_mask,velec);
 697             velecsum         = _mm256_add_ps(velecsum,velec);
 698
 699             fscal            = felec;
 700
 701             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 702
 703             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm256_mul_ps(fscal,dx20);
 707             ty               = _mm256_mul_ps(fscal,dy20);
 708             tz               = _mm256_mul_ps(fscal,dz20);
 709
 710             /* Update vectorial force */
 711             fix2             = _mm256_add_ps(fix2,tx);
 712             fiy2             = _mm256_add_ps(fiy2,ty);
 713             fiz2             = _mm256_add_ps(fiz2,tz);
 714
 715             fjx0             = _mm256_add_ps(fjx0,tx);
 716             fjy0             = _mm256_add_ps(fjy0,ty);
 717             fjz0             = _mm256_add_ps(fjz0,tz);
 718
 719             }
 720
 721             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 722             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 723             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 724             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 725             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 726             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 727             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 728             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 729
 730             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 731
 732             /* Inner loop uses 351 flops */
 733         }
 734
 735         /* End of innermost loop */
 736
 737         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 738                                                  f+i_coord_offset,fshift+i_shift_offset);
 739
 740         ggid                        = gid[iidx];
 741         /* Update potential energies */
 742         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 743         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 744
 745         /* Increment number of inner iterations */
 746         inneriter                  += j_index_end - j_index_start;
 747
 748         /* Outer loop uses 20 flops */
 749     }
 750
 751     /* Increment number of outer iterations */
 752     outeriter        += nri;
 753
 754     /* Update outer/inner flops */
 755
 756     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*351);
 757 }
 758 /*
 759  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_256_single
 760  * Electrostatics interaction: Ewald
 761  * VdW interaction:            LennardJones
 762  * Geometry:                   Water3-Particle
 763  * Calculate force/pot:        Force
 764  */
 765 void
 766 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_256_single
 767                     (t_nblist * gmx_restrict                nlist,
 768                      rvec * gmx_restrict                    xx,
 769                      rvec * gmx_restrict                    ff,
 770                      t_forcerec * gmx_restrict              fr,
 771                      t_mdatoms * gmx_restrict               mdatoms,
 772                      nb_kernel_data_t * gmx_restrict        kernel_data,
 773                      t_nrnb * gmx_restrict                  nrnb)
 774 {
 775     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 776      * just 0 for non-waters.
 777      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 778      * jnr indices corresponding to data put in the four positions in the SIMD register.
 779      */
 780     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 781     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 782     int              jnrA,jnrB,jnrC,jnrD;
 783     int              jnrE,jnrF,jnrG,jnrH;
 784     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 785     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 786     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 787     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 788     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 789     real             rcutoff_scalar;
 790     real             *shiftvec,*fshift,*x,*f;
 791     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 792     real             scratch[4*DIM];
 793     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 794     real *           vdwioffsetptr0;
 795     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 796     real *           vdwioffsetptr1;
 797     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 798     real *           vdwioffsetptr2;
 799     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 800     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 801     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 802     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 803     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 804     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 805     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 806     real             *charge;
 807     int              nvdwtype;
 808     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 809     int              *vdwtype;
 810     real             *vdwparam;
 811     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 812     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 813     __m256i          ewitab;
 814     __m128i          ewitab_lo,ewitab_hi;
 815     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 816     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 817     real             *ewtab;
 818     __m256           dummy_mask,cutoff_mask;
 819     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 820     __m256           one     = _mm256_set1_ps(1.0);
 821     __m256           two     = _mm256_set1_ps(2.0);
 822     x                = xx[0];
 823     f                = ff[0];
 824
 825     nri              = nlist->nri;
 826     iinr             = nlist->iinr;
 827     jindex           = nlist->jindex;
 828     jjnr             = nlist->jjnr;
 829     shiftidx         = nlist->shift;
 830     gid              = nlist->gid;
 831     shiftvec         = fr->shift_vec[0];
 832     fshift           = fr->fshift[0];
 833     facel            = _mm256_set1_ps(fr->epsfac);
 834     charge           = mdatoms->chargeA;
 835     nvdwtype         = fr->ntype;
 836     vdwparam         = fr->nbfp;
 837     vdwtype          = mdatoms->typeA;
 838
 839     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 840     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 841     beta2            = _mm256_mul_ps(beta,beta);
 842     beta3            = _mm256_mul_ps(beta,beta2);
 843
 844     ewtab            = fr->ic->tabq_coul_F;
 845     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 846     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 847
 848     /* Setup water-specific parameters */
 849     inr              = nlist->iinr[0];
 850     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 851     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 852     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 853     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 854
 855     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 856     rcutoff_scalar   = fr->rcoulomb;
 857     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 858     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 859
 860     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 861     rvdw             = _mm256_set1_ps(fr->rvdw);
 862
 863     /* Avoid stupid compiler warnings */
 864     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 865     j_coord_offsetA = 0;
 866     j_coord_offsetB = 0;
 867     j_coord_offsetC = 0;
 868     j_coord_offsetD = 0;
 869     j_coord_offsetE = 0;
 870     j_coord_offsetF = 0;
 871     j_coord_offsetG = 0;
 872     j_coord_offsetH = 0;
 873
 874     outeriter        = 0;
 875     inneriter        = 0;
 876
 877     for(iidx=0;iidx<4*DIM;iidx++)
 878     {
 879         scratch[iidx] = 0.0;
 880     }
 881
 882     /* Start outer loop over neighborlists */
 883     for(iidx=0; iidx<nri; iidx++)
 884     {
 885         /* Load shift vector for this list */
 886         i_shift_offset   = DIM*shiftidx[iidx];
 887
 888         /* Load limits for loop over neighbors */
 889         j_index_start    = jindex[iidx];
 890         j_index_end      = jindex[iidx+1];
 891
 892         /* Get outer coordinate index */
 893         inr              = iinr[iidx];
 894         i_coord_offset   = DIM*inr;
 895
 896         /* Load i particle coords and add shift vector */
 897         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 898                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 899
 900         fix0             = _mm256_setzero_ps();
 901         fiy0             = _mm256_setzero_ps();
 902         fiz0             = _mm256_setzero_ps();
 903         fix1             = _mm256_setzero_ps();
 904         fiy1             = _mm256_setzero_ps();
 905         fiz1             = _mm256_setzero_ps();
 906         fix2             = _mm256_setzero_ps();
 907         fiy2             = _mm256_setzero_ps();
 908         fiz2             = _mm256_setzero_ps();
 909
 910         /* Start inner kernel loop */
 911         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 912         {
 913
 914             /* Get j neighbor index, and coordinate index */
 915             jnrA             = jjnr[jidx];
 916             jnrB             = jjnr[jidx+1];
 917             jnrC             = jjnr[jidx+2];
 918             jnrD             = jjnr[jidx+3];
 919             jnrE             = jjnr[jidx+4];
 920             jnrF             = jjnr[jidx+5];
 921             jnrG             = jjnr[jidx+6];
 922             jnrH             = jjnr[jidx+7];
 923             j_coord_offsetA  = DIM*jnrA;
 924             j_coord_offsetB  = DIM*jnrB;
 925             j_coord_offsetC  = DIM*jnrC;
 926             j_coord_offsetD  = DIM*jnrD;
 927             j_coord_offsetE  = DIM*jnrE;
 928             j_coord_offsetF  = DIM*jnrF;
 929             j_coord_offsetG  = DIM*jnrG;
 930             j_coord_offsetH  = DIM*jnrH;
 931
 932             /* load j atom coordinates */
 933             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 934                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 935                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 936                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 937                                                  &jx0,&jy0,&jz0);
 938
 939             /* Calculate displacement vector */
 940             dx00             = _mm256_sub_ps(ix0,jx0);
 941             dy00             = _mm256_sub_ps(iy0,jy0);
 942             dz00             = _mm256_sub_ps(iz0,jz0);
 943             dx10             = _mm256_sub_ps(ix1,jx0);
 944             dy10             = _mm256_sub_ps(iy1,jy0);
 945             dz10             = _mm256_sub_ps(iz1,jz0);
 946             dx20             = _mm256_sub_ps(ix2,jx0);
 947             dy20             = _mm256_sub_ps(iy2,jy0);
 948             dz20             = _mm256_sub_ps(iz2,jz0);
 949
 950             /* Calculate squared distance and things based on it */
 951             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 952             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 953             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 954
 955             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 956             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 957             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 958
 959             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 960             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 961             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 962
 963             /* Load parameters for j particles */
 964             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 965                                                                  charge+jnrC+0,charge+jnrD+0,
 966                                                                  charge+jnrE+0,charge+jnrF+0,
 967                                                                  charge+jnrG+0,charge+jnrH+0);
 968             vdwjidx0A        = 2*vdwtype[jnrA+0];
 969             vdwjidx0B        = 2*vdwtype[jnrB+0];
 970             vdwjidx0C        = 2*vdwtype[jnrC+0];
 971             vdwjidx0D        = 2*vdwtype[jnrD+0];
 972             vdwjidx0E        = 2*vdwtype[jnrE+0];
 973             vdwjidx0F        = 2*vdwtype[jnrF+0];
 974             vdwjidx0G        = 2*vdwtype[jnrG+0];
 975             vdwjidx0H        = 2*vdwtype[jnrH+0];
 976
 977             fjx0             = _mm256_setzero_ps();
 978             fjy0             = _mm256_setzero_ps();
 979             fjz0             = _mm256_setzero_ps();
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 986             {
 987
 988             r00              = _mm256_mul_ps(rsq00,rinv00);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq00             = _mm256_mul_ps(iq0,jq0);
 992             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 993                                             vdwioffsetptr0+vdwjidx0B,
 994                                             vdwioffsetptr0+vdwjidx0C,
 995                                             vdwioffsetptr0+vdwjidx0D,
 996                                             vdwioffsetptr0+vdwjidx0E,
 997                                             vdwioffsetptr0+vdwjidx0F,
 998                                             vdwioffsetptr0+vdwjidx0G,
 999                                             vdwioffsetptr0+vdwjidx0H,
1000                                             &c6_00,&c12_00);
1001
1002             /* EWALD ELECTROSTATICS */
1003
1004             /* Analytical PME correction */
1005             zeta2            = _mm256_mul_ps(beta2,rsq00);
1006             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1007             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1008             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1009             felec            = _mm256_mul_ps(qq00,felec);
1010
1011             /* LENNARD-JONES DISPERSION/REPULSION */
1012
1013             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1014             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1015
1016             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1017
1018             fscal            = _mm256_add_ps(felec,fvdw);
1019
1020             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1021
1022             /* Calculate temporary vectorial force */
1023             tx               = _mm256_mul_ps(fscal,dx00);
1024             ty               = _mm256_mul_ps(fscal,dy00);
1025             tz               = _mm256_mul_ps(fscal,dz00);
1026
1027             /* Update vectorial force */
1028             fix0             = _mm256_add_ps(fix0,tx);
1029             fiy0             = _mm256_add_ps(fiy0,ty);
1030             fiz0             = _mm256_add_ps(fiz0,tz);
1031
1032             fjx0             = _mm256_add_ps(fjx0,tx);
1033             fjy0             = _mm256_add_ps(fjy0,ty);
1034             fjz0             = _mm256_add_ps(fjz0,tz);
1035
1036             }
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1043             {
1044
1045             r10              = _mm256_mul_ps(rsq10,rinv10);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq10             = _mm256_mul_ps(iq1,jq0);
1049
1050             /* EWALD ELECTROSTATICS */
1051
1052             /* Analytical PME correction */
1053             zeta2            = _mm256_mul_ps(beta2,rsq10);
1054             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1055             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1056             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1057             felec            = _mm256_mul_ps(qq10,felec);
1058
1059             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1060
1061             fscal            = felec;
1062
1063             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1064
1065             /* Calculate temporary vectorial force */
1066             tx               = _mm256_mul_ps(fscal,dx10);
1067             ty               = _mm256_mul_ps(fscal,dy10);
1068             tz               = _mm256_mul_ps(fscal,dz10);
1069
1070             /* Update vectorial force */
1071             fix1             = _mm256_add_ps(fix1,tx);
1072             fiy1             = _mm256_add_ps(fiy1,ty);
1073             fiz1             = _mm256_add_ps(fiz1,tz);
1074
1075             fjx0             = _mm256_add_ps(fjx0,tx);
1076             fjy0             = _mm256_add_ps(fjy0,ty);
1077             fjz0             = _mm256_add_ps(fjz0,tz);
1078
1079             }
1080
1081             /**************************
1082              * CALCULATE INTERACTIONS *
1083              **************************/
1084
1085             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1086             {
1087
1088             r20              = _mm256_mul_ps(rsq20,rinv20);
1089
1090             /* Compute parameters for interactions between i and j atoms */
1091             qq20             = _mm256_mul_ps(iq2,jq0);
1092
1093             /* EWALD ELECTROSTATICS */
1094
1095             /* Analytical PME correction */
1096             zeta2            = _mm256_mul_ps(beta2,rsq20);
1097             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1098             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1099             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1100             felec            = _mm256_mul_ps(qq20,felec);
1101
1102             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1107
1108             /* Calculate temporary vectorial force */
1109             tx               = _mm256_mul_ps(fscal,dx20);
1110             ty               = _mm256_mul_ps(fscal,dy20);
1111             tz               = _mm256_mul_ps(fscal,dz20);
1112
1113             /* Update vectorial force */
1114             fix2             = _mm256_add_ps(fix2,tx);
1115             fiy2             = _mm256_add_ps(fiy2,ty);
1116             fiz2             = _mm256_add_ps(fiz2,tz);
1117
1118             fjx0             = _mm256_add_ps(fjx0,tx);
1119             fjy0             = _mm256_add_ps(fjy0,ty);
1120             fjz0             = _mm256_add_ps(fjz0,tz);
1121
1122             }
1123
1124             fjptrA             = f+j_coord_offsetA;
1125             fjptrB             = f+j_coord_offsetB;
1126             fjptrC             = f+j_coord_offsetC;
1127             fjptrD             = f+j_coord_offsetD;
1128             fjptrE             = f+j_coord_offsetE;
1129             fjptrF             = f+j_coord_offsetF;
1130             fjptrG             = f+j_coord_offsetG;
1131             fjptrH             = f+j_coord_offsetH;
1132
1133             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1134
1135             /* Inner loop uses 187 flops */
1136         }
1137
1138         if(jidx<j_index_end)
1139         {
1140
1141             /* Get j neighbor index, and coordinate index */
1142             jnrlistA         = jjnr[jidx];
1143             jnrlistB         = jjnr[jidx+1];
1144             jnrlistC         = jjnr[jidx+2];
1145             jnrlistD         = jjnr[jidx+3];
1146             jnrlistE         = jjnr[jidx+4];
1147             jnrlistF         = jjnr[jidx+5];
1148             jnrlistG         = jjnr[jidx+6];
1149             jnrlistH         = jjnr[jidx+7];
1150             /* Sign of each element will be negative for non-real atoms.
1151              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1152              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1153              */
1154             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1155                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1156
1157             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1158             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1159             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1160             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1161             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1162             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1163             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1164             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1165             j_coord_offsetA  = DIM*jnrA;
1166             j_coord_offsetB  = DIM*jnrB;
1167             j_coord_offsetC  = DIM*jnrC;
1168             j_coord_offsetD  = DIM*jnrD;
1169             j_coord_offsetE  = DIM*jnrE;
1170             j_coord_offsetF  = DIM*jnrF;
1171             j_coord_offsetG  = DIM*jnrG;
1172             j_coord_offsetH  = DIM*jnrH;
1173
1174             /* load j atom coordinates */
1175             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1176                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1177                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1178                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1179                                                  &jx0,&jy0,&jz0);
1180
1181             /* Calculate displacement vector */
1182             dx00             = _mm256_sub_ps(ix0,jx0);
1183             dy00             = _mm256_sub_ps(iy0,jy0);
1184             dz00             = _mm256_sub_ps(iz0,jz0);
1185             dx10             = _mm256_sub_ps(ix1,jx0);
1186             dy10             = _mm256_sub_ps(iy1,jy0);
1187             dz10             = _mm256_sub_ps(iz1,jz0);
1188             dx20             = _mm256_sub_ps(ix2,jx0);
1189             dy20             = _mm256_sub_ps(iy2,jy0);
1190             dz20             = _mm256_sub_ps(iz2,jz0);
1191
1192             /* Calculate squared distance and things based on it */
1193             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1194             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1195             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1196
1197             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1198             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1199             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1200
1201             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1202             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1203             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1204
1205             /* Load parameters for j particles */
1206             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1207                                                                  charge+jnrC+0,charge+jnrD+0,
1208                                                                  charge+jnrE+0,charge+jnrF+0,
1209                                                                  charge+jnrG+0,charge+jnrH+0);
1210             vdwjidx0A        = 2*vdwtype[jnrA+0];
1211             vdwjidx0B        = 2*vdwtype[jnrB+0];
1212             vdwjidx0C        = 2*vdwtype[jnrC+0];
1213             vdwjidx0D        = 2*vdwtype[jnrD+0];
1214             vdwjidx0E        = 2*vdwtype[jnrE+0];
1215             vdwjidx0F        = 2*vdwtype[jnrF+0];
1216             vdwjidx0G        = 2*vdwtype[jnrG+0];
1217             vdwjidx0H        = 2*vdwtype[jnrH+0];
1218
1219             fjx0             = _mm256_setzero_ps();
1220             fjy0             = _mm256_setzero_ps();
1221             fjz0             = _mm256_setzero_ps();
1222
1223             /**************************
1224              * CALCULATE INTERACTIONS *
1225              **************************/
1226
1227             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1228             {
1229
1230             r00              = _mm256_mul_ps(rsq00,rinv00);
1231             r00              = _mm256_andnot_ps(dummy_mask,r00);
1232
1233             /* Compute parameters for interactions between i and j atoms */
1234             qq00             = _mm256_mul_ps(iq0,jq0);
1235             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1236                                             vdwioffsetptr0+vdwjidx0B,
1237                                             vdwioffsetptr0+vdwjidx0C,
1238                                             vdwioffsetptr0+vdwjidx0D,
1239                                             vdwioffsetptr0+vdwjidx0E,
1240                                             vdwioffsetptr0+vdwjidx0F,
1241                                             vdwioffsetptr0+vdwjidx0G,
1242                                             vdwioffsetptr0+vdwjidx0H,
1243                                             &c6_00,&c12_00);
1244
1245             /* EWALD ELECTROSTATICS */
1246
1247             /* Analytical PME correction */
1248             zeta2            = _mm256_mul_ps(beta2,rsq00);
1249             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1250             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1251             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1252             felec            = _mm256_mul_ps(qq00,felec);
1253
1254             /* LENNARD-JONES DISPERSION/REPULSION */
1255
1256             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1257             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1258
1259             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1260
1261             fscal            = _mm256_add_ps(felec,fvdw);
1262
1263             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1264
1265             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1266
1267             /* Calculate temporary vectorial force */
1268             tx               = _mm256_mul_ps(fscal,dx00);
1269             ty               = _mm256_mul_ps(fscal,dy00);
1270             tz               = _mm256_mul_ps(fscal,dz00);
1271
1272             /* Update vectorial force */
1273             fix0             = _mm256_add_ps(fix0,tx);
1274             fiy0             = _mm256_add_ps(fiy0,ty);
1275             fiz0             = _mm256_add_ps(fiz0,tz);
1276
1277             fjx0             = _mm256_add_ps(fjx0,tx);
1278             fjy0             = _mm256_add_ps(fjy0,ty);
1279             fjz0             = _mm256_add_ps(fjz0,tz);
1280
1281             }
1282
1283             /**************************
1284              * CALCULATE INTERACTIONS *
1285              **************************/
1286
1287             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1288             {
1289
1290             r10              = _mm256_mul_ps(rsq10,rinv10);
1291             r10              = _mm256_andnot_ps(dummy_mask,r10);
1292
1293             /* Compute parameters for interactions between i and j atoms */
1294             qq10             = _mm256_mul_ps(iq1,jq0);
1295
1296             /* EWALD ELECTROSTATICS */
1297
1298             /* Analytical PME correction */
1299             zeta2            = _mm256_mul_ps(beta2,rsq10);
1300             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1301             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1302             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1303             felec            = _mm256_mul_ps(qq10,felec);
1304
1305             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1306
1307             fscal            = felec;
1308
1309             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1310
1311             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1312
1313             /* Calculate temporary vectorial force */
1314             tx               = _mm256_mul_ps(fscal,dx10);
1315             ty               = _mm256_mul_ps(fscal,dy10);
1316             tz               = _mm256_mul_ps(fscal,dz10);
1317
1318             /* Update vectorial force */
1319             fix1             = _mm256_add_ps(fix1,tx);
1320             fiy1             = _mm256_add_ps(fiy1,ty);
1321             fiz1             = _mm256_add_ps(fiz1,tz);
1322
1323             fjx0             = _mm256_add_ps(fjx0,tx);
1324             fjy0             = _mm256_add_ps(fjy0,ty);
1325             fjz0             = _mm256_add_ps(fjz0,tz);
1326
1327             }
1328
1329             /**************************
1330              * CALCULATE INTERACTIONS *
1331              **************************/
1332
1333             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1334             {
1335
1336             r20              = _mm256_mul_ps(rsq20,rinv20);
1337             r20              = _mm256_andnot_ps(dummy_mask,r20);
1338
1339             /* Compute parameters for interactions between i and j atoms */
1340             qq20             = _mm256_mul_ps(iq2,jq0);
1341
1342             /* EWALD ELECTROSTATICS */
1343
1344             /* Analytical PME correction */
1345             zeta2            = _mm256_mul_ps(beta2,rsq20);
1346             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1347             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1348             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1349             felec            = _mm256_mul_ps(qq20,felec);
1350
1351             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1352
1353             fscal            = felec;
1354
1355             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1356
1357             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1358
1359             /* Calculate temporary vectorial force */
1360             tx               = _mm256_mul_ps(fscal,dx20);
1361             ty               = _mm256_mul_ps(fscal,dy20);
1362             tz               = _mm256_mul_ps(fscal,dz20);
1363
1364             /* Update vectorial force */
1365             fix2             = _mm256_add_ps(fix2,tx);
1366             fiy2             = _mm256_add_ps(fiy2,ty);
1367             fiz2             = _mm256_add_ps(fiz2,tz);
1368
1369             fjx0             = _mm256_add_ps(fjx0,tx);
1370             fjy0             = _mm256_add_ps(fjy0,ty);
1371             fjz0             = _mm256_add_ps(fjz0,tz);
1372
1373             }
1374
1375             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1376             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1377             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1378             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1379             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1380             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1381             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1382             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1383
1384             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1385
1386             /* Inner loop uses 190 flops */
1387         }
1388
1389         /* End of innermost loop */
1390
1391         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1392                                                  f+i_coord_offset,fshift+i_shift_offset);
1393
1394         /* Increment number of inner iterations */
1395         inneriter                  += j_index_end - j_index_start;
1396
1397         /* Outer loop uses 18 flops */
1398     }
1399
1400     /* Increment number of outer iterations */
1401     outeriter        += nri;
1402
1403     /* Update outer/inner flops */
1404
1405     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*190);
1406 }