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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     real *           vdwioffsetptr3;
  92     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  94     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256           dummy_mask,cutoff_mask;
 108     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 109     __m256           one     = _mm256_set1_ps(1.0);
 110     __m256           two     = _mm256_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_ps(fr->ic->epsfac);
 123     charge           = mdatoms->chargeA;
 124     krf              = _mm256_set1_ps(fr->ic->k_rf);
 125     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 126     crf              = _mm256_set1_ps(fr->ic->c_rf);
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 134     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 135     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 136     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 137
 138     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 139     rcutoff_scalar   = fr->ic->rcoulomb;
 140     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 141     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 142
 143     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 144     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152     j_coord_offsetE = 0;
 153     j_coord_offsetF = 0;
 154     j_coord_offsetG = 0;
 155     j_coord_offsetH = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 181                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 182
 183         fix0             = _mm256_setzero_ps();
 184         fiy0             = _mm256_setzero_ps();
 185         fiz0             = _mm256_setzero_ps();
 186         fix1             = _mm256_setzero_ps();
 187         fiy1             = _mm256_setzero_ps();
 188         fiz1             = _mm256_setzero_ps();
 189         fix2             = _mm256_setzero_ps();
 190         fiy2             = _mm256_setzero_ps();
 191         fiz2             = _mm256_setzero_ps();
 192         fix3             = _mm256_setzero_ps();
 193         fiy3             = _mm256_setzero_ps();
 194         fiz3             = _mm256_setzero_ps();
 195
 196         /* Reset potential sums */
 197         velecsum         = _mm256_setzero_ps();
 198         vvdwsum          = _mm256_setzero_ps();
 199
 200         /* Start inner kernel loop */
 201         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 202         {
 203
 204             /* Get j neighbor index, and coordinate index */
 205             jnrA             = jjnr[jidx];
 206             jnrB             = jjnr[jidx+1];
 207             jnrC             = jjnr[jidx+2];
 208             jnrD             = jjnr[jidx+3];
 209             jnrE             = jjnr[jidx+4];
 210             jnrF             = jjnr[jidx+5];
 211             jnrG             = jjnr[jidx+6];
 212             jnrH             = jjnr[jidx+7];
 213             j_coord_offsetA  = DIM*jnrA;
 214             j_coord_offsetB  = DIM*jnrB;
 215             j_coord_offsetC  = DIM*jnrC;
 216             j_coord_offsetD  = DIM*jnrD;
 217             j_coord_offsetE  = DIM*jnrE;
 218             j_coord_offsetF  = DIM*jnrF;
 219             j_coord_offsetG  = DIM*jnrG;
 220             j_coord_offsetH  = DIM*jnrH;
 221
 222             /* load j atom coordinates */
 223             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 224                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 225                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 226                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 227                                                  &jx0,&jy0,&jz0);
 228
 229             /* Calculate displacement vector */
 230             dx00             = _mm256_sub_ps(ix0,jx0);
 231             dy00             = _mm256_sub_ps(iy0,jy0);
 232             dz00             = _mm256_sub_ps(iz0,jz0);
 233             dx10             = _mm256_sub_ps(ix1,jx0);
 234             dy10             = _mm256_sub_ps(iy1,jy0);
 235             dz10             = _mm256_sub_ps(iz1,jz0);
 236             dx20             = _mm256_sub_ps(ix2,jx0);
 237             dy20             = _mm256_sub_ps(iy2,jy0);
 238             dz20             = _mm256_sub_ps(iz2,jz0);
 239             dx30             = _mm256_sub_ps(ix3,jx0);
 240             dy30             = _mm256_sub_ps(iy3,jy0);
 241             dz30             = _mm256_sub_ps(iz3,jz0);
 242
 243             /* Calculate squared distance and things based on it */
 244             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 245             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 246             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 247             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 248
 249             rinv10           = avx256_invsqrt_f(rsq10);
 250             rinv20           = avx256_invsqrt_f(rsq20);
 251             rinv30           = avx256_invsqrt_f(rsq30);
 252
 253             rinvsq00         = avx256_inv_f(rsq00);
 254             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 255             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 256             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 257
 258             /* Load parameters for j particles */
 259             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 260                                                                  charge+jnrC+0,charge+jnrD+0,
 261                                                                  charge+jnrE+0,charge+jnrF+0,
 262                                                                  charge+jnrG+0,charge+jnrH+0);
 263             vdwjidx0A        = 2*vdwtype[jnrA+0];
 264             vdwjidx0B        = 2*vdwtype[jnrB+0];
 265             vdwjidx0C        = 2*vdwtype[jnrC+0];
 266             vdwjidx0D        = 2*vdwtype[jnrD+0];
 267             vdwjidx0E        = 2*vdwtype[jnrE+0];
 268             vdwjidx0F        = 2*vdwtype[jnrF+0];
 269             vdwjidx0G        = 2*vdwtype[jnrG+0];
 270             vdwjidx0H        = 2*vdwtype[jnrH+0];
 271
 272             fjx0             = _mm256_setzero_ps();
 273             fjy0             = _mm256_setzero_ps();
 274             fjz0             = _mm256_setzero_ps();
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 281             {
 282
 283             /* Compute parameters for interactions between i and j atoms */
 284             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 285                                             vdwioffsetptr0+vdwjidx0B,
 286                                             vdwioffsetptr0+vdwjidx0C,
 287                                             vdwioffsetptr0+vdwjidx0D,
 288                                             vdwioffsetptr0+vdwjidx0E,
 289                                             vdwioffsetptr0+vdwjidx0F,
 290                                             vdwioffsetptr0+vdwjidx0G,
 291                                             vdwioffsetptr0+vdwjidx0H,
 292                                             &c6_00,&c12_00);
 293
 294             /* LENNARD-JONES DISPERSION/REPULSION */
 295
 296             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 297             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 298             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 299             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) ,
 300                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 301             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 302
 303             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 307             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 308
 309             fscal            = fvdw;
 310
 311             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 312
 313             /* Calculate temporary vectorial force */
 314             tx               = _mm256_mul_ps(fscal,dx00);
 315             ty               = _mm256_mul_ps(fscal,dy00);
 316             tz               = _mm256_mul_ps(fscal,dz00);
 317
 318             /* Update vectorial force */
 319             fix0             = _mm256_add_ps(fix0,tx);
 320             fiy0             = _mm256_add_ps(fiy0,ty);
 321             fiz0             = _mm256_add_ps(fiz0,tz);
 322
 323             fjx0             = _mm256_add_ps(fjx0,tx);
 324             fjy0             = _mm256_add_ps(fjy0,ty);
 325             fjz0             = _mm256_add_ps(fjz0,tz);
 326
 327             }
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 334             {
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq10             = _mm256_mul_ps(iq1,jq0);
 338
 339             /* REACTION-FIELD ELECTROSTATICS */
 340             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 341             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 342
 343             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm256_and_ps(velec,cutoff_mask);
 347             velecsum         = _mm256_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm256_mul_ps(fscal,dx10);
 355             ty               = _mm256_mul_ps(fscal,dy10);
 356             tz               = _mm256_mul_ps(fscal,dz10);
 357
 358             /* Update vectorial force */
 359             fix1             = _mm256_add_ps(fix1,tx);
 360             fiy1             = _mm256_add_ps(fiy1,ty);
 361             fiz1             = _mm256_add_ps(fiz1,tz);
 362
 363             fjx0             = _mm256_add_ps(fjx0,tx);
 364             fjy0             = _mm256_add_ps(fjy0,ty);
 365             fjz0             = _mm256_add_ps(fjz0,tz);
 366
 367             }
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 374             {
 375
 376             /* Compute parameters for interactions between i and j atoms */
 377             qq20             = _mm256_mul_ps(iq2,jq0);
 378
 379             /* REACTION-FIELD ELECTROSTATICS */
 380             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 381             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 382
 383             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velec            = _mm256_and_ps(velec,cutoff_mask);
 387             velecsum         = _mm256_add_ps(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm256_mul_ps(fscal,dx20);
 395             ty               = _mm256_mul_ps(fscal,dy20);
 396             tz               = _mm256_mul_ps(fscal,dz20);
 397
 398             /* Update vectorial force */
 399             fix2             = _mm256_add_ps(fix2,tx);
 400             fiy2             = _mm256_add_ps(fiy2,ty);
 401             fiz2             = _mm256_add_ps(fiz2,tz);
 402
 403             fjx0             = _mm256_add_ps(fjx0,tx);
 404             fjy0             = _mm256_add_ps(fjy0,ty);
 405             fjz0             = _mm256_add_ps(fjz0,tz);
 406
 407             }
 408
 409             /**************************
 410              * CALCULATE INTERACTIONS *
 411              **************************/
 412
 413             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 414             {
 415
 416             /* Compute parameters for interactions between i and j atoms */
 417             qq30             = _mm256_mul_ps(iq3,jq0);
 418
 419             /* REACTION-FIELD ELECTROSTATICS */
 420             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 421             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 422
 423             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velec            = _mm256_and_ps(velec,cutoff_mask);
 427             velecsum         = _mm256_add_ps(velecsum,velec);
 428
 429             fscal            = felec;
 430
 431             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 432
 433             /* Calculate temporary vectorial force */
 434             tx               = _mm256_mul_ps(fscal,dx30);
 435             ty               = _mm256_mul_ps(fscal,dy30);
 436             tz               = _mm256_mul_ps(fscal,dz30);
 437
 438             /* Update vectorial force */
 439             fix3             = _mm256_add_ps(fix3,tx);
 440             fiy3             = _mm256_add_ps(fiy3,ty);
 441             fiz3             = _mm256_add_ps(fiz3,tz);
 442
 443             fjx0             = _mm256_add_ps(fjx0,tx);
 444             fjy0             = _mm256_add_ps(fjy0,ty);
 445             fjz0             = _mm256_add_ps(fjz0,tz);
 446
 447             }
 448
 449             fjptrA             = f+j_coord_offsetA;
 450             fjptrB             = f+j_coord_offsetB;
 451             fjptrC             = f+j_coord_offsetC;
 452             fjptrD             = f+j_coord_offsetD;
 453             fjptrE             = f+j_coord_offsetE;
 454             fjptrF             = f+j_coord_offsetF;
 455             fjptrG             = f+j_coord_offsetG;
 456             fjptrH             = f+j_coord_offsetH;
 457
 458             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 459
 460             /* Inner loop uses 152 flops */
 461         }
 462
 463         if(jidx<j_index_end)
 464         {
 465
 466             /* Get j neighbor index, and coordinate index */
 467             jnrlistA         = jjnr[jidx];
 468             jnrlistB         = jjnr[jidx+1];
 469             jnrlistC         = jjnr[jidx+2];
 470             jnrlistD         = jjnr[jidx+3];
 471             jnrlistE         = jjnr[jidx+4];
 472             jnrlistF         = jjnr[jidx+5];
 473             jnrlistG         = jjnr[jidx+6];
 474             jnrlistH         = jjnr[jidx+7];
 475             /* Sign of each element will be negative for non-real atoms.
 476              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 477              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 478              */
 479             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 480                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 481
 482             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 483             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 484             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 485             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 486             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 487             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 488             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 489             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 490             j_coord_offsetA  = DIM*jnrA;
 491             j_coord_offsetB  = DIM*jnrB;
 492             j_coord_offsetC  = DIM*jnrC;
 493             j_coord_offsetD  = DIM*jnrD;
 494             j_coord_offsetE  = DIM*jnrE;
 495             j_coord_offsetF  = DIM*jnrF;
 496             j_coord_offsetG  = DIM*jnrG;
 497             j_coord_offsetH  = DIM*jnrH;
 498
 499             /* load j atom coordinates */
 500             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 501                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 502                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 503                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 504                                                  &jx0,&jy0,&jz0);
 505
 506             /* Calculate displacement vector */
 507             dx00             = _mm256_sub_ps(ix0,jx0);
 508             dy00             = _mm256_sub_ps(iy0,jy0);
 509             dz00             = _mm256_sub_ps(iz0,jz0);
 510             dx10             = _mm256_sub_ps(ix1,jx0);
 511             dy10             = _mm256_sub_ps(iy1,jy0);
 512             dz10             = _mm256_sub_ps(iz1,jz0);
 513             dx20             = _mm256_sub_ps(ix2,jx0);
 514             dy20             = _mm256_sub_ps(iy2,jy0);
 515             dz20             = _mm256_sub_ps(iz2,jz0);
 516             dx30             = _mm256_sub_ps(ix3,jx0);
 517             dy30             = _mm256_sub_ps(iy3,jy0);
 518             dz30             = _mm256_sub_ps(iz3,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 522             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 523             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 524             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 525
 526             rinv10           = avx256_invsqrt_f(rsq10);
 527             rinv20           = avx256_invsqrt_f(rsq20);
 528             rinv30           = avx256_invsqrt_f(rsq30);
 529
 530             rinvsq00         = avx256_inv_f(rsq00);
 531             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 532             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 533             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 534
 535             /* Load parameters for j particles */
 536             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 537                                                                  charge+jnrC+0,charge+jnrD+0,
 538                                                                  charge+jnrE+0,charge+jnrF+0,
 539                                                                  charge+jnrG+0,charge+jnrH+0);
 540             vdwjidx0A        = 2*vdwtype[jnrA+0];
 541             vdwjidx0B        = 2*vdwtype[jnrB+0];
 542             vdwjidx0C        = 2*vdwtype[jnrC+0];
 543             vdwjidx0D        = 2*vdwtype[jnrD+0];
 544             vdwjidx0E        = 2*vdwtype[jnrE+0];
 545             vdwjidx0F        = 2*vdwtype[jnrF+0];
 546             vdwjidx0G        = 2*vdwtype[jnrG+0];
 547             vdwjidx0H        = 2*vdwtype[jnrH+0];
 548
 549             fjx0             = _mm256_setzero_ps();
 550             fjy0             = _mm256_setzero_ps();
 551             fjz0             = _mm256_setzero_ps();
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 558             {
 559
 560             /* Compute parameters for interactions between i and j atoms */
 561             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 562                                             vdwioffsetptr0+vdwjidx0B,
 563                                             vdwioffsetptr0+vdwjidx0C,
 564                                             vdwioffsetptr0+vdwjidx0D,
 565                                             vdwioffsetptr0+vdwjidx0E,
 566                                             vdwioffsetptr0+vdwjidx0F,
 567                                             vdwioffsetptr0+vdwjidx0G,
 568                                             vdwioffsetptr0+vdwjidx0H,
 569                                             &c6_00,&c12_00);
 570
 571             /* LENNARD-JONES DISPERSION/REPULSION */
 572
 573             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 574             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 575             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 576             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) ,
 577                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 578             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 579
 580             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 584             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 585             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 586
 587             fscal            = fvdw;
 588
 589             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 590
 591             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm256_mul_ps(fscal,dx00);
 595             ty               = _mm256_mul_ps(fscal,dy00);
 596             tz               = _mm256_mul_ps(fscal,dz00);
 597
 598             /* Update vectorial force */
 599             fix0             = _mm256_add_ps(fix0,tx);
 600             fiy0             = _mm256_add_ps(fiy0,ty);
 601             fiz0             = _mm256_add_ps(fiz0,tz);
 602
 603             fjx0             = _mm256_add_ps(fjx0,tx);
 604             fjy0             = _mm256_add_ps(fjy0,ty);
 605             fjz0             = _mm256_add_ps(fjz0,tz);
 606
 607             }
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 614             {
 615
 616             /* Compute parameters for interactions between i and j atoms */
 617             qq10             = _mm256_mul_ps(iq1,jq0);
 618
 619             /* REACTION-FIELD ELECTROSTATICS */
 620             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 621             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 622
 623             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 624
 625             /* Update potential sum for this i atom from the interaction with this j atom. */
 626             velec            = _mm256_and_ps(velec,cutoff_mask);
 627             velec            = _mm256_andnot_ps(dummy_mask,velec);
 628             velecsum         = _mm256_add_ps(velecsum,velec);
 629
 630             fscal            = felec;
 631
 632             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 633
 634             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 635
 636             /* Calculate temporary vectorial force */
 637             tx               = _mm256_mul_ps(fscal,dx10);
 638             ty               = _mm256_mul_ps(fscal,dy10);
 639             tz               = _mm256_mul_ps(fscal,dz10);
 640
 641             /* Update vectorial force */
 642             fix1             = _mm256_add_ps(fix1,tx);
 643             fiy1             = _mm256_add_ps(fiy1,ty);
 644             fiz1             = _mm256_add_ps(fiz1,tz);
 645
 646             fjx0             = _mm256_add_ps(fjx0,tx);
 647             fjy0             = _mm256_add_ps(fjy0,ty);
 648             fjz0             = _mm256_add_ps(fjz0,tz);
 649
 650             }
 651
 652             /**************************
 653              * CALCULATE INTERACTIONS *
 654              **************************/
 655
 656             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 657             {
 658
 659             /* Compute parameters for interactions between i and j atoms */
 660             qq20             = _mm256_mul_ps(iq2,jq0);
 661
 662             /* REACTION-FIELD ELECTROSTATICS */
 663             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 664             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 665
 666             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 667
 668             /* Update potential sum for this i atom from the interaction with this j atom. */
 669             velec            = _mm256_and_ps(velec,cutoff_mask);
 670             velec            = _mm256_andnot_ps(dummy_mask,velec);
 671             velecsum         = _mm256_add_ps(velecsum,velec);
 672
 673             fscal            = felec;
 674
 675             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 676
 677             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 678
 679             /* Calculate temporary vectorial force */
 680             tx               = _mm256_mul_ps(fscal,dx20);
 681             ty               = _mm256_mul_ps(fscal,dy20);
 682             tz               = _mm256_mul_ps(fscal,dz20);
 683
 684             /* Update vectorial force */
 685             fix2             = _mm256_add_ps(fix2,tx);
 686             fiy2             = _mm256_add_ps(fiy2,ty);
 687             fiz2             = _mm256_add_ps(fiz2,tz);
 688
 689             fjx0             = _mm256_add_ps(fjx0,tx);
 690             fjy0             = _mm256_add_ps(fjy0,ty);
 691             fjz0             = _mm256_add_ps(fjz0,tz);
 692
 693             }
 694
 695             /**************************
 696              * CALCULATE INTERACTIONS *
 697              **************************/
 698
 699             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 700             {
 701
 702             /* Compute parameters for interactions between i and j atoms */
 703             qq30             = _mm256_mul_ps(iq3,jq0);
 704
 705             /* REACTION-FIELD ELECTROSTATICS */
 706             velec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_add_ps(rinv30,_mm256_mul_ps(krf,rsq30)),crf));
 707             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
 708
 709             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 710
 711             /* Update potential sum for this i atom from the interaction with this j atom. */
 712             velec            = _mm256_and_ps(velec,cutoff_mask);
 713             velec            = _mm256_andnot_ps(dummy_mask,velec);
 714             velecsum         = _mm256_add_ps(velecsum,velec);
 715
 716             fscal            = felec;
 717
 718             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 719
 720             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 721
 722             /* Calculate temporary vectorial force */
 723             tx               = _mm256_mul_ps(fscal,dx30);
 724             ty               = _mm256_mul_ps(fscal,dy30);
 725             tz               = _mm256_mul_ps(fscal,dz30);
 726
 727             /* Update vectorial force */
 728             fix3             = _mm256_add_ps(fix3,tx);
 729             fiy3             = _mm256_add_ps(fiy3,ty);
 730             fiz3             = _mm256_add_ps(fiz3,tz);
 731
 732             fjx0             = _mm256_add_ps(fjx0,tx);
 733             fjy0             = _mm256_add_ps(fjy0,ty);
 734             fjz0             = _mm256_add_ps(fjz0,tz);
 735
 736             }
 737
 738             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 739             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 740             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 741             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 742             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 743             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 744             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 745             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 746
 747             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 748
 749             /* Inner loop uses 152 flops */
 750         }
 751
 752         /* End of innermost loop */
 753
 754         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 755                                                  f+i_coord_offset,fshift+i_shift_offset);
 756
 757         ggid                        = gid[iidx];
 758         /* Update potential energies */
 759         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 760         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 761
 762         /* Increment number of inner iterations */
 763         inneriter                  += j_index_end - j_index_start;
 764
 765         /* Outer loop uses 26 flops */
 766     }
 767
 768     /* Increment number of outer iterations */
 769     outeriter        += nri;
 770
 771     /* Update outer/inner flops */
 772
 773     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*152);
 774 }
 775 /*
 776  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_single
 777  * Electrostatics interaction: ReactionField
 778  * VdW interaction:            LennardJones
 779  * Geometry:                   Water4-Particle
 780  * Calculate force/pot:        Force
 781  */
 782 void
 783 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_single
 784                     (t_nblist                    * gmx_restrict       nlist,
 785                      rvec                        * gmx_restrict          xx,
 786                      rvec                        * gmx_restrict          ff,
 787                      struct t_forcerec           * gmx_restrict          fr,
 788                      t_mdatoms                   * gmx_restrict     mdatoms,
 789                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 790                      t_nrnb                      * gmx_restrict        nrnb)
 791 {
 792     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 793      * just 0 for non-waters.
 794      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 795      * jnr indices corresponding to data put in the four positions in the SIMD register.
 796      */
 797     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 798     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 799     int              jnrA,jnrB,jnrC,jnrD;
 800     int              jnrE,jnrF,jnrG,jnrH;
 801     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 802     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 803     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 804     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 805     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 806     real             rcutoff_scalar;
 807     real             *shiftvec,*fshift,*x,*f;
 808     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 809     real             scratch[4*DIM];
 810     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 811     real *           vdwioffsetptr0;
 812     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 813     real *           vdwioffsetptr1;
 814     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 815     real *           vdwioffsetptr2;
 816     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 817     real *           vdwioffsetptr3;
 818     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 819     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 820     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 821     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 822     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 823     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 824     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 825     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 826     real             *charge;
 827     int              nvdwtype;
 828     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 829     int              *vdwtype;
 830     real             *vdwparam;
 831     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 832     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 833     __m256           dummy_mask,cutoff_mask;
 834     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 835     __m256           one     = _mm256_set1_ps(1.0);
 836     __m256           two     = _mm256_set1_ps(2.0);
 837     x                = xx[0];
 838     f                = ff[0];
 839
 840     nri              = nlist->nri;
 841     iinr             = nlist->iinr;
 842     jindex           = nlist->jindex;
 843     jjnr             = nlist->jjnr;
 844     shiftidx         = nlist->shift;
 845     gid              = nlist->gid;
 846     shiftvec         = fr->shift_vec[0];
 847     fshift           = fr->fshift[0];
 848     facel            = _mm256_set1_ps(fr->ic->epsfac);
 849     charge           = mdatoms->chargeA;
 850     krf              = _mm256_set1_ps(fr->ic->k_rf);
 851     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 852     crf              = _mm256_set1_ps(fr->ic->c_rf);
 853     nvdwtype         = fr->ntype;
 854     vdwparam         = fr->nbfp;
 855     vdwtype          = mdatoms->typeA;
 856
 857     /* Setup water-specific parameters */
 858     inr              = nlist->iinr[0];
 859     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 860     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 861     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 862     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 863
 864     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 865     rcutoff_scalar   = fr->ic->rcoulomb;
 866     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 867     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 868
 869     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 870     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 871
 872     /* Avoid stupid compiler warnings */
 873     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 874     j_coord_offsetA = 0;
 875     j_coord_offsetB = 0;
 876     j_coord_offsetC = 0;
 877     j_coord_offsetD = 0;
 878     j_coord_offsetE = 0;
 879     j_coord_offsetF = 0;
 880     j_coord_offsetG = 0;
 881     j_coord_offsetH = 0;
 882
 883     outeriter        = 0;
 884     inneriter        = 0;
 885
 886     for(iidx=0;iidx<4*DIM;iidx++)
 887     {
 888         scratch[iidx] = 0.0;
 889     }
 890
 891     /* Start outer loop over neighborlists */
 892     for(iidx=0; iidx<nri; iidx++)
 893     {
 894         /* Load shift vector for this list */
 895         i_shift_offset   = DIM*shiftidx[iidx];
 896
 897         /* Load limits for loop over neighbors */
 898         j_index_start    = jindex[iidx];
 899         j_index_end      = jindex[iidx+1];
 900
 901         /* Get outer coordinate index */
 902         inr              = iinr[iidx];
 903         i_coord_offset   = DIM*inr;
 904
 905         /* Load i particle coords and add shift vector */
 906         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 907                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 908
 909         fix0             = _mm256_setzero_ps();
 910         fiy0             = _mm256_setzero_ps();
 911         fiz0             = _mm256_setzero_ps();
 912         fix1             = _mm256_setzero_ps();
 913         fiy1             = _mm256_setzero_ps();
 914         fiz1             = _mm256_setzero_ps();
 915         fix2             = _mm256_setzero_ps();
 916         fiy2             = _mm256_setzero_ps();
 917         fiz2             = _mm256_setzero_ps();
 918         fix3             = _mm256_setzero_ps();
 919         fiy3             = _mm256_setzero_ps();
 920         fiz3             = _mm256_setzero_ps();
 921
 922         /* Start inner kernel loop */
 923         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 924         {
 925
 926             /* Get j neighbor index, and coordinate index */
 927             jnrA             = jjnr[jidx];
 928             jnrB             = jjnr[jidx+1];
 929             jnrC             = jjnr[jidx+2];
 930             jnrD             = jjnr[jidx+3];
 931             jnrE             = jjnr[jidx+4];
 932             jnrF             = jjnr[jidx+5];
 933             jnrG             = jjnr[jidx+6];
 934             jnrH             = jjnr[jidx+7];
 935             j_coord_offsetA  = DIM*jnrA;
 936             j_coord_offsetB  = DIM*jnrB;
 937             j_coord_offsetC  = DIM*jnrC;
 938             j_coord_offsetD  = DIM*jnrD;
 939             j_coord_offsetE  = DIM*jnrE;
 940             j_coord_offsetF  = DIM*jnrF;
 941             j_coord_offsetG  = DIM*jnrG;
 942             j_coord_offsetH  = DIM*jnrH;
 943
 944             /* load j atom coordinates */
 945             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 946                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 947                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 948                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 949                                                  &jx0,&jy0,&jz0);
 950
 951             /* Calculate displacement vector */
 952             dx00             = _mm256_sub_ps(ix0,jx0);
 953             dy00             = _mm256_sub_ps(iy0,jy0);
 954             dz00             = _mm256_sub_ps(iz0,jz0);
 955             dx10             = _mm256_sub_ps(ix1,jx0);
 956             dy10             = _mm256_sub_ps(iy1,jy0);
 957             dz10             = _mm256_sub_ps(iz1,jz0);
 958             dx20             = _mm256_sub_ps(ix2,jx0);
 959             dy20             = _mm256_sub_ps(iy2,jy0);
 960             dz20             = _mm256_sub_ps(iz2,jz0);
 961             dx30             = _mm256_sub_ps(ix3,jx0);
 962             dy30             = _mm256_sub_ps(iy3,jy0);
 963             dz30             = _mm256_sub_ps(iz3,jz0);
 964
 965             /* Calculate squared distance and things based on it */
 966             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 967             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 968             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 969             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 970
 971             rinv10           = avx256_invsqrt_f(rsq10);
 972             rinv20           = avx256_invsqrt_f(rsq20);
 973             rinv30           = avx256_invsqrt_f(rsq30);
 974
 975             rinvsq00         = avx256_inv_f(rsq00);
 976             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 977             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 978             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 979
 980             /* Load parameters for j particles */
 981             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 982                                                                  charge+jnrC+0,charge+jnrD+0,
 983                                                                  charge+jnrE+0,charge+jnrF+0,
 984                                                                  charge+jnrG+0,charge+jnrH+0);
 985             vdwjidx0A        = 2*vdwtype[jnrA+0];
 986             vdwjidx0B        = 2*vdwtype[jnrB+0];
 987             vdwjidx0C        = 2*vdwtype[jnrC+0];
 988             vdwjidx0D        = 2*vdwtype[jnrD+0];
 989             vdwjidx0E        = 2*vdwtype[jnrE+0];
 990             vdwjidx0F        = 2*vdwtype[jnrF+0];
 991             vdwjidx0G        = 2*vdwtype[jnrG+0];
 992             vdwjidx0H        = 2*vdwtype[jnrH+0];
 993
 994             fjx0             = _mm256_setzero_ps();
 995             fjy0             = _mm256_setzero_ps();
 996             fjz0             = _mm256_setzero_ps();
 997
 998             /**************************
 999              * CALCULATE INTERACTIONS *
1000              **************************/
1001
1002             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1003             {
1004
1005             /* Compute parameters for interactions between i and j atoms */
1006             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1007                                             vdwioffsetptr0+vdwjidx0B,
1008                                             vdwioffsetptr0+vdwjidx0C,
1009                                             vdwioffsetptr0+vdwjidx0D,
1010                                             vdwioffsetptr0+vdwjidx0E,
1011                                             vdwioffsetptr0+vdwjidx0F,
1012                                             vdwioffsetptr0+vdwjidx0G,
1013                                             vdwioffsetptr0+vdwjidx0H,
1014                                             &c6_00,&c12_00);
1015
1016             /* LENNARD-JONES DISPERSION/REPULSION */
1017
1018             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1019             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1020
1021             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1022
1023             fscal            = fvdw;
1024
1025             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm256_mul_ps(fscal,dx00);
1029             ty               = _mm256_mul_ps(fscal,dy00);
1030             tz               = _mm256_mul_ps(fscal,dz00);
1031
1032             /* Update vectorial force */
1033             fix0             = _mm256_add_ps(fix0,tx);
1034             fiy0             = _mm256_add_ps(fiy0,ty);
1035             fiz0             = _mm256_add_ps(fiz0,tz);
1036
1037             fjx0             = _mm256_add_ps(fjx0,tx);
1038             fjy0             = _mm256_add_ps(fjy0,ty);
1039             fjz0             = _mm256_add_ps(fjz0,tz);
1040
1041             }
1042
1043             /**************************
1044              * CALCULATE INTERACTIONS *
1045              **************************/
1046
1047             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1048             {
1049
1050             /* Compute parameters for interactions between i and j atoms */
1051             qq10             = _mm256_mul_ps(iq1,jq0);
1052
1053             /* REACTION-FIELD ELECTROSTATICS */
1054             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1055
1056             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1057
1058             fscal            = felec;
1059
1060             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1061
1062             /* Calculate temporary vectorial force */
1063             tx               = _mm256_mul_ps(fscal,dx10);
1064             ty               = _mm256_mul_ps(fscal,dy10);
1065             tz               = _mm256_mul_ps(fscal,dz10);
1066
1067             /* Update vectorial force */
1068             fix1             = _mm256_add_ps(fix1,tx);
1069             fiy1             = _mm256_add_ps(fiy1,ty);
1070             fiz1             = _mm256_add_ps(fiz1,tz);
1071
1072             fjx0             = _mm256_add_ps(fjx0,tx);
1073             fjy0             = _mm256_add_ps(fjy0,ty);
1074             fjz0             = _mm256_add_ps(fjz0,tz);
1075
1076             }
1077
1078             /**************************
1079              * CALCULATE INTERACTIONS *
1080              **************************/
1081
1082             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1083             {
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             qq20             = _mm256_mul_ps(iq2,jq0);
1087
1088             /* REACTION-FIELD ELECTROSTATICS */
1089             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1090
1091             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1092
1093             fscal            = felec;
1094
1095             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1096
1097             /* Calculate temporary vectorial force */
1098             tx               = _mm256_mul_ps(fscal,dx20);
1099             ty               = _mm256_mul_ps(fscal,dy20);
1100             tz               = _mm256_mul_ps(fscal,dz20);
1101
1102             /* Update vectorial force */
1103             fix2             = _mm256_add_ps(fix2,tx);
1104             fiy2             = _mm256_add_ps(fiy2,ty);
1105             fiz2             = _mm256_add_ps(fiz2,tz);
1106
1107             fjx0             = _mm256_add_ps(fjx0,tx);
1108             fjy0             = _mm256_add_ps(fjy0,ty);
1109             fjz0             = _mm256_add_ps(fjz0,tz);
1110
1111             }
1112
1113             /**************************
1114              * CALCULATE INTERACTIONS *
1115              **************************/
1116
1117             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1118             {
1119
1120             /* Compute parameters for interactions between i and j atoms */
1121             qq30             = _mm256_mul_ps(iq3,jq0);
1122
1123             /* REACTION-FIELD ELECTROSTATICS */
1124             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1125
1126             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1127
1128             fscal            = felec;
1129
1130             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1131
1132             /* Calculate temporary vectorial force */
1133             tx               = _mm256_mul_ps(fscal,dx30);
1134             ty               = _mm256_mul_ps(fscal,dy30);
1135             tz               = _mm256_mul_ps(fscal,dz30);
1136
1137             /* Update vectorial force */
1138             fix3             = _mm256_add_ps(fix3,tx);
1139             fiy3             = _mm256_add_ps(fiy3,ty);
1140             fiz3             = _mm256_add_ps(fiz3,tz);
1141
1142             fjx0             = _mm256_add_ps(fjx0,tx);
1143             fjy0             = _mm256_add_ps(fjy0,ty);
1144             fjz0             = _mm256_add_ps(fjz0,tz);
1145
1146             }
1147
1148             fjptrA             = f+j_coord_offsetA;
1149             fjptrB             = f+j_coord_offsetB;
1150             fjptrC             = f+j_coord_offsetC;
1151             fjptrD             = f+j_coord_offsetD;
1152             fjptrE             = f+j_coord_offsetE;
1153             fjptrF             = f+j_coord_offsetF;
1154             fjptrG             = f+j_coord_offsetG;
1155             fjptrH             = f+j_coord_offsetH;
1156
1157             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1158
1159             /* Inner loop uses 123 flops */
1160         }
1161
1162         if(jidx<j_index_end)
1163         {
1164
1165             /* Get j neighbor index, and coordinate index */
1166             jnrlistA         = jjnr[jidx];
1167             jnrlistB         = jjnr[jidx+1];
1168             jnrlistC         = jjnr[jidx+2];
1169             jnrlistD         = jjnr[jidx+3];
1170             jnrlistE         = jjnr[jidx+4];
1171             jnrlistF         = jjnr[jidx+5];
1172             jnrlistG         = jjnr[jidx+6];
1173             jnrlistH         = jjnr[jidx+7];
1174             /* Sign of each element will be negative for non-real atoms.
1175              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1176              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1177              */
1178             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1179                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1180
1181             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1182             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1183             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1184             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1185             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1186             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1187             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1188             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1189             j_coord_offsetA  = DIM*jnrA;
1190             j_coord_offsetB  = DIM*jnrB;
1191             j_coord_offsetC  = DIM*jnrC;
1192             j_coord_offsetD  = DIM*jnrD;
1193             j_coord_offsetE  = DIM*jnrE;
1194             j_coord_offsetF  = DIM*jnrF;
1195             j_coord_offsetG  = DIM*jnrG;
1196             j_coord_offsetH  = DIM*jnrH;
1197
1198             /* load j atom coordinates */
1199             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1200                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1201                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1202                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1203                                                  &jx0,&jy0,&jz0);
1204
1205             /* Calculate displacement vector */
1206             dx00             = _mm256_sub_ps(ix0,jx0);
1207             dy00             = _mm256_sub_ps(iy0,jy0);
1208             dz00             = _mm256_sub_ps(iz0,jz0);
1209             dx10             = _mm256_sub_ps(ix1,jx0);
1210             dy10             = _mm256_sub_ps(iy1,jy0);
1211             dz10             = _mm256_sub_ps(iz1,jz0);
1212             dx20             = _mm256_sub_ps(ix2,jx0);
1213             dy20             = _mm256_sub_ps(iy2,jy0);
1214             dz20             = _mm256_sub_ps(iz2,jz0);
1215             dx30             = _mm256_sub_ps(ix3,jx0);
1216             dy30             = _mm256_sub_ps(iy3,jy0);
1217             dz30             = _mm256_sub_ps(iz3,jz0);
1218
1219             /* Calculate squared distance and things based on it */
1220             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1221             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1222             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1223             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1224
1225             rinv10           = avx256_invsqrt_f(rsq10);
1226             rinv20           = avx256_invsqrt_f(rsq20);
1227             rinv30           = avx256_invsqrt_f(rsq30);
1228
1229             rinvsq00         = avx256_inv_f(rsq00);
1230             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1231             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1232             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1233
1234             /* Load parameters for j particles */
1235             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1236                                                                  charge+jnrC+0,charge+jnrD+0,
1237                                                                  charge+jnrE+0,charge+jnrF+0,
1238                                                                  charge+jnrG+0,charge+jnrH+0);
1239             vdwjidx0A        = 2*vdwtype[jnrA+0];
1240             vdwjidx0B        = 2*vdwtype[jnrB+0];
1241             vdwjidx0C        = 2*vdwtype[jnrC+0];
1242             vdwjidx0D        = 2*vdwtype[jnrD+0];
1243             vdwjidx0E        = 2*vdwtype[jnrE+0];
1244             vdwjidx0F        = 2*vdwtype[jnrF+0];
1245             vdwjidx0G        = 2*vdwtype[jnrG+0];
1246             vdwjidx0H        = 2*vdwtype[jnrH+0];
1247
1248             fjx0             = _mm256_setzero_ps();
1249             fjy0             = _mm256_setzero_ps();
1250             fjz0             = _mm256_setzero_ps();
1251
1252             /**************************
1253              * CALCULATE INTERACTIONS *
1254              **************************/
1255
1256             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1257             {
1258
1259             /* Compute parameters for interactions between i and j atoms */
1260             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1261                                             vdwioffsetptr0+vdwjidx0B,
1262                                             vdwioffsetptr0+vdwjidx0C,
1263                                             vdwioffsetptr0+vdwjidx0D,
1264                                             vdwioffsetptr0+vdwjidx0E,
1265                                             vdwioffsetptr0+vdwjidx0F,
1266                                             vdwioffsetptr0+vdwjidx0G,
1267                                             vdwioffsetptr0+vdwjidx0H,
1268                                             &c6_00,&c12_00);
1269
1270             /* LENNARD-JONES DISPERSION/REPULSION */
1271
1272             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1273             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1274
1275             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1276
1277             fscal            = fvdw;
1278
1279             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1280
1281             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1282
1283             /* Calculate temporary vectorial force */
1284             tx               = _mm256_mul_ps(fscal,dx00);
1285             ty               = _mm256_mul_ps(fscal,dy00);
1286             tz               = _mm256_mul_ps(fscal,dz00);
1287
1288             /* Update vectorial force */
1289             fix0             = _mm256_add_ps(fix0,tx);
1290             fiy0             = _mm256_add_ps(fiy0,ty);
1291             fiz0             = _mm256_add_ps(fiz0,tz);
1292
1293             fjx0             = _mm256_add_ps(fjx0,tx);
1294             fjy0             = _mm256_add_ps(fjy0,ty);
1295             fjz0             = _mm256_add_ps(fjz0,tz);
1296
1297             }
1298
1299             /**************************
1300              * CALCULATE INTERACTIONS *
1301              **************************/
1302
1303             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1304             {
1305
1306             /* Compute parameters for interactions between i and j atoms */
1307             qq10             = _mm256_mul_ps(iq1,jq0);
1308
1309             /* REACTION-FIELD ELECTROSTATICS */
1310             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1311
1312             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1313
1314             fscal            = felec;
1315
1316             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1317
1318             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1319
1320             /* Calculate temporary vectorial force */
1321             tx               = _mm256_mul_ps(fscal,dx10);
1322             ty               = _mm256_mul_ps(fscal,dy10);
1323             tz               = _mm256_mul_ps(fscal,dz10);
1324
1325             /* Update vectorial force */
1326             fix1             = _mm256_add_ps(fix1,tx);
1327             fiy1             = _mm256_add_ps(fiy1,ty);
1328             fiz1             = _mm256_add_ps(fiz1,tz);
1329
1330             fjx0             = _mm256_add_ps(fjx0,tx);
1331             fjy0             = _mm256_add_ps(fjy0,ty);
1332             fjz0             = _mm256_add_ps(fjz0,tz);
1333
1334             }
1335
1336             /**************************
1337              * CALCULATE INTERACTIONS *
1338              **************************/
1339
1340             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1341             {
1342
1343             /* Compute parameters for interactions between i and j atoms */
1344             qq20             = _mm256_mul_ps(iq2,jq0);
1345
1346             /* REACTION-FIELD ELECTROSTATICS */
1347             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1348
1349             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1350
1351             fscal            = felec;
1352
1353             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1354
1355             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1356
1357             /* Calculate temporary vectorial force */
1358             tx               = _mm256_mul_ps(fscal,dx20);
1359             ty               = _mm256_mul_ps(fscal,dy20);
1360             tz               = _mm256_mul_ps(fscal,dz20);
1361
1362             /* Update vectorial force */
1363             fix2             = _mm256_add_ps(fix2,tx);
1364             fiy2             = _mm256_add_ps(fiy2,ty);
1365             fiz2             = _mm256_add_ps(fiz2,tz);
1366
1367             fjx0             = _mm256_add_ps(fjx0,tx);
1368             fjy0             = _mm256_add_ps(fjy0,ty);
1369             fjz0             = _mm256_add_ps(fjz0,tz);
1370
1371             }
1372
1373             /**************************
1374              * CALCULATE INTERACTIONS *
1375              **************************/
1376
1377             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1378             {
1379
1380             /* Compute parameters for interactions between i and j atoms */
1381             qq30             = _mm256_mul_ps(iq3,jq0);
1382
1383             /* REACTION-FIELD ELECTROSTATICS */
1384             felec            = _mm256_mul_ps(qq30,_mm256_sub_ps(_mm256_mul_ps(rinv30,rinvsq30),krf2));
1385
1386             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1387
1388             fscal            = felec;
1389
1390             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1391
1392             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1393
1394             /* Calculate temporary vectorial force */
1395             tx               = _mm256_mul_ps(fscal,dx30);
1396             ty               = _mm256_mul_ps(fscal,dy30);
1397             tz               = _mm256_mul_ps(fscal,dz30);
1398
1399             /* Update vectorial force */
1400             fix3             = _mm256_add_ps(fix3,tx);
1401             fiy3             = _mm256_add_ps(fiy3,ty);
1402             fiz3             = _mm256_add_ps(fiz3,tz);
1403
1404             fjx0             = _mm256_add_ps(fjx0,tx);
1405             fjy0             = _mm256_add_ps(fjy0,ty);
1406             fjz0             = _mm256_add_ps(fjz0,tz);
1407
1408             }
1409
1410             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1411             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1412             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1413             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1414             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1415             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1416             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1417             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1418
1419             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1420
1421             /* Inner loop uses 123 flops */
1422         }
1423
1424         /* End of innermost loop */
1425
1426         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1427                                                  f+i_coord_offset,fshift+i_shift_offset);
1428
1429         /* Increment number of inner iterations */
1430         inneriter                  += j_index_end - j_index_start;
1431
1432         /* Outer loop uses 24 flops */
1433     }
1434
1435     /* Increment number of outer iterations */
1436     outeriter        += nri;
1437
1438     /* Update outer/inner flops */
1439
1440     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
1441 }