src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwLJ_GeomW4P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 107     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 108     __m256i          ewitab;
 109     __m128i          ewitab_lo,ewitab_hi;
 110     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 112     real             *ewtab;
 113     __m256           dummy_mask,cutoff_mask;
 114     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 115     __m256           one     = _mm256_set1_ps(1.0);
 116     __m256           two     = _mm256_set1_ps(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_ps(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 135     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 136     beta2            = _mm256_mul_ps(beta,beta);
 137     beta3            = _mm256_mul_ps(beta,beta2);
 138
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 146     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 147     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 148     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154     j_coord_offsetC = 0;
 155     j_coord_offsetD = 0;
 156     j_coord_offsetE = 0;
 157     j_coord_offsetF = 0;
 158     j_coord_offsetG = 0;
 159     j_coord_offsetH = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     for(iidx=0;iidx<4*DIM;iidx++)
 165     {
 166         scratch[iidx] = 0.0;
 167     }
 168
 169     /* Start outer loop over neighborlists */
 170     for(iidx=0; iidx<nri; iidx++)
 171     {
 172         /* Load shift vector for this list */
 173         i_shift_offset   = DIM*shiftidx[iidx];
 174
 175         /* Load limits for loop over neighbors */
 176         j_index_start    = jindex[iidx];
 177         j_index_end      = jindex[iidx+1];
 178
 179         /* Get outer coordinate index */
 180         inr              = iinr[iidx];
 181         i_coord_offset   = DIM*inr;
 182
 183         /* Load i particle coords and add shift vector */
 184         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 185                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 186
 187         fix0             = _mm256_setzero_ps();
 188         fiy0             = _mm256_setzero_ps();
 189         fiz0             = _mm256_setzero_ps();
 190         fix1             = _mm256_setzero_ps();
 191         fiy1             = _mm256_setzero_ps();
 192         fiz1             = _mm256_setzero_ps();
 193         fix2             = _mm256_setzero_ps();
 194         fiy2             = _mm256_setzero_ps();
 195         fiz2             = _mm256_setzero_ps();
 196         fix3             = _mm256_setzero_ps();
 197         fiy3             = _mm256_setzero_ps();
 198         fiz3             = _mm256_setzero_ps();
 199
 200         /* Reset potential sums */
 201         velecsum         = _mm256_setzero_ps();
 202         vvdwsum          = _mm256_setzero_ps();
 203
 204         /* Start inner kernel loop */
 205         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 206         {
 207
 208             /* Get j neighbor index, and coordinate index */
 209             jnrA             = jjnr[jidx];
 210             jnrB             = jjnr[jidx+1];
 211             jnrC             = jjnr[jidx+2];
 212             jnrD             = jjnr[jidx+3];
 213             jnrE             = jjnr[jidx+4];
 214             jnrF             = jjnr[jidx+5];
 215             jnrG             = jjnr[jidx+6];
 216             jnrH             = jjnr[jidx+7];
 217             j_coord_offsetA  = DIM*jnrA;
 218             j_coord_offsetB  = DIM*jnrB;
 219             j_coord_offsetC  = DIM*jnrC;
 220             j_coord_offsetD  = DIM*jnrD;
 221             j_coord_offsetE  = DIM*jnrE;
 222             j_coord_offsetF  = DIM*jnrF;
 223             j_coord_offsetG  = DIM*jnrG;
 224             j_coord_offsetH  = DIM*jnrH;
 225
 226             /* load j atom coordinates */
 227             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 228                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 229                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 230                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 231                                                  &jx0,&jy0,&jz0);
 232
 233             /* Calculate displacement vector */
 234             dx00             = _mm256_sub_ps(ix0,jx0);
 235             dy00             = _mm256_sub_ps(iy0,jy0);
 236             dz00             = _mm256_sub_ps(iz0,jz0);
 237             dx10             = _mm256_sub_ps(ix1,jx0);
 238             dy10             = _mm256_sub_ps(iy1,jy0);
 239             dz10             = _mm256_sub_ps(iz1,jz0);
 240             dx20             = _mm256_sub_ps(ix2,jx0);
 241             dy20             = _mm256_sub_ps(iy2,jy0);
 242             dz20             = _mm256_sub_ps(iz2,jz0);
 243             dx30             = _mm256_sub_ps(ix3,jx0);
 244             dy30             = _mm256_sub_ps(iy3,jy0);
 245             dz30             = _mm256_sub_ps(iz3,jz0);
 246
 247             /* Calculate squared distance and things based on it */
 248             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 249             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 250             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 251             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 252
 253             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 254             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 255             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 256
 257             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 258             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 259             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 260             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 261
 262             /* Load parameters for j particles */
 263             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 264                                                                  charge+jnrC+0,charge+jnrD+0,
 265                                                                  charge+jnrE+0,charge+jnrF+0,
 266                                                                  charge+jnrG+0,charge+jnrH+0);
 267             vdwjidx0A        = 2*vdwtype[jnrA+0];
 268             vdwjidx0B        = 2*vdwtype[jnrB+0];
 269             vdwjidx0C        = 2*vdwtype[jnrC+0];
 270             vdwjidx0D        = 2*vdwtype[jnrD+0];
 271             vdwjidx0E        = 2*vdwtype[jnrE+0];
 272             vdwjidx0F        = 2*vdwtype[jnrF+0];
 273             vdwjidx0G        = 2*vdwtype[jnrG+0];
 274             vdwjidx0H        = 2*vdwtype[jnrH+0];
 275
 276             fjx0             = _mm256_setzero_ps();
 277             fjy0             = _mm256_setzero_ps();
 278             fjz0             = _mm256_setzero_ps();
 279
 280             /**************************
 281              * CALCULATE INTERACTIONS *
 282              **************************/
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 286                                             vdwioffsetptr0+vdwjidx0B,
 287                                             vdwioffsetptr0+vdwjidx0C,
 288                                             vdwioffsetptr0+vdwjidx0D,
 289                                             vdwioffsetptr0+vdwjidx0E,
 290                                             vdwioffsetptr0+vdwjidx0F,
 291                                             vdwioffsetptr0+vdwjidx0G,
 292                                             vdwioffsetptr0+vdwjidx0H,
 293                                             &c6_00,&c12_00);
 294
 295             /* LENNARD-JONES DISPERSION/REPULSION */
 296
 297             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 298             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 299             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 300             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 301             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 305
 306             fscal            = fvdw;
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm256_mul_ps(fscal,dx00);
 310             ty               = _mm256_mul_ps(fscal,dy00);
 311             tz               = _mm256_mul_ps(fscal,dz00);
 312
 313             /* Update vectorial force */
 314             fix0             = _mm256_add_ps(fix0,tx);
 315             fiy0             = _mm256_add_ps(fiy0,ty);
 316             fiz0             = _mm256_add_ps(fiz0,tz);
 317
 318             fjx0             = _mm256_add_ps(fjx0,tx);
 319             fjy0             = _mm256_add_ps(fjy0,ty);
 320             fjz0             = _mm256_add_ps(fjz0,tz);
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             r10              = _mm256_mul_ps(rsq10,rinv10);
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq10             = _mm256_mul_ps(iq1,jq0);
 330
 331             /* EWALD ELECTROSTATICS */
 332
 333             /* Analytical PME correction */
 334             zeta2            = _mm256_mul_ps(beta2,rsq10);
 335             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 336             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 337             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 338             felec            = _mm256_mul_ps(qq10,felec);
 339             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 340             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 341             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 342             velec            = _mm256_mul_ps(qq10,velec);
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm256_add_ps(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_ps(fscal,dx10);
 351             ty               = _mm256_mul_ps(fscal,dy10);
 352             tz               = _mm256_mul_ps(fscal,dz10);
 353
 354             /* Update vectorial force */
 355             fix1             = _mm256_add_ps(fix1,tx);
 356             fiy1             = _mm256_add_ps(fiy1,ty);
 357             fiz1             = _mm256_add_ps(fiz1,tz);
 358
 359             fjx0             = _mm256_add_ps(fjx0,tx);
 360             fjy0             = _mm256_add_ps(fjy0,ty);
 361             fjz0             = _mm256_add_ps(fjz0,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r20              = _mm256_mul_ps(rsq20,rinv20);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq20             = _mm256_mul_ps(iq2,jq0);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Analytical PME correction */
 375             zeta2            = _mm256_mul_ps(beta2,rsq20);
 376             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 377             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 378             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 379             felec            = _mm256_mul_ps(qq20,felec);
 380             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 381             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 382             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 383             velec            = _mm256_mul_ps(qq20,velec);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velecsum         = _mm256_add_ps(velecsum,velec);
 387
 388             fscal            = felec;
 389
 390             /* Calculate temporary vectorial force */
 391             tx               = _mm256_mul_ps(fscal,dx20);
 392             ty               = _mm256_mul_ps(fscal,dy20);
 393             tz               = _mm256_mul_ps(fscal,dz20);
 394
 395             /* Update vectorial force */
 396             fix2             = _mm256_add_ps(fix2,tx);
 397             fiy2             = _mm256_add_ps(fiy2,ty);
 398             fiz2             = _mm256_add_ps(fiz2,tz);
 399
 400             fjx0             = _mm256_add_ps(fjx0,tx);
 401             fjy0             = _mm256_add_ps(fjy0,ty);
 402             fjz0             = _mm256_add_ps(fjz0,tz);
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             r30              = _mm256_mul_ps(rsq30,rinv30);
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq30             = _mm256_mul_ps(iq3,jq0);
 412
 413             /* EWALD ELECTROSTATICS */
 414
 415             /* Analytical PME correction */
 416             zeta2            = _mm256_mul_ps(beta2,rsq30);
 417             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 418             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 419             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 420             felec            = _mm256_mul_ps(qq30,felec);
 421             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 422             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 423             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 424             velec            = _mm256_mul_ps(qq30,velec);
 425
 426             /* Update potential sum for this i atom from the interaction with this j atom. */
 427             velecsum         = _mm256_add_ps(velecsum,velec);
 428
 429             fscal            = felec;
 430
 431             /* Calculate temporary vectorial force */
 432             tx               = _mm256_mul_ps(fscal,dx30);
 433             ty               = _mm256_mul_ps(fscal,dy30);
 434             tz               = _mm256_mul_ps(fscal,dz30);
 435
 436             /* Update vectorial force */
 437             fix3             = _mm256_add_ps(fix3,tx);
 438             fiy3             = _mm256_add_ps(fiy3,ty);
 439             fiz3             = _mm256_add_ps(fiz3,tz);
 440
 441             fjx0             = _mm256_add_ps(fjx0,tx);
 442             fjy0             = _mm256_add_ps(fjy0,ty);
 443             fjz0             = _mm256_add_ps(fjz0,tz);
 444
 445             fjptrA             = f+j_coord_offsetA;
 446             fjptrB             = f+j_coord_offsetB;
 447             fjptrC             = f+j_coord_offsetC;
 448             fjptrD             = f+j_coord_offsetD;
 449             fjptrE             = f+j_coord_offsetE;
 450             fjptrF             = f+j_coord_offsetF;
 451             fjptrG             = f+j_coord_offsetG;
 452             fjptrH             = f+j_coord_offsetH;
 453
 454             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 455
 456             /* Inner loop uses 287 flops */
 457         }
 458
 459         if(jidx<j_index_end)
 460         {
 461
 462             /* Get j neighbor index, and coordinate index */
 463             jnrlistA         = jjnr[jidx];
 464             jnrlistB         = jjnr[jidx+1];
 465             jnrlistC         = jjnr[jidx+2];
 466             jnrlistD         = jjnr[jidx+3];
 467             jnrlistE         = jjnr[jidx+4];
 468             jnrlistF         = jjnr[jidx+5];
 469             jnrlistG         = jjnr[jidx+6];
 470             jnrlistH         = jjnr[jidx+7];
 471             /* Sign of each element will be negative for non-real atoms.
 472              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 473              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 474              */
 475             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 476                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 477
 478             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 479             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 480             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 481             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 482             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 483             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 484             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 485             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 486             j_coord_offsetA  = DIM*jnrA;
 487             j_coord_offsetB  = DIM*jnrB;
 488             j_coord_offsetC  = DIM*jnrC;
 489             j_coord_offsetD  = DIM*jnrD;
 490             j_coord_offsetE  = DIM*jnrE;
 491             j_coord_offsetF  = DIM*jnrF;
 492             j_coord_offsetG  = DIM*jnrG;
 493             j_coord_offsetH  = DIM*jnrH;
 494
 495             /* load j atom coordinates */
 496             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 497                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 498                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 499                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 500                                                  &jx0,&jy0,&jz0);
 501
 502             /* Calculate displacement vector */
 503             dx00             = _mm256_sub_ps(ix0,jx0);
 504             dy00             = _mm256_sub_ps(iy0,jy0);
 505             dz00             = _mm256_sub_ps(iz0,jz0);
 506             dx10             = _mm256_sub_ps(ix1,jx0);
 507             dy10             = _mm256_sub_ps(iy1,jy0);
 508             dz10             = _mm256_sub_ps(iz1,jz0);
 509             dx20             = _mm256_sub_ps(ix2,jx0);
 510             dy20             = _mm256_sub_ps(iy2,jy0);
 511             dz20             = _mm256_sub_ps(iz2,jz0);
 512             dx30             = _mm256_sub_ps(ix3,jx0);
 513             dy30             = _mm256_sub_ps(iy3,jy0);
 514             dz30             = _mm256_sub_ps(iz3,jz0);
 515
 516             /* Calculate squared distance and things based on it */
 517             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 518             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 519             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 520             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 521
 522             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 523             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 524             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 525
 526             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 527             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 528             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 529             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 530
 531             /* Load parameters for j particles */
 532             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 533                                                                  charge+jnrC+0,charge+jnrD+0,
 534                                                                  charge+jnrE+0,charge+jnrF+0,
 535                                                                  charge+jnrG+0,charge+jnrH+0);
 536             vdwjidx0A        = 2*vdwtype[jnrA+0];
 537             vdwjidx0B        = 2*vdwtype[jnrB+0];
 538             vdwjidx0C        = 2*vdwtype[jnrC+0];
 539             vdwjidx0D        = 2*vdwtype[jnrD+0];
 540             vdwjidx0E        = 2*vdwtype[jnrE+0];
 541             vdwjidx0F        = 2*vdwtype[jnrF+0];
 542             vdwjidx0G        = 2*vdwtype[jnrG+0];
 543             vdwjidx0H        = 2*vdwtype[jnrH+0];
 544
 545             fjx0             = _mm256_setzero_ps();
 546             fjy0             = _mm256_setzero_ps();
 547             fjz0             = _mm256_setzero_ps();
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             /* Compute parameters for interactions between i and j atoms */
 554             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 555                                             vdwioffsetptr0+vdwjidx0B,
 556                                             vdwioffsetptr0+vdwjidx0C,
 557                                             vdwioffsetptr0+vdwjidx0D,
 558                                             vdwioffsetptr0+vdwjidx0E,
 559                                             vdwioffsetptr0+vdwjidx0F,
 560                                             vdwioffsetptr0+vdwjidx0G,
 561                                             vdwioffsetptr0+vdwjidx0H,
 562                                             &c6_00,&c12_00);
 563
 564             /* LENNARD-JONES DISPERSION/REPULSION */
 565
 566             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 567             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 568             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 569             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 570             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 571
 572             /* Update potential sum for this i atom from the interaction with this j atom. */
 573             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 574             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 575
 576             fscal            = fvdw;
 577
 578             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 579
 580             /* Calculate temporary vectorial force */
 581             tx               = _mm256_mul_ps(fscal,dx00);
 582             ty               = _mm256_mul_ps(fscal,dy00);
 583             tz               = _mm256_mul_ps(fscal,dz00);
 584
 585             /* Update vectorial force */
 586             fix0             = _mm256_add_ps(fix0,tx);
 587             fiy0             = _mm256_add_ps(fiy0,ty);
 588             fiz0             = _mm256_add_ps(fiz0,tz);
 589
 590             fjx0             = _mm256_add_ps(fjx0,tx);
 591             fjy0             = _mm256_add_ps(fjy0,ty);
 592             fjz0             = _mm256_add_ps(fjz0,tz);
 593
 594             /**************************
 595              * CALCULATE INTERACTIONS *
 596              **************************/
 597
 598             r10              = _mm256_mul_ps(rsq10,rinv10);
 599             r10              = _mm256_andnot_ps(dummy_mask,r10);
 600
 601             /* Compute parameters for interactions between i and j atoms */
 602             qq10             = _mm256_mul_ps(iq1,jq0);
 603
 604             /* EWALD ELECTROSTATICS */
 605
 606             /* Analytical PME correction */
 607             zeta2            = _mm256_mul_ps(beta2,rsq10);
 608             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 609             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 610             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 611             felec            = _mm256_mul_ps(qq10,felec);
 612             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 613             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 614             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 615             velec            = _mm256_mul_ps(qq10,velec);
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             velec            = _mm256_andnot_ps(dummy_mask,velec);
 619             velecsum         = _mm256_add_ps(velecsum,velec);
 620
 621             fscal            = felec;
 622
 623             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm256_mul_ps(fscal,dx10);
 627             ty               = _mm256_mul_ps(fscal,dy10);
 628             tz               = _mm256_mul_ps(fscal,dz10);
 629
 630             /* Update vectorial force */
 631             fix1             = _mm256_add_ps(fix1,tx);
 632             fiy1             = _mm256_add_ps(fiy1,ty);
 633             fiz1             = _mm256_add_ps(fiz1,tz);
 634
 635             fjx0             = _mm256_add_ps(fjx0,tx);
 636             fjy0             = _mm256_add_ps(fjy0,ty);
 637             fjz0             = _mm256_add_ps(fjz0,tz);
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             r20              = _mm256_mul_ps(rsq20,rinv20);
 644             r20              = _mm256_andnot_ps(dummy_mask,r20);
 645
 646             /* Compute parameters for interactions between i and j atoms */
 647             qq20             = _mm256_mul_ps(iq2,jq0);
 648
 649             /* EWALD ELECTROSTATICS */
 650
 651             /* Analytical PME correction */
 652             zeta2            = _mm256_mul_ps(beta2,rsq20);
 653             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 654             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 655             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 656             felec            = _mm256_mul_ps(qq20,felec);
 657             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 658             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 659             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 660             velec            = _mm256_mul_ps(qq20,velec);
 661
 662             /* Update potential sum for this i atom from the interaction with this j atom. */
 663             velec            = _mm256_andnot_ps(dummy_mask,velec);
 664             velecsum         = _mm256_add_ps(velecsum,velec);
 665
 666             fscal            = felec;
 667
 668             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 669
 670             /* Calculate temporary vectorial force */
 671             tx               = _mm256_mul_ps(fscal,dx20);
 672             ty               = _mm256_mul_ps(fscal,dy20);
 673             tz               = _mm256_mul_ps(fscal,dz20);
 674
 675             /* Update vectorial force */
 676             fix2             = _mm256_add_ps(fix2,tx);
 677             fiy2             = _mm256_add_ps(fiy2,ty);
 678             fiz2             = _mm256_add_ps(fiz2,tz);
 679
 680             fjx0             = _mm256_add_ps(fjx0,tx);
 681             fjy0             = _mm256_add_ps(fjy0,ty);
 682             fjz0             = _mm256_add_ps(fjz0,tz);
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             r30              = _mm256_mul_ps(rsq30,rinv30);
 689             r30              = _mm256_andnot_ps(dummy_mask,r30);
 690
 691             /* Compute parameters for interactions between i and j atoms */
 692             qq30             = _mm256_mul_ps(iq3,jq0);
 693
 694             /* EWALD ELECTROSTATICS */
 695
 696             /* Analytical PME correction */
 697             zeta2            = _mm256_mul_ps(beta2,rsq30);
 698             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 699             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 700             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 701             felec            = _mm256_mul_ps(qq30,felec);
 702             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 703             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 704             velec            = _mm256_sub_ps(rinv30,pmecorrV);
 705             velec            = _mm256_mul_ps(qq30,velec);
 706
 707             /* Update potential sum for this i atom from the interaction with this j atom. */
 708             velec            = _mm256_andnot_ps(dummy_mask,velec);
 709             velecsum         = _mm256_add_ps(velecsum,velec);
 710
 711             fscal            = felec;
 712
 713             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 714
 715             /* Calculate temporary vectorial force */
 716             tx               = _mm256_mul_ps(fscal,dx30);
 717             ty               = _mm256_mul_ps(fscal,dy30);
 718             tz               = _mm256_mul_ps(fscal,dz30);
 719
 720             /* Update vectorial force */
 721             fix3             = _mm256_add_ps(fix3,tx);
 722             fiy3             = _mm256_add_ps(fiy3,ty);
 723             fiz3             = _mm256_add_ps(fiz3,tz);
 724
 725             fjx0             = _mm256_add_ps(fjx0,tx);
 726             fjy0             = _mm256_add_ps(fjy0,ty);
 727             fjz0             = _mm256_add_ps(fjz0,tz);
 728
 729             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 730             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 731             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 732             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 733             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 734             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 735             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 736             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 737
 738             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 739
 740             /* Inner loop uses 290 flops */
 741         }
 742
 743         /* End of innermost loop */
 744
 745         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 746                                                  f+i_coord_offset,fshift+i_shift_offset);
 747
 748         ggid                        = gid[iidx];
 749         /* Update potential energies */
 750         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 751         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 752
 753         /* Increment number of inner iterations */
 754         inneriter                  += j_index_end - j_index_start;
 755
 756         /* Outer loop uses 26 flops */
 757     }
 758
 759     /* Increment number of outer iterations */
 760     outeriter        += nri;
 761
 762     /* Update outer/inner flops */
 763
 764     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*290);
 765 }
 766 /*
 767  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_256_single
 768  * Electrostatics interaction: Ewald
 769  * VdW interaction:            LennardJones
 770  * Geometry:                   Water4-Particle
 771  * Calculate force/pot:        Force
 772  */
 773 void
 774 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_256_single
 775                     (t_nblist                    * gmx_restrict       nlist,
 776                      rvec                        * gmx_restrict          xx,
 777                      rvec                        * gmx_restrict          ff,
 778                      t_forcerec                  * gmx_restrict          fr,
 779                      t_mdatoms                   * gmx_restrict     mdatoms,
 780                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 781                      t_nrnb                      * gmx_restrict        nrnb)
 782 {
 783     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 784      * just 0 for non-waters.
 785      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 786      * jnr indices corresponding to data put in the four positions in the SIMD register.
 787      */
 788     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 789     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 790     int              jnrA,jnrB,jnrC,jnrD;
 791     int              jnrE,jnrF,jnrG,jnrH;
 792     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 793     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 794     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 795     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 796     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 797     real             rcutoff_scalar;
 798     real             *shiftvec,*fshift,*x,*f;
 799     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 800     real             scratch[4*DIM];
 801     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 802     real *           vdwioffsetptr0;
 803     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 804     real *           vdwioffsetptr1;
 805     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 806     real *           vdwioffsetptr2;
 807     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 808     real *           vdwioffsetptr3;
 809     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 810     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 811     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 812     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 813     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 814     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 815     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 816     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 817     real             *charge;
 818     int              nvdwtype;
 819     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 820     int              *vdwtype;
 821     real             *vdwparam;
 822     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 823     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 824     __m256i          ewitab;
 825     __m128i          ewitab_lo,ewitab_hi;
 826     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 827     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 828     real             *ewtab;
 829     __m256           dummy_mask,cutoff_mask;
 830     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 831     __m256           one     = _mm256_set1_ps(1.0);
 832     __m256           two     = _mm256_set1_ps(2.0);
 833     x                = xx[0];
 834     f                = ff[0];
 835
 836     nri              = nlist->nri;
 837     iinr             = nlist->iinr;
 838     jindex           = nlist->jindex;
 839     jjnr             = nlist->jjnr;
 840     shiftidx         = nlist->shift;
 841     gid              = nlist->gid;
 842     shiftvec         = fr->shift_vec[0];
 843     fshift           = fr->fshift[0];
 844     facel            = _mm256_set1_ps(fr->epsfac);
 845     charge           = mdatoms->chargeA;
 846     nvdwtype         = fr->ntype;
 847     vdwparam         = fr->nbfp;
 848     vdwtype          = mdatoms->typeA;
 849
 850     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 851     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 852     beta2            = _mm256_mul_ps(beta,beta);
 853     beta3            = _mm256_mul_ps(beta,beta2);
 854
 855     ewtab            = fr->ic->tabq_coul_F;
 856     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 857     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 858
 859     /* Setup water-specific parameters */
 860     inr              = nlist->iinr[0];
 861     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 862     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 863     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 864     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 865
 866     /* Avoid stupid compiler warnings */
 867     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 868     j_coord_offsetA = 0;
 869     j_coord_offsetB = 0;
 870     j_coord_offsetC = 0;
 871     j_coord_offsetD = 0;
 872     j_coord_offsetE = 0;
 873     j_coord_offsetF = 0;
 874     j_coord_offsetG = 0;
 875     j_coord_offsetH = 0;
 876
 877     outeriter        = 0;
 878     inneriter        = 0;
 879
 880     for(iidx=0;iidx<4*DIM;iidx++)
 881     {
 882         scratch[iidx] = 0.0;
 883     }
 884
 885     /* Start outer loop over neighborlists */
 886     for(iidx=0; iidx<nri; iidx++)
 887     {
 888         /* Load shift vector for this list */
 889         i_shift_offset   = DIM*shiftidx[iidx];
 890
 891         /* Load limits for loop over neighbors */
 892         j_index_start    = jindex[iidx];
 893         j_index_end      = jindex[iidx+1];
 894
 895         /* Get outer coordinate index */
 896         inr              = iinr[iidx];
 897         i_coord_offset   = DIM*inr;
 898
 899         /* Load i particle coords and add shift vector */
 900         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 901                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 902
 903         fix0             = _mm256_setzero_ps();
 904         fiy0             = _mm256_setzero_ps();
 905         fiz0             = _mm256_setzero_ps();
 906         fix1             = _mm256_setzero_ps();
 907         fiy1             = _mm256_setzero_ps();
 908         fiz1             = _mm256_setzero_ps();
 909         fix2             = _mm256_setzero_ps();
 910         fiy2             = _mm256_setzero_ps();
 911         fiz2             = _mm256_setzero_ps();
 912         fix3             = _mm256_setzero_ps();
 913         fiy3             = _mm256_setzero_ps();
 914         fiz3             = _mm256_setzero_ps();
 915
 916         /* Start inner kernel loop */
 917         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 918         {
 919
 920             /* Get j neighbor index, and coordinate index */
 921             jnrA             = jjnr[jidx];
 922             jnrB             = jjnr[jidx+1];
 923             jnrC             = jjnr[jidx+2];
 924             jnrD             = jjnr[jidx+3];
 925             jnrE             = jjnr[jidx+4];
 926             jnrF             = jjnr[jidx+5];
 927             jnrG             = jjnr[jidx+6];
 928             jnrH             = jjnr[jidx+7];
 929             j_coord_offsetA  = DIM*jnrA;
 930             j_coord_offsetB  = DIM*jnrB;
 931             j_coord_offsetC  = DIM*jnrC;
 932             j_coord_offsetD  = DIM*jnrD;
 933             j_coord_offsetE  = DIM*jnrE;
 934             j_coord_offsetF  = DIM*jnrF;
 935             j_coord_offsetG  = DIM*jnrG;
 936             j_coord_offsetH  = DIM*jnrH;
 937
 938             /* load j atom coordinates */
 939             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 940                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 941                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 942                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 943                                                  &jx0,&jy0,&jz0);
 944
 945             /* Calculate displacement vector */
 946             dx00             = _mm256_sub_ps(ix0,jx0);
 947             dy00             = _mm256_sub_ps(iy0,jy0);
 948             dz00             = _mm256_sub_ps(iz0,jz0);
 949             dx10             = _mm256_sub_ps(ix1,jx0);
 950             dy10             = _mm256_sub_ps(iy1,jy0);
 951             dz10             = _mm256_sub_ps(iz1,jz0);
 952             dx20             = _mm256_sub_ps(ix2,jx0);
 953             dy20             = _mm256_sub_ps(iy2,jy0);
 954             dz20             = _mm256_sub_ps(iz2,jz0);
 955             dx30             = _mm256_sub_ps(ix3,jx0);
 956             dy30             = _mm256_sub_ps(iy3,jy0);
 957             dz30             = _mm256_sub_ps(iz3,jz0);
 958
 959             /* Calculate squared distance and things based on it */
 960             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 961             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 962             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 963             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 964
 965             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 966             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 967             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
 968
 969             rinvsq00         = gmx_mm256_inv_ps(rsq00);
 970             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 971             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 972             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 973
 974             /* Load parameters for j particles */
 975             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 976                                                                  charge+jnrC+0,charge+jnrD+0,
 977                                                                  charge+jnrE+0,charge+jnrF+0,
 978                                                                  charge+jnrG+0,charge+jnrH+0);
 979             vdwjidx0A        = 2*vdwtype[jnrA+0];
 980             vdwjidx0B        = 2*vdwtype[jnrB+0];
 981             vdwjidx0C        = 2*vdwtype[jnrC+0];
 982             vdwjidx0D        = 2*vdwtype[jnrD+0];
 983             vdwjidx0E        = 2*vdwtype[jnrE+0];
 984             vdwjidx0F        = 2*vdwtype[jnrF+0];
 985             vdwjidx0G        = 2*vdwtype[jnrG+0];
 986             vdwjidx0H        = 2*vdwtype[jnrH+0];
 987
 988             fjx0             = _mm256_setzero_ps();
 989             fjy0             = _mm256_setzero_ps();
 990             fjz0             = _mm256_setzero_ps();
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             /* Compute parameters for interactions between i and j atoms */
 997             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 998                                             vdwioffsetptr0+vdwjidx0B,
 999                                             vdwioffsetptr0+vdwjidx0C,
1000                                             vdwioffsetptr0+vdwjidx0D,
1001                                             vdwioffsetptr0+vdwjidx0E,
1002                                             vdwioffsetptr0+vdwjidx0F,
1003                                             vdwioffsetptr0+vdwjidx0G,
1004                                             vdwioffsetptr0+vdwjidx0H,
1005                                             &c6_00,&c12_00);
1006
1007             /* LENNARD-JONES DISPERSION/REPULSION */
1008
1009             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1010             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1011
1012             fscal            = fvdw;
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm256_mul_ps(fscal,dx00);
1016             ty               = _mm256_mul_ps(fscal,dy00);
1017             tz               = _mm256_mul_ps(fscal,dz00);
1018
1019             /* Update vectorial force */
1020             fix0             = _mm256_add_ps(fix0,tx);
1021             fiy0             = _mm256_add_ps(fiy0,ty);
1022             fiz0             = _mm256_add_ps(fiz0,tz);
1023
1024             fjx0             = _mm256_add_ps(fjx0,tx);
1025             fjy0             = _mm256_add_ps(fjy0,ty);
1026             fjz0             = _mm256_add_ps(fjz0,tz);
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             r10              = _mm256_mul_ps(rsq10,rinv10);
1033
1034             /* Compute parameters for interactions between i and j atoms */
1035             qq10             = _mm256_mul_ps(iq1,jq0);
1036
1037             /* EWALD ELECTROSTATICS */
1038
1039             /* Analytical PME correction */
1040             zeta2            = _mm256_mul_ps(beta2,rsq10);
1041             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1042             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1043             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1044             felec            = _mm256_mul_ps(qq10,felec);
1045
1046             fscal            = felec;
1047
1048             /* Calculate temporary vectorial force */
1049             tx               = _mm256_mul_ps(fscal,dx10);
1050             ty               = _mm256_mul_ps(fscal,dy10);
1051             tz               = _mm256_mul_ps(fscal,dz10);
1052
1053             /* Update vectorial force */
1054             fix1             = _mm256_add_ps(fix1,tx);
1055             fiy1             = _mm256_add_ps(fiy1,ty);
1056             fiz1             = _mm256_add_ps(fiz1,tz);
1057
1058             fjx0             = _mm256_add_ps(fjx0,tx);
1059             fjy0             = _mm256_add_ps(fjy0,ty);
1060             fjz0             = _mm256_add_ps(fjz0,tz);
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             r20              = _mm256_mul_ps(rsq20,rinv20);
1067
1068             /* Compute parameters for interactions between i and j atoms */
1069             qq20             = _mm256_mul_ps(iq2,jq0);
1070
1071             /* EWALD ELECTROSTATICS */
1072
1073             /* Analytical PME correction */
1074             zeta2            = _mm256_mul_ps(beta2,rsq20);
1075             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1076             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1077             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1078             felec            = _mm256_mul_ps(qq20,felec);
1079
1080             fscal            = felec;
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm256_mul_ps(fscal,dx20);
1084             ty               = _mm256_mul_ps(fscal,dy20);
1085             tz               = _mm256_mul_ps(fscal,dz20);
1086
1087             /* Update vectorial force */
1088             fix2             = _mm256_add_ps(fix2,tx);
1089             fiy2             = _mm256_add_ps(fiy2,ty);
1090             fiz2             = _mm256_add_ps(fiz2,tz);
1091
1092             fjx0             = _mm256_add_ps(fjx0,tx);
1093             fjy0             = _mm256_add_ps(fjy0,ty);
1094             fjz0             = _mm256_add_ps(fjz0,tz);
1095
1096             /**************************
1097              * CALCULATE INTERACTIONS *
1098              **************************/
1099
1100             r30              = _mm256_mul_ps(rsq30,rinv30);
1101
1102             /* Compute parameters for interactions between i and j atoms */
1103             qq30             = _mm256_mul_ps(iq3,jq0);
1104
1105             /* EWALD ELECTROSTATICS */
1106
1107             /* Analytical PME correction */
1108             zeta2            = _mm256_mul_ps(beta2,rsq30);
1109             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1110             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1111             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1112             felec            = _mm256_mul_ps(qq30,felec);
1113
1114             fscal            = felec;
1115
1116             /* Calculate temporary vectorial force */
1117             tx               = _mm256_mul_ps(fscal,dx30);
1118             ty               = _mm256_mul_ps(fscal,dy30);
1119             tz               = _mm256_mul_ps(fscal,dz30);
1120
1121             /* Update vectorial force */
1122             fix3             = _mm256_add_ps(fix3,tx);
1123             fiy3             = _mm256_add_ps(fiy3,ty);
1124             fiz3             = _mm256_add_ps(fiz3,tz);
1125
1126             fjx0             = _mm256_add_ps(fjx0,tx);
1127             fjy0             = _mm256_add_ps(fjy0,ty);
1128             fjz0             = _mm256_add_ps(fjz0,tz);
1129
1130             fjptrA             = f+j_coord_offsetA;
1131             fjptrB             = f+j_coord_offsetB;
1132             fjptrC             = f+j_coord_offsetC;
1133             fjptrD             = f+j_coord_offsetD;
1134             fjptrE             = f+j_coord_offsetE;
1135             fjptrF             = f+j_coord_offsetF;
1136             fjptrG             = f+j_coord_offsetG;
1137             fjptrH             = f+j_coord_offsetH;
1138
1139             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1140
1141             /* Inner loop uses 198 flops */
1142         }
1143
1144         if(jidx<j_index_end)
1145         {
1146
1147             /* Get j neighbor index, and coordinate index */
1148             jnrlistA         = jjnr[jidx];
1149             jnrlistB         = jjnr[jidx+1];
1150             jnrlistC         = jjnr[jidx+2];
1151             jnrlistD         = jjnr[jidx+3];
1152             jnrlistE         = jjnr[jidx+4];
1153             jnrlistF         = jjnr[jidx+5];
1154             jnrlistG         = jjnr[jidx+6];
1155             jnrlistH         = jjnr[jidx+7];
1156             /* Sign of each element will be negative for non-real atoms.
1157              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1158              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1159              */
1160             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1161                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1162
1163             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1164             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1165             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1166             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1167             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1168             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1169             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1170             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1171             j_coord_offsetA  = DIM*jnrA;
1172             j_coord_offsetB  = DIM*jnrB;
1173             j_coord_offsetC  = DIM*jnrC;
1174             j_coord_offsetD  = DIM*jnrD;
1175             j_coord_offsetE  = DIM*jnrE;
1176             j_coord_offsetF  = DIM*jnrF;
1177             j_coord_offsetG  = DIM*jnrG;
1178             j_coord_offsetH  = DIM*jnrH;
1179
1180             /* load j atom coordinates */
1181             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1182                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1183                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1184                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1185                                                  &jx0,&jy0,&jz0);
1186
1187             /* Calculate displacement vector */
1188             dx00             = _mm256_sub_ps(ix0,jx0);
1189             dy00             = _mm256_sub_ps(iy0,jy0);
1190             dz00             = _mm256_sub_ps(iz0,jz0);
1191             dx10             = _mm256_sub_ps(ix1,jx0);
1192             dy10             = _mm256_sub_ps(iy1,jy0);
1193             dz10             = _mm256_sub_ps(iz1,jz0);
1194             dx20             = _mm256_sub_ps(ix2,jx0);
1195             dy20             = _mm256_sub_ps(iy2,jy0);
1196             dz20             = _mm256_sub_ps(iz2,jz0);
1197             dx30             = _mm256_sub_ps(ix3,jx0);
1198             dy30             = _mm256_sub_ps(iy3,jy0);
1199             dz30             = _mm256_sub_ps(iz3,jz0);
1200
1201             /* Calculate squared distance and things based on it */
1202             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1203             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1204             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1205             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1206
1207             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1208             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1209             rinv30           = gmx_mm256_invsqrt_ps(rsq30);
1210
1211             rinvsq00         = gmx_mm256_inv_ps(rsq00);
1212             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1213             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1214             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1215
1216             /* Load parameters for j particles */
1217             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1218                                                                  charge+jnrC+0,charge+jnrD+0,
1219                                                                  charge+jnrE+0,charge+jnrF+0,
1220                                                                  charge+jnrG+0,charge+jnrH+0);
1221             vdwjidx0A        = 2*vdwtype[jnrA+0];
1222             vdwjidx0B        = 2*vdwtype[jnrB+0];
1223             vdwjidx0C        = 2*vdwtype[jnrC+0];
1224             vdwjidx0D        = 2*vdwtype[jnrD+0];
1225             vdwjidx0E        = 2*vdwtype[jnrE+0];
1226             vdwjidx0F        = 2*vdwtype[jnrF+0];
1227             vdwjidx0G        = 2*vdwtype[jnrG+0];
1228             vdwjidx0H        = 2*vdwtype[jnrH+0];
1229
1230             fjx0             = _mm256_setzero_ps();
1231             fjy0             = _mm256_setzero_ps();
1232             fjz0             = _mm256_setzero_ps();
1233
1234             /**************************
1235              * CALCULATE INTERACTIONS *
1236              **************************/
1237
1238             /* Compute parameters for interactions between i and j atoms */
1239             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1240                                             vdwioffsetptr0+vdwjidx0B,
1241                                             vdwioffsetptr0+vdwjidx0C,
1242                                             vdwioffsetptr0+vdwjidx0D,
1243                                             vdwioffsetptr0+vdwjidx0E,
1244                                             vdwioffsetptr0+vdwjidx0F,
1245                                             vdwioffsetptr0+vdwjidx0G,
1246                                             vdwioffsetptr0+vdwjidx0H,
1247                                             &c6_00,&c12_00);
1248
1249             /* LENNARD-JONES DISPERSION/REPULSION */
1250
1251             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1252             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1253
1254             fscal            = fvdw;
1255
1256             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1257
1258             /* Calculate temporary vectorial force */
1259             tx               = _mm256_mul_ps(fscal,dx00);
1260             ty               = _mm256_mul_ps(fscal,dy00);
1261             tz               = _mm256_mul_ps(fscal,dz00);
1262
1263             /* Update vectorial force */
1264             fix0             = _mm256_add_ps(fix0,tx);
1265             fiy0             = _mm256_add_ps(fiy0,ty);
1266             fiz0             = _mm256_add_ps(fiz0,tz);
1267
1268             fjx0             = _mm256_add_ps(fjx0,tx);
1269             fjy0             = _mm256_add_ps(fjy0,ty);
1270             fjz0             = _mm256_add_ps(fjz0,tz);
1271
1272             /**************************
1273              * CALCULATE INTERACTIONS *
1274              **************************/
1275
1276             r10              = _mm256_mul_ps(rsq10,rinv10);
1277             r10              = _mm256_andnot_ps(dummy_mask,r10);
1278
1279             /* Compute parameters for interactions between i and j atoms */
1280             qq10             = _mm256_mul_ps(iq1,jq0);
1281
1282             /* EWALD ELECTROSTATICS */
1283
1284             /* Analytical PME correction */
1285             zeta2            = _mm256_mul_ps(beta2,rsq10);
1286             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1287             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1288             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1289             felec            = _mm256_mul_ps(qq10,felec);
1290
1291             fscal            = felec;
1292
1293             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1294
1295             /* Calculate temporary vectorial force */
1296             tx               = _mm256_mul_ps(fscal,dx10);
1297             ty               = _mm256_mul_ps(fscal,dy10);
1298             tz               = _mm256_mul_ps(fscal,dz10);
1299
1300             /* Update vectorial force */
1301             fix1             = _mm256_add_ps(fix1,tx);
1302             fiy1             = _mm256_add_ps(fiy1,ty);
1303             fiz1             = _mm256_add_ps(fiz1,tz);
1304
1305             fjx0             = _mm256_add_ps(fjx0,tx);
1306             fjy0             = _mm256_add_ps(fjy0,ty);
1307             fjz0             = _mm256_add_ps(fjz0,tz);
1308
1309             /**************************
1310              * CALCULATE INTERACTIONS *
1311              **************************/
1312
1313             r20              = _mm256_mul_ps(rsq20,rinv20);
1314             r20              = _mm256_andnot_ps(dummy_mask,r20);
1315
1316             /* Compute parameters for interactions between i and j atoms */
1317             qq20             = _mm256_mul_ps(iq2,jq0);
1318
1319             /* EWALD ELECTROSTATICS */
1320
1321             /* Analytical PME correction */
1322             zeta2            = _mm256_mul_ps(beta2,rsq20);
1323             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1324             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1325             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1326             felec            = _mm256_mul_ps(qq20,felec);
1327
1328             fscal            = felec;
1329
1330             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1331
1332             /* Calculate temporary vectorial force */
1333             tx               = _mm256_mul_ps(fscal,dx20);
1334             ty               = _mm256_mul_ps(fscal,dy20);
1335             tz               = _mm256_mul_ps(fscal,dz20);
1336
1337             /* Update vectorial force */
1338             fix2             = _mm256_add_ps(fix2,tx);
1339             fiy2             = _mm256_add_ps(fiy2,ty);
1340             fiz2             = _mm256_add_ps(fiz2,tz);
1341
1342             fjx0             = _mm256_add_ps(fjx0,tx);
1343             fjy0             = _mm256_add_ps(fjy0,ty);
1344             fjz0             = _mm256_add_ps(fjz0,tz);
1345
1346             /**************************
1347              * CALCULATE INTERACTIONS *
1348              **************************/
1349
1350             r30              = _mm256_mul_ps(rsq30,rinv30);
1351             r30              = _mm256_andnot_ps(dummy_mask,r30);
1352
1353             /* Compute parameters for interactions between i and j atoms */
1354             qq30             = _mm256_mul_ps(iq3,jq0);
1355
1356             /* EWALD ELECTROSTATICS */
1357
1358             /* Analytical PME correction */
1359             zeta2            = _mm256_mul_ps(beta2,rsq30);
1360             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1361             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1362             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1363             felec            = _mm256_mul_ps(qq30,felec);
1364
1365             fscal            = felec;
1366
1367             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1368
1369             /* Calculate temporary vectorial force */
1370             tx               = _mm256_mul_ps(fscal,dx30);
1371             ty               = _mm256_mul_ps(fscal,dy30);
1372             tz               = _mm256_mul_ps(fscal,dz30);
1373
1374             /* Update vectorial force */
1375             fix3             = _mm256_add_ps(fix3,tx);
1376             fiy3             = _mm256_add_ps(fiy3,ty);
1377             fiz3             = _mm256_add_ps(fiz3,tz);
1378
1379             fjx0             = _mm256_add_ps(fjx0,tx);
1380             fjy0             = _mm256_add_ps(fjy0,ty);
1381             fjz0             = _mm256_add_ps(fjz0,tz);
1382
1383             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1384             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1385             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1386             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1387             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1388             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1389             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1390             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1391
1392             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1393
1394             /* Inner loop uses 201 flops */
1395         }
1396
1397         /* End of innermost loop */
1398
1399         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1400                                                  f+i_coord_offset,fshift+i_shift_offset);
1401
1402         /* Increment number of inner iterations */
1403         inneriter                  += j_index_end - j_index_start;
1404
1405         /* Outer loop uses 24 flops */
1406     }
1407
1408     /* Increment number of outer iterations */
1409     outeriter        += nri;
1410
1411     /* Update outer/inner flops */
1412
1413     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*201);
1414 }