src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_double 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_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_VF_avx_256_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water4-Water4
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     real *           vdwgridioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     real *           vdwgridioffsetptr1;
  88     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     real *           vdwgridioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     real *           vdwgridioffsetptr3;
  94     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  95     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  96     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  97     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  98     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  99     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 100     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 101     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
 102     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 103     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 104     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 105     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 106     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 107     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 108     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 109     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 110     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 111     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 112     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 113     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 114     real             *charge;
 115     int              nvdwtype;
 116     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 117     int              *vdwtype;
 118     real             *vdwparam;
 119     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 120     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 121     __m256d           c6grid_00;
 122     __m256d           c6grid_11;
 123     __m256d           c6grid_12;
 124     __m256d           c6grid_13;
 125     __m256d           c6grid_21;
 126     __m256d           c6grid_22;
 127     __m256d           c6grid_23;
 128     __m256d           c6grid_31;
 129     __m256d           c6grid_32;
 130     __m256d           c6grid_33;
 131     real             *vdwgridparam;
 132     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 133     __m256d           one_half  = _mm256_set1_pd(0.5);
 134     __m256d           minus_one = _mm256_set1_pd(-1.0);
 135     __m128i          ewitab;
 136     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 137     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 138     real             *ewtab;
 139     __m256d          dummy_mask,cutoff_mask;
 140     __m128           tmpmask0,tmpmask1;
 141     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 142     __m256d          one     = _mm256_set1_pd(1.0);
 143     __m256d          two     = _mm256_set1_pd(2.0);
 144     x                = xx[0];
 145     f                = ff[0];
 146
 147     nri              = nlist->nri;
 148     iinr             = nlist->iinr;
 149     jindex           = nlist->jindex;
 150     jjnr             = nlist->jjnr;
 151     shiftidx         = nlist->shift;
 152     gid              = nlist->gid;
 153     shiftvec         = fr->shift_vec[0];
 154     fshift           = fr->fshift[0];
 155     facel            = _mm256_set1_pd(fr->ic->epsfac);
 156     charge           = mdatoms->chargeA;
 157     nvdwtype         = fr->ntype;
 158     vdwparam         = fr->nbfp;
 159     vdwtype          = mdatoms->typeA;
 160     vdwgridparam     = fr->ljpme_c6grid;
 161     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 162     ewclj            = _mm256_set1_pd(fr->ic->ewaldcoeff_lj);
 163     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 164
 165     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 166     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 167     beta2            = _mm256_mul_pd(beta,beta);
 168     beta3            = _mm256_mul_pd(beta,beta2);
 169
 170     ewtab            = fr->ic->tabq_coul_FDV0;
 171     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 172     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 173
 174     /* Setup water-specific parameters */
 175     inr              = nlist->iinr[0];
 176     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 177     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 178     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 179     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 180     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 181
 182     jq1              = _mm256_set1_pd(charge[inr+1]);
 183     jq2              = _mm256_set1_pd(charge[inr+2]);
 184     jq3              = _mm256_set1_pd(charge[inr+3]);
 185     vdwjidx0A        = 2*vdwtype[inr+0];
 186     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 187     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 188     c6grid_00        = _mm256_set1_pd(vdwgridioffsetptr0[vdwjidx0A]);
 189     qq11             = _mm256_mul_pd(iq1,jq1);
 190     qq12             = _mm256_mul_pd(iq1,jq2);
 191     qq13             = _mm256_mul_pd(iq1,jq3);
 192     qq21             = _mm256_mul_pd(iq2,jq1);
 193     qq22             = _mm256_mul_pd(iq2,jq2);
 194     qq23             = _mm256_mul_pd(iq2,jq3);
 195     qq31             = _mm256_mul_pd(iq3,jq1);
 196     qq32             = _mm256_mul_pd(iq3,jq2);
 197     qq33             = _mm256_mul_pd(iq3,jq3);
 198
 199     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 200     rcutoff_scalar   = fr->ic->rcoulomb;
 201     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 202     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 203
 204     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 205     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 206
 207     /* Avoid stupid compiler warnings */
 208     jnrA = jnrB = jnrC = jnrD = 0;
 209     j_coord_offsetA = 0;
 210     j_coord_offsetB = 0;
 211     j_coord_offsetC = 0;
 212     j_coord_offsetD = 0;
 213
 214     outeriter        = 0;
 215     inneriter        = 0;
 216
 217     for(iidx=0;iidx<4*DIM;iidx++)
 218     {
 219         scratch[iidx] = 0.0;
 220     }
 221
 222     /* Start outer loop over neighborlists */
 223     for(iidx=0; iidx<nri; iidx++)
 224     {
 225         /* Load shift vector for this list */
 226         i_shift_offset   = DIM*shiftidx[iidx];
 227
 228         /* Load limits for loop over neighbors */
 229         j_index_start    = jindex[iidx];
 230         j_index_end      = jindex[iidx+1];
 231
 232         /* Get outer coordinate index */
 233         inr              = iinr[iidx];
 234         i_coord_offset   = DIM*inr;
 235
 236         /* Load i particle coords and add shift vector */
 237         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 238                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 239
 240         fix0             = _mm256_setzero_pd();
 241         fiy0             = _mm256_setzero_pd();
 242         fiz0             = _mm256_setzero_pd();
 243         fix1             = _mm256_setzero_pd();
 244         fiy1             = _mm256_setzero_pd();
 245         fiz1             = _mm256_setzero_pd();
 246         fix2             = _mm256_setzero_pd();
 247         fiy2             = _mm256_setzero_pd();
 248         fiz2             = _mm256_setzero_pd();
 249         fix3             = _mm256_setzero_pd();
 250         fiy3             = _mm256_setzero_pd();
 251         fiz3             = _mm256_setzero_pd();
 252
 253         /* Reset potential sums */
 254         velecsum         = _mm256_setzero_pd();
 255         vvdwsum          = _mm256_setzero_pd();
 256
 257         /* Start inner kernel loop */
 258         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 259         {
 260
 261             /* Get j neighbor index, and coordinate index */
 262             jnrA             = jjnr[jidx];
 263             jnrB             = jjnr[jidx+1];
 264             jnrC             = jjnr[jidx+2];
 265             jnrD             = jjnr[jidx+3];
 266             j_coord_offsetA  = DIM*jnrA;
 267             j_coord_offsetB  = DIM*jnrB;
 268             j_coord_offsetC  = DIM*jnrC;
 269             j_coord_offsetD  = DIM*jnrD;
 270
 271             /* load j atom coordinates */
 272             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 273                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 274                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 275                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 276
 277             /* Calculate displacement vector */
 278             dx00             = _mm256_sub_pd(ix0,jx0);
 279             dy00             = _mm256_sub_pd(iy0,jy0);
 280             dz00             = _mm256_sub_pd(iz0,jz0);
 281             dx11             = _mm256_sub_pd(ix1,jx1);
 282             dy11             = _mm256_sub_pd(iy1,jy1);
 283             dz11             = _mm256_sub_pd(iz1,jz1);
 284             dx12             = _mm256_sub_pd(ix1,jx2);
 285             dy12             = _mm256_sub_pd(iy1,jy2);
 286             dz12             = _mm256_sub_pd(iz1,jz2);
 287             dx13             = _mm256_sub_pd(ix1,jx3);
 288             dy13             = _mm256_sub_pd(iy1,jy3);
 289             dz13             = _mm256_sub_pd(iz1,jz3);
 290             dx21             = _mm256_sub_pd(ix2,jx1);
 291             dy21             = _mm256_sub_pd(iy2,jy1);
 292             dz21             = _mm256_sub_pd(iz2,jz1);
 293             dx22             = _mm256_sub_pd(ix2,jx2);
 294             dy22             = _mm256_sub_pd(iy2,jy2);
 295             dz22             = _mm256_sub_pd(iz2,jz2);
 296             dx23             = _mm256_sub_pd(ix2,jx3);
 297             dy23             = _mm256_sub_pd(iy2,jy3);
 298             dz23             = _mm256_sub_pd(iz2,jz3);
 299             dx31             = _mm256_sub_pd(ix3,jx1);
 300             dy31             = _mm256_sub_pd(iy3,jy1);
 301             dz31             = _mm256_sub_pd(iz3,jz1);
 302             dx32             = _mm256_sub_pd(ix3,jx2);
 303             dy32             = _mm256_sub_pd(iy3,jy2);
 304             dz32             = _mm256_sub_pd(iz3,jz2);
 305             dx33             = _mm256_sub_pd(ix3,jx3);
 306             dy33             = _mm256_sub_pd(iy3,jy3);
 307             dz33             = _mm256_sub_pd(iz3,jz3);
 308
 309             /* Calculate squared distance and things based on it */
 310             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 311             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 312             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 313             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 314             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 315             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 316             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 317             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 318             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 319             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 320
 321             rinv00           = avx256_invsqrt_d(rsq00);
 322             rinv11           = avx256_invsqrt_d(rsq11);
 323             rinv12           = avx256_invsqrt_d(rsq12);
 324             rinv13           = avx256_invsqrt_d(rsq13);
 325             rinv21           = avx256_invsqrt_d(rsq21);
 326             rinv22           = avx256_invsqrt_d(rsq22);
 327             rinv23           = avx256_invsqrt_d(rsq23);
 328             rinv31           = avx256_invsqrt_d(rsq31);
 329             rinv32           = avx256_invsqrt_d(rsq32);
 330             rinv33           = avx256_invsqrt_d(rsq33);
 331
 332             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 333             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 334             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 335             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 336             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 337             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 338             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 339             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 340             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 341             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 342
 343             fjx0             = _mm256_setzero_pd();
 344             fjy0             = _mm256_setzero_pd();
 345             fjz0             = _mm256_setzero_pd();
 346             fjx1             = _mm256_setzero_pd();
 347             fjy1             = _mm256_setzero_pd();
 348             fjz1             = _mm256_setzero_pd();
 349             fjx2             = _mm256_setzero_pd();
 350             fjy2             = _mm256_setzero_pd();
 351             fjz2             = _mm256_setzero_pd();
 352             fjx3             = _mm256_setzero_pd();
 353             fjy3             = _mm256_setzero_pd();
 354             fjz3             = _mm256_setzero_pd();
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 361             {
 362
 363             r00              = _mm256_mul_pd(rsq00,rinv00);
 364
 365             /* Analytical LJ-PME */
 366             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 367             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 368             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 369             exponent         = avx256_exp_d(ewcljrsq);
 370             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 371             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 372             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 373             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 374             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 375             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 376                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_add_pd(_mm256_mul_pd(c6_00,sh_vdw_invrcut6),_mm256_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 377             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 378             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 379
 380             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 384             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 385
 386             fscal            = fvdw;
 387
 388             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 389
 390             /* Calculate temporary vectorial force */
 391             tx               = _mm256_mul_pd(fscal,dx00);
 392             ty               = _mm256_mul_pd(fscal,dy00);
 393             tz               = _mm256_mul_pd(fscal,dz00);
 394
 395             /* Update vectorial force */
 396             fix0             = _mm256_add_pd(fix0,tx);
 397             fiy0             = _mm256_add_pd(fiy0,ty);
 398             fiz0             = _mm256_add_pd(fiz0,tz);
 399
 400             fjx0             = _mm256_add_pd(fjx0,tx);
 401             fjy0             = _mm256_add_pd(fjy0,ty);
 402             fjz0             = _mm256_add_pd(fjz0,tz);
 403
 404             }
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 411             {
 412
 413             r11              = _mm256_mul_pd(rsq11,rinv11);
 414
 415             /* EWALD ELECTROSTATICS */
 416
 417             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 418             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 419             ewitab           = _mm256_cvttpd_epi32(ewrt);
 420             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 421             ewitab           = _mm_slli_epi32(ewitab,2);
 422             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 423             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 424             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 425             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 426             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 427             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 428             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 429             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_sub_pd(rinv11,sh_ewald),velec));
 430             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 431
 432             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velec            = _mm256_and_pd(velec,cutoff_mask);
 436             velecsum         = _mm256_add_pd(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 441
 442             /* Calculate temporary vectorial force */
 443             tx               = _mm256_mul_pd(fscal,dx11);
 444             ty               = _mm256_mul_pd(fscal,dy11);
 445             tz               = _mm256_mul_pd(fscal,dz11);
 446
 447             /* Update vectorial force */
 448             fix1             = _mm256_add_pd(fix1,tx);
 449             fiy1             = _mm256_add_pd(fiy1,ty);
 450             fiz1             = _mm256_add_pd(fiz1,tz);
 451
 452             fjx1             = _mm256_add_pd(fjx1,tx);
 453             fjy1             = _mm256_add_pd(fjy1,ty);
 454             fjz1             = _mm256_add_pd(fjz1,tz);
 455
 456             }
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 463             {
 464
 465             r12              = _mm256_mul_pd(rsq12,rinv12);
 466
 467             /* EWALD ELECTROSTATICS */
 468
 469             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 470             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 471             ewitab           = _mm256_cvttpd_epi32(ewrt);
 472             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 473             ewitab           = _mm_slli_epi32(ewitab,2);
 474             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 475             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 476             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 477             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 478             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 479             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 480             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 481             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_sub_pd(rinv12,sh_ewald),velec));
 482             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 483
 484             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             velec            = _mm256_and_pd(velec,cutoff_mask);
 488             velecsum         = _mm256_add_pd(velecsum,velec);
 489
 490             fscal            = felec;
 491
 492             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 493
 494             /* Calculate temporary vectorial force */
 495             tx               = _mm256_mul_pd(fscal,dx12);
 496             ty               = _mm256_mul_pd(fscal,dy12);
 497             tz               = _mm256_mul_pd(fscal,dz12);
 498
 499             /* Update vectorial force */
 500             fix1             = _mm256_add_pd(fix1,tx);
 501             fiy1             = _mm256_add_pd(fiy1,ty);
 502             fiz1             = _mm256_add_pd(fiz1,tz);
 503
 504             fjx2             = _mm256_add_pd(fjx2,tx);
 505             fjy2             = _mm256_add_pd(fjy2,ty);
 506             fjz2             = _mm256_add_pd(fjz2,tz);
 507
 508             }
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 515             {
 516
 517             r13              = _mm256_mul_pd(rsq13,rinv13);
 518
 519             /* EWALD ELECTROSTATICS */
 520
 521             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 522             ewrt             = _mm256_mul_pd(r13,ewtabscale);
 523             ewitab           = _mm256_cvttpd_epi32(ewrt);
 524             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 525             ewitab           = _mm_slli_epi32(ewitab,2);
 526             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 527             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 528             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 529             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 530             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 531             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 532             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 533             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_sub_pd(rinv13,sh_ewald),velec));
 534             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
 535
 536             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
 537
 538             /* Update potential sum for this i atom from the interaction with this j atom. */
 539             velec            = _mm256_and_pd(velec,cutoff_mask);
 540             velecsum         = _mm256_add_pd(velecsum,velec);
 541
 542             fscal            = felec;
 543
 544             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 545
 546             /* Calculate temporary vectorial force */
 547             tx               = _mm256_mul_pd(fscal,dx13);
 548             ty               = _mm256_mul_pd(fscal,dy13);
 549             tz               = _mm256_mul_pd(fscal,dz13);
 550
 551             /* Update vectorial force */
 552             fix1             = _mm256_add_pd(fix1,tx);
 553             fiy1             = _mm256_add_pd(fiy1,ty);
 554             fiz1             = _mm256_add_pd(fiz1,tz);
 555
 556             fjx3             = _mm256_add_pd(fjx3,tx);
 557             fjy3             = _mm256_add_pd(fjy3,ty);
 558             fjz3             = _mm256_add_pd(fjz3,tz);
 559
 560             }
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 567             {
 568
 569             r21              = _mm256_mul_pd(rsq21,rinv21);
 570
 571             /* EWALD ELECTROSTATICS */
 572
 573             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 574             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 575             ewitab           = _mm256_cvttpd_epi32(ewrt);
 576             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 577             ewitab           = _mm_slli_epi32(ewitab,2);
 578             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 579             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 580             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 581             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 582             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 583             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 584             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 585             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_sub_pd(rinv21,sh_ewald),velec));
 586             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 587
 588             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             velec            = _mm256_and_pd(velec,cutoff_mask);
 592             velecsum         = _mm256_add_pd(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm256_mul_pd(fscal,dx21);
 600             ty               = _mm256_mul_pd(fscal,dy21);
 601             tz               = _mm256_mul_pd(fscal,dz21);
 602
 603             /* Update vectorial force */
 604             fix2             = _mm256_add_pd(fix2,tx);
 605             fiy2             = _mm256_add_pd(fiy2,ty);
 606             fiz2             = _mm256_add_pd(fiz2,tz);
 607
 608             fjx1             = _mm256_add_pd(fjx1,tx);
 609             fjy1             = _mm256_add_pd(fjy1,ty);
 610             fjz1             = _mm256_add_pd(fjz1,tz);
 611
 612             }
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 619             {
 620
 621             r22              = _mm256_mul_pd(rsq22,rinv22);
 622
 623             /* EWALD ELECTROSTATICS */
 624
 625             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 626             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 627             ewitab           = _mm256_cvttpd_epi32(ewrt);
 628             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 629             ewitab           = _mm_slli_epi32(ewitab,2);
 630             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 631             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 632             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 633             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 634             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 635             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 636             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 637             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_sub_pd(rinv22,sh_ewald),velec));
 638             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 639
 640             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velec            = _mm256_and_pd(velec,cutoff_mask);
 644             velecsum         = _mm256_add_pd(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 649
 650             /* Calculate temporary vectorial force */
 651             tx               = _mm256_mul_pd(fscal,dx22);
 652             ty               = _mm256_mul_pd(fscal,dy22);
 653             tz               = _mm256_mul_pd(fscal,dz22);
 654
 655             /* Update vectorial force */
 656             fix2             = _mm256_add_pd(fix2,tx);
 657             fiy2             = _mm256_add_pd(fiy2,ty);
 658             fiz2             = _mm256_add_pd(fiz2,tz);
 659
 660             fjx2             = _mm256_add_pd(fjx2,tx);
 661             fjy2             = _mm256_add_pd(fjy2,ty);
 662             fjz2             = _mm256_add_pd(fjz2,tz);
 663
 664             }
 665
 666             /**************************
 667              * CALCULATE INTERACTIONS *
 668              **************************/
 669
 670             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 671             {
 672
 673             r23              = _mm256_mul_pd(rsq23,rinv23);
 674
 675             /* EWALD ELECTROSTATICS */
 676
 677             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 678             ewrt             = _mm256_mul_pd(r23,ewtabscale);
 679             ewitab           = _mm256_cvttpd_epi32(ewrt);
 680             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 681             ewitab           = _mm_slli_epi32(ewitab,2);
 682             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 683             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 684             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 685             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 686             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 687             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 688             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 689             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_sub_pd(rinv23,sh_ewald),velec));
 690             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
 691
 692             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
 693
 694             /* Update potential sum for this i atom from the interaction with this j atom. */
 695             velec            = _mm256_and_pd(velec,cutoff_mask);
 696             velecsum         = _mm256_add_pd(velecsum,velec);
 697
 698             fscal            = felec;
 699
 700             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 701
 702             /* Calculate temporary vectorial force */
 703             tx               = _mm256_mul_pd(fscal,dx23);
 704             ty               = _mm256_mul_pd(fscal,dy23);
 705             tz               = _mm256_mul_pd(fscal,dz23);
 706
 707             /* Update vectorial force */
 708             fix2             = _mm256_add_pd(fix2,tx);
 709             fiy2             = _mm256_add_pd(fiy2,ty);
 710             fiz2             = _mm256_add_pd(fiz2,tz);
 711
 712             fjx3             = _mm256_add_pd(fjx3,tx);
 713             fjy3             = _mm256_add_pd(fjy3,ty);
 714             fjz3             = _mm256_add_pd(fjz3,tz);
 715
 716             }
 717
 718             /**************************
 719              * CALCULATE INTERACTIONS *
 720              **************************/
 721
 722             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 723             {
 724
 725             r31              = _mm256_mul_pd(rsq31,rinv31);
 726
 727             /* EWALD ELECTROSTATICS */
 728
 729             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 730             ewrt             = _mm256_mul_pd(r31,ewtabscale);
 731             ewitab           = _mm256_cvttpd_epi32(ewrt);
 732             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 733             ewitab           = _mm_slli_epi32(ewitab,2);
 734             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 735             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 736             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 737             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 738             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 739             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 740             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 741             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_sub_pd(rinv31,sh_ewald),velec));
 742             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
 743
 744             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
 745
 746             /* Update potential sum for this i atom from the interaction with this j atom. */
 747             velec            = _mm256_and_pd(velec,cutoff_mask);
 748             velecsum         = _mm256_add_pd(velecsum,velec);
 749
 750             fscal            = felec;
 751
 752             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 753
 754             /* Calculate temporary vectorial force */
 755             tx               = _mm256_mul_pd(fscal,dx31);
 756             ty               = _mm256_mul_pd(fscal,dy31);
 757             tz               = _mm256_mul_pd(fscal,dz31);
 758
 759             /* Update vectorial force */
 760             fix3             = _mm256_add_pd(fix3,tx);
 761             fiy3             = _mm256_add_pd(fiy3,ty);
 762             fiz3             = _mm256_add_pd(fiz3,tz);
 763
 764             fjx1             = _mm256_add_pd(fjx1,tx);
 765             fjy1             = _mm256_add_pd(fjy1,ty);
 766             fjz1             = _mm256_add_pd(fjz1,tz);
 767
 768             }
 769
 770             /**************************
 771              * CALCULATE INTERACTIONS *
 772              **************************/
 773
 774             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 775             {
 776
 777             r32              = _mm256_mul_pd(rsq32,rinv32);
 778
 779             /* EWALD ELECTROSTATICS */
 780
 781             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 782             ewrt             = _mm256_mul_pd(r32,ewtabscale);
 783             ewitab           = _mm256_cvttpd_epi32(ewrt);
 784             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 785             ewitab           = _mm_slli_epi32(ewitab,2);
 786             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 787             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 788             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 789             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 790             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 791             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 792             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 793             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_sub_pd(rinv32,sh_ewald),velec));
 794             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
 795
 796             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
 797
 798             /* Update potential sum for this i atom from the interaction with this j atom. */
 799             velec            = _mm256_and_pd(velec,cutoff_mask);
 800             velecsum         = _mm256_add_pd(velecsum,velec);
 801
 802             fscal            = felec;
 803
 804             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 805
 806             /* Calculate temporary vectorial force */
 807             tx               = _mm256_mul_pd(fscal,dx32);
 808             ty               = _mm256_mul_pd(fscal,dy32);
 809             tz               = _mm256_mul_pd(fscal,dz32);
 810
 811             /* Update vectorial force */
 812             fix3             = _mm256_add_pd(fix3,tx);
 813             fiy3             = _mm256_add_pd(fiy3,ty);
 814             fiz3             = _mm256_add_pd(fiz3,tz);
 815
 816             fjx2             = _mm256_add_pd(fjx2,tx);
 817             fjy2             = _mm256_add_pd(fjy2,ty);
 818             fjz2             = _mm256_add_pd(fjz2,tz);
 819
 820             }
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 827             {
 828
 829             r33              = _mm256_mul_pd(rsq33,rinv33);
 830
 831             /* EWALD ELECTROSTATICS */
 832
 833             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 834             ewrt             = _mm256_mul_pd(r33,ewtabscale);
 835             ewitab           = _mm256_cvttpd_epi32(ewrt);
 836             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 837             ewitab           = _mm_slli_epi32(ewitab,2);
 838             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 839             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 840             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 841             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 842             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 843             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 844             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 845             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_sub_pd(rinv33,sh_ewald),velec));
 846             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
 847
 848             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
 849
 850             /* Update potential sum for this i atom from the interaction with this j atom. */
 851             velec            = _mm256_and_pd(velec,cutoff_mask);
 852             velecsum         = _mm256_add_pd(velecsum,velec);
 853
 854             fscal            = felec;
 855
 856             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 857
 858             /* Calculate temporary vectorial force */
 859             tx               = _mm256_mul_pd(fscal,dx33);
 860             ty               = _mm256_mul_pd(fscal,dy33);
 861             tz               = _mm256_mul_pd(fscal,dz33);
 862
 863             /* Update vectorial force */
 864             fix3             = _mm256_add_pd(fix3,tx);
 865             fiy3             = _mm256_add_pd(fiy3,ty);
 866             fiz3             = _mm256_add_pd(fiz3,tz);
 867
 868             fjx3             = _mm256_add_pd(fjx3,tx);
 869             fjy3             = _mm256_add_pd(fjy3,ty);
 870             fjz3             = _mm256_add_pd(fjz3,tz);
 871
 872             }
 873
 874             fjptrA             = f+j_coord_offsetA;
 875             fjptrB             = f+j_coord_offsetB;
 876             fjptrC             = f+j_coord_offsetC;
 877             fjptrD             = f+j_coord_offsetD;
 878
 879             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 880                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 881                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 882
 883             /* Inner loop uses 479 flops */
 884         }
 885
 886         if(jidx<j_index_end)
 887         {
 888
 889             /* Get j neighbor index, and coordinate index */
 890             jnrlistA         = jjnr[jidx];
 891             jnrlistB         = jjnr[jidx+1];
 892             jnrlistC         = jjnr[jidx+2];
 893             jnrlistD         = jjnr[jidx+3];
 894             /* Sign of each element will be negative for non-real atoms.
 895              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 896              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 897              */
 898             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 899
 900             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 901             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 902             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 903
 904             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 905             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 906             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 907             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 908             j_coord_offsetA  = DIM*jnrA;
 909             j_coord_offsetB  = DIM*jnrB;
 910             j_coord_offsetC  = DIM*jnrC;
 911             j_coord_offsetD  = DIM*jnrD;
 912
 913             /* load j atom coordinates */
 914             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 915                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 916                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 917                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 918
 919             /* Calculate displacement vector */
 920             dx00             = _mm256_sub_pd(ix0,jx0);
 921             dy00             = _mm256_sub_pd(iy0,jy0);
 922             dz00             = _mm256_sub_pd(iz0,jz0);
 923             dx11             = _mm256_sub_pd(ix1,jx1);
 924             dy11             = _mm256_sub_pd(iy1,jy1);
 925             dz11             = _mm256_sub_pd(iz1,jz1);
 926             dx12             = _mm256_sub_pd(ix1,jx2);
 927             dy12             = _mm256_sub_pd(iy1,jy2);
 928             dz12             = _mm256_sub_pd(iz1,jz2);
 929             dx13             = _mm256_sub_pd(ix1,jx3);
 930             dy13             = _mm256_sub_pd(iy1,jy3);
 931             dz13             = _mm256_sub_pd(iz1,jz3);
 932             dx21             = _mm256_sub_pd(ix2,jx1);
 933             dy21             = _mm256_sub_pd(iy2,jy1);
 934             dz21             = _mm256_sub_pd(iz2,jz1);
 935             dx22             = _mm256_sub_pd(ix2,jx2);
 936             dy22             = _mm256_sub_pd(iy2,jy2);
 937             dz22             = _mm256_sub_pd(iz2,jz2);
 938             dx23             = _mm256_sub_pd(ix2,jx3);
 939             dy23             = _mm256_sub_pd(iy2,jy3);
 940             dz23             = _mm256_sub_pd(iz2,jz3);
 941             dx31             = _mm256_sub_pd(ix3,jx1);
 942             dy31             = _mm256_sub_pd(iy3,jy1);
 943             dz31             = _mm256_sub_pd(iz3,jz1);
 944             dx32             = _mm256_sub_pd(ix3,jx2);
 945             dy32             = _mm256_sub_pd(iy3,jy2);
 946             dz32             = _mm256_sub_pd(iz3,jz2);
 947             dx33             = _mm256_sub_pd(ix3,jx3);
 948             dy33             = _mm256_sub_pd(iy3,jy3);
 949             dz33             = _mm256_sub_pd(iz3,jz3);
 950
 951             /* Calculate squared distance and things based on it */
 952             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 953             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 954             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 955             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 956             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 957             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 958             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 959             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 960             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 961             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 962
 963             rinv00           = avx256_invsqrt_d(rsq00);
 964             rinv11           = avx256_invsqrt_d(rsq11);
 965             rinv12           = avx256_invsqrt_d(rsq12);
 966             rinv13           = avx256_invsqrt_d(rsq13);
 967             rinv21           = avx256_invsqrt_d(rsq21);
 968             rinv22           = avx256_invsqrt_d(rsq22);
 969             rinv23           = avx256_invsqrt_d(rsq23);
 970             rinv31           = avx256_invsqrt_d(rsq31);
 971             rinv32           = avx256_invsqrt_d(rsq32);
 972             rinv33           = avx256_invsqrt_d(rsq33);
 973
 974             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 975             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 976             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 977             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 978             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 979             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 980             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 981             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 982             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 983             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 984
 985             fjx0             = _mm256_setzero_pd();
 986             fjy0             = _mm256_setzero_pd();
 987             fjz0             = _mm256_setzero_pd();
 988             fjx1             = _mm256_setzero_pd();
 989             fjy1             = _mm256_setzero_pd();
 990             fjz1             = _mm256_setzero_pd();
 991             fjx2             = _mm256_setzero_pd();
 992             fjy2             = _mm256_setzero_pd();
 993             fjz2             = _mm256_setzero_pd();
 994             fjx3             = _mm256_setzero_pd();
 995             fjy3             = _mm256_setzero_pd();
 996             fjz3             = _mm256_setzero_pd();
 997
 998             /**************************
 999              * CALCULATE INTERACTIONS *
1000              **************************/
1001
1002             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1003             {
1004
1005             r00              = _mm256_mul_pd(rsq00,rinv00);
1006             r00              = _mm256_andnot_pd(dummy_mask,r00);
1007
1008             /* Analytical LJ-PME */
1009             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1010             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1011             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1012             exponent         = avx256_exp_d(ewcljrsq);
1013             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1014             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1015             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
1016             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
1017             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1018             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
1019                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_add_pd(_mm256_mul_pd(c6_00,sh_vdw_invrcut6),_mm256_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
1020             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
1021             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
1022
1023             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1024
1025             /* Update potential sum for this i atom from the interaction with this j atom. */
1026             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
1027             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
1028             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
1029
1030             fscal            = fvdw;
1031
1032             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1033
1034             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm256_mul_pd(fscal,dx00);
1038             ty               = _mm256_mul_pd(fscal,dy00);
1039             tz               = _mm256_mul_pd(fscal,dz00);
1040
1041             /* Update vectorial force */
1042             fix0             = _mm256_add_pd(fix0,tx);
1043             fiy0             = _mm256_add_pd(fiy0,ty);
1044             fiz0             = _mm256_add_pd(fiz0,tz);
1045
1046             fjx0             = _mm256_add_pd(fjx0,tx);
1047             fjy0             = _mm256_add_pd(fjy0,ty);
1048             fjz0             = _mm256_add_pd(fjz0,tz);
1049
1050             }
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1057             {
1058
1059             r11              = _mm256_mul_pd(rsq11,rinv11);
1060             r11              = _mm256_andnot_pd(dummy_mask,r11);
1061
1062             /* EWALD ELECTROSTATICS */
1063
1064             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1065             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1066             ewitab           = _mm256_cvttpd_epi32(ewrt);
1067             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1068             ewitab           = _mm_slli_epi32(ewitab,2);
1069             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1070             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1071             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1072             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1073             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1074             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1075             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1076             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_sub_pd(rinv11,sh_ewald),velec));
1077             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1078
1079             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1080
1081             /* Update potential sum for this i atom from the interaction with this j atom. */
1082             velec            = _mm256_and_pd(velec,cutoff_mask);
1083             velec            = _mm256_andnot_pd(dummy_mask,velec);
1084             velecsum         = _mm256_add_pd(velecsum,velec);
1085
1086             fscal            = felec;
1087
1088             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1089
1090             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm256_mul_pd(fscal,dx11);
1094             ty               = _mm256_mul_pd(fscal,dy11);
1095             tz               = _mm256_mul_pd(fscal,dz11);
1096
1097             /* Update vectorial force */
1098             fix1             = _mm256_add_pd(fix1,tx);
1099             fiy1             = _mm256_add_pd(fiy1,ty);
1100             fiz1             = _mm256_add_pd(fiz1,tz);
1101
1102             fjx1             = _mm256_add_pd(fjx1,tx);
1103             fjy1             = _mm256_add_pd(fjy1,ty);
1104             fjz1             = _mm256_add_pd(fjz1,tz);
1105
1106             }
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1113             {
1114
1115             r12              = _mm256_mul_pd(rsq12,rinv12);
1116             r12              = _mm256_andnot_pd(dummy_mask,r12);
1117
1118             /* EWALD ELECTROSTATICS */
1119
1120             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1121             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1122             ewitab           = _mm256_cvttpd_epi32(ewrt);
1123             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1124             ewitab           = _mm_slli_epi32(ewitab,2);
1125             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1126             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1127             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1128             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1129             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1130             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1131             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1132             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_sub_pd(rinv12,sh_ewald),velec));
1133             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1134
1135             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1136
1137             /* Update potential sum for this i atom from the interaction with this j atom. */
1138             velec            = _mm256_and_pd(velec,cutoff_mask);
1139             velec            = _mm256_andnot_pd(dummy_mask,velec);
1140             velecsum         = _mm256_add_pd(velecsum,velec);
1141
1142             fscal            = felec;
1143
1144             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1145
1146             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1147
1148             /* Calculate temporary vectorial force */
1149             tx               = _mm256_mul_pd(fscal,dx12);
1150             ty               = _mm256_mul_pd(fscal,dy12);
1151             tz               = _mm256_mul_pd(fscal,dz12);
1152
1153             /* Update vectorial force */
1154             fix1             = _mm256_add_pd(fix1,tx);
1155             fiy1             = _mm256_add_pd(fiy1,ty);
1156             fiz1             = _mm256_add_pd(fiz1,tz);
1157
1158             fjx2             = _mm256_add_pd(fjx2,tx);
1159             fjy2             = _mm256_add_pd(fjy2,ty);
1160             fjz2             = _mm256_add_pd(fjz2,tz);
1161
1162             }
1163
1164             /**************************
1165              * CALCULATE INTERACTIONS *
1166              **************************/
1167
1168             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1169             {
1170
1171             r13              = _mm256_mul_pd(rsq13,rinv13);
1172             r13              = _mm256_andnot_pd(dummy_mask,r13);
1173
1174             /* EWALD ELECTROSTATICS */
1175
1176             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1177             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1178             ewitab           = _mm256_cvttpd_epi32(ewrt);
1179             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1180             ewitab           = _mm_slli_epi32(ewitab,2);
1181             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1182             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1183             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1184             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1185             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1186             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1187             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1188             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_sub_pd(rinv13,sh_ewald),velec));
1189             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1190
1191             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1192
1193             /* Update potential sum for this i atom from the interaction with this j atom. */
1194             velec            = _mm256_and_pd(velec,cutoff_mask);
1195             velec            = _mm256_andnot_pd(dummy_mask,velec);
1196             velecsum         = _mm256_add_pd(velecsum,velec);
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1201
1202             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm256_mul_pd(fscal,dx13);
1206             ty               = _mm256_mul_pd(fscal,dy13);
1207             tz               = _mm256_mul_pd(fscal,dz13);
1208
1209             /* Update vectorial force */
1210             fix1             = _mm256_add_pd(fix1,tx);
1211             fiy1             = _mm256_add_pd(fiy1,ty);
1212             fiz1             = _mm256_add_pd(fiz1,tz);
1213
1214             fjx3             = _mm256_add_pd(fjx3,tx);
1215             fjy3             = _mm256_add_pd(fjy3,ty);
1216             fjz3             = _mm256_add_pd(fjz3,tz);
1217
1218             }
1219
1220             /**************************
1221              * CALCULATE INTERACTIONS *
1222              **************************/
1223
1224             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1225             {
1226
1227             r21              = _mm256_mul_pd(rsq21,rinv21);
1228             r21              = _mm256_andnot_pd(dummy_mask,r21);
1229
1230             /* EWALD ELECTROSTATICS */
1231
1232             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1233             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1234             ewitab           = _mm256_cvttpd_epi32(ewrt);
1235             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1236             ewitab           = _mm_slli_epi32(ewitab,2);
1237             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1238             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1239             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1240             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1241             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1242             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1243             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1244             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_sub_pd(rinv21,sh_ewald),velec));
1245             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1246
1247             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1248
1249             /* Update potential sum for this i atom from the interaction with this j atom. */
1250             velec            = _mm256_and_pd(velec,cutoff_mask);
1251             velec            = _mm256_andnot_pd(dummy_mask,velec);
1252             velecsum         = _mm256_add_pd(velecsum,velec);
1253
1254             fscal            = felec;
1255
1256             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1257
1258             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm256_mul_pd(fscal,dx21);
1262             ty               = _mm256_mul_pd(fscal,dy21);
1263             tz               = _mm256_mul_pd(fscal,dz21);
1264
1265             /* Update vectorial force */
1266             fix2             = _mm256_add_pd(fix2,tx);
1267             fiy2             = _mm256_add_pd(fiy2,ty);
1268             fiz2             = _mm256_add_pd(fiz2,tz);
1269
1270             fjx1             = _mm256_add_pd(fjx1,tx);
1271             fjy1             = _mm256_add_pd(fjy1,ty);
1272             fjz1             = _mm256_add_pd(fjz1,tz);
1273
1274             }
1275
1276             /**************************
1277              * CALCULATE INTERACTIONS *
1278              **************************/
1279
1280             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1281             {
1282
1283             r22              = _mm256_mul_pd(rsq22,rinv22);
1284             r22              = _mm256_andnot_pd(dummy_mask,r22);
1285
1286             /* EWALD ELECTROSTATICS */
1287
1288             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1289             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1290             ewitab           = _mm256_cvttpd_epi32(ewrt);
1291             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1292             ewitab           = _mm_slli_epi32(ewitab,2);
1293             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1294             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1295             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1296             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1297             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1298             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1299             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1300             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_sub_pd(rinv22,sh_ewald),velec));
1301             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1302
1303             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1304
1305             /* Update potential sum for this i atom from the interaction with this j atom. */
1306             velec            = _mm256_and_pd(velec,cutoff_mask);
1307             velec            = _mm256_andnot_pd(dummy_mask,velec);
1308             velecsum         = _mm256_add_pd(velecsum,velec);
1309
1310             fscal            = felec;
1311
1312             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1313
1314             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1315
1316             /* Calculate temporary vectorial force */
1317             tx               = _mm256_mul_pd(fscal,dx22);
1318             ty               = _mm256_mul_pd(fscal,dy22);
1319             tz               = _mm256_mul_pd(fscal,dz22);
1320
1321             /* Update vectorial force */
1322             fix2             = _mm256_add_pd(fix2,tx);
1323             fiy2             = _mm256_add_pd(fiy2,ty);
1324             fiz2             = _mm256_add_pd(fiz2,tz);
1325
1326             fjx2             = _mm256_add_pd(fjx2,tx);
1327             fjy2             = _mm256_add_pd(fjy2,ty);
1328             fjz2             = _mm256_add_pd(fjz2,tz);
1329
1330             }
1331
1332             /**************************
1333              * CALCULATE INTERACTIONS *
1334              **************************/
1335
1336             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1337             {
1338
1339             r23              = _mm256_mul_pd(rsq23,rinv23);
1340             r23              = _mm256_andnot_pd(dummy_mask,r23);
1341
1342             /* EWALD ELECTROSTATICS */
1343
1344             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1345             ewrt             = _mm256_mul_pd(r23,ewtabscale);
1346             ewitab           = _mm256_cvttpd_epi32(ewrt);
1347             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1348             ewitab           = _mm_slli_epi32(ewitab,2);
1349             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1350             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1351             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1352             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1353             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1354             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1355             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1356             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_sub_pd(rinv23,sh_ewald),velec));
1357             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
1358
1359             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1360
1361             /* Update potential sum for this i atom from the interaction with this j atom. */
1362             velec            = _mm256_and_pd(velec,cutoff_mask);
1363             velec            = _mm256_andnot_pd(dummy_mask,velec);
1364             velecsum         = _mm256_add_pd(velecsum,velec);
1365
1366             fscal            = felec;
1367
1368             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1369
1370             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1371
1372             /* Calculate temporary vectorial force */
1373             tx               = _mm256_mul_pd(fscal,dx23);
1374             ty               = _mm256_mul_pd(fscal,dy23);
1375             tz               = _mm256_mul_pd(fscal,dz23);
1376
1377             /* Update vectorial force */
1378             fix2             = _mm256_add_pd(fix2,tx);
1379             fiy2             = _mm256_add_pd(fiy2,ty);
1380             fiz2             = _mm256_add_pd(fiz2,tz);
1381
1382             fjx3             = _mm256_add_pd(fjx3,tx);
1383             fjy3             = _mm256_add_pd(fjy3,ty);
1384             fjz3             = _mm256_add_pd(fjz3,tz);
1385
1386             }
1387
1388             /**************************
1389              * CALCULATE INTERACTIONS *
1390              **************************/
1391
1392             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1393             {
1394
1395             r31              = _mm256_mul_pd(rsq31,rinv31);
1396             r31              = _mm256_andnot_pd(dummy_mask,r31);
1397
1398             /* EWALD ELECTROSTATICS */
1399
1400             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1401             ewrt             = _mm256_mul_pd(r31,ewtabscale);
1402             ewitab           = _mm256_cvttpd_epi32(ewrt);
1403             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1404             ewitab           = _mm_slli_epi32(ewitab,2);
1405             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1406             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1407             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1408             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1409             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1410             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1411             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1412             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_sub_pd(rinv31,sh_ewald),velec));
1413             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
1414
1415             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1416
1417             /* Update potential sum for this i atom from the interaction with this j atom. */
1418             velec            = _mm256_and_pd(velec,cutoff_mask);
1419             velec            = _mm256_andnot_pd(dummy_mask,velec);
1420             velecsum         = _mm256_add_pd(velecsum,velec);
1421
1422             fscal            = felec;
1423
1424             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1425
1426             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1427
1428             /* Calculate temporary vectorial force */
1429             tx               = _mm256_mul_pd(fscal,dx31);
1430             ty               = _mm256_mul_pd(fscal,dy31);
1431             tz               = _mm256_mul_pd(fscal,dz31);
1432
1433             /* Update vectorial force */
1434             fix3             = _mm256_add_pd(fix3,tx);
1435             fiy3             = _mm256_add_pd(fiy3,ty);
1436             fiz3             = _mm256_add_pd(fiz3,tz);
1437
1438             fjx1             = _mm256_add_pd(fjx1,tx);
1439             fjy1             = _mm256_add_pd(fjy1,ty);
1440             fjz1             = _mm256_add_pd(fjz1,tz);
1441
1442             }
1443
1444             /**************************
1445              * CALCULATE INTERACTIONS *
1446              **************************/
1447
1448             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1449             {
1450
1451             r32              = _mm256_mul_pd(rsq32,rinv32);
1452             r32              = _mm256_andnot_pd(dummy_mask,r32);
1453
1454             /* EWALD ELECTROSTATICS */
1455
1456             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1457             ewrt             = _mm256_mul_pd(r32,ewtabscale);
1458             ewitab           = _mm256_cvttpd_epi32(ewrt);
1459             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1460             ewitab           = _mm_slli_epi32(ewitab,2);
1461             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1462             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1463             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1464             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1465             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1466             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1467             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1468             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_sub_pd(rinv32,sh_ewald),velec));
1469             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
1470
1471             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1472
1473             /* Update potential sum for this i atom from the interaction with this j atom. */
1474             velec            = _mm256_and_pd(velec,cutoff_mask);
1475             velec            = _mm256_andnot_pd(dummy_mask,velec);
1476             velecsum         = _mm256_add_pd(velecsum,velec);
1477
1478             fscal            = felec;
1479
1480             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1481
1482             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1483
1484             /* Calculate temporary vectorial force */
1485             tx               = _mm256_mul_pd(fscal,dx32);
1486             ty               = _mm256_mul_pd(fscal,dy32);
1487             tz               = _mm256_mul_pd(fscal,dz32);
1488
1489             /* Update vectorial force */
1490             fix3             = _mm256_add_pd(fix3,tx);
1491             fiy3             = _mm256_add_pd(fiy3,ty);
1492             fiz3             = _mm256_add_pd(fiz3,tz);
1493
1494             fjx2             = _mm256_add_pd(fjx2,tx);
1495             fjy2             = _mm256_add_pd(fjy2,ty);
1496             fjz2             = _mm256_add_pd(fjz2,tz);
1497
1498             }
1499
1500             /**************************
1501              * CALCULATE INTERACTIONS *
1502              **************************/
1503
1504             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1505             {
1506
1507             r33              = _mm256_mul_pd(rsq33,rinv33);
1508             r33              = _mm256_andnot_pd(dummy_mask,r33);
1509
1510             /* EWALD ELECTROSTATICS */
1511
1512             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1513             ewrt             = _mm256_mul_pd(r33,ewtabscale);
1514             ewitab           = _mm256_cvttpd_epi32(ewrt);
1515             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1516             ewitab           = _mm_slli_epi32(ewitab,2);
1517             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1518             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1519             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1520             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1521             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1522             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1523             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1524             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_sub_pd(rinv33,sh_ewald),velec));
1525             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
1526
1527             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1528
1529             /* Update potential sum for this i atom from the interaction with this j atom. */
1530             velec            = _mm256_and_pd(velec,cutoff_mask);
1531             velec            = _mm256_andnot_pd(dummy_mask,velec);
1532             velecsum         = _mm256_add_pd(velecsum,velec);
1533
1534             fscal            = felec;
1535
1536             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1537
1538             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1539
1540             /* Calculate temporary vectorial force */
1541             tx               = _mm256_mul_pd(fscal,dx33);
1542             ty               = _mm256_mul_pd(fscal,dy33);
1543             tz               = _mm256_mul_pd(fscal,dz33);
1544
1545             /* Update vectorial force */
1546             fix3             = _mm256_add_pd(fix3,tx);
1547             fiy3             = _mm256_add_pd(fiy3,ty);
1548             fiz3             = _mm256_add_pd(fiz3,tz);
1549
1550             fjx3             = _mm256_add_pd(fjx3,tx);
1551             fjy3             = _mm256_add_pd(fjy3,ty);
1552             fjz3             = _mm256_add_pd(fjz3,tz);
1553
1554             }
1555
1556             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1557             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1558             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1559             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1560
1561             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1562                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1563                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1564
1565             /* Inner loop uses 489 flops */
1566         }
1567
1568         /* End of innermost loop */
1569
1570         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1571                                                  f+i_coord_offset,fshift+i_shift_offset);
1572
1573         ggid                        = gid[iidx];
1574         /* Update potential energies */
1575         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1576         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1577
1578         /* Increment number of inner iterations */
1579         inneriter                  += j_index_end - j_index_start;
1580
1581         /* Outer loop uses 26 flops */
1582     }
1583
1584     /* Increment number of outer iterations */
1585     outeriter        += nri;
1586
1587     /* Update outer/inner flops */
1588
1589     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*489);
1590 }
1591 /*
1592  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_F_avx_256_double
1593  * Electrostatics interaction: Ewald
1594  * VdW interaction:            LJEwald
1595  * Geometry:                   Water4-Water4
1596  * Calculate force/pot:        Force
1597  */
1598 void
1599 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_F_avx_256_double
1600                     (t_nblist                    * gmx_restrict       nlist,
1601                      rvec                        * gmx_restrict          xx,
1602                      rvec                        * gmx_restrict          ff,
1603                      struct t_forcerec           * gmx_restrict          fr,
1604                      t_mdatoms                   * gmx_restrict     mdatoms,
1605                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1606                      t_nrnb                      * gmx_restrict        nrnb)
1607 {
1608     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1609      * just 0 for non-waters.
1610      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1611      * jnr indices corresponding to data put in the four positions in the SIMD register.
1612      */
1613     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1614     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1615     int              jnrA,jnrB,jnrC,jnrD;
1616     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1617     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1618     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1619     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1620     real             rcutoff_scalar;
1621     real             *shiftvec,*fshift,*x,*f;
1622     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1623     real             scratch[4*DIM];
1624     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1625     real *           vdwioffsetptr0;
1626     real *           vdwgridioffsetptr0;
1627     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1628     real *           vdwioffsetptr1;
1629     real *           vdwgridioffsetptr1;
1630     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1631     real *           vdwioffsetptr2;
1632     real *           vdwgridioffsetptr2;
1633     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1634     real *           vdwioffsetptr3;
1635     real *           vdwgridioffsetptr3;
1636     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1637     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1638     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1639     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1640     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1641     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1642     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1643     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1644     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1645     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1646     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1647     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1648     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1649     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1650     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1651     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1652     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1653     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1654     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1655     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1656     real             *charge;
1657     int              nvdwtype;
1658     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1659     int              *vdwtype;
1660     real             *vdwparam;
1661     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1662     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1663     __m256d           c6grid_00;
1664     __m256d           c6grid_11;
1665     __m256d           c6grid_12;
1666     __m256d           c6grid_13;
1667     __m256d           c6grid_21;
1668     __m256d           c6grid_22;
1669     __m256d           c6grid_23;
1670     __m256d           c6grid_31;
1671     __m256d           c6grid_32;
1672     __m256d           c6grid_33;
1673     real             *vdwgridparam;
1674     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1675     __m256d           one_half  = _mm256_set1_pd(0.5);
1676     __m256d           minus_one = _mm256_set1_pd(-1.0);
1677     __m128i          ewitab;
1678     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1679     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1680     real             *ewtab;
1681     __m256d          dummy_mask,cutoff_mask;
1682     __m128           tmpmask0,tmpmask1;
1683     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1684     __m256d          one     = _mm256_set1_pd(1.0);
1685     __m256d          two     = _mm256_set1_pd(2.0);
1686     x                = xx[0];
1687     f                = ff[0];
1688
1689     nri              = nlist->nri;
1690     iinr             = nlist->iinr;
1691     jindex           = nlist->jindex;
1692     jjnr             = nlist->jjnr;
1693     shiftidx         = nlist->shift;
1694     gid              = nlist->gid;
1695     shiftvec         = fr->shift_vec[0];
1696     fshift           = fr->fshift[0];
1697     facel            = _mm256_set1_pd(fr->ic->epsfac);
1698     charge           = mdatoms->chargeA;
1699     nvdwtype         = fr->ntype;
1700     vdwparam         = fr->nbfp;
1701     vdwtype          = mdatoms->typeA;
1702     vdwgridparam     = fr->ljpme_c6grid;
1703     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
1704     ewclj            = _mm256_set1_pd(fr->ic->ewaldcoeff_lj);
1705     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
1706
1707     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1708     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
1709     beta2            = _mm256_mul_pd(beta,beta);
1710     beta3            = _mm256_mul_pd(beta,beta2);
1711
1712     ewtab            = fr->ic->tabq_coul_F;
1713     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1714     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1715
1716     /* Setup water-specific parameters */
1717     inr              = nlist->iinr[0];
1718     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1719     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1720     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1721     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1722     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
1723
1724     jq1              = _mm256_set1_pd(charge[inr+1]);
1725     jq2              = _mm256_set1_pd(charge[inr+2]);
1726     jq3              = _mm256_set1_pd(charge[inr+3]);
1727     vdwjidx0A        = 2*vdwtype[inr+0];
1728     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1729     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1730     c6grid_00        = _mm256_set1_pd(vdwgridioffsetptr0[vdwjidx0A]);
1731     qq11             = _mm256_mul_pd(iq1,jq1);
1732     qq12             = _mm256_mul_pd(iq1,jq2);
1733     qq13             = _mm256_mul_pd(iq1,jq3);
1734     qq21             = _mm256_mul_pd(iq2,jq1);
1735     qq22             = _mm256_mul_pd(iq2,jq2);
1736     qq23             = _mm256_mul_pd(iq2,jq3);
1737     qq31             = _mm256_mul_pd(iq3,jq1);
1738     qq32             = _mm256_mul_pd(iq3,jq2);
1739     qq33             = _mm256_mul_pd(iq3,jq3);
1740
1741     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1742     rcutoff_scalar   = fr->ic->rcoulomb;
1743     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1744     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1745
1746     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
1747     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
1748
1749     /* Avoid stupid compiler warnings */
1750     jnrA = jnrB = jnrC = jnrD = 0;
1751     j_coord_offsetA = 0;
1752     j_coord_offsetB = 0;
1753     j_coord_offsetC = 0;
1754     j_coord_offsetD = 0;
1755
1756     outeriter        = 0;
1757     inneriter        = 0;
1758
1759     for(iidx=0;iidx<4*DIM;iidx++)
1760     {
1761         scratch[iidx] = 0.0;
1762     }
1763
1764     /* Start outer loop over neighborlists */
1765     for(iidx=0; iidx<nri; iidx++)
1766     {
1767         /* Load shift vector for this list */
1768         i_shift_offset   = DIM*shiftidx[iidx];
1769
1770         /* Load limits for loop over neighbors */
1771         j_index_start    = jindex[iidx];
1772         j_index_end      = jindex[iidx+1];
1773
1774         /* Get outer coordinate index */
1775         inr              = iinr[iidx];
1776         i_coord_offset   = DIM*inr;
1777
1778         /* Load i particle coords and add shift vector */
1779         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1780                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1781
1782         fix0             = _mm256_setzero_pd();
1783         fiy0             = _mm256_setzero_pd();
1784         fiz0             = _mm256_setzero_pd();
1785         fix1             = _mm256_setzero_pd();
1786         fiy1             = _mm256_setzero_pd();
1787         fiz1             = _mm256_setzero_pd();
1788         fix2             = _mm256_setzero_pd();
1789         fiy2             = _mm256_setzero_pd();
1790         fiz2             = _mm256_setzero_pd();
1791         fix3             = _mm256_setzero_pd();
1792         fiy3             = _mm256_setzero_pd();
1793         fiz3             = _mm256_setzero_pd();
1794
1795         /* Start inner kernel loop */
1796         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1797         {
1798
1799             /* Get j neighbor index, and coordinate index */
1800             jnrA             = jjnr[jidx];
1801             jnrB             = jjnr[jidx+1];
1802             jnrC             = jjnr[jidx+2];
1803             jnrD             = jjnr[jidx+3];
1804             j_coord_offsetA  = DIM*jnrA;
1805             j_coord_offsetB  = DIM*jnrB;
1806             j_coord_offsetC  = DIM*jnrC;
1807             j_coord_offsetD  = DIM*jnrD;
1808
1809             /* load j atom coordinates */
1810             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1811                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1812                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1813                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1814
1815             /* Calculate displacement vector */
1816             dx00             = _mm256_sub_pd(ix0,jx0);
1817             dy00             = _mm256_sub_pd(iy0,jy0);
1818             dz00             = _mm256_sub_pd(iz0,jz0);
1819             dx11             = _mm256_sub_pd(ix1,jx1);
1820             dy11             = _mm256_sub_pd(iy1,jy1);
1821             dz11             = _mm256_sub_pd(iz1,jz1);
1822             dx12             = _mm256_sub_pd(ix1,jx2);
1823             dy12             = _mm256_sub_pd(iy1,jy2);
1824             dz12             = _mm256_sub_pd(iz1,jz2);
1825             dx13             = _mm256_sub_pd(ix1,jx3);
1826             dy13             = _mm256_sub_pd(iy1,jy3);
1827             dz13             = _mm256_sub_pd(iz1,jz3);
1828             dx21             = _mm256_sub_pd(ix2,jx1);
1829             dy21             = _mm256_sub_pd(iy2,jy1);
1830             dz21             = _mm256_sub_pd(iz2,jz1);
1831             dx22             = _mm256_sub_pd(ix2,jx2);
1832             dy22             = _mm256_sub_pd(iy2,jy2);
1833             dz22             = _mm256_sub_pd(iz2,jz2);
1834             dx23             = _mm256_sub_pd(ix2,jx3);
1835             dy23             = _mm256_sub_pd(iy2,jy3);
1836             dz23             = _mm256_sub_pd(iz2,jz3);
1837             dx31             = _mm256_sub_pd(ix3,jx1);
1838             dy31             = _mm256_sub_pd(iy3,jy1);
1839             dz31             = _mm256_sub_pd(iz3,jz1);
1840             dx32             = _mm256_sub_pd(ix3,jx2);
1841             dy32             = _mm256_sub_pd(iy3,jy2);
1842             dz32             = _mm256_sub_pd(iz3,jz2);
1843             dx33             = _mm256_sub_pd(ix3,jx3);
1844             dy33             = _mm256_sub_pd(iy3,jy3);
1845             dz33             = _mm256_sub_pd(iz3,jz3);
1846
1847             /* Calculate squared distance and things based on it */
1848             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1849             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1850             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1851             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1852             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1853             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1854             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1855             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1856             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1857             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1858
1859             rinv00           = avx256_invsqrt_d(rsq00);
1860             rinv11           = avx256_invsqrt_d(rsq11);
1861             rinv12           = avx256_invsqrt_d(rsq12);
1862             rinv13           = avx256_invsqrt_d(rsq13);
1863             rinv21           = avx256_invsqrt_d(rsq21);
1864             rinv22           = avx256_invsqrt_d(rsq22);
1865             rinv23           = avx256_invsqrt_d(rsq23);
1866             rinv31           = avx256_invsqrt_d(rsq31);
1867             rinv32           = avx256_invsqrt_d(rsq32);
1868             rinv33           = avx256_invsqrt_d(rsq33);
1869
1870             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1871             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1872             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1873             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1874             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1875             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1876             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1877             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1878             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1879             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1880
1881             fjx0             = _mm256_setzero_pd();
1882             fjy0             = _mm256_setzero_pd();
1883             fjz0             = _mm256_setzero_pd();
1884             fjx1             = _mm256_setzero_pd();
1885             fjy1             = _mm256_setzero_pd();
1886             fjz1             = _mm256_setzero_pd();
1887             fjx2             = _mm256_setzero_pd();
1888             fjy2             = _mm256_setzero_pd();
1889             fjz2             = _mm256_setzero_pd();
1890             fjx3             = _mm256_setzero_pd();
1891             fjy3             = _mm256_setzero_pd();
1892             fjz3             = _mm256_setzero_pd();
1893
1894             /**************************
1895              * CALCULATE INTERACTIONS *
1896              **************************/
1897
1898             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1899             {
1900
1901             r00              = _mm256_mul_pd(rsq00,rinv00);
1902
1903             /* Analytical LJ-PME */
1904             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1905             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1906             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1907             exponent         = avx256_exp_d(ewcljrsq);
1908             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1909             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1910             /* f6A = 6 * C6grid * (1 - poly) */
1911             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1912             /* f6B = C6grid * exponent * beta^6 */
1913             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1914             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1915             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1916
1917             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1918
1919             fscal            = fvdw;
1920
1921             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1922
1923             /* Calculate temporary vectorial force */
1924             tx               = _mm256_mul_pd(fscal,dx00);
1925             ty               = _mm256_mul_pd(fscal,dy00);
1926             tz               = _mm256_mul_pd(fscal,dz00);
1927
1928             /* Update vectorial force */
1929             fix0             = _mm256_add_pd(fix0,tx);
1930             fiy0             = _mm256_add_pd(fiy0,ty);
1931             fiz0             = _mm256_add_pd(fiz0,tz);
1932
1933             fjx0             = _mm256_add_pd(fjx0,tx);
1934             fjy0             = _mm256_add_pd(fjy0,ty);
1935             fjz0             = _mm256_add_pd(fjz0,tz);
1936
1937             }
1938
1939             /**************************
1940              * CALCULATE INTERACTIONS *
1941              **************************/
1942
1943             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1944             {
1945
1946             r11              = _mm256_mul_pd(rsq11,rinv11);
1947
1948             /* EWALD ELECTROSTATICS */
1949
1950             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1951             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1952             ewitab           = _mm256_cvttpd_epi32(ewrt);
1953             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1954             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1955                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1956                                             &ewtabF,&ewtabFn);
1957             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1958             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1959
1960             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1961
1962             fscal            = felec;
1963
1964             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1965
1966             /* Calculate temporary vectorial force */
1967             tx               = _mm256_mul_pd(fscal,dx11);
1968             ty               = _mm256_mul_pd(fscal,dy11);
1969             tz               = _mm256_mul_pd(fscal,dz11);
1970
1971             /* Update vectorial force */
1972             fix1             = _mm256_add_pd(fix1,tx);
1973             fiy1             = _mm256_add_pd(fiy1,ty);
1974             fiz1             = _mm256_add_pd(fiz1,tz);
1975
1976             fjx1             = _mm256_add_pd(fjx1,tx);
1977             fjy1             = _mm256_add_pd(fjy1,ty);
1978             fjz1             = _mm256_add_pd(fjz1,tz);
1979
1980             }
1981
1982             /**************************
1983              * CALCULATE INTERACTIONS *
1984              **************************/
1985
1986             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1987             {
1988
1989             r12              = _mm256_mul_pd(rsq12,rinv12);
1990
1991             /* EWALD ELECTROSTATICS */
1992
1993             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1994             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1995             ewitab           = _mm256_cvttpd_epi32(ewrt);
1996             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1997             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1998                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1999                                             &ewtabF,&ewtabFn);
2000             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2001             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2002
2003             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2004
2005             fscal            = felec;
2006
2007             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2008
2009             /* Calculate temporary vectorial force */
2010             tx               = _mm256_mul_pd(fscal,dx12);
2011             ty               = _mm256_mul_pd(fscal,dy12);
2012             tz               = _mm256_mul_pd(fscal,dz12);
2013
2014             /* Update vectorial force */
2015             fix1             = _mm256_add_pd(fix1,tx);
2016             fiy1             = _mm256_add_pd(fiy1,ty);
2017             fiz1             = _mm256_add_pd(fiz1,tz);
2018
2019             fjx2             = _mm256_add_pd(fjx2,tx);
2020             fjy2             = _mm256_add_pd(fjy2,ty);
2021             fjz2             = _mm256_add_pd(fjz2,tz);
2022
2023             }
2024
2025             /**************************
2026              * CALCULATE INTERACTIONS *
2027              **************************/
2028
2029             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2030             {
2031
2032             r13              = _mm256_mul_pd(rsq13,rinv13);
2033
2034             /* EWALD ELECTROSTATICS */
2035
2036             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2037             ewrt             = _mm256_mul_pd(r13,ewtabscale);
2038             ewitab           = _mm256_cvttpd_epi32(ewrt);
2039             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2040             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2041                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2042                                             &ewtabF,&ewtabFn);
2043             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2044             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
2045
2046             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
2047
2048             fscal            = felec;
2049
2050             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2051
2052             /* Calculate temporary vectorial force */
2053             tx               = _mm256_mul_pd(fscal,dx13);
2054             ty               = _mm256_mul_pd(fscal,dy13);
2055             tz               = _mm256_mul_pd(fscal,dz13);
2056
2057             /* Update vectorial force */
2058             fix1             = _mm256_add_pd(fix1,tx);
2059             fiy1             = _mm256_add_pd(fiy1,ty);
2060             fiz1             = _mm256_add_pd(fiz1,tz);
2061
2062             fjx3             = _mm256_add_pd(fjx3,tx);
2063             fjy3             = _mm256_add_pd(fjy3,ty);
2064             fjz3             = _mm256_add_pd(fjz3,tz);
2065
2066             }
2067
2068             /**************************
2069              * CALCULATE INTERACTIONS *
2070              **************************/
2071
2072             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2073             {
2074
2075             r21              = _mm256_mul_pd(rsq21,rinv21);
2076
2077             /* EWALD ELECTROSTATICS */
2078
2079             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2080             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2081             ewitab           = _mm256_cvttpd_epi32(ewrt);
2082             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2083             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2084                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2085                                             &ewtabF,&ewtabFn);
2086             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2087             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2088
2089             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2090
2091             fscal            = felec;
2092
2093             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2094
2095             /* Calculate temporary vectorial force */
2096             tx               = _mm256_mul_pd(fscal,dx21);
2097             ty               = _mm256_mul_pd(fscal,dy21);
2098             tz               = _mm256_mul_pd(fscal,dz21);
2099
2100             /* Update vectorial force */
2101             fix2             = _mm256_add_pd(fix2,tx);
2102             fiy2             = _mm256_add_pd(fiy2,ty);
2103             fiz2             = _mm256_add_pd(fiz2,tz);
2104
2105             fjx1             = _mm256_add_pd(fjx1,tx);
2106             fjy1             = _mm256_add_pd(fjy1,ty);
2107             fjz1             = _mm256_add_pd(fjz1,tz);
2108
2109             }
2110
2111             /**************************
2112              * CALCULATE INTERACTIONS *
2113              **************************/
2114
2115             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2116             {
2117
2118             r22              = _mm256_mul_pd(rsq22,rinv22);
2119
2120             /* EWALD ELECTROSTATICS */
2121
2122             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2123             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2124             ewitab           = _mm256_cvttpd_epi32(ewrt);
2125             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2126             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2127                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2128                                             &ewtabF,&ewtabFn);
2129             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2130             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2131
2132             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2133
2134             fscal            = felec;
2135
2136             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2137
2138             /* Calculate temporary vectorial force */
2139             tx               = _mm256_mul_pd(fscal,dx22);
2140             ty               = _mm256_mul_pd(fscal,dy22);
2141             tz               = _mm256_mul_pd(fscal,dz22);
2142
2143             /* Update vectorial force */
2144             fix2             = _mm256_add_pd(fix2,tx);
2145             fiy2             = _mm256_add_pd(fiy2,ty);
2146             fiz2             = _mm256_add_pd(fiz2,tz);
2147
2148             fjx2             = _mm256_add_pd(fjx2,tx);
2149             fjy2             = _mm256_add_pd(fjy2,ty);
2150             fjz2             = _mm256_add_pd(fjz2,tz);
2151
2152             }
2153
2154             /**************************
2155              * CALCULATE INTERACTIONS *
2156              **************************/
2157
2158             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2159             {
2160
2161             r23              = _mm256_mul_pd(rsq23,rinv23);
2162
2163             /* EWALD ELECTROSTATICS */
2164
2165             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2166             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2167             ewitab           = _mm256_cvttpd_epi32(ewrt);
2168             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2169             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2170                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2171                                             &ewtabF,&ewtabFn);
2172             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2173             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2174
2175             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2176
2177             fscal            = felec;
2178
2179             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2180
2181             /* Calculate temporary vectorial force */
2182             tx               = _mm256_mul_pd(fscal,dx23);
2183             ty               = _mm256_mul_pd(fscal,dy23);
2184             tz               = _mm256_mul_pd(fscal,dz23);
2185
2186             /* Update vectorial force */
2187             fix2             = _mm256_add_pd(fix2,tx);
2188             fiy2             = _mm256_add_pd(fiy2,ty);
2189             fiz2             = _mm256_add_pd(fiz2,tz);
2190
2191             fjx3             = _mm256_add_pd(fjx3,tx);
2192             fjy3             = _mm256_add_pd(fjy3,ty);
2193             fjz3             = _mm256_add_pd(fjz3,tz);
2194
2195             }
2196
2197             /**************************
2198              * CALCULATE INTERACTIONS *
2199              **************************/
2200
2201             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2202             {
2203
2204             r31              = _mm256_mul_pd(rsq31,rinv31);
2205
2206             /* EWALD ELECTROSTATICS */
2207
2208             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2209             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2210             ewitab           = _mm256_cvttpd_epi32(ewrt);
2211             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2212             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2213                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2214                                             &ewtabF,&ewtabFn);
2215             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2216             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2217
2218             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2219
2220             fscal            = felec;
2221
2222             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2223
2224             /* Calculate temporary vectorial force */
2225             tx               = _mm256_mul_pd(fscal,dx31);
2226             ty               = _mm256_mul_pd(fscal,dy31);
2227             tz               = _mm256_mul_pd(fscal,dz31);
2228
2229             /* Update vectorial force */
2230             fix3             = _mm256_add_pd(fix3,tx);
2231             fiy3             = _mm256_add_pd(fiy3,ty);
2232             fiz3             = _mm256_add_pd(fiz3,tz);
2233
2234             fjx1             = _mm256_add_pd(fjx1,tx);
2235             fjy1             = _mm256_add_pd(fjy1,ty);
2236             fjz1             = _mm256_add_pd(fjz1,tz);
2237
2238             }
2239
2240             /**************************
2241              * CALCULATE INTERACTIONS *
2242              **************************/
2243
2244             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2245             {
2246
2247             r32              = _mm256_mul_pd(rsq32,rinv32);
2248
2249             /* EWALD ELECTROSTATICS */
2250
2251             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2252             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2253             ewitab           = _mm256_cvttpd_epi32(ewrt);
2254             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2255             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2256                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2257                                             &ewtabF,&ewtabFn);
2258             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2259             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
2260
2261             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2262
2263             fscal            = felec;
2264
2265             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2266
2267             /* Calculate temporary vectorial force */
2268             tx               = _mm256_mul_pd(fscal,dx32);
2269             ty               = _mm256_mul_pd(fscal,dy32);
2270             tz               = _mm256_mul_pd(fscal,dz32);
2271
2272             /* Update vectorial force */
2273             fix3             = _mm256_add_pd(fix3,tx);
2274             fiy3             = _mm256_add_pd(fiy3,ty);
2275             fiz3             = _mm256_add_pd(fiz3,tz);
2276
2277             fjx2             = _mm256_add_pd(fjx2,tx);
2278             fjy2             = _mm256_add_pd(fjy2,ty);
2279             fjz2             = _mm256_add_pd(fjz2,tz);
2280
2281             }
2282
2283             /**************************
2284              * CALCULATE INTERACTIONS *
2285              **************************/
2286
2287             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2288             {
2289
2290             r33              = _mm256_mul_pd(rsq33,rinv33);
2291
2292             /* EWALD ELECTROSTATICS */
2293
2294             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2295             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2296             ewitab           = _mm256_cvttpd_epi32(ewrt);
2297             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2298             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2299                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2300                                             &ewtabF,&ewtabFn);
2301             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2302             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2303
2304             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2305
2306             fscal            = felec;
2307
2308             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2309
2310             /* Calculate temporary vectorial force */
2311             tx               = _mm256_mul_pd(fscal,dx33);
2312             ty               = _mm256_mul_pd(fscal,dy33);
2313             tz               = _mm256_mul_pd(fscal,dz33);
2314
2315             /* Update vectorial force */
2316             fix3             = _mm256_add_pd(fix3,tx);
2317             fiy3             = _mm256_add_pd(fiy3,ty);
2318             fiz3             = _mm256_add_pd(fiz3,tz);
2319
2320             fjx3             = _mm256_add_pd(fjx3,tx);
2321             fjy3             = _mm256_add_pd(fjy3,ty);
2322             fjz3             = _mm256_add_pd(fjz3,tz);
2323
2324             }
2325
2326             fjptrA             = f+j_coord_offsetA;
2327             fjptrB             = f+j_coord_offsetB;
2328             fjptrC             = f+j_coord_offsetC;
2329             fjptrD             = f+j_coord_offsetD;
2330
2331             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2332                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2333                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2334
2335             /* Inner loop uses 403 flops */
2336         }
2337
2338         if(jidx<j_index_end)
2339         {
2340
2341             /* Get j neighbor index, and coordinate index */
2342             jnrlistA         = jjnr[jidx];
2343             jnrlistB         = jjnr[jidx+1];
2344             jnrlistC         = jjnr[jidx+2];
2345             jnrlistD         = jjnr[jidx+3];
2346             /* Sign of each element will be negative for non-real atoms.
2347              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2348              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
2349              */
2350             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2351
2352             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
2353             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
2354             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
2355
2356             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2357             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2358             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2359             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2360             j_coord_offsetA  = DIM*jnrA;
2361             j_coord_offsetB  = DIM*jnrB;
2362             j_coord_offsetC  = DIM*jnrC;
2363             j_coord_offsetD  = DIM*jnrD;
2364
2365             /* load j atom coordinates */
2366             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
2367                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2368                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2369                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2370
2371             /* Calculate displacement vector */
2372             dx00             = _mm256_sub_pd(ix0,jx0);
2373             dy00             = _mm256_sub_pd(iy0,jy0);
2374             dz00             = _mm256_sub_pd(iz0,jz0);
2375             dx11             = _mm256_sub_pd(ix1,jx1);
2376             dy11             = _mm256_sub_pd(iy1,jy1);
2377             dz11             = _mm256_sub_pd(iz1,jz1);
2378             dx12             = _mm256_sub_pd(ix1,jx2);
2379             dy12             = _mm256_sub_pd(iy1,jy2);
2380             dz12             = _mm256_sub_pd(iz1,jz2);
2381             dx13             = _mm256_sub_pd(ix1,jx3);
2382             dy13             = _mm256_sub_pd(iy1,jy3);
2383             dz13             = _mm256_sub_pd(iz1,jz3);
2384             dx21             = _mm256_sub_pd(ix2,jx1);
2385             dy21             = _mm256_sub_pd(iy2,jy1);
2386             dz21             = _mm256_sub_pd(iz2,jz1);
2387             dx22             = _mm256_sub_pd(ix2,jx2);
2388             dy22             = _mm256_sub_pd(iy2,jy2);
2389             dz22             = _mm256_sub_pd(iz2,jz2);
2390             dx23             = _mm256_sub_pd(ix2,jx3);
2391             dy23             = _mm256_sub_pd(iy2,jy3);
2392             dz23             = _mm256_sub_pd(iz2,jz3);
2393             dx31             = _mm256_sub_pd(ix3,jx1);
2394             dy31             = _mm256_sub_pd(iy3,jy1);
2395             dz31             = _mm256_sub_pd(iz3,jz1);
2396             dx32             = _mm256_sub_pd(ix3,jx2);
2397             dy32             = _mm256_sub_pd(iy3,jy2);
2398             dz32             = _mm256_sub_pd(iz3,jz2);
2399             dx33             = _mm256_sub_pd(ix3,jx3);
2400             dy33             = _mm256_sub_pd(iy3,jy3);
2401             dz33             = _mm256_sub_pd(iz3,jz3);
2402
2403             /* Calculate squared distance and things based on it */
2404             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
2405             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2406             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2407             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
2408             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2409             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2410             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
2411             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
2412             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
2413             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
2414
2415             rinv00           = avx256_invsqrt_d(rsq00);
2416             rinv11           = avx256_invsqrt_d(rsq11);
2417             rinv12           = avx256_invsqrt_d(rsq12);
2418             rinv13           = avx256_invsqrt_d(rsq13);
2419             rinv21           = avx256_invsqrt_d(rsq21);
2420             rinv22           = avx256_invsqrt_d(rsq22);
2421             rinv23           = avx256_invsqrt_d(rsq23);
2422             rinv31           = avx256_invsqrt_d(rsq31);
2423             rinv32           = avx256_invsqrt_d(rsq32);
2424             rinv33           = avx256_invsqrt_d(rsq33);
2425
2426             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
2427             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
2428             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
2429             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
2430             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
2431             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
2432             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
2433             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
2434             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
2435             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
2436
2437             fjx0             = _mm256_setzero_pd();
2438             fjy0             = _mm256_setzero_pd();
2439             fjz0             = _mm256_setzero_pd();
2440             fjx1             = _mm256_setzero_pd();
2441             fjy1             = _mm256_setzero_pd();
2442             fjz1             = _mm256_setzero_pd();
2443             fjx2             = _mm256_setzero_pd();
2444             fjy2             = _mm256_setzero_pd();
2445             fjz2             = _mm256_setzero_pd();
2446             fjx3             = _mm256_setzero_pd();
2447             fjy3             = _mm256_setzero_pd();
2448             fjz3             = _mm256_setzero_pd();
2449
2450             /**************************
2451              * CALCULATE INTERACTIONS *
2452              **************************/
2453
2454             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2455             {
2456
2457             r00              = _mm256_mul_pd(rsq00,rinv00);
2458             r00              = _mm256_andnot_pd(dummy_mask,r00);
2459
2460             /* Analytical LJ-PME */
2461             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2462             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
2463             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
2464             exponent         = avx256_exp_d(ewcljrsq);
2465             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2466             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
2467             /* f6A = 6 * C6grid * (1 - poly) */
2468             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
2469             /* f6B = C6grid * exponent * beta^6 */
2470             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
2471             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2472             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
2473
2474             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
2475
2476             fscal            = fvdw;
2477
2478             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2479
2480             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2481
2482             /* Calculate temporary vectorial force */
2483             tx               = _mm256_mul_pd(fscal,dx00);
2484             ty               = _mm256_mul_pd(fscal,dy00);
2485             tz               = _mm256_mul_pd(fscal,dz00);
2486
2487             /* Update vectorial force */
2488             fix0             = _mm256_add_pd(fix0,tx);
2489             fiy0             = _mm256_add_pd(fiy0,ty);
2490             fiz0             = _mm256_add_pd(fiz0,tz);
2491
2492             fjx0             = _mm256_add_pd(fjx0,tx);
2493             fjy0             = _mm256_add_pd(fjy0,ty);
2494             fjz0             = _mm256_add_pd(fjz0,tz);
2495
2496             }
2497
2498             /**************************
2499              * CALCULATE INTERACTIONS *
2500              **************************/
2501
2502             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2503             {
2504
2505             r11              = _mm256_mul_pd(rsq11,rinv11);
2506             r11              = _mm256_andnot_pd(dummy_mask,r11);
2507
2508             /* EWALD ELECTROSTATICS */
2509
2510             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2511             ewrt             = _mm256_mul_pd(r11,ewtabscale);
2512             ewitab           = _mm256_cvttpd_epi32(ewrt);
2513             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2514             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2515                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2516                                             &ewtabF,&ewtabFn);
2517             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2518             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2519
2520             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2521
2522             fscal            = felec;
2523
2524             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2525
2526             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2527
2528             /* Calculate temporary vectorial force */
2529             tx               = _mm256_mul_pd(fscal,dx11);
2530             ty               = _mm256_mul_pd(fscal,dy11);
2531             tz               = _mm256_mul_pd(fscal,dz11);
2532
2533             /* Update vectorial force */
2534             fix1             = _mm256_add_pd(fix1,tx);
2535             fiy1             = _mm256_add_pd(fiy1,ty);
2536             fiz1             = _mm256_add_pd(fiz1,tz);
2537
2538             fjx1             = _mm256_add_pd(fjx1,tx);
2539             fjy1             = _mm256_add_pd(fjy1,ty);
2540             fjz1             = _mm256_add_pd(fjz1,tz);
2541
2542             }
2543
2544             /**************************
2545              * CALCULATE INTERACTIONS *
2546              **************************/
2547
2548             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2549             {
2550
2551             r12              = _mm256_mul_pd(rsq12,rinv12);
2552             r12              = _mm256_andnot_pd(dummy_mask,r12);
2553
2554             /* EWALD ELECTROSTATICS */
2555
2556             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2557             ewrt             = _mm256_mul_pd(r12,ewtabscale);
2558             ewitab           = _mm256_cvttpd_epi32(ewrt);
2559             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2560             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2561                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2562                                             &ewtabF,&ewtabFn);
2563             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2564             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2565
2566             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2567
2568             fscal            = felec;
2569
2570             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2571
2572             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2573
2574             /* Calculate temporary vectorial force */
2575             tx               = _mm256_mul_pd(fscal,dx12);
2576             ty               = _mm256_mul_pd(fscal,dy12);
2577             tz               = _mm256_mul_pd(fscal,dz12);
2578
2579             /* Update vectorial force */
2580             fix1             = _mm256_add_pd(fix1,tx);
2581             fiy1             = _mm256_add_pd(fiy1,ty);
2582             fiz1             = _mm256_add_pd(fiz1,tz);
2583
2584             fjx2             = _mm256_add_pd(fjx2,tx);
2585             fjy2             = _mm256_add_pd(fjy2,ty);
2586             fjz2             = _mm256_add_pd(fjz2,tz);
2587
2588             }
2589
2590             /**************************
2591              * CALCULATE INTERACTIONS *
2592              **************************/
2593
2594             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2595             {
2596
2597             r13              = _mm256_mul_pd(rsq13,rinv13);
2598             r13              = _mm256_andnot_pd(dummy_mask,r13);
2599
2600             /* EWALD ELECTROSTATICS */
2601
2602             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2603             ewrt             = _mm256_mul_pd(r13,ewtabscale);
2604             ewitab           = _mm256_cvttpd_epi32(ewrt);
2605             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2606             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2607                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2608                                             &ewtabF,&ewtabFn);
2609             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2610             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
2611
2612             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
2613
2614             fscal            = felec;
2615
2616             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2617
2618             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2619
2620             /* Calculate temporary vectorial force */
2621             tx               = _mm256_mul_pd(fscal,dx13);
2622             ty               = _mm256_mul_pd(fscal,dy13);
2623             tz               = _mm256_mul_pd(fscal,dz13);
2624
2625             /* Update vectorial force */
2626             fix1             = _mm256_add_pd(fix1,tx);
2627             fiy1             = _mm256_add_pd(fiy1,ty);
2628             fiz1             = _mm256_add_pd(fiz1,tz);
2629
2630             fjx3             = _mm256_add_pd(fjx3,tx);
2631             fjy3             = _mm256_add_pd(fjy3,ty);
2632             fjz3             = _mm256_add_pd(fjz3,tz);
2633
2634             }
2635
2636             /**************************
2637              * CALCULATE INTERACTIONS *
2638              **************************/
2639
2640             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2641             {
2642
2643             r21              = _mm256_mul_pd(rsq21,rinv21);
2644             r21              = _mm256_andnot_pd(dummy_mask,r21);
2645
2646             /* EWALD ELECTROSTATICS */
2647
2648             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2649             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2650             ewitab           = _mm256_cvttpd_epi32(ewrt);
2651             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2652             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2653                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2654                                             &ewtabF,&ewtabFn);
2655             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2656             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2657
2658             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2659
2660             fscal            = felec;
2661
2662             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2663
2664             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2665
2666             /* Calculate temporary vectorial force */
2667             tx               = _mm256_mul_pd(fscal,dx21);
2668             ty               = _mm256_mul_pd(fscal,dy21);
2669             tz               = _mm256_mul_pd(fscal,dz21);
2670
2671             /* Update vectorial force */
2672             fix2             = _mm256_add_pd(fix2,tx);
2673             fiy2             = _mm256_add_pd(fiy2,ty);
2674             fiz2             = _mm256_add_pd(fiz2,tz);
2675
2676             fjx1             = _mm256_add_pd(fjx1,tx);
2677             fjy1             = _mm256_add_pd(fjy1,ty);
2678             fjz1             = _mm256_add_pd(fjz1,tz);
2679
2680             }
2681
2682             /**************************
2683              * CALCULATE INTERACTIONS *
2684              **************************/
2685
2686             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2687             {
2688
2689             r22              = _mm256_mul_pd(rsq22,rinv22);
2690             r22              = _mm256_andnot_pd(dummy_mask,r22);
2691
2692             /* EWALD ELECTROSTATICS */
2693
2694             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2695             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2696             ewitab           = _mm256_cvttpd_epi32(ewrt);
2697             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2698             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2699                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2700                                             &ewtabF,&ewtabFn);
2701             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2702             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2703
2704             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2705
2706             fscal            = felec;
2707
2708             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2709
2710             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2711
2712             /* Calculate temporary vectorial force */
2713             tx               = _mm256_mul_pd(fscal,dx22);
2714             ty               = _mm256_mul_pd(fscal,dy22);
2715             tz               = _mm256_mul_pd(fscal,dz22);
2716
2717             /* Update vectorial force */
2718             fix2             = _mm256_add_pd(fix2,tx);
2719             fiy2             = _mm256_add_pd(fiy2,ty);
2720             fiz2             = _mm256_add_pd(fiz2,tz);
2721
2722             fjx2             = _mm256_add_pd(fjx2,tx);
2723             fjy2             = _mm256_add_pd(fjy2,ty);
2724             fjz2             = _mm256_add_pd(fjz2,tz);
2725
2726             }
2727
2728             /**************************
2729              * CALCULATE INTERACTIONS *
2730              **************************/
2731
2732             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2733             {
2734
2735             r23              = _mm256_mul_pd(rsq23,rinv23);
2736             r23              = _mm256_andnot_pd(dummy_mask,r23);
2737
2738             /* EWALD ELECTROSTATICS */
2739
2740             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2741             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2742             ewitab           = _mm256_cvttpd_epi32(ewrt);
2743             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2744             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2745                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2746                                             &ewtabF,&ewtabFn);
2747             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2748             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2749
2750             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2751
2752             fscal            = felec;
2753
2754             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2755
2756             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2757
2758             /* Calculate temporary vectorial force */
2759             tx               = _mm256_mul_pd(fscal,dx23);
2760             ty               = _mm256_mul_pd(fscal,dy23);
2761             tz               = _mm256_mul_pd(fscal,dz23);
2762
2763             /* Update vectorial force */
2764             fix2             = _mm256_add_pd(fix2,tx);
2765             fiy2             = _mm256_add_pd(fiy2,ty);
2766             fiz2             = _mm256_add_pd(fiz2,tz);
2767
2768             fjx3             = _mm256_add_pd(fjx3,tx);
2769             fjy3             = _mm256_add_pd(fjy3,ty);
2770             fjz3             = _mm256_add_pd(fjz3,tz);
2771
2772             }
2773
2774             /**************************
2775              * CALCULATE INTERACTIONS *
2776              **************************/
2777
2778             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2779             {
2780
2781             r31              = _mm256_mul_pd(rsq31,rinv31);
2782             r31              = _mm256_andnot_pd(dummy_mask,r31);
2783
2784             /* EWALD ELECTROSTATICS */
2785
2786             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2787             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2788             ewitab           = _mm256_cvttpd_epi32(ewrt);
2789             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2790             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2791                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2792                                             &ewtabF,&ewtabFn);
2793             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2794             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2795
2796             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2797
2798             fscal            = felec;
2799
2800             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2801
2802             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2803
2804             /* Calculate temporary vectorial force */
2805             tx               = _mm256_mul_pd(fscal,dx31);
2806             ty               = _mm256_mul_pd(fscal,dy31);
2807             tz               = _mm256_mul_pd(fscal,dz31);
2808
2809             /* Update vectorial force */
2810             fix3             = _mm256_add_pd(fix3,tx);
2811             fiy3             = _mm256_add_pd(fiy3,ty);
2812             fiz3             = _mm256_add_pd(fiz3,tz);
2813
2814             fjx1             = _mm256_add_pd(fjx1,tx);
2815             fjy1             = _mm256_add_pd(fjy1,ty);
2816             fjz1             = _mm256_add_pd(fjz1,tz);
2817
2818             }
2819
2820             /**************************
2821              * CALCULATE INTERACTIONS *
2822              **************************/
2823
2824             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2825             {
2826
2827             r32              = _mm256_mul_pd(rsq32,rinv32);
2828             r32              = _mm256_andnot_pd(dummy_mask,r32);
2829
2830             /* EWALD ELECTROSTATICS */
2831
2832             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2833             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2834             ewitab           = _mm256_cvttpd_epi32(ewrt);
2835             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2836             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2837                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2838                                             &ewtabF,&ewtabFn);
2839             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2840             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
2841
2842             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2843
2844             fscal            = felec;
2845
2846             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2847
2848             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2849
2850             /* Calculate temporary vectorial force */
2851             tx               = _mm256_mul_pd(fscal,dx32);
2852             ty               = _mm256_mul_pd(fscal,dy32);
2853             tz               = _mm256_mul_pd(fscal,dz32);
2854
2855             /* Update vectorial force */
2856             fix3             = _mm256_add_pd(fix3,tx);
2857             fiy3             = _mm256_add_pd(fiy3,ty);
2858             fiz3             = _mm256_add_pd(fiz3,tz);
2859
2860             fjx2             = _mm256_add_pd(fjx2,tx);
2861             fjy2             = _mm256_add_pd(fjy2,ty);
2862             fjz2             = _mm256_add_pd(fjz2,tz);
2863
2864             }
2865
2866             /**************************
2867              * CALCULATE INTERACTIONS *
2868              **************************/
2869
2870             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2871             {
2872
2873             r33              = _mm256_mul_pd(rsq33,rinv33);
2874             r33              = _mm256_andnot_pd(dummy_mask,r33);
2875
2876             /* EWALD ELECTROSTATICS */
2877
2878             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2879             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2880             ewitab           = _mm256_cvttpd_epi32(ewrt);
2881             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2882             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2883                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2884                                             &ewtabF,&ewtabFn);
2885             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2886             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2887
2888             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2889
2890             fscal            = felec;
2891
2892             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2893
2894             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2895
2896             /* Calculate temporary vectorial force */
2897             tx               = _mm256_mul_pd(fscal,dx33);
2898             ty               = _mm256_mul_pd(fscal,dy33);
2899             tz               = _mm256_mul_pd(fscal,dz33);
2900
2901             /* Update vectorial force */
2902             fix3             = _mm256_add_pd(fix3,tx);
2903             fiy3             = _mm256_add_pd(fiy3,ty);
2904             fiz3             = _mm256_add_pd(fiz3,tz);
2905
2906             fjx3             = _mm256_add_pd(fjx3,tx);
2907             fjy3             = _mm256_add_pd(fjy3,ty);
2908             fjz3             = _mm256_add_pd(fjz3,tz);
2909
2910             }
2911
2912             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2913             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2914             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2915             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2916
2917             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2918                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2919                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2920
2921             /* Inner loop uses 413 flops */
2922         }
2923
2924         /* End of innermost loop */
2925
2926         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2927                                                  f+i_coord_offset,fshift+i_shift_offset);
2928
2929         /* Increment number of inner iterations */
2930         inneriter                  += j_index_end - j_index_start;
2931
2932         /* Outer loop uses 24 flops */
2933     }
2934
2935     /* Increment number of outer iterations */
2936     outeriter        += nri;
2937
2938     /* Update outer/inner flops */
2939
2940     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*413);
2941 }