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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     real *           vdwgridioffsetptr0;
  88     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  89     real *           vdwioffsetptr1;
  90     real *           vdwgridioffsetptr1;
  91     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     real *           vdwgridioffsetptr2;
  94     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  95     real *           vdwioffsetptr3;
  96     real *           vdwgridioffsetptr3;
  97     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  98     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  99     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 100     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 101     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 102     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 103     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 104     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
 105     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 106     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 107     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 108     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 109     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 110     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 111     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 112     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 113     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 114     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 115     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 116     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 117     real             *charge;
 118     int              nvdwtype;
 119     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 120     int              *vdwtype;
 121     real             *vdwparam;
 122     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 123     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 124     __m256d           c6grid_00;
 125     __m256d           c6grid_11;
 126     __m256d           c6grid_12;
 127     __m256d           c6grid_13;
 128     __m256d           c6grid_21;
 129     __m256d           c6grid_22;
 130     __m256d           c6grid_23;
 131     __m256d           c6grid_31;
 132     __m256d           c6grid_32;
 133     __m256d           c6grid_33;
 134     real             *vdwgridparam;
 135     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 136     __m256d           one_half  = _mm256_set1_pd(0.5);
 137     __m256d           minus_one = _mm256_set1_pd(-1.0);
 138     __m128i          ewitab;
 139     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 140     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 141     real             *ewtab;
 142     __m256d          dummy_mask,cutoff_mask;
 143     __m128           tmpmask0,tmpmask1;
 144     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 145     __m256d          one     = _mm256_set1_pd(1.0);
 146     __m256d          two     = _mm256_set1_pd(2.0);
 147     x                = xx[0];
 148     f                = ff[0];
 149
 150     nri              = nlist->nri;
 151     iinr             = nlist->iinr;
 152     jindex           = nlist->jindex;
 153     jjnr             = nlist->jjnr;
 154     shiftidx         = nlist->shift;
 155     gid              = nlist->gid;
 156     shiftvec         = fr->shift_vec[0];
 157     fshift           = fr->fshift[0];
 158     facel            = _mm256_set1_pd(fr->epsfac);
 159     charge           = mdatoms->chargeA;
 160     nvdwtype         = fr->ntype;
 161     vdwparam         = fr->nbfp;
 162     vdwtype          = mdatoms->typeA;
 163     vdwgridparam     = fr->ljpme_c6grid;
 164     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 165     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 166     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 167
 168     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 169     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 170     beta2            = _mm256_mul_pd(beta,beta);
 171     beta3            = _mm256_mul_pd(beta,beta2);
 172
 173     ewtab            = fr->ic->tabq_coul_FDV0;
 174     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 175     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 176
 177     /* Setup water-specific parameters */
 178     inr              = nlist->iinr[0];
 179     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 180     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 181     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 182     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 183     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 184
 185     jq1              = _mm256_set1_pd(charge[inr+1]);
 186     jq2              = _mm256_set1_pd(charge[inr+2]);
 187     jq3              = _mm256_set1_pd(charge[inr+3]);
 188     vdwjidx0A        = 2*vdwtype[inr+0];
 189     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 190     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 191     c6grid_00        = _mm256_set1_pd(vdwgridioffsetptr0[vdwjidx0A]);
 192     qq11             = _mm256_mul_pd(iq1,jq1);
 193     qq12             = _mm256_mul_pd(iq1,jq2);
 194     qq13             = _mm256_mul_pd(iq1,jq3);
 195     qq21             = _mm256_mul_pd(iq2,jq1);
 196     qq22             = _mm256_mul_pd(iq2,jq2);
 197     qq23             = _mm256_mul_pd(iq2,jq3);
 198     qq31             = _mm256_mul_pd(iq3,jq1);
 199     qq32             = _mm256_mul_pd(iq3,jq2);
 200     qq33             = _mm256_mul_pd(iq3,jq3);
 201
 202     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 203     rcutoff_scalar   = fr->rcoulomb;
 204     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 205     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 206
 207     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 208     rvdw             = _mm256_set1_pd(fr->rvdw);
 209
 210     /* Avoid stupid compiler warnings */
 211     jnrA = jnrB = jnrC = jnrD = 0;
 212     j_coord_offsetA = 0;
 213     j_coord_offsetB = 0;
 214     j_coord_offsetC = 0;
 215     j_coord_offsetD = 0;
 216
 217     outeriter        = 0;
 218     inneriter        = 0;
 219
 220     for(iidx=0;iidx<4*DIM;iidx++)
 221     {
 222         scratch[iidx] = 0.0;
 223     }
 224
 225     /* Start outer loop over neighborlists */
 226     for(iidx=0; iidx<nri; iidx++)
 227     {
 228         /* Load shift vector for this list */
 229         i_shift_offset   = DIM*shiftidx[iidx];
 230
 231         /* Load limits for loop over neighbors */
 232         j_index_start    = jindex[iidx];
 233         j_index_end      = jindex[iidx+1];
 234
 235         /* Get outer coordinate index */
 236         inr              = iinr[iidx];
 237         i_coord_offset   = DIM*inr;
 238
 239         /* Load i particle coords and add shift vector */
 240         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 241                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 242
 243         fix0             = _mm256_setzero_pd();
 244         fiy0             = _mm256_setzero_pd();
 245         fiz0             = _mm256_setzero_pd();
 246         fix1             = _mm256_setzero_pd();
 247         fiy1             = _mm256_setzero_pd();
 248         fiz1             = _mm256_setzero_pd();
 249         fix2             = _mm256_setzero_pd();
 250         fiy2             = _mm256_setzero_pd();
 251         fiz2             = _mm256_setzero_pd();
 252         fix3             = _mm256_setzero_pd();
 253         fiy3             = _mm256_setzero_pd();
 254         fiz3             = _mm256_setzero_pd();
 255
 256         /* Reset potential sums */
 257         velecsum         = _mm256_setzero_pd();
 258         vvdwsum          = _mm256_setzero_pd();
 259
 260         /* Start inner kernel loop */
 261         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 262         {
 263
 264             /* Get j neighbor index, and coordinate index */
 265             jnrA             = jjnr[jidx];
 266             jnrB             = jjnr[jidx+1];
 267             jnrC             = jjnr[jidx+2];
 268             jnrD             = jjnr[jidx+3];
 269             j_coord_offsetA  = DIM*jnrA;
 270             j_coord_offsetB  = DIM*jnrB;
 271             j_coord_offsetC  = DIM*jnrC;
 272             j_coord_offsetD  = DIM*jnrD;
 273
 274             /* load j atom coordinates */
 275             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 276                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 277                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 278                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 279
 280             /* Calculate displacement vector */
 281             dx00             = _mm256_sub_pd(ix0,jx0);
 282             dy00             = _mm256_sub_pd(iy0,jy0);
 283             dz00             = _mm256_sub_pd(iz0,jz0);
 284             dx11             = _mm256_sub_pd(ix1,jx1);
 285             dy11             = _mm256_sub_pd(iy1,jy1);
 286             dz11             = _mm256_sub_pd(iz1,jz1);
 287             dx12             = _mm256_sub_pd(ix1,jx2);
 288             dy12             = _mm256_sub_pd(iy1,jy2);
 289             dz12             = _mm256_sub_pd(iz1,jz2);
 290             dx13             = _mm256_sub_pd(ix1,jx3);
 291             dy13             = _mm256_sub_pd(iy1,jy3);
 292             dz13             = _mm256_sub_pd(iz1,jz3);
 293             dx21             = _mm256_sub_pd(ix2,jx1);
 294             dy21             = _mm256_sub_pd(iy2,jy1);
 295             dz21             = _mm256_sub_pd(iz2,jz1);
 296             dx22             = _mm256_sub_pd(ix2,jx2);
 297             dy22             = _mm256_sub_pd(iy2,jy2);
 298             dz22             = _mm256_sub_pd(iz2,jz2);
 299             dx23             = _mm256_sub_pd(ix2,jx3);
 300             dy23             = _mm256_sub_pd(iy2,jy3);
 301             dz23             = _mm256_sub_pd(iz2,jz3);
 302             dx31             = _mm256_sub_pd(ix3,jx1);
 303             dy31             = _mm256_sub_pd(iy3,jy1);
 304             dz31             = _mm256_sub_pd(iz3,jz1);
 305             dx32             = _mm256_sub_pd(ix3,jx2);
 306             dy32             = _mm256_sub_pd(iy3,jy2);
 307             dz32             = _mm256_sub_pd(iz3,jz2);
 308             dx33             = _mm256_sub_pd(ix3,jx3);
 309             dy33             = _mm256_sub_pd(iy3,jy3);
 310             dz33             = _mm256_sub_pd(iz3,jz3);
 311
 312             /* Calculate squared distance and things based on it */
 313             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 314             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 315             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 316             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 317             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 318             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 319             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 320             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 321             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 322             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 323
 324             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 325             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 326             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 327             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 328             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 329             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 330             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 331             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 332             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 333             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 334
 335             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 336             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 337             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 338             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 339             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 340             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 341             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 342             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 343             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 344             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 345
 346             fjx0             = _mm256_setzero_pd();
 347             fjy0             = _mm256_setzero_pd();
 348             fjz0             = _mm256_setzero_pd();
 349             fjx1             = _mm256_setzero_pd();
 350             fjy1             = _mm256_setzero_pd();
 351             fjz1             = _mm256_setzero_pd();
 352             fjx2             = _mm256_setzero_pd();
 353             fjy2             = _mm256_setzero_pd();
 354             fjz2             = _mm256_setzero_pd();
 355             fjx3             = _mm256_setzero_pd();
 356             fjy3             = _mm256_setzero_pd();
 357             fjz3             = _mm256_setzero_pd();
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 364             {
 365
 366             r00              = _mm256_mul_pd(rsq00,rinv00);
 367
 368             /* Analytical LJ-PME */
 369             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 370             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 371             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 372             exponent         = gmx_simd_exp_d(ewcljrsq);
 373             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 374             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 375             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 376             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 377             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 378             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) ,
 379                                           _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));
 380             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 381             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);
 382
 383             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 387             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 388
 389             fscal            = fvdw;
 390
 391             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm256_mul_pd(fscal,dx00);
 395             ty               = _mm256_mul_pd(fscal,dy00);
 396             tz               = _mm256_mul_pd(fscal,dz00);
 397
 398             /* Update vectorial force */
 399             fix0             = _mm256_add_pd(fix0,tx);
 400             fiy0             = _mm256_add_pd(fiy0,ty);
 401             fiz0             = _mm256_add_pd(fiz0,tz);
 402
 403             fjx0             = _mm256_add_pd(fjx0,tx);
 404             fjy0             = _mm256_add_pd(fjy0,ty);
 405             fjz0             = _mm256_add_pd(fjz0,tz);
 406
 407             }
 408
 409             /**************************
 410              * CALCULATE INTERACTIONS *
 411              **************************/
 412
 413             if (gmx_mm256_any_lt(rsq11,rcutoff2))
 414             {
 415
 416             r11              = _mm256_mul_pd(rsq11,rinv11);
 417
 418             /* EWALD ELECTROSTATICS */
 419
 420             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 421             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 422             ewitab           = _mm256_cvttpd_epi32(ewrt);
 423             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 424             ewitab           = _mm_slli_epi32(ewitab,2);
 425             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 426             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 427             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 428             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 429             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 430             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 431             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 432             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_sub_pd(rinv11,sh_ewald),velec));
 433             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 434
 435             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             velec            = _mm256_and_pd(velec,cutoff_mask);
 439             velecsum         = _mm256_add_pd(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm256_mul_pd(fscal,dx11);
 447             ty               = _mm256_mul_pd(fscal,dy11);
 448             tz               = _mm256_mul_pd(fscal,dz11);
 449
 450             /* Update vectorial force */
 451             fix1             = _mm256_add_pd(fix1,tx);
 452             fiy1             = _mm256_add_pd(fiy1,ty);
 453             fiz1             = _mm256_add_pd(fiz1,tz);
 454
 455             fjx1             = _mm256_add_pd(fjx1,tx);
 456             fjy1             = _mm256_add_pd(fjy1,ty);
 457             fjz1             = _mm256_add_pd(fjz1,tz);
 458
 459             }
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             if (gmx_mm256_any_lt(rsq12,rcutoff2))
 466             {
 467
 468             r12              = _mm256_mul_pd(rsq12,rinv12);
 469
 470             /* EWALD ELECTROSTATICS */
 471
 472             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 473             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 474             ewitab           = _mm256_cvttpd_epi32(ewrt);
 475             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 476             ewitab           = _mm_slli_epi32(ewitab,2);
 477             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 478             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 479             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 480             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 481             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 482             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 483             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 484             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_sub_pd(rinv12,sh_ewald),velec));
 485             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 486
 487             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velec            = _mm256_and_pd(velec,cutoff_mask);
 491             velecsum         = _mm256_add_pd(velecsum,velec);
 492
 493             fscal            = felec;
 494
 495             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 496
 497             /* Calculate temporary vectorial force */
 498             tx               = _mm256_mul_pd(fscal,dx12);
 499             ty               = _mm256_mul_pd(fscal,dy12);
 500             tz               = _mm256_mul_pd(fscal,dz12);
 501
 502             /* Update vectorial force */
 503             fix1             = _mm256_add_pd(fix1,tx);
 504             fiy1             = _mm256_add_pd(fiy1,ty);
 505             fiz1             = _mm256_add_pd(fiz1,tz);
 506
 507             fjx2             = _mm256_add_pd(fjx2,tx);
 508             fjy2             = _mm256_add_pd(fjy2,ty);
 509             fjz2             = _mm256_add_pd(fjz2,tz);
 510
 511             }
 512
 513             /**************************
 514              * CALCULATE INTERACTIONS *
 515              **************************/
 516
 517             if (gmx_mm256_any_lt(rsq13,rcutoff2))
 518             {
 519
 520             r13              = _mm256_mul_pd(rsq13,rinv13);
 521
 522             /* EWALD ELECTROSTATICS */
 523
 524             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 525             ewrt             = _mm256_mul_pd(r13,ewtabscale);
 526             ewitab           = _mm256_cvttpd_epi32(ewrt);
 527             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 528             ewitab           = _mm_slli_epi32(ewitab,2);
 529             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 530             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 531             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 532             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 533             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 534             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 535             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 536             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_sub_pd(rinv13,sh_ewald),velec));
 537             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
 538
 539             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velec            = _mm256_and_pd(velec,cutoff_mask);
 543             velecsum         = _mm256_add_pd(velecsum,velec);
 544
 545             fscal            = felec;
 546
 547             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 548
 549             /* Calculate temporary vectorial force */
 550             tx               = _mm256_mul_pd(fscal,dx13);
 551             ty               = _mm256_mul_pd(fscal,dy13);
 552             tz               = _mm256_mul_pd(fscal,dz13);
 553
 554             /* Update vectorial force */
 555             fix1             = _mm256_add_pd(fix1,tx);
 556             fiy1             = _mm256_add_pd(fiy1,ty);
 557             fiz1             = _mm256_add_pd(fiz1,tz);
 558
 559             fjx3             = _mm256_add_pd(fjx3,tx);
 560             fjy3             = _mm256_add_pd(fjy3,ty);
 561             fjz3             = _mm256_add_pd(fjz3,tz);
 562
 563             }
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             if (gmx_mm256_any_lt(rsq21,rcutoff2))
 570             {
 571
 572             r21              = _mm256_mul_pd(rsq21,rinv21);
 573
 574             /* EWALD ELECTROSTATICS */
 575
 576             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 577             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 578             ewitab           = _mm256_cvttpd_epi32(ewrt);
 579             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 580             ewitab           = _mm_slli_epi32(ewitab,2);
 581             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 582             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 583             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 584             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 585             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 586             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 587             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 588             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_sub_pd(rinv21,sh_ewald),velec));
 589             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 590
 591             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm256_and_pd(velec,cutoff_mask);
 595             velecsum         = _mm256_add_pd(velecsum,velec);
 596
 597             fscal            = felec;
 598
 599             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm256_mul_pd(fscal,dx21);
 603             ty               = _mm256_mul_pd(fscal,dy21);
 604             tz               = _mm256_mul_pd(fscal,dz21);
 605
 606             /* Update vectorial force */
 607             fix2             = _mm256_add_pd(fix2,tx);
 608             fiy2             = _mm256_add_pd(fiy2,ty);
 609             fiz2             = _mm256_add_pd(fiz2,tz);
 610
 611             fjx1             = _mm256_add_pd(fjx1,tx);
 612             fjy1             = _mm256_add_pd(fjy1,ty);
 613             fjz1             = _mm256_add_pd(fjz1,tz);
 614
 615             }
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             if (gmx_mm256_any_lt(rsq22,rcutoff2))
 622             {
 623
 624             r22              = _mm256_mul_pd(rsq22,rinv22);
 625
 626             /* EWALD ELECTROSTATICS */
 627
 628             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 629             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 630             ewitab           = _mm256_cvttpd_epi32(ewrt);
 631             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 632             ewitab           = _mm_slli_epi32(ewitab,2);
 633             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 634             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 635             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 636             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 637             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 638             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 639             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 640             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_sub_pd(rinv22,sh_ewald),velec));
 641             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 642
 643             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
 644
 645             /* Update potential sum for this i atom from the interaction with this j atom. */
 646             velec            = _mm256_and_pd(velec,cutoff_mask);
 647             velecsum         = _mm256_add_pd(velecsum,velec);
 648
 649             fscal            = felec;
 650
 651             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 652
 653             /* Calculate temporary vectorial force */
 654             tx               = _mm256_mul_pd(fscal,dx22);
 655             ty               = _mm256_mul_pd(fscal,dy22);
 656             tz               = _mm256_mul_pd(fscal,dz22);
 657
 658             /* Update vectorial force */
 659             fix2             = _mm256_add_pd(fix2,tx);
 660             fiy2             = _mm256_add_pd(fiy2,ty);
 661             fiz2             = _mm256_add_pd(fiz2,tz);
 662
 663             fjx2             = _mm256_add_pd(fjx2,tx);
 664             fjy2             = _mm256_add_pd(fjy2,ty);
 665             fjz2             = _mm256_add_pd(fjz2,tz);
 666
 667             }
 668
 669             /**************************
 670              * CALCULATE INTERACTIONS *
 671              **************************/
 672
 673             if (gmx_mm256_any_lt(rsq23,rcutoff2))
 674             {
 675
 676             r23              = _mm256_mul_pd(rsq23,rinv23);
 677
 678             /* EWALD ELECTROSTATICS */
 679
 680             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 681             ewrt             = _mm256_mul_pd(r23,ewtabscale);
 682             ewitab           = _mm256_cvttpd_epi32(ewrt);
 683             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 684             ewitab           = _mm_slli_epi32(ewitab,2);
 685             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 686             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 687             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 688             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 689             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 690             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 691             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 692             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_sub_pd(rinv23,sh_ewald),velec));
 693             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
 694
 695             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
 696
 697             /* Update potential sum for this i atom from the interaction with this j atom. */
 698             velec            = _mm256_and_pd(velec,cutoff_mask);
 699             velecsum         = _mm256_add_pd(velecsum,velec);
 700
 701             fscal            = felec;
 702
 703             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm256_mul_pd(fscal,dx23);
 707             ty               = _mm256_mul_pd(fscal,dy23);
 708             tz               = _mm256_mul_pd(fscal,dz23);
 709
 710             /* Update vectorial force */
 711             fix2             = _mm256_add_pd(fix2,tx);
 712             fiy2             = _mm256_add_pd(fiy2,ty);
 713             fiz2             = _mm256_add_pd(fiz2,tz);
 714
 715             fjx3             = _mm256_add_pd(fjx3,tx);
 716             fjy3             = _mm256_add_pd(fjy3,ty);
 717             fjz3             = _mm256_add_pd(fjz3,tz);
 718
 719             }
 720
 721             /**************************
 722              * CALCULATE INTERACTIONS *
 723              **************************/
 724
 725             if (gmx_mm256_any_lt(rsq31,rcutoff2))
 726             {
 727
 728             r31              = _mm256_mul_pd(rsq31,rinv31);
 729
 730             /* EWALD ELECTROSTATICS */
 731
 732             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 733             ewrt             = _mm256_mul_pd(r31,ewtabscale);
 734             ewitab           = _mm256_cvttpd_epi32(ewrt);
 735             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 736             ewitab           = _mm_slli_epi32(ewitab,2);
 737             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 738             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 739             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 740             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 741             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 742             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 743             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 744             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_sub_pd(rinv31,sh_ewald),velec));
 745             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
 746
 747             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
 748
 749             /* Update potential sum for this i atom from the interaction with this j atom. */
 750             velec            = _mm256_and_pd(velec,cutoff_mask);
 751             velecsum         = _mm256_add_pd(velecsum,velec);
 752
 753             fscal            = felec;
 754
 755             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 756
 757             /* Calculate temporary vectorial force */
 758             tx               = _mm256_mul_pd(fscal,dx31);
 759             ty               = _mm256_mul_pd(fscal,dy31);
 760             tz               = _mm256_mul_pd(fscal,dz31);
 761
 762             /* Update vectorial force */
 763             fix3             = _mm256_add_pd(fix3,tx);
 764             fiy3             = _mm256_add_pd(fiy3,ty);
 765             fiz3             = _mm256_add_pd(fiz3,tz);
 766
 767             fjx1             = _mm256_add_pd(fjx1,tx);
 768             fjy1             = _mm256_add_pd(fjy1,ty);
 769             fjz1             = _mm256_add_pd(fjz1,tz);
 770
 771             }
 772
 773             /**************************
 774              * CALCULATE INTERACTIONS *
 775              **************************/
 776
 777             if (gmx_mm256_any_lt(rsq32,rcutoff2))
 778             {
 779
 780             r32              = _mm256_mul_pd(rsq32,rinv32);
 781
 782             /* EWALD ELECTROSTATICS */
 783
 784             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 785             ewrt             = _mm256_mul_pd(r32,ewtabscale);
 786             ewitab           = _mm256_cvttpd_epi32(ewrt);
 787             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 788             ewitab           = _mm_slli_epi32(ewitab,2);
 789             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 790             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 791             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 792             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 793             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 794             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 795             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 796             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_sub_pd(rinv32,sh_ewald),velec));
 797             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
 798
 799             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
 800
 801             /* Update potential sum for this i atom from the interaction with this j atom. */
 802             velec            = _mm256_and_pd(velec,cutoff_mask);
 803             velecsum         = _mm256_add_pd(velecsum,velec);
 804
 805             fscal            = felec;
 806
 807             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 808
 809             /* Calculate temporary vectorial force */
 810             tx               = _mm256_mul_pd(fscal,dx32);
 811             ty               = _mm256_mul_pd(fscal,dy32);
 812             tz               = _mm256_mul_pd(fscal,dz32);
 813
 814             /* Update vectorial force */
 815             fix3             = _mm256_add_pd(fix3,tx);
 816             fiy3             = _mm256_add_pd(fiy3,ty);
 817             fiz3             = _mm256_add_pd(fiz3,tz);
 818
 819             fjx2             = _mm256_add_pd(fjx2,tx);
 820             fjy2             = _mm256_add_pd(fjy2,ty);
 821             fjz2             = _mm256_add_pd(fjz2,tz);
 822
 823             }
 824
 825             /**************************
 826              * CALCULATE INTERACTIONS *
 827              **************************/
 828
 829             if (gmx_mm256_any_lt(rsq33,rcutoff2))
 830             {
 831
 832             r33              = _mm256_mul_pd(rsq33,rinv33);
 833
 834             /* EWALD ELECTROSTATICS */
 835
 836             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 837             ewrt             = _mm256_mul_pd(r33,ewtabscale);
 838             ewitab           = _mm256_cvttpd_epi32(ewrt);
 839             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 840             ewitab           = _mm_slli_epi32(ewitab,2);
 841             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 842             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 843             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 844             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 845             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 846             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 847             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 848             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_sub_pd(rinv33,sh_ewald),velec));
 849             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
 850
 851             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
 852
 853             /* Update potential sum for this i atom from the interaction with this j atom. */
 854             velec            = _mm256_and_pd(velec,cutoff_mask);
 855             velecsum         = _mm256_add_pd(velecsum,velec);
 856
 857             fscal            = felec;
 858
 859             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 860
 861             /* Calculate temporary vectorial force */
 862             tx               = _mm256_mul_pd(fscal,dx33);
 863             ty               = _mm256_mul_pd(fscal,dy33);
 864             tz               = _mm256_mul_pd(fscal,dz33);
 865
 866             /* Update vectorial force */
 867             fix3             = _mm256_add_pd(fix3,tx);
 868             fiy3             = _mm256_add_pd(fiy3,ty);
 869             fiz3             = _mm256_add_pd(fiz3,tz);
 870
 871             fjx3             = _mm256_add_pd(fjx3,tx);
 872             fjy3             = _mm256_add_pd(fjy3,ty);
 873             fjz3             = _mm256_add_pd(fjz3,tz);
 874
 875             }
 876
 877             fjptrA             = f+j_coord_offsetA;
 878             fjptrB             = f+j_coord_offsetB;
 879             fjptrC             = f+j_coord_offsetC;
 880             fjptrD             = f+j_coord_offsetD;
 881
 882             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 883                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 884                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 885
 886             /* Inner loop uses 479 flops */
 887         }
 888
 889         if(jidx<j_index_end)
 890         {
 891
 892             /* Get j neighbor index, and coordinate index */
 893             jnrlistA         = jjnr[jidx];
 894             jnrlistB         = jjnr[jidx+1];
 895             jnrlistC         = jjnr[jidx+2];
 896             jnrlistD         = jjnr[jidx+3];
 897             /* Sign of each element will be negative for non-real atoms.
 898              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 899              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 900              */
 901             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 902
 903             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 904             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 905             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 906
 907             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 908             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 909             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 910             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 911             j_coord_offsetA  = DIM*jnrA;
 912             j_coord_offsetB  = DIM*jnrB;
 913             j_coord_offsetC  = DIM*jnrC;
 914             j_coord_offsetD  = DIM*jnrD;
 915
 916             /* load j atom coordinates */
 917             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 918                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 919                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 920                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 921
 922             /* Calculate displacement vector */
 923             dx00             = _mm256_sub_pd(ix0,jx0);
 924             dy00             = _mm256_sub_pd(iy0,jy0);
 925             dz00             = _mm256_sub_pd(iz0,jz0);
 926             dx11             = _mm256_sub_pd(ix1,jx1);
 927             dy11             = _mm256_sub_pd(iy1,jy1);
 928             dz11             = _mm256_sub_pd(iz1,jz1);
 929             dx12             = _mm256_sub_pd(ix1,jx2);
 930             dy12             = _mm256_sub_pd(iy1,jy2);
 931             dz12             = _mm256_sub_pd(iz1,jz2);
 932             dx13             = _mm256_sub_pd(ix1,jx3);
 933             dy13             = _mm256_sub_pd(iy1,jy3);
 934             dz13             = _mm256_sub_pd(iz1,jz3);
 935             dx21             = _mm256_sub_pd(ix2,jx1);
 936             dy21             = _mm256_sub_pd(iy2,jy1);
 937             dz21             = _mm256_sub_pd(iz2,jz1);
 938             dx22             = _mm256_sub_pd(ix2,jx2);
 939             dy22             = _mm256_sub_pd(iy2,jy2);
 940             dz22             = _mm256_sub_pd(iz2,jz2);
 941             dx23             = _mm256_sub_pd(ix2,jx3);
 942             dy23             = _mm256_sub_pd(iy2,jy3);
 943             dz23             = _mm256_sub_pd(iz2,jz3);
 944             dx31             = _mm256_sub_pd(ix3,jx1);
 945             dy31             = _mm256_sub_pd(iy3,jy1);
 946             dz31             = _mm256_sub_pd(iz3,jz1);
 947             dx32             = _mm256_sub_pd(ix3,jx2);
 948             dy32             = _mm256_sub_pd(iy3,jy2);
 949             dz32             = _mm256_sub_pd(iz3,jz2);
 950             dx33             = _mm256_sub_pd(ix3,jx3);
 951             dy33             = _mm256_sub_pd(iy3,jy3);
 952             dz33             = _mm256_sub_pd(iz3,jz3);
 953
 954             /* Calculate squared distance and things based on it */
 955             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 956             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 957             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 958             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 959             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 960             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 961             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 962             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 963             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 964             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 965
 966             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 967             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 968             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 969             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 970             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 971             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 972             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 973             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 974             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 975             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 976
 977             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 978             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 979             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 980             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 981             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 982             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 983             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 984             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 985             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 986             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 987
 988             fjx0             = _mm256_setzero_pd();
 989             fjy0             = _mm256_setzero_pd();
 990             fjz0             = _mm256_setzero_pd();
 991             fjx1             = _mm256_setzero_pd();
 992             fjy1             = _mm256_setzero_pd();
 993             fjz1             = _mm256_setzero_pd();
 994             fjx2             = _mm256_setzero_pd();
 995             fjy2             = _mm256_setzero_pd();
 996             fjz2             = _mm256_setzero_pd();
 997             fjx3             = _mm256_setzero_pd();
 998             fjy3             = _mm256_setzero_pd();
 999             fjz3             = _mm256_setzero_pd();
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1006             {
1007
1008             r00              = _mm256_mul_pd(rsq00,rinv00);
1009             r00              = _mm256_andnot_pd(dummy_mask,r00);
1010
1011             /* Analytical LJ-PME */
1012             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1013             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1014             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1015             exponent         = gmx_simd_exp_d(ewcljrsq);
1016             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1017             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1018             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
1019             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
1020             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
1021             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) ,
1022                                           _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));
1023             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
1024             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);
1025
1026             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1027
1028             /* Update potential sum for this i atom from the interaction with this j atom. */
1029             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
1030             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
1031             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
1032
1033             fscal            = fvdw;
1034
1035             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1036
1037             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm256_mul_pd(fscal,dx00);
1041             ty               = _mm256_mul_pd(fscal,dy00);
1042             tz               = _mm256_mul_pd(fscal,dz00);
1043
1044             /* Update vectorial force */
1045             fix0             = _mm256_add_pd(fix0,tx);
1046             fiy0             = _mm256_add_pd(fiy0,ty);
1047             fiz0             = _mm256_add_pd(fiz0,tz);
1048
1049             fjx0             = _mm256_add_pd(fjx0,tx);
1050             fjy0             = _mm256_add_pd(fjy0,ty);
1051             fjz0             = _mm256_add_pd(fjz0,tz);
1052
1053             }
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1060             {
1061
1062             r11              = _mm256_mul_pd(rsq11,rinv11);
1063             r11              = _mm256_andnot_pd(dummy_mask,r11);
1064
1065             /* EWALD ELECTROSTATICS */
1066
1067             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1068             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1069             ewitab           = _mm256_cvttpd_epi32(ewrt);
1070             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1071             ewitab           = _mm_slli_epi32(ewitab,2);
1072             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1073             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1074             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1075             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1076             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1077             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1078             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1079             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(_mm256_sub_pd(rinv11,sh_ewald),velec));
1080             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1081
1082             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1083
1084             /* Update potential sum for this i atom from the interaction with this j atom. */
1085             velec            = _mm256_and_pd(velec,cutoff_mask);
1086             velec            = _mm256_andnot_pd(dummy_mask,velec);
1087             velecsum         = _mm256_add_pd(velecsum,velec);
1088
1089             fscal            = felec;
1090
1091             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1092
1093             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm256_mul_pd(fscal,dx11);
1097             ty               = _mm256_mul_pd(fscal,dy11);
1098             tz               = _mm256_mul_pd(fscal,dz11);
1099
1100             /* Update vectorial force */
1101             fix1             = _mm256_add_pd(fix1,tx);
1102             fiy1             = _mm256_add_pd(fiy1,ty);
1103             fiz1             = _mm256_add_pd(fiz1,tz);
1104
1105             fjx1             = _mm256_add_pd(fjx1,tx);
1106             fjy1             = _mm256_add_pd(fjy1,ty);
1107             fjz1             = _mm256_add_pd(fjz1,tz);
1108
1109             }
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1116             {
1117
1118             r12              = _mm256_mul_pd(rsq12,rinv12);
1119             r12              = _mm256_andnot_pd(dummy_mask,r12);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1124             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1125             ewitab           = _mm256_cvttpd_epi32(ewrt);
1126             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1127             ewitab           = _mm_slli_epi32(ewitab,2);
1128             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1129             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1130             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1131             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1132             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1133             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1134             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1135             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(_mm256_sub_pd(rinv12,sh_ewald),velec));
1136             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1137
1138             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
1139
1140             /* Update potential sum for this i atom from the interaction with this j atom. */
1141             velec            = _mm256_and_pd(velec,cutoff_mask);
1142             velec            = _mm256_andnot_pd(dummy_mask,velec);
1143             velecsum         = _mm256_add_pd(velecsum,velec);
1144
1145             fscal            = felec;
1146
1147             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1148
1149             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1150
1151             /* Calculate temporary vectorial force */
1152             tx               = _mm256_mul_pd(fscal,dx12);
1153             ty               = _mm256_mul_pd(fscal,dy12);
1154             tz               = _mm256_mul_pd(fscal,dz12);
1155
1156             /* Update vectorial force */
1157             fix1             = _mm256_add_pd(fix1,tx);
1158             fiy1             = _mm256_add_pd(fiy1,ty);
1159             fiz1             = _mm256_add_pd(fiz1,tz);
1160
1161             fjx2             = _mm256_add_pd(fjx2,tx);
1162             fjy2             = _mm256_add_pd(fjy2,ty);
1163             fjz2             = _mm256_add_pd(fjz2,tz);
1164
1165             }
1166
1167             /**************************
1168              * CALCULATE INTERACTIONS *
1169              **************************/
1170
1171             if (gmx_mm256_any_lt(rsq13,rcutoff2))
1172             {
1173
1174             r13              = _mm256_mul_pd(rsq13,rinv13);
1175             r13              = _mm256_andnot_pd(dummy_mask,r13);
1176
1177             /* EWALD ELECTROSTATICS */
1178
1179             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1180             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1181             ewitab           = _mm256_cvttpd_epi32(ewrt);
1182             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1183             ewitab           = _mm_slli_epi32(ewitab,2);
1184             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1185             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1186             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1187             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1188             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1189             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1190             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1191             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(_mm256_sub_pd(rinv13,sh_ewald),velec));
1192             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1193
1194             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
1195
1196             /* Update potential sum for this i atom from the interaction with this j atom. */
1197             velec            = _mm256_and_pd(velec,cutoff_mask);
1198             velec            = _mm256_andnot_pd(dummy_mask,velec);
1199             velecsum         = _mm256_add_pd(velecsum,velec);
1200
1201             fscal            = felec;
1202
1203             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1204
1205             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1206
1207             /* Calculate temporary vectorial force */
1208             tx               = _mm256_mul_pd(fscal,dx13);
1209             ty               = _mm256_mul_pd(fscal,dy13);
1210             tz               = _mm256_mul_pd(fscal,dz13);
1211
1212             /* Update vectorial force */
1213             fix1             = _mm256_add_pd(fix1,tx);
1214             fiy1             = _mm256_add_pd(fiy1,ty);
1215             fiz1             = _mm256_add_pd(fiz1,tz);
1216
1217             fjx3             = _mm256_add_pd(fjx3,tx);
1218             fjy3             = _mm256_add_pd(fjy3,ty);
1219             fjz3             = _mm256_add_pd(fjz3,tz);
1220
1221             }
1222
1223             /**************************
1224              * CALCULATE INTERACTIONS *
1225              **************************/
1226
1227             if (gmx_mm256_any_lt(rsq21,rcutoff2))
1228             {
1229
1230             r21              = _mm256_mul_pd(rsq21,rinv21);
1231             r21              = _mm256_andnot_pd(dummy_mask,r21);
1232
1233             /* EWALD ELECTROSTATICS */
1234
1235             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1236             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1237             ewitab           = _mm256_cvttpd_epi32(ewrt);
1238             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1239             ewitab           = _mm_slli_epi32(ewitab,2);
1240             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1241             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1242             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1243             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1244             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1245             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1246             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1247             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(_mm256_sub_pd(rinv21,sh_ewald),velec));
1248             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1249
1250             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
1251
1252             /* Update potential sum for this i atom from the interaction with this j atom. */
1253             velec            = _mm256_and_pd(velec,cutoff_mask);
1254             velec            = _mm256_andnot_pd(dummy_mask,velec);
1255             velecsum         = _mm256_add_pd(velecsum,velec);
1256
1257             fscal            = felec;
1258
1259             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1260
1261             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1262
1263             /* Calculate temporary vectorial force */
1264             tx               = _mm256_mul_pd(fscal,dx21);
1265             ty               = _mm256_mul_pd(fscal,dy21);
1266             tz               = _mm256_mul_pd(fscal,dz21);
1267
1268             /* Update vectorial force */
1269             fix2             = _mm256_add_pd(fix2,tx);
1270             fiy2             = _mm256_add_pd(fiy2,ty);
1271             fiz2             = _mm256_add_pd(fiz2,tz);
1272
1273             fjx1             = _mm256_add_pd(fjx1,tx);
1274             fjy1             = _mm256_add_pd(fjy1,ty);
1275             fjz1             = _mm256_add_pd(fjz1,tz);
1276
1277             }
1278
1279             /**************************
1280              * CALCULATE INTERACTIONS *
1281              **************************/
1282
1283             if (gmx_mm256_any_lt(rsq22,rcutoff2))
1284             {
1285
1286             r22              = _mm256_mul_pd(rsq22,rinv22);
1287             r22              = _mm256_andnot_pd(dummy_mask,r22);
1288
1289             /* EWALD ELECTROSTATICS */
1290
1291             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1292             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1293             ewitab           = _mm256_cvttpd_epi32(ewrt);
1294             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1295             ewitab           = _mm_slli_epi32(ewitab,2);
1296             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1297             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1298             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1299             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1300             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1301             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1302             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1303             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(_mm256_sub_pd(rinv22,sh_ewald),velec));
1304             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1305
1306             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
1307
1308             /* Update potential sum for this i atom from the interaction with this j atom. */
1309             velec            = _mm256_and_pd(velec,cutoff_mask);
1310             velec            = _mm256_andnot_pd(dummy_mask,velec);
1311             velecsum         = _mm256_add_pd(velecsum,velec);
1312
1313             fscal            = felec;
1314
1315             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1316
1317             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1318
1319             /* Calculate temporary vectorial force */
1320             tx               = _mm256_mul_pd(fscal,dx22);
1321             ty               = _mm256_mul_pd(fscal,dy22);
1322             tz               = _mm256_mul_pd(fscal,dz22);
1323
1324             /* Update vectorial force */
1325             fix2             = _mm256_add_pd(fix2,tx);
1326             fiy2             = _mm256_add_pd(fiy2,ty);
1327             fiz2             = _mm256_add_pd(fiz2,tz);
1328
1329             fjx2             = _mm256_add_pd(fjx2,tx);
1330             fjy2             = _mm256_add_pd(fjy2,ty);
1331             fjz2             = _mm256_add_pd(fjz2,tz);
1332
1333             }
1334
1335             /**************************
1336              * CALCULATE INTERACTIONS *
1337              **************************/
1338
1339             if (gmx_mm256_any_lt(rsq23,rcutoff2))
1340             {
1341
1342             r23              = _mm256_mul_pd(rsq23,rinv23);
1343             r23              = _mm256_andnot_pd(dummy_mask,r23);
1344
1345             /* EWALD ELECTROSTATICS */
1346
1347             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1348             ewrt             = _mm256_mul_pd(r23,ewtabscale);
1349             ewitab           = _mm256_cvttpd_epi32(ewrt);
1350             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1351             ewitab           = _mm_slli_epi32(ewitab,2);
1352             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1353             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1354             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1355             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1356             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1357             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1358             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1359             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(_mm256_sub_pd(rinv23,sh_ewald),velec));
1360             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
1361
1362             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
1363
1364             /* Update potential sum for this i atom from the interaction with this j atom. */
1365             velec            = _mm256_and_pd(velec,cutoff_mask);
1366             velec            = _mm256_andnot_pd(dummy_mask,velec);
1367             velecsum         = _mm256_add_pd(velecsum,velec);
1368
1369             fscal            = felec;
1370
1371             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1372
1373             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1374
1375             /* Calculate temporary vectorial force */
1376             tx               = _mm256_mul_pd(fscal,dx23);
1377             ty               = _mm256_mul_pd(fscal,dy23);
1378             tz               = _mm256_mul_pd(fscal,dz23);
1379
1380             /* Update vectorial force */
1381             fix2             = _mm256_add_pd(fix2,tx);
1382             fiy2             = _mm256_add_pd(fiy2,ty);
1383             fiz2             = _mm256_add_pd(fiz2,tz);
1384
1385             fjx3             = _mm256_add_pd(fjx3,tx);
1386             fjy3             = _mm256_add_pd(fjy3,ty);
1387             fjz3             = _mm256_add_pd(fjz3,tz);
1388
1389             }
1390
1391             /**************************
1392              * CALCULATE INTERACTIONS *
1393              **************************/
1394
1395             if (gmx_mm256_any_lt(rsq31,rcutoff2))
1396             {
1397
1398             r31              = _mm256_mul_pd(rsq31,rinv31);
1399             r31              = _mm256_andnot_pd(dummy_mask,r31);
1400
1401             /* EWALD ELECTROSTATICS */
1402
1403             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1404             ewrt             = _mm256_mul_pd(r31,ewtabscale);
1405             ewitab           = _mm256_cvttpd_epi32(ewrt);
1406             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1407             ewitab           = _mm_slli_epi32(ewitab,2);
1408             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1409             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1410             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1411             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1412             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1413             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1414             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1415             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(_mm256_sub_pd(rinv31,sh_ewald),velec));
1416             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
1417
1418             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
1419
1420             /* Update potential sum for this i atom from the interaction with this j atom. */
1421             velec            = _mm256_and_pd(velec,cutoff_mask);
1422             velec            = _mm256_andnot_pd(dummy_mask,velec);
1423             velecsum         = _mm256_add_pd(velecsum,velec);
1424
1425             fscal            = felec;
1426
1427             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1428
1429             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1430
1431             /* Calculate temporary vectorial force */
1432             tx               = _mm256_mul_pd(fscal,dx31);
1433             ty               = _mm256_mul_pd(fscal,dy31);
1434             tz               = _mm256_mul_pd(fscal,dz31);
1435
1436             /* Update vectorial force */
1437             fix3             = _mm256_add_pd(fix3,tx);
1438             fiy3             = _mm256_add_pd(fiy3,ty);
1439             fiz3             = _mm256_add_pd(fiz3,tz);
1440
1441             fjx1             = _mm256_add_pd(fjx1,tx);
1442             fjy1             = _mm256_add_pd(fjy1,ty);
1443             fjz1             = _mm256_add_pd(fjz1,tz);
1444
1445             }
1446
1447             /**************************
1448              * CALCULATE INTERACTIONS *
1449              **************************/
1450
1451             if (gmx_mm256_any_lt(rsq32,rcutoff2))
1452             {
1453
1454             r32              = _mm256_mul_pd(rsq32,rinv32);
1455             r32              = _mm256_andnot_pd(dummy_mask,r32);
1456
1457             /* EWALD ELECTROSTATICS */
1458
1459             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1460             ewrt             = _mm256_mul_pd(r32,ewtabscale);
1461             ewitab           = _mm256_cvttpd_epi32(ewrt);
1462             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1463             ewitab           = _mm_slli_epi32(ewitab,2);
1464             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1465             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1466             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1467             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1468             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1469             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1470             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1471             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(_mm256_sub_pd(rinv32,sh_ewald),velec));
1472             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
1473
1474             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
1475
1476             /* Update potential sum for this i atom from the interaction with this j atom. */
1477             velec            = _mm256_and_pd(velec,cutoff_mask);
1478             velec            = _mm256_andnot_pd(dummy_mask,velec);
1479             velecsum         = _mm256_add_pd(velecsum,velec);
1480
1481             fscal            = felec;
1482
1483             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1484
1485             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1486
1487             /* Calculate temporary vectorial force */
1488             tx               = _mm256_mul_pd(fscal,dx32);
1489             ty               = _mm256_mul_pd(fscal,dy32);
1490             tz               = _mm256_mul_pd(fscal,dz32);
1491
1492             /* Update vectorial force */
1493             fix3             = _mm256_add_pd(fix3,tx);
1494             fiy3             = _mm256_add_pd(fiy3,ty);
1495             fiz3             = _mm256_add_pd(fiz3,tz);
1496
1497             fjx2             = _mm256_add_pd(fjx2,tx);
1498             fjy2             = _mm256_add_pd(fjy2,ty);
1499             fjz2             = _mm256_add_pd(fjz2,tz);
1500
1501             }
1502
1503             /**************************
1504              * CALCULATE INTERACTIONS *
1505              **************************/
1506
1507             if (gmx_mm256_any_lt(rsq33,rcutoff2))
1508             {
1509
1510             r33              = _mm256_mul_pd(rsq33,rinv33);
1511             r33              = _mm256_andnot_pd(dummy_mask,r33);
1512
1513             /* EWALD ELECTROSTATICS */
1514
1515             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1516             ewrt             = _mm256_mul_pd(r33,ewtabscale);
1517             ewitab           = _mm256_cvttpd_epi32(ewrt);
1518             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1519             ewitab           = _mm_slli_epi32(ewitab,2);
1520             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1521             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1522             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1523             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1524             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1525             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1526             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1527             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(_mm256_sub_pd(rinv33,sh_ewald),velec));
1528             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
1529
1530             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
1531
1532             /* Update potential sum for this i atom from the interaction with this j atom. */
1533             velec            = _mm256_and_pd(velec,cutoff_mask);
1534             velec            = _mm256_andnot_pd(dummy_mask,velec);
1535             velecsum         = _mm256_add_pd(velecsum,velec);
1536
1537             fscal            = felec;
1538
1539             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1540
1541             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1542
1543             /* Calculate temporary vectorial force */
1544             tx               = _mm256_mul_pd(fscal,dx33);
1545             ty               = _mm256_mul_pd(fscal,dy33);
1546             tz               = _mm256_mul_pd(fscal,dz33);
1547
1548             /* Update vectorial force */
1549             fix3             = _mm256_add_pd(fix3,tx);
1550             fiy3             = _mm256_add_pd(fiy3,ty);
1551             fiz3             = _mm256_add_pd(fiz3,tz);
1552
1553             fjx3             = _mm256_add_pd(fjx3,tx);
1554             fjy3             = _mm256_add_pd(fjy3,ty);
1555             fjz3             = _mm256_add_pd(fjz3,tz);
1556
1557             }
1558
1559             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1560             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1561             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1562             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1563
1564             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1565                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1566                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1567
1568             /* Inner loop uses 489 flops */
1569         }
1570
1571         /* End of innermost loop */
1572
1573         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1574                                                  f+i_coord_offset,fshift+i_shift_offset);
1575
1576         ggid                        = gid[iidx];
1577         /* Update potential energies */
1578         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1579         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1580
1581         /* Increment number of inner iterations */
1582         inneriter                  += j_index_end - j_index_start;
1583
1584         /* Outer loop uses 26 flops */
1585     }
1586
1587     /* Increment number of outer iterations */
1588     outeriter        += nri;
1589
1590     /* Update outer/inner flops */
1591
1592     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*489);
1593 }
1594 /*
1595  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_F_avx_256_double
1596  * Electrostatics interaction: Ewald
1597  * VdW interaction:            LJEwald
1598  * Geometry:                   Water4-Water4
1599  * Calculate force/pot:        Force
1600  */
1601 void
1602 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_F_avx_256_double
1603                     (t_nblist                    * gmx_restrict       nlist,
1604                      rvec                        * gmx_restrict          xx,
1605                      rvec                        * gmx_restrict          ff,
1606                      t_forcerec                  * gmx_restrict          fr,
1607                      t_mdatoms                   * gmx_restrict     mdatoms,
1608                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1609                      t_nrnb                      * gmx_restrict        nrnb)
1610 {
1611     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1612      * just 0 for non-waters.
1613      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1614      * jnr indices corresponding to data put in the four positions in the SIMD register.
1615      */
1616     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1617     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1618     int              jnrA,jnrB,jnrC,jnrD;
1619     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1620     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1621     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1622     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1623     real             rcutoff_scalar;
1624     real             *shiftvec,*fshift,*x,*f;
1625     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1626     real             scratch[4*DIM];
1627     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1628     real *           vdwioffsetptr0;
1629     real *           vdwgridioffsetptr0;
1630     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1631     real *           vdwioffsetptr1;
1632     real *           vdwgridioffsetptr1;
1633     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1634     real *           vdwioffsetptr2;
1635     real *           vdwgridioffsetptr2;
1636     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1637     real *           vdwioffsetptr3;
1638     real *           vdwgridioffsetptr3;
1639     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1640     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1641     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1642     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1643     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1644     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1645     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1646     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1647     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1648     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1649     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1650     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1651     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1652     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1653     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1654     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1655     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1656     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1657     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1658     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1659     real             *charge;
1660     int              nvdwtype;
1661     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1662     int              *vdwtype;
1663     real             *vdwparam;
1664     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1665     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1666     __m256d           c6grid_00;
1667     __m256d           c6grid_11;
1668     __m256d           c6grid_12;
1669     __m256d           c6grid_13;
1670     __m256d           c6grid_21;
1671     __m256d           c6grid_22;
1672     __m256d           c6grid_23;
1673     __m256d           c6grid_31;
1674     __m256d           c6grid_32;
1675     __m256d           c6grid_33;
1676     real             *vdwgridparam;
1677     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1678     __m256d           one_half  = _mm256_set1_pd(0.5);
1679     __m256d           minus_one = _mm256_set1_pd(-1.0);
1680     __m128i          ewitab;
1681     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1682     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1683     real             *ewtab;
1684     __m256d          dummy_mask,cutoff_mask;
1685     __m128           tmpmask0,tmpmask1;
1686     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1687     __m256d          one     = _mm256_set1_pd(1.0);
1688     __m256d          two     = _mm256_set1_pd(2.0);
1689     x                = xx[0];
1690     f                = ff[0];
1691
1692     nri              = nlist->nri;
1693     iinr             = nlist->iinr;
1694     jindex           = nlist->jindex;
1695     jjnr             = nlist->jjnr;
1696     shiftidx         = nlist->shift;
1697     gid              = nlist->gid;
1698     shiftvec         = fr->shift_vec[0];
1699     fshift           = fr->fshift[0];
1700     facel            = _mm256_set1_pd(fr->epsfac);
1701     charge           = mdatoms->chargeA;
1702     nvdwtype         = fr->ntype;
1703     vdwparam         = fr->nbfp;
1704     vdwtype          = mdatoms->typeA;
1705     vdwgridparam     = fr->ljpme_c6grid;
1706     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
1707     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
1708     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
1709
1710     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1711     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
1712     beta2            = _mm256_mul_pd(beta,beta);
1713     beta3            = _mm256_mul_pd(beta,beta2);
1714
1715     ewtab            = fr->ic->tabq_coul_F;
1716     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1717     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1718
1719     /* Setup water-specific parameters */
1720     inr              = nlist->iinr[0];
1721     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1722     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1723     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1724     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1725     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
1726
1727     jq1              = _mm256_set1_pd(charge[inr+1]);
1728     jq2              = _mm256_set1_pd(charge[inr+2]);
1729     jq3              = _mm256_set1_pd(charge[inr+3]);
1730     vdwjidx0A        = 2*vdwtype[inr+0];
1731     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1732     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1733     c6grid_00        = _mm256_set1_pd(vdwgridioffsetptr0[vdwjidx0A]);
1734     qq11             = _mm256_mul_pd(iq1,jq1);
1735     qq12             = _mm256_mul_pd(iq1,jq2);
1736     qq13             = _mm256_mul_pd(iq1,jq3);
1737     qq21             = _mm256_mul_pd(iq2,jq1);
1738     qq22             = _mm256_mul_pd(iq2,jq2);
1739     qq23             = _mm256_mul_pd(iq2,jq3);
1740     qq31             = _mm256_mul_pd(iq3,jq1);
1741     qq32             = _mm256_mul_pd(iq3,jq2);
1742     qq33             = _mm256_mul_pd(iq3,jq3);
1743
1744     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1745     rcutoff_scalar   = fr->rcoulomb;
1746     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
1747     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
1748
1749     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
1750     rvdw             = _mm256_set1_pd(fr->rvdw);
1751
1752     /* Avoid stupid compiler warnings */
1753     jnrA = jnrB = jnrC = jnrD = 0;
1754     j_coord_offsetA = 0;
1755     j_coord_offsetB = 0;
1756     j_coord_offsetC = 0;
1757     j_coord_offsetD = 0;
1758
1759     outeriter        = 0;
1760     inneriter        = 0;
1761
1762     for(iidx=0;iidx<4*DIM;iidx++)
1763     {
1764         scratch[iidx] = 0.0;
1765     }
1766
1767     /* Start outer loop over neighborlists */
1768     for(iidx=0; iidx<nri; iidx++)
1769     {
1770         /* Load shift vector for this list */
1771         i_shift_offset   = DIM*shiftidx[iidx];
1772
1773         /* Load limits for loop over neighbors */
1774         j_index_start    = jindex[iidx];
1775         j_index_end      = jindex[iidx+1];
1776
1777         /* Get outer coordinate index */
1778         inr              = iinr[iidx];
1779         i_coord_offset   = DIM*inr;
1780
1781         /* Load i particle coords and add shift vector */
1782         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1783                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1784
1785         fix0             = _mm256_setzero_pd();
1786         fiy0             = _mm256_setzero_pd();
1787         fiz0             = _mm256_setzero_pd();
1788         fix1             = _mm256_setzero_pd();
1789         fiy1             = _mm256_setzero_pd();
1790         fiz1             = _mm256_setzero_pd();
1791         fix2             = _mm256_setzero_pd();
1792         fiy2             = _mm256_setzero_pd();
1793         fiz2             = _mm256_setzero_pd();
1794         fix3             = _mm256_setzero_pd();
1795         fiy3             = _mm256_setzero_pd();
1796         fiz3             = _mm256_setzero_pd();
1797
1798         /* Start inner kernel loop */
1799         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1800         {
1801
1802             /* Get j neighbor index, and coordinate index */
1803             jnrA             = jjnr[jidx];
1804             jnrB             = jjnr[jidx+1];
1805             jnrC             = jjnr[jidx+2];
1806             jnrD             = jjnr[jidx+3];
1807             j_coord_offsetA  = DIM*jnrA;
1808             j_coord_offsetB  = DIM*jnrB;
1809             j_coord_offsetC  = DIM*jnrC;
1810             j_coord_offsetD  = DIM*jnrD;
1811
1812             /* load j atom coordinates */
1813             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1814                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1815                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1816                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1817
1818             /* Calculate displacement vector */
1819             dx00             = _mm256_sub_pd(ix0,jx0);
1820             dy00             = _mm256_sub_pd(iy0,jy0);
1821             dz00             = _mm256_sub_pd(iz0,jz0);
1822             dx11             = _mm256_sub_pd(ix1,jx1);
1823             dy11             = _mm256_sub_pd(iy1,jy1);
1824             dz11             = _mm256_sub_pd(iz1,jz1);
1825             dx12             = _mm256_sub_pd(ix1,jx2);
1826             dy12             = _mm256_sub_pd(iy1,jy2);
1827             dz12             = _mm256_sub_pd(iz1,jz2);
1828             dx13             = _mm256_sub_pd(ix1,jx3);
1829             dy13             = _mm256_sub_pd(iy1,jy3);
1830             dz13             = _mm256_sub_pd(iz1,jz3);
1831             dx21             = _mm256_sub_pd(ix2,jx1);
1832             dy21             = _mm256_sub_pd(iy2,jy1);
1833             dz21             = _mm256_sub_pd(iz2,jz1);
1834             dx22             = _mm256_sub_pd(ix2,jx2);
1835             dy22             = _mm256_sub_pd(iy2,jy2);
1836             dz22             = _mm256_sub_pd(iz2,jz2);
1837             dx23             = _mm256_sub_pd(ix2,jx3);
1838             dy23             = _mm256_sub_pd(iy2,jy3);
1839             dz23             = _mm256_sub_pd(iz2,jz3);
1840             dx31             = _mm256_sub_pd(ix3,jx1);
1841             dy31             = _mm256_sub_pd(iy3,jy1);
1842             dz31             = _mm256_sub_pd(iz3,jz1);
1843             dx32             = _mm256_sub_pd(ix3,jx2);
1844             dy32             = _mm256_sub_pd(iy3,jy2);
1845             dz32             = _mm256_sub_pd(iz3,jz2);
1846             dx33             = _mm256_sub_pd(ix3,jx3);
1847             dy33             = _mm256_sub_pd(iy3,jy3);
1848             dz33             = _mm256_sub_pd(iz3,jz3);
1849
1850             /* Calculate squared distance and things based on it */
1851             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1852             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1853             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1854             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1855             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1856             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1857             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1858             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1859             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1860             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1861
1862             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1863             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1864             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1865             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1866             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1867             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1868             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1869             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1870             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1871             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1872
1873             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1874             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1875             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1876             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1877             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1878             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1879             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1880             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1881             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1882             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1883
1884             fjx0             = _mm256_setzero_pd();
1885             fjy0             = _mm256_setzero_pd();
1886             fjz0             = _mm256_setzero_pd();
1887             fjx1             = _mm256_setzero_pd();
1888             fjy1             = _mm256_setzero_pd();
1889             fjz1             = _mm256_setzero_pd();
1890             fjx2             = _mm256_setzero_pd();
1891             fjy2             = _mm256_setzero_pd();
1892             fjz2             = _mm256_setzero_pd();
1893             fjx3             = _mm256_setzero_pd();
1894             fjy3             = _mm256_setzero_pd();
1895             fjz3             = _mm256_setzero_pd();
1896
1897             /**************************
1898              * CALCULATE INTERACTIONS *
1899              **************************/
1900
1901             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1902             {
1903
1904             r00              = _mm256_mul_pd(rsq00,rinv00);
1905
1906             /* Analytical LJ-PME */
1907             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1908             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1909             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1910             exponent         = gmx_simd_exp_d(ewcljrsq);
1911             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1912             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1913             /* f6A = 6 * C6grid * (1 - poly) */
1914             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1915             /* f6B = C6grid * exponent * beta^6 */
1916             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1917             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1918             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);
1919
1920             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1921
1922             fscal            = fvdw;
1923
1924             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1925
1926             /* Calculate temporary vectorial force */
1927             tx               = _mm256_mul_pd(fscal,dx00);
1928             ty               = _mm256_mul_pd(fscal,dy00);
1929             tz               = _mm256_mul_pd(fscal,dz00);
1930
1931             /* Update vectorial force */
1932             fix0             = _mm256_add_pd(fix0,tx);
1933             fiy0             = _mm256_add_pd(fiy0,ty);
1934             fiz0             = _mm256_add_pd(fiz0,tz);
1935
1936             fjx0             = _mm256_add_pd(fjx0,tx);
1937             fjy0             = _mm256_add_pd(fjy0,ty);
1938             fjz0             = _mm256_add_pd(fjz0,tz);
1939
1940             }
1941
1942             /**************************
1943              * CALCULATE INTERACTIONS *
1944              **************************/
1945
1946             if (gmx_mm256_any_lt(rsq11,rcutoff2))
1947             {
1948
1949             r11              = _mm256_mul_pd(rsq11,rinv11);
1950
1951             /* EWALD ELECTROSTATICS */
1952
1953             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1954             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1955             ewitab           = _mm256_cvttpd_epi32(ewrt);
1956             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1957             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1958                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1959                                             &ewtabF,&ewtabFn);
1960             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1961             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1962
1963             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
1964
1965             fscal            = felec;
1966
1967             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1968
1969             /* Calculate temporary vectorial force */
1970             tx               = _mm256_mul_pd(fscal,dx11);
1971             ty               = _mm256_mul_pd(fscal,dy11);
1972             tz               = _mm256_mul_pd(fscal,dz11);
1973
1974             /* Update vectorial force */
1975             fix1             = _mm256_add_pd(fix1,tx);
1976             fiy1             = _mm256_add_pd(fiy1,ty);
1977             fiz1             = _mm256_add_pd(fiz1,tz);
1978
1979             fjx1             = _mm256_add_pd(fjx1,tx);
1980             fjy1             = _mm256_add_pd(fjy1,ty);
1981             fjz1             = _mm256_add_pd(fjz1,tz);
1982
1983             }
1984
1985             /**************************
1986              * CALCULATE INTERACTIONS *
1987              **************************/
1988
1989             if (gmx_mm256_any_lt(rsq12,rcutoff2))
1990             {
1991
1992             r12              = _mm256_mul_pd(rsq12,rinv12);
1993
1994             /* EWALD ELECTROSTATICS */
1995
1996             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1997             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1998             ewitab           = _mm256_cvttpd_epi32(ewrt);
1999             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2000             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2001                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2002                                             &ewtabF,&ewtabFn);
2003             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2004             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2005
2006             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2007
2008             fscal            = felec;
2009
2010             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2011
2012             /* Calculate temporary vectorial force */
2013             tx               = _mm256_mul_pd(fscal,dx12);
2014             ty               = _mm256_mul_pd(fscal,dy12);
2015             tz               = _mm256_mul_pd(fscal,dz12);
2016
2017             /* Update vectorial force */
2018             fix1             = _mm256_add_pd(fix1,tx);
2019             fiy1             = _mm256_add_pd(fiy1,ty);
2020             fiz1             = _mm256_add_pd(fiz1,tz);
2021
2022             fjx2             = _mm256_add_pd(fjx2,tx);
2023             fjy2             = _mm256_add_pd(fjy2,ty);
2024             fjz2             = _mm256_add_pd(fjz2,tz);
2025
2026             }
2027
2028             /**************************
2029              * CALCULATE INTERACTIONS *
2030              **************************/
2031
2032             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2033             {
2034
2035             r13              = _mm256_mul_pd(rsq13,rinv13);
2036
2037             /* EWALD ELECTROSTATICS */
2038
2039             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2040             ewrt             = _mm256_mul_pd(r13,ewtabscale);
2041             ewitab           = _mm256_cvttpd_epi32(ewrt);
2042             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2043             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2044                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2045                                             &ewtabF,&ewtabFn);
2046             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2047             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
2048
2049             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
2050
2051             fscal            = felec;
2052
2053             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2054
2055             /* Calculate temporary vectorial force */
2056             tx               = _mm256_mul_pd(fscal,dx13);
2057             ty               = _mm256_mul_pd(fscal,dy13);
2058             tz               = _mm256_mul_pd(fscal,dz13);
2059
2060             /* Update vectorial force */
2061             fix1             = _mm256_add_pd(fix1,tx);
2062             fiy1             = _mm256_add_pd(fiy1,ty);
2063             fiz1             = _mm256_add_pd(fiz1,tz);
2064
2065             fjx3             = _mm256_add_pd(fjx3,tx);
2066             fjy3             = _mm256_add_pd(fjy3,ty);
2067             fjz3             = _mm256_add_pd(fjz3,tz);
2068
2069             }
2070
2071             /**************************
2072              * CALCULATE INTERACTIONS *
2073              **************************/
2074
2075             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2076             {
2077
2078             r21              = _mm256_mul_pd(rsq21,rinv21);
2079
2080             /* EWALD ELECTROSTATICS */
2081
2082             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2083             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2084             ewitab           = _mm256_cvttpd_epi32(ewrt);
2085             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2086             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2087                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2088                                             &ewtabF,&ewtabFn);
2089             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2090             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2091
2092             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2093
2094             fscal            = felec;
2095
2096             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2097
2098             /* Calculate temporary vectorial force */
2099             tx               = _mm256_mul_pd(fscal,dx21);
2100             ty               = _mm256_mul_pd(fscal,dy21);
2101             tz               = _mm256_mul_pd(fscal,dz21);
2102
2103             /* Update vectorial force */
2104             fix2             = _mm256_add_pd(fix2,tx);
2105             fiy2             = _mm256_add_pd(fiy2,ty);
2106             fiz2             = _mm256_add_pd(fiz2,tz);
2107
2108             fjx1             = _mm256_add_pd(fjx1,tx);
2109             fjy1             = _mm256_add_pd(fjy1,ty);
2110             fjz1             = _mm256_add_pd(fjz1,tz);
2111
2112             }
2113
2114             /**************************
2115              * CALCULATE INTERACTIONS *
2116              **************************/
2117
2118             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2119             {
2120
2121             r22              = _mm256_mul_pd(rsq22,rinv22);
2122
2123             /* EWALD ELECTROSTATICS */
2124
2125             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2126             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2127             ewitab           = _mm256_cvttpd_epi32(ewrt);
2128             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2129             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2130                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2131                                             &ewtabF,&ewtabFn);
2132             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2133             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2134
2135             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2136
2137             fscal            = felec;
2138
2139             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2140
2141             /* Calculate temporary vectorial force */
2142             tx               = _mm256_mul_pd(fscal,dx22);
2143             ty               = _mm256_mul_pd(fscal,dy22);
2144             tz               = _mm256_mul_pd(fscal,dz22);
2145
2146             /* Update vectorial force */
2147             fix2             = _mm256_add_pd(fix2,tx);
2148             fiy2             = _mm256_add_pd(fiy2,ty);
2149             fiz2             = _mm256_add_pd(fiz2,tz);
2150
2151             fjx2             = _mm256_add_pd(fjx2,tx);
2152             fjy2             = _mm256_add_pd(fjy2,ty);
2153             fjz2             = _mm256_add_pd(fjz2,tz);
2154
2155             }
2156
2157             /**************************
2158              * CALCULATE INTERACTIONS *
2159              **************************/
2160
2161             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2162             {
2163
2164             r23              = _mm256_mul_pd(rsq23,rinv23);
2165
2166             /* EWALD ELECTROSTATICS */
2167
2168             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2169             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2170             ewitab           = _mm256_cvttpd_epi32(ewrt);
2171             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2172             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2173                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2174                                             &ewtabF,&ewtabFn);
2175             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2176             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2177
2178             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2179
2180             fscal            = felec;
2181
2182             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2183
2184             /* Calculate temporary vectorial force */
2185             tx               = _mm256_mul_pd(fscal,dx23);
2186             ty               = _mm256_mul_pd(fscal,dy23);
2187             tz               = _mm256_mul_pd(fscal,dz23);
2188
2189             /* Update vectorial force */
2190             fix2             = _mm256_add_pd(fix2,tx);
2191             fiy2             = _mm256_add_pd(fiy2,ty);
2192             fiz2             = _mm256_add_pd(fiz2,tz);
2193
2194             fjx3             = _mm256_add_pd(fjx3,tx);
2195             fjy3             = _mm256_add_pd(fjy3,ty);
2196             fjz3             = _mm256_add_pd(fjz3,tz);
2197
2198             }
2199
2200             /**************************
2201              * CALCULATE INTERACTIONS *
2202              **************************/
2203
2204             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2205             {
2206
2207             r31              = _mm256_mul_pd(rsq31,rinv31);
2208
2209             /* EWALD ELECTROSTATICS */
2210
2211             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2212             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2213             ewitab           = _mm256_cvttpd_epi32(ewrt);
2214             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2215             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2216                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2217                                             &ewtabF,&ewtabFn);
2218             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2219             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2220
2221             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2222
2223             fscal            = felec;
2224
2225             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2226
2227             /* Calculate temporary vectorial force */
2228             tx               = _mm256_mul_pd(fscal,dx31);
2229             ty               = _mm256_mul_pd(fscal,dy31);
2230             tz               = _mm256_mul_pd(fscal,dz31);
2231
2232             /* Update vectorial force */
2233             fix3             = _mm256_add_pd(fix3,tx);
2234             fiy3             = _mm256_add_pd(fiy3,ty);
2235             fiz3             = _mm256_add_pd(fiz3,tz);
2236
2237             fjx1             = _mm256_add_pd(fjx1,tx);
2238             fjy1             = _mm256_add_pd(fjy1,ty);
2239             fjz1             = _mm256_add_pd(fjz1,tz);
2240
2241             }
2242
2243             /**************************
2244              * CALCULATE INTERACTIONS *
2245              **************************/
2246
2247             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2248             {
2249
2250             r32              = _mm256_mul_pd(rsq32,rinv32);
2251
2252             /* EWALD ELECTROSTATICS */
2253
2254             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2255             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2256             ewitab           = _mm256_cvttpd_epi32(ewrt);
2257             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2258             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2259                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2260                                             &ewtabF,&ewtabFn);
2261             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2262             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
2263
2264             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2265
2266             fscal            = felec;
2267
2268             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2269
2270             /* Calculate temporary vectorial force */
2271             tx               = _mm256_mul_pd(fscal,dx32);
2272             ty               = _mm256_mul_pd(fscal,dy32);
2273             tz               = _mm256_mul_pd(fscal,dz32);
2274
2275             /* Update vectorial force */
2276             fix3             = _mm256_add_pd(fix3,tx);
2277             fiy3             = _mm256_add_pd(fiy3,ty);
2278             fiz3             = _mm256_add_pd(fiz3,tz);
2279
2280             fjx2             = _mm256_add_pd(fjx2,tx);
2281             fjy2             = _mm256_add_pd(fjy2,ty);
2282             fjz2             = _mm256_add_pd(fjz2,tz);
2283
2284             }
2285
2286             /**************************
2287              * CALCULATE INTERACTIONS *
2288              **************************/
2289
2290             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2291             {
2292
2293             r33              = _mm256_mul_pd(rsq33,rinv33);
2294
2295             /* EWALD ELECTROSTATICS */
2296
2297             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2298             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2299             ewitab           = _mm256_cvttpd_epi32(ewrt);
2300             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2301             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2302                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2303                                             &ewtabF,&ewtabFn);
2304             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2305             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2306
2307             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2308
2309             fscal            = felec;
2310
2311             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2312
2313             /* Calculate temporary vectorial force */
2314             tx               = _mm256_mul_pd(fscal,dx33);
2315             ty               = _mm256_mul_pd(fscal,dy33);
2316             tz               = _mm256_mul_pd(fscal,dz33);
2317
2318             /* Update vectorial force */
2319             fix3             = _mm256_add_pd(fix3,tx);
2320             fiy3             = _mm256_add_pd(fiy3,ty);
2321             fiz3             = _mm256_add_pd(fiz3,tz);
2322
2323             fjx3             = _mm256_add_pd(fjx3,tx);
2324             fjy3             = _mm256_add_pd(fjy3,ty);
2325             fjz3             = _mm256_add_pd(fjz3,tz);
2326
2327             }
2328
2329             fjptrA             = f+j_coord_offsetA;
2330             fjptrB             = f+j_coord_offsetB;
2331             fjptrC             = f+j_coord_offsetC;
2332             fjptrD             = f+j_coord_offsetD;
2333
2334             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2335                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2336                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2337
2338             /* Inner loop uses 403 flops */
2339         }
2340
2341         if(jidx<j_index_end)
2342         {
2343
2344             /* Get j neighbor index, and coordinate index */
2345             jnrlistA         = jjnr[jidx];
2346             jnrlistB         = jjnr[jidx+1];
2347             jnrlistC         = jjnr[jidx+2];
2348             jnrlistD         = jjnr[jidx+3];
2349             /* Sign of each element will be negative for non-real atoms.
2350              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2351              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
2352              */
2353             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2354
2355             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
2356             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
2357             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
2358
2359             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2360             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2361             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2362             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2363             j_coord_offsetA  = DIM*jnrA;
2364             j_coord_offsetB  = DIM*jnrB;
2365             j_coord_offsetC  = DIM*jnrC;
2366             j_coord_offsetD  = DIM*jnrD;
2367
2368             /* load j atom coordinates */
2369             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
2370                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2371                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2372                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2373
2374             /* Calculate displacement vector */
2375             dx00             = _mm256_sub_pd(ix0,jx0);
2376             dy00             = _mm256_sub_pd(iy0,jy0);
2377             dz00             = _mm256_sub_pd(iz0,jz0);
2378             dx11             = _mm256_sub_pd(ix1,jx1);
2379             dy11             = _mm256_sub_pd(iy1,jy1);
2380             dz11             = _mm256_sub_pd(iz1,jz1);
2381             dx12             = _mm256_sub_pd(ix1,jx2);
2382             dy12             = _mm256_sub_pd(iy1,jy2);
2383             dz12             = _mm256_sub_pd(iz1,jz2);
2384             dx13             = _mm256_sub_pd(ix1,jx3);
2385             dy13             = _mm256_sub_pd(iy1,jy3);
2386             dz13             = _mm256_sub_pd(iz1,jz3);
2387             dx21             = _mm256_sub_pd(ix2,jx1);
2388             dy21             = _mm256_sub_pd(iy2,jy1);
2389             dz21             = _mm256_sub_pd(iz2,jz1);
2390             dx22             = _mm256_sub_pd(ix2,jx2);
2391             dy22             = _mm256_sub_pd(iy2,jy2);
2392             dz22             = _mm256_sub_pd(iz2,jz2);
2393             dx23             = _mm256_sub_pd(ix2,jx3);
2394             dy23             = _mm256_sub_pd(iy2,jy3);
2395             dz23             = _mm256_sub_pd(iz2,jz3);
2396             dx31             = _mm256_sub_pd(ix3,jx1);
2397             dy31             = _mm256_sub_pd(iy3,jy1);
2398             dz31             = _mm256_sub_pd(iz3,jz1);
2399             dx32             = _mm256_sub_pd(ix3,jx2);
2400             dy32             = _mm256_sub_pd(iy3,jy2);
2401             dz32             = _mm256_sub_pd(iz3,jz2);
2402             dx33             = _mm256_sub_pd(ix3,jx3);
2403             dy33             = _mm256_sub_pd(iy3,jy3);
2404             dz33             = _mm256_sub_pd(iz3,jz3);
2405
2406             /* Calculate squared distance and things based on it */
2407             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
2408             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2409             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2410             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
2411             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2412             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2413             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
2414             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
2415             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
2416             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
2417
2418             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
2419             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
2420             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
2421             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
2422             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
2423             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
2424             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
2425             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
2426             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
2427             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
2428
2429             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
2430             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
2431             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
2432             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
2433             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
2434             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
2435             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
2436             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
2437             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
2438             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
2439
2440             fjx0             = _mm256_setzero_pd();
2441             fjy0             = _mm256_setzero_pd();
2442             fjz0             = _mm256_setzero_pd();
2443             fjx1             = _mm256_setzero_pd();
2444             fjy1             = _mm256_setzero_pd();
2445             fjz1             = _mm256_setzero_pd();
2446             fjx2             = _mm256_setzero_pd();
2447             fjy2             = _mm256_setzero_pd();
2448             fjz2             = _mm256_setzero_pd();
2449             fjx3             = _mm256_setzero_pd();
2450             fjy3             = _mm256_setzero_pd();
2451             fjz3             = _mm256_setzero_pd();
2452
2453             /**************************
2454              * CALCULATE INTERACTIONS *
2455              **************************/
2456
2457             if (gmx_mm256_any_lt(rsq00,rcutoff2))
2458             {
2459
2460             r00              = _mm256_mul_pd(rsq00,rinv00);
2461             r00              = _mm256_andnot_pd(dummy_mask,r00);
2462
2463             /* Analytical LJ-PME */
2464             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2465             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
2466             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
2467             exponent         = gmx_simd_exp_d(ewcljrsq);
2468             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2469             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
2470             /* f6A = 6 * C6grid * (1 - poly) */
2471             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
2472             /* f6B = C6grid * exponent * beta^6 */
2473             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
2474             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2475             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);
2476
2477             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
2478
2479             fscal            = fvdw;
2480
2481             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2482
2483             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2484
2485             /* Calculate temporary vectorial force */
2486             tx               = _mm256_mul_pd(fscal,dx00);
2487             ty               = _mm256_mul_pd(fscal,dy00);
2488             tz               = _mm256_mul_pd(fscal,dz00);
2489
2490             /* Update vectorial force */
2491             fix0             = _mm256_add_pd(fix0,tx);
2492             fiy0             = _mm256_add_pd(fiy0,ty);
2493             fiz0             = _mm256_add_pd(fiz0,tz);
2494
2495             fjx0             = _mm256_add_pd(fjx0,tx);
2496             fjy0             = _mm256_add_pd(fjy0,ty);
2497             fjz0             = _mm256_add_pd(fjz0,tz);
2498
2499             }
2500
2501             /**************************
2502              * CALCULATE INTERACTIONS *
2503              **************************/
2504
2505             if (gmx_mm256_any_lt(rsq11,rcutoff2))
2506             {
2507
2508             r11              = _mm256_mul_pd(rsq11,rinv11);
2509             r11              = _mm256_andnot_pd(dummy_mask,r11);
2510
2511             /* EWALD ELECTROSTATICS */
2512
2513             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2514             ewrt             = _mm256_mul_pd(r11,ewtabscale);
2515             ewitab           = _mm256_cvttpd_epi32(ewrt);
2516             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2517             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2518                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2519                                             &ewtabF,&ewtabFn);
2520             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2521             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2522
2523             cutoff_mask      = _mm256_cmp_pd(rsq11,rcutoff2,_CMP_LT_OQ);
2524
2525             fscal            = felec;
2526
2527             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2528
2529             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2530
2531             /* Calculate temporary vectorial force */
2532             tx               = _mm256_mul_pd(fscal,dx11);
2533             ty               = _mm256_mul_pd(fscal,dy11);
2534             tz               = _mm256_mul_pd(fscal,dz11);
2535
2536             /* Update vectorial force */
2537             fix1             = _mm256_add_pd(fix1,tx);
2538             fiy1             = _mm256_add_pd(fiy1,ty);
2539             fiz1             = _mm256_add_pd(fiz1,tz);
2540
2541             fjx1             = _mm256_add_pd(fjx1,tx);
2542             fjy1             = _mm256_add_pd(fjy1,ty);
2543             fjz1             = _mm256_add_pd(fjz1,tz);
2544
2545             }
2546
2547             /**************************
2548              * CALCULATE INTERACTIONS *
2549              **************************/
2550
2551             if (gmx_mm256_any_lt(rsq12,rcutoff2))
2552             {
2553
2554             r12              = _mm256_mul_pd(rsq12,rinv12);
2555             r12              = _mm256_andnot_pd(dummy_mask,r12);
2556
2557             /* EWALD ELECTROSTATICS */
2558
2559             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2560             ewrt             = _mm256_mul_pd(r12,ewtabscale);
2561             ewitab           = _mm256_cvttpd_epi32(ewrt);
2562             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2563             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2564                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2565                                             &ewtabF,&ewtabFn);
2566             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2567             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2568
2569             cutoff_mask      = _mm256_cmp_pd(rsq12,rcutoff2,_CMP_LT_OQ);
2570
2571             fscal            = felec;
2572
2573             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2574
2575             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2576
2577             /* Calculate temporary vectorial force */
2578             tx               = _mm256_mul_pd(fscal,dx12);
2579             ty               = _mm256_mul_pd(fscal,dy12);
2580             tz               = _mm256_mul_pd(fscal,dz12);
2581
2582             /* Update vectorial force */
2583             fix1             = _mm256_add_pd(fix1,tx);
2584             fiy1             = _mm256_add_pd(fiy1,ty);
2585             fiz1             = _mm256_add_pd(fiz1,tz);
2586
2587             fjx2             = _mm256_add_pd(fjx2,tx);
2588             fjy2             = _mm256_add_pd(fjy2,ty);
2589             fjz2             = _mm256_add_pd(fjz2,tz);
2590
2591             }
2592
2593             /**************************
2594              * CALCULATE INTERACTIONS *
2595              **************************/
2596
2597             if (gmx_mm256_any_lt(rsq13,rcutoff2))
2598             {
2599
2600             r13              = _mm256_mul_pd(rsq13,rinv13);
2601             r13              = _mm256_andnot_pd(dummy_mask,r13);
2602
2603             /* EWALD ELECTROSTATICS */
2604
2605             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2606             ewrt             = _mm256_mul_pd(r13,ewtabscale);
2607             ewitab           = _mm256_cvttpd_epi32(ewrt);
2608             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2609             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2610                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2611                                             &ewtabF,&ewtabFn);
2612             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2613             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
2614
2615             cutoff_mask      = _mm256_cmp_pd(rsq13,rcutoff2,_CMP_LT_OQ);
2616
2617             fscal            = felec;
2618
2619             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2620
2621             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2622
2623             /* Calculate temporary vectorial force */
2624             tx               = _mm256_mul_pd(fscal,dx13);
2625             ty               = _mm256_mul_pd(fscal,dy13);
2626             tz               = _mm256_mul_pd(fscal,dz13);
2627
2628             /* Update vectorial force */
2629             fix1             = _mm256_add_pd(fix1,tx);
2630             fiy1             = _mm256_add_pd(fiy1,ty);
2631             fiz1             = _mm256_add_pd(fiz1,tz);
2632
2633             fjx3             = _mm256_add_pd(fjx3,tx);
2634             fjy3             = _mm256_add_pd(fjy3,ty);
2635             fjz3             = _mm256_add_pd(fjz3,tz);
2636
2637             }
2638
2639             /**************************
2640              * CALCULATE INTERACTIONS *
2641              **************************/
2642
2643             if (gmx_mm256_any_lt(rsq21,rcutoff2))
2644             {
2645
2646             r21              = _mm256_mul_pd(rsq21,rinv21);
2647             r21              = _mm256_andnot_pd(dummy_mask,r21);
2648
2649             /* EWALD ELECTROSTATICS */
2650
2651             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2652             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2653             ewitab           = _mm256_cvttpd_epi32(ewrt);
2654             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2655             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2656                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2657                                             &ewtabF,&ewtabFn);
2658             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2659             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2660
2661             cutoff_mask      = _mm256_cmp_pd(rsq21,rcutoff2,_CMP_LT_OQ);
2662
2663             fscal            = felec;
2664
2665             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2666
2667             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2668
2669             /* Calculate temporary vectorial force */
2670             tx               = _mm256_mul_pd(fscal,dx21);
2671             ty               = _mm256_mul_pd(fscal,dy21);
2672             tz               = _mm256_mul_pd(fscal,dz21);
2673
2674             /* Update vectorial force */
2675             fix2             = _mm256_add_pd(fix2,tx);
2676             fiy2             = _mm256_add_pd(fiy2,ty);
2677             fiz2             = _mm256_add_pd(fiz2,tz);
2678
2679             fjx1             = _mm256_add_pd(fjx1,tx);
2680             fjy1             = _mm256_add_pd(fjy1,ty);
2681             fjz1             = _mm256_add_pd(fjz1,tz);
2682
2683             }
2684
2685             /**************************
2686              * CALCULATE INTERACTIONS *
2687              **************************/
2688
2689             if (gmx_mm256_any_lt(rsq22,rcutoff2))
2690             {
2691
2692             r22              = _mm256_mul_pd(rsq22,rinv22);
2693             r22              = _mm256_andnot_pd(dummy_mask,r22);
2694
2695             /* EWALD ELECTROSTATICS */
2696
2697             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2698             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2699             ewitab           = _mm256_cvttpd_epi32(ewrt);
2700             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2701             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2702                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2703                                             &ewtabF,&ewtabFn);
2704             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2705             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2706
2707             cutoff_mask      = _mm256_cmp_pd(rsq22,rcutoff2,_CMP_LT_OQ);
2708
2709             fscal            = felec;
2710
2711             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2712
2713             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2714
2715             /* Calculate temporary vectorial force */
2716             tx               = _mm256_mul_pd(fscal,dx22);
2717             ty               = _mm256_mul_pd(fscal,dy22);
2718             tz               = _mm256_mul_pd(fscal,dz22);
2719
2720             /* Update vectorial force */
2721             fix2             = _mm256_add_pd(fix2,tx);
2722             fiy2             = _mm256_add_pd(fiy2,ty);
2723             fiz2             = _mm256_add_pd(fiz2,tz);
2724
2725             fjx2             = _mm256_add_pd(fjx2,tx);
2726             fjy2             = _mm256_add_pd(fjy2,ty);
2727             fjz2             = _mm256_add_pd(fjz2,tz);
2728
2729             }
2730
2731             /**************************
2732              * CALCULATE INTERACTIONS *
2733              **************************/
2734
2735             if (gmx_mm256_any_lt(rsq23,rcutoff2))
2736             {
2737
2738             r23              = _mm256_mul_pd(rsq23,rinv23);
2739             r23              = _mm256_andnot_pd(dummy_mask,r23);
2740
2741             /* EWALD ELECTROSTATICS */
2742
2743             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2744             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2745             ewitab           = _mm256_cvttpd_epi32(ewrt);
2746             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2747             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2748                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2749                                             &ewtabF,&ewtabFn);
2750             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2751             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2752
2753             cutoff_mask      = _mm256_cmp_pd(rsq23,rcutoff2,_CMP_LT_OQ);
2754
2755             fscal            = felec;
2756
2757             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2758
2759             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2760
2761             /* Calculate temporary vectorial force */
2762             tx               = _mm256_mul_pd(fscal,dx23);
2763             ty               = _mm256_mul_pd(fscal,dy23);
2764             tz               = _mm256_mul_pd(fscal,dz23);
2765
2766             /* Update vectorial force */
2767             fix2             = _mm256_add_pd(fix2,tx);
2768             fiy2             = _mm256_add_pd(fiy2,ty);
2769             fiz2             = _mm256_add_pd(fiz2,tz);
2770
2771             fjx3             = _mm256_add_pd(fjx3,tx);
2772             fjy3             = _mm256_add_pd(fjy3,ty);
2773             fjz3             = _mm256_add_pd(fjz3,tz);
2774
2775             }
2776
2777             /**************************
2778              * CALCULATE INTERACTIONS *
2779              **************************/
2780
2781             if (gmx_mm256_any_lt(rsq31,rcutoff2))
2782             {
2783
2784             r31              = _mm256_mul_pd(rsq31,rinv31);
2785             r31              = _mm256_andnot_pd(dummy_mask,r31);
2786
2787             /* EWALD ELECTROSTATICS */
2788
2789             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2790             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2791             ewitab           = _mm256_cvttpd_epi32(ewrt);
2792             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2793             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2794                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2795                                             &ewtabF,&ewtabFn);
2796             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2797             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2798
2799             cutoff_mask      = _mm256_cmp_pd(rsq31,rcutoff2,_CMP_LT_OQ);
2800
2801             fscal            = felec;
2802
2803             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2804
2805             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2806
2807             /* Calculate temporary vectorial force */
2808             tx               = _mm256_mul_pd(fscal,dx31);
2809             ty               = _mm256_mul_pd(fscal,dy31);
2810             tz               = _mm256_mul_pd(fscal,dz31);
2811
2812             /* Update vectorial force */
2813             fix3             = _mm256_add_pd(fix3,tx);
2814             fiy3             = _mm256_add_pd(fiy3,ty);
2815             fiz3             = _mm256_add_pd(fiz3,tz);
2816
2817             fjx1             = _mm256_add_pd(fjx1,tx);
2818             fjy1             = _mm256_add_pd(fjy1,ty);
2819             fjz1             = _mm256_add_pd(fjz1,tz);
2820
2821             }
2822
2823             /**************************
2824              * CALCULATE INTERACTIONS *
2825              **************************/
2826
2827             if (gmx_mm256_any_lt(rsq32,rcutoff2))
2828             {
2829
2830             r32              = _mm256_mul_pd(rsq32,rinv32);
2831             r32              = _mm256_andnot_pd(dummy_mask,r32);
2832
2833             /* EWALD ELECTROSTATICS */
2834
2835             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2836             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2837             ewitab           = _mm256_cvttpd_epi32(ewrt);
2838             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2839             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2840                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2841                                             &ewtabF,&ewtabFn);
2842             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2843             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
2844
2845             cutoff_mask      = _mm256_cmp_pd(rsq32,rcutoff2,_CMP_LT_OQ);
2846
2847             fscal            = felec;
2848
2849             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2850
2851             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2852
2853             /* Calculate temporary vectorial force */
2854             tx               = _mm256_mul_pd(fscal,dx32);
2855             ty               = _mm256_mul_pd(fscal,dy32);
2856             tz               = _mm256_mul_pd(fscal,dz32);
2857
2858             /* Update vectorial force */
2859             fix3             = _mm256_add_pd(fix3,tx);
2860             fiy3             = _mm256_add_pd(fiy3,ty);
2861             fiz3             = _mm256_add_pd(fiz3,tz);
2862
2863             fjx2             = _mm256_add_pd(fjx2,tx);
2864             fjy2             = _mm256_add_pd(fjy2,ty);
2865             fjz2             = _mm256_add_pd(fjz2,tz);
2866
2867             }
2868
2869             /**************************
2870              * CALCULATE INTERACTIONS *
2871              **************************/
2872
2873             if (gmx_mm256_any_lt(rsq33,rcutoff2))
2874             {
2875
2876             r33              = _mm256_mul_pd(rsq33,rinv33);
2877             r33              = _mm256_andnot_pd(dummy_mask,r33);
2878
2879             /* EWALD ELECTROSTATICS */
2880
2881             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2882             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2883             ewitab           = _mm256_cvttpd_epi32(ewrt);
2884             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2885             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2886                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2887                                             &ewtabF,&ewtabFn);
2888             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2889             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2890
2891             cutoff_mask      = _mm256_cmp_pd(rsq33,rcutoff2,_CMP_LT_OQ);
2892
2893             fscal            = felec;
2894
2895             fscal            = _mm256_and_pd(fscal,cutoff_mask);
2896
2897             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2898
2899             /* Calculate temporary vectorial force */
2900             tx               = _mm256_mul_pd(fscal,dx33);
2901             ty               = _mm256_mul_pd(fscal,dy33);
2902             tz               = _mm256_mul_pd(fscal,dz33);
2903
2904             /* Update vectorial force */
2905             fix3             = _mm256_add_pd(fix3,tx);
2906             fiy3             = _mm256_add_pd(fiy3,ty);
2907             fiz3             = _mm256_add_pd(fiz3,tz);
2908
2909             fjx3             = _mm256_add_pd(fjx3,tx);
2910             fjy3             = _mm256_add_pd(fjy3,ty);
2911             fjz3             = _mm256_add_pd(fjz3,tz);
2912
2913             }
2914
2915             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2916             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2917             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2918             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2919
2920             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2921                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2922                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2923
2924             /* Inner loop uses 413 flops */
2925         }
2926
2927         /* End of innermost loop */
2928
2929         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2930                                                  f+i_coord_offset,fshift+i_shift_offset);
2931
2932         /* Increment number of inner iterations */
2933         inneriter                  += j_index_end - j_index_start;
2934
2935         /* Outer loop uses 24 flops */
2936     }
2937
2938     /* Increment number of outer iterations */
2939     outeriter        += nri;
2940
2941     /* Update outer/inner flops */
2942
2943     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*413);
2944 }