src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_avx_256_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_VF_avx_256_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Water4
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     real *           vdwioffsetptr1;
  86     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     real *           vdwioffsetptr3;
  90     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  94     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  95     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  96     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  97     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  98     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  99     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 100     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 101     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 102     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 103     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 104     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 105     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 106     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 107     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 108     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 109     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     int              nvdwtype;
 112     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 113     int              *vdwtype;
 114     real             *vdwparam;
 115     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 116     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 117     __m128i          vfitab;
 118     __m128i          ifour       = _mm_set1_epi32(4);
 119     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 120     real             *vftab;
 121     __m256d          dummy_mask,cutoff_mask;
 122     __m128           tmpmask0,tmpmask1;
 123     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 124     __m256d          one     = _mm256_set1_pd(1.0);
 125     __m256d          two     = _mm256_set1_pd(2.0);
 126     x                = xx[0];
 127     f                = ff[0];
 128
 129     nri              = nlist->nri;
 130     iinr             = nlist->iinr;
 131     jindex           = nlist->jindex;
 132     jjnr             = nlist->jjnr;
 133     shiftidx         = nlist->shift;
 134     gid              = nlist->gid;
 135     shiftvec         = fr->shift_vec[0];
 136     fshift           = fr->fshift[0];
 137     facel            = _mm256_set1_pd(fr->ic->epsfac);
 138     charge           = mdatoms->chargeA;
 139     nvdwtype         = fr->ntype;
 140     vdwparam         = fr->nbfp;
 141     vdwtype          = mdatoms->typeA;
 142
 143     vftab            = kernel_data->table_elec->data;
 144     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 145
 146     /* Setup water-specific parameters */
 147     inr              = nlist->iinr[0];
 148     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 149     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 150     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 151     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 152
 153     jq1              = _mm256_set1_pd(charge[inr+1]);
 154     jq2              = _mm256_set1_pd(charge[inr+2]);
 155     jq3              = _mm256_set1_pd(charge[inr+3]);
 156     vdwjidx0A        = 2*vdwtype[inr+0];
 157     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 158     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 159     qq11             = _mm256_mul_pd(iq1,jq1);
 160     qq12             = _mm256_mul_pd(iq1,jq2);
 161     qq13             = _mm256_mul_pd(iq1,jq3);
 162     qq21             = _mm256_mul_pd(iq2,jq1);
 163     qq22             = _mm256_mul_pd(iq2,jq2);
 164     qq23             = _mm256_mul_pd(iq2,jq3);
 165     qq31             = _mm256_mul_pd(iq3,jq1);
 166     qq32             = _mm256_mul_pd(iq3,jq2);
 167     qq33             = _mm256_mul_pd(iq3,jq3);
 168
 169     /* Avoid stupid compiler warnings */
 170     jnrA = jnrB = jnrC = jnrD = 0;
 171     j_coord_offsetA = 0;
 172     j_coord_offsetB = 0;
 173     j_coord_offsetC = 0;
 174     j_coord_offsetD = 0;
 175
 176     outeriter        = 0;
 177     inneriter        = 0;
 178
 179     for(iidx=0;iidx<4*DIM;iidx++)
 180     {
 181         scratch[iidx] = 0.0;
 182     }
 183
 184     /* Start outer loop over neighborlists */
 185     for(iidx=0; iidx<nri; iidx++)
 186     {
 187         /* Load shift vector for this list */
 188         i_shift_offset   = DIM*shiftidx[iidx];
 189
 190         /* Load limits for loop over neighbors */
 191         j_index_start    = jindex[iidx];
 192         j_index_end      = jindex[iidx+1];
 193
 194         /* Get outer coordinate index */
 195         inr              = iinr[iidx];
 196         i_coord_offset   = DIM*inr;
 197
 198         /* Load i particle coords and add shift vector */
 199         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 200                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 201
 202         fix0             = _mm256_setzero_pd();
 203         fiy0             = _mm256_setzero_pd();
 204         fiz0             = _mm256_setzero_pd();
 205         fix1             = _mm256_setzero_pd();
 206         fiy1             = _mm256_setzero_pd();
 207         fiz1             = _mm256_setzero_pd();
 208         fix2             = _mm256_setzero_pd();
 209         fiy2             = _mm256_setzero_pd();
 210         fiz2             = _mm256_setzero_pd();
 211         fix3             = _mm256_setzero_pd();
 212         fiy3             = _mm256_setzero_pd();
 213         fiz3             = _mm256_setzero_pd();
 214
 215         /* Reset potential sums */
 216         velecsum         = _mm256_setzero_pd();
 217         vvdwsum          = _mm256_setzero_pd();
 218
 219         /* Start inner kernel loop */
 220         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 221         {
 222
 223             /* Get j neighbor index, and coordinate index */
 224             jnrA             = jjnr[jidx];
 225             jnrB             = jjnr[jidx+1];
 226             jnrC             = jjnr[jidx+2];
 227             jnrD             = jjnr[jidx+3];
 228             j_coord_offsetA  = DIM*jnrA;
 229             j_coord_offsetB  = DIM*jnrB;
 230             j_coord_offsetC  = DIM*jnrC;
 231             j_coord_offsetD  = DIM*jnrD;
 232
 233             /* load j atom coordinates */
 234             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 235                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 236                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 237                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 238
 239             /* Calculate displacement vector */
 240             dx00             = _mm256_sub_pd(ix0,jx0);
 241             dy00             = _mm256_sub_pd(iy0,jy0);
 242             dz00             = _mm256_sub_pd(iz0,jz0);
 243             dx11             = _mm256_sub_pd(ix1,jx1);
 244             dy11             = _mm256_sub_pd(iy1,jy1);
 245             dz11             = _mm256_sub_pd(iz1,jz1);
 246             dx12             = _mm256_sub_pd(ix1,jx2);
 247             dy12             = _mm256_sub_pd(iy1,jy2);
 248             dz12             = _mm256_sub_pd(iz1,jz2);
 249             dx13             = _mm256_sub_pd(ix1,jx3);
 250             dy13             = _mm256_sub_pd(iy1,jy3);
 251             dz13             = _mm256_sub_pd(iz1,jz3);
 252             dx21             = _mm256_sub_pd(ix2,jx1);
 253             dy21             = _mm256_sub_pd(iy2,jy1);
 254             dz21             = _mm256_sub_pd(iz2,jz1);
 255             dx22             = _mm256_sub_pd(ix2,jx2);
 256             dy22             = _mm256_sub_pd(iy2,jy2);
 257             dz22             = _mm256_sub_pd(iz2,jz2);
 258             dx23             = _mm256_sub_pd(ix2,jx3);
 259             dy23             = _mm256_sub_pd(iy2,jy3);
 260             dz23             = _mm256_sub_pd(iz2,jz3);
 261             dx31             = _mm256_sub_pd(ix3,jx1);
 262             dy31             = _mm256_sub_pd(iy3,jy1);
 263             dz31             = _mm256_sub_pd(iz3,jz1);
 264             dx32             = _mm256_sub_pd(ix3,jx2);
 265             dy32             = _mm256_sub_pd(iy3,jy2);
 266             dz32             = _mm256_sub_pd(iz3,jz2);
 267             dx33             = _mm256_sub_pd(ix3,jx3);
 268             dy33             = _mm256_sub_pd(iy3,jy3);
 269             dz33             = _mm256_sub_pd(iz3,jz3);
 270
 271             /* Calculate squared distance and things based on it */
 272             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 273             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 274             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 275             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 276             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 277             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 278             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 279             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 280             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 281             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 282
 283             rinv11           = avx256_invsqrt_d(rsq11);
 284             rinv12           = avx256_invsqrt_d(rsq12);
 285             rinv13           = avx256_invsqrt_d(rsq13);
 286             rinv21           = avx256_invsqrt_d(rsq21);
 287             rinv22           = avx256_invsqrt_d(rsq22);
 288             rinv23           = avx256_invsqrt_d(rsq23);
 289             rinv31           = avx256_invsqrt_d(rsq31);
 290             rinv32           = avx256_invsqrt_d(rsq32);
 291             rinv33           = avx256_invsqrt_d(rsq33);
 292
 293             rinvsq00         = avx256_inv_d(rsq00);
 294
 295             fjx0             = _mm256_setzero_pd();
 296             fjy0             = _mm256_setzero_pd();
 297             fjz0             = _mm256_setzero_pd();
 298             fjx1             = _mm256_setzero_pd();
 299             fjy1             = _mm256_setzero_pd();
 300             fjz1             = _mm256_setzero_pd();
 301             fjx2             = _mm256_setzero_pd();
 302             fjy2             = _mm256_setzero_pd();
 303             fjz2             = _mm256_setzero_pd();
 304             fjx3             = _mm256_setzero_pd();
 305             fjy3             = _mm256_setzero_pd();
 306             fjz3             = _mm256_setzero_pd();
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             /* LENNARD-JONES DISPERSION/REPULSION */
 313
 314             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 315             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 316             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 317             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 318             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 322
 323             fscal            = fvdw;
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm256_mul_pd(fscal,dx00);
 327             ty               = _mm256_mul_pd(fscal,dy00);
 328             tz               = _mm256_mul_pd(fscal,dz00);
 329
 330             /* Update vectorial force */
 331             fix0             = _mm256_add_pd(fix0,tx);
 332             fiy0             = _mm256_add_pd(fiy0,ty);
 333             fiz0             = _mm256_add_pd(fiz0,tz);
 334
 335             fjx0             = _mm256_add_pd(fjx0,tx);
 336             fjy0             = _mm256_add_pd(fjy0,ty);
 337             fjz0             = _mm256_add_pd(fjz0,tz);
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             r11              = _mm256_mul_pd(rsq11,rinv11);
 344
 345             /* Calculate table index by multiplying r with table scale and truncate to integer */
 346             rt               = _mm256_mul_pd(r11,vftabscale);
 347             vfitab           = _mm256_cvttpd_epi32(rt);
 348             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 349             vfitab           = _mm_slli_epi32(vfitab,2);
 350
 351             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 352             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 353             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 354             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 355             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 356             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 357             Heps             = _mm256_mul_pd(vfeps,H);
 358             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 359             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 360             velec            = _mm256_mul_pd(qq11,VV);
 361             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 362             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velecsum         = _mm256_add_pd(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm256_mul_pd(fscal,dx11);
 371             ty               = _mm256_mul_pd(fscal,dy11);
 372             tz               = _mm256_mul_pd(fscal,dz11);
 373
 374             /* Update vectorial force */
 375             fix1             = _mm256_add_pd(fix1,tx);
 376             fiy1             = _mm256_add_pd(fiy1,ty);
 377             fiz1             = _mm256_add_pd(fiz1,tz);
 378
 379             fjx1             = _mm256_add_pd(fjx1,tx);
 380             fjy1             = _mm256_add_pd(fjy1,ty);
 381             fjz1             = _mm256_add_pd(fjz1,tz);
 382
 383             /**************************
 384              * CALCULATE INTERACTIONS *
 385              **************************/
 386
 387             r12              = _mm256_mul_pd(rsq12,rinv12);
 388
 389             /* Calculate table index by multiplying r with table scale and truncate to integer */
 390             rt               = _mm256_mul_pd(r12,vftabscale);
 391             vfitab           = _mm256_cvttpd_epi32(rt);
 392             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 393             vfitab           = _mm_slli_epi32(vfitab,2);
 394
 395             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 396             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 397             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 398             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 399             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 400             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 401             Heps             = _mm256_mul_pd(vfeps,H);
 402             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 403             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 404             velec            = _mm256_mul_pd(qq12,VV);
 405             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 406             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
 407
 408             /* Update potential sum for this i atom from the interaction with this j atom. */
 409             velecsum         = _mm256_add_pd(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413             /* Calculate temporary vectorial force */
 414             tx               = _mm256_mul_pd(fscal,dx12);
 415             ty               = _mm256_mul_pd(fscal,dy12);
 416             tz               = _mm256_mul_pd(fscal,dz12);
 417
 418             /* Update vectorial force */
 419             fix1             = _mm256_add_pd(fix1,tx);
 420             fiy1             = _mm256_add_pd(fiy1,ty);
 421             fiz1             = _mm256_add_pd(fiz1,tz);
 422
 423             fjx2             = _mm256_add_pd(fjx2,tx);
 424             fjy2             = _mm256_add_pd(fjy2,ty);
 425             fjz2             = _mm256_add_pd(fjz2,tz);
 426
 427             /**************************
 428              * CALCULATE INTERACTIONS *
 429              **************************/
 430
 431             r13              = _mm256_mul_pd(rsq13,rinv13);
 432
 433             /* Calculate table index by multiplying r with table scale and truncate to integer */
 434             rt               = _mm256_mul_pd(r13,vftabscale);
 435             vfitab           = _mm256_cvttpd_epi32(rt);
 436             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 437             vfitab           = _mm_slli_epi32(vfitab,2);
 438
 439             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 440             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 441             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 442             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 443             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 444             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 445             Heps             = _mm256_mul_pd(vfeps,H);
 446             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 447             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 448             velec            = _mm256_mul_pd(qq13,VV);
 449             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 450             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
 451
 452             /* Update potential sum for this i atom from the interaction with this j atom. */
 453             velecsum         = _mm256_add_pd(velecsum,velec);
 454
 455             fscal            = felec;
 456
 457             /* Calculate temporary vectorial force */
 458             tx               = _mm256_mul_pd(fscal,dx13);
 459             ty               = _mm256_mul_pd(fscal,dy13);
 460             tz               = _mm256_mul_pd(fscal,dz13);
 461
 462             /* Update vectorial force */
 463             fix1             = _mm256_add_pd(fix1,tx);
 464             fiy1             = _mm256_add_pd(fiy1,ty);
 465             fiz1             = _mm256_add_pd(fiz1,tz);
 466
 467             fjx3             = _mm256_add_pd(fjx3,tx);
 468             fjy3             = _mm256_add_pd(fjy3,ty);
 469             fjz3             = _mm256_add_pd(fjz3,tz);
 470
 471             /**************************
 472              * CALCULATE INTERACTIONS *
 473              **************************/
 474
 475             r21              = _mm256_mul_pd(rsq21,rinv21);
 476
 477             /* Calculate table index by multiplying r with table scale and truncate to integer */
 478             rt               = _mm256_mul_pd(r21,vftabscale);
 479             vfitab           = _mm256_cvttpd_epi32(rt);
 480             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 481             vfitab           = _mm_slli_epi32(vfitab,2);
 482
 483             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 484             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 485             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 486             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 487             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 488             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 489             Heps             = _mm256_mul_pd(vfeps,H);
 490             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 491             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 492             velec            = _mm256_mul_pd(qq21,VV);
 493             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 494             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
 495
 496             /* Update potential sum for this i atom from the interaction with this j atom. */
 497             velecsum         = _mm256_add_pd(velecsum,velec);
 498
 499             fscal            = felec;
 500
 501             /* Calculate temporary vectorial force */
 502             tx               = _mm256_mul_pd(fscal,dx21);
 503             ty               = _mm256_mul_pd(fscal,dy21);
 504             tz               = _mm256_mul_pd(fscal,dz21);
 505
 506             /* Update vectorial force */
 507             fix2             = _mm256_add_pd(fix2,tx);
 508             fiy2             = _mm256_add_pd(fiy2,ty);
 509             fiz2             = _mm256_add_pd(fiz2,tz);
 510
 511             fjx1             = _mm256_add_pd(fjx1,tx);
 512             fjy1             = _mm256_add_pd(fjy1,ty);
 513             fjz1             = _mm256_add_pd(fjz1,tz);
 514
 515             /**************************
 516              * CALCULATE INTERACTIONS *
 517              **************************/
 518
 519             r22              = _mm256_mul_pd(rsq22,rinv22);
 520
 521             /* Calculate table index by multiplying r with table scale and truncate to integer */
 522             rt               = _mm256_mul_pd(r22,vftabscale);
 523             vfitab           = _mm256_cvttpd_epi32(rt);
 524             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 525             vfitab           = _mm_slli_epi32(vfitab,2);
 526
 527             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 528             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 529             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 530             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 531             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 532             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 533             Heps             = _mm256_mul_pd(vfeps,H);
 534             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 535             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 536             velec            = _mm256_mul_pd(qq22,VV);
 537             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 538             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velecsum         = _mm256_add_pd(velecsum,velec);
 542
 543             fscal            = felec;
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm256_mul_pd(fscal,dx22);
 547             ty               = _mm256_mul_pd(fscal,dy22);
 548             tz               = _mm256_mul_pd(fscal,dz22);
 549
 550             /* Update vectorial force */
 551             fix2             = _mm256_add_pd(fix2,tx);
 552             fiy2             = _mm256_add_pd(fiy2,ty);
 553             fiz2             = _mm256_add_pd(fiz2,tz);
 554
 555             fjx2             = _mm256_add_pd(fjx2,tx);
 556             fjy2             = _mm256_add_pd(fjy2,ty);
 557             fjz2             = _mm256_add_pd(fjz2,tz);
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r23              = _mm256_mul_pd(rsq23,rinv23);
 564
 565             /* Calculate table index by multiplying r with table scale and truncate to integer */
 566             rt               = _mm256_mul_pd(r23,vftabscale);
 567             vfitab           = _mm256_cvttpd_epi32(rt);
 568             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 569             vfitab           = _mm_slli_epi32(vfitab,2);
 570
 571             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 572             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 573             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 574             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 575             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 576             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 577             Heps             = _mm256_mul_pd(vfeps,H);
 578             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 579             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 580             velec            = _mm256_mul_pd(qq23,VV);
 581             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 582             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velecsum         = _mm256_add_pd(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             /* Calculate temporary vectorial force */
 590             tx               = _mm256_mul_pd(fscal,dx23);
 591             ty               = _mm256_mul_pd(fscal,dy23);
 592             tz               = _mm256_mul_pd(fscal,dz23);
 593
 594             /* Update vectorial force */
 595             fix2             = _mm256_add_pd(fix2,tx);
 596             fiy2             = _mm256_add_pd(fiy2,ty);
 597             fiz2             = _mm256_add_pd(fiz2,tz);
 598
 599             fjx3             = _mm256_add_pd(fjx3,tx);
 600             fjy3             = _mm256_add_pd(fjy3,ty);
 601             fjz3             = _mm256_add_pd(fjz3,tz);
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             r31              = _mm256_mul_pd(rsq31,rinv31);
 608
 609             /* Calculate table index by multiplying r with table scale and truncate to integer */
 610             rt               = _mm256_mul_pd(r31,vftabscale);
 611             vfitab           = _mm256_cvttpd_epi32(rt);
 612             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 613             vfitab           = _mm_slli_epi32(vfitab,2);
 614
 615             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 616             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 617             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 618             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 619             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 620             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 621             Heps             = _mm256_mul_pd(vfeps,H);
 622             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 623             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 624             velec            = _mm256_mul_pd(qq31,VV);
 625             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 626             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
 627
 628             /* Update potential sum for this i atom from the interaction with this j atom. */
 629             velecsum         = _mm256_add_pd(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm256_mul_pd(fscal,dx31);
 635             ty               = _mm256_mul_pd(fscal,dy31);
 636             tz               = _mm256_mul_pd(fscal,dz31);
 637
 638             /* Update vectorial force */
 639             fix3             = _mm256_add_pd(fix3,tx);
 640             fiy3             = _mm256_add_pd(fiy3,ty);
 641             fiz3             = _mm256_add_pd(fiz3,tz);
 642
 643             fjx1             = _mm256_add_pd(fjx1,tx);
 644             fjy1             = _mm256_add_pd(fjy1,ty);
 645             fjz1             = _mm256_add_pd(fjz1,tz);
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             r32              = _mm256_mul_pd(rsq32,rinv32);
 652
 653             /* Calculate table index by multiplying r with table scale and truncate to integer */
 654             rt               = _mm256_mul_pd(r32,vftabscale);
 655             vfitab           = _mm256_cvttpd_epi32(rt);
 656             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 657             vfitab           = _mm_slli_epi32(vfitab,2);
 658
 659             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 660             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 661             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 662             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 663             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 664             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 665             Heps             = _mm256_mul_pd(vfeps,H);
 666             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 667             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 668             velec            = _mm256_mul_pd(qq32,VV);
 669             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 670             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
 671
 672             /* Update potential sum for this i atom from the interaction with this j atom. */
 673             velecsum         = _mm256_add_pd(velecsum,velec);
 674
 675             fscal            = felec;
 676
 677             /* Calculate temporary vectorial force */
 678             tx               = _mm256_mul_pd(fscal,dx32);
 679             ty               = _mm256_mul_pd(fscal,dy32);
 680             tz               = _mm256_mul_pd(fscal,dz32);
 681
 682             /* Update vectorial force */
 683             fix3             = _mm256_add_pd(fix3,tx);
 684             fiy3             = _mm256_add_pd(fiy3,ty);
 685             fiz3             = _mm256_add_pd(fiz3,tz);
 686
 687             fjx2             = _mm256_add_pd(fjx2,tx);
 688             fjy2             = _mm256_add_pd(fjy2,ty);
 689             fjz2             = _mm256_add_pd(fjz2,tz);
 690
 691             /**************************
 692              * CALCULATE INTERACTIONS *
 693              **************************/
 694
 695             r33              = _mm256_mul_pd(rsq33,rinv33);
 696
 697             /* Calculate table index by multiplying r with table scale and truncate to integer */
 698             rt               = _mm256_mul_pd(r33,vftabscale);
 699             vfitab           = _mm256_cvttpd_epi32(rt);
 700             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 701             vfitab           = _mm_slli_epi32(vfitab,2);
 702
 703             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 704             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 705             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 706             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 707             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 708             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 709             Heps             = _mm256_mul_pd(vfeps,H);
 710             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 711             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 712             velec            = _mm256_mul_pd(qq33,VV);
 713             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 714             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
 715
 716             /* Update potential sum for this i atom from the interaction with this j atom. */
 717             velecsum         = _mm256_add_pd(velecsum,velec);
 718
 719             fscal            = felec;
 720
 721             /* Calculate temporary vectorial force */
 722             tx               = _mm256_mul_pd(fscal,dx33);
 723             ty               = _mm256_mul_pd(fscal,dy33);
 724             tz               = _mm256_mul_pd(fscal,dz33);
 725
 726             /* Update vectorial force */
 727             fix3             = _mm256_add_pd(fix3,tx);
 728             fiy3             = _mm256_add_pd(fiy3,ty);
 729             fiz3             = _mm256_add_pd(fiz3,tz);
 730
 731             fjx3             = _mm256_add_pd(fjx3,tx);
 732             fjy3             = _mm256_add_pd(fjy3,ty);
 733             fjz3             = _mm256_add_pd(fjz3,tz);
 734
 735             fjptrA             = f+j_coord_offsetA;
 736             fjptrB             = f+j_coord_offsetB;
 737             fjptrC             = f+j_coord_offsetC;
 738             fjptrD             = f+j_coord_offsetD;
 739
 740             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 741                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 742                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 743
 744             /* Inner loop uses 422 flops */
 745         }
 746
 747         if(jidx<j_index_end)
 748         {
 749
 750             /* Get j neighbor index, and coordinate index */
 751             jnrlistA         = jjnr[jidx];
 752             jnrlistB         = jjnr[jidx+1];
 753             jnrlistC         = jjnr[jidx+2];
 754             jnrlistD         = jjnr[jidx+3];
 755             /* Sign of each element will be negative for non-real atoms.
 756              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 757              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 758              */
 759             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 760
 761             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 762             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 763             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 764
 765             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 766             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 767             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 768             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 769             j_coord_offsetA  = DIM*jnrA;
 770             j_coord_offsetB  = DIM*jnrB;
 771             j_coord_offsetC  = DIM*jnrC;
 772             j_coord_offsetD  = DIM*jnrD;
 773
 774             /* load j atom coordinates */
 775             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 776                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 777                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 778                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 779
 780             /* Calculate displacement vector */
 781             dx00             = _mm256_sub_pd(ix0,jx0);
 782             dy00             = _mm256_sub_pd(iy0,jy0);
 783             dz00             = _mm256_sub_pd(iz0,jz0);
 784             dx11             = _mm256_sub_pd(ix1,jx1);
 785             dy11             = _mm256_sub_pd(iy1,jy1);
 786             dz11             = _mm256_sub_pd(iz1,jz1);
 787             dx12             = _mm256_sub_pd(ix1,jx2);
 788             dy12             = _mm256_sub_pd(iy1,jy2);
 789             dz12             = _mm256_sub_pd(iz1,jz2);
 790             dx13             = _mm256_sub_pd(ix1,jx3);
 791             dy13             = _mm256_sub_pd(iy1,jy3);
 792             dz13             = _mm256_sub_pd(iz1,jz3);
 793             dx21             = _mm256_sub_pd(ix2,jx1);
 794             dy21             = _mm256_sub_pd(iy2,jy1);
 795             dz21             = _mm256_sub_pd(iz2,jz1);
 796             dx22             = _mm256_sub_pd(ix2,jx2);
 797             dy22             = _mm256_sub_pd(iy2,jy2);
 798             dz22             = _mm256_sub_pd(iz2,jz2);
 799             dx23             = _mm256_sub_pd(ix2,jx3);
 800             dy23             = _mm256_sub_pd(iy2,jy3);
 801             dz23             = _mm256_sub_pd(iz2,jz3);
 802             dx31             = _mm256_sub_pd(ix3,jx1);
 803             dy31             = _mm256_sub_pd(iy3,jy1);
 804             dz31             = _mm256_sub_pd(iz3,jz1);
 805             dx32             = _mm256_sub_pd(ix3,jx2);
 806             dy32             = _mm256_sub_pd(iy3,jy2);
 807             dz32             = _mm256_sub_pd(iz3,jz2);
 808             dx33             = _mm256_sub_pd(ix3,jx3);
 809             dy33             = _mm256_sub_pd(iy3,jy3);
 810             dz33             = _mm256_sub_pd(iz3,jz3);
 811
 812             /* Calculate squared distance and things based on it */
 813             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 814             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 815             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 816             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 817             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 818             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 819             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 820             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 821             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 822             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 823
 824             rinv11           = avx256_invsqrt_d(rsq11);
 825             rinv12           = avx256_invsqrt_d(rsq12);
 826             rinv13           = avx256_invsqrt_d(rsq13);
 827             rinv21           = avx256_invsqrt_d(rsq21);
 828             rinv22           = avx256_invsqrt_d(rsq22);
 829             rinv23           = avx256_invsqrt_d(rsq23);
 830             rinv31           = avx256_invsqrt_d(rsq31);
 831             rinv32           = avx256_invsqrt_d(rsq32);
 832             rinv33           = avx256_invsqrt_d(rsq33);
 833
 834             rinvsq00         = avx256_inv_d(rsq00);
 835
 836             fjx0             = _mm256_setzero_pd();
 837             fjy0             = _mm256_setzero_pd();
 838             fjz0             = _mm256_setzero_pd();
 839             fjx1             = _mm256_setzero_pd();
 840             fjy1             = _mm256_setzero_pd();
 841             fjz1             = _mm256_setzero_pd();
 842             fjx2             = _mm256_setzero_pd();
 843             fjy2             = _mm256_setzero_pd();
 844             fjz2             = _mm256_setzero_pd();
 845             fjx3             = _mm256_setzero_pd();
 846             fjy3             = _mm256_setzero_pd();
 847             fjz3             = _mm256_setzero_pd();
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             /* LENNARD-JONES DISPERSION/REPULSION */
 854
 855             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 856             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 857             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 858             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 859             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 860
 861             /* Update potential sum for this i atom from the interaction with this j atom. */
 862             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 863             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 864
 865             fscal            = fvdw;
 866
 867             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 868
 869             /* Calculate temporary vectorial force */
 870             tx               = _mm256_mul_pd(fscal,dx00);
 871             ty               = _mm256_mul_pd(fscal,dy00);
 872             tz               = _mm256_mul_pd(fscal,dz00);
 873
 874             /* Update vectorial force */
 875             fix0             = _mm256_add_pd(fix0,tx);
 876             fiy0             = _mm256_add_pd(fiy0,ty);
 877             fiz0             = _mm256_add_pd(fiz0,tz);
 878
 879             fjx0             = _mm256_add_pd(fjx0,tx);
 880             fjy0             = _mm256_add_pd(fjy0,ty);
 881             fjz0             = _mm256_add_pd(fjz0,tz);
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             r11              = _mm256_mul_pd(rsq11,rinv11);
 888             r11              = _mm256_andnot_pd(dummy_mask,r11);
 889
 890             /* Calculate table index by multiplying r with table scale and truncate to integer */
 891             rt               = _mm256_mul_pd(r11,vftabscale);
 892             vfitab           = _mm256_cvttpd_epi32(rt);
 893             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 894             vfitab           = _mm_slli_epi32(vfitab,2);
 895
 896             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 897             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 898             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 899             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 900             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 901             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 902             Heps             = _mm256_mul_pd(vfeps,H);
 903             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 904             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 905             velec            = _mm256_mul_pd(qq11,VV);
 906             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 907             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 908
 909             /* Update potential sum for this i atom from the interaction with this j atom. */
 910             velec            = _mm256_andnot_pd(dummy_mask,velec);
 911             velecsum         = _mm256_add_pd(velecsum,velec);
 912
 913             fscal            = felec;
 914
 915             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 916
 917             /* Calculate temporary vectorial force */
 918             tx               = _mm256_mul_pd(fscal,dx11);
 919             ty               = _mm256_mul_pd(fscal,dy11);
 920             tz               = _mm256_mul_pd(fscal,dz11);
 921
 922             /* Update vectorial force */
 923             fix1             = _mm256_add_pd(fix1,tx);
 924             fiy1             = _mm256_add_pd(fiy1,ty);
 925             fiz1             = _mm256_add_pd(fiz1,tz);
 926
 927             fjx1             = _mm256_add_pd(fjx1,tx);
 928             fjy1             = _mm256_add_pd(fjy1,ty);
 929             fjz1             = _mm256_add_pd(fjz1,tz);
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             r12              = _mm256_mul_pd(rsq12,rinv12);
 936             r12              = _mm256_andnot_pd(dummy_mask,r12);
 937
 938             /* Calculate table index by multiplying r with table scale and truncate to integer */
 939             rt               = _mm256_mul_pd(r12,vftabscale);
 940             vfitab           = _mm256_cvttpd_epi32(rt);
 941             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 942             vfitab           = _mm_slli_epi32(vfitab,2);
 943
 944             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 945             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 946             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 947             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 948             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 949             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 950             Heps             = _mm256_mul_pd(vfeps,H);
 951             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 952             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 953             velec            = _mm256_mul_pd(qq12,VV);
 954             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 955             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
 956
 957             /* Update potential sum for this i atom from the interaction with this j atom. */
 958             velec            = _mm256_andnot_pd(dummy_mask,velec);
 959             velecsum         = _mm256_add_pd(velecsum,velec);
 960
 961             fscal            = felec;
 962
 963             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 964
 965             /* Calculate temporary vectorial force */
 966             tx               = _mm256_mul_pd(fscal,dx12);
 967             ty               = _mm256_mul_pd(fscal,dy12);
 968             tz               = _mm256_mul_pd(fscal,dz12);
 969
 970             /* Update vectorial force */
 971             fix1             = _mm256_add_pd(fix1,tx);
 972             fiy1             = _mm256_add_pd(fiy1,ty);
 973             fiz1             = _mm256_add_pd(fiz1,tz);
 974
 975             fjx2             = _mm256_add_pd(fjx2,tx);
 976             fjy2             = _mm256_add_pd(fjy2,ty);
 977             fjz2             = _mm256_add_pd(fjz2,tz);
 978
 979             /**************************
 980              * CALCULATE INTERACTIONS *
 981              **************************/
 982
 983             r13              = _mm256_mul_pd(rsq13,rinv13);
 984             r13              = _mm256_andnot_pd(dummy_mask,r13);
 985
 986             /* Calculate table index by multiplying r with table scale and truncate to integer */
 987             rt               = _mm256_mul_pd(r13,vftabscale);
 988             vfitab           = _mm256_cvttpd_epi32(rt);
 989             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 990             vfitab           = _mm_slli_epi32(vfitab,2);
 991
 992             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 993             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 994             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 995             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 996             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 997             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 998             Heps             = _mm256_mul_pd(vfeps,H);
 999             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1000             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1001             velec            = _mm256_mul_pd(qq13,VV);
1002             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1003             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
1004
1005             /* Update potential sum for this i atom from the interaction with this j atom. */
1006             velec            = _mm256_andnot_pd(dummy_mask,velec);
1007             velecsum         = _mm256_add_pd(velecsum,velec);
1008
1009             fscal            = felec;
1010
1011             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1012
1013             /* Calculate temporary vectorial force */
1014             tx               = _mm256_mul_pd(fscal,dx13);
1015             ty               = _mm256_mul_pd(fscal,dy13);
1016             tz               = _mm256_mul_pd(fscal,dz13);
1017
1018             /* Update vectorial force */
1019             fix1             = _mm256_add_pd(fix1,tx);
1020             fiy1             = _mm256_add_pd(fiy1,ty);
1021             fiz1             = _mm256_add_pd(fiz1,tz);
1022
1023             fjx3             = _mm256_add_pd(fjx3,tx);
1024             fjy3             = _mm256_add_pd(fjy3,ty);
1025             fjz3             = _mm256_add_pd(fjz3,tz);
1026
1027             /**************************
1028              * CALCULATE INTERACTIONS *
1029              **************************/
1030
1031             r21              = _mm256_mul_pd(rsq21,rinv21);
1032             r21              = _mm256_andnot_pd(dummy_mask,r21);
1033
1034             /* Calculate table index by multiplying r with table scale and truncate to integer */
1035             rt               = _mm256_mul_pd(r21,vftabscale);
1036             vfitab           = _mm256_cvttpd_epi32(rt);
1037             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1038             vfitab           = _mm_slli_epi32(vfitab,2);
1039
1040             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1041             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1042             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1043             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1044             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1045             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1046             Heps             = _mm256_mul_pd(vfeps,H);
1047             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1048             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1049             velec            = _mm256_mul_pd(qq21,VV);
1050             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1051             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1052
1053             /* Update potential sum for this i atom from the interaction with this j atom. */
1054             velec            = _mm256_andnot_pd(dummy_mask,velec);
1055             velecsum         = _mm256_add_pd(velecsum,velec);
1056
1057             fscal            = felec;
1058
1059             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1060
1061             /* Calculate temporary vectorial force */
1062             tx               = _mm256_mul_pd(fscal,dx21);
1063             ty               = _mm256_mul_pd(fscal,dy21);
1064             tz               = _mm256_mul_pd(fscal,dz21);
1065
1066             /* Update vectorial force */
1067             fix2             = _mm256_add_pd(fix2,tx);
1068             fiy2             = _mm256_add_pd(fiy2,ty);
1069             fiz2             = _mm256_add_pd(fiz2,tz);
1070
1071             fjx1             = _mm256_add_pd(fjx1,tx);
1072             fjy1             = _mm256_add_pd(fjy1,ty);
1073             fjz1             = _mm256_add_pd(fjz1,tz);
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             r22              = _mm256_mul_pd(rsq22,rinv22);
1080             r22              = _mm256_andnot_pd(dummy_mask,r22);
1081
1082             /* Calculate table index by multiplying r with table scale and truncate to integer */
1083             rt               = _mm256_mul_pd(r22,vftabscale);
1084             vfitab           = _mm256_cvttpd_epi32(rt);
1085             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1086             vfitab           = _mm_slli_epi32(vfitab,2);
1087
1088             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1089             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1090             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1091             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1092             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1093             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1094             Heps             = _mm256_mul_pd(vfeps,H);
1095             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1096             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1097             velec            = _mm256_mul_pd(qq22,VV);
1098             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1099             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1100
1101             /* Update potential sum for this i atom from the interaction with this j atom. */
1102             velec            = _mm256_andnot_pd(dummy_mask,velec);
1103             velecsum         = _mm256_add_pd(velecsum,velec);
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm256_mul_pd(fscal,dx22);
1111             ty               = _mm256_mul_pd(fscal,dy22);
1112             tz               = _mm256_mul_pd(fscal,dz22);
1113
1114             /* Update vectorial force */
1115             fix2             = _mm256_add_pd(fix2,tx);
1116             fiy2             = _mm256_add_pd(fiy2,ty);
1117             fiz2             = _mm256_add_pd(fiz2,tz);
1118
1119             fjx2             = _mm256_add_pd(fjx2,tx);
1120             fjy2             = _mm256_add_pd(fjy2,ty);
1121             fjz2             = _mm256_add_pd(fjz2,tz);
1122
1123             /**************************
1124              * CALCULATE INTERACTIONS *
1125              **************************/
1126
1127             r23              = _mm256_mul_pd(rsq23,rinv23);
1128             r23              = _mm256_andnot_pd(dummy_mask,r23);
1129
1130             /* Calculate table index by multiplying r with table scale and truncate to integer */
1131             rt               = _mm256_mul_pd(r23,vftabscale);
1132             vfitab           = _mm256_cvttpd_epi32(rt);
1133             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1134             vfitab           = _mm_slli_epi32(vfitab,2);
1135
1136             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1137             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1138             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1139             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1140             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1141             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1142             Heps             = _mm256_mul_pd(vfeps,H);
1143             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1144             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1145             velec            = _mm256_mul_pd(qq23,VV);
1146             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1147             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
1148
1149             /* Update potential sum for this i atom from the interaction with this j atom. */
1150             velec            = _mm256_andnot_pd(dummy_mask,velec);
1151             velecsum         = _mm256_add_pd(velecsum,velec);
1152
1153             fscal            = felec;
1154
1155             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1156
1157             /* Calculate temporary vectorial force */
1158             tx               = _mm256_mul_pd(fscal,dx23);
1159             ty               = _mm256_mul_pd(fscal,dy23);
1160             tz               = _mm256_mul_pd(fscal,dz23);
1161
1162             /* Update vectorial force */
1163             fix2             = _mm256_add_pd(fix2,tx);
1164             fiy2             = _mm256_add_pd(fiy2,ty);
1165             fiz2             = _mm256_add_pd(fiz2,tz);
1166
1167             fjx3             = _mm256_add_pd(fjx3,tx);
1168             fjy3             = _mm256_add_pd(fjy3,ty);
1169             fjz3             = _mm256_add_pd(fjz3,tz);
1170
1171             /**************************
1172              * CALCULATE INTERACTIONS *
1173              **************************/
1174
1175             r31              = _mm256_mul_pd(rsq31,rinv31);
1176             r31              = _mm256_andnot_pd(dummy_mask,r31);
1177
1178             /* Calculate table index by multiplying r with table scale and truncate to integer */
1179             rt               = _mm256_mul_pd(r31,vftabscale);
1180             vfitab           = _mm256_cvttpd_epi32(rt);
1181             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1182             vfitab           = _mm_slli_epi32(vfitab,2);
1183
1184             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1185             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1186             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1187             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1188             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1189             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1190             Heps             = _mm256_mul_pd(vfeps,H);
1191             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1192             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1193             velec            = _mm256_mul_pd(qq31,VV);
1194             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1195             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
1196
1197             /* Update potential sum for this i atom from the interaction with this j atom. */
1198             velec            = _mm256_andnot_pd(dummy_mask,velec);
1199             velecsum         = _mm256_add_pd(velecsum,velec);
1200
1201             fscal            = felec;
1202
1203             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1204
1205             /* Calculate temporary vectorial force */
1206             tx               = _mm256_mul_pd(fscal,dx31);
1207             ty               = _mm256_mul_pd(fscal,dy31);
1208             tz               = _mm256_mul_pd(fscal,dz31);
1209
1210             /* Update vectorial force */
1211             fix3             = _mm256_add_pd(fix3,tx);
1212             fiy3             = _mm256_add_pd(fiy3,ty);
1213             fiz3             = _mm256_add_pd(fiz3,tz);
1214
1215             fjx1             = _mm256_add_pd(fjx1,tx);
1216             fjy1             = _mm256_add_pd(fjy1,ty);
1217             fjz1             = _mm256_add_pd(fjz1,tz);
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             r32              = _mm256_mul_pd(rsq32,rinv32);
1224             r32              = _mm256_andnot_pd(dummy_mask,r32);
1225
1226             /* Calculate table index by multiplying r with table scale and truncate to integer */
1227             rt               = _mm256_mul_pd(r32,vftabscale);
1228             vfitab           = _mm256_cvttpd_epi32(rt);
1229             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1230             vfitab           = _mm_slli_epi32(vfitab,2);
1231
1232             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1233             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1234             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1235             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1236             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1237             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1238             Heps             = _mm256_mul_pd(vfeps,H);
1239             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1240             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1241             velec            = _mm256_mul_pd(qq32,VV);
1242             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1243             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
1244
1245             /* Update potential sum for this i atom from the interaction with this j atom. */
1246             velec            = _mm256_andnot_pd(dummy_mask,velec);
1247             velecsum         = _mm256_add_pd(velecsum,velec);
1248
1249             fscal            = felec;
1250
1251             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1252
1253             /* Calculate temporary vectorial force */
1254             tx               = _mm256_mul_pd(fscal,dx32);
1255             ty               = _mm256_mul_pd(fscal,dy32);
1256             tz               = _mm256_mul_pd(fscal,dz32);
1257
1258             /* Update vectorial force */
1259             fix3             = _mm256_add_pd(fix3,tx);
1260             fiy3             = _mm256_add_pd(fiy3,ty);
1261             fiz3             = _mm256_add_pd(fiz3,tz);
1262
1263             fjx2             = _mm256_add_pd(fjx2,tx);
1264             fjy2             = _mm256_add_pd(fjy2,ty);
1265             fjz2             = _mm256_add_pd(fjz2,tz);
1266
1267             /**************************
1268              * CALCULATE INTERACTIONS *
1269              **************************/
1270
1271             r33              = _mm256_mul_pd(rsq33,rinv33);
1272             r33              = _mm256_andnot_pd(dummy_mask,r33);
1273
1274             /* Calculate table index by multiplying r with table scale and truncate to integer */
1275             rt               = _mm256_mul_pd(r33,vftabscale);
1276             vfitab           = _mm256_cvttpd_epi32(rt);
1277             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1278             vfitab           = _mm_slli_epi32(vfitab,2);
1279
1280             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1281             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1282             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1283             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1284             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1285             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1286             Heps             = _mm256_mul_pd(vfeps,H);
1287             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1288             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1289             velec            = _mm256_mul_pd(qq33,VV);
1290             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1291             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
1292
1293             /* Update potential sum for this i atom from the interaction with this j atom. */
1294             velec            = _mm256_andnot_pd(dummy_mask,velec);
1295             velecsum         = _mm256_add_pd(velecsum,velec);
1296
1297             fscal            = felec;
1298
1299             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1300
1301             /* Calculate temporary vectorial force */
1302             tx               = _mm256_mul_pd(fscal,dx33);
1303             ty               = _mm256_mul_pd(fscal,dy33);
1304             tz               = _mm256_mul_pd(fscal,dz33);
1305
1306             /* Update vectorial force */
1307             fix3             = _mm256_add_pd(fix3,tx);
1308             fiy3             = _mm256_add_pd(fiy3,ty);
1309             fiz3             = _mm256_add_pd(fiz3,tz);
1310
1311             fjx3             = _mm256_add_pd(fjx3,tx);
1312             fjy3             = _mm256_add_pd(fjy3,ty);
1313             fjz3             = _mm256_add_pd(fjz3,tz);
1314
1315             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1316             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1317             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1318             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1319
1320             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1321                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1322                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1323
1324             /* Inner loop uses 431 flops */
1325         }
1326
1327         /* End of innermost loop */
1328
1329         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1330                                                  f+i_coord_offset,fshift+i_shift_offset);
1331
1332         ggid                        = gid[iidx];
1333         /* Update potential energies */
1334         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1335         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1336
1337         /* Increment number of inner iterations */
1338         inneriter                  += j_index_end - j_index_start;
1339
1340         /* Outer loop uses 26 flops */
1341     }
1342
1343     /* Increment number of outer iterations */
1344     outeriter        += nri;
1345
1346     /* Update outer/inner flops */
1347
1348     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*431);
1349 }
1350 /*
1351  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_F_avx_256_double
1352  * Electrostatics interaction: CubicSplineTable
1353  * VdW interaction:            LennardJones
1354  * Geometry:                   Water4-Water4
1355  * Calculate force/pot:        Force
1356  */
1357 void
1358 nb_kernel_ElecCSTab_VdwLJ_GeomW4W4_F_avx_256_double
1359                     (t_nblist                    * gmx_restrict       nlist,
1360                      rvec                        * gmx_restrict          xx,
1361                      rvec                        * gmx_restrict          ff,
1362                      struct t_forcerec           * gmx_restrict          fr,
1363                      t_mdatoms                   * gmx_restrict     mdatoms,
1364                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1365                      t_nrnb                      * gmx_restrict        nrnb)
1366 {
1367     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1368      * just 0 for non-waters.
1369      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1370      * jnr indices corresponding to data put in the four positions in the SIMD register.
1371      */
1372     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1373     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1374     int              jnrA,jnrB,jnrC,jnrD;
1375     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1376     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1377     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1378     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1379     real             rcutoff_scalar;
1380     real             *shiftvec,*fshift,*x,*f;
1381     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1382     real             scratch[4*DIM];
1383     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1384     real *           vdwioffsetptr0;
1385     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1386     real *           vdwioffsetptr1;
1387     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1388     real *           vdwioffsetptr2;
1389     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1390     real *           vdwioffsetptr3;
1391     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1392     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1393     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1394     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1395     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1396     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1397     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1398     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1399     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1400     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1401     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1402     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1403     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1404     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1405     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1406     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1407     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1408     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1409     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1410     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1411     real             *charge;
1412     int              nvdwtype;
1413     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1414     int              *vdwtype;
1415     real             *vdwparam;
1416     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1417     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1418     __m128i          vfitab;
1419     __m128i          ifour       = _mm_set1_epi32(4);
1420     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1421     real             *vftab;
1422     __m256d          dummy_mask,cutoff_mask;
1423     __m128           tmpmask0,tmpmask1;
1424     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1425     __m256d          one     = _mm256_set1_pd(1.0);
1426     __m256d          two     = _mm256_set1_pd(2.0);
1427     x                = xx[0];
1428     f                = ff[0];
1429
1430     nri              = nlist->nri;
1431     iinr             = nlist->iinr;
1432     jindex           = nlist->jindex;
1433     jjnr             = nlist->jjnr;
1434     shiftidx         = nlist->shift;
1435     gid              = nlist->gid;
1436     shiftvec         = fr->shift_vec[0];
1437     fshift           = fr->fshift[0];
1438     facel            = _mm256_set1_pd(fr->ic->epsfac);
1439     charge           = mdatoms->chargeA;
1440     nvdwtype         = fr->ntype;
1441     vdwparam         = fr->nbfp;
1442     vdwtype          = mdatoms->typeA;
1443
1444     vftab            = kernel_data->table_elec->data;
1445     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
1446
1447     /* Setup water-specific parameters */
1448     inr              = nlist->iinr[0];
1449     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1450     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1451     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1452     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1453
1454     jq1              = _mm256_set1_pd(charge[inr+1]);
1455     jq2              = _mm256_set1_pd(charge[inr+2]);
1456     jq3              = _mm256_set1_pd(charge[inr+3]);
1457     vdwjidx0A        = 2*vdwtype[inr+0];
1458     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1459     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1460     qq11             = _mm256_mul_pd(iq1,jq1);
1461     qq12             = _mm256_mul_pd(iq1,jq2);
1462     qq13             = _mm256_mul_pd(iq1,jq3);
1463     qq21             = _mm256_mul_pd(iq2,jq1);
1464     qq22             = _mm256_mul_pd(iq2,jq2);
1465     qq23             = _mm256_mul_pd(iq2,jq3);
1466     qq31             = _mm256_mul_pd(iq3,jq1);
1467     qq32             = _mm256_mul_pd(iq3,jq2);
1468     qq33             = _mm256_mul_pd(iq3,jq3);
1469
1470     /* Avoid stupid compiler warnings */
1471     jnrA = jnrB = jnrC = jnrD = 0;
1472     j_coord_offsetA = 0;
1473     j_coord_offsetB = 0;
1474     j_coord_offsetC = 0;
1475     j_coord_offsetD = 0;
1476
1477     outeriter        = 0;
1478     inneriter        = 0;
1479
1480     for(iidx=0;iidx<4*DIM;iidx++)
1481     {
1482         scratch[iidx] = 0.0;
1483     }
1484
1485     /* Start outer loop over neighborlists */
1486     for(iidx=0; iidx<nri; iidx++)
1487     {
1488         /* Load shift vector for this list */
1489         i_shift_offset   = DIM*shiftidx[iidx];
1490
1491         /* Load limits for loop over neighbors */
1492         j_index_start    = jindex[iidx];
1493         j_index_end      = jindex[iidx+1];
1494
1495         /* Get outer coordinate index */
1496         inr              = iinr[iidx];
1497         i_coord_offset   = DIM*inr;
1498
1499         /* Load i particle coords and add shift vector */
1500         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1501                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1502
1503         fix0             = _mm256_setzero_pd();
1504         fiy0             = _mm256_setzero_pd();
1505         fiz0             = _mm256_setzero_pd();
1506         fix1             = _mm256_setzero_pd();
1507         fiy1             = _mm256_setzero_pd();
1508         fiz1             = _mm256_setzero_pd();
1509         fix2             = _mm256_setzero_pd();
1510         fiy2             = _mm256_setzero_pd();
1511         fiz2             = _mm256_setzero_pd();
1512         fix3             = _mm256_setzero_pd();
1513         fiy3             = _mm256_setzero_pd();
1514         fiz3             = _mm256_setzero_pd();
1515
1516         /* Start inner kernel loop */
1517         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1518         {
1519
1520             /* Get j neighbor index, and coordinate index */
1521             jnrA             = jjnr[jidx];
1522             jnrB             = jjnr[jidx+1];
1523             jnrC             = jjnr[jidx+2];
1524             jnrD             = jjnr[jidx+3];
1525             j_coord_offsetA  = DIM*jnrA;
1526             j_coord_offsetB  = DIM*jnrB;
1527             j_coord_offsetC  = DIM*jnrC;
1528             j_coord_offsetD  = DIM*jnrD;
1529
1530             /* load j atom coordinates */
1531             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1532                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1533                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1534                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1535
1536             /* Calculate displacement vector */
1537             dx00             = _mm256_sub_pd(ix0,jx0);
1538             dy00             = _mm256_sub_pd(iy0,jy0);
1539             dz00             = _mm256_sub_pd(iz0,jz0);
1540             dx11             = _mm256_sub_pd(ix1,jx1);
1541             dy11             = _mm256_sub_pd(iy1,jy1);
1542             dz11             = _mm256_sub_pd(iz1,jz1);
1543             dx12             = _mm256_sub_pd(ix1,jx2);
1544             dy12             = _mm256_sub_pd(iy1,jy2);
1545             dz12             = _mm256_sub_pd(iz1,jz2);
1546             dx13             = _mm256_sub_pd(ix1,jx3);
1547             dy13             = _mm256_sub_pd(iy1,jy3);
1548             dz13             = _mm256_sub_pd(iz1,jz3);
1549             dx21             = _mm256_sub_pd(ix2,jx1);
1550             dy21             = _mm256_sub_pd(iy2,jy1);
1551             dz21             = _mm256_sub_pd(iz2,jz1);
1552             dx22             = _mm256_sub_pd(ix2,jx2);
1553             dy22             = _mm256_sub_pd(iy2,jy2);
1554             dz22             = _mm256_sub_pd(iz2,jz2);
1555             dx23             = _mm256_sub_pd(ix2,jx3);
1556             dy23             = _mm256_sub_pd(iy2,jy3);
1557             dz23             = _mm256_sub_pd(iz2,jz3);
1558             dx31             = _mm256_sub_pd(ix3,jx1);
1559             dy31             = _mm256_sub_pd(iy3,jy1);
1560             dz31             = _mm256_sub_pd(iz3,jz1);
1561             dx32             = _mm256_sub_pd(ix3,jx2);
1562             dy32             = _mm256_sub_pd(iy3,jy2);
1563             dz32             = _mm256_sub_pd(iz3,jz2);
1564             dx33             = _mm256_sub_pd(ix3,jx3);
1565             dy33             = _mm256_sub_pd(iy3,jy3);
1566             dz33             = _mm256_sub_pd(iz3,jz3);
1567
1568             /* Calculate squared distance and things based on it */
1569             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1570             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1571             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1572             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1573             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1574             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1575             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1576             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1577             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1578             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1579
1580             rinv11           = avx256_invsqrt_d(rsq11);
1581             rinv12           = avx256_invsqrt_d(rsq12);
1582             rinv13           = avx256_invsqrt_d(rsq13);
1583             rinv21           = avx256_invsqrt_d(rsq21);
1584             rinv22           = avx256_invsqrt_d(rsq22);
1585             rinv23           = avx256_invsqrt_d(rsq23);
1586             rinv31           = avx256_invsqrt_d(rsq31);
1587             rinv32           = avx256_invsqrt_d(rsq32);
1588             rinv33           = avx256_invsqrt_d(rsq33);
1589
1590             rinvsq00         = avx256_inv_d(rsq00);
1591
1592             fjx0             = _mm256_setzero_pd();
1593             fjy0             = _mm256_setzero_pd();
1594             fjz0             = _mm256_setzero_pd();
1595             fjx1             = _mm256_setzero_pd();
1596             fjy1             = _mm256_setzero_pd();
1597             fjz1             = _mm256_setzero_pd();
1598             fjx2             = _mm256_setzero_pd();
1599             fjy2             = _mm256_setzero_pd();
1600             fjz2             = _mm256_setzero_pd();
1601             fjx3             = _mm256_setzero_pd();
1602             fjy3             = _mm256_setzero_pd();
1603             fjz3             = _mm256_setzero_pd();
1604
1605             /**************************
1606              * CALCULATE INTERACTIONS *
1607              **************************/
1608
1609             /* LENNARD-JONES DISPERSION/REPULSION */
1610
1611             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1612             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1613
1614             fscal            = fvdw;
1615
1616             /* Calculate temporary vectorial force */
1617             tx               = _mm256_mul_pd(fscal,dx00);
1618             ty               = _mm256_mul_pd(fscal,dy00);
1619             tz               = _mm256_mul_pd(fscal,dz00);
1620
1621             /* Update vectorial force */
1622             fix0             = _mm256_add_pd(fix0,tx);
1623             fiy0             = _mm256_add_pd(fiy0,ty);
1624             fiz0             = _mm256_add_pd(fiz0,tz);
1625
1626             fjx0             = _mm256_add_pd(fjx0,tx);
1627             fjy0             = _mm256_add_pd(fjy0,ty);
1628             fjz0             = _mm256_add_pd(fjz0,tz);
1629
1630             /**************************
1631              * CALCULATE INTERACTIONS *
1632              **************************/
1633
1634             r11              = _mm256_mul_pd(rsq11,rinv11);
1635
1636             /* Calculate table index by multiplying r with table scale and truncate to integer */
1637             rt               = _mm256_mul_pd(r11,vftabscale);
1638             vfitab           = _mm256_cvttpd_epi32(rt);
1639             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1640             vfitab           = _mm_slli_epi32(vfitab,2);
1641
1642             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1643             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1644             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1645             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1646             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1647             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1648             Heps             = _mm256_mul_pd(vfeps,H);
1649             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1650             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1651             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1652
1653             fscal            = felec;
1654
1655             /* Calculate temporary vectorial force */
1656             tx               = _mm256_mul_pd(fscal,dx11);
1657             ty               = _mm256_mul_pd(fscal,dy11);
1658             tz               = _mm256_mul_pd(fscal,dz11);
1659
1660             /* Update vectorial force */
1661             fix1             = _mm256_add_pd(fix1,tx);
1662             fiy1             = _mm256_add_pd(fiy1,ty);
1663             fiz1             = _mm256_add_pd(fiz1,tz);
1664
1665             fjx1             = _mm256_add_pd(fjx1,tx);
1666             fjy1             = _mm256_add_pd(fjy1,ty);
1667             fjz1             = _mm256_add_pd(fjz1,tz);
1668
1669             /**************************
1670              * CALCULATE INTERACTIONS *
1671              **************************/
1672
1673             r12              = _mm256_mul_pd(rsq12,rinv12);
1674
1675             /* Calculate table index by multiplying r with table scale and truncate to integer */
1676             rt               = _mm256_mul_pd(r12,vftabscale);
1677             vfitab           = _mm256_cvttpd_epi32(rt);
1678             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1679             vfitab           = _mm_slli_epi32(vfitab,2);
1680
1681             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1682             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1683             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1684             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1685             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1686             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1687             Heps             = _mm256_mul_pd(vfeps,H);
1688             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1689             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1690             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1691
1692             fscal            = felec;
1693
1694             /* Calculate temporary vectorial force */
1695             tx               = _mm256_mul_pd(fscal,dx12);
1696             ty               = _mm256_mul_pd(fscal,dy12);
1697             tz               = _mm256_mul_pd(fscal,dz12);
1698
1699             /* Update vectorial force */
1700             fix1             = _mm256_add_pd(fix1,tx);
1701             fiy1             = _mm256_add_pd(fiy1,ty);
1702             fiz1             = _mm256_add_pd(fiz1,tz);
1703
1704             fjx2             = _mm256_add_pd(fjx2,tx);
1705             fjy2             = _mm256_add_pd(fjy2,ty);
1706             fjz2             = _mm256_add_pd(fjz2,tz);
1707
1708             /**************************
1709              * CALCULATE INTERACTIONS *
1710              **************************/
1711
1712             r13              = _mm256_mul_pd(rsq13,rinv13);
1713
1714             /* Calculate table index by multiplying r with table scale and truncate to integer */
1715             rt               = _mm256_mul_pd(r13,vftabscale);
1716             vfitab           = _mm256_cvttpd_epi32(rt);
1717             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1718             vfitab           = _mm_slli_epi32(vfitab,2);
1719
1720             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1721             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1722             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1723             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1724             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1725             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1726             Heps             = _mm256_mul_pd(vfeps,H);
1727             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1728             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1729             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
1730
1731             fscal            = felec;
1732
1733             /* Calculate temporary vectorial force */
1734             tx               = _mm256_mul_pd(fscal,dx13);
1735             ty               = _mm256_mul_pd(fscal,dy13);
1736             tz               = _mm256_mul_pd(fscal,dz13);
1737
1738             /* Update vectorial force */
1739             fix1             = _mm256_add_pd(fix1,tx);
1740             fiy1             = _mm256_add_pd(fiy1,ty);
1741             fiz1             = _mm256_add_pd(fiz1,tz);
1742
1743             fjx3             = _mm256_add_pd(fjx3,tx);
1744             fjy3             = _mm256_add_pd(fjy3,ty);
1745             fjz3             = _mm256_add_pd(fjz3,tz);
1746
1747             /**************************
1748              * CALCULATE INTERACTIONS *
1749              **************************/
1750
1751             r21              = _mm256_mul_pd(rsq21,rinv21);
1752
1753             /* Calculate table index by multiplying r with table scale and truncate to integer */
1754             rt               = _mm256_mul_pd(r21,vftabscale);
1755             vfitab           = _mm256_cvttpd_epi32(rt);
1756             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1757             vfitab           = _mm_slli_epi32(vfitab,2);
1758
1759             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1760             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1761             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1762             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1763             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1764             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1765             Heps             = _mm256_mul_pd(vfeps,H);
1766             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1767             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1768             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1769
1770             fscal            = felec;
1771
1772             /* Calculate temporary vectorial force */
1773             tx               = _mm256_mul_pd(fscal,dx21);
1774             ty               = _mm256_mul_pd(fscal,dy21);
1775             tz               = _mm256_mul_pd(fscal,dz21);
1776
1777             /* Update vectorial force */
1778             fix2             = _mm256_add_pd(fix2,tx);
1779             fiy2             = _mm256_add_pd(fiy2,ty);
1780             fiz2             = _mm256_add_pd(fiz2,tz);
1781
1782             fjx1             = _mm256_add_pd(fjx1,tx);
1783             fjy1             = _mm256_add_pd(fjy1,ty);
1784             fjz1             = _mm256_add_pd(fjz1,tz);
1785
1786             /**************************
1787              * CALCULATE INTERACTIONS *
1788              **************************/
1789
1790             r22              = _mm256_mul_pd(rsq22,rinv22);
1791
1792             /* Calculate table index by multiplying r with table scale and truncate to integer */
1793             rt               = _mm256_mul_pd(r22,vftabscale);
1794             vfitab           = _mm256_cvttpd_epi32(rt);
1795             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1796             vfitab           = _mm_slli_epi32(vfitab,2);
1797
1798             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1799             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1800             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1801             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1802             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1803             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1804             Heps             = _mm256_mul_pd(vfeps,H);
1805             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1806             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1807             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1808
1809             fscal            = felec;
1810
1811             /* Calculate temporary vectorial force */
1812             tx               = _mm256_mul_pd(fscal,dx22);
1813             ty               = _mm256_mul_pd(fscal,dy22);
1814             tz               = _mm256_mul_pd(fscal,dz22);
1815
1816             /* Update vectorial force */
1817             fix2             = _mm256_add_pd(fix2,tx);
1818             fiy2             = _mm256_add_pd(fiy2,ty);
1819             fiz2             = _mm256_add_pd(fiz2,tz);
1820
1821             fjx2             = _mm256_add_pd(fjx2,tx);
1822             fjy2             = _mm256_add_pd(fjy2,ty);
1823             fjz2             = _mm256_add_pd(fjz2,tz);
1824
1825             /**************************
1826              * CALCULATE INTERACTIONS *
1827              **************************/
1828
1829             r23              = _mm256_mul_pd(rsq23,rinv23);
1830
1831             /* Calculate table index by multiplying r with table scale and truncate to integer */
1832             rt               = _mm256_mul_pd(r23,vftabscale);
1833             vfitab           = _mm256_cvttpd_epi32(rt);
1834             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1835             vfitab           = _mm_slli_epi32(vfitab,2);
1836
1837             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1838             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1839             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1840             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1841             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1842             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1843             Heps             = _mm256_mul_pd(vfeps,H);
1844             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1845             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1846             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
1847
1848             fscal            = felec;
1849
1850             /* Calculate temporary vectorial force */
1851             tx               = _mm256_mul_pd(fscal,dx23);
1852             ty               = _mm256_mul_pd(fscal,dy23);
1853             tz               = _mm256_mul_pd(fscal,dz23);
1854
1855             /* Update vectorial force */
1856             fix2             = _mm256_add_pd(fix2,tx);
1857             fiy2             = _mm256_add_pd(fiy2,ty);
1858             fiz2             = _mm256_add_pd(fiz2,tz);
1859
1860             fjx3             = _mm256_add_pd(fjx3,tx);
1861             fjy3             = _mm256_add_pd(fjy3,ty);
1862             fjz3             = _mm256_add_pd(fjz3,tz);
1863
1864             /**************************
1865              * CALCULATE INTERACTIONS *
1866              **************************/
1867
1868             r31              = _mm256_mul_pd(rsq31,rinv31);
1869
1870             /* Calculate table index by multiplying r with table scale and truncate to integer */
1871             rt               = _mm256_mul_pd(r31,vftabscale);
1872             vfitab           = _mm256_cvttpd_epi32(rt);
1873             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1874             vfitab           = _mm_slli_epi32(vfitab,2);
1875
1876             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1877             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1878             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1879             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1880             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1881             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1882             Heps             = _mm256_mul_pd(vfeps,H);
1883             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1884             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1885             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
1886
1887             fscal            = felec;
1888
1889             /* Calculate temporary vectorial force */
1890             tx               = _mm256_mul_pd(fscal,dx31);
1891             ty               = _mm256_mul_pd(fscal,dy31);
1892             tz               = _mm256_mul_pd(fscal,dz31);
1893
1894             /* Update vectorial force */
1895             fix3             = _mm256_add_pd(fix3,tx);
1896             fiy3             = _mm256_add_pd(fiy3,ty);
1897             fiz3             = _mm256_add_pd(fiz3,tz);
1898
1899             fjx1             = _mm256_add_pd(fjx1,tx);
1900             fjy1             = _mm256_add_pd(fjy1,ty);
1901             fjz1             = _mm256_add_pd(fjz1,tz);
1902
1903             /**************************
1904              * CALCULATE INTERACTIONS *
1905              **************************/
1906
1907             r32              = _mm256_mul_pd(rsq32,rinv32);
1908
1909             /* Calculate table index by multiplying r with table scale and truncate to integer */
1910             rt               = _mm256_mul_pd(r32,vftabscale);
1911             vfitab           = _mm256_cvttpd_epi32(rt);
1912             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1913             vfitab           = _mm_slli_epi32(vfitab,2);
1914
1915             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1916             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1917             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1918             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1919             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1920             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1921             Heps             = _mm256_mul_pd(vfeps,H);
1922             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1923             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1924             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
1925
1926             fscal            = felec;
1927
1928             /* Calculate temporary vectorial force */
1929             tx               = _mm256_mul_pd(fscal,dx32);
1930             ty               = _mm256_mul_pd(fscal,dy32);
1931             tz               = _mm256_mul_pd(fscal,dz32);
1932
1933             /* Update vectorial force */
1934             fix3             = _mm256_add_pd(fix3,tx);
1935             fiy3             = _mm256_add_pd(fiy3,ty);
1936             fiz3             = _mm256_add_pd(fiz3,tz);
1937
1938             fjx2             = _mm256_add_pd(fjx2,tx);
1939             fjy2             = _mm256_add_pd(fjy2,ty);
1940             fjz2             = _mm256_add_pd(fjz2,tz);
1941
1942             /**************************
1943              * CALCULATE INTERACTIONS *
1944              **************************/
1945
1946             r33              = _mm256_mul_pd(rsq33,rinv33);
1947
1948             /* Calculate table index by multiplying r with table scale and truncate to integer */
1949             rt               = _mm256_mul_pd(r33,vftabscale);
1950             vfitab           = _mm256_cvttpd_epi32(rt);
1951             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1952             vfitab           = _mm_slli_epi32(vfitab,2);
1953
1954             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1955             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1956             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1957             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1958             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1959             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1960             Heps             = _mm256_mul_pd(vfeps,H);
1961             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1962             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1963             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
1964
1965             fscal            = felec;
1966
1967             /* Calculate temporary vectorial force */
1968             tx               = _mm256_mul_pd(fscal,dx33);
1969             ty               = _mm256_mul_pd(fscal,dy33);
1970             tz               = _mm256_mul_pd(fscal,dz33);
1971
1972             /* Update vectorial force */
1973             fix3             = _mm256_add_pd(fix3,tx);
1974             fiy3             = _mm256_add_pd(fiy3,ty);
1975             fiz3             = _mm256_add_pd(fiz3,tz);
1976
1977             fjx3             = _mm256_add_pd(fjx3,tx);
1978             fjy3             = _mm256_add_pd(fjy3,ty);
1979             fjz3             = _mm256_add_pd(fjz3,tz);
1980
1981             fjptrA             = f+j_coord_offsetA;
1982             fjptrB             = f+j_coord_offsetB;
1983             fjptrC             = f+j_coord_offsetC;
1984             fjptrD             = f+j_coord_offsetD;
1985
1986             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1987                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1988                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1989
1990             /* Inner loop uses 381 flops */
1991         }
1992
1993         if(jidx<j_index_end)
1994         {
1995
1996             /* Get j neighbor index, and coordinate index */
1997             jnrlistA         = jjnr[jidx];
1998             jnrlistB         = jjnr[jidx+1];
1999             jnrlistC         = jjnr[jidx+2];
2000             jnrlistD         = jjnr[jidx+3];
2001             /* Sign of each element will be negative for non-real atoms.
2002              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2003              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
2004              */
2005             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2006
2007             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
2008             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
2009             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
2010
2011             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2012             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2013             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2014             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2015             j_coord_offsetA  = DIM*jnrA;
2016             j_coord_offsetB  = DIM*jnrB;
2017             j_coord_offsetC  = DIM*jnrC;
2018             j_coord_offsetD  = DIM*jnrD;
2019
2020             /* load j atom coordinates */
2021             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
2022                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2023                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2024                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2025
2026             /* Calculate displacement vector */
2027             dx00             = _mm256_sub_pd(ix0,jx0);
2028             dy00             = _mm256_sub_pd(iy0,jy0);
2029             dz00             = _mm256_sub_pd(iz0,jz0);
2030             dx11             = _mm256_sub_pd(ix1,jx1);
2031             dy11             = _mm256_sub_pd(iy1,jy1);
2032             dz11             = _mm256_sub_pd(iz1,jz1);
2033             dx12             = _mm256_sub_pd(ix1,jx2);
2034             dy12             = _mm256_sub_pd(iy1,jy2);
2035             dz12             = _mm256_sub_pd(iz1,jz2);
2036             dx13             = _mm256_sub_pd(ix1,jx3);
2037             dy13             = _mm256_sub_pd(iy1,jy3);
2038             dz13             = _mm256_sub_pd(iz1,jz3);
2039             dx21             = _mm256_sub_pd(ix2,jx1);
2040             dy21             = _mm256_sub_pd(iy2,jy1);
2041             dz21             = _mm256_sub_pd(iz2,jz1);
2042             dx22             = _mm256_sub_pd(ix2,jx2);
2043             dy22             = _mm256_sub_pd(iy2,jy2);
2044             dz22             = _mm256_sub_pd(iz2,jz2);
2045             dx23             = _mm256_sub_pd(ix2,jx3);
2046             dy23             = _mm256_sub_pd(iy2,jy3);
2047             dz23             = _mm256_sub_pd(iz2,jz3);
2048             dx31             = _mm256_sub_pd(ix3,jx1);
2049             dy31             = _mm256_sub_pd(iy3,jy1);
2050             dz31             = _mm256_sub_pd(iz3,jz1);
2051             dx32             = _mm256_sub_pd(ix3,jx2);
2052             dy32             = _mm256_sub_pd(iy3,jy2);
2053             dz32             = _mm256_sub_pd(iz3,jz2);
2054             dx33             = _mm256_sub_pd(ix3,jx3);
2055             dy33             = _mm256_sub_pd(iy3,jy3);
2056             dz33             = _mm256_sub_pd(iz3,jz3);
2057
2058             /* Calculate squared distance and things based on it */
2059             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
2060             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2061             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2062             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
2063             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2064             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2065             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
2066             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
2067             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
2068             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
2069
2070             rinv11           = avx256_invsqrt_d(rsq11);
2071             rinv12           = avx256_invsqrt_d(rsq12);
2072             rinv13           = avx256_invsqrt_d(rsq13);
2073             rinv21           = avx256_invsqrt_d(rsq21);
2074             rinv22           = avx256_invsqrt_d(rsq22);
2075             rinv23           = avx256_invsqrt_d(rsq23);
2076             rinv31           = avx256_invsqrt_d(rsq31);
2077             rinv32           = avx256_invsqrt_d(rsq32);
2078             rinv33           = avx256_invsqrt_d(rsq33);
2079
2080             rinvsq00         = avx256_inv_d(rsq00);
2081
2082             fjx0             = _mm256_setzero_pd();
2083             fjy0             = _mm256_setzero_pd();
2084             fjz0             = _mm256_setzero_pd();
2085             fjx1             = _mm256_setzero_pd();
2086             fjy1             = _mm256_setzero_pd();
2087             fjz1             = _mm256_setzero_pd();
2088             fjx2             = _mm256_setzero_pd();
2089             fjy2             = _mm256_setzero_pd();
2090             fjz2             = _mm256_setzero_pd();
2091             fjx3             = _mm256_setzero_pd();
2092             fjy3             = _mm256_setzero_pd();
2093             fjz3             = _mm256_setzero_pd();
2094
2095             /**************************
2096              * CALCULATE INTERACTIONS *
2097              **************************/
2098
2099             /* LENNARD-JONES DISPERSION/REPULSION */
2100
2101             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2102             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
2103
2104             fscal            = fvdw;
2105
2106             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2107
2108             /* Calculate temporary vectorial force */
2109             tx               = _mm256_mul_pd(fscal,dx00);
2110             ty               = _mm256_mul_pd(fscal,dy00);
2111             tz               = _mm256_mul_pd(fscal,dz00);
2112
2113             /* Update vectorial force */
2114             fix0             = _mm256_add_pd(fix0,tx);
2115             fiy0             = _mm256_add_pd(fiy0,ty);
2116             fiz0             = _mm256_add_pd(fiz0,tz);
2117
2118             fjx0             = _mm256_add_pd(fjx0,tx);
2119             fjy0             = _mm256_add_pd(fjy0,ty);
2120             fjz0             = _mm256_add_pd(fjz0,tz);
2121
2122             /**************************
2123              * CALCULATE INTERACTIONS *
2124              **************************/
2125
2126             r11              = _mm256_mul_pd(rsq11,rinv11);
2127             r11              = _mm256_andnot_pd(dummy_mask,r11);
2128
2129             /* Calculate table index by multiplying r with table scale and truncate to integer */
2130             rt               = _mm256_mul_pd(r11,vftabscale);
2131             vfitab           = _mm256_cvttpd_epi32(rt);
2132             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2133             vfitab           = _mm_slli_epi32(vfitab,2);
2134
2135             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2136             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2137             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2138             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2139             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2140             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2141             Heps             = _mm256_mul_pd(vfeps,H);
2142             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2143             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2144             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
2145
2146             fscal            = felec;
2147
2148             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2149
2150             /* Calculate temporary vectorial force */
2151             tx               = _mm256_mul_pd(fscal,dx11);
2152             ty               = _mm256_mul_pd(fscal,dy11);
2153             tz               = _mm256_mul_pd(fscal,dz11);
2154
2155             /* Update vectorial force */
2156             fix1             = _mm256_add_pd(fix1,tx);
2157             fiy1             = _mm256_add_pd(fiy1,ty);
2158             fiz1             = _mm256_add_pd(fiz1,tz);
2159
2160             fjx1             = _mm256_add_pd(fjx1,tx);
2161             fjy1             = _mm256_add_pd(fjy1,ty);
2162             fjz1             = _mm256_add_pd(fjz1,tz);
2163
2164             /**************************
2165              * CALCULATE INTERACTIONS *
2166              **************************/
2167
2168             r12              = _mm256_mul_pd(rsq12,rinv12);
2169             r12              = _mm256_andnot_pd(dummy_mask,r12);
2170
2171             /* Calculate table index by multiplying r with table scale and truncate to integer */
2172             rt               = _mm256_mul_pd(r12,vftabscale);
2173             vfitab           = _mm256_cvttpd_epi32(rt);
2174             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2175             vfitab           = _mm_slli_epi32(vfitab,2);
2176
2177             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2178             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2179             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2180             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2181             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2182             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2183             Heps             = _mm256_mul_pd(vfeps,H);
2184             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2185             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2186             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
2187
2188             fscal            = felec;
2189
2190             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2191
2192             /* Calculate temporary vectorial force */
2193             tx               = _mm256_mul_pd(fscal,dx12);
2194             ty               = _mm256_mul_pd(fscal,dy12);
2195             tz               = _mm256_mul_pd(fscal,dz12);
2196
2197             /* Update vectorial force */
2198             fix1             = _mm256_add_pd(fix1,tx);
2199             fiy1             = _mm256_add_pd(fiy1,ty);
2200             fiz1             = _mm256_add_pd(fiz1,tz);
2201
2202             fjx2             = _mm256_add_pd(fjx2,tx);
2203             fjy2             = _mm256_add_pd(fjy2,ty);
2204             fjz2             = _mm256_add_pd(fjz2,tz);
2205
2206             /**************************
2207              * CALCULATE INTERACTIONS *
2208              **************************/
2209
2210             r13              = _mm256_mul_pd(rsq13,rinv13);
2211             r13              = _mm256_andnot_pd(dummy_mask,r13);
2212
2213             /* Calculate table index by multiplying r with table scale and truncate to integer */
2214             rt               = _mm256_mul_pd(r13,vftabscale);
2215             vfitab           = _mm256_cvttpd_epi32(rt);
2216             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2217             vfitab           = _mm_slli_epi32(vfitab,2);
2218
2219             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2220             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2221             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2222             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2223             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2224             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2225             Heps             = _mm256_mul_pd(vfeps,H);
2226             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2227             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2228             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq13,FF),_mm256_mul_pd(vftabscale,rinv13)));
2229
2230             fscal            = felec;
2231
2232             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2233
2234             /* Calculate temporary vectorial force */
2235             tx               = _mm256_mul_pd(fscal,dx13);
2236             ty               = _mm256_mul_pd(fscal,dy13);
2237             tz               = _mm256_mul_pd(fscal,dz13);
2238
2239             /* Update vectorial force */
2240             fix1             = _mm256_add_pd(fix1,tx);
2241             fiy1             = _mm256_add_pd(fiy1,ty);
2242             fiz1             = _mm256_add_pd(fiz1,tz);
2243
2244             fjx3             = _mm256_add_pd(fjx3,tx);
2245             fjy3             = _mm256_add_pd(fjy3,ty);
2246             fjz3             = _mm256_add_pd(fjz3,tz);
2247
2248             /**************************
2249              * CALCULATE INTERACTIONS *
2250              **************************/
2251
2252             r21              = _mm256_mul_pd(rsq21,rinv21);
2253             r21              = _mm256_andnot_pd(dummy_mask,r21);
2254
2255             /* Calculate table index by multiplying r with table scale and truncate to integer */
2256             rt               = _mm256_mul_pd(r21,vftabscale);
2257             vfitab           = _mm256_cvttpd_epi32(rt);
2258             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2259             vfitab           = _mm_slli_epi32(vfitab,2);
2260
2261             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2262             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2263             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2264             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2265             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2266             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2267             Heps             = _mm256_mul_pd(vfeps,H);
2268             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2269             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2270             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
2271
2272             fscal            = felec;
2273
2274             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2275
2276             /* Calculate temporary vectorial force */
2277             tx               = _mm256_mul_pd(fscal,dx21);
2278             ty               = _mm256_mul_pd(fscal,dy21);
2279             tz               = _mm256_mul_pd(fscal,dz21);
2280
2281             /* Update vectorial force */
2282             fix2             = _mm256_add_pd(fix2,tx);
2283             fiy2             = _mm256_add_pd(fiy2,ty);
2284             fiz2             = _mm256_add_pd(fiz2,tz);
2285
2286             fjx1             = _mm256_add_pd(fjx1,tx);
2287             fjy1             = _mm256_add_pd(fjy1,ty);
2288             fjz1             = _mm256_add_pd(fjz1,tz);
2289
2290             /**************************
2291              * CALCULATE INTERACTIONS *
2292              **************************/
2293
2294             r22              = _mm256_mul_pd(rsq22,rinv22);
2295             r22              = _mm256_andnot_pd(dummy_mask,r22);
2296
2297             /* Calculate table index by multiplying r with table scale and truncate to integer */
2298             rt               = _mm256_mul_pd(r22,vftabscale);
2299             vfitab           = _mm256_cvttpd_epi32(rt);
2300             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2301             vfitab           = _mm_slli_epi32(vfitab,2);
2302
2303             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2304             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2305             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2306             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2307             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2308             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2309             Heps             = _mm256_mul_pd(vfeps,H);
2310             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2311             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2312             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
2313
2314             fscal            = felec;
2315
2316             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2317
2318             /* Calculate temporary vectorial force */
2319             tx               = _mm256_mul_pd(fscal,dx22);
2320             ty               = _mm256_mul_pd(fscal,dy22);
2321             tz               = _mm256_mul_pd(fscal,dz22);
2322
2323             /* Update vectorial force */
2324             fix2             = _mm256_add_pd(fix2,tx);
2325             fiy2             = _mm256_add_pd(fiy2,ty);
2326             fiz2             = _mm256_add_pd(fiz2,tz);
2327
2328             fjx2             = _mm256_add_pd(fjx2,tx);
2329             fjy2             = _mm256_add_pd(fjy2,ty);
2330             fjz2             = _mm256_add_pd(fjz2,tz);
2331
2332             /**************************
2333              * CALCULATE INTERACTIONS *
2334              **************************/
2335
2336             r23              = _mm256_mul_pd(rsq23,rinv23);
2337             r23              = _mm256_andnot_pd(dummy_mask,r23);
2338
2339             /* Calculate table index by multiplying r with table scale and truncate to integer */
2340             rt               = _mm256_mul_pd(r23,vftabscale);
2341             vfitab           = _mm256_cvttpd_epi32(rt);
2342             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2343             vfitab           = _mm_slli_epi32(vfitab,2);
2344
2345             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2346             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2347             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2348             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2349             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2350             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2351             Heps             = _mm256_mul_pd(vfeps,H);
2352             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2353             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2354             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq23,FF),_mm256_mul_pd(vftabscale,rinv23)));
2355
2356             fscal            = felec;
2357
2358             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2359
2360             /* Calculate temporary vectorial force */
2361             tx               = _mm256_mul_pd(fscal,dx23);
2362             ty               = _mm256_mul_pd(fscal,dy23);
2363             tz               = _mm256_mul_pd(fscal,dz23);
2364
2365             /* Update vectorial force */
2366             fix2             = _mm256_add_pd(fix2,tx);
2367             fiy2             = _mm256_add_pd(fiy2,ty);
2368             fiz2             = _mm256_add_pd(fiz2,tz);
2369
2370             fjx3             = _mm256_add_pd(fjx3,tx);
2371             fjy3             = _mm256_add_pd(fjy3,ty);
2372             fjz3             = _mm256_add_pd(fjz3,tz);
2373
2374             /**************************
2375              * CALCULATE INTERACTIONS *
2376              **************************/
2377
2378             r31              = _mm256_mul_pd(rsq31,rinv31);
2379             r31              = _mm256_andnot_pd(dummy_mask,r31);
2380
2381             /* Calculate table index by multiplying r with table scale and truncate to integer */
2382             rt               = _mm256_mul_pd(r31,vftabscale);
2383             vfitab           = _mm256_cvttpd_epi32(rt);
2384             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2385             vfitab           = _mm_slli_epi32(vfitab,2);
2386
2387             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2388             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2389             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2390             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2391             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2392             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2393             Heps             = _mm256_mul_pd(vfeps,H);
2394             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2395             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2396             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq31,FF),_mm256_mul_pd(vftabscale,rinv31)));
2397
2398             fscal            = felec;
2399
2400             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2401
2402             /* Calculate temporary vectorial force */
2403             tx               = _mm256_mul_pd(fscal,dx31);
2404             ty               = _mm256_mul_pd(fscal,dy31);
2405             tz               = _mm256_mul_pd(fscal,dz31);
2406
2407             /* Update vectorial force */
2408             fix3             = _mm256_add_pd(fix3,tx);
2409             fiy3             = _mm256_add_pd(fiy3,ty);
2410             fiz3             = _mm256_add_pd(fiz3,tz);
2411
2412             fjx1             = _mm256_add_pd(fjx1,tx);
2413             fjy1             = _mm256_add_pd(fjy1,ty);
2414             fjz1             = _mm256_add_pd(fjz1,tz);
2415
2416             /**************************
2417              * CALCULATE INTERACTIONS *
2418              **************************/
2419
2420             r32              = _mm256_mul_pd(rsq32,rinv32);
2421             r32              = _mm256_andnot_pd(dummy_mask,r32);
2422
2423             /* Calculate table index by multiplying r with table scale and truncate to integer */
2424             rt               = _mm256_mul_pd(r32,vftabscale);
2425             vfitab           = _mm256_cvttpd_epi32(rt);
2426             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2427             vfitab           = _mm_slli_epi32(vfitab,2);
2428
2429             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2430             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2431             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2432             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2433             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2434             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2435             Heps             = _mm256_mul_pd(vfeps,H);
2436             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2437             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2438             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq32,FF),_mm256_mul_pd(vftabscale,rinv32)));
2439
2440             fscal            = felec;
2441
2442             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2443
2444             /* Calculate temporary vectorial force */
2445             tx               = _mm256_mul_pd(fscal,dx32);
2446             ty               = _mm256_mul_pd(fscal,dy32);
2447             tz               = _mm256_mul_pd(fscal,dz32);
2448
2449             /* Update vectorial force */
2450             fix3             = _mm256_add_pd(fix3,tx);
2451             fiy3             = _mm256_add_pd(fiy3,ty);
2452             fiz3             = _mm256_add_pd(fiz3,tz);
2453
2454             fjx2             = _mm256_add_pd(fjx2,tx);
2455             fjy2             = _mm256_add_pd(fjy2,ty);
2456             fjz2             = _mm256_add_pd(fjz2,tz);
2457
2458             /**************************
2459              * CALCULATE INTERACTIONS *
2460              **************************/
2461
2462             r33              = _mm256_mul_pd(rsq33,rinv33);
2463             r33              = _mm256_andnot_pd(dummy_mask,r33);
2464
2465             /* Calculate table index by multiplying r with table scale and truncate to integer */
2466             rt               = _mm256_mul_pd(r33,vftabscale);
2467             vfitab           = _mm256_cvttpd_epi32(rt);
2468             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2469             vfitab           = _mm_slli_epi32(vfitab,2);
2470
2471             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2472             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2473             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2474             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2475             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2476             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2477             Heps             = _mm256_mul_pd(vfeps,H);
2478             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2479             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2480             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq33,FF),_mm256_mul_pd(vftabscale,rinv33)));
2481
2482             fscal            = felec;
2483
2484             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2485
2486             /* Calculate temporary vectorial force */
2487             tx               = _mm256_mul_pd(fscal,dx33);
2488             ty               = _mm256_mul_pd(fscal,dy33);
2489             tz               = _mm256_mul_pd(fscal,dz33);
2490
2491             /* Update vectorial force */
2492             fix3             = _mm256_add_pd(fix3,tx);
2493             fiy3             = _mm256_add_pd(fiy3,ty);
2494             fiz3             = _mm256_add_pd(fiz3,tz);
2495
2496             fjx3             = _mm256_add_pd(fjx3,tx);
2497             fjy3             = _mm256_add_pd(fjy3,ty);
2498             fjz3             = _mm256_add_pd(fjz3,tz);
2499
2500             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2501             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2502             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2503             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2504
2505             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2506                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2507                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2508
2509             /* Inner loop uses 390 flops */
2510         }
2511
2512         /* End of innermost loop */
2513
2514         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2515                                                  f+i_coord_offset,fshift+i_shift_offset);
2516
2517         /* Increment number of inner iterations */
2518         inneriter                  += j_index_end - j_index_start;
2519
2520         /* Outer loop uses 24 flops */
2521     }
2522
2523     /* Increment number of outer iterations */
2524     outeriter        += nri;
2525
2526     /* Update outer/inner flops */
2527
2528     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*390);
2529 }