src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwCSTab_GeomW4W4_avx_256_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_double.h"
  34 #include "kernelutil_x86_avx_256_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4W4_VF_avx_256_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Water4
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW4W4_VF_avx_256_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  63     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  64     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  65     real             rcutoff_scalar;
  66     real             *shiftvec,*fshift,*x,*f;
  67     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  68     real             scratch[4*DIM];
  69     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  70     real *           vdwioffsetptr0;
  71     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  72     real *           vdwioffsetptr1;
  73     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  74     real *           vdwioffsetptr2;
  75     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  76     real *           vdwioffsetptr3;
  77     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  79     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  81     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  82     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  83     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  84     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  85     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  86     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  88     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  89     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
  90     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  91     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  92     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
  93     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
  94     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
  95     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
  96     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 103     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128i          ewitab;
 109     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 111     real             *ewtab;
 112     __m256d          dummy_mask,cutoff_mask;
 113     __m128           tmpmask0,tmpmask1;
 114     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 115     __m256d          one     = _mm256_set1_pd(1.0);
 116     __m256d          two     = _mm256_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     vftab            = kernel_data->table_vdw->data;
 135     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 136
 137     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 138     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 139     beta2            = _mm256_mul_pd(beta,beta);
 140     beta3            = _mm256_mul_pd(beta,beta2);
 141
 142     ewtab            = fr->ic->tabq_coul_FDV0;
 143     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 144     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_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             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 284             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 285             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 286             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 287             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 288             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 289             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 290             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 291             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 292             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 293
 294             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 295             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 296             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 297             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 298             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 299             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 300             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 301             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 302             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 303
 304             fjx0             = _mm256_setzero_pd();
 305             fjy0             = _mm256_setzero_pd();
 306             fjz0             = _mm256_setzero_pd();
 307             fjx1             = _mm256_setzero_pd();
 308             fjy1             = _mm256_setzero_pd();
 309             fjz1             = _mm256_setzero_pd();
 310             fjx2             = _mm256_setzero_pd();
 311             fjy2             = _mm256_setzero_pd();
 312             fjz2             = _mm256_setzero_pd();
 313             fjx3             = _mm256_setzero_pd();
 314             fjy3             = _mm256_setzero_pd();
 315             fjz3             = _mm256_setzero_pd();
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             r00              = _mm256_mul_pd(rsq00,rinv00);
 322
 323             /* Calculate table index by multiplying r with table scale and truncate to integer */
 324             rt               = _mm256_mul_pd(r00,vftabscale);
 325             vfitab           = _mm256_cvttpd_epi32(rt);
 326             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 327             vfitab           = _mm_slli_epi32(vfitab,3);
 328
 329             /* CUBIC SPLINE TABLE DISPERSION */
 330             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 331             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 332             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 333             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 334             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 335             Heps             = _mm256_mul_pd(vfeps,H);
 336             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 337             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 338             vvdw6            = _mm256_mul_pd(c6_00,VV);
 339             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 340             fvdw6            = _mm256_mul_pd(c6_00,FF);
 341
 342             /* CUBIC SPLINE TABLE REPULSION */
 343             vfitab           = _mm_add_epi32(vfitab,ifour);
 344             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 345             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 346             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 347             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 348             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 349             Heps             = _mm256_mul_pd(vfeps,H);
 350             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 351             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 352             vvdw12           = _mm256_mul_pd(c12_00,VV);
 353             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 354             fvdw12           = _mm256_mul_pd(c12_00,FF);
 355             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 356             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 360
 361             fscal            = fvdw;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm256_mul_pd(fscal,dx00);
 365             ty               = _mm256_mul_pd(fscal,dy00);
 366             tz               = _mm256_mul_pd(fscal,dz00);
 367
 368             /* Update vectorial force */
 369             fix0             = _mm256_add_pd(fix0,tx);
 370             fiy0             = _mm256_add_pd(fiy0,ty);
 371             fiz0             = _mm256_add_pd(fiz0,tz);
 372
 373             fjx0             = _mm256_add_pd(fjx0,tx);
 374             fjy0             = _mm256_add_pd(fjy0,ty);
 375             fjz0             = _mm256_add_pd(fjz0,tz);
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             r11              = _mm256_mul_pd(rsq11,rinv11);
 382
 383             /* EWALD ELECTROSTATICS */
 384
 385             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 386             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 387             ewitab           = _mm256_cvttpd_epi32(ewrt);
 388             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 389             ewitab           = _mm_slli_epi32(ewitab,2);
 390             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 391             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 392             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 393             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 394             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 395             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 396             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 397             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
 398             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velecsum         = _mm256_add_pd(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405             /* Calculate temporary vectorial force */
 406             tx               = _mm256_mul_pd(fscal,dx11);
 407             ty               = _mm256_mul_pd(fscal,dy11);
 408             tz               = _mm256_mul_pd(fscal,dz11);
 409
 410             /* Update vectorial force */
 411             fix1             = _mm256_add_pd(fix1,tx);
 412             fiy1             = _mm256_add_pd(fiy1,ty);
 413             fiz1             = _mm256_add_pd(fiz1,tz);
 414
 415             fjx1             = _mm256_add_pd(fjx1,tx);
 416             fjy1             = _mm256_add_pd(fjy1,ty);
 417             fjz1             = _mm256_add_pd(fjz1,tz);
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             r12              = _mm256_mul_pd(rsq12,rinv12);
 424
 425             /* EWALD ELECTROSTATICS */
 426
 427             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 428             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 429             ewitab           = _mm256_cvttpd_epi32(ewrt);
 430             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 431             ewitab           = _mm_slli_epi32(ewitab,2);
 432             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 433             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 434             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 435             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 436             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 437             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 438             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 439             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
 440             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
 441
 442             /* Update potential sum for this i atom from the interaction with this j atom. */
 443             velecsum         = _mm256_add_pd(velecsum,velec);
 444
 445             fscal            = felec;
 446
 447             /* Calculate temporary vectorial force */
 448             tx               = _mm256_mul_pd(fscal,dx12);
 449             ty               = _mm256_mul_pd(fscal,dy12);
 450             tz               = _mm256_mul_pd(fscal,dz12);
 451
 452             /* Update vectorial force */
 453             fix1             = _mm256_add_pd(fix1,tx);
 454             fiy1             = _mm256_add_pd(fiy1,ty);
 455             fiz1             = _mm256_add_pd(fiz1,tz);
 456
 457             fjx2             = _mm256_add_pd(fjx2,tx);
 458             fjy2             = _mm256_add_pd(fjy2,ty);
 459             fjz2             = _mm256_add_pd(fjz2,tz);
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             r13              = _mm256_mul_pd(rsq13,rinv13);
 466
 467             /* EWALD ELECTROSTATICS */
 468
 469             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 470             ewrt             = _mm256_mul_pd(r13,ewtabscale);
 471             ewitab           = _mm256_cvttpd_epi32(ewrt);
 472             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 473             ewitab           = _mm_slli_epi32(ewitab,2);
 474             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 475             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 476             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 477             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 478             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 479             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 480             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 481             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(rinv13,velec));
 482             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
 483
 484             /* Update potential sum for this i atom from the interaction with this j atom. */
 485             velecsum         = _mm256_add_pd(velecsum,velec);
 486
 487             fscal            = felec;
 488
 489             /* Calculate temporary vectorial force */
 490             tx               = _mm256_mul_pd(fscal,dx13);
 491             ty               = _mm256_mul_pd(fscal,dy13);
 492             tz               = _mm256_mul_pd(fscal,dz13);
 493
 494             /* Update vectorial force */
 495             fix1             = _mm256_add_pd(fix1,tx);
 496             fiy1             = _mm256_add_pd(fiy1,ty);
 497             fiz1             = _mm256_add_pd(fiz1,tz);
 498
 499             fjx3             = _mm256_add_pd(fjx3,tx);
 500             fjy3             = _mm256_add_pd(fjy3,ty);
 501             fjz3             = _mm256_add_pd(fjz3,tz);
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             r21              = _mm256_mul_pd(rsq21,rinv21);
 508
 509             /* EWALD ELECTROSTATICS */
 510
 511             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 512             ewrt             = _mm256_mul_pd(r21,ewtabscale);
 513             ewitab           = _mm256_cvttpd_epi32(ewrt);
 514             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 515             ewitab           = _mm_slli_epi32(ewitab,2);
 516             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 517             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 518             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 519             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 520             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 521             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 522             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 523             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
 524             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
 525
 526             /* Update potential sum for this i atom from the interaction with this j atom. */
 527             velecsum         = _mm256_add_pd(velecsum,velec);
 528
 529             fscal            = felec;
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm256_mul_pd(fscal,dx21);
 533             ty               = _mm256_mul_pd(fscal,dy21);
 534             tz               = _mm256_mul_pd(fscal,dz21);
 535
 536             /* Update vectorial force */
 537             fix2             = _mm256_add_pd(fix2,tx);
 538             fiy2             = _mm256_add_pd(fiy2,ty);
 539             fiz2             = _mm256_add_pd(fiz2,tz);
 540
 541             fjx1             = _mm256_add_pd(fjx1,tx);
 542             fjy1             = _mm256_add_pd(fjy1,ty);
 543             fjz1             = _mm256_add_pd(fjz1,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             r22              = _mm256_mul_pd(rsq22,rinv22);
 550
 551             /* EWALD ELECTROSTATICS */
 552
 553             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 554             ewrt             = _mm256_mul_pd(r22,ewtabscale);
 555             ewitab           = _mm256_cvttpd_epi32(ewrt);
 556             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 557             ewitab           = _mm_slli_epi32(ewitab,2);
 558             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 559             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 560             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 561             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 562             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 563             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 564             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 565             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
 566             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velecsum         = _mm256_add_pd(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm256_mul_pd(fscal,dx22);
 575             ty               = _mm256_mul_pd(fscal,dy22);
 576             tz               = _mm256_mul_pd(fscal,dz22);
 577
 578             /* Update vectorial force */
 579             fix2             = _mm256_add_pd(fix2,tx);
 580             fiy2             = _mm256_add_pd(fiy2,ty);
 581             fiz2             = _mm256_add_pd(fiz2,tz);
 582
 583             fjx2             = _mm256_add_pd(fjx2,tx);
 584             fjy2             = _mm256_add_pd(fjy2,ty);
 585             fjz2             = _mm256_add_pd(fjz2,tz);
 586
 587             /**************************
 588              * CALCULATE INTERACTIONS *
 589              **************************/
 590
 591             r23              = _mm256_mul_pd(rsq23,rinv23);
 592
 593             /* EWALD ELECTROSTATICS */
 594
 595             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 596             ewrt             = _mm256_mul_pd(r23,ewtabscale);
 597             ewitab           = _mm256_cvttpd_epi32(ewrt);
 598             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 599             ewitab           = _mm_slli_epi32(ewitab,2);
 600             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 601             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 602             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 603             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 604             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 605             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 606             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 607             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(rinv23,velec));
 608             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
 609
 610             /* Update potential sum for this i atom from the interaction with this j atom. */
 611             velecsum         = _mm256_add_pd(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm256_mul_pd(fscal,dx23);
 617             ty               = _mm256_mul_pd(fscal,dy23);
 618             tz               = _mm256_mul_pd(fscal,dz23);
 619
 620             /* Update vectorial force */
 621             fix2             = _mm256_add_pd(fix2,tx);
 622             fiy2             = _mm256_add_pd(fiy2,ty);
 623             fiz2             = _mm256_add_pd(fiz2,tz);
 624
 625             fjx3             = _mm256_add_pd(fjx3,tx);
 626             fjy3             = _mm256_add_pd(fjy3,ty);
 627             fjz3             = _mm256_add_pd(fjz3,tz);
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             r31              = _mm256_mul_pd(rsq31,rinv31);
 634
 635             /* EWALD ELECTROSTATICS */
 636
 637             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 638             ewrt             = _mm256_mul_pd(r31,ewtabscale);
 639             ewitab           = _mm256_cvttpd_epi32(ewrt);
 640             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 641             ewitab           = _mm_slli_epi32(ewitab,2);
 642             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 643             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 644             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 645             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 646             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 647             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 648             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 649             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(rinv31,velec));
 650             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
 651
 652             /* Update potential sum for this i atom from the interaction with this j atom. */
 653             velecsum         = _mm256_add_pd(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             /* Calculate temporary vectorial force */
 658             tx               = _mm256_mul_pd(fscal,dx31);
 659             ty               = _mm256_mul_pd(fscal,dy31);
 660             tz               = _mm256_mul_pd(fscal,dz31);
 661
 662             /* Update vectorial force */
 663             fix3             = _mm256_add_pd(fix3,tx);
 664             fiy3             = _mm256_add_pd(fiy3,ty);
 665             fiz3             = _mm256_add_pd(fiz3,tz);
 666
 667             fjx1             = _mm256_add_pd(fjx1,tx);
 668             fjy1             = _mm256_add_pd(fjy1,ty);
 669             fjz1             = _mm256_add_pd(fjz1,tz);
 670
 671             /**************************
 672              * CALCULATE INTERACTIONS *
 673              **************************/
 674
 675             r32              = _mm256_mul_pd(rsq32,rinv32);
 676
 677             /* EWALD ELECTROSTATICS */
 678
 679             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 680             ewrt             = _mm256_mul_pd(r32,ewtabscale);
 681             ewitab           = _mm256_cvttpd_epi32(ewrt);
 682             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 683             ewitab           = _mm_slli_epi32(ewitab,2);
 684             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 685             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 686             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 687             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 688             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 689             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 690             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 691             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(rinv32,velec));
 692             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
 693
 694             /* Update potential sum for this i atom from the interaction with this j atom. */
 695             velecsum         = _mm256_add_pd(velecsum,velec);
 696
 697             fscal            = felec;
 698
 699             /* Calculate temporary vectorial force */
 700             tx               = _mm256_mul_pd(fscal,dx32);
 701             ty               = _mm256_mul_pd(fscal,dy32);
 702             tz               = _mm256_mul_pd(fscal,dz32);
 703
 704             /* Update vectorial force */
 705             fix3             = _mm256_add_pd(fix3,tx);
 706             fiy3             = _mm256_add_pd(fiy3,ty);
 707             fiz3             = _mm256_add_pd(fiz3,tz);
 708
 709             fjx2             = _mm256_add_pd(fjx2,tx);
 710             fjy2             = _mm256_add_pd(fjy2,ty);
 711             fjz2             = _mm256_add_pd(fjz2,tz);
 712
 713             /**************************
 714              * CALCULATE INTERACTIONS *
 715              **************************/
 716
 717             r33              = _mm256_mul_pd(rsq33,rinv33);
 718
 719             /* EWALD ELECTROSTATICS */
 720
 721             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 722             ewrt             = _mm256_mul_pd(r33,ewtabscale);
 723             ewitab           = _mm256_cvttpd_epi32(ewrt);
 724             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 725             ewitab           = _mm_slli_epi32(ewitab,2);
 726             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 727             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 728             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 729             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 730             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 731             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 732             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 733             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(rinv33,velec));
 734             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
 735
 736             /* Update potential sum for this i atom from the interaction with this j atom. */
 737             velecsum         = _mm256_add_pd(velecsum,velec);
 738
 739             fscal            = felec;
 740
 741             /* Calculate temporary vectorial force */
 742             tx               = _mm256_mul_pd(fscal,dx33);
 743             ty               = _mm256_mul_pd(fscal,dy33);
 744             tz               = _mm256_mul_pd(fscal,dz33);
 745
 746             /* Update vectorial force */
 747             fix3             = _mm256_add_pd(fix3,tx);
 748             fiy3             = _mm256_add_pd(fiy3,ty);
 749             fiz3             = _mm256_add_pd(fiz3,tz);
 750
 751             fjx3             = _mm256_add_pd(fjx3,tx);
 752             fjy3             = _mm256_add_pd(fjy3,ty);
 753             fjz3             = _mm256_add_pd(fjz3,tz);
 754
 755             fjptrA             = f+j_coord_offsetA;
 756             fjptrB             = f+j_coord_offsetB;
 757             fjptrC             = f+j_coord_offsetC;
 758             fjptrD             = f+j_coord_offsetD;
 759
 760             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 761                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 762                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 763
 764             /* Inner loop uses 428 flops */
 765         }
 766
 767         if(jidx<j_index_end)
 768         {
 769
 770             /* Get j neighbor index, and coordinate index */
 771             jnrlistA         = jjnr[jidx];
 772             jnrlistB         = jjnr[jidx+1];
 773             jnrlistC         = jjnr[jidx+2];
 774             jnrlistD         = jjnr[jidx+3];
 775             /* Sign of each element will be negative for non-real atoms.
 776              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 777              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 778              */
 779             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 780
 781             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 782             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 783             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 784
 785             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 786             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 787             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 788             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 789             j_coord_offsetA  = DIM*jnrA;
 790             j_coord_offsetB  = DIM*jnrB;
 791             j_coord_offsetC  = DIM*jnrC;
 792             j_coord_offsetD  = DIM*jnrD;
 793
 794             /* load j atom coordinates */
 795             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 796                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 797                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 798                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 799
 800             /* Calculate displacement vector */
 801             dx00             = _mm256_sub_pd(ix0,jx0);
 802             dy00             = _mm256_sub_pd(iy0,jy0);
 803             dz00             = _mm256_sub_pd(iz0,jz0);
 804             dx11             = _mm256_sub_pd(ix1,jx1);
 805             dy11             = _mm256_sub_pd(iy1,jy1);
 806             dz11             = _mm256_sub_pd(iz1,jz1);
 807             dx12             = _mm256_sub_pd(ix1,jx2);
 808             dy12             = _mm256_sub_pd(iy1,jy2);
 809             dz12             = _mm256_sub_pd(iz1,jz2);
 810             dx13             = _mm256_sub_pd(ix1,jx3);
 811             dy13             = _mm256_sub_pd(iy1,jy3);
 812             dz13             = _mm256_sub_pd(iz1,jz3);
 813             dx21             = _mm256_sub_pd(ix2,jx1);
 814             dy21             = _mm256_sub_pd(iy2,jy1);
 815             dz21             = _mm256_sub_pd(iz2,jz1);
 816             dx22             = _mm256_sub_pd(ix2,jx2);
 817             dy22             = _mm256_sub_pd(iy2,jy2);
 818             dz22             = _mm256_sub_pd(iz2,jz2);
 819             dx23             = _mm256_sub_pd(ix2,jx3);
 820             dy23             = _mm256_sub_pd(iy2,jy3);
 821             dz23             = _mm256_sub_pd(iz2,jz3);
 822             dx31             = _mm256_sub_pd(ix3,jx1);
 823             dy31             = _mm256_sub_pd(iy3,jy1);
 824             dz31             = _mm256_sub_pd(iz3,jz1);
 825             dx32             = _mm256_sub_pd(ix3,jx2);
 826             dy32             = _mm256_sub_pd(iy3,jy2);
 827             dz32             = _mm256_sub_pd(iz3,jz2);
 828             dx33             = _mm256_sub_pd(ix3,jx3);
 829             dy33             = _mm256_sub_pd(iy3,jy3);
 830             dz33             = _mm256_sub_pd(iz3,jz3);
 831
 832             /* Calculate squared distance and things based on it */
 833             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 834             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 835             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 836             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
 837             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 838             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 839             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
 840             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
 841             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
 842             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
 843
 844             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 845             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 846             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 847             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
 848             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 849             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 850             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
 851             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
 852             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
 853             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
 854
 855             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
 856             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
 857             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
 858             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
 859             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
 860             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
 861             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
 862             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
 863             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
 864
 865             fjx0             = _mm256_setzero_pd();
 866             fjy0             = _mm256_setzero_pd();
 867             fjz0             = _mm256_setzero_pd();
 868             fjx1             = _mm256_setzero_pd();
 869             fjy1             = _mm256_setzero_pd();
 870             fjz1             = _mm256_setzero_pd();
 871             fjx2             = _mm256_setzero_pd();
 872             fjy2             = _mm256_setzero_pd();
 873             fjz2             = _mm256_setzero_pd();
 874             fjx3             = _mm256_setzero_pd();
 875             fjy3             = _mm256_setzero_pd();
 876             fjz3             = _mm256_setzero_pd();
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             r00              = _mm256_mul_pd(rsq00,rinv00);
 883             r00              = _mm256_andnot_pd(dummy_mask,r00);
 884
 885             /* Calculate table index by multiplying r with table scale and truncate to integer */
 886             rt               = _mm256_mul_pd(r00,vftabscale);
 887             vfitab           = _mm256_cvttpd_epi32(rt);
 888             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 889             vfitab           = _mm_slli_epi32(vfitab,3);
 890
 891             /* CUBIC SPLINE TABLE DISPERSION */
 892             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 893             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 894             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 895             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 896             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 897             Heps             = _mm256_mul_pd(vfeps,H);
 898             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 899             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 900             vvdw6            = _mm256_mul_pd(c6_00,VV);
 901             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 902             fvdw6            = _mm256_mul_pd(c6_00,FF);
 903
 904             /* CUBIC SPLINE TABLE REPULSION */
 905             vfitab           = _mm_add_epi32(vfitab,ifour);
 906             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 907             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 908             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 909             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 910             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 911             Heps             = _mm256_mul_pd(vfeps,H);
 912             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 913             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 914             vvdw12           = _mm256_mul_pd(c12_00,VV);
 915             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 916             fvdw12           = _mm256_mul_pd(c12_00,FF);
 917             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 918             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 919
 920             /* Update potential sum for this i atom from the interaction with this j atom. */
 921             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 922             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 923
 924             fscal            = fvdw;
 925
 926             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 927
 928             /* Calculate temporary vectorial force */
 929             tx               = _mm256_mul_pd(fscal,dx00);
 930             ty               = _mm256_mul_pd(fscal,dy00);
 931             tz               = _mm256_mul_pd(fscal,dz00);
 932
 933             /* Update vectorial force */
 934             fix0             = _mm256_add_pd(fix0,tx);
 935             fiy0             = _mm256_add_pd(fiy0,ty);
 936             fiz0             = _mm256_add_pd(fiz0,tz);
 937
 938             fjx0             = _mm256_add_pd(fjx0,tx);
 939             fjy0             = _mm256_add_pd(fjy0,ty);
 940             fjz0             = _mm256_add_pd(fjz0,tz);
 941
 942             /**************************
 943              * CALCULATE INTERACTIONS *
 944              **************************/
 945
 946             r11              = _mm256_mul_pd(rsq11,rinv11);
 947             r11              = _mm256_andnot_pd(dummy_mask,r11);
 948
 949             /* EWALD ELECTROSTATICS */
 950
 951             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 952             ewrt             = _mm256_mul_pd(r11,ewtabscale);
 953             ewitab           = _mm256_cvttpd_epi32(ewrt);
 954             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 955             ewitab           = _mm_slli_epi32(ewitab,2);
 956             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 957             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 958             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 959             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 960             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 961             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 962             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 963             velec            = _mm256_mul_pd(qq11,_mm256_sub_pd(rinv11,velec));
 964             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
 965
 966             /* Update potential sum for this i atom from the interaction with this j atom. */
 967             velec            = _mm256_andnot_pd(dummy_mask,velec);
 968             velecsum         = _mm256_add_pd(velecsum,velec);
 969
 970             fscal            = felec;
 971
 972             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm256_mul_pd(fscal,dx11);
 976             ty               = _mm256_mul_pd(fscal,dy11);
 977             tz               = _mm256_mul_pd(fscal,dz11);
 978
 979             /* Update vectorial force */
 980             fix1             = _mm256_add_pd(fix1,tx);
 981             fiy1             = _mm256_add_pd(fiy1,ty);
 982             fiz1             = _mm256_add_pd(fiz1,tz);
 983
 984             fjx1             = _mm256_add_pd(fjx1,tx);
 985             fjy1             = _mm256_add_pd(fjy1,ty);
 986             fjz1             = _mm256_add_pd(fjz1,tz);
 987
 988             /**************************
 989              * CALCULATE INTERACTIONS *
 990              **************************/
 991
 992             r12              = _mm256_mul_pd(rsq12,rinv12);
 993             r12              = _mm256_andnot_pd(dummy_mask,r12);
 994
 995             /* EWALD ELECTROSTATICS */
 996
 997             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 998             ewrt             = _mm256_mul_pd(r12,ewtabscale);
 999             ewitab           = _mm256_cvttpd_epi32(ewrt);
1000             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1001             ewitab           = _mm_slli_epi32(ewitab,2);
1002             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1003             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1004             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1005             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1006             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1007             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1008             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1009             velec            = _mm256_mul_pd(qq12,_mm256_sub_pd(rinv12,velec));
1010             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1011
1012             /* Update potential sum for this i atom from the interaction with this j atom. */
1013             velec            = _mm256_andnot_pd(dummy_mask,velec);
1014             velecsum         = _mm256_add_pd(velecsum,velec);
1015
1016             fscal            = felec;
1017
1018             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1019
1020             /* Calculate temporary vectorial force */
1021             tx               = _mm256_mul_pd(fscal,dx12);
1022             ty               = _mm256_mul_pd(fscal,dy12);
1023             tz               = _mm256_mul_pd(fscal,dz12);
1024
1025             /* Update vectorial force */
1026             fix1             = _mm256_add_pd(fix1,tx);
1027             fiy1             = _mm256_add_pd(fiy1,ty);
1028             fiz1             = _mm256_add_pd(fiz1,tz);
1029
1030             fjx2             = _mm256_add_pd(fjx2,tx);
1031             fjy2             = _mm256_add_pd(fjy2,ty);
1032             fjz2             = _mm256_add_pd(fjz2,tz);
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             r13              = _mm256_mul_pd(rsq13,rinv13);
1039             r13              = _mm256_andnot_pd(dummy_mask,r13);
1040
1041             /* EWALD ELECTROSTATICS */
1042
1043             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1044             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1045             ewitab           = _mm256_cvttpd_epi32(ewrt);
1046             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1047             ewitab           = _mm_slli_epi32(ewitab,2);
1048             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1049             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1050             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1051             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1052             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1053             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1054             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1055             velec            = _mm256_mul_pd(qq13,_mm256_sub_pd(rinv13,velec));
1056             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1057
1058             /* Update potential sum for this i atom from the interaction with this j atom. */
1059             velec            = _mm256_andnot_pd(dummy_mask,velec);
1060             velecsum         = _mm256_add_pd(velecsum,velec);
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1065
1066             /* Calculate temporary vectorial force */
1067             tx               = _mm256_mul_pd(fscal,dx13);
1068             ty               = _mm256_mul_pd(fscal,dy13);
1069             tz               = _mm256_mul_pd(fscal,dz13);
1070
1071             /* Update vectorial force */
1072             fix1             = _mm256_add_pd(fix1,tx);
1073             fiy1             = _mm256_add_pd(fiy1,ty);
1074             fiz1             = _mm256_add_pd(fiz1,tz);
1075
1076             fjx3             = _mm256_add_pd(fjx3,tx);
1077             fjy3             = _mm256_add_pd(fjy3,ty);
1078             fjz3             = _mm256_add_pd(fjz3,tz);
1079
1080             /**************************
1081              * CALCULATE INTERACTIONS *
1082              **************************/
1083
1084             r21              = _mm256_mul_pd(rsq21,rinv21);
1085             r21              = _mm256_andnot_pd(dummy_mask,r21);
1086
1087             /* EWALD ELECTROSTATICS */
1088
1089             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1090             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1091             ewitab           = _mm256_cvttpd_epi32(ewrt);
1092             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1093             ewitab           = _mm_slli_epi32(ewitab,2);
1094             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1095             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1096             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1097             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1098             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1099             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1100             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1101             velec            = _mm256_mul_pd(qq21,_mm256_sub_pd(rinv21,velec));
1102             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1103
1104             /* Update potential sum for this i atom from the interaction with this j atom. */
1105             velec            = _mm256_andnot_pd(dummy_mask,velec);
1106             velecsum         = _mm256_add_pd(velecsum,velec);
1107
1108             fscal            = felec;
1109
1110             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm256_mul_pd(fscal,dx21);
1114             ty               = _mm256_mul_pd(fscal,dy21);
1115             tz               = _mm256_mul_pd(fscal,dz21);
1116
1117             /* Update vectorial force */
1118             fix2             = _mm256_add_pd(fix2,tx);
1119             fiy2             = _mm256_add_pd(fiy2,ty);
1120             fiz2             = _mm256_add_pd(fiz2,tz);
1121
1122             fjx1             = _mm256_add_pd(fjx1,tx);
1123             fjy1             = _mm256_add_pd(fjy1,ty);
1124             fjz1             = _mm256_add_pd(fjz1,tz);
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             r22              = _mm256_mul_pd(rsq22,rinv22);
1131             r22              = _mm256_andnot_pd(dummy_mask,r22);
1132
1133             /* EWALD ELECTROSTATICS */
1134
1135             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1136             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1137             ewitab           = _mm256_cvttpd_epi32(ewrt);
1138             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1139             ewitab           = _mm_slli_epi32(ewitab,2);
1140             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1141             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1142             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1143             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1144             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1145             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1146             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1147             velec            = _mm256_mul_pd(qq22,_mm256_sub_pd(rinv22,velec));
1148             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1149
1150             /* Update potential sum for this i atom from the interaction with this j atom. */
1151             velec            = _mm256_andnot_pd(dummy_mask,velec);
1152             velecsum         = _mm256_add_pd(velecsum,velec);
1153
1154             fscal            = felec;
1155
1156             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1157
1158             /* Calculate temporary vectorial force */
1159             tx               = _mm256_mul_pd(fscal,dx22);
1160             ty               = _mm256_mul_pd(fscal,dy22);
1161             tz               = _mm256_mul_pd(fscal,dz22);
1162
1163             /* Update vectorial force */
1164             fix2             = _mm256_add_pd(fix2,tx);
1165             fiy2             = _mm256_add_pd(fiy2,ty);
1166             fiz2             = _mm256_add_pd(fiz2,tz);
1167
1168             fjx2             = _mm256_add_pd(fjx2,tx);
1169             fjy2             = _mm256_add_pd(fjy2,ty);
1170             fjz2             = _mm256_add_pd(fjz2,tz);
1171
1172             /**************************
1173              * CALCULATE INTERACTIONS *
1174              **************************/
1175
1176             r23              = _mm256_mul_pd(rsq23,rinv23);
1177             r23              = _mm256_andnot_pd(dummy_mask,r23);
1178
1179             /* EWALD ELECTROSTATICS */
1180
1181             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1182             ewrt             = _mm256_mul_pd(r23,ewtabscale);
1183             ewitab           = _mm256_cvttpd_epi32(ewrt);
1184             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1185             ewitab           = _mm_slli_epi32(ewitab,2);
1186             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1187             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1188             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1189             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1190             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1191             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1192             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1193             velec            = _mm256_mul_pd(qq23,_mm256_sub_pd(rinv23,velec));
1194             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
1195
1196             /* Update potential sum for this i atom from the interaction with this j atom. */
1197             velec            = _mm256_andnot_pd(dummy_mask,velec);
1198             velecsum         = _mm256_add_pd(velecsum,velec);
1199
1200             fscal            = felec;
1201
1202             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm256_mul_pd(fscal,dx23);
1206             ty               = _mm256_mul_pd(fscal,dy23);
1207             tz               = _mm256_mul_pd(fscal,dz23);
1208
1209             /* Update vectorial force */
1210             fix2             = _mm256_add_pd(fix2,tx);
1211             fiy2             = _mm256_add_pd(fiy2,ty);
1212             fiz2             = _mm256_add_pd(fiz2,tz);
1213
1214             fjx3             = _mm256_add_pd(fjx3,tx);
1215             fjy3             = _mm256_add_pd(fjy3,ty);
1216             fjz3             = _mm256_add_pd(fjz3,tz);
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             r31              = _mm256_mul_pd(rsq31,rinv31);
1223             r31              = _mm256_andnot_pd(dummy_mask,r31);
1224
1225             /* EWALD ELECTROSTATICS */
1226
1227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1228             ewrt             = _mm256_mul_pd(r31,ewtabscale);
1229             ewitab           = _mm256_cvttpd_epi32(ewrt);
1230             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1231             ewitab           = _mm_slli_epi32(ewitab,2);
1232             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1233             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1234             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1235             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1236             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1237             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1238             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1239             velec            = _mm256_mul_pd(qq31,_mm256_sub_pd(rinv31,velec));
1240             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
1241
1242             /* Update potential sum for this i atom from the interaction with this j atom. */
1243             velec            = _mm256_andnot_pd(dummy_mask,velec);
1244             velecsum         = _mm256_add_pd(velecsum,velec);
1245
1246             fscal            = felec;
1247
1248             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1249
1250             /* Calculate temporary vectorial force */
1251             tx               = _mm256_mul_pd(fscal,dx31);
1252             ty               = _mm256_mul_pd(fscal,dy31);
1253             tz               = _mm256_mul_pd(fscal,dz31);
1254
1255             /* Update vectorial force */
1256             fix3             = _mm256_add_pd(fix3,tx);
1257             fiy3             = _mm256_add_pd(fiy3,ty);
1258             fiz3             = _mm256_add_pd(fiz3,tz);
1259
1260             fjx1             = _mm256_add_pd(fjx1,tx);
1261             fjy1             = _mm256_add_pd(fjy1,ty);
1262             fjz1             = _mm256_add_pd(fjz1,tz);
1263
1264             /**************************
1265              * CALCULATE INTERACTIONS *
1266              **************************/
1267
1268             r32              = _mm256_mul_pd(rsq32,rinv32);
1269             r32              = _mm256_andnot_pd(dummy_mask,r32);
1270
1271             /* EWALD ELECTROSTATICS */
1272
1273             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1274             ewrt             = _mm256_mul_pd(r32,ewtabscale);
1275             ewitab           = _mm256_cvttpd_epi32(ewrt);
1276             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1277             ewitab           = _mm_slli_epi32(ewitab,2);
1278             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1279             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1280             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1281             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1282             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1283             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1284             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1285             velec            = _mm256_mul_pd(qq32,_mm256_sub_pd(rinv32,velec));
1286             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
1287
1288             /* Update potential sum for this i atom from the interaction with this j atom. */
1289             velec            = _mm256_andnot_pd(dummy_mask,velec);
1290             velecsum         = _mm256_add_pd(velecsum,velec);
1291
1292             fscal            = felec;
1293
1294             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1295
1296             /* Calculate temporary vectorial force */
1297             tx               = _mm256_mul_pd(fscal,dx32);
1298             ty               = _mm256_mul_pd(fscal,dy32);
1299             tz               = _mm256_mul_pd(fscal,dz32);
1300
1301             /* Update vectorial force */
1302             fix3             = _mm256_add_pd(fix3,tx);
1303             fiy3             = _mm256_add_pd(fiy3,ty);
1304             fiz3             = _mm256_add_pd(fiz3,tz);
1305
1306             fjx2             = _mm256_add_pd(fjx2,tx);
1307             fjy2             = _mm256_add_pd(fjy2,ty);
1308             fjz2             = _mm256_add_pd(fjz2,tz);
1309
1310             /**************************
1311              * CALCULATE INTERACTIONS *
1312              **************************/
1313
1314             r33              = _mm256_mul_pd(rsq33,rinv33);
1315             r33              = _mm256_andnot_pd(dummy_mask,r33);
1316
1317             /* EWALD ELECTROSTATICS */
1318
1319             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1320             ewrt             = _mm256_mul_pd(r33,ewtabscale);
1321             ewitab           = _mm256_cvttpd_epi32(ewrt);
1322             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1323             ewitab           = _mm_slli_epi32(ewitab,2);
1324             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1325             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1326             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
1327             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
1328             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
1329             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
1330             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
1331             velec            = _mm256_mul_pd(qq33,_mm256_sub_pd(rinv33,velec));
1332             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
1333
1334             /* Update potential sum for this i atom from the interaction with this j atom. */
1335             velec            = _mm256_andnot_pd(dummy_mask,velec);
1336             velecsum         = _mm256_add_pd(velecsum,velec);
1337
1338             fscal            = felec;
1339
1340             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1341
1342             /* Calculate temporary vectorial force */
1343             tx               = _mm256_mul_pd(fscal,dx33);
1344             ty               = _mm256_mul_pd(fscal,dy33);
1345             tz               = _mm256_mul_pd(fscal,dz33);
1346
1347             /* Update vectorial force */
1348             fix3             = _mm256_add_pd(fix3,tx);
1349             fiy3             = _mm256_add_pd(fiy3,ty);
1350             fiz3             = _mm256_add_pd(fiz3,tz);
1351
1352             fjx3             = _mm256_add_pd(fjx3,tx);
1353             fjy3             = _mm256_add_pd(fjy3,ty);
1354             fjz3             = _mm256_add_pd(fjz3,tz);
1355
1356             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1357             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1358             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1359             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1360
1361             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1362                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1363                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1364
1365             /* Inner loop uses 438 flops */
1366         }
1367
1368         /* End of innermost loop */
1369
1370         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1371                                                  f+i_coord_offset,fshift+i_shift_offset);
1372
1373         ggid                        = gid[iidx];
1374         /* Update potential energies */
1375         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1376         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1377
1378         /* Increment number of inner iterations */
1379         inneriter                  += j_index_end - j_index_start;
1380
1381         /* Outer loop uses 26 flops */
1382     }
1383
1384     /* Increment number of outer iterations */
1385     outeriter        += nri;
1386
1387     /* Update outer/inner flops */
1388
1389     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*438);
1390 }
1391 /*
1392  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4W4_F_avx_256_double
1393  * Electrostatics interaction: Ewald
1394  * VdW interaction:            CubicSplineTable
1395  * Geometry:                   Water4-Water4
1396  * Calculate force/pot:        Force
1397  */
1398 void
1399 nb_kernel_ElecEw_VdwCSTab_GeomW4W4_F_avx_256_double
1400                     (t_nblist * gmx_restrict                nlist,
1401                      rvec * gmx_restrict                    xx,
1402                      rvec * gmx_restrict                    ff,
1403                      t_forcerec * gmx_restrict              fr,
1404                      t_mdatoms * gmx_restrict               mdatoms,
1405                      nb_kernel_data_t * gmx_restrict        kernel_data,
1406                      t_nrnb * gmx_restrict                  nrnb)
1407 {
1408     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1409      * just 0 for non-waters.
1410      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1411      * jnr indices corresponding to data put in the four positions in the SIMD register.
1412      */
1413     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1414     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1415     int              jnrA,jnrB,jnrC,jnrD;
1416     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1417     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1418     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1419     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1420     real             rcutoff_scalar;
1421     real             *shiftvec,*fshift,*x,*f;
1422     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1423     real             scratch[4*DIM];
1424     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1425     real *           vdwioffsetptr0;
1426     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1427     real *           vdwioffsetptr1;
1428     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1429     real *           vdwioffsetptr2;
1430     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1431     real *           vdwioffsetptr3;
1432     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1433     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1434     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1435     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1436     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1437     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1438     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1439     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1440     __m256d          jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1441     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1442     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1443     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1444     __m256d          dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1445     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1446     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1447     __m256d          dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1448     __m256d          dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1449     __m256d          dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1450     __m256d          dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1451     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1452     real             *charge;
1453     int              nvdwtype;
1454     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1455     int              *vdwtype;
1456     real             *vdwparam;
1457     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1458     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1459     __m128i          vfitab;
1460     __m128i          ifour       = _mm_set1_epi32(4);
1461     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1462     real             *vftab;
1463     __m128i          ewitab;
1464     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1465     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1466     real             *ewtab;
1467     __m256d          dummy_mask,cutoff_mask;
1468     __m128           tmpmask0,tmpmask1;
1469     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1470     __m256d          one     = _mm256_set1_pd(1.0);
1471     __m256d          two     = _mm256_set1_pd(2.0);
1472     x                = xx[0];
1473     f                = ff[0];
1474
1475     nri              = nlist->nri;
1476     iinr             = nlist->iinr;
1477     jindex           = nlist->jindex;
1478     jjnr             = nlist->jjnr;
1479     shiftidx         = nlist->shift;
1480     gid              = nlist->gid;
1481     shiftvec         = fr->shift_vec[0];
1482     fshift           = fr->fshift[0];
1483     facel            = _mm256_set1_pd(fr->epsfac);
1484     charge           = mdatoms->chargeA;
1485     nvdwtype         = fr->ntype;
1486     vdwparam         = fr->nbfp;
1487     vdwtype          = mdatoms->typeA;
1488
1489     vftab            = kernel_data->table_vdw->data;
1490     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
1491
1492     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
1493     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
1494     beta2            = _mm256_mul_pd(beta,beta);
1495     beta3            = _mm256_mul_pd(beta,beta2);
1496
1497     ewtab            = fr->ic->tabq_coul_F;
1498     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
1499     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
1500
1501     /* Setup water-specific parameters */
1502     inr              = nlist->iinr[0];
1503     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1504     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1505     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
1506     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1507
1508     jq1              = _mm256_set1_pd(charge[inr+1]);
1509     jq2              = _mm256_set1_pd(charge[inr+2]);
1510     jq3              = _mm256_set1_pd(charge[inr+3]);
1511     vdwjidx0A        = 2*vdwtype[inr+0];
1512     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1513     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1514     qq11             = _mm256_mul_pd(iq1,jq1);
1515     qq12             = _mm256_mul_pd(iq1,jq2);
1516     qq13             = _mm256_mul_pd(iq1,jq3);
1517     qq21             = _mm256_mul_pd(iq2,jq1);
1518     qq22             = _mm256_mul_pd(iq2,jq2);
1519     qq23             = _mm256_mul_pd(iq2,jq3);
1520     qq31             = _mm256_mul_pd(iq3,jq1);
1521     qq32             = _mm256_mul_pd(iq3,jq2);
1522     qq33             = _mm256_mul_pd(iq3,jq3);
1523
1524     /* Avoid stupid compiler warnings */
1525     jnrA = jnrB = jnrC = jnrD = 0;
1526     j_coord_offsetA = 0;
1527     j_coord_offsetB = 0;
1528     j_coord_offsetC = 0;
1529     j_coord_offsetD = 0;
1530
1531     outeriter        = 0;
1532     inneriter        = 0;
1533
1534     for(iidx=0;iidx<4*DIM;iidx++)
1535     {
1536         scratch[iidx] = 0.0;
1537     }
1538
1539     /* Start outer loop over neighborlists */
1540     for(iidx=0; iidx<nri; iidx++)
1541     {
1542         /* Load shift vector for this list */
1543         i_shift_offset   = DIM*shiftidx[iidx];
1544
1545         /* Load limits for loop over neighbors */
1546         j_index_start    = jindex[iidx];
1547         j_index_end      = jindex[iidx+1];
1548
1549         /* Get outer coordinate index */
1550         inr              = iinr[iidx];
1551         i_coord_offset   = DIM*inr;
1552
1553         /* Load i particle coords and add shift vector */
1554         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1555                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1556
1557         fix0             = _mm256_setzero_pd();
1558         fiy0             = _mm256_setzero_pd();
1559         fiz0             = _mm256_setzero_pd();
1560         fix1             = _mm256_setzero_pd();
1561         fiy1             = _mm256_setzero_pd();
1562         fiz1             = _mm256_setzero_pd();
1563         fix2             = _mm256_setzero_pd();
1564         fiy2             = _mm256_setzero_pd();
1565         fiz2             = _mm256_setzero_pd();
1566         fix3             = _mm256_setzero_pd();
1567         fiy3             = _mm256_setzero_pd();
1568         fiz3             = _mm256_setzero_pd();
1569
1570         /* Start inner kernel loop */
1571         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1572         {
1573
1574             /* Get j neighbor index, and coordinate index */
1575             jnrA             = jjnr[jidx];
1576             jnrB             = jjnr[jidx+1];
1577             jnrC             = jjnr[jidx+2];
1578             jnrD             = jjnr[jidx+3];
1579             j_coord_offsetA  = DIM*jnrA;
1580             j_coord_offsetB  = DIM*jnrB;
1581             j_coord_offsetC  = DIM*jnrC;
1582             j_coord_offsetD  = DIM*jnrD;
1583
1584             /* load j atom coordinates */
1585             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1586                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1587                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1588                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1589
1590             /* Calculate displacement vector */
1591             dx00             = _mm256_sub_pd(ix0,jx0);
1592             dy00             = _mm256_sub_pd(iy0,jy0);
1593             dz00             = _mm256_sub_pd(iz0,jz0);
1594             dx11             = _mm256_sub_pd(ix1,jx1);
1595             dy11             = _mm256_sub_pd(iy1,jy1);
1596             dz11             = _mm256_sub_pd(iz1,jz1);
1597             dx12             = _mm256_sub_pd(ix1,jx2);
1598             dy12             = _mm256_sub_pd(iy1,jy2);
1599             dz12             = _mm256_sub_pd(iz1,jz2);
1600             dx13             = _mm256_sub_pd(ix1,jx3);
1601             dy13             = _mm256_sub_pd(iy1,jy3);
1602             dz13             = _mm256_sub_pd(iz1,jz3);
1603             dx21             = _mm256_sub_pd(ix2,jx1);
1604             dy21             = _mm256_sub_pd(iy2,jy1);
1605             dz21             = _mm256_sub_pd(iz2,jz1);
1606             dx22             = _mm256_sub_pd(ix2,jx2);
1607             dy22             = _mm256_sub_pd(iy2,jy2);
1608             dz22             = _mm256_sub_pd(iz2,jz2);
1609             dx23             = _mm256_sub_pd(ix2,jx3);
1610             dy23             = _mm256_sub_pd(iy2,jy3);
1611             dz23             = _mm256_sub_pd(iz2,jz3);
1612             dx31             = _mm256_sub_pd(ix3,jx1);
1613             dy31             = _mm256_sub_pd(iy3,jy1);
1614             dz31             = _mm256_sub_pd(iz3,jz1);
1615             dx32             = _mm256_sub_pd(ix3,jx2);
1616             dy32             = _mm256_sub_pd(iy3,jy2);
1617             dz32             = _mm256_sub_pd(iz3,jz2);
1618             dx33             = _mm256_sub_pd(ix3,jx3);
1619             dy33             = _mm256_sub_pd(iy3,jy3);
1620             dz33             = _mm256_sub_pd(iz3,jz3);
1621
1622             /* Calculate squared distance and things based on it */
1623             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1624             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1625             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1626             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
1627             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1628             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1629             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
1630             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
1631             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
1632             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
1633
1634             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1635             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1636             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1637             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
1638             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1639             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1640             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
1641             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
1642             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
1643             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
1644
1645             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
1646             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
1647             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
1648             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
1649             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
1650             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
1651             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
1652             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
1653             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
1654
1655             fjx0             = _mm256_setzero_pd();
1656             fjy0             = _mm256_setzero_pd();
1657             fjz0             = _mm256_setzero_pd();
1658             fjx1             = _mm256_setzero_pd();
1659             fjy1             = _mm256_setzero_pd();
1660             fjz1             = _mm256_setzero_pd();
1661             fjx2             = _mm256_setzero_pd();
1662             fjy2             = _mm256_setzero_pd();
1663             fjz2             = _mm256_setzero_pd();
1664             fjx3             = _mm256_setzero_pd();
1665             fjy3             = _mm256_setzero_pd();
1666             fjz3             = _mm256_setzero_pd();
1667
1668             /**************************
1669              * CALCULATE INTERACTIONS *
1670              **************************/
1671
1672             r00              = _mm256_mul_pd(rsq00,rinv00);
1673
1674             /* Calculate table index by multiplying r with table scale and truncate to integer */
1675             rt               = _mm256_mul_pd(r00,vftabscale);
1676             vfitab           = _mm256_cvttpd_epi32(rt);
1677             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1678             vfitab           = _mm_slli_epi32(vfitab,3);
1679
1680             /* CUBIC SPLINE TABLE DISPERSION */
1681             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1682             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1683             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1684             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1685             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1686             Heps             = _mm256_mul_pd(vfeps,H);
1687             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1688             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1689             fvdw6            = _mm256_mul_pd(c6_00,FF);
1690
1691             /* CUBIC SPLINE TABLE REPULSION */
1692             vfitab           = _mm_add_epi32(vfitab,ifour);
1693             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1694             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1695             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1696             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1697             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1698             Heps             = _mm256_mul_pd(vfeps,H);
1699             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1700             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1701             fvdw12           = _mm256_mul_pd(c12_00,FF);
1702             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1703
1704             fscal            = fvdw;
1705
1706             /* Calculate temporary vectorial force */
1707             tx               = _mm256_mul_pd(fscal,dx00);
1708             ty               = _mm256_mul_pd(fscal,dy00);
1709             tz               = _mm256_mul_pd(fscal,dz00);
1710
1711             /* Update vectorial force */
1712             fix0             = _mm256_add_pd(fix0,tx);
1713             fiy0             = _mm256_add_pd(fiy0,ty);
1714             fiz0             = _mm256_add_pd(fiz0,tz);
1715
1716             fjx0             = _mm256_add_pd(fjx0,tx);
1717             fjy0             = _mm256_add_pd(fjy0,ty);
1718             fjz0             = _mm256_add_pd(fjz0,tz);
1719
1720             /**************************
1721              * CALCULATE INTERACTIONS *
1722              **************************/
1723
1724             r11              = _mm256_mul_pd(rsq11,rinv11);
1725
1726             /* EWALD ELECTROSTATICS */
1727
1728             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1729             ewrt             = _mm256_mul_pd(r11,ewtabscale);
1730             ewitab           = _mm256_cvttpd_epi32(ewrt);
1731             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1732             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1733                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1734                                             &ewtabF,&ewtabFn);
1735             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1736             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
1737
1738             fscal            = felec;
1739
1740             /* Calculate temporary vectorial force */
1741             tx               = _mm256_mul_pd(fscal,dx11);
1742             ty               = _mm256_mul_pd(fscal,dy11);
1743             tz               = _mm256_mul_pd(fscal,dz11);
1744
1745             /* Update vectorial force */
1746             fix1             = _mm256_add_pd(fix1,tx);
1747             fiy1             = _mm256_add_pd(fiy1,ty);
1748             fiz1             = _mm256_add_pd(fiz1,tz);
1749
1750             fjx1             = _mm256_add_pd(fjx1,tx);
1751             fjy1             = _mm256_add_pd(fjy1,ty);
1752             fjz1             = _mm256_add_pd(fjz1,tz);
1753
1754             /**************************
1755              * CALCULATE INTERACTIONS *
1756              **************************/
1757
1758             r12              = _mm256_mul_pd(rsq12,rinv12);
1759
1760             /* EWALD ELECTROSTATICS */
1761
1762             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1763             ewrt             = _mm256_mul_pd(r12,ewtabscale);
1764             ewitab           = _mm256_cvttpd_epi32(ewrt);
1765             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1766             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1767                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1768                                             &ewtabF,&ewtabFn);
1769             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1770             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
1771
1772             fscal            = felec;
1773
1774             /* Calculate temporary vectorial force */
1775             tx               = _mm256_mul_pd(fscal,dx12);
1776             ty               = _mm256_mul_pd(fscal,dy12);
1777             tz               = _mm256_mul_pd(fscal,dz12);
1778
1779             /* Update vectorial force */
1780             fix1             = _mm256_add_pd(fix1,tx);
1781             fiy1             = _mm256_add_pd(fiy1,ty);
1782             fiz1             = _mm256_add_pd(fiz1,tz);
1783
1784             fjx2             = _mm256_add_pd(fjx2,tx);
1785             fjy2             = _mm256_add_pd(fjy2,ty);
1786             fjz2             = _mm256_add_pd(fjz2,tz);
1787
1788             /**************************
1789              * CALCULATE INTERACTIONS *
1790              **************************/
1791
1792             r13              = _mm256_mul_pd(rsq13,rinv13);
1793
1794             /* EWALD ELECTROSTATICS */
1795
1796             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1797             ewrt             = _mm256_mul_pd(r13,ewtabscale);
1798             ewitab           = _mm256_cvttpd_epi32(ewrt);
1799             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1800             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1801                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1802                                             &ewtabF,&ewtabFn);
1803             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1804             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
1805
1806             fscal            = felec;
1807
1808             /* Calculate temporary vectorial force */
1809             tx               = _mm256_mul_pd(fscal,dx13);
1810             ty               = _mm256_mul_pd(fscal,dy13);
1811             tz               = _mm256_mul_pd(fscal,dz13);
1812
1813             /* Update vectorial force */
1814             fix1             = _mm256_add_pd(fix1,tx);
1815             fiy1             = _mm256_add_pd(fiy1,ty);
1816             fiz1             = _mm256_add_pd(fiz1,tz);
1817
1818             fjx3             = _mm256_add_pd(fjx3,tx);
1819             fjy3             = _mm256_add_pd(fjy3,ty);
1820             fjz3             = _mm256_add_pd(fjz3,tz);
1821
1822             /**************************
1823              * CALCULATE INTERACTIONS *
1824              **************************/
1825
1826             r21              = _mm256_mul_pd(rsq21,rinv21);
1827
1828             /* EWALD ELECTROSTATICS */
1829
1830             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1831             ewrt             = _mm256_mul_pd(r21,ewtabscale);
1832             ewitab           = _mm256_cvttpd_epi32(ewrt);
1833             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1834             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1835                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1836                                             &ewtabF,&ewtabFn);
1837             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1838             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
1839
1840             fscal            = felec;
1841
1842             /* Calculate temporary vectorial force */
1843             tx               = _mm256_mul_pd(fscal,dx21);
1844             ty               = _mm256_mul_pd(fscal,dy21);
1845             tz               = _mm256_mul_pd(fscal,dz21);
1846
1847             /* Update vectorial force */
1848             fix2             = _mm256_add_pd(fix2,tx);
1849             fiy2             = _mm256_add_pd(fiy2,ty);
1850             fiz2             = _mm256_add_pd(fiz2,tz);
1851
1852             fjx1             = _mm256_add_pd(fjx1,tx);
1853             fjy1             = _mm256_add_pd(fjy1,ty);
1854             fjz1             = _mm256_add_pd(fjz1,tz);
1855
1856             /**************************
1857              * CALCULATE INTERACTIONS *
1858              **************************/
1859
1860             r22              = _mm256_mul_pd(rsq22,rinv22);
1861
1862             /* EWALD ELECTROSTATICS */
1863
1864             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1865             ewrt             = _mm256_mul_pd(r22,ewtabscale);
1866             ewitab           = _mm256_cvttpd_epi32(ewrt);
1867             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1868             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1869                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1870                                             &ewtabF,&ewtabFn);
1871             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1872             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
1873
1874             fscal            = felec;
1875
1876             /* Calculate temporary vectorial force */
1877             tx               = _mm256_mul_pd(fscal,dx22);
1878             ty               = _mm256_mul_pd(fscal,dy22);
1879             tz               = _mm256_mul_pd(fscal,dz22);
1880
1881             /* Update vectorial force */
1882             fix2             = _mm256_add_pd(fix2,tx);
1883             fiy2             = _mm256_add_pd(fiy2,ty);
1884             fiz2             = _mm256_add_pd(fiz2,tz);
1885
1886             fjx2             = _mm256_add_pd(fjx2,tx);
1887             fjy2             = _mm256_add_pd(fjy2,ty);
1888             fjz2             = _mm256_add_pd(fjz2,tz);
1889
1890             /**************************
1891              * CALCULATE INTERACTIONS *
1892              **************************/
1893
1894             r23              = _mm256_mul_pd(rsq23,rinv23);
1895
1896             /* EWALD ELECTROSTATICS */
1897
1898             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1899             ewrt             = _mm256_mul_pd(r23,ewtabscale);
1900             ewitab           = _mm256_cvttpd_epi32(ewrt);
1901             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1902             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1903                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1904                                             &ewtabF,&ewtabFn);
1905             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1906             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
1907
1908             fscal            = felec;
1909
1910             /* Calculate temporary vectorial force */
1911             tx               = _mm256_mul_pd(fscal,dx23);
1912             ty               = _mm256_mul_pd(fscal,dy23);
1913             tz               = _mm256_mul_pd(fscal,dz23);
1914
1915             /* Update vectorial force */
1916             fix2             = _mm256_add_pd(fix2,tx);
1917             fiy2             = _mm256_add_pd(fiy2,ty);
1918             fiz2             = _mm256_add_pd(fiz2,tz);
1919
1920             fjx3             = _mm256_add_pd(fjx3,tx);
1921             fjy3             = _mm256_add_pd(fjy3,ty);
1922             fjz3             = _mm256_add_pd(fjz3,tz);
1923
1924             /**************************
1925              * CALCULATE INTERACTIONS *
1926              **************************/
1927
1928             r31              = _mm256_mul_pd(rsq31,rinv31);
1929
1930             /* EWALD ELECTROSTATICS */
1931
1932             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1933             ewrt             = _mm256_mul_pd(r31,ewtabscale);
1934             ewitab           = _mm256_cvttpd_epi32(ewrt);
1935             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1936             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1937                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1938                                             &ewtabF,&ewtabFn);
1939             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1940             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
1941
1942             fscal            = felec;
1943
1944             /* Calculate temporary vectorial force */
1945             tx               = _mm256_mul_pd(fscal,dx31);
1946             ty               = _mm256_mul_pd(fscal,dy31);
1947             tz               = _mm256_mul_pd(fscal,dz31);
1948
1949             /* Update vectorial force */
1950             fix3             = _mm256_add_pd(fix3,tx);
1951             fiy3             = _mm256_add_pd(fiy3,ty);
1952             fiz3             = _mm256_add_pd(fiz3,tz);
1953
1954             fjx1             = _mm256_add_pd(fjx1,tx);
1955             fjy1             = _mm256_add_pd(fjy1,ty);
1956             fjz1             = _mm256_add_pd(fjz1,tz);
1957
1958             /**************************
1959              * CALCULATE INTERACTIONS *
1960              **************************/
1961
1962             r32              = _mm256_mul_pd(rsq32,rinv32);
1963
1964             /* EWALD ELECTROSTATICS */
1965
1966             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1967             ewrt             = _mm256_mul_pd(r32,ewtabscale);
1968             ewitab           = _mm256_cvttpd_epi32(ewrt);
1969             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1970             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1971                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1972                                             &ewtabF,&ewtabFn);
1973             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1974             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
1975
1976             fscal            = felec;
1977
1978             /* Calculate temporary vectorial force */
1979             tx               = _mm256_mul_pd(fscal,dx32);
1980             ty               = _mm256_mul_pd(fscal,dy32);
1981             tz               = _mm256_mul_pd(fscal,dz32);
1982
1983             /* Update vectorial force */
1984             fix3             = _mm256_add_pd(fix3,tx);
1985             fiy3             = _mm256_add_pd(fiy3,ty);
1986             fiz3             = _mm256_add_pd(fiz3,tz);
1987
1988             fjx2             = _mm256_add_pd(fjx2,tx);
1989             fjy2             = _mm256_add_pd(fjy2,ty);
1990             fjz2             = _mm256_add_pd(fjz2,tz);
1991
1992             /**************************
1993              * CALCULATE INTERACTIONS *
1994              **************************/
1995
1996             r33              = _mm256_mul_pd(rsq33,rinv33);
1997
1998             /* EWALD ELECTROSTATICS */
1999
2000             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2001             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2002             ewitab           = _mm256_cvttpd_epi32(ewrt);
2003             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2004             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2005                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2006                                             &ewtabF,&ewtabFn);
2007             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2008             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2009
2010             fscal            = felec;
2011
2012             /* Calculate temporary vectorial force */
2013             tx               = _mm256_mul_pd(fscal,dx33);
2014             ty               = _mm256_mul_pd(fscal,dy33);
2015             tz               = _mm256_mul_pd(fscal,dz33);
2016
2017             /* Update vectorial force */
2018             fix3             = _mm256_add_pd(fix3,tx);
2019             fiy3             = _mm256_add_pd(fiy3,ty);
2020             fiz3             = _mm256_add_pd(fiz3,tz);
2021
2022             fjx3             = _mm256_add_pd(fjx3,tx);
2023             fjy3             = _mm256_add_pd(fjy3,ty);
2024             fjz3             = _mm256_add_pd(fjz3,tz);
2025
2026             fjptrA             = f+j_coord_offsetA;
2027             fjptrB             = f+j_coord_offsetB;
2028             fjptrC             = f+j_coord_offsetC;
2029             fjptrD             = f+j_coord_offsetD;
2030
2031             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2032                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2033                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2034
2035             /* Inner loop uses 375 flops */
2036         }
2037
2038         if(jidx<j_index_end)
2039         {
2040
2041             /* Get j neighbor index, and coordinate index */
2042             jnrlistA         = jjnr[jidx];
2043             jnrlistB         = jjnr[jidx+1];
2044             jnrlistC         = jjnr[jidx+2];
2045             jnrlistD         = jjnr[jidx+3];
2046             /* Sign of each element will be negative for non-real atoms.
2047              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2048              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
2049              */
2050             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2051
2052             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
2053             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
2054             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
2055
2056             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2057             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2058             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2059             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2060             j_coord_offsetA  = DIM*jnrA;
2061             j_coord_offsetB  = DIM*jnrB;
2062             j_coord_offsetC  = DIM*jnrC;
2063             j_coord_offsetD  = DIM*jnrD;
2064
2065             /* load j atom coordinates */
2066             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
2067                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2068                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2069                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2070
2071             /* Calculate displacement vector */
2072             dx00             = _mm256_sub_pd(ix0,jx0);
2073             dy00             = _mm256_sub_pd(iy0,jy0);
2074             dz00             = _mm256_sub_pd(iz0,jz0);
2075             dx11             = _mm256_sub_pd(ix1,jx1);
2076             dy11             = _mm256_sub_pd(iy1,jy1);
2077             dz11             = _mm256_sub_pd(iz1,jz1);
2078             dx12             = _mm256_sub_pd(ix1,jx2);
2079             dy12             = _mm256_sub_pd(iy1,jy2);
2080             dz12             = _mm256_sub_pd(iz1,jz2);
2081             dx13             = _mm256_sub_pd(ix1,jx3);
2082             dy13             = _mm256_sub_pd(iy1,jy3);
2083             dz13             = _mm256_sub_pd(iz1,jz3);
2084             dx21             = _mm256_sub_pd(ix2,jx1);
2085             dy21             = _mm256_sub_pd(iy2,jy1);
2086             dz21             = _mm256_sub_pd(iz2,jz1);
2087             dx22             = _mm256_sub_pd(ix2,jx2);
2088             dy22             = _mm256_sub_pd(iy2,jy2);
2089             dz22             = _mm256_sub_pd(iz2,jz2);
2090             dx23             = _mm256_sub_pd(ix2,jx3);
2091             dy23             = _mm256_sub_pd(iy2,jy3);
2092             dz23             = _mm256_sub_pd(iz2,jz3);
2093             dx31             = _mm256_sub_pd(ix3,jx1);
2094             dy31             = _mm256_sub_pd(iy3,jy1);
2095             dz31             = _mm256_sub_pd(iz3,jz1);
2096             dx32             = _mm256_sub_pd(ix3,jx2);
2097             dy32             = _mm256_sub_pd(iy3,jy2);
2098             dz32             = _mm256_sub_pd(iz3,jz2);
2099             dx33             = _mm256_sub_pd(ix3,jx3);
2100             dy33             = _mm256_sub_pd(iy3,jy3);
2101             dz33             = _mm256_sub_pd(iz3,jz3);
2102
2103             /* Calculate squared distance and things based on it */
2104             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
2105             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
2106             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
2107             rsq13            = gmx_mm256_calc_rsq_pd(dx13,dy13,dz13);
2108             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
2109             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
2110             rsq23            = gmx_mm256_calc_rsq_pd(dx23,dy23,dz23);
2111             rsq31            = gmx_mm256_calc_rsq_pd(dx31,dy31,dz31);
2112             rsq32            = gmx_mm256_calc_rsq_pd(dx32,dy32,dz32);
2113             rsq33            = gmx_mm256_calc_rsq_pd(dx33,dy33,dz33);
2114
2115             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
2116             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
2117             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
2118             rinv13           = gmx_mm256_invsqrt_pd(rsq13);
2119             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
2120             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
2121             rinv23           = gmx_mm256_invsqrt_pd(rsq23);
2122             rinv31           = gmx_mm256_invsqrt_pd(rsq31);
2123             rinv32           = gmx_mm256_invsqrt_pd(rsq32);
2124             rinv33           = gmx_mm256_invsqrt_pd(rsq33);
2125
2126             rinvsq11         = _mm256_mul_pd(rinv11,rinv11);
2127             rinvsq12         = _mm256_mul_pd(rinv12,rinv12);
2128             rinvsq13         = _mm256_mul_pd(rinv13,rinv13);
2129             rinvsq21         = _mm256_mul_pd(rinv21,rinv21);
2130             rinvsq22         = _mm256_mul_pd(rinv22,rinv22);
2131             rinvsq23         = _mm256_mul_pd(rinv23,rinv23);
2132             rinvsq31         = _mm256_mul_pd(rinv31,rinv31);
2133             rinvsq32         = _mm256_mul_pd(rinv32,rinv32);
2134             rinvsq33         = _mm256_mul_pd(rinv33,rinv33);
2135
2136             fjx0             = _mm256_setzero_pd();
2137             fjy0             = _mm256_setzero_pd();
2138             fjz0             = _mm256_setzero_pd();
2139             fjx1             = _mm256_setzero_pd();
2140             fjy1             = _mm256_setzero_pd();
2141             fjz1             = _mm256_setzero_pd();
2142             fjx2             = _mm256_setzero_pd();
2143             fjy2             = _mm256_setzero_pd();
2144             fjz2             = _mm256_setzero_pd();
2145             fjx3             = _mm256_setzero_pd();
2146             fjy3             = _mm256_setzero_pd();
2147             fjz3             = _mm256_setzero_pd();
2148
2149             /**************************
2150              * CALCULATE INTERACTIONS *
2151              **************************/
2152
2153             r00              = _mm256_mul_pd(rsq00,rinv00);
2154             r00              = _mm256_andnot_pd(dummy_mask,r00);
2155
2156             /* Calculate table index by multiplying r with table scale and truncate to integer */
2157             rt               = _mm256_mul_pd(r00,vftabscale);
2158             vfitab           = _mm256_cvttpd_epi32(rt);
2159             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2160             vfitab           = _mm_slli_epi32(vfitab,3);
2161
2162             /* CUBIC SPLINE TABLE DISPERSION */
2163             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2164             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2165             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2166             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2167             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2168             Heps             = _mm256_mul_pd(vfeps,H);
2169             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2170             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2171             fvdw6            = _mm256_mul_pd(c6_00,FF);
2172
2173             /* CUBIC SPLINE TABLE REPULSION */
2174             vfitab           = _mm_add_epi32(vfitab,ifour);
2175             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2176             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2177             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2178             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2179             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2180             Heps             = _mm256_mul_pd(vfeps,H);
2181             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2182             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2183             fvdw12           = _mm256_mul_pd(c12_00,FF);
2184             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
2185
2186             fscal            = fvdw;
2187
2188             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2189
2190             /* Calculate temporary vectorial force */
2191             tx               = _mm256_mul_pd(fscal,dx00);
2192             ty               = _mm256_mul_pd(fscal,dy00);
2193             tz               = _mm256_mul_pd(fscal,dz00);
2194
2195             /* Update vectorial force */
2196             fix0             = _mm256_add_pd(fix0,tx);
2197             fiy0             = _mm256_add_pd(fiy0,ty);
2198             fiz0             = _mm256_add_pd(fiz0,tz);
2199
2200             fjx0             = _mm256_add_pd(fjx0,tx);
2201             fjy0             = _mm256_add_pd(fjy0,ty);
2202             fjz0             = _mm256_add_pd(fjz0,tz);
2203
2204             /**************************
2205              * CALCULATE INTERACTIONS *
2206              **************************/
2207
2208             r11              = _mm256_mul_pd(rsq11,rinv11);
2209             r11              = _mm256_andnot_pd(dummy_mask,r11);
2210
2211             /* EWALD ELECTROSTATICS */
2212
2213             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2214             ewrt             = _mm256_mul_pd(r11,ewtabscale);
2215             ewitab           = _mm256_cvttpd_epi32(ewrt);
2216             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2217             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2218                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2219                                             &ewtabF,&ewtabFn);
2220             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2221             felec            = _mm256_mul_pd(_mm256_mul_pd(qq11,rinv11),_mm256_sub_pd(rinvsq11,felec));
2222
2223             fscal            = felec;
2224
2225             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2226
2227             /* Calculate temporary vectorial force */
2228             tx               = _mm256_mul_pd(fscal,dx11);
2229             ty               = _mm256_mul_pd(fscal,dy11);
2230             tz               = _mm256_mul_pd(fscal,dz11);
2231
2232             /* Update vectorial force */
2233             fix1             = _mm256_add_pd(fix1,tx);
2234             fiy1             = _mm256_add_pd(fiy1,ty);
2235             fiz1             = _mm256_add_pd(fiz1,tz);
2236
2237             fjx1             = _mm256_add_pd(fjx1,tx);
2238             fjy1             = _mm256_add_pd(fjy1,ty);
2239             fjz1             = _mm256_add_pd(fjz1,tz);
2240
2241             /**************************
2242              * CALCULATE INTERACTIONS *
2243              **************************/
2244
2245             r12              = _mm256_mul_pd(rsq12,rinv12);
2246             r12              = _mm256_andnot_pd(dummy_mask,r12);
2247
2248             /* EWALD ELECTROSTATICS */
2249
2250             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2251             ewrt             = _mm256_mul_pd(r12,ewtabscale);
2252             ewitab           = _mm256_cvttpd_epi32(ewrt);
2253             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2254             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2255                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2256                                             &ewtabF,&ewtabFn);
2257             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2258             felec            = _mm256_mul_pd(_mm256_mul_pd(qq12,rinv12),_mm256_sub_pd(rinvsq12,felec));
2259
2260             fscal            = felec;
2261
2262             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2263
2264             /* Calculate temporary vectorial force */
2265             tx               = _mm256_mul_pd(fscal,dx12);
2266             ty               = _mm256_mul_pd(fscal,dy12);
2267             tz               = _mm256_mul_pd(fscal,dz12);
2268
2269             /* Update vectorial force */
2270             fix1             = _mm256_add_pd(fix1,tx);
2271             fiy1             = _mm256_add_pd(fiy1,ty);
2272             fiz1             = _mm256_add_pd(fiz1,tz);
2273
2274             fjx2             = _mm256_add_pd(fjx2,tx);
2275             fjy2             = _mm256_add_pd(fjy2,ty);
2276             fjz2             = _mm256_add_pd(fjz2,tz);
2277
2278             /**************************
2279              * CALCULATE INTERACTIONS *
2280              **************************/
2281
2282             r13              = _mm256_mul_pd(rsq13,rinv13);
2283             r13              = _mm256_andnot_pd(dummy_mask,r13);
2284
2285             /* EWALD ELECTROSTATICS */
2286
2287             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2288             ewrt             = _mm256_mul_pd(r13,ewtabscale);
2289             ewitab           = _mm256_cvttpd_epi32(ewrt);
2290             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2291             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2292                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2293                                             &ewtabF,&ewtabFn);
2294             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2295             felec            = _mm256_mul_pd(_mm256_mul_pd(qq13,rinv13),_mm256_sub_pd(rinvsq13,felec));
2296
2297             fscal            = felec;
2298
2299             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2300
2301             /* Calculate temporary vectorial force */
2302             tx               = _mm256_mul_pd(fscal,dx13);
2303             ty               = _mm256_mul_pd(fscal,dy13);
2304             tz               = _mm256_mul_pd(fscal,dz13);
2305
2306             /* Update vectorial force */
2307             fix1             = _mm256_add_pd(fix1,tx);
2308             fiy1             = _mm256_add_pd(fiy1,ty);
2309             fiz1             = _mm256_add_pd(fiz1,tz);
2310
2311             fjx3             = _mm256_add_pd(fjx3,tx);
2312             fjy3             = _mm256_add_pd(fjy3,ty);
2313             fjz3             = _mm256_add_pd(fjz3,tz);
2314
2315             /**************************
2316              * CALCULATE INTERACTIONS *
2317              **************************/
2318
2319             r21              = _mm256_mul_pd(rsq21,rinv21);
2320             r21              = _mm256_andnot_pd(dummy_mask,r21);
2321
2322             /* EWALD ELECTROSTATICS */
2323
2324             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2325             ewrt             = _mm256_mul_pd(r21,ewtabscale);
2326             ewitab           = _mm256_cvttpd_epi32(ewrt);
2327             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2328             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2329                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2330                                             &ewtabF,&ewtabFn);
2331             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2332             felec            = _mm256_mul_pd(_mm256_mul_pd(qq21,rinv21),_mm256_sub_pd(rinvsq21,felec));
2333
2334             fscal            = felec;
2335
2336             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2337
2338             /* Calculate temporary vectorial force */
2339             tx               = _mm256_mul_pd(fscal,dx21);
2340             ty               = _mm256_mul_pd(fscal,dy21);
2341             tz               = _mm256_mul_pd(fscal,dz21);
2342
2343             /* Update vectorial force */
2344             fix2             = _mm256_add_pd(fix2,tx);
2345             fiy2             = _mm256_add_pd(fiy2,ty);
2346             fiz2             = _mm256_add_pd(fiz2,tz);
2347
2348             fjx1             = _mm256_add_pd(fjx1,tx);
2349             fjy1             = _mm256_add_pd(fjy1,ty);
2350             fjz1             = _mm256_add_pd(fjz1,tz);
2351
2352             /**************************
2353              * CALCULATE INTERACTIONS *
2354              **************************/
2355
2356             r22              = _mm256_mul_pd(rsq22,rinv22);
2357             r22              = _mm256_andnot_pd(dummy_mask,r22);
2358
2359             /* EWALD ELECTROSTATICS */
2360
2361             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2362             ewrt             = _mm256_mul_pd(r22,ewtabscale);
2363             ewitab           = _mm256_cvttpd_epi32(ewrt);
2364             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2365             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2366                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2367                                             &ewtabF,&ewtabFn);
2368             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2369             felec            = _mm256_mul_pd(_mm256_mul_pd(qq22,rinv22),_mm256_sub_pd(rinvsq22,felec));
2370
2371             fscal            = felec;
2372
2373             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2374
2375             /* Calculate temporary vectorial force */
2376             tx               = _mm256_mul_pd(fscal,dx22);
2377             ty               = _mm256_mul_pd(fscal,dy22);
2378             tz               = _mm256_mul_pd(fscal,dz22);
2379
2380             /* Update vectorial force */
2381             fix2             = _mm256_add_pd(fix2,tx);
2382             fiy2             = _mm256_add_pd(fiy2,ty);
2383             fiz2             = _mm256_add_pd(fiz2,tz);
2384
2385             fjx2             = _mm256_add_pd(fjx2,tx);
2386             fjy2             = _mm256_add_pd(fjy2,ty);
2387             fjz2             = _mm256_add_pd(fjz2,tz);
2388
2389             /**************************
2390              * CALCULATE INTERACTIONS *
2391              **************************/
2392
2393             r23              = _mm256_mul_pd(rsq23,rinv23);
2394             r23              = _mm256_andnot_pd(dummy_mask,r23);
2395
2396             /* EWALD ELECTROSTATICS */
2397
2398             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2399             ewrt             = _mm256_mul_pd(r23,ewtabscale);
2400             ewitab           = _mm256_cvttpd_epi32(ewrt);
2401             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2402             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2403                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2404                                             &ewtabF,&ewtabFn);
2405             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2406             felec            = _mm256_mul_pd(_mm256_mul_pd(qq23,rinv23),_mm256_sub_pd(rinvsq23,felec));
2407
2408             fscal            = felec;
2409
2410             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2411
2412             /* Calculate temporary vectorial force */
2413             tx               = _mm256_mul_pd(fscal,dx23);
2414             ty               = _mm256_mul_pd(fscal,dy23);
2415             tz               = _mm256_mul_pd(fscal,dz23);
2416
2417             /* Update vectorial force */
2418             fix2             = _mm256_add_pd(fix2,tx);
2419             fiy2             = _mm256_add_pd(fiy2,ty);
2420             fiz2             = _mm256_add_pd(fiz2,tz);
2421
2422             fjx3             = _mm256_add_pd(fjx3,tx);
2423             fjy3             = _mm256_add_pd(fjy3,ty);
2424             fjz3             = _mm256_add_pd(fjz3,tz);
2425
2426             /**************************
2427              * CALCULATE INTERACTIONS *
2428              **************************/
2429
2430             r31              = _mm256_mul_pd(rsq31,rinv31);
2431             r31              = _mm256_andnot_pd(dummy_mask,r31);
2432
2433             /* EWALD ELECTROSTATICS */
2434
2435             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2436             ewrt             = _mm256_mul_pd(r31,ewtabscale);
2437             ewitab           = _mm256_cvttpd_epi32(ewrt);
2438             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2439             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2440                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2441                                             &ewtabF,&ewtabFn);
2442             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2443             felec            = _mm256_mul_pd(_mm256_mul_pd(qq31,rinv31),_mm256_sub_pd(rinvsq31,felec));
2444
2445             fscal            = felec;
2446
2447             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2448
2449             /* Calculate temporary vectorial force */
2450             tx               = _mm256_mul_pd(fscal,dx31);
2451             ty               = _mm256_mul_pd(fscal,dy31);
2452             tz               = _mm256_mul_pd(fscal,dz31);
2453
2454             /* Update vectorial force */
2455             fix3             = _mm256_add_pd(fix3,tx);
2456             fiy3             = _mm256_add_pd(fiy3,ty);
2457             fiz3             = _mm256_add_pd(fiz3,tz);
2458
2459             fjx1             = _mm256_add_pd(fjx1,tx);
2460             fjy1             = _mm256_add_pd(fjy1,ty);
2461             fjz1             = _mm256_add_pd(fjz1,tz);
2462
2463             /**************************
2464              * CALCULATE INTERACTIONS *
2465              **************************/
2466
2467             r32              = _mm256_mul_pd(rsq32,rinv32);
2468             r32              = _mm256_andnot_pd(dummy_mask,r32);
2469
2470             /* EWALD ELECTROSTATICS */
2471
2472             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2473             ewrt             = _mm256_mul_pd(r32,ewtabscale);
2474             ewitab           = _mm256_cvttpd_epi32(ewrt);
2475             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2476             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2477                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2478                                             &ewtabF,&ewtabFn);
2479             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2480             felec            = _mm256_mul_pd(_mm256_mul_pd(qq32,rinv32),_mm256_sub_pd(rinvsq32,felec));
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,dx32);
2488             ty               = _mm256_mul_pd(fscal,dy32);
2489             tz               = _mm256_mul_pd(fscal,dz32);
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             fjx2             = _mm256_add_pd(fjx2,tx);
2497             fjy2             = _mm256_add_pd(fjy2,ty);
2498             fjz2             = _mm256_add_pd(fjz2,tz);
2499
2500             /**************************
2501              * CALCULATE INTERACTIONS *
2502              **************************/
2503
2504             r33              = _mm256_mul_pd(rsq33,rinv33);
2505             r33              = _mm256_andnot_pd(dummy_mask,r33);
2506
2507             /* EWALD ELECTROSTATICS */
2508
2509             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2510             ewrt             = _mm256_mul_pd(r33,ewtabscale);
2511             ewitab           = _mm256_cvttpd_epi32(ewrt);
2512             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
2513             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
2514                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
2515                                             &ewtabF,&ewtabFn);
2516             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
2517             felec            = _mm256_mul_pd(_mm256_mul_pd(qq33,rinv33),_mm256_sub_pd(rinvsq33,felec));
2518
2519             fscal            = felec;
2520
2521             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2522
2523             /* Calculate temporary vectorial force */
2524             tx               = _mm256_mul_pd(fscal,dx33);
2525             ty               = _mm256_mul_pd(fscal,dy33);
2526             tz               = _mm256_mul_pd(fscal,dz33);
2527
2528             /* Update vectorial force */
2529             fix3             = _mm256_add_pd(fix3,tx);
2530             fiy3             = _mm256_add_pd(fiy3,ty);
2531             fiz3             = _mm256_add_pd(fiz3,tz);
2532
2533             fjx3             = _mm256_add_pd(fjx3,tx);
2534             fjy3             = _mm256_add_pd(fjy3,ty);
2535             fjz3             = _mm256_add_pd(fjz3,tz);
2536
2537             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2538             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2539             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2540             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2541
2542             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2543                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2544                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2545
2546             /* Inner loop uses 385 flops */
2547         }
2548
2549         /* End of innermost loop */
2550
2551         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2552                                                  f+i_coord_offset,fshift+i_shift_offset);
2553
2554         /* Increment number of inner iterations */
2555         inneriter                  += j_index_end - j_index_start;
2556
2557         /* Outer loop uses 24 flops */
2558     }
2559
2560     /* Increment number of outer iterations */
2561     outeriter        += nri;
2562
2563     /* Update outer/inner flops */
2564
2565     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*385);
2566 }