src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_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_ElecCSTab_VdwLJ_GeomW3W3_VF_avx_256_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  77     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  78     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  79     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  80     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  81     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  82     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  83     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  84     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  85     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  86     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  87     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  88     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  89     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  90     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  98     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m256d          dummy_mask,cutoff_mask;
 104     __m128           tmpmask0,tmpmask1;
 105     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 106     __m256d          one     = _mm256_set1_pd(1.0);
 107     __m256d          two     = _mm256_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     vftab            = kernel_data->table_elec->data;
 126     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 131     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 132     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 133     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 134
 135     jq0              = _mm256_set1_pd(charge[inr+0]);
 136     jq1              = _mm256_set1_pd(charge[inr+1]);
 137     jq2              = _mm256_set1_pd(charge[inr+2]);
 138     vdwjidx0A        = 2*vdwtype[inr+0];
 139     qq00             = _mm256_mul_pd(iq0,jq0);
 140     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 141     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 142     qq01             = _mm256_mul_pd(iq0,jq1);
 143     qq02             = _mm256_mul_pd(iq0,jq2);
 144     qq10             = _mm256_mul_pd(iq1,jq0);
 145     qq11             = _mm256_mul_pd(iq1,jq1);
 146     qq12             = _mm256_mul_pd(iq1,jq2);
 147     qq20             = _mm256_mul_pd(iq2,jq0);
 148     qq21             = _mm256_mul_pd(iq2,jq1);
 149     qq22             = _mm256_mul_pd(iq2,jq2);
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 183
 184         fix0             = _mm256_setzero_pd();
 185         fiy0             = _mm256_setzero_pd();
 186         fiz0             = _mm256_setzero_pd();
 187         fix1             = _mm256_setzero_pd();
 188         fiy1             = _mm256_setzero_pd();
 189         fiz1             = _mm256_setzero_pd();
 190         fix2             = _mm256_setzero_pd();
 191         fiy2             = _mm256_setzero_pd();
 192         fiz2             = _mm256_setzero_pd();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_pd();
 196         vvdwsum          = _mm256_setzero_pd();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm256_sub_pd(ix0,jx0);
 219             dy00             = _mm256_sub_pd(iy0,jy0);
 220             dz00             = _mm256_sub_pd(iz0,jz0);
 221             dx01             = _mm256_sub_pd(ix0,jx1);
 222             dy01             = _mm256_sub_pd(iy0,jy1);
 223             dz01             = _mm256_sub_pd(iz0,jz1);
 224             dx02             = _mm256_sub_pd(ix0,jx2);
 225             dy02             = _mm256_sub_pd(iy0,jy2);
 226             dz02             = _mm256_sub_pd(iz0,jz2);
 227             dx10             = _mm256_sub_pd(ix1,jx0);
 228             dy10             = _mm256_sub_pd(iy1,jy0);
 229             dz10             = _mm256_sub_pd(iz1,jz0);
 230             dx11             = _mm256_sub_pd(ix1,jx1);
 231             dy11             = _mm256_sub_pd(iy1,jy1);
 232             dz11             = _mm256_sub_pd(iz1,jz1);
 233             dx12             = _mm256_sub_pd(ix1,jx2);
 234             dy12             = _mm256_sub_pd(iy1,jy2);
 235             dz12             = _mm256_sub_pd(iz1,jz2);
 236             dx20             = _mm256_sub_pd(ix2,jx0);
 237             dy20             = _mm256_sub_pd(iy2,jy0);
 238             dz20             = _mm256_sub_pd(iz2,jz0);
 239             dx21             = _mm256_sub_pd(ix2,jx1);
 240             dy21             = _mm256_sub_pd(iy2,jy1);
 241             dz21             = _mm256_sub_pd(iz2,jz1);
 242             dx22             = _mm256_sub_pd(ix2,jx2);
 243             dy22             = _mm256_sub_pd(iy2,jy2);
 244             dz22             = _mm256_sub_pd(iz2,jz2);
 245
 246             /* Calculate squared distance and things based on it */
 247             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 248             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 249             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 250             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 251             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 252             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 253             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 254             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 255             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 256
 257             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 258             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 259             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 260             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 261             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 262             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 263             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 264             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 265             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 266
 267             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 268
 269             fjx0             = _mm256_setzero_pd();
 270             fjy0             = _mm256_setzero_pd();
 271             fjz0             = _mm256_setzero_pd();
 272             fjx1             = _mm256_setzero_pd();
 273             fjy1             = _mm256_setzero_pd();
 274             fjz1             = _mm256_setzero_pd();
 275             fjx2             = _mm256_setzero_pd();
 276             fjy2             = _mm256_setzero_pd();
 277             fjz2             = _mm256_setzero_pd();
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r00              = _mm256_mul_pd(rsq00,rinv00);
 284
 285             /* Calculate table index by multiplying r with table scale and truncate to integer */
 286             rt               = _mm256_mul_pd(r00,vftabscale);
 287             vfitab           = _mm256_cvttpd_epi32(rt);
 288             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 289             vfitab           = _mm_slli_epi32(vfitab,2);
 290
 291             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 292             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 293             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 294             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 295             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 296             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 297             Heps             = _mm256_mul_pd(vfeps,H);
 298             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 299             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 300             velec            = _mm256_mul_pd(qq00,VV);
 301             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 302             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 303
 304             /* LENNARD-JONES DISPERSION/REPULSION */
 305
 306             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 307             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 308             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 309             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 310             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm256_add_pd(velecsum,velec);
 314             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 315
 316             fscal            = _mm256_add_pd(felec,fvdw);
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm256_mul_pd(fscal,dx00);
 320             ty               = _mm256_mul_pd(fscal,dy00);
 321             tz               = _mm256_mul_pd(fscal,dz00);
 322
 323             /* Update vectorial force */
 324             fix0             = _mm256_add_pd(fix0,tx);
 325             fiy0             = _mm256_add_pd(fiy0,ty);
 326             fiz0             = _mm256_add_pd(fiz0,tz);
 327
 328             fjx0             = _mm256_add_pd(fjx0,tx);
 329             fjy0             = _mm256_add_pd(fjy0,ty);
 330             fjz0             = _mm256_add_pd(fjz0,tz);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             r01              = _mm256_mul_pd(rsq01,rinv01);
 337
 338             /* Calculate table index by multiplying r with table scale and truncate to integer */
 339             rt               = _mm256_mul_pd(r01,vftabscale);
 340             vfitab           = _mm256_cvttpd_epi32(rt);
 341             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 342             vfitab           = _mm_slli_epi32(vfitab,2);
 343
 344             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 345             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 346             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 347             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 348             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 349             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 350             Heps             = _mm256_mul_pd(vfeps,H);
 351             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 352             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 353             velec            = _mm256_mul_pd(qq01,VV);
 354             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 355             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velecsum         = _mm256_add_pd(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm256_mul_pd(fscal,dx01);
 364             ty               = _mm256_mul_pd(fscal,dy01);
 365             tz               = _mm256_mul_pd(fscal,dz01);
 366
 367             /* Update vectorial force */
 368             fix0             = _mm256_add_pd(fix0,tx);
 369             fiy0             = _mm256_add_pd(fiy0,ty);
 370             fiz0             = _mm256_add_pd(fiz0,tz);
 371
 372             fjx1             = _mm256_add_pd(fjx1,tx);
 373             fjy1             = _mm256_add_pd(fjy1,ty);
 374             fjz1             = _mm256_add_pd(fjz1,tz);
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             r02              = _mm256_mul_pd(rsq02,rinv02);
 381
 382             /* Calculate table index by multiplying r with table scale and truncate to integer */
 383             rt               = _mm256_mul_pd(r02,vftabscale);
 384             vfitab           = _mm256_cvttpd_epi32(rt);
 385             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 386             vfitab           = _mm_slli_epi32(vfitab,2);
 387
 388             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 389             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 390             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 391             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 392             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 393             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 394             Heps             = _mm256_mul_pd(vfeps,H);
 395             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 396             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 397             velec            = _mm256_mul_pd(qq02,VV);
 398             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 399             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             velecsum         = _mm256_add_pd(velecsum,velec);
 403
 404             fscal            = felec;
 405
 406             /* Calculate temporary vectorial force */
 407             tx               = _mm256_mul_pd(fscal,dx02);
 408             ty               = _mm256_mul_pd(fscal,dy02);
 409             tz               = _mm256_mul_pd(fscal,dz02);
 410
 411             /* Update vectorial force */
 412             fix0             = _mm256_add_pd(fix0,tx);
 413             fiy0             = _mm256_add_pd(fiy0,ty);
 414             fiz0             = _mm256_add_pd(fiz0,tz);
 415
 416             fjx2             = _mm256_add_pd(fjx2,tx);
 417             fjy2             = _mm256_add_pd(fjy2,ty);
 418             fjz2             = _mm256_add_pd(fjz2,tz);
 419
 420             /**************************
 421              * CALCULATE INTERACTIONS *
 422              **************************/
 423
 424             r10              = _mm256_mul_pd(rsq10,rinv10);
 425
 426             /* Calculate table index by multiplying r with table scale and truncate to integer */
 427             rt               = _mm256_mul_pd(r10,vftabscale);
 428             vfitab           = _mm256_cvttpd_epi32(rt);
 429             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 430             vfitab           = _mm_slli_epi32(vfitab,2);
 431
 432             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 433             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 434             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 435             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 436             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 437             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 438             Heps             = _mm256_mul_pd(vfeps,H);
 439             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 440             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 441             velec            = _mm256_mul_pd(qq10,VV);
 442             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 443             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 444
 445             /* Update potential sum for this i atom from the interaction with this j atom. */
 446             velecsum         = _mm256_add_pd(velecsum,velec);
 447
 448             fscal            = felec;
 449
 450             /* Calculate temporary vectorial force */
 451             tx               = _mm256_mul_pd(fscal,dx10);
 452             ty               = _mm256_mul_pd(fscal,dy10);
 453             tz               = _mm256_mul_pd(fscal,dz10);
 454
 455             /* Update vectorial force */
 456             fix1             = _mm256_add_pd(fix1,tx);
 457             fiy1             = _mm256_add_pd(fiy1,ty);
 458             fiz1             = _mm256_add_pd(fiz1,tz);
 459
 460             fjx0             = _mm256_add_pd(fjx0,tx);
 461             fjy0             = _mm256_add_pd(fjy0,ty);
 462             fjz0             = _mm256_add_pd(fjz0,tz);
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             r11              = _mm256_mul_pd(rsq11,rinv11);
 469
 470             /* Calculate table index by multiplying r with table scale and truncate to integer */
 471             rt               = _mm256_mul_pd(r11,vftabscale);
 472             vfitab           = _mm256_cvttpd_epi32(rt);
 473             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 474             vfitab           = _mm_slli_epi32(vfitab,2);
 475
 476             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 477             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 478             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 479             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 480             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 481             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 482             Heps             = _mm256_mul_pd(vfeps,H);
 483             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 484             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 485             velec            = _mm256_mul_pd(qq11,VV);
 486             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 487             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velecsum         = _mm256_add_pd(velecsum,velec);
 491
 492             fscal            = felec;
 493
 494             /* Calculate temporary vectorial force */
 495             tx               = _mm256_mul_pd(fscal,dx11);
 496             ty               = _mm256_mul_pd(fscal,dy11);
 497             tz               = _mm256_mul_pd(fscal,dz11);
 498
 499             /* Update vectorial force */
 500             fix1             = _mm256_add_pd(fix1,tx);
 501             fiy1             = _mm256_add_pd(fiy1,ty);
 502             fiz1             = _mm256_add_pd(fiz1,tz);
 503
 504             fjx1             = _mm256_add_pd(fjx1,tx);
 505             fjy1             = _mm256_add_pd(fjy1,ty);
 506             fjz1             = _mm256_add_pd(fjz1,tz);
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             r12              = _mm256_mul_pd(rsq12,rinv12);
 513
 514             /* Calculate table index by multiplying r with table scale and truncate to integer */
 515             rt               = _mm256_mul_pd(r12,vftabscale);
 516             vfitab           = _mm256_cvttpd_epi32(rt);
 517             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 518             vfitab           = _mm_slli_epi32(vfitab,2);
 519
 520             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 521             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 522             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 523             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 524             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 525             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 526             Heps             = _mm256_mul_pd(vfeps,H);
 527             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 528             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 529             velec            = _mm256_mul_pd(qq12,VV);
 530             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 531             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             velecsum         = _mm256_add_pd(velecsum,velec);
 535
 536             fscal            = felec;
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm256_mul_pd(fscal,dx12);
 540             ty               = _mm256_mul_pd(fscal,dy12);
 541             tz               = _mm256_mul_pd(fscal,dz12);
 542
 543             /* Update vectorial force */
 544             fix1             = _mm256_add_pd(fix1,tx);
 545             fiy1             = _mm256_add_pd(fiy1,ty);
 546             fiz1             = _mm256_add_pd(fiz1,tz);
 547
 548             fjx2             = _mm256_add_pd(fjx2,tx);
 549             fjy2             = _mm256_add_pd(fjy2,ty);
 550             fjz2             = _mm256_add_pd(fjz2,tz);
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             r20              = _mm256_mul_pd(rsq20,rinv20);
 557
 558             /* Calculate table index by multiplying r with table scale and truncate to integer */
 559             rt               = _mm256_mul_pd(r20,vftabscale);
 560             vfitab           = _mm256_cvttpd_epi32(rt);
 561             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 562             vfitab           = _mm_slli_epi32(vfitab,2);
 563
 564             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 565             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 566             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 567             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 568             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 569             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 570             Heps             = _mm256_mul_pd(vfeps,H);
 571             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 572             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 573             velec            = _mm256_mul_pd(qq20,VV);
 574             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 575             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velecsum         = _mm256_add_pd(velecsum,velec);
 579
 580             fscal            = felec;
 581
 582             /* Calculate temporary vectorial force */
 583             tx               = _mm256_mul_pd(fscal,dx20);
 584             ty               = _mm256_mul_pd(fscal,dy20);
 585             tz               = _mm256_mul_pd(fscal,dz20);
 586
 587             /* Update vectorial force */
 588             fix2             = _mm256_add_pd(fix2,tx);
 589             fiy2             = _mm256_add_pd(fiy2,ty);
 590             fiz2             = _mm256_add_pd(fiz2,tz);
 591
 592             fjx0             = _mm256_add_pd(fjx0,tx);
 593             fjy0             = _mm256_add_pd(fjy0,ty);
 594             fjz0             = _mm256_add_pd(fjz0,tz);
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             r21              = _mm256_mul_pd(rsq21,rinv21);
 601
 602             /* Calculate table index by multiplying r with table scale and truncate to integer */
 603             rt               = _mm256_mul_pd(r21,vftabscale);
 604             vfitab           = _mm256_cvttpd_epi32(rt);
 605             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 606             vfitab           = _mm_slli_epi32(vfitab,2);
 607
 608             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 609             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 610             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 611             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 612             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 613             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 614             Heps             = _mm256_mul_pd(vfeps,H);
 615             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 616             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 617             velec            = _mm256_mul_pd(qq21,VV);
 618             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 619             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velecsum         = _mm256_add_pd(velecsum,velec);
 623
 624             fscal            = felec;
 625
 626             /* Calculate temporary vectorial force */
 627             tx               = _mm256_mul_pd(fscal,dx21);
 628             ty               = _mm256_mul_pd(fscal,dy21);
 629             tz               = _mm256_mul_pd(fscal,dz21);
 630
 631             /* Update vectorial force */
 632             fix2             = _mm256_add_pd(fix2,tx);
 633             fiy2             = _mm256_add_pd(fiy2,ty);
 634             fiz2             = _mm256_add_pd(fiz2,tz);
 635
 636             fjx1             = _mm256_add_pd(fjx1,tx);
 637             fjy1             = _mm256_add_pd(fjy1,ty);
 638             fjz1             = _mm256_add_pd(fjz1,tz);
 639
 640             /**************************
 641              * CALCULATE INTERACTIONS *
 642              **************************/
 643
 644             r22              = _mm256_mul_pd(rsq22,rinv22);
 645
 646             /* Calculate table index by multiplying r with table scale and truncate to integer */
 647             rt               = _mm256_mul_pd(r22,vftabscale);
 648             vfitab           = _mm256_cvttpd_epi32(rt);
 649             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 650             vfitab           = _mm_slli_epi32(vfitab,2);
 651
 652             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 653             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 654             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 655             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 656             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 657             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 658             Heps             = _mm256_mul_pd(vfeps,H);
 659             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 660             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 661             velec            = _mm256_mul_pd(qq22,VV);
 662             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 663             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
 664
 665             /* Update potential sum for this i atom from the interaction with this j atom. */
 666             velecsum         = _mm256_add_pd(velecsum,velec);
 667
 668             fscal            = felec;
 669
 670             /* Calculate temporary vectorial force */
 671             tx               = _mm256_mul_pd(fscal,dx22);
 672             ty               = _mm256_mul_pd(fscal,dy22);
 673             tz               = _mm256_mul_pd(fscal,dz22);
 674
 675             /* Update vectorial force */
 676             fix2             = _mm256_add_pd(fix2,tx);
 677             fiy2             = _mm256_add_pd(fiy2,ty);
 678             fiz2             = _mm256_add_pd(fiz2,tz);
 679
 680             fjx2             = _mm256_add_pd(fjx2,tx);
 681             fjy2             = _mm256_add_pd(fjy2,ty);
 682             fjz2             = _mm256_add_pd(fjz2,tz);
 683
 684             fjptrA             = f+j_coord_offsetA;
 685             fjptrB             = f+j_coord_offsetB;
 686             fjptrC             = f+j_coord_offsetC;
 687             fjptrD             = f+j_coord_offsetD;
 688
 689             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 690                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 691
 692             /* Inner loop uses 400 flops */
 693         }
 694
 695         if(jidx<j_index_end)
 696         {
 697
 698             /* Get j neighbor index, and coordinate index */
 699             jnrlistA         = jjnr[jidx];
 700             jnrlistB         = jjnr[jidx+1];
 701             jnrlistC         = jjnr[jidx+2];
 702             jnrlistD         = jjnr[jidx+3];
 703             /* Sign of each element will be negative for non-real atoms.
 704              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 705              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 706              */
 707             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 708
 709             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 710             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 711             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 712
 713             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 714             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 715             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 716             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 717             j_coord_offsetA  = DIM*jnrA;
 718             j_coord_offsetB  = DIM*jnrB;
 719             j_coord_offsetC  = DIM*jnrC;
 720             j_coord_offsetD  = DIM*jnrD;
 721
 722             /* load j atom coordinates */
 723             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 724                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 725                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 726
 727             /* Calculate displacement vector */
 728             dx00             = _mm256_sub_pd(ix0,jx0);
 729             dy00             = _mm256_sub_pd(iy0,jy0);
 730             dz00             = _mm256_sub_pd(iz0,jz0);
 731             dx01             = _mm256_sub_pd(ix0,jx1);
 732             dy01             = _mm256_sub_pd(iy0,jy1);
 733             dz01             = _mm256_sub_pd(iz0,jz1);
 734             dx02             = _mm256_sub_pd(ix0,jx2);
 735             dy02             = _mm256_sub_pd(iy0,jy2);
 736             dz02             = _mm256_sub_pd(iz0,jz2);
 737             dx10             = _mm256_sub_pd(ix1,jx0);
 738             dy10             = _mm256_sub_pd(iy1,jy0);
 739             dz10             = _mm256_sub_pd(iz1,jz0);
 740             dx11             = _mm256_sub_pd(ix1,jx1);
 741             dy11             = _mm256_sub_pd(iy1,jy1);
 742             dz11             = _mm256_sub_pd(iz1,jz1);
 743             dx12             = _mm256_sub_pd(ix1,jx2);
 744             dy12             = _mm256_sub_pd(iy1,jy2);
 745             dz12             = _mm256_sub_pd(iz1,jz2);
 746             dx20             = _mm256_sub_pd(ix2,jx0);
 747             dy20             = _mm256_sub_pd(iy2,jy0);
 748             dz20             = _mm256_sub_pd(iz2,jz0);
 749             dx21             = _mm256_sub_pd(ix2,jx1);
 750             dy21             = _mm256_sub_pd(iy2,jy1);
 751             dz21             = _mm256_sub_pd(iz2,jz1);
 752             dx22             = _mm256_sub_pd(ix2,jx2);
 753             dy22             = _mm256_sub_pd(iy2,jy2);
 754             dz22             = _mm256_sub_pd(iz2,jz2);
 755
 756             /* Calculate squared distance and things based on it */
 757             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 758             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 759             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 760             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 761             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 762             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 763             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 764             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 765             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 766
 767             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 768             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 769             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 770             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 771             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 772             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 773             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 774             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 775             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 776
 777             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 778
 779             fjx0             = _mm256_setzero_pd();
 780             fjy0             = _mm256_setzero_pd();
 781             fjz0             = _mm256_setzero_pd();
 782             fjx1             = _mm256_setzero_pd();
 783             fjy1             = _mm256_setzero_pd();
 784             fjz1             = _mm256_setzero_pd();
 785             fjx2             = _mm256_setzero_pd();
 786             fjy2             = _mm256_setzero_pd();
 787             fjz2             = _mm256_setzero_pd();
 788
 789             /**************************
 790              * CALCULATE INTERACTIONS *
 791              **************************/
 792
 793             r00              = _mm256_mul_pd(rsq00,rinv00);
 794             r00              = _mm256_andnot_pd(dummy_mask,r00);
 795
 796             /* Calculate table index by multiplying r with table scale and truncate to integer */
 797             rt               = _mm256_mul_pd(r00,vftabscale);
 798             vfitab           = _mm256_cvttpd_epi32(rt);
 799             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 800             vfitab           = _mm_slli_epi32(vfitab,2);
 801
 802             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 803             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 804             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 805             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 806             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 807             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 808             Heps             = _mm256_mul_pd(vfeps,H);
 809             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 810             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 811             velec            = _mm256_mul_pd(qq00,VV);
 812             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 813             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 814
 815             /* LENNARD-JONES DISPERSION/REPULSION */
 816
 817             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 818             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 819             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 820             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 821             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 822
 823             /* Update potential sum for this i atom from the interaction with this j atom. */
 824             velec            = _mm256_andnot_pd(dummy_mask,velec);
 825             velecsum         = _mm256_add_pd(velecsum,velec);
 826             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 827             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 828
 829             fscal            = _mm256_add_pd(felec,fvdw);
 830
 831             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 832
 833             /* Calculate temporary vectorial force */
 834             tx               = _mm256_mul_pd(fscal,dx00);
 835             ty               = _mm256_mul_pd(fscal,dy00);
 836             tz               = _mm256_mul_pd(fscal,dz00);
 837
 838             /* Update vectorial force */
 839             fix0             = _mm256_add_pd(fix0,tx);
 840             fiy0             = _mm256_add_pd(fiy0,ty);
 841             fiz0             = _mm256_add_pd(fiz0,tz);
 842
 843             fjx0             = _mm256_add_pd(fjx0,tx);
 844             fjy0             = _mm256_add_pd(fjy0,ty);
 845             fjz0             = _mm256_add_pd(fjz0,tz);
 846
 847             /**************************
 848              * CALCULATE INTERACTIONS *
 849              **************************/
 850
 851             r01              = _mm256_mul_pd(rsq01,rinv01);
 852             r01              = _mm256_andnot_pd(dummy_mask,r01);
 853
 854             /* Calculate table index by multiplying r with table scale and truncate to integer */
 855             rt               = _mm256_mul_pd(r01,vftabscale);
 856             vfitab           = _mm256_cvttpd_epi32(rt);
 857             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 858             vfitab           = _mm_slli_epi32(vfitab,2);
 859
 860             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 861             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 862             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 863             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 864             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 865             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 866             Heps             = _mm256_mul_pd(vfeps,H);
 867             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 868             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 869             velec            = _mm256_mul_pd(qq01,VV);
 870             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 871             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
 872
 873             /* Update potential sum for this i atom from the interaction with this j atom. */
 874             velec            = _mm256_andnot_pd(dummy_mask,velec);
 875             velecsum         = _mm256_add_pd(velecsum,velec);
 876
 877             fscal            = felec;
 878
 879             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 880
 881             /* Calculate temporary vectorial force */
 882             tx               = _mm256_mul_pd(fscal,dx01);
 883             ty               = _mm256_mul_pd(fscal,dy01);
 884             tz               = _mm256_mul_pd(fscal,dz01);
 885
 886             /* Update vectorial force */
 887             fix0             = _mm256_add_pd(fix0,tx);
 888             fiy0             = _mm256_add_pd(fiy0,ty);
 889             fiz0             = _mm256_add_pd(fiz0,tz);
 890
 891             fjx1             = _mm256_add_pd(fjx1,tx);
 892             fjy1             = _mm256_add_pd(fjy1,ty);
 893             fjz1             = _mm256_add_pd(fjz1,tz);
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             r02              = _mm256_mul_pd(rsq02,rinv02);
 900             r02              = _mm256_andnot_pd(dummy_mask,r02);
 901
 902             /* Calculate table index by multiplying r with table scale and truncate to integer */
 903             rt               = _mm256_mul_pd(r02,vftabscale);
 904             vfitab           = _mm256_cvttpd_epi32(rt);
 905             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 906             vfitab           = _mm_slli_epi32(vfitab,2);
 907
 908             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 909             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 910             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 911             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 912             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 913             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 914             Heps             = _mm256_mul_pd(vfeps,H);
 915             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 916             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 917             velec            = _mm256_mul_pd(qq02,VV);
 918             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 919             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
 920
 921             /* Update potential sum for this i atom from the interaction with this j atom. */
 922             velec            = _mm256_andnot_pd(dummy_mask,velec);
 923             velecsum         = _mm256_add_pd(velecsum,velec);
 924
 925             fscal            = felec;
 926
 927             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 928
 929             /* Calculate temporary vectorial force */
 930             tx               = _mm256_mul_pd(fscal,dx02);
 931             ty               = _mm256_mul_pd(fscal,dy02);
 932             tz               = _mm256_mul_pd(fscal,dz02);
 933
 934             /* Update vectorial force */
 935             fix0             = _mm256_add_pd(fix0,tx);
 936             fiy0             = _mm256_add_pd(fiy0,ty);
 937             fiz0             = _mm256_add_pd(fiz0,tz);
 938
 939             fjx2             = _mm256_add_pd(fjx2,tx);
 940             fjy2             = _mm256_add_pd(fjy2,ty);
 941             fjz2             = _mm256_add_pd(fjz2,tz);
 942
 943             /**************************
 944              * CALCULATE INTERACTIONS *
 945              **************************/
 946
 947             r10              = _mm256_mul_pd(rsq10,rinv10);
 948             r10              = _mm256_andnot_pd(dummy_mask,r10);
 949
 950             /* Calculate table index by multiplying r with table scale and truncate to integer */
 951             rt               = _mm256_mul_pd(r10,vftabscale);
 952             vfitab           = _mm256_cvttpd_epi32(rt);
 953             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 954             vfitab           = _mm_slli_epi32(vfitab,2);
 955
 956             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 957             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 958             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 959             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 960             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 961             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 962             Heps             = _mm256_mul_pd(vfeps,H);
 963             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 964             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 965             velec            = _mm256_mul_pd(qq10,VV);
 966             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 967             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 968
 969             /* Update potential sum for this i atom from the interaction with this j atom. */
 970             velec            = _mm256_andnot_pd(dummy_mask,velec);
 971             velecsum         = _mm256_add_pd(velecsum,velec);
 972
 973             fscal            = felec;
 974
 975             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 976
 977             /* Calculate temporary vectorial force */
 978             tx               = _mm256_mul_pd(fscal,dx10);
 979             ty               = _mm256_mul_pd(fscal,dy10);
 980             tz               = _mm256_mul_pd(fscal,dz10);
 981
 982             /* Update vectorial force */
 983             fix1             = _mm256_add_pd(fix1,tx);
 984             fiy1             = _mm256_add_pd(fiy1,ty);
 985             fiz1             = _mm256_add_pd(fiz1,tz);
 986
 987             fjx0             = _mm256_add_pd(fjx0,tx);
 988             fjy0             = _mm256_add_pd(fjy0,ty);
 989             fjz0             = _mm256_add_pd(fjz0,tz);
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             r11              = _mm256_mul_pd(rsq11,rinv11);
 996             r11              = _mm256_andnot_pd(dummy_mask,r11);
 997
 998             /* Calculate table index by multiplying r with table scale and truncate to integer */
 999             rt               = _mm256_mul_pd(r11,vftabscale);
1000             vfitab           = _mm256_cvttpd_epi32(rt);
1001             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1002             vfitab           = _mm_slli_epi32(vfitab,2);
1003
1004             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1005             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1006             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1007             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1008             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1009             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1010             Heps             = _mm256_mul_pd(vfeps,H);
1011             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1012             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1013             velec            = _mm256_mul_pd(qq11,VV);
1014             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1015             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1016
1017             /* Update potential sum for this i atom from the interaction with this j atom. */
1018             velec            = _mm256_andnot_pd(dummy_mask,velec);
1019             velecsum         = _mm256_add_pd(velecsum,velec);
1020
1021             fscal            = felec;
1022
1023             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1024
1025             /* Calculate temporary vectorial force */
1026             tx               = _mm256_mul_pd(fscal,dx11);
1027             ty               = _mm256_mul_pd(fscal,dy11);
1028             tz               = _mm256_mul_pd(fscal,dz11);
1029
1030             /* Update vectorial force */
1031             fix1             = _mm256_add_pd(fix1,tx);
1032             fiy1             = _mm256_add_pd(fiy1,ty);
1033             fiz1             = _mm256_add_pd(fiz1,tz);
1034
1035             fjx1             = _mm256_add_pd(fjx1,tx);
1036             fjy1             = _mm256_add_pd(fjy1,ty);
1037             fjz1             = _mm256_add_pd(fjz1,tz);
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             r12              = _mm256_mul_pd(rsq12,rinv12);
1044             r12              = _mm256_andnot_pd(dummy_mask,r12);
1045
1046             /* Calculate table index by multiplying r with table scale and truncate to integer */
1047             rt               = _mm256_mul_pd(r12,vftabscale);
1048             vfitab           = _mm256_cvttpd_epi32(rt);
1049             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1050             vfitab           = _mm_slli_epi32(vfitab,2);
1051
1052             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1053             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1054             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1055             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1056             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1057             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1058             Heps             = _mm256_mul_pd(vfeps,H);
1059             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1060             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1061             velec            = _mm256_mul_pd(qq12,VV);
1062             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1063             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1064
1065             /* Update potential sum for this i atom from the interaction with this j atom. */
1066             velec            = _mm256_andnot_pd(dummy_mask,velec);
1067             velecsum         = _mm256_add_pd(velecsum,velec);
1068
1069             fscal            = felec;
1070
1071             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1072
1073             /* Calculate temporary vectorial force */
1074             tx               = _mm256_mul_pd(fscal,dx12);
1075             ty               = _mm256_mul_pd(fscal,dy12);
1076             tz               = _mm256_mul_pd(fscal,dz12);
1077
1078             /* Update vectorial force */
1079             fix1             = _mm256_add_pd(fix1,tx);
1080             fiy1             = _mm256_add_pd(fiy1,ty);
1081             fiz1             = _mm256_add_pd(fiz1,tz);
1082
1083             fjx2             = _mm256_add_pd(fjx2,tx);
1084             fjy2             = _mm256_add_pd(fjy2,ty);
1085             fjz2             = _mm256_add_pd(fjz2,tz);
1086
1087             /**************************
1088              * CALCULATE INTERACTIONS *
1089              **************************/
1090
1091             r20              = _mm256_mul_pd(rsq20,rinv20);
1092             r20              = _mm256_andnot_pd(dummy_mask,r20);
1093
1094             /* Calculate table index by multiplying r with table scale and truncate to integer */
1095             rt               = _mm256_mul_pd(r20,vftabscale);
1096             vfitab           = _mm256_cvttpd_epi32(rt);
1097             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1098             vfitab           = _mm_slli_epi32(vfitab,2);
1099
1100             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1101             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1102             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1103             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1104             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1105             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1106             Heps             = _mm256_mul_pd(vfeps,H);
1107             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1108             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1109             velec            = _mm256_mul_pd(qq20,VV);
1110             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1111             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1112
1113             /* Update potential sum for this i atom from the interaction with this j atom. */
1114             velec            = _mm256_andnot_pd(dummy_mask,velec);
1115             velecsum         = _mm256_add_pd(velecsum,velec);
1116
1117             fscal            = felec;
1118
1119             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1120
1121             /* Calculate temporary vectorial force */
1122             tx               = _mm256_mul_pd(fscal,dx20);
1123             ty               = _mm256_mul_pd(fscal,dy20);
1124             tz               = _mm256_mul_pd(fscal,dz20);
1125
1126             /* Update vectorial force */
1127             fix2             = _mm256_add_pd(fix2,tx);
1128             fiy2             = _mm256_add_pd(fiy2,ty);
1129             fiz2             = _mm256_add_pd(fiz2,tz);
1130
1131             fjx0             = _mm256_add_pd(fjx0,tx);
1132             fjy0             = _mm256_add_pd(fjy0,ty);
1133             fjz0             = _mm256_add_pd(fjz0,tz);
1134
1135             /**************************
1136              * CALCULATE INTERACTIONS *
1137              **************************/
1138
1139             r21              = _mm256_mul_pd(rsq21,rinv21);
1140             r21              = _mm256_andnot_pd(dummy_mask,r21);
1141
1142             /* Calculate table index by multiplying r with table scale and truncate to integer */
1143             rt               = _mm256_mul_pd(r21,vftabscale);
1144             vfitab           = _mm256_cvttpd_epi32(rt);
1145             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1146             vfitab           = _mm_slli_epi32(vfitab,2);
1147
1148             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1149             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1150             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1151             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1152             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1153             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1154             Heps             = _mm256_mul_pd(vfeps,H);
1155             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1156             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1157             velec            = _mm256_mul_pd(qq21,VV);
1158             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1159             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1160
1161             /* Update potential sum for this i atom from the interaction with this j atom. */
1162             velec            = _mm256_andnot_pd(dummy_mask,velec);
1163             velecsum         = _mm256_add_pd(velecsum,velec);
1164
1165             fscal            = felec;
1166
1167             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1168
1169             /* Calculate temporary vectorial force */
1170             tx               = _mm256_mul_pd(fscal,dx21);
1171             ty               = _mm256_mul_pd(fscal,dy21);
1172             tz               = _mm256_mul_pd(fscal,dz21);
1173
1174             /* Update vectorial force */
1175             fix2             = _mm256_add_pd(fix2,tx);
1176             fiy2             = _mm256_add_pd(fiy2,ty);
1177             fiz2             = _mm256_add_pd(fiz2,tz);
1178
1179             fjx1             = _mm256_add_pd(fjx1,tx);
1180             fjy1             = _mm256_add_pd(fjy1,ty);
1181             fjz1             = _mm256_add_pd(fjz1,tz);
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             r22              = _mm256_mul_pd(rsq22,rinv22);
1188             r22              = _mm256_andnot_pd(dummy_mask,r22);
1189
1190             /* Calculate table index by multiplying r with table scale and truncate to integer */
1191             rt               = _mm256_mul_pd(r22,vftabscale);
1192             vfitab           = _mm256_cvttpd_epi32(rt);
1193             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1194             vfitab           = _mm_slli_epi32(vfitab,2);
1195
1196             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1197             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1198             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1199             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1200             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1201             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1202             Heps             = _mm256_mul_pd(vfeps,H);
1203             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1204             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1205             velec            = _mm256_mul_pd(qq22,VV);
1206             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1207             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1208
1209             /* Update potential sum for this i atom from the interaction with this j atom. */
1210             velec            = _mm256_andnot_pd(dummy_mask,velec);
1211             velecsum         = _mm256_add_pd(velecsum,velec);
1212
1213             fscal            = felec;
1214
1215             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1216
1217             /* Calculate temporary vectorial force */
1218             tx               = _mm256_mul_pd(fscal,dx22);
1219             ty               = _mm256_mul_pd(fscal,dy22);
1220             tz               = _mm256_mul_pd(fscal,dz22);
1221
1222             /* Update vectorial force */
1223             fix2             = _mm256_add_pd(fix2,tx);
1224             fiy2             = _mm256_add_pd(fiy2,ty);
1225             fiz2             = _mm256_add_pd(fiz2,tz);
1226
1227             fjx2             = _mm256_add_pd(fjx2,tx);
1228             fjy2             = _mm256_add_pd(fjy2,ty);
1229             fjz2             = _mm256_add_pd(fjz2,tz);
1230
1231             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1232             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1233             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1234             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1235
1236             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1237                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1238
1239             /* Inner loop uses 409 flops */
1240         }
1241
1242         /* End of innermost loop */
1243
1244         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1245                                                  f+i_coord_offset,fshift+i_shift_offset);
1246
1247         ggid                        = gid[iidx];
1248         /* Update potential energies */
1249         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1250         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1251
1252         /* Increment number of inner iterations */
1253         inneriter                  += j_index_end - j_index_start;
1254
1255         /* Outer loop uses 20 flops */
1256     }
1257
1258     /* Increment number of outer iterations */
1259     outeriter        += nri;
1260
1261     /* Update outer/inner flops */
1262
1263     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*409);
1264 }
1265 /*
1266  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_F_avx_256_double
1267  * Electrostatics interaction: CubicSplineTable
1268  * VdW interaction:            LennardJones
1269  * Geometry:                   Water3-Water3
1270  * Calculate force/pot:        Force
1271  */
1272 void
1273 nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_F_avx_256_double
1274                     (t_nblist * gmx_restrict                nlist,
1275                      rvec * gmx_restrict                    xx,
1276                      rvec * gmx_restrict                    ff,
1277                      t_forcerec * gmx_restrict              fr,
1278                      t_mdatoms * gmx_restrict               mdatoms,
1279                      nb_kernel_data_t * gmx_restrict        kernel_data,
1280                      t_nrnb * gmx_restrict                  nrnb)
1281 {
1282     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1283      * just 0 for non-waters.
1284      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1285      * jnr indices corresponding to data put in the four positions in the SIMD register.
1286      */
1287     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1288     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1289     int              jnrA,jnrB,jnrC,jnrD;
1290     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1291     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1292     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1293     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1294     real             rcutoff_scalar;
1295     real             *shiftvec,*fshift,*x,*f;
1296     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1297     real             scratch[4*DIM];
1298     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1299     real *           vdwioffsetptr0;
1300     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1301     real *           vdwioffsetptr1;
1302     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1303     real *           vdwioffsetptr2;
1304     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1305     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1306     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1307     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1308     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1309     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1310     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1311     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1312     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1313     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1314     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1315     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1316     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1317     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1318     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1319     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1320     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1321     real             *charge;
1322     int              nvdwtype;
1323     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1324     int              *vdwtype;
1325     real             *vdwparam;
1326     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1327     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1328     __m128i          vfitab;
1329     __m128i          ifour       = _mm_set1_epi32(4);
1330     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1331     real             *vftab;
1332     __m256d          dummy_mask,cutoff_mask;
1333     __m128           tmpmask0,tmpmask1;
1334     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1335     __m256d          one     = _mm256_set1_pd(1.0);
1336     __m256d          two     = _mm256_set1_pd(2.0);
1337     x                = xx[0];
1338     f                = ff[0];
1339
1340     nri              = nlist->nri;
1341     iinr             = nlist->iinr;
1342     jindex           = nlist->jindex;
1343     jjnr             = nlist->jjnr;
1344     shiftidx         = nlist->shift;
1345     gid              = nlist->gid;
1346     shiftvec         = fr->shift_vec[0];
1347     fshift           = fr->fshift[0];
1348     facel            = _mm256_set1_pd(fr->epsfac);
1349     charge           = mdatoms->chargeA;
1350     nvdwtype         = fr->ntype;
1351     vdwparam         = fr->nbfp;
1352     vdwtype          = mdatoms->typeA;
1353
1354     vftab            = kernel_data->table_elec->data;
1355     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
1356
1357     /* Setup water-specific parameters */
1358     inr              = nlist->iinr[0];
1359     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1360     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1361     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1362     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1363
1364     jq0              = _mm256_set1_pd(charge[inr+0]);
1365     jq1              = _mm256_set1_pd(charge[inr+1]);
1366     jq2              = _mm256_set1_pd(charge[inr+2]);
1367     vdwjidx0A        = 2*vdwtype[inr+0];
1368     qq00             = _mm256_mul_pd(iq0,jq0);
1369     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1370     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1371     qq01             = _mm256_mul_pd(iq0,jq1);
1372     qq02             = _mm256_mul_pd(iq0,jq2);
1373     qq10             = _mm256_mul_pd(iq1,jq0);
1374     qq11             = _mm256_mul_pd(iq1,jq1);
1375     qq12             = _mm256_mul_pd(iq1,jq2);
1376     qq20             = _mm256_mul_pd(iq2,jq0);
1377     qq21             = _mm256_mul_pd(iq2,jq1);
1378     qq22             = _mm256_mul_pd(iq2,jq2);
1379
1380     /* Avoid stupid compiler warnings */
1381     jnrA = jnrB = jnrC = jnrD = 0;
1382     j_coord_offsetA = 0;
1383     j_coord_offsetB = 0;
1384     j_coord_offsetC = 0;
1385     j_coord_offsetD = 0;
1386
1387     outeriter        = 0;
1388     inneriter        = 0;
1389
1390     for(iidx=0;iidx<4*DIM;iidx++)
1391     {
1392         scratch[iidx] = 0.0;
1393     }
1394
1395     /* Start outer loop over neighborlists */
1396     for(iidx=0; iidx<nri; iidx++)
1397     {
1398         /* Load shift vector for this list */
1399         i_shift_offset   = DIM*shiftidx[iidx];
1400
1401         /* Load limits for loop over neighbors */
1402         j_index_start    = jindex[iidx];
1403         j_index_end      = jindex[iidx+1];
1404
1405         /* Get outer coordinate index */
1406         inr              = iinr[iidx];
1407         i_coord_offset   = DIM*inr;
1408
1409         /* Load i particle coords and add shift vector */
1410         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1411                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1412
1413         fix0             = _mm256_setzero_pd();
1414         fiy0             = _mm256_setzero_pd();
1415         fiz0             = _mm256_setzero_pd();
1416         fix1             = _mm256_setzero_pd();
1417         fiy1             = _mm256_setzero_pd();
1418         fiz1             = _mm256_setzero_pd();
1419         fix2             = _mm256_setzero_pd();
1420         fiy2             = _mm256_setzero_pd();
1421         fiz2             = _mm256_setzero_pd();
1422
1423         /* Start inner kernel loop */
1424         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1425         {
1426
1427             /* Get j neighbor index, and coordinate index */
1428             jnrA             = jjnr[jidx];
1429             jnrB             = jjnr[jidx+1];
1430             jnrC             = jjnr[jidx+2];
1431             jnrD             = jjnr[jidx+3];
1432             j_coord_offsetA  = DIM*jnrA;
1433             j_coord_offsetB  = DIM*jnrB;
1434             j_coord_offsetC  = DIM*jnrC;
1435             j_coord_offsetD  = DIM*jnrD;
1436
1437             /* load j atom coordinates */
1438             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1439                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1440                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1441
1442             /* Calculate displacement vector */
1443             dx00             = _mm256_sub_pd(ix0,jx0);
1444             dy00             = _mm256_sub_pd(iy0,jy0);
1445             dz00             = _mm256_sub_pd(iz0,jz0);
1446             dx01             = _mm256_sub_pd(ix0,jx1);
1447             dy01             = _mm256_sub_pd(iy0,jy1);
1448             dz01             = _mm256_sub_pd(iz0,jz1);
1449             dx02             = _mm256_sub_pd(ix0,jx2);
1450             dy02             = _mm256_sub_pd(iy0,jy2);
1451             dz02             = _mm256_sub_pd(iz0,jz2);
1452             dx10             = _mm256_sub_pd(ix1,jx0);
1453             dy10             = _mm256_sub_pd(iy1,jy0);
1454             dz10             = _mm256_sub_pd(iz1,jz0);
1455             dx11             = _mm256_sub_pd(ix1,jx1);
1456             dy11             = _mm256_sub_pd(iy1,jy1);
1457             dz11             = _mm256_sub_pd(iz1,jz1);
1458             dx12             = _mm256_sub_pd(ix1,jx2);
1459             dy12             = _mm256_sub_pd(iy1,jy2);
1460             dz12             = _mm256_sub_pd(iz1,jz2);
1461             dx20             = _mm256_sub_pd(ix2,jx0);
1462             dy20             = _mm256_sub_pd(iy2,jy0);
1463             dz20             = _mm256_sub_pd(iz2,jz0);
1464             dx21             = _mm256_sub_pd(ix2,jx1);
1465             dy21             = _mm256_sub_pd(iy2,jy1);
1466             dz21             = _mm256_sub_pd(iz2,jz1);
1467             dx22             = _mm256_sub_pd(ix2,jx2);
1468             dy22             = _mm256_sub_pd(iy2,jy2);
1469             dz22             = _mm256_sub_pd(iz2,jz2);
1470
1471             /* Calculate squared distance and things based on it */
1472             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1473             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1474             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1475             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1476             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1477             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1478             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1479             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1480             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1481
1482             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1483             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1484             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1485             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1486             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1487             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1488             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1489             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1490             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1491
1492             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1493
1494             fjx0             = _mm256_setzero_pd();
1495             fjy0             = _mm256_setzero_pd();
1496             fjz0             = _mm256_setzero_pd();
1497             fjx1             = _mm256_setzero_pd();
1498             fjy1             = _mm256_setzero_pd();
1499             fjz1             = _mm256_setzero_pd();
1500             fjx2             = _mm256_setzero_pd();
1501             fjy2             = _mm256_setzero_pd();
1502             fjz2             = _mm256_setzero_pd();
1503
1504             /**************************
1505              * CALCULATE INTERACTIONS *
1506              **************************/
1507
1508             r00              = _mm256_mul_pd(rsq00,rinv00);
1509
1510             /* Calculate table index by multiplying r with table scale and truncate to integer */
1511             rt               = _mm256_mul_pd(r00,vftabscale);
1512             vfitab           = _mm256_cvttpd_epi32(rt);
1513             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1514             vfitab           = _mm_slli_epi32(vfitab,2);
1515
1516             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1517             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1518             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1519             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1520             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1521             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1522             Heps             = _mm256_mul_pd(vfeps,H);
1523             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1524             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1525             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1526
1527             /* LENNARD-JONES DISPERSION/REPULSION */
1528
1529             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1530             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1531
1532             fscal            = _mm256_add_pd(felec,fvdw);
1533
1534             /* Calculate temporary vectorial force */
1535             tx               = _mm256_mul_pd(fscal,dx00);
1536             ty               = _mm256_mul_pd(fscal,dy00);
1537             tz               = _mm256_mul_pd(fscal,dz00);
1538
1539             /* Update vectorial force */
1540             fix0             = _mm256_add_pd(fix0,tx);
1541             fiy0             = _mm256_add_pd(fiy0,ty);
1542             fiz0             = _mm256_add_pd(fiz0,tz);
1543
1544             fjx0             = _mm256_add_pd(fjx0,tx);
1545             fjy0             = _mm256_add_pd(fjy0,ty);
1546             fjz0             = _mm256_add_pd(fjz0,tz);
1547
1548             /**************************
1549              * CALCULATE INTERACTIONS *
1550              **************************/
1551
1552             r01              = _mm256_mul_pd(rsq01,rinv01);
1553
1554             /* Calculate table index by multiplying r with table scale and truncate to integer */
1555             rt               = _mm256_mul_pd(r01,vftabscale);
1556             vfitab           = _mm256_cvttpd_epi32(rt);
1557             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1558             vfitab           = _mm_slli_epi32(vfitab,2);
1559
1560             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1561             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1562             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1563             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1564             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1565             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1566             Heps             = _mm256_mul_pd(vfeps,H);
1567             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1568             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1569             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
1570
1571             fscal            = felec;
1572
1573             /* Calculate temporary vectorial force */
1574             tx               = _mm256_mul_pd(fscal,dx01);
1575             ty               = _mm256_mul_pd(fscal,dy01);
1576             tz               = _mm256_mul_pd(fscal,dz01);
1577
1578             /* Update vectorial force */
1579             fix0             = _mm256_add_pd(fix0,tx);
1580             fiy0             = _mm256_add_pd(fiy0,ty);
1581             fiz0             = _mm256_add_pd(fiz0,tz);
1582
1583             fjx1             = _mm256_add_pd(fjx1,tx);
1584             fjy1             = _mm256_add_pd(fjy1,ty);
1585             fjz1             = _mm256_add_pd(fjz1,tz);
1586
1587             /**************************
1588              * CALCULATE INTERACTIONS *
1589              **************************/
1590
1591             r02              = _mm256_mul_pd(rsq02,rinv02);
1592
1593             /* Calculate table index by multiplying r with table scale and truncate to integer */
1594             rt               = _mm256_mul_pd(r02,vftabscale);
1595             vfitab           = _mm256_cvttpd_epi32(rt);
1596             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1597             vfitab           = _mm_slli_epi32(vfitab,2);
1598
1599             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1600             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1601             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1602             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1603             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1604             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1605             Heps             = _mm256_mul_pd(vfeps,H);
1606             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1607             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1608             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
1609
1610             fscal            = felec;
1611
1612             /* Calculate temporary vectorial force */
1613             tx               = _mm256_mul_pd(fscal,dx02);
1614             ty               = _mm256_mul_pd(fscal,dy02);
1615             tz               = _mm256_mul_pd(fscal,dz02);
1616
1617             /* Update vectorial force */
1618             fix0             = _mm256_add_pd(fix0,tx);
1619             fiy0             = _mm256_add_pd(fiy0,ty);
1620             fiz0             = _mm256_add_pd(fiz0,tz);
1621
1622             fjx2             = _mm256_add_pd(fjx2,tx);
1623             fjy2             = _mm256_add_pd(fjy2,ty);
1624             fjz2             = _mm256_add_pd(fjz2,tz);
1625
1626             /**************************
1627              * CALCULATE INTERACTIONS *
1628              **************************/
1629
1630             r10              = _mm256_mul_pd(rsq10,rinv10);
1631
1632             /* Calculate table index by multiplying r with table scale and truncate to integer */
1633             rt               = _mm256_mul_pd(r10,vftabscale);
1634             vfitab           = _mm256_cvttpd_epi32(rt);
1635             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1636             vfitab           = _mm_slli_epi32(vfitab,2);
1637
1638             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1639             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1640             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1641             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1642             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1643             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1644             Heps             = _mm256_mul_pd(vfeps,H);
1645             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1646             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1647             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1648
1649             fscal            = felec;
1650
1651             /* Calculate temporary vectorial force */
1652             tx               = _mm256_mul_pd(fscal,dx10);
1653             ty               = _mm256_mul_pd(fscal,dy10);
1654             tz               = _mm256_mul_pd(fscal,dz10);
1655
1656             /* Update vectorial force */
1657             fix1             = _mm256_add_pd(fix1,tx);
1658             fiy1             = _mm256_add_pd(fiy1,ty);
1659             fiz1             = _mm256_add_pd(fiz1,tz);
1660
1661             fjx0             = _mm256_add_pd(fjx0,tx);
1662             fjy0             = _mm256_add_pd(fjy0,ty);
1663             fjz0             = _mm256_add_pd(fjz0,tz);
1664
1665             /**************************
1666              * CALCULATE INTERACTIONS *
1667              **************************/
1668
1669             r11              = _mm256_mul_pd(rsq11,rinv11);
1670
1671             /* Calculate table index by multiplying r with table scale and truncate to integer */
1672             rt               = _mm256_mul_pd(r11,vftabscale);
1673             vfitab           = _mm256_cvttpd_epi32(rt);
1674             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1675             vfitab           = _mm_slli_epi32(vfitab,2);
1676
1677             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1678             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1679             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1680             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1681             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1682             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1683             Heps             = _mm256_mul_pd(vfeps,H);
1684             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1685             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1686             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1687
1688             fscal            = felec;
1689
1690             /* Calculate temporary vectorial force */
1691             tx               = _mm256_mul_pd(fscal,dx11);
1692             ty               = _mm256_mul_pd(fscal,dy11);
1693             tz               = _mm256_mul_pd(fscal,dz11);
1694
1695             /* Update vectorial force */
1696             fix1             = _mm256_add_pd(fix1,tx);
1697             fiy1             = _mm256_add_pd(fiy1,ty);
1698             fiz1             = _mm256_add_pd(fiz1,tz);
1699
1700             fjx1             = _mm256_add_pd(fjx1,tx);
1701             fjy1             = _mm256_add_pd(fjy1,ty);
1702             fjz1             = _mm256_add_pd(fjz1,tz);
1703
1704             /**************************
1705              * CALCULATE INTERACTIONS *
1706              **************************/
1707
1708             r12              = _mm256_mul_pd(rsq12,rinv12);
1709
1710             /* Calculate table index by multiplying r with table scale and truncate to integer */
1711             rt               = _mm256_mul_pd(r12,vftabscale);
1712             vfitab           = _mm256_cvttpd_epi32(rt);
1713             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1714             vfitab           = _mm_slli_epi32(vfitab,2);
1715
1716             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1717             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1718             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1719             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1720             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1721             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1722             Heps             = _mm256_mul_pd(vfeps,H);
1723             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1724             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1725             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1726
1727             fscal            = felec;
1728
1729             /* Calculate temporary vectorial force */
1730             tx               = _mm256_mul_pd(fscal,dx12);
1731             ty               = _mm256_mul_pd(fscal,dy12);
1732             tz               = _mm256_mul_pd(fscal,dz12);
1733
1734             /* Update vectorial force */
1735             fix1             = _mm256_add_pd(fix1,tx);
1736             fiy1             = _mm256_add_pd(fiy1,ty);
1737             fiz1             = _mm256_add_pd(fiz1,tz);
1738
1739             fjx2             = _mm256_add_pd(fjx2,tx);
1740             fjy2             = _mm256_add_pd(fjy2,ty);
1741             fjz2             = _mm256_add_pd(fjz2,tz);
1742
1743             /**************************
1744              * CALCULATE INTERACTIONS *
1745              **************************/
1746
1747             r20              = _mm256_mul_pd(rsq20,rinv20);
1748
1749             /* Calculate table index by multiplying r with table scale and truncate to integer */
1750             rt               = _mm256_mul_pd(r20,vftabscale);
1751             vfitab           = _mm256_cvttpd_epi32(rt);
1752             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1753             vfitab           = _mm_slli_epi32(vfitab,2);
1754
1755             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1756             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1757             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1758             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1759             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1760             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1761             Heps             = _mm256_mul_pd(vfeps,H);
1762             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1763             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1764             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1765
1766             fscal            = felec;
1767
1768             /* Calculate temporary vectorial force */
1769             tx               = _mm256_mul_pd(fscal,dx20);
1770             ty               = _mm256_mul_pd(fscal,dy20);
1771             tz               = _mm256_mul_pd(fscal,dz20);
1772
1773             /* Update vectorial force */
1774             fix2             = _mm256_add_pd(fix2,tx);
1775             fiy2             = _mm256_add_pd(fiy2,ty);
1776             fiz2             = _mm256_add_pd(fiz2,tz);
1777
1778             fjx0             = _mm256_add_pd(fjx0,tx);
1779             fjy0             = _mm256_add_pd(fjy0,ty);
1780             fjz0             = _mm256_add_pd(fjz0,tz);
1781
1782             /**************************
1783              * CALCULATE INTERACTIONS *
1784              **************************/
1785
1786             r21              = _mm256_mul_pd(rsq21,rinv21);
1787
1788             /* Calculate table index by multiplying r with table scale and truncate to integer */
1789             rt               = _mm256_mul_pd(r21,vftabscale);
1790             vfitab           = _mm256_cvttpd_epi32(rt);
1791             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1792             vfitab           = _mm_slli_epi32(vfitab,2);
1793
1794             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1795             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1796             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1797             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1798             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1799             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1800             Heps             = _mm256_mul_pd(vfeps,H);
1801             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1802             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1803             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1804
1805             fscal            = felec;
1806
1807             /* Calculate temporary vectorial force */
1808             tx               = _mm256_mul_pd(fscal,dx21);
1809             ty               = _mm256_mul_pd(fscal,dy21);
1810             tz               = _mm256_mul_pd(fscal,dz21);
1811
1812             /* Update vectorial force */
1813             fix2             = _mm256_add_pd(fix2,tx);
1814             fiy2             = _mm256_add_pd(fiy2,ty);
1815             fiz2             = _mm256_add_pd(fiz2,tz);
1816
1817             fjx1             = _mm256_add_pd(fjx1,tx);
1818             fjy1             = _mm256_add_pd(fjy1,ty);
1819             fjz1             = _mm256_add_pd(fjz1,tz);
1820
1821             /**************************
1822              * CALCULATE INTERACTIONS *
1823              **************************/
1824
1825             r22              = _mm256_mul_pd(rsq22,rinv22);
1826
1827             /* Calculate table index by multiplying r with table scale and truncate to integer */
1828             rt               = _mm256_mul_pd(r22,vftabscale);
1829             vfitab           = _mm256_cvttpd_epi32(rt);
1830             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1831             vfitab           = _mm_slli_epi32(vfitab,2);
1832
1833             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1834             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1835             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1836             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1837             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1838             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1839             Heps             = _mm256_mul_pd(vfeps,H);
1840             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1841             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1842             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1843
1844             fscal            = felec;
1845
1846             /* Calculate temporary vectorial force */
1847             tx               = _mm256_mul_pd(fscal,dx22);
1848             ty               = _mm256_mul_pd(fscal,dy22);
1849             tz               = _mm256_mul_pd(fscal,dz22);
1850
1851             /* Update vectorial force */
1852             fix2             = _mm256_add_pd(fix2,tx);
1853             fiy2             = _mm256_add_pd(fiy2,ty);
1854             fiz2             = _mm256_add_pd(fiz2,tz);
1855
1856             fjx2             = _mm256_add_pd(fjx2,tx);
1857             fjy2             = _mm256_add_pd(fjy2,ty);
1858             fjz2             = _mm256_add_pd(fjz2,tz);
1859
1860             fjptrA             = f+j_coord_offsetA;
1861             fjptrB             = f+j_coord_offsetB;
1862             fjptrC             = f+j_coord_offsetC;
1863             fjptrD             = f+j_coord_offsetD;
1864
1865             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1866                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1867
1868             /* Inner loop uses 359 flops */
1869         }
1870
1871         if(jidx<j_index_end)
1872         {
1873
1874             /* Get j neighbor index, and coordinate index */
1875             jnrlistA         = jjnr[jidx];
1876             jnrlistB         = jjnr[jidx+1];
1877             jnrlistC         = jjnr[jidx+2];
1878             jnrlistD         = jjnr[jidx+3];
1879             /* Sign of each element will be negative for non-real atoms.
1880              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1881              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1882              */
1883             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1884
1885             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1886             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1887             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1888
1889             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1890             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1891             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1892             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1893             j_coord_offsetA  = DIM*jnrA;
1894             j_coord_offsetB  = DIM*jnrB;
1895             j_coord_offsetC  = DIM*jnrC;
1896             j_coord_offsetD  = DIM*jnrD;
1897
1898             /* load j atom coordinates */
1899             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1900                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1901                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1902
1903             /* Calculate displacement vector */
1904             dx00             = _mm256_sub_pd(ix0,jx0);
1905             dy00             = _mm256_sub_pd(iy0,jy0);
1906             dz00             = _mm256_sub_pd(iz0,jz0);
1907             dx01             = _mm256_sub_pd(ix0,jx1);
1908             dy01             = _mm256_sub_pd(iy0,jy1);
1909             dz01             = _mm256_sub_pd(iz0,jz1);
1910             dx02             = _mm256_sub_pd(ix0,jx2);
1911             dy02             = _mm256_sub_pd(iy0,jy2);
1912             dz02             = _mm256_sub_pd(iz0,jz2);
1913             dx10             = _mm256_sub_pd(ix1,jx0);
1914             dy10             = _mm256_sub_pd(iy1,jy0);
1915             dz10             = _mm256_sub_pd(iz1,jz0);
1916             dx11             = _mm256_sub_pd(ix1,jx1);
1917             dy11             = _mm256_sub_pd(iy1,jy1);
1918             dz11             = _mm256_sub_pd(iz1,jz1);
1919             dx12             = _mm256_sub_pd(ix1,jx2);
1920             dy12             = _mm256_sub_pd(iy1,jy2);
1921             dz12             = _mm256_sub_pd(iz1,jz2);
1922             dx20             = _mm256_sub_pd(ix2,jx0);
1923             dy20             = _mm256_sub_pd(iy2,jy0);
1924             dz20             = _mm256_sub_pd(iz2,jz0);
1925             dx21             = _mm256_sub_pd(ix2,jx1);
1926             dy21             = _mm256_sub_pd(iy2,jy1);
1927             dz21             = _mm256_sub_pd(iz2,jz1);
1928             dx22             = _mm256_sub_pd(ix2,jx2);
1929             dy22             = _mm256_sub_pd(iy2,jy2);
1930             dz22             = _mm256_sub_pd(iz2,jz2);
1931
1932             /* Calculate squared distance and things based on it */
1933             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1934             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1935             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1936             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1937             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1938             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1939             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1940             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1941             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1942
1943             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1944             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1945             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1946             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1947             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1948             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1949             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1950             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1951             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1952
1953             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1954
1955             fjx0             = _mm256_setzero_pd();
1956             fjy0             = _mm256_setzero_pd();
1957             fjz0             = _mm256_setzero_pd();
1958             fjx1             = _mm256_setzero_pd();
1959             fjy1             = _mm256_setzero_pd();
1960             fjz1             = _mm256_setzero_pd();
1961             fjx2             = _mm256_setzero_pd();
1962             fjy2             = _mm256_setzero_pd();
1963             fjz2             = _mm256_setzero_pd();
1964
1965             /**************************
1966              * CALCULATE INTERACTIONS *
1967              **************************/
1968
1969             r00              = _mm256_mul_pd(rsq00,rinv00);
1970             r00              = _mm256_andnot_pd(dummy_mask,r00);
1971
1972             /* Calculate table index by multiplying r with table scale and truncate to integer */
1973             rt               = _mm256_mul_pd(r00,vftabscale);
1974             vfitab           = _mm256_cvttpd_epi32(rt);
1975             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1976             vfitab           = _mm_slli_epi32(vfitab,2);
1977
1978             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1979             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1980             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1981             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1982             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1983             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1984             Heps             = _mm256_mul_pd(vfeps,H);
1985             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1986             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1987             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1988
1989             /* LENNARD-JONES DISPERSION/REPULSION */
1990
1991             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1992             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1993
1994             fscal            = _mm256_add_pd(felec,fvdw);
1995
1996             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1997
1998             /* Calculate temporary vectorial force */
1999             tx               = _mm256_mul_pd(fscal,dx00);
2000             ty               = _mm256_mul_pd(fscal,dy00);
2001             tz               = _mm256_mul_pd(fscal,dz00);
2002
2003             /* Update vectorial force */
2004             fix0             = _mm256_add_pd(fix0,tx);
2005             fiy0             = _mm256_add_pd(fiy0,ty);
2006             fiz0             = _mm256_add_pd(fiz0,tz);
2007
2008             fjx0             = _mm256_add_pd(fjx0,tx);
2009             fjy0             = _mm256_add_pd(fjy0,ty);
2010             fjz0             = _mm256_add_pd(fjz0,tz);
2011
2012             /**************************
2013              * CALCULATE INTERACTIONS *
2014              **************************/
2015
2016             r01              = _mm256_mul_pd(rsq01,rinv01);
2017             r01              = _mm256_andnot_pd(dummy_mask,r01);
2018
2019             /* Calculate table index by multiplying r with table scale and truncate to integer */
2020             rt               = _mm256_mul_pd(r01,vftabscale);
2021             vfitab           = _mm256_cvttpd_epi32(rt);
2022             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2023             vfitab           = _mm_slli_epi32(vfitab,2);
2024
2025             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2026             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2027             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2028             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2029             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2030             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2031             Heps             = _mm256_mul_pd(vfeps,H);
2032             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2033             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2034             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
2035
2036             fscal            = felec;
2037
2038             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2039
2040             /* Calculate temporary vectorial force */
2041             tx               = _mm256_mul_pd(fscal,dx01);
2042             ty               = _mm256_mul_pd(fscal,dy01);
2043             tz               = _mm256_mul_pd(fscal,dz01);
2044
2045             /* Update vectorial force */
2046             fix0             = _mm256_add_pd(fix0,tx);
2047             fiy0             = _mm256_add_pd(fiy0,ty);
2048             fiz0             = _mm256_add_pd(fiz0,tz);
2049
2050             fjx1             = _mm256_add_pd(fjx1,tx);
2051             fjy1             = _mm256_add_pd(fjy1,ty);
2052             fjz1             = _mm256_add_pd(fjz1,tz);
2053
2054             /**************************
2055              * CALCULATE INTERACTIONS *
2056              **************************/
2057
2058             r02              = _mm256_mul_pd(rsq02,rinv02);
2059             r02              = _mm256_andnot_pd(dummy_mask,r02);
2060
2061             /* Calculate table index by multiplying r with table scale and truncate to integer */
2062             rt               = _mm256_mul_pd(r02,vftabscale);
2063             vfitab           = _mm256_cvttpd_epi32(rt);
2064             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2065             vfitab           = _mm_slli_epi32(vfitab,2);
2066
2067             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2068             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2069             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2070             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2071             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2072             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2073             Heps             = _mm256_mul_pd(vfeps,H);
2074             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2075             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2076             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
2077
2078             fscal            = felec;
2079
2080             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2081
2082             /* Calculate temporary vectorial force */
2083             tx               = _mm256_mul_pd(fscal,dx02);
2084             ty               = _mm256_mul_pd(fscal,dy02);
2085             tz               = _mm256_mul_pd(fscal,dz02);
2086
2087             /* Update vectorial force */
2088             fix0             = _mm256_add_pd(fix0,tx);
2089             fiy0             = _mm256_add_pd(fiy0,ty);
2090             fiz0             = _mm256_add_pd(fiz0,tz);
2091
2092             fjx2             = _mm256_add_pd(fjx2,tx);
2093             fjy2             = _mm256_add_pd(fjy2,ty);
2094             fjz2             = _mm256_add_pd(fjz2,tz);
2095
2096             /**************************
2097              * CALCULATE INTERACTIONS *
2098              **************************/
2099
2100             r10              = _mm256_mul_pd(rsq10,rinv10);
2101             r10              = _mm256_andnot_pd(dummy_mask,r10);
2102
2103             /* Calculate table index by multiplying r with table scale and truncate to integer */
2104             rt               = _mm256_mul_pd(r10,vftabscale);
2105             vfitab           = _mm256_cvttpd_epi32(rt);
2106             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2107             vfitab           = _mm_slli_epi32(vfitab,2);
2108
2109             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2110             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2111             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2112             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2113             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2114             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2115             Heps             = _mm256_mul_pd(vfeps,H);
2116             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2117             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2118             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
2119
2120             fscal            = felec;
2121
2122             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2123
2124             /* Calculate temporary vectorial force */
2125             tx               = _mm256_mul_pd(fscal,dx10);
2126             ty               = _mm256_mul_pd(fscal,dy10);
2127             tz               = _mm256_mul_pd(fscal,dz10);
2128
2129             /* Update vectorial force */
2130             fix1             = _mm256_add_pd(fix1,tx);
2131             fiy1             = _mm256_add_pd(fiy1,ty);
2132             fiz1             = _mm256_add_pd(fiz1,tz);
2133
2134             fjx0             = _mm256_add_pd(fjx0,tx);
2135             fjy0             = _mm256_add_pd(fjy0,ty);
2136             fjz0             = _mm256_add_pd(fjz0,tz);
2137
2138             /**************************
2139              * CALCULATE INTERACTIONS *
2140              **************************/
2141
2142             r11              = _mm256_mul_pd(rsq11,rinv11);
2143             r11              = _mm256_andnot_pd(dummy_mask,r11);
2144
2145             /* Calculate table index by multiplying r with table scale and truncate to integer */
2146             rt               = _mm256_mul_pd(r11,vftabscale);
2147             vfitab           = _mm256_cvttpd_epi32(rt);
2148             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2149             vfitab           = _mm_slli_epi32(vfitab,2);
2150
2151             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2152             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2153             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2154             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2155             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2156             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2157             Heps             = _mm256_mul_pd(vfeps,H);
2158             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2159             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2160             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
2161
2162             fscal            = felec;
2163
2164             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2165
2166             /* Calculate temporary vectorial force */
2167             tx               = _mm256_mul_pd(fscal,dx11);
2168             ty               = _mm256_mul_pd(fscal,dy11);
2169             tz               = _mm256_mul_pd(fscal,dz11);
2170
2171             /* Update vectorial force */
2172             fix1             = _mm256_add_pd(fix1,tx);
2173             fiy1             = _mm256_add_pd(fiy1,ty);
2174             fiz1             = _mm256_add_pd(fiz1,tz);
2175
2176             fjx1             = _mm256_add_pd(fjx1,tx);
2177             fjy1             = _mm256_add_pd(fjy1,ty);
2178             fjz1             = _mm256_add_pd(fjz1,tz);
2179
2180             /**************************
2181              * CALCULATE INTERACTIONS *
2182              **************************/
2183
2184             r12              = _mm256_mul_pd(rsq12,rinv12);
2185             r12              = _mm256_andnot_pd(dummy_mask,r12);
2186
2187             /* Calculate table index by multiplying r with table scale and truncate to integer */
2188             rt               = _mm256_mul_pd(r12,vftabscale);
2189             vfitab           = _mm256_cvttpd_epi32(rt);
2190             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2191             vfitab           = _mm_slli_epi32(vfitab,2);
2192
2193             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2194             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2195             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2196             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2197             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2198             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2199             Heps             = _mm256_mul_pd(vfeps,H);
2200             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2201             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2202             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
2203
2204             fscal            = felec;
2205
2206             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2207
2208             /* Calculate temporary vectorial force */
2209             tx               = _mm256_mul_pd(fscal,dx12);
2210             ty               = _mm256_mul_pd(fscal,dy12);
2211             tz               = _mm256_mul_pd(fscal,dz12);
2212
2213             /* Update vectorial force */
2214             fix1             = _mm256_add_pd(fix1,tx);
2215             fiy1             = _mm256_add_pd(fiy1,ty);
2216             fiz1             = _mm256_add_pd(fiz1,tz);
2217
2218             fjx2             = _mm256_add_pd(fjx2,tx);
2219             fjy2             = _mm256_add_pd(fjy2,ty);
2220             fjz2             = _mm256_add_pd(fjz2,tz);
2221
2222             /**************************
2223              * CALCULATE INTERACTIONS *
2224              **************************/
2225
2226             r20              = _mm256_mul_pd(rsq20,rinv20);
2227             r20              = _mm256_andnot_pd(dummy_mask,r20);
2228
2229             /* Calculate table index by multiplying r with table scale and truncate to integer */
2230             rt               = _mm256_mul_pd(r20,vftabscale);
2231             vfitab           = _mm256_cvttpd_epi32(rt);
2232             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2233             vfitab           = _mm_slli_epi32(vfitab,2);
2234
2235             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2236             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2237             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2238             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2239             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2240             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2241             Heps             = _mm256_mul_pd(vfeps,H);
2242             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2243             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2244             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
2245
2246             fscal            = felec;
2247
2248             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2249
2250             /* Calculate temporary vectorial force */
2251             tx               = _mm256_mul_pd(fscal,dx20);
2252             ty               = _mm256_mul_pd(fscal,dy20);
2253             tz               = _mm256_mul_pd(fscal,dz20);
2254
2255             /* Update vectorial force */
2256             fix2             = _mm256_add_pd(fix2,tx);
2257             fiy2             = _mm256_add_pd(fiy2,ty);
2258             fiz2             = _mm256_add_pd(fiz2,tz);
2259
2260             fjx0             = _mm256_add_pd(fjx0,tx);
2261             fjy0             = _mm256_add_pd(fjy0,ty);
2262             fjz0             = _mm256_add_pd(fjz0,tz);
2263
2264             /**************************
2265              * CALCULATE INTERACTIONS *
2266              **************************/
2267
2268             r21              = _mm256_mul_pd(rsq21,rinv21);
2269             r21              = _mm256_andnot_pd(dummy_mask,r21);
2270
2271             /* Calculate table index by multiplying r with table scale and truncate to integer */
2272             rt               = _mm256_mul_pd(r21,vftabscale);
2273             vfitab           = _mm256_cvttpd_epi32(rt);
2274             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2275             vfitab           = _mm_slli_epi32(vfitab,2);
2276
2277             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2278             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2279             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2280             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2281             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2282             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2283             Heps             = _mm256_mul_pd(vfeps,H);
2284             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2285             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2286             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
2287
2288             fscal            = felec;
2289
2290             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2291
2292             /* Calculate temporary vectorial force */
2293             tx               = _mm256_mul_pd(fscal,dx21);
2294             ty               = _mm256_mul_pd(fscal,dy21);
2295             tz               = _mm256_mul_pd(fscal,dz21);
2296
2297             /* Update vectorial force */
2298             fix2             = _mm256_add_pd(fix2,tx);
2299             fiy2             = _mm256_add_pd(fiy2,ty);
2300             fiz2             = _mm256_add_pd(fiz2,tz);
2301
2302             fjx1             = _mm256_add_pd(fjx1,tx);
2303             fjy1             = _mm256_add_pd(fjy1,ty);
2304             fjz1             = _mm256_add_pd(fjz1,tz);
2305
2306             /**************************
2307              * CALCULATE INTERACTIONS *
2308              **************************/
2309
2310             r22              = _mm256_mul_pd(rsq22,rinv22);
2311             r22              = _mm256_andnot_pd(dummy_mask,r22);
2312
2313             /* Calculate table index by multiplying r with table scale and truncate to integer */
2314             rt               = _mm256_mul_pd(r22,vftabscale);
2315             vfitab           = _mm256_cvttpd_epi32(rt);
2316             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2317             vfitab           = _mm_slli_epi32(vfitab,2);
2318
2319             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2320             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2321             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2322             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2323             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2324             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2325             Heps             = _mm256_mul_pd(vfeps,H);
2326             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2327             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2328             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
2329
2330             fscal            = felec;
2331
2332             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2333
2334             /* Calculate temporary vectorial force */
2335             tx               = _mm256_mul_pd(fscal,dx22);
2336             ty               = _mm256_mul_pd(fscal,dy22);
2337             tz               = _mm256_mul_pd(fscal,dz22);
2338
2339             /* Update vectorial force */
2340             fix2             = _mm256_add_pd(fix2,tx);
2341             fiy2             = _mm256_add_pd(fiy2,ty);
2342             fiz2             = _mm256_add_pd(fiz2,tz);
2343
2344             fjx2             = _mm256_add_pd(fjx2,tx);
2345             fjy2             = _mm256_add_pd(fjy2,ty);
2346             fjz2             = _mm256_add_pd(fjz2,tz);
2347
2348             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2349             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2350             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2351             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2352
2353             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2354                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2355
2356             /* Inner loop uses 368 flops */
2357         }
2358
2359         /* End of innermost loop */
2360
2361         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2362                                                  f+i_coord_offset,fshift+i_shift_offset);
2363
2364         /* Increment number of inner iterations */
2365         inneriter                  += j_index_end - j_index_start;
2366
2367         /* Outer loop uses 18 flops */
2368     }
2369
2370     /* Increment number of outer iterations */
2371     outeriter        += nri;
2372
2373     /* Update outer/inner flops */
2374
2375     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*368);
2376 }