src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_VF_avx_256_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  95     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  97     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 100     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 101     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 105     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     int              nvdwtype;
 110     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 111     int              *vdwtype;
 112     real             *vdwparam;
 113     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 114     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 115     __m128i          vfitab;
 116     __m128i          ifour       = _mm_set1_epi32(4);
 117     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 118     real             *vftab;
 119     __m256d          dummy_mask,cutoff_mask;
 120     __m128           tmpmask0,tmpmask1;
 121     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 122     __m256d          one     = _mm256_set1_pd(1.0);
 123     __m256d          two     = _mm256_set1_pd(2.0);
 124     x                = xx[0];
 125     f                = ff[0];
 126
 127     nri              = nlist->nri;
 128     iinr             = nlist->iinr;
 129     jindex           = nlist->jindex;
 130     jjnr             = nlist->jjnr;
 131     shiftidx         = nlist->shift;
 132     gid              = nlist->gid;
 133     shiftvec         = fr->shift_vec[0];
 134     fshift           = fr->fshift[0];
 135     facel            = _mm256_set1_pd(fr->epsfac);
 136     charge           = mdatoms->chargeA;
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140
 141     vftab            = kernel_data->table_elec->data;
 142     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 147     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 148     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 149     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 150
 151     jq0              = _mm256_set1_pd(charge[inr+0]);
 152     jq1              = _mm256_set1_pd(charge[inr+1]);
 153     jq2              = _mm256_set1_pd(charge[inr+2]);
 154     vdwjidx0A        = 2*vdwtype[inr+0];
 155     qq00             = _mm256_mul_pd(iq0,jq0);
 156     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
 157     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
 158     qq01             = _mm256_mul_pd(iq0,jq1);
 159     qq02             = _mm256_mul_pd(iq0,jq2);
 160     qq10             = _mm256_mul_pd(iq1,jq0);
 161     qq11             = _mm256_mul_pd(iq1,jq1);
 162     qq12             = _mm256_mul_pd(iq1,jq2);
 163     qq20             = _mm256_mul_pd(iq2,jq0);
 164     qq21             = _mm256_mul_pd(iq2,jq1);
 165     qq22             = _mm256_mul_pd(iq2,jq2);
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173
 174     outeriter        = 0;
 175     inneriter        = 0;
 176
 177     for(iidx=0;iidx<4*DIM;iidx++)
 178     {
 179         scratch[iidx] = 0.0;
 180     }
 181
 182     /* Start outer loop over neighborlists */
 183     for(iidx=0; iidx<nri; iidx++)
 184     {
 185         /* Load shift vector for this list */
 186         i_shift_offset   = DIM*shiftidx[iidx];
 187
 188         /* Load limits for loop over neighbors */
 189         j_index_start    = jindex[iidx];
 190         j_index_end      = jindex[iidx+1];
 191
 192         /* Get outer coordinate index */
 193         inr              = iinr[iidx];
 194         i_coord_offset   = DIM*inr;
 195
 196         /* Load i particle coords and add shift vector */
 197         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 198                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 199
 200         fix0             = _mm256_setzero_pd();
 201         fiy0             = _mm256_setzero_pd();
 202         fiz0             = _mm256_setzero_pd();
 203         fix1             = _mm256_setzero_pd();
 204         fiy1             = _mm256_setzero_pd();
 205         fiz1             = _mm256_setzero_pd();
 206         fix2             = _mm256_setzero_pd();
 207         fiy2             = _mm256_setzero_pd();
 208         fiz2             = _mm256_setzero_pd();
 209
 210         /* Reset potential sums */
 211         velecsum         = _mm256_setzero_pd();
 212         vvdwsum          = _mm256_setzero_pd();
 213
 214         /* Start inner kernel loop */
 215         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 216         {
 217
 218             /* Get j neighbor index, and coordinate index */
 219             jnrA             = jjnr[jidx];
 220             jnrB             = jjnr[jidx+1];
 221             jnrC             = jjnr[jidx+2];
 222             jnrD             = jjnr[jidx+3];
 223             j_coord_offsetA  = DIM*jnrA;
 224             j_coord_offsetB  = DIM*jnrB;
 225             j_coord_offsetC  = DIM*jnrC;
 226             j_coord_offsetD  = DIM*jnrD;
 227
 228             /* load j atom coordinates */
 229             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 230                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 231                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 232
 233             /* Calculate displacement vector */
 234             dx00             = _mm256_sub_pd(ix0,jx0);
 235             dy00             = _mm256_sub_pd(iy0,jy0);
 236             dz00             = _mm256_sub_pd(iz0,jz0);
 237             dx01             = _mm256_sub_pd(ix0,jx1);
 238             dy01             = _mm256_sub_pd(iy0,jy1);
 239             dz01             = _mm256_sub_pd(iz0,jz1);
 240             dx02             = _mm256_sub_pd(ix0,jx2);
 241             dy02             = _mm256_sub_pd(iy0,jy2);
 242             dz02             = _mm256_sub_pd(iz0,jz2);
 243             dx10             = _mm256_sub_pd(ix1,jx0);
 244             dy10             = _mm256_sub_pd(iy1,jy0);
 245             dz10             = _mm256_sub_pd(iz1,jz0);
 246             dx11             = _mm256_sub_pd(ix1,jx1);
 247             dy11             = _mm256_sub_pd(iy1,jy1);
 248             dz11             = _mm256_sub_pd(iz1,jz1);
 249             dx12             = _mm256_sub_pd(ix1,jx2);
 250             dy12             = _mm256_sub_pd(iy1,jy2);
 251             dz12             = _mm256_sub_pd(iz1,jz2);
 252             dx20             = _mm256_sub_pd(ix2,jx0);
 253             dy20             = _mm256_sub_pd(iy2,jy0);
 254             dz20             = _mm256_sub_pd(iz2,jz0);
 255             dx21             = _mm256_sub_pd(ix2,jx1);
 256             dy21             = _mm256_sub_pd(iy2,jy1);
 257             dz21             = _mm256_sub_pd(iz2,jz1);
 258             dx22             = _mm256_sub_pd(ix2,jx2);
 259             dy22             = _mm256_sub_pd(iy2,jy2);
 260             dz22             = _mm256_sub_pd(iz2,jz2);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 264             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 265             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 266             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 267             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 268             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 269             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 270             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 271             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 272
 273             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 274             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 275             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 276             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 277             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 278             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 279             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 280             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 281             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 282
 283             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 284
 285             fjx0             = _mm256_setzero_pd();
 286             fjy0             = _mm256_setzero_pd();
 287             fjz0             = _mm256_setzero_pd();
 288             fjx1             = _mm256_setzero_pd();
 289             fjy1             = _mm256_setzero_pd();
 290             fjz1             = _mm256_setzero_pd();
 291             fjx2             = _mm256_setzero_pd();
 292             fjy2             = _mm256_setzero_pd();
 293             fjz2             = _mm256_setzero_pd();
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             r00              = _mm256_mul_pd(rsq00,rinv00);
 300
 301             /* Calculate table index by multiplying r with table scale and truncate to integer */
 302             rt               = _mm256_mul_pd(r00,vftabscale);
 303             vfitab           = _mm256_cvttpd_epi32(rt);
 304             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 305             vfitab           = _mm_slli_epi32(vfitab,2);
 306
 307             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 308             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 309             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 310             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 311             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 312             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 313             Heps             = _mm256_mul_pd(vfeps,H);
 314             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 315             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 316             velec            = _mm256_mul_pd(qq00,VV);
 317             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 318             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 319
 320             /* LENNARD-JONES DISPERSION/REPULSION */
 321
 322             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 323             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 324             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 325             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 326             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm256_add_pd(velecsum,velec);
 330             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 331
 332             fscal            = _mm256_add_pd(felec,fvdw);
 333
 334             /* Calculate temporary vectorial force */
 335             tx               = _mm256_mul_pd(fscal,dx00);
 336             ty               = _mm256_mul_pd(fscal,dy00);
 337             tz               = _mm256_mul_pd(fscal,dz00);
 338
 339             /* Update vectorial force */
 340             fix0             = _mm256_add_pd(fix0,tx);
 341             fiy0             = _mm256_add_pd(fiy0,ty);
 342             fiz0             = _mm256_add_pd(fiz0,tz);
 343
 344             fjx0             = _mm256_add_pd(fjx0,tx);
 345             fjy0             = _mm256_add_pd(fjy0,ty);
 346             fjz0             = _mm256_add_pd(fjz0,tz);
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             r01              = _mm256_mul_pd(rsq01,rinv01);
 353
 354             /* Calculate table index by multiplying r with table scale and truncate to integer */
 355             rt               = _mm256_mul_pd(r01,vftabscale);
 356             vfitab           = _mm256_cvttpd_epi32(rt);
 357             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 358             vfitab           = _mm_slli_epi32(vfitab,2);
 359
 360             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 361             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 362             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 363             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 364             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 365             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 366             Heps             = _mm256_mul_pd(vfeps,H);
 367             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 368             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 369             velec            = _mm256_mul_pd(qq01,VV);
 370             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 371             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velecsum         = _mm256_add_pd(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm256_mul_pd(fscal,dx01);
 380             ty               = _mm256_mul_pd(fscal,dy01);
 381             tz               = _mm256_mul_pd(fscal,dz01);
 382
 383             /* Update vectorial force */
 384             fix0             = _mm256_add_pd(fix0,tx);
 385             fiy0             = _mm256_add_pd(fiy0,ty);
 386             fiz0             = _mm256_add_pd(fiz0,tz);
 387
 388             fjx1             = _mm256_add_pd(fjx1,tx);
 389             fjy1             = _mm256_add_pd(fjy1,ty);
 390             fjz1             = _mm256_add_pd(fjz1,tz);
 391
 392             /**************************
 393              * CALCULATE INTERACTIONS *
 394              **************************/
 395
 396             r02              = _mm256_mul_pd(rsq02,rinv02);
 397
 398             /* Calculate table index by multiplying r with table scale and truncate to integer */
 399             rt               = _mm256_mul_pd(r02,vftabscale);
 400             vfitab           = _mm256_cvttpd_epi32(rt);
 401             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 402             vfitab           = _mm_slli_epi32(vfitab,2);
 403
 404             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 405             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 406             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 407             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 408             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 409             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 410             Heps             = _mm256_mul_pd(vfeps,H);
 411             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 412             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 413             velec            = _mm256_mul_pd(qq02,VV);
 414             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 415             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
 416
 417             /* Update potential sum for this i atom from the interaction with this j atom. */
 418             velecsum         = _mm256_add_pd(velecsum,velec);
 419
 420             fscal            = felec;
 421
 422             /* Calculate temporary vectorial force */
 423             tx               = _mm256_mul_pd(fscal,dx02);
 424             ty               = _mm256_mul_pd(fscal,dy02);
 425             tz               = _mm256_mul_pd(fscal,dz02);
 426
 427             /* Update vectorial force */
 428             fix0             = _mm256_add_pd(fix0,tx);
 429             fiy0             = _mm256_add_pd(fiy0,ty);
 430             fiz0             = _mm256_add_pd(fiz0,tz);
 431
 432             fjx2             = _mm256_add_pd(fjx2,tx);
 433             fjy2             = _mm256_add_pd(fjy2,ty);
 434             fjz2             = _mm256_add_pd(fjz2,tz);
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             r10              = _mm256_mul_pd(rsq10,rinv10);
 441
 442             /* Calculate table index by multiplying r with table scale and truncate to integer */
 443             rt               = _mm256_mul_pd(r10,vftabscale);
 444             vfitab           = _mm256_cvttpd_epi32(rt);
 445             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 446             vfitab           = _mm_slli_epi32(vfitab,2);
 447
 448             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 449             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 450             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 451             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 452             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 453             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 454             Heps             = _mm256_mul_pd(vfeps,H);
 455             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 456             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 457             velec            = _mm256_mul_pd(qq10,VV);
 458             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 459             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             velecsum         = _mm256_add_pd(velecsum,velec);
 463
 464             fscal            = felec;
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm256_mul_pd(fscal,dx10);
 468             ty               = _mm256_mul_pd(fscal,dy10);
 469             tz               = _mm256_mul_pd(fscal,dz10);
 470
 471             /* Update vectorial force */
 472             fix1             = _mm256_add_pd(fix1,tx);
 473             fiy1             = _mm256_add_pd(fiy1,ty);
 474             fiz1             = _mm256_add_pd(fiz1,tz);
 475
 476             fjx0             = _mm256_add_pd(fjx0,tx);
 477             fjy0             = _mm256_add_pd(fjy0,ty);
 478             fjz0             = _mm256_add_pd(fjz0,tz);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r11              = _mm256_mul_pd(rsq11,rinv11);
 485
 486             /* Calculate table index by multiplying r with table scale and truncate to integer */
 487             rt               = _mm256_mul_pd(r11,vftabscale);
 488             vfitab           = _mm256_cvttpd_epi32(rt);
 489             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 490             vfitab           = _mm_slli_epi32(vfitab,2);
 491
 492             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 493             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 494             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 495             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 496             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 497             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 498             Heps             = _mm256_mul_pd(vfeps,H);
 499             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 500             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 501             velec            = _mm256_mul_pd(qq11,VV);
 502             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 503             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velecsum         = _mm256_add_pd(velecsum,velec);
 507
 508             fscal            = felec;
 509
 510             /* Calculate temporary vectorial force */
 511             tx               = _mm256_mul_pd(fscal,dx11);
 512             ty               = _mm256_mul_pd(fscal,dy11);
 513             tz               = _mm256_mul_pd(fscal,dz11);
 514
 515             /* Update vectorial force */
 516             fix1             = _mm256_add_pd(fix1,tx);
 517             fiy1             = _mm256_add_pd(fiy1,ty);
 518             fiz1             = _mm256_add_pd(fiz1,tz);
 519
 520             fjx1             = _mm256_add_pd(fjx1,tx);
 521             fjy1             = _mm256_add_pd(fjy1,ty);
 522             fjz1             = _mm256_add_pd(fjz1,tz);
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r12              = _mm256_mul_pd(rsq12,rinv12);
 529
 530             /* Calculate table index by multiplying r with table scale and truncate to integer */
 531             rt               = _mm256_mul_pd(r12,vftabscale);
 532             vfitab           = _mm256_cvttpd_epi32(rt);
 533             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 534             vfitab           = _mm_slli_epi32(vfitab,2);
 535
 536             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 537             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 538             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 539             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 540             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 541             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 542             Heps             = _mm256_mul_pd(vfeps,H);
 543             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 544             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 545             velec            = _mm256_mul_pd(qq12,VV);
 546             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 547             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velecsum         = _mm256_add_pd(velecsum,velec);
 551
 552             fscal            = felec;
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm256_mul_pd(fscal,dx12);
 556             ty               = _mm256_mul_pd(fscal,dy12);
 557             tz               = _mm256_mul_pd(fscal,dz12);
 558
 559             /* Update vectorial force */
 560             fix1             = _mm256_add_pd(fix1,tx);
 561             fiy1             = _mm256_add_pd(fiy1,ty);
 562             fiz1             = _mm256_add_pd(fiz1,tz);
 563
 564             fjx2             = _mm256_add_pd(fjx2,tx);
 565             fjy2             = _mm256_add_pd(fjy2,ty);
 566             fjz2             = _mm256_add_pd(fjz2,tz);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             r20              = _mm256_mul_pd(rsq20,rinv20);
 573
 574             /* Calculate table index by multiplying r with table scale and truncate to integer */
 575             rt               = _mm256_mul_pd(r20,vftabscale);
 576             vfitab           = _mm256_cvttpd_epi32(rt);
 577             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 578             vfitab           = _mm_slli_epi32(vfitab,2);
 579
 580             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 581             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 582             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 583             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 584             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 585             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 586             Heps             = _mm256_mul_pd(vfeps,H);
 587             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 588             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 589             velec            = _mm256_mul_pd(qq20,VV);
 590             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 591             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velecsum         = _mm256_add_pd(velecsum,velec);
 595
 596             fscal            = felec;
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm256_mul_pd(fscal,dx20);
 600             ty               = _mm256_mul_pd(fscal,dy20);
 601             tz               = _mm256_mul_pd(fscal,dz20);
 602
 603             /* Update vectorial force */
 604             fix2             = _mm256_add_pd(fix2,tx);
 605             fiy2             = _mm256_add_pd(fiy2,ty);
 606             fiz2             = _mm256_add_pd(fiz2,tz);
 607
 608             fjx0             = _mm256_add_pd(fjx0,tx);
 609             fjy0             = _mm256_add_pd(fjy0,ty);
 610             fjz0             = _mm256_add_pd(fjz0,tz);
 611
 612             /**************************
 613              * CALCULATE INTERACTIONS *
 614              **************************/
 615
 616             r21              = _mm256_mul_pd(rsq21,rinv21);
 617
 618             /* Calculate table index by multiplying r with table scale and truncate to integer */
 619             rt               = _mm256_mul_pd(r21,vftabscale);
 620             vfitab           = _mm256_cvttpd_epi32(rt);
 621             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 622             vfitab           = _mm_slli_epi32(vfitab,2);
 623
 624             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 625             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 626             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 627             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 628             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 629             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 630             Heps             = _mm256_mul_pd(vfeps,H);
 631             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 632             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 633             velec            = _mm256_mul_pd(qq21,VV);
 634             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 635             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
 636
 637             /* Update potential sum for this i atom from the interaction with this j atom. */
 638             velecsum         = _mm256_add_pd(velecsum,velec);
 639
 640             fscal            = felec;
 641
 642             /* Calculate temporary vectorial force */
 643             tx               = _mm256_mul_pd(fscal,dx21);
 644             ty               = _mm256_mul_pd(fscal,dy21);
 645             tz               = _mm256_mul_pd(fscal,dz21);
 646
 647             /* Update vectorial force */
 648             fix2             = _mm256_add_pd(fix2,tx);
 649             fiy2             = _mm256_add_pd(fiy2,ty);
 650             fiz2             = _mm256_add_pd(fiz2,tz);
 651
 652             fjx1             = _mm256_add_pd(fjx1,tx);
 653             fjy1             = _mm256_add_pd(fjy1,ty);
 654             fjz1             = _mm256_add_pd(fjz1,tz);
 655
 656             /**************************
 657              * CALCULATE INTERACTIONS *
 658              **************************/
 659
 660             r22              = _mm256_mul_pd(rsq22,rinv22);
 661
 662             /* Calculate table index by multiplying r with table scale and truncate to integer */
 663             rt               = _mm256_mul_pd(r22,vftabscale);
 664             vfitab           = _mm256_cvttpd_epi32(rt);
 665             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 666             vfitab           = _mm_slli_epi32(vfitab,2);
 667
 668             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 669             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 670             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 671             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 672             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 673             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 674             Heps             = _mm256_mul_pd(vfeps,H);
 675             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 676             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 677             velec            = _mm256_mul_pd(qq22,VV);
 678             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 679             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
 680
 681             /* Update potential sum for this i atom from the interaction with this j atom. */
 682             velecsum         = _mm256_add_pd(velecsum,velec);
 683
 684             fscal            = felec;
 685
 686             /* Calculate temporary vectorial force */
 687             tx               = _mm256_mul_pd(fscal,dx22);
 688             ty               = _mm256_mul_pd(fscal,dy22);
 689             tz               = _mm256_mul_pd(fscal,dz22);
 690
 691             /* Update vectorial force */
 692             fix2             = _mm256_add_pd(fix2,tx);
 693             fiy2             = _mm256_add_pd(fiy2,ty);
 694             fiz2             = _mm256_add_pd(fiz2,tz);
 695
 696             fjx2             = _mm256_add_pd(fjx2,tx);
 697             fjy2             = _mm256_add_pd(fjy2,ty);
 698             fjz2             = _mm256_add_pd(fjz2,tz);
 699
 700             fjptrA             = f+j_coord_offsetA;
 701             fjptrB             = f+j_coord_offsetB;
 702             fjptrC             = f+j_coord_offsetC;
 703             fjptrD             = f+j_coord_offsetD;
 704
 705             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 706                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 707
 708             /* Inner loop uses 400 flops */
 709         }
 710
 711         if(jidx<j_index_end)
 712         {
 713
 714             /* Get j neighbor index, and coordinate index */
 715             jnrlistA         = jjnr[jidx];
 716             jnrlistB         = jjnr[jidx+1];
 717             jnrlistC         = jjnr[jidx+2];
 718             jnrlistD         = jjnr[jidx+3];
 719             /* Sign of each element will be negative for non-real atoms.
 720              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 721              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 722              */
 723             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 724
 725             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 726             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 727             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 728
 729             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 730             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 731             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 732             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 733             j_coord_offsetA  = DIM*jnrA;
 734             j_coord_offsetB  = DIM*jnrB;
 735             j_coord_offsetC  = DIM*jnrC;
 736             j_coord_offsetD  = DIM*jnrD;
 737
 738             /* load j atom coordinates */
 739             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 740                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 741                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 742
 743             /* Calculate displacement vector */
 744             dx00             = _mm256_sub_pd(ix0,jx0);
 745             dy00             = _mm256_sub_pd(iy0,jy0);
 746             dz00             = _mm256_sub_pd(iz0,jz0);
 747             dx01             = _mm256_sub_pd(ix0,jx1);
 748             dy01             = _mm256_sub_pd(iy0,jy1);
 749             dz01             = _mm256_sub_pd(iz0,jz1);
 750             dx02             = _mm256_sub_pd(ix0,jx2);
 751             dy02             = _mm256_sub_pd(iy0,jy2);
 752             dz02             = _mm256_sub_pd(iz0,jz2);
 753             dx10             = _mm256_sub_pd(ix1,jx0);
 754             dy10             = _mm256_sub_pd(iy1,jy0);
 755             dz10             = _mm256_sub_pd(iz1,jz0);
 756             dx11             = _mm256_sub_pd(ix1,jx1);
 757             dy11             = _mm256_sub_pd(iy1,jy1);
 758             dz11             = _mm256_sub_pd(iz1,jz1);
 759             dx12             = _mm256_sub_pd(ix1,jx2);
 760             dy12             = _mm256_sub_pd(iy1,jy2);
 761             dz12             = _mm256_sub_pd(iz1,jz2);
 762             dx20             = _mm256_sub_pd(ix2,jx0);
 763             dy20             = _mm256_sub_pd(iy2,jy0);
 764             dz20             = _mm256_sub_pd(iz2,jz0);
 765             dx21             = _mm256_sub_pd(ix2,jx1);
 766             dy21             = _mm256_sub_pd(iy2,jy1);
 767             dz21             = _mm256_sub_pd(iz2,jz1);
 768             dx22             = _mm256_sub_pd(ix2,jx2);
 769             dy22             = _mm256_sub_pd(iy2,jy2);
 770             dz22             = _mm256_sub_pd(iz2,jz2);
 771
 772             /* Calculate squared distance and things based on it */
 773             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 774             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
 775             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
 776             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 777             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
 778             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
 779             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 780             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
 781             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
 782
 783             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 784             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
 785             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
 786             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 787             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
 788             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
 789             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 790             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
 791             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
 792
 793             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 794
 795             fjx0             = _mm256_setzero_pd();
 796             fjy0             = _mm256_setzero_pd();
 797             fjz0             = _mm256_setzero_pd();
 798             fjx1             = _mm256_setzero_pd();
 799             fjy1             = _mm256_setzero_pd();
 800             fjz1             = _mm256_setzero_pd();
 801             fjx2             = _mm256_setzero_pd();
 802             fjy2             = _mm256_setzero_pd();
 803             fjz2             = _mm256_setzero_pd();
 804
 805             /**************************
 806              * CALCULATE INTERACTIONS *
 807              **************************/
 808
 809             r00              = _mm256_mul_pd(rsq00,rinv00);
 810             r00              = _mm256_andnot_pd(dummy_mask,r00);
 811
 812             /* Calculate table index by multiplying r with table scale and truncate to integer */
 813             rt               = _mm256_mul_pd(r00,vftabscale);
 814             vfitab           = _mm256_cvttpd_epi32(rt);
 815             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 816             vfitab           = _mm_slli_epi32(vfitab,2);
 817
 818             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 819             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 820             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 821             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 822             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 823             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 824             Heps             = _mm256_mul_pd(vfeps,H);
 825             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 826             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 827             velec            = _mm256_mul_pd(qq00,VV);
 828             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 829             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 830
 831             /* LENNARD-JONES DISPERSION/REPULSION */
 832
 833             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 834             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 835             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 836             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 837             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 838
 839             /* Update potential sum for this i atom from the interaction with this j atom. */
 840             velec            = _mm256_andnot_pd(dummy_mask,velec);
 841             velecsum         = _mm256_add_pd(velecsum,velec);
 842             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 843             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 844
 845             fscal            = _mm256_add_pd(felec,fvdw);
 846
 847             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 848
 849             /* Calculate temporary vectorial force */
 850             tx               = _mm256_mul_pd(fscal,dx00);
 851             ty               = _mm256_mul_pd(fscal,dy00);
 852             tz               = _mm256_mul_pd(fscal,dz00);
 853
 854             /* Update vectorial force */
 855             fix0             = _mm256_add_pd(fix0,tx);
 856             fiy0             = _mm256_add_pd(fiy0,ty);
 857             fiz0             = _mm256_add_pd(fiz0,tz);
 858
 859             fjx0             = _mm256_add_pd(fjx0,tx);
 860             fjy0             = _mm256_add_pd(fjy0,ty);
 861             fjz0             = _mm256_add_pd(fjz0,tz);
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             r01              = _mm256_mul_pd(rsq01,rinv01);
 868             r01              = _mm256_andnot_pd(dummy_mask,r01);
 869
 870             /* Calculate table index by multiplying r with table scale and truncate to integer */
 871             rt               = _mm256_mul_pd(r01,vftabscale);
 872             vfitab           = _mm256_cvttpd_epi32(rt);
 873             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 874             vfitab           = _mm_slli_epi32(vfitab,2);
 875
 876             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 877             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 878             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 879             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 880             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 881             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 882             Heps             = _mm256_mul_pd(vfeps,H);
 883             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 884             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 885             velec            = _mm256_mul_pd(qq01,VV);
 886             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 887             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
 888
 889             /* Update potential sum for this i atom from the interaction with this j atom. */
 890             velec            = _mm256_andnot_pd(dummy_mask,velec);
 891             velecsum         = _mm256_add_pd(velecsum,velec);
 892
 893             fscal            = felec;
 894
 895             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm256_mul_pd(fscal,dx01);
 899             ty               = _mm256_mul_pd(fscal,dy01);
 900             tz               = _mm256_mul_pd(fscal,dz01);
 901
 902             /* Update vectorial force */
 903             fix0             = _mm256_add_pd(fix0,tx);
 904             fiy0             = _mm256_add_pd(fiy0,ty);
 905             fiz0             = _mm256_add_pd(fiz0,tz);
 906
 907             fjx1             = _mm256_add_pd(fjx1,tx);
 908             fjy1             = _mm256_add_pd(fjy1,ty);
 909             fjz1             = _mm256_add_pd(fjz1,tz);
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             r02              = _mm256_mul_pd(rsq02,rinv02);
 916             r02              = _mm256_andnot_pd(dummy_mask,r02);
 917
 918             /* Calculate table index by multiplying r with table scale and truncate to integer */
 919             rt               = _mm256_mul_pd(r02,vftabscale);
 920             vfitab           = _mm256_cvttpd_epi32(rt);
 921             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 922             vfitab           = _mm_slli_epi32(vfitab,2);
 923
 924             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 925             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 926             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 927             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 928             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 929             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 930             Heps             = _mm256_mul_pd(vfeps,H);
 931             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 932             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 933             velec            = _mm256_mul_pd(qq02,VV);
 934             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 935             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
 936
 937             /* Update potential sum for this i atom from the interaction with this j atom. */
 938             velec            = _mm256_andnot_pd(dummy_mask,velec);
 939             velecsum         = _mm256_add_pd(velecsum,velec);
 940
 941             fscal            = felec;
 942
 943             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 944
 945             /* Calculate temporary vectorial force */
 946             tx               = _mm256_mul_pd(fscal,dx02);
 947             ty               = _mm256_mul_pd(fscal,dy02);
 948             tz               = _mm256_mul_pd(fscal,dz02);
 949
 950             /* Update vectorial force */
 951             fix0             = _mm256_add_pd(fix0,tx);
 952             fiy0             = _mm256_add_pd(fiy0,ty);
 953             fiz0             = _mm256_add_pd(fiz0,tz);
 954
 955             fjx2             = _mm256_add_pd(fjx2,tx);
 956             fjy2             = _mm256_add_pd(fjy2,ty);
 957             fjz2             = _mm256_add_pd(fjz2,tz);
 958
 959             /**************************
 960              * CALCULATE INTERACTIONS *
 961              **************************/
 962
 963             r10              = _mm256_mul_pd(rsq10,rinv10);
 964             r10              = _mm256_andnot_pd(dummy_mask,r10);
 965
 966             /* Calculate table index by multiplying r with table scale and truncate to integer */
 967             rt               = _mm256_mul_pd(r10,vftabscale);
 968             vfitab           = _mm256_cvttpd_epi32(rt);
 969             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 970             vfitab           = _mm_slli_epi32(vfitab,2);
 971
 972             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 973             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 974             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 975             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 976             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 977             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 978             Heps             = _mm256_mul_pd(vfeps,H);
 979             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 980             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 981             velec            = _mm256_mul_pd(qq10,VV);
 982             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 983             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 984
 985             /* Update potential sum for this i atom from the interaction with this j atom. */
 986             velec            = _mm256_andnot_pd(dummy_mask,velec);
 987             velecsum         = _mm256_add_pd(velecsum,velec);
 988
 989             fscal            = felec;
 990
 991             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm256_mul_pd(fscal,dx10);
 995             ty               = _mm256_mul_pd(fscal,dy10);
 996             tz               = _mm256_mul_pd(fscal,dz10);
 997
 998             /* Update vectorial force */
 999             fix1             = _mm256_add_pd(fix1,tx);
1000             fiy1             = _mm256_add_pd(fiy1,ty);
1001             fiz1             = _mm256_add_pd(fiz1,tz);
1002
1003             fjx0             = _mm256_add_pd(fjx0,tx);
1004             fjy0             = _mm256_add_pd(fjy0,ty);
1005             fjz0             = _mm256_add_pd(fjz0,tz);
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             r11              = _mm256_mul_pd(rsq11,rinv11);
1012             r11              = _mm256_andnot_pd(dummy_mask,r11);
1013
1014             /* Calculate table index by multiplying r with table scale and truncate to integer */
1015             rt               = _mm256_mul_pd(r11,vftabscale);
1016             vfitab           = _mm256_cvttpd_epi32(rt);
1017             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1018             vfitab           = _mm_slli_epi32(vfitab,2);
1019
1020             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1021             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1022             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1023             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1024             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1025             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1026             Heps             = _mm256_mul_pd(vfeps,H);
1027             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1028             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1029             velec            = _mm256_mul_pd(qq11,VV);
1030             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1031             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1032
1033             /* Update potential sum for this i atom from the interaction with this j atom. */
1034             velec            = _mm256_andnot_pd(dummy_mask,velec);
1035             velecsum         = _mm256_add_pd(velecsum,velec);
1036
1037             fscal            = felec;
1038
1039             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1040
1041             /* Calculate temporary vectorial force */
1042             tx               = _mm256_mul_pd(fscal,dx11);
1043             ty               = _mm256_mul_pd(fscal,dy11);
1044             tz               = _mm256_mul_pd(fscal,dz11);
1045
1046             /* Update vectorial force */
1047             fix1             = _mm256_add_pd(fix1,tx);
1048             fiy1             = _mm256_add_pd(fiy1,ty);
1049             fiz1             = _mm256_add_pd(fiz1,tz);
1050
1051             fjx1             = _mm256_add_pd(fjx1,tx);
1052             fjy1             = _mm256_add_pd(fjy1,ty);
1053             fjz1             = _mm256_add_pd(fjz1,tz);
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             r12              = _mm256_mul_pd(rsq12,rinv12);
1060             r12              = _mm256_andnot_pd(dummy_mask,r12);
1061
1062             /* Calculate table index by multiplying r with table scale and truncate to integer */
1063             rt               = _mm256_mul_pd(r12,vftabscale);
1064             vfitab           = _mm256_cvttpd_epi32(rt);
1065             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1066             vfitab           = _mm_slli_epi32(vfitab,2);
1067
1068             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1069             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1070             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1071             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1072             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1073             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1074             Heps             = _mm256_mul_pd(vfeps,H);
1075             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1076             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1077             velec            = _mm256_mul_pd(qq12,VV);
1078             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1079             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1080
1081             /* Update potential sum for this i atom from the interaction with this j atom. */
1082             velec            = _mm256_andnot_pd(dummy_mask,velec);
1083             velecsum         = _mm256_add_pd(velecsum,velec);
1084
1085             fscal            = felec;
1086
1087             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm256_mul_pd(fscal,dx12);
1091             ty               = _mm256_mul_pd(fscal,dy12);
1092             tz               = _mm256_mul_pd(fscal,dz12);
1093
1094             /* Update vectorial force */
1095             fix1             = _mm256_add_pd(fix1,tx);
1096             fiy1             = _mm256_add_pd(fiy1,ty);
1097             fiz1             = _mm256_add_pd(fiz1,tz);
1098
1099             fjx2             = _mm256_add_pd(fjx2,tx);
1100             fjy2             = _mm256_add_pd(fjy2,ty);
1101             fjz2             = _mm256_add_pd(fjz2,tz);
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             r20              = _mm256_mul_pd(rsq20,rinv20);
1108             r20              = _mm256_andnot_pd(dummy_mask,r20);
1109
1110             /* Calculate table index by multiplying r with table scale and truncate to integer */
1111             rt               = _mm256_mul_pd(r20,vftabscale);
1112             vfitab           = _mm256_cvttpd_epi32(rt);
1113             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1114             vfitab           = _mm_slli_epi32(vfitab,2);
1115
1116             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1117             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1118             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1119             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1120             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1121             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1122             Heps             = _mm256_mul_pd(vfeps,H);
1123             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1124             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1125             velec            = _mm256_mul_pd(qq20,VV);
1126             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1127             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1128
1129             /* Update potential sum for this i atom from the interaction with this j atom. */
1130             velec            = _mm256_andnot_pd(dummy_mask,velec);
1131             velecsum         = _mm256_add_pd(velecsum,velec);
1132
1133             fscal            = felec;
1134
1135             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1136
1137             /* Calculate temporary vectorial force */
1138             tx               = _mm256_mul_pd(fscal,dx20);
1139             ty               = _mm256_mul_pd(fscal,dy20);
1140             tz               = _mm256_mul_pd(fscal,dz20);
1141
1142             /* Update vectorial force */
1143             fix2             = _mm256_add_pd(fix2,tx);
1144             fiy2             = _mm256_add_pd(fiy2,ty);
1145             fiz2             = _mm256_add_pd(fiz2,tz);
1146
1147             fjx0             = _mm256_add_pd(fjx0,tx);
1148             fjy0             = _mm256_add_pd(fjy0,ty);
1149             fjz0             = _mm256_add_pd(fjz0,tz);
1150
1151             /**************************
1152              * CALCULATE INTERACTIONS *
1153              **************************/
1154
1155             r21              = _mm256_mul_pd(rsq21,rinv21);
1156             r21              = _mm256_andnot_pd(dummy_mask,r21);
1157
1158             /* Calculate table index by multiplying r with table scale and truncate to integer */
1159             rt               = _mm256_mul_pd(r21,vftabscale);
1160             vfitab           = _mm256_cvttpd_epi32(rt);
1161             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1162             vfitab           = _mm_slli_epi32(vfitab,2);
1163
1164             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1165             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1166             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1167             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1168             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1169             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1170             Heps             = _mm256_mul_pd(vfeps,H);
1171             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1172             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1173             velec            = _mm256_mul_pd(qq21,VV);
1174             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1175             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1176
1177             /* Update potential sum for this i atom from the interaction with this j atom. */
1178             velec            = _mm256_andnot_pd(dummy_mask,velec);
1179             velecsum         = _mm256_add_pd(velecsum,velec);
1180
1181             fscal            = felec;
1182
1183             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1184
1185             /* Calculate temporary vectorial force */
1186             tx               = _mm256_mul_pd(fscal,dx21);
1187             ty               = _mm256_mul_pd(fscal,dy21);
1188             tz               = _mm256_mul_pd(fscal,dz21);
1189
1190             /* Update vectorial force */
1191             fix2             = _mm256_add_pd(fix2,tx);
1192             fiy2             = _mm256_add_pd(fiy2,ty);
1193             fiz2             = _mm256_add_pd(fiz2,tz);
1194
1195             fjx1             = _mm256_add_pd(fjx1,tx);
1196             fjy1             = _mm256_add_pd(fjy1,ty);
1197             fjz1             = _mm256_add_pd(fjz1,tz);
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             r22              = _mm256_mul_pd(rsq22,rinv22);
1204             r22              = _mm256_andnot_pd(dummy_mask,r22);
1205
1206             /* Calculate table index by multiplying r with table scale and truncate to integer */
1207             rt               = _mm256_mul_pd(r22,vftabscale);
1208             vfitab           = _mm256_cvttpd_epi32(rt);
1209             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1210             vfitab           = _mm_slli_epi32(vfitab,2);
1211
1212             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1213             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1214             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1215             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1216             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1217             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1218             Heps             = _mm256_mul_pd(vfeps,H);
1219             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1220             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
1221             velec            = _mm256_mul_pd(qq22,VV);
1222             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1223             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1224
1225             /* Update potential sum for this i atom from the interaction with this j atom. */
1226             velec            = _mm256_andnot_pd(dummy_mask,velec);
1227             velecsum         = _mm256_add_pd(velecsum,velec);
1228
1229             fscal            = felec;
1230
1231             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1232
1233             /* Calculate temporary vectorial force */
1234             tx               = _mm256_mul_pd(fscal,dx22);
1235             ty               = _mm256_mul_pd(fscal,dy22);
1236             tz               = _mm256_mul_pd(fscal,dz22);
1237
1238             /* Update vectorial force */
1239             fix2             = _mm256_add_pd(fix2,tx);
1240             fiy2             = _mm256_add_pd(fiy2,ty);
1241             fiz2             = _mm256_add_pd(fiz2,tz);
1242
1243             fjx2             = _mm256_add_pd(fjx2,tx);
1244             fjy2             = _mm256_add_pd(fjy2,ty);
1245             fjz2             = _mm256_add_pd(fjz2,tz);
1246
1247             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1248             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1249             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1250             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1251
1252             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1253                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1254
1255             /* Inner loop uses 409 flops */
1256         }
1257
1258         /* End of innermost loop */
1259
1260         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1261                                                  f+i_coord_offset,fshift+i_shift_offset);
1262
1263         ggid                        = gid[iidx];
1264         /* Update potential energies */
1265         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1266         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1267
1268         /* Increment number of inner iterations */
1269         inneriter                  += j_index_end - j_index_start;
1270
1271         /* Outer loop uses 20 flops */
1272     }
1273
1274     /* Increment number of outer iterations */
1275     outeriter        += nri;
1276
1277     /* Update outer/inner flops */
1278
1279     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*409);
1280 }
1281 /*
1282  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_F_avx_256_double
1283  * Electrostatics interaction: CubicSplineTable
1284  * VdW interaction:            LennardJones
1285  * Geometry:                   Water3-Water3
1286  * Calculate force/pot:        Force
1287  */
1288 void
1289 nb_kernel_ElecCSTab_VdwLJ_GeomW3W3_F_avx_256_double
1290                     (t_nblist                    * gmx_restrict       nlist,
1291                      rvec                        * gmx_restrict          xx,
1292                      rvec                        * gmx_restrict          ff,
1293                      t_forcerec                  * gmx_restrict          fr,
1294                      t_mdatoms                   * gmx_restrict     mdatoms,
1295                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1296                      t_nrnb                      * gmx_restrict        nrnb)
1297 {
1298     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1299      * just 0 for non-waters.
1300      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
1301      * jnr indices corresponding to data put in the four positions in the SIMD register.
1302      */
1303     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1304     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1305     int              jnrA,jnrB,jnrC,jnrD;
1306     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1307     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1308     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1309     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1310     real             rcutoff_scalar;
1311     real             *shiftvec,*fshift,*x,*f;
1312     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1313     real             scratch[4*DIM];
1314     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1315     real *           vdwioffsetptr0;
1316     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1317     real *           vdwioffsetptr1;
1318     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1319     real *           vdwioffsetptr2;
1320     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1321     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1322     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1323     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1324     __m256d          jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1325     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1326     __m256d          jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1327     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1328     __m256d          dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1329     __m256d          dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1330     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1331     __m256d          dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1332     __m256d          dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1333     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1334     __m256d          dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1335     __m256d          dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1336     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
1337     real             *charge;
1338     int              nvdwtype;
1339     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1340     int              *vdwtype;
1341     real             *vdwparam;
1342     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
1343     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
1344     __m128i          vfitab;
1345     __m128i          ifour       = _mm_set1_epi32(4);
1346     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1347     real             *vftab;
1348     __m256d          dummy_mask,cutoff_mask;
1349     __m128           tmpmask0,tmpmask1;
1350     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
1351     __m256d          one     = _mm256_set1_pd(1.0);
1352     __m256d          two     = _mm256_set1_pd(2.0);
1353     x                = xx[0];
1354     f                = ff[0];
1355
1356     nri              = nlist->nri;
1357     iinr             = nlist->iinr;
1358     jindex           = nlist->jindex;
1359     jjnr             = nlist->jjnr;
1360     shiftidx         = nlist->shift;
1361     gid              = nlist->gid;
1362     shiftvec         = fr->shift_vec[0];
1363     fshift           = fr->fshift[0];
1364     facel            = _mm256_set1_pd(fr->epsfac);
1365     charge           = mdatoms->chargeA;
1366     nvdwtype         = fr->ntype;
1367     vdwparam         = fr->nbfp;
1368     vdwtype          = mdatoms->typeA;
1369
1370     vftab            = kernel_data->table_elec->data;
1371     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
1372
1373     /* Setup water-specific parameters */
1374     inr              = nlist->iinr[0];
1375     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
1376     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
1377     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
1378     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1379
1380     jq0              = _mm256_set1_pd(charge[inr+0]);
1381     jq1              = _mm256_set1_pd(charge[inr+1]);
1382     jq2              = _mm256_set1_pd(charge[inr+2]);
1383     vdwjidx0A        = 2*vdwtype[inr+0];
1384     qq00             = _mm256_mul_pd(iq0,jq0);
1385     c6_00            = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A]);
1386     c12_00           = _mm256_set1_pd(vdwioffsetptr0[vdwjidx0A+1]);
1387     qq01             = _mm256_mul_pd(iq0,jq1);
1388     qq02             = _mm256_mul_pd(iq0,jq2);
1389     qq10             = _mm256_mul_pd(iq1,jq0);
1390     qq11             = _mm256_mul_pd(iq1,jq1);
1391     qq12             = _mm256_mul_pd(iq1,jq2);
1392     qq20             = _mm256_mul_pd(iq2,jq0);
1393     qq21             = _mm256_mul_pd(iq2,jq1);
1394     qq22             = _mm256_mul_pd(iq2,jq2);
1395
1396     /* Avoid stupid compiler warnings */
1397     jnrA = jnrB = jnrC = jnrD = 0;
1398     j_coord_offsetA = 0;
1399     j_coord_offsetB = 0;
1400     j_coord_offsetC = 0;
1401     j_coord_offsetD = 0;
1402
1403     outeriter        = 0;
1404     inneriter        = 0;
1405
1406     for(iidx=0;iidx<4*DIM;iidx++)
1407     {
1408         scratch[iidx] = 0.0;
1409     }
1410
1411     /* Start outer loop over neighborlists */
1412     for(iidx=0; iidx<nri; iidx++)
1413     {
1414         /* Load shift vector for this list */
1415         i_shift_offset   = DIM*shiftidx[iidx];
1416
1417         /* Load limits for loop over neighbors */
1418         j_index_start    = jindex[iidx];
1419         j_index_end      = jindex[iidx+1];
1420
1421         /* Get outer coordinate index */
1422         inr              = iinr[iidx];
1423         i_coord_offset   = DIM*inr;
1424
1425         /* Load i particle coords and add shift vector */
1426         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1427                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1428
1429         fix0             = _mm256_setzero_pd();
1430         fiy0             = _mm256_setzero_pd();
1431         fiz0             = _mm256_setzero_pd();
1432         fix1             = _mm256_setzero_pd();
1433         fiy1             = _mm256_setzero_pd();
1434         fiz1             = _mm256_setzero_pd();
1435         fix2             = _mm256_setzero_pd();
1436         fiy2             = _mm256_setzero_pd();
1437         fiz2             = _mm256_setzero_pd();
1438
1439         /* Start inner kernel loop */
1440         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1441         {
1442
1443             /* Get j neighbor index, and coordinate index */
1444             jnrA             = jjnr[jidx];
1445             jnrB             = jjnr[jidx+1];
1446             jnrC             = jjnr[jidx+2];
1447             jnrD             = jjnr[jidx+3];
1448             j_coord_offsetA  = DIM*jnrA;
1449             j_coord_offsetB  = DIM*jnrB;
1450             j_coord_offsetC  = DIM*jnrC;
1451             j_coord_offsetD  = DIM*jnrD;
1452
1453             /* load j atom coordinates */
1454             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1455                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1456                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1457
1458             /* Calculate displacement vector */
1459             dx00             = _mm256_sub_pd(ix0,jx0);
1460             dy00             = _mm256_sub_pd(iy0,jy0);
1461             dz00             = _mm256_sub_pd(iz0,jz0);
1462             dx01             = _mm256_sub_pd(ix0,jx1);
1463             dy01             = _mm256_sub_pd(iy0,jy1);
1464             dz01             = _mm256_sub_pd(iz0,jz1);
1465             dx02             = _mm256_sub_pd(ix0,jx2);
1466             dy02             = _mm256_sub_pd(iy0,jy2);
1467             dz02             = _mm256_sub_pd(iz0,jz2);
1468             dx10             = _mm256_sub_pd(ix1,jx0);
1469             dy10             = _mm256_sub_pd(iy1,jy0);
1470             dz10             = _mm256_sub_pd(iz1,jz0);
1471             dx11             = _mm256_sub_pd(ix1,jx1);
1472             dy11             = _mm256_sub_pd(iy1,jy1);
1473             dz11             = _mm256_sub_pd(iz1,jz1);
1474             dx12             = _mm256_sub_pd(ix1,jx2);
1475             dy12             = _mm256_sub_pd(iy1,jy2);
1476             dz12             = _mm256_sub_pd(iz1,jz2);
1477             dx20             = _mm256_sub_pd(ix2,jx0);
1478             dy20             = _mm256_sub_pd(iy2,jy0);
1479             dz20             = _mm256_sub_pd(iz2,jz0);
1480             dx21             = _mm256_sub_pd(ix2,jx1);
1481             dy21             = _mm256_sub_pd(iy2,jy1);
1482             dz21             = _mm256_sub_pd(iz2,jz1);
1483             dx22             = _mm256_sub_pd(ix2,jx2);
1484             dy22             = _mm256_sub_pd(iy2,jy2);
1485             dz22             = _mm256_sub_pd(iz2,jz2);
1486
1487             /* Calculate squared distance and things based on it */
1488             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1489             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1490             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1491             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1492             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1493             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1494             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1495             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1496             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1497
1498             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1499             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1500             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1501             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1502             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1503             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1504             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1505             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1506             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1507
1508             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1509
1510             fjx0             = _mm256_setzero_pd();
1511             fjy0             = _mm256_setzero_pd();
1512             fjz0             = _mm256_setzero_pd();
1513             fjx1             = _mm256_setzero_pd();
1514             fjy1             = _mm256_setzero_pd();
1515             fjz1             = _mm256_setzero_pd();
1516             fjx2             = _mm256_setzero_pd();
1517             fjy2             = _mm256_setzero_pd();
1518             fjz2             = _mm256_setzero_pd();
1519
1520             /**************************
1521              * CALCULATE INTERACTIONS *
1522              **************************/
1523
1524             r00              = _mm256_mul_pd(rsq00,rinv00);
1525
1526             /* Calculate table index by multiplying r with table scale and truncate to integer */
1527             rt               = _mm256_mul_pd(r00,vftabscale);
1528             vfitab           = _mm256_cvttpd_epi32(rt);
1529             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1530             vfitab           = _mm_slli_epi32(vfitab,2);
1531
1532             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1533             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1534             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1535             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1536             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1537             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1538             Heps             = _mm256_mul_pd(vfeps,H);
1539             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1540             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1541             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1542
1543             /* LENNARD-JONES DISPERSION/REPULSION */
1544
1545             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1546             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1547
1548             fscal            = _mm256_add_pd(felec,fvdw);
1549
1550             /* Calculate temporary vectorial force */
1551             tx               = _mm256_mul_pd(fscal,dx00);
1552             ty               = _mm256_mul_pd(fscal,dy00);
1553             tz               = _mm256_mul_pd(fscal,dz00);
1554
1555             /* Update vectorial force */
1556             fix0             = _mm256_add_pd(fix0,tx);
1557             fiy0             = _mm256_add_pd(fiy0,ty);
1558             fiz0             = _mm256_add_pd(fiz0,tz);
1559
1560             fjx0             = _mm256_add_pd(fjx0,tx);
1561             fjy0             = _mm256_add_pd(fjy0,ty);
1562             fjz0             = _mm256_add_pd(fjz0,tz);
1563
1564             /**************************
1565              * CALCULATE INTERACTIONS *
1566              **************************/
1567
1568             r01              = _mm256_mul_pd(rsq01,rinv01);
1569
1570             /* Calculate table index by multiplying r with table scale and truncate to integer */
1571             rt               = _mm256_mul_pd(r01,vftabscale);
1572             vfitab           = _mm256_cvttpd_epi32(rt);
1573             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1574             vfitab           = _mm_slli_epi32(vfitab,2);
1575
1576             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1577             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1578             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1579             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1580             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1581             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1582             Heps             = _mm256_mul_pd(vfeps,H);
1583             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1584             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1585             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
1586
1587             fscal            = felec;
1588
1589             /* Calculate temporary vectorial force */
1590             tx               = _mm256_mul_pd(fscal,dx01);
1591             ty               = _mm256_mul_pd(fscal,dy01);
1592             tz               = _mm256_mul_pd(fscal,dz01);
1593
1594             /* Update vectorial force */
1595             fix0             = _mm256_add_pd(fix0,tx);
1596             fiy0             = _mm256_add_pd(fiy0,ty);
1597             fiz0             = _mm256_add_pd(fiz0,tz);
1598
1599             fjx1             = _mm256_add_pd(fjx1,tx);
1600             fjy1             = _mm256_add_pd(fjy1,ty);
1601             fjz1             = _mm256_add_pd(fjz1,tz);
1602
1603             /**************************
1604              * CALCULATE INTERACTIONS *
1605              **************************/
1606
1607             r02              = _mm256_mul_pd(rsq02,rinv02);
1608
1609             /* Calculate table index by multiplying r with table scale and truncate to integer */
1610             rt               = _mm256_mul_pd(r02,vftabscale);
1611             vfitab           = _mm256_cvttpd_epi32(rt);
1612             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1613             vfitab           = _mm_slli_epi32(vfitab,2);
1614
1615             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1616             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1617             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1618             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1619             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1620             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1621             Heps             = _mm256_mul_pd(vfeps,H);
1622             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1623             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1624             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
1625
1626             fscal            = felec;
1627
1628             /* Calculate temporary vectorial force */
1629             tx               = _mm256_mul_pd(fscal,dx02);
1630             ty               = _mm256_mul_pd(fscal,dy02);
1631             tz               = _mm256_mul_pd(fscal,dz02);
1632
1633             /* Update vectorial force */
1634             fix0             = _mm256_add_pd(fix0,tx);
1635             fiy0             = _mm256_add_pd(fiy0,ty);
1636             fiz0             = _mm256_add_pd(fiz0,tz);
1637
1638             fjx2             = _mm256_add_pd(fjx2,tx);
1639             fjy2             = _mm256_add_pd(fjy2,ty);
1640             fjz2             = _mm256_add_pd(fjz2,tz);
1641
1642             /**************************
1643              * CALCULATE INTERACTIONS *
1644              **************************/
1645
1646             r10              = _mm256_mul_pd(rsq10,rinv10);
1647
1648             /* Calculate table index by multiplying r with table scale and truncate to integer */
1649             rt               = _mm256_mul_pd(r10,vftabscale);
1650             vfitab           = _mm256_cvttpd_epi32(rt);
1651             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1652             vfitab           = _mm_slli_epi32(vfitab,2);
1653
1654             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1655             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1656             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1657             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1658             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1659             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1660             Heps             = _mm256_mul_pd(vfeps,H);
1661             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1662             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1663             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1664
1665             fscal            = felec;
1666
1667             /* Calculate temporary vectorial force */
1668             tx               = _mm256_mul_pd(fscal,dx10);
1669             ty               = _mm256_mul_pd(fscal,dy10);
1670             tz               = _mm256_mul_pd(fscal,dz10);
1671
1672             /* Update vectorial force */
1673             fix1             = _mm256_add_pd(fix1,tx);
1674             fiy1             = _mm256_add_pd(fiy1,ty);
1675             fiz1             = _mm256_add_pd(fiz1,tz);
1676
1677             fjx0             = _mm256_add_pd(fjx0,tx);
1678             fjy0             = _mm256_add_pd(fjy0,ty);
1679             fjz0             = _mm256_add_pd(fjz0,tz);
1680
1681             /**************************
1682              * CALCULATE INTERACTIONS *
1683              **************************/
1684
1685             r11              = _mm256_mul_pd(rsq11,rinv11);
1686
1687             /* Calculate table index by multiplying r with table scale and truncate to integer */
1688             rt               = _mm256_mul_pd(r11,vftabscale);
1689             vfitab           = _mm256_cvttpd_epi32(rt);
1690             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1691             vfitab           = _mm_slli_epi32(vfitab,2);
1692
1693             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1694             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1695             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1696             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1697             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1698             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1699             Heps             = _mm256_mul_pd(vfeps,H);
1700             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1701             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1702             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
1703
1704             fscal            = felec;
1705
1706             /* Calculate temporary vectorial force */
1707             tx               = _mm256_mul_pd(fscal,dx11);
1708             ty               = _mm256_mul_pd(fscal,dy11);
1709             tz               = _mm256_mul_pd(fscal,dz11);
1710
1711             /* Update vectorial force */
1712             fix1             = _mm256_add_pd(fix1,tx);
1713             fiy1             = _mm256_add_pd(fiy1,ty);
1714             fiz1             = _mm256_add_pd(fiz1,tz);
1715
1716             fjx1             = _mm256_add_pd(fjx1,tx);
1717             fjy1             = _mm256_add_pd(fjy1,ty);
1718             fjz1             = _mm256_add_pd(fjz1,tz);
1719
1720             /**************************
1721              * CALCULATE INTERACTIONS *
1722              **************************/
1723
1724             r12              = _mm256_mul_pd(rsq12,rinv12);
1725
1726             /* Calculate table index by multiplying r with table scale and truncate to integer */
1727             rt               = _mm256_mul_pd(r12,vftabscale);
1728             vfitab           = _mm256_cvttpd_epi32(rt);
1729             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1730             vfitab           = _mm_slli_epi32(vfitab,2);
1731
1732             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1733             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1734             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1735             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1736             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1737             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1738             Heps             = _mm256_mul_pd(vfeps,H);
1739             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1740             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1741             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
1742
1743             fscal            = felec;
1744
1745             /* Calculate temporary vectorial force */
1746             tx               = _mm256_mul_pd(fscal,dx12);
1747             ty               = _mm256_mul_pd(fscal,dy12);
1748             tz               = _mm256_mul_pd(fscal,dz12);
1749
1750             /* Update vectorial force */
1751             fix1             = _mm256_add_pd(fix1,tx);
1752             fiy1             = _mm256_add_pd(fiy1,ty);
1753             fiz1             = _mm256_add_pd(fiz1,tz);
1754
1755             fjx2             = _mm256_add_pd(fjx2,tx);
1756             fjy2             = _mm256_add_pd(fjy2,ty);
1757             fjz2             = _mm256_add_pd(fjz2,tz);
1758
1759             /**************************
1760              * CALCULATE INTERACTIONS *
1761              **************************/
1762
1763             r20              = _mm256_mul_pd(rsq20,rinv20);
1764
1765             /* Calculate table index by multiplying r with table scale and truncate to integer */
1766             rt               = _mm256_mul_pd(r20,vftabscale);
1767             vfitab           = _mm256_cvttpd_epi32(rt);
1768             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1769             vfitab           = _mm_slli_epi32(vfitab,2);
1770
1771             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1772             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1773             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1774             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1775             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1776             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1777             Heps             = _mm256_mul_pd(vfeps,H);
1778             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1779             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1780             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1781
1782             fscal            = felec;
1783
1784             /* Calculate temporary vectorial force */
1785             tx               = _mm256_mul_pd(fscal,dx20);
1786             ty               = _mm256_mul_pd(fscal,dy20);
1787             tz               = _mm256_mul_pd(fscal,dz20);
1788
1789             /* Update vectorial force */
1790             fix2             = _mm256_add_pd(fix2,tx);
1791             fiy2             = _mm256_add_pd(fiy2,ty);
1792             fiz2             = _mm256_add_pd(fiz2,tz);
1793
1794             fjx0             = _mm256_add_pd(fjx0,tx);
1795             fjy0             = _mm256_add_pd(fjy0,ty);
1796             fjz0             = _mm256_add_pd(fjz0,tz);
1797
1798             /**************************
1799              * CALCULATE INTERACTIONS *
1800              **************************/
1801
1802             r21              = _mm256_mul_pd(rsq21,rinv21);
1803
1804             /* Calculate table index by multiplying r with table scale and truncate to integer */
1805             rt               = _mm256_mul_pd(r21,vftabscale);
1806             vfitab           = _mm256_cvttpd_epi32(rt);
1807             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1808             vfitab           = _mm_slli_epi32(vfitab,2);
1809
1810             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1811             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1812             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1813             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1814             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1815             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1816             Heps             = _mm256_mul_pd(vfeps,H);
1817             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1818             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1819             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
1820
1821             fscal            = felec;
1822
1823             /* Calculate temporary vectorial force */
1824             tx               = _mm256_mul_pd(fscal,dx21);
1825             ty               = _mm256_mul_pd(fscal,dy21);
1826             tz               = _mm256_mul_pd(fscal,dz21);
1827
1828             /* Update vectorial force */
1829             fix2             = _mm256_add_pd(fix2,tx);
1830             fiy2             = _mm256_add_pd(fiy2,ty);
1831             fiz2             = _mm256_add_pd(fiz2,tz);
1832
1833             fjx1             = _mm256_add_pd(fjx1,tx);
1834             fjy1             = _mm256_add_pd(fjy1,ty);
1835             fjz1             = _mm256_add_pd(fjz1,tz);
1836
1837             /**************************
1838              * CALCULATE INTERACTIONS *
1839              **************************/
1840
1841             r22              = _mm256_mul_pd(rsq22,rinv22);
1842
1843             /* Calculate table index by multiplying r with table scale and truncate to integer */
1844             rt               = _mm256_mul_pd(r22,vftabscale);
1845             vfitab           = _mm256_cvttpd_epi32(rt);
1846             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1847             vfitab           = _mm_slli_epi32(vfitab,2);
1848
1849             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1850             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1851             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1852             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1853             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1854             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1855             Heps             = _mm256_mul_pd(vfeps,H);
1856             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1857             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1858             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
1859
1860             fscal            = felec;
1861
1862             /* Calculate temporary vectorial force */
1863             tx               = _mm256_mul_pd(fscal,dx22);
1864             ty               = _mm256_mul_pd(fscal,dy22);
1865             tz               = _mm256_mul_pd(fscal,dz22);
1866
1867             /* Update vectorial force */
1868             fix2             = _mm256_add_pd(fix2,tx);
1869             fiy2             = _mm256_add_pd(fiy2,ty);
1870             fiz2             = _mm256_add_pd(fiz2,tz);
1871
1872             fjx2             = _mm256_add_pd(fjx2,tx);
1873             fjy2             = _mm256_add_pd(fjy2,ty);
1874             fjz2             = _mm256_add_pd(fjz2,tz);
1875
1876             fjptrA             = f+j_coord_offsetA;
1877             fjptrB             = f+j_coord_offsetB;
1878             fjptrC             = f+j_coord_offsetC;
1879             fjptrD             = f+j_coord_offsetD;
1880
1881             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
1882                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1883
1884             /* Inner loop uses 359 flops */
1885         }
1886
1887         if(jidx<j_index_end)
1888         {
1889
1890             /* Get j neighbor index, and coordinate index */
1891             jnrlistA         = jjnr[jidx];
1892             jnrlistB         = jjnr[jidx+1];
1893             jnrlistC         = jjnr[jidx+2];
1894             jnrlistD         = jjnr[jidx+3];
1895             /* Sign of each element will be negative for non-real atoms.
1896              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1897              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1898              */
1899             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1900
1901             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1902             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1903             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1904
1905             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1906             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1907             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1908             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1909             j_coord_offsetA  = DIM*jnrA;
1910             j_coord_offsetB  = DIM*jnrB;
1911             j_coord_offsetC  = DIM*jnrC;
1912             j_coord_offsetD  = DIM*jnrD;
1913
1914             /* load j atom coordinates */
1915             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1916                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1917                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1918
1919             /* Calculate displacement vector */
1920             dx00             = _mm256_sub_pd(ix0,jx0);
1921             dy00             = _mm256_sub_pd(iy0,jy0);
1922             dz00             = _mm256_sub_pd(iz0,jz0);
1923             dx01             = _mm256_sub_pd(ix0,jx1);
1924             dy01             = _mm256_sub_pd(iy0,jy1);
1925             dz01             = _mm256_sub_pd(iz0,jz1);
1926             dx02             = _mm256_sub_pd(ix0,jx2);
1927             dy02             = _mm256_sub_pd(iy0,jy2);
1928             dz02             = _mm256_sub_pd(iz0,jz2);
1929             dx10             = _mm256_sub_pd(ix1,jx0);
1930             dy10             = _mm256_sub_pd(iy1,jy0);
1931             dz10             = _mm256_sub_pd(iz1,jz0);
1932             dx11             = _mm256_sub_pd(ix1,jx1);
1933             dy11             = _mm256_sub_pd(iy1,jy1);
1934             dz11             = _mm256_sub_pd(iz1,jz1);
1935             dx12             = _mm256_sub_pd(ix1,jx2);
1936             dy12             = _mm256_sub_pd(iy1,jy2);
1937             dz12             = _mm256_sub_pd(iz1,jz2);
1938             dx20             = _mm256_sub_pd(ix2,jx0);
1939             dy20             = _mm256_sub_pd(iy2,jy0);
1940             dz20             = _mm256_sub_pd(iz2,jz0);
1941             dx21             = _mm256_sub_pd(ix2,jx1);
1942             dy21             = _mm256_sub_pd(iy2,jy1);
1943             dz21             = _mm256_sub_pd(iz2,jz1);
1944             dx22             = _mm256_sub_pd(ix2,jx2);
1945             dy22             = _mm256_sub_pd(iy2,jy2);
1946             dz22             = _mm256_sub_pd(iz2,jz2);
1947
1948             /* Calculate squared distance and things based on it */
1949             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1950             rsq01            = gmx_mm256_calc_rsq_pd(dx01,dy01,dz01);
1951             rsq02            = gmx_mm256_calc_rsq_pd(dx02,dy02,dz02);
1952             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1953             rsq11            = gmx_mm256_calc_rsq_pd(dx11,dy11,dz11);
1954             rsq12            = gmx_mm256_calc_rsq_pd(dx12,dy12,dz12);
1955             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1956             rsq21            = gmx_mm256_calc_rsq_pd(dx21,dy21,dz21);
1957             rsq22            = gmx_mm256_calc_rsq_pd(dx22,dy22,dz22);
1958
1959             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1960             rinv01           = gmx_mm256_invsqrt_pd(rsq01);
1961             rinv02           = gmx_mm256_invsqrt_pd(rsq02);
1962             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1963             rinv11           = gmx_mm256_invsqrt_pd(rsq11);
1964             rinv12           = gmx_mm256_invsqrt_pd(rsq12);
1965             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1966             rinv21           = gmx_mm256_invsqrt_pd(rsq21);
1967             rinv22           = gmx_mm256_invsqrt_pd(rsq22);
1968
1969             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1970
1971             fjx0             = _mm256_setzero_pd();
1972             fjy0             = _mm256_setzero_pd();
1973             fjz0             = _mm256_setzero_pd();
1974             fjx1             = _mm256_setzero_pd();
1975             fjy1             = _mm256_setzero_pd();
1976             fjz1             = _mm256_setzero_pd();
1977             fjx2             = _mm256_setzero_pd();
1978             fjy2             = _mm256_setzero_pd();
1979             fjz2             = _mm256_setzero_pd();
1980
1981             /**************************
1982              * CALCULATE INTERACTIONS *
1983              **************************/
1984
1985             r00              = _mm256_mul_pd(rsq00,rinv00);
1986             r00              = _mm256_andnot_pd(dummy_mask,r00);
1987
1988             /* Calculate table index by multiplying r with table scale and truncate to integer */
1989             rt               = _mm256_mul_pd(r00,vftabscale);
1990             vfitab           = _mm256_cvttpd_epi32(rt);
1991             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1992             vfitab           = _mm_slli_epi32(vfitab,2);
1993
1994             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1995             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1996             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1997             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1998             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1999             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2000             Heps             = _mm256_mul_pd(vfeps,H);
2001             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2002             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2003             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
2004
2005             /* LENNARD-JONES DISPERSION/REPULSION */
2006
2007             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
2008             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
2009
2010             fscal            = _mm256_add_pd(felec,fvdw);
2011
2012             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2013
2014             /* Calculate temporary vectorial force */
2015             tx               = _mm256_mul_pd(fscal,dx00);
2016             ty               = _mm256_mul_pd(fscal,dy00);
2017             tz               = _mm256_mul_pd(fscal,dz00);
2018
2019             /* Update vectorial force */
2020             fix0             = _mm256_add_pd(fix0,tx);
2021             fiy0             = _mm256_add_pd(fiy0,ty);
2022             fiz0             = _mm256_add_pd(fiz0,tz);
2023
2024             fjx0             = _mm256_add_pd(fjx0,tx);
2025             fjy0             = _mm256_add_pd(fjy0,ty);
2026             fjz0             = _mm256_add_pd(fjz0,tz);
2027
2028             /**************************
2029              * CALCULATE INTERACTIONS *
2030              **************************/
2031
2032             r01              = _mm256_mul_pd(rsq01,rinv01);
2033             r01              = _mm256_andnot_pd(dummy_mask,r01);
2034
2035             /* Calculate table index by multiplying r with table scale and truncate to integer */
2036             rt               = _mm256_mul_pd(r01,vftabscale);
2037             vfitab           = _mm256_cvttpd_epi32(rt);
2038             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2039             vfitab           = _mm_slli_epi32(vfitab,2);
2040
2041             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2042             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2043             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2044             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2045             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2046             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2047             Heps             = _mm256_mul_pd(vfeps,H);
2048             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2049             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2050             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq01,FF),_mm256_mul_pd(vftabscale,rinv01)));
2051
2052             fscal            = felec;
2053
2054             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2055
2056             /* Calculate temporary vectorial force */
2057             tx               = _mm256_mul_pd(fscal,dx01);
2058             ty               = _mm256_mul_pd(fscal,dy01);
2059             tz               = _mm256_mul_pd(fscal,dz01);
2060
2061             /* Update vectorial force */
2062             fix0             = _mm256_add_pd(fix0,tx);
2063             fiy0             = _mm256_add_pd(fiy0,ty);
2064             fiz0             = _mm256_add_pd(fiz0,tz);
2065
2066             fjx1             = _mm256_add_pd(fjx1,tx);
2067             fjy1             = _mm256_add_pd(fjy1,ty);
2068             fjz1             = _mm256_add_pd(fjz1,tz);
2069
2070             /**************************
2071              * CALCULATE INTERACTIONS *
2072              **************************/
2073
2074             r02              = _mm256_mul_pd(rsq02,rinv02);
2075             r02              = _mm256_andnot_pd(dummy_mask,r02);
2076
2077             /* Calculate table index by multiplying r with table scale and truncate to integer */
2078             rt               = _mm256_mul_pd(r02,vftabscale);
2079             vfitab           = _mm256_cvttpd_epi32(rt);
2080             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2081             vfitab           = _mm_slli_epi32(vfitab,2);
2082
2083             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2084             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2085             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2086             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2087             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2088             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2089             Heps             = _mm256_mul_pd(vfeps,H);
2090             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2091             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2092             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq02,FF),_mm256_mul_pd(vftabscale,rinv02)));
2093
2094             fscal            = felec;
2095
2096             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2097
2098             /* Calculate temporary vectorial force */
2099             tx               = _mm256_mul_pd(fscal,dx02);
2100             ty               = _mm256_mul_pd(fscal,dy02);
2101             tz               = _mm256_mul_pd(fscal,dz02);
2102
2103             /* Update vectorial force */
2104             fix0             = _mm256_add_pd(fix0,tx);
2105             fiy0             = _mm256_add_pd(fiy0,ty);
2106             fiz0             = _mm256_add_pd(fiz0,tz);
2107
2108             fjx2             = _mm256_add_pd(fjx2,tx);
2109             fjy2             = _mm256_add_pd(fjy2,ty);
2110             fjz2             = _mm256_add_pd(fjz2,tz);
2111
2112             /**************************
2113              * CALCULATE INTERACTIONS *
2114              **************************/
2115
2116             r10              = _mm256_mul_pd(rsq10,rinv10);
2117             r10              = _mm256_andnot_pd(dummy_mask,r10);
2118
2119             /* Calculate table index by multiplying r with table scale and truncate to integer */
2120             rt               = _mm256_mul_pd(r10,vftabscale);
2121             vfitab           = _mm256_cvttpd_epi32(rt);
2122             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2123             vfitab           = _mm_slli_epi32(vfitab,2);
2124
2125             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2126             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2127             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2128             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2129             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2130             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2131             Heps             = _mm256_mul_pd(vfeps,H);
2132             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2133             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2134             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
2135
2136             fscal            = felec;
2137
2138             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2139
2140             /* Calculate temporary vectorial force */
2141             tx               = _mm256_mul_pd(fscal,dx10);
2142             ty               = _mm256_mul_pd(fscal,dy10);
2143             tz               = _mm256_mul_pd(fscal,dz10);
2144
2145             /* Update vectorial force */
2146             fix1             = _mm256_add_pd(fix1,tx);
2147             fiy1             = _mm256_add_pd(fiy1,ty);
2148             fiz1             = _mm256_add_pd(fiz1,tz);
2149
2150             fjx0             = _mm256_add_pd(fjx0,tx);
2151             fjy0             = _mm256_add_pd(fjy0,ty);
2152             fjz0             = _mm256_add_pd(fjz0,tz);
2153
2154             /**************************
2155              * CALCULATE INTERACTIONS *
2156              **************************/
2157
2158             r11              = _mm256_mul_pd(rsq11,rinv11);
2159             r11              = _mm256_andnot_pd(dummy_mask,r11);
2160
2161             /* Calculate table index by multiplying r with table scale and truncate to integer */
2162             rt               = _mm256_mul_pd(r11,vftabscale);
2163             vfitab           = _mm256_cvttpd_epi32(rt);
2164             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2165             vfitab           = _mm_slli_epi32(vfitab,2);
2166
2167             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2168             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2169             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2170             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2171             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2172             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2173             Heps             = _mm256_mul_pd(vfeps,H);
2174             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2175             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2176             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq11,FF),_mm256_mul_pd(vftabscale,rinv11)));
2177
2178             fscal            = felec;
2179
2180             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2181
2182             /* Calculate temporary vectorial force */
2183             tx               = _mm256_mul_pd(fscal,dx11);
2184             ty               = _mm256_mul_pd(fscal,dy11);
2185             tz               = _mm256_mul_pd(fscal,dz11);
2186
2187             /* Update vectorial force */
2188             fix1             = _mm256_add_pd(fix1,tx);
2189             fiy1             = _mm256_add_pd(fiy1,ty);
2190             fiz1             = _mm256_add_pd(fiz1,tz);
2191
2192             fjx1             = _mm256_add_pd(fjx1,tx);
2193             fjy1             = _mm256_add_pd(fjy1,ty);
2194             fjz1             = _mm256_add_pd(fjz1,tz);
2195
2196             /**************************
2197              * CALCULATE INTERACTIONS *
2198              **************************/
2199
2200             r12              = _mm256_mul_pd(rsq12,rinv12);
2201             r12              = _mm256_andnot_pd(dummy_mask,r12);
2202
2203             /* Calculate table index by multiplying r with table scale and truncate to integer */
2204             rt               = _mm256_mul_pd(r12,vftabscale);
2205             vfitab           = _mm256_cvttpd_epi32(rt);
2206             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2207             vfitab           = _mm_slli_epi32(vfitab,2);
2208
2209             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2210             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2211             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2212             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2213             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2214             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2215             Heps             = _mm256_mul_pd(vfeps,H);
2216             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2217             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2218             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq12,FF),_mm256_mul_pd(vftabscale,rinv12)));
2219
2220             fscal            = felec;
2221
2222             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2223
2224             /* Calculate temporary vectorial force */
2225             tx               = _mm256_mul_pd(fscal,dx12);
2226             ty               = _mm256_mul_pd(fscal,dy12);
2227             tz               = _mm256_mul_pd(fscal,dz12);
2228
2229             /* Update vectorial force */
2230             fix1             = _mm256_add_pd(fix1,tx);
2231             fiy1             = _mm256_add_pd(fiy1,ty);
2232             fiz1             = _mm256_add_pd(fiz1,tz);
2233
2234             fjx2             = _mm256_add_pd(fjx2,tx);
2235             fjy2             = _mm256_add_pd(fjy2,ty);
2236             fjz2             = _mm256_add_pd(fjz2,tz);
2237
2238             /**************************
2239              * CALCULATE INTERACTIONS *
2240              **************************/
2241
2242             r20              = _mm256_mul_pd(rsq20,rinv20);
2243             r20              = _mm256_andnot_pd(dummy_mask,r20);
2244
2245             /* Calculate table index by multiplying r with table scale and truncate to integer */
2246             rt               = _mm256_mul_pd(r20,vftabscale);
2247             vfitab           = _mm256_cvttpd_epi32(rt);
2248             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2249             vfitab           = _mm_slli_epi32(vfitab,2);
2250
2251             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2252             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2253             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2254             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2255             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2256             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2257             Heps             = _mm256_mul_pd(vfeps,H);
2258             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2259             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2260             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
2261
2262             fscal            = felec;
2263
2264             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2265
2266             /* Calculate temporary vectorial force */
2267             tx               = _mm256_mul_pd(fscal,dx20);
2268             ty               = _mm256_mul_pd(fscal,dy20);
2269             tz               = _mm256_mul_pd(fscal,dz20);
2270
2271             /* Update vectorial force */
2272             fix2             = _mm256_add_pd(fix2,tx);
2273             fiy2             = _mm256_add_pd(fiy2,ty);
2274             fiz2             = _mm256_add_pd(fiz2,tz);
2275
2276             fjx0             = _mm256_add_pd(fjx0,tx);
2277             fjy0             = _mm256_add_pd(fjy0,ty);
2278             fjz0             = _mm256_add_pd(fjz0,tz);
2279
2280             /**************************
2281              * CALCULATE INTERACTIONS *
2282              **************************/
2283
2284             r21              = _mm256_mul_pd(rsq21,rinv21);
2285             r21              = _mm256_andnot_pd(dummy_mask,r21);
2286
2287             /* Calculate table index by multiplying r with table scale and truncate to integer */
2288             rt               = _mm256_mul_pd(r21,vftabscale);
2289             vfitab           = _mm256_cvttpd_epi32(rt);
2290             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2291             vfitab           = _mm_slli_epi32(vfitab,2);
2292
2293             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2294             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2295             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2296             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2297             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2298             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2299             Heps             = _mm256_mul_pd(vfeps,H);
2300             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2301             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2302             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq21,FF),_mm256_mul_pd(vftabscale,rinv21)));
2303
2304             fscal            = felec;
2305
2306             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2307
2308             /* Calculate temporary vectorial force */
2309             tx               = _mm256_mul_pd(fscal,dx21);
2310             ty               = _mm256_mul_pd(fscal,dy21);
2311             tz               = _mm256_mul_pd(fscal,dz21);
2312
2313             /* Update vectorial force */
2314             fix2             = _mm256_add_pd(fix2,tx);
2315             fiy2             = _mm256_add_pd(fiy2,ty);
2316             fiz2             = _mm256_add_pd(fiz2,tz);
2317
2318             fjx1             = _mm256_add_pd(fjx1,tx);
2319             fjy1             = _mm256_add_pd(fjy1,ty);
2320             fjz1             = _mm256_add_pd(fjz1,tz);
2321
2322             /**************************
2323              * CALCULATE INTERACTIONS *
2324              **************************/
2325
2326             r22              = _mm256_mul_pd(rsq22,rinv22);
2327             r22              = _mm256_andnot_pd(dummy_mask,r22);
2328
2329             /* Calculate table index by multiplying r with table scale and truncate to integer */
2330             rt               = _mm256_mul_pd(r22,vftabscale);
2331             vfitab           = _mm256_cvttpd_epi32(rt);
2332             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
2333             vfitab           = _mm_slli_epi32(vfitab,2);
2334
2335             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2336             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
2337             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
2338             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
2339             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
2340             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
2341             Heps             = _mm256_mul_pd(vfeps,H);
2342             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
2343             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
2344             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq22,FF),_mm256_mul_pd(vftabscale,rinv22)));
2345
2346             fscal            = felec;
2347
2348             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
2349
2350             /* Calculate temporary vectorial force */
2351             tx               = _mm256_mul_pd(fscal,dx22);
2352             ty               = _mm256_mul_pd(fscal,dy22);
2353             tz               = _mm256_mul_pd(fscal,dz22);
2354
2355             /* Update vectorial force */
2356             fix2             = _mm256_add_pd(fix2,tx);
2357             fiy2             = _mm256_add_pd(fiy2,ty);
2358             fiz2             = _mm256_add_pd(fiz2,tz);
2359
2360             fjx2             = _mm256_add_pd(fjx2,tx);
2361             fjy2             = _mm256_add_pd(fjy2,ty);
2362             fjz2             = _mm256_add_pd(fjz2,tz);
2363
2364             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2365             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2366             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2367             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2368
2369             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
2370                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2371
2372             /* Inner loop uses 368 flops */
2373         }
2374
2375         /* End of innermost loop */
2376
2377         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2378                                                  f+i_coord_offset,fshift+i_shift_offset);
2379
2380         /* Increment number of inner iterations */
2381         inneriter                  += j_index_end - j_index_start;
2382
2383         /* Outer loop uses 18 flops */
2384     }
2385
2386     /* Increment number of outer iterations */
2387     outeriter        += nri;
2388
2389     /* Update outer/inner flops */
2390
2391     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*368);
2392 }