src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_VF_avx_256_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  95     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  97     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 100     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 101     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 103     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 104     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 105     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 106     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 107     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 108     real             *charge;
 109     int              nvdwtype;
 110     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 111     int              *vdwtype;
 112     real             *vdwparam;
 113     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 114     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 115     __m256i          vfitab;
 116     __m128i          vfitab_lo,vfitab_hi;
 117     __m128i          ifour       = _mm_set1_epi32(4);
 118     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 119     real             *vftab;
 120     __m256           dummy_mask,cutoff_mask;
 121     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 122     __m256           one     = _mm256_set1_ps(1.0);
 123     __m256           two     = _mm256_set1_ps(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_ps(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_vdw->data;
 142     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 147     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 148     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 149     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 150
 151     jq0              = _mm256_set1_ps(charge[inr+0]);
 152     jq1              = _mm256_set1_ps(charge[inr+1]);
 153     jq2              = _mm256_set1_ps(charge[inr+2]);
 154     vdwjidx0A        = 2*vdwtype[inr+0];
 155     qq00             = _mm256_mul_ps(iq0,jq0);
 156     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 157     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 158     qq01             = _mm256_mul_ps(iq0,jq1);
 159     qq02             = _mm256_mul_ps(iq0,jq2);
 160     qq10             = _mm256_mul_ps(iq1,jq0);
 161     qq11             = _mm256_mul_ps(iq1,jq1);
 162     qq12             = _mm256_mul_ps(iq1,jq2);
 163     qq20             = _mm256_mul_ps(iq2,jq0);
 164     qq21             = _mm256_mul_ps(iq2,jq1);
 165     qq22             = _mm256_mul_ps(iq2,jq2);
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173     j_coord_offsetE = 0;
 174     j_coord_offsetF = 0;
 175     j_coord_offsetG = 0;
 176     j_coord_offsetH = 0;
 177
 178     outeriter        = 0;
 179     inneriter        = 0;
 180
 181     for(iidx=0;iidx<4*DIM;iidx++)
 182     {
 183         scratch[iidx] = 0.0;
 184     }
 185
 186     /* Start outer loop over neighborlists */
 187     for(iidx=0; iidx<nri; iidx++)
 188     {
 189         /* Load shift vector for this list */
 190         i_shift_offset   = DIM*shiftidx[iidx];
 191
 192         /* Load limits for loop over neighbors */
 193         j_index_start    = jindex[iidx];
 194         j_index_end      = jindex[iidx+1];
 195
 196         /* Get outer coordinate index */
 197         inr              = iinr[iidx];
 198         i_coord_offset   = DIM*inr;
 199
 200         /* Load i particle coords and add shift vector */
 201         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 202                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 203
 204         fix0             = _mm256_setzero_ps();
 205         fiy0             = _mm256_setzero_ps();
 206         fiz0             = _mm256_setzero_ps();
 207         fix1             = _mm256_setzero_ps();
 208         fiy1             = _mm256_setzero_ps();
 209         fiz1             = _mm256_setzero_ps();
 210         fix2             = _mm256_setzero_ps();
 211         fiy2             = _mm256_setzero_ps();
 212         fiz2             = _mm256_setzero_ps();
 213
 214         /* Reset potential sums */
 215         velecsum         = _mm256_setzero_ps();
 216         vvdwsum          = _mm256_setzero_ps();
 217
 218         /* Start inner kernel loop */
 219         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 220         {
 221
 222             /* Get j neighbor index, and coordinate index */
 223             jnrA             = jjnr[jidx];
 224             jnrB             = jjnr[jidx+1];
 225             jnrC             = jjnr[jidx+2];
 226             jnrD             = jjnr[jidx+3];
 227             jnrE             = jjnr[jidx+4];
 228             jnrF             = jjnr[jidx+5];
 229             jnrG             = jjnr[jidx+6];
 230             jnrH             = jjnr[jidx+7];
 231             j_coord_offsetA  = DIM*jnrA;
 232             j_coord_offsetB  = DIM*jnrB;
 233             j_coord_offsetC  = DIM*jnrC;
 234             j_coord_offsetD  = DIM*jnrD;
 235             j_coord_offsetE  = DIM*jnrE;
 236             j_coord_offsetF  = DIM*jnrF;
 237             j_coord_offsetG  = DIM*jnrG;
 238             j_coord_offsetH  = DIM*jnrH;
 239
 240             /* load j atom coordinates */
 241             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 242                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 243                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 244                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 245                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 246
 247             /* Calculate displacement vector */
 248             dx00             = _mm256_sub_ps(ix0,jx0);
 249             dy00             = _mm256_sub_ps(iy0,jy0);
 250             dz00             = _mm256_sub_ps(iz0,jz0);
 251             dx01             = _mm256_sub_ps(ix0,jx1);
 252             dy01             = _mm256_sub_ps(iy0,jy1);
 253             dz01             = _mm256_sub_ps(iz0,jz1);
 254             dx02             = _mm256_sub_ps(ix0,jx2);
 255             dy02             = _mm256_sub_ps(iy0,jy2);
 256             dz02             = _mm256_sub_ps(iz0,jz2);
 257             dx10             = _mm256_sub_ps(ix1,jx0);
 258             dy10             = _mm256_sub_ps(iy1,jy0);
 259             dz10             = _mm256_sub_ps(iz1,jz0);
 260             dx11             = _mm256_sub_ps(ix1,jx1);
 261             dy11             = _mm256_sub_ps(iy1,jy1);
 262             dz11             = _mm256_sub_ps(iz1,jz1);
 263             dx12             = _mm256_sub_ps(ix1,jx2);
 264             dy12             = _mm256_sub_ps(iy1,jy2);
 265             dz12             = _mm256_sub_ps(iz1,jz2);
 266             dx20             = _mm256_sub_ps(ix2,jx0);
 267             dy20             = _mm256_sub_ps(iy2,jy0);
 268             dz20             = _mm256_sub_ps(iz2,jz0);
 269             dx21             = _mm256_sub_ps(ix2,jx1);
 270             dy21             = _mm256_sub_ps(iy2,jy1);
 271             dz21             = _mm256_sub_ps(iz2,jz1);
 272             dx22             = _mm256_sub_ps(ix2,jx2);
 273             dy22             = _mm256_sub_ps(iy2,jy2);
 274             dz22             = _mm256_sub_ps(iz2,jz2);
 275
 276             /* Calculate squared distance and things based on it */
 277             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 278             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 279             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 280             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 281             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 282             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 283             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 284             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 285             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 286
 287             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 288             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 289             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 290             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 291             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 292             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 293             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 294             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 295             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 296
 297             fjx0             = _mm256_setzero_ps();
 298             fjy0             = _mm256_setzero_ps();
 299             fjz0             = _mm256_setzero_ps();
 300             fjx1             = _mm256_setzero_ps();
 301             fjy1             = _mm256_setzero_ps();
 302             fjz1             = _mm256_setzero_ps();
 303             fjx2             = _mm256_setzero_ps();
 304             fjy2             = _mm256_setzero_ps();
 305             fjz2             = _mm256_setzero_ps();
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             r00              = _mm256_mul_ps(rsq00,rinv00);
 312
 313             /* Calculate table index by multiplying r with table scale and truncate to integer */
 314             rt               = _mm256_mul_ps(r00,vftabscale);
 315             vfitab           = _mm256_cvttps_epi32(rt);
 316             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 317             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 318             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 319             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 320             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 321             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 322
 323             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 324             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 326             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 327                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 328             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 329                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 330             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 331                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 332             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 333             Heps             = _mm256_mul_ps(vfeps,H);
 334             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 335             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 336             velec            = _mm256_mul_ps(qq00,VV);
 337             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 338             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 339
 340             /* CUBIC SPLINE TABLE DISPERSION */
 341             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 342             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 343             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 344                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 345             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 346                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 347             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 348                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 349             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 350                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 351             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 352             Heps             = _mm256_mul_ps(vfeps,H);
 353             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 354             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 355             vvdw6            = _mm256_mul_ps(c6_00,VV);
 356             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 357             fvdw6            = _mm256_mul_ps(c6_00,FF);
 358
 359             /* CUBIC SPLINE TABLE REPULSION */
 360             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 361             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 362             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 363                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 364             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 365                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 366             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 367                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 368             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 369                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 370             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 371             Heps             = _mm256_mul_ps(vfeps,H);
 372             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 373             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 374             vvdw12           = _mm256_mul_ps(c12_00,VV);
 375             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 376             fvdw12           = _mm256_mul_ps(c12_00,FF);
 377             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 378             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velecsum         = _mm256_add_ps(velecsum,velec);
 382             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 383
 384             fscal            = _mm256_add_ps(felec,fvdw);
 385
 386             /* Calculate temporary vectorial force */
 387             tx               = _mm256_mul_ps(fscal,dx00);
 388             ty               = _mm256_mul_ps(fscal,dy00);
 389             tz               = _mm256_mul_ps(fscal,dz00);
 390
 391             /* Update vectorial force */
 392             fix0             = _mm256_add_ps(fix0,tx);
 393             fiy0             = _mm256_add_ps(fiy0,ty);
 394             fiz0             = _mm256_add_ps(fiz0,tz);
 395
 396             fjx0             = _mm256_add_ps(fjx0,tx);
 397             fjy0             = _mm256_add_ps(fjy0,ty);
 398             fjz0             = _mm256_add_ps(fjz0,tz);
 399
 400             /**************************
 401              * CALCULATE INTERACTIONS *
 402              **************************/
 403
 404             r01              = _mm256_mul_ps(rsq01,rinv01);
 405
 406             /* Calculate table index by multiplying r with table scale and truncate to integer */
 407             rt               = _mm256_mul_ps(r01,vftabscale);
 408             vfitab           = _mm256_cvttps_epi32(rt);
 409             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 410             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 411             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 412             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 413             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 414             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 415
 416             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 417             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 418                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 419             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 420                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 421             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 422                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 423             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 424                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 425             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 426             Heps             = _mm256_mul_ps(vfeps,H);
 427             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 428             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 429             velec            = _mm256_mul_ps(qq01,VV);
 430             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 431             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velecsum         = _mm256_add_ps(velecsum,velec);
 435
 436             fscal            = felec;
 437
 438             /* Calculate temporary vectorial force */
 439             tx               = _mm256_mul_ps(fscal,dx01);
 440             ty               = _mm256_mul_ps(fscal,dy01);
 441             tz               = _mm256_mul_ps(fscal,dz01);
 442
 443             /* Update vectorial force */
 444             fix0             = _mm256_add_ps(fix0,tx);
 445             fiy0             = _mm256_add_ps(fiy0,ty);
 446             fiz0             = _mm256_add_ps(fiz0,tz);
 447
 448             fjx1             = _mm256_add_ps(fjx1,tx);
 449             fjy1             = _mm256_add_ps(fjy1,ty);
 450             fjz1             = _mm256_add_ps(fjz1,tz);
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             r02              = _mm256_mul_ps(rsq02,rinv02);
 457
 458             /* Calculate table index by multiplying r with table scale and truncate to integer */
 459             rt               = _mm256_mul_ps(r02,vftabscale);
 460             vfitab           = _mm256_cvttps_epi32(rt);
 461             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 462             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 463             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 464             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 465             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 466             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 467
 468             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 469             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 470                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 471             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 472                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 473             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 474                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 475             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 476                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 477             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 478             Heps             = _mm256_mul_ps(vfeps,H);
 479             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 480             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 481             velec            = _mm256_mul_ps(qq02,VV);
 482             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 483             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
 484
 485             /* Update potential sum for this i atom from the interaction with this j atom. */
 486             velecsum         = _mm256_add_ps(velecsum,velec);
 487
 488             fscal            = felec;
 489
 490             /* Calculate temporary vectorial force */
 491             tx               = _mm256_mul_ps(fscal,dx02);
 492             ty               = _mm256_mul_ps(fscal,dy02);
 493             tz               = _mm256_mul_ps(fscal,dz02);
 494
 495             /* Update vectorial force */
 496             fix0             = _mm256_add_ps(fix0,tx);
 497             fiy0             = _mm256_add_ps(fiy0,ty);
 498             fiz0             = _mm256_add_ps(fiz0,tz);
 499
 500             fjx2             = _mm256_add_ps(fjx2,tx);
 501             fjy2             = _mm256_add_ps(fjy2,ty);
 502             fjz2             = _mm256_add_ps(fjz2,tz);
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             r10              = _mm256_mul_ps(rsq10,rinv10);
 509
 510             /* Calculate table index by multiplying r with table scale and truncate to integer */
 511             rt               = _mm256_mul_ps(r10,vftabscale);
 512             vfitab           = _mm256_cvttps_epi32(rt);
 513             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 514             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 515             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 516             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 517             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 518             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 519
 520             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 521             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 522                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 523             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 524                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 525             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 526                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 527             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 528                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 529             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 530             Heps             = _mm256_mul_ps(vfeps,H);
 531             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 532             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 533             velec            = _mm256_mul_ps(qq10,VV);
 534             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 535             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velecsum         = _mm256_add_ps(velecsum,velec);
 539
 540             fscal            = felec;
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm256_mul_ps(fscal,dx10);
 544             ty               = _mm256_mul_ps(fscal,dy10);
 545             tz               = _mm256_mul_ps(fscal,dz10);
 546
 547             /* Update vectorial force */
 548             fix1             = _mm256_add_ps(fix1,tx);
 549             fiy1             = _mm256_add_ps(fiy1,ty);
 550             fiz1             = _mm256_add_ps(fiz1,tz);
 551
 552             fjx0             = _mm256_add_ps(fjx0,tx);
 553             fjy0             = _mm256_add_ps(fjy0,ty);
 554             fjz0             = _mm256_add_ps(fjz0,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r11              = _mm256_mul_ps(rsq11,rinv11);
 561
 562             /* Calculate table index by multiplying r with table scale and truncate to integer */
 563             rt               = _mm256_mul_ps(r11,vftabscale);
 564             vfitab           = _mm256_cvttps_epi32(rt);
 565             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 566             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 567             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 568             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 569             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 570             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 571
 572             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 573             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 574                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 575             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 576                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 577             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 578                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 579             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 580                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 581             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 582             Heps             = _mm256_mul_ps(vfeps,H);
 583             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 584             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 585             velec            = _mm256_mul_ps(qq11,VV);
 586             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 587             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
 588
 589             /* Update potential sum for this i atom from the interaction with this j atom. */
 590             velecsum         = _mm256_add_ps(velecsum,velec);
 591
 592             fscal            = felec;
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm256_mul_ps(fscal,dx11);
 596             ty               = _mm256_mul_ps(fscal,dy11);
 597             tz               = _mm256_mul_ps(fscal,dz11);
 598
 599             /* Update vectorial force */
 600             fix1             = _mm256_add_ps(fix1,tx);
 601             fiy1             = _mm256_add_ps(fiy1,ty);
 602             fiz1             = _mm256_add_ps(fiz1,tz);
 603
 604             fjx1             = _mm256_add_ps(fjx1,tx);
 605             fjy1             = _mm256_add_ps(fjy1,ty);
 606             fjz1             = _mm256_add_ps(fjz1,tz);
 607
 608             /**************************
 609              * CALCULATE INTERACTIONS *
 610              **************************/
 611
 612             r12              = _mm256_mul_ps(rsq12,rinv12);
 613
 614             /* Calculate table index by multiplying r with table scale and truncate to integer */
 615             rt               = _mm256_mul_ps(r12,vftabscale);
 616             vfitab           = _mm256_cvttps_epi32(rt);
 617             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 618             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 619             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 620             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 621             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 622             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 623
 624             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 625             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 626                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 627             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 628                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 629             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 630                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 631             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 632                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 633             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 634             Heps             = _mm256_mul_ps(vfeps,H);
 635             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 636             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 637             velec            = _mm256_mul_ps(qq12,VV);
 638             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 639             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
 640
 641             /* Update potential sum for this i atom from the interaction with this j atom. */
 642             velecsum         = _mm256_add_ps(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             /* Calculate temporary vectorial force */
 647             tx               = _mm256_mul_ps(fscal,dx12);
 648             ty               = _mm256_mul_ps(fscal,dy12);
 649             tz               = _mm256_mul_ps(fscal,dz12);
 650
 651             /* Update vectorial force */
 652             fix1             = _mm256_add_ps(fix1,tx);
 653             fiy1             = _mm256_add_ps(fiy1,ty);
 654             fiz1             = _mm256_add_ps(fiz1,tz);
 655
 656             fjx2             = _mm256_add_ps(fjx2,tx);
 657             fjy2             = _mm256_add_ps(fjy2,ty);
 658             fjz2             = _mm256_add_ps(fjz2,tz);
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             r20              = _mm256_mul_ps(rsq20,rinv20);
 665
 666             /* Calculate table index by multiplying r with table scale and truncate to integer */
 667             rt               = _mm256_mul_ps(r20,vftabscale);
 668             vfitab           = _mm256_cvttps_epi32(rt);
 669             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 670             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 671             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 672             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 673             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 674             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 675
 676             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 677             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 678                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 679             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 680                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 681             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 682                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 683             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 684                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 685             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 686             Heps             = _mm256_mul_ps(vfeps,H);
 687             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 688             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 689             velec            = _mm256_mul_ps(qq20,VV);
 690             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 691             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 692
 693             /* Update potential sum for this i atom from the interaction with this j atom. */
 694             velecsum         = _mm256_add_ps(velecsum,velec);
 695
 696             fscal            = felec;
 697
 698             /* Calculate temporary vectorial force */
 699             tx               = _mm256_mul_ps(fscal,dx20);
 700             ty               = _mm256_mul_ps(fscal,dy20);
 701             tz               = _mm256_mul_ps(fscal,dz20);
 702
 703             /* Update vectorial force */
 704             fix2             = _mm256_add_ps(fix2,tx);
 705             fiy2             = _mm256_add_ps(fiy2,ty);
 706             fiz2             = _mm256_add_ps(fiz2,tz);
 707
 708             fjx0             = _mm256_add_ps(fjx0,tx);
 709             fjy0             = _mm256_add_ps(fjy0,ty);
 710             fjz0             = _mm256_add_ps(fjz0,tz);
 711
 712             /**************************
 713              * CALCULATE INTERACTIONS *
 714              **************************/
 715
 716             r21              = _mm256_mul_ps(rsq21,rinv21);
 717
 718             /* Calculate table index by multiplying r with table scale and truncate to integer */
 719             rt               = _mm256_mul_ps(r21,vftabscale);
 720             vfitab           = _mm256_cvttps_epi32(rt);
 721             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 722             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 723             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 724             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 725             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 726             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 727
 728             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 729             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 730                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 731             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 732                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 733             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 734                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 735             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 736                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 737             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 738             Heps             = _mm256_mul_ps(vfeps,H);
 739             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 740             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 741             velec            = _mm256_mul_ps(qq21,VV);
 742             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 743             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
 744
 745             /* Update potential sum for this i atom from the interaction with this j atom. */
 746             velecsum         = _mm256_add_ps(velecsum,velec);
 747
 748             fscal            = felec;
 749
 750             /* Calculate temporary vectorial force */
 751             tx               = _mm256_mul_ps(fscal,dx21);
 752             ty               = _mm256_mul_ps(fscal,dy21);
 753             tz               = _mm256_mul_ps(fscal,dz21);
 754
 755             /* Update vectorial force */
 756             fix2             = _mm256_add_ps(fix2,tx);
 757             fiy2             = _mm256_add_ps(fiy2,ty);
 758             fiz2             = _mm256_add_ps(fiz2,tz);
 759
 760             fjx1             = _mm256_add_ps(fjx1,tx);
 761             fjy1             = _mm256_add_ps(fjy1,ty);
 762             fjz1             = _mm256_add_ps(fjz1,tz);
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             r22              = _mm256_mul_ps(rsq22,rinv22);
 769
 770             /* Calculate table index by multiplying r with table scale and truncate to integer */
 771             rt               = _mm256_mul_ps(r22,vftabscale);
 772             vfitab           = _mm256_cvttps_epi32(rt);
 773             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 774             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 775             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 776             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 777             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 778             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 779
 780             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 781             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 782                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 783             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 784                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 785             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 786                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 787             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 788                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 789             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 790             Heps             = _mm256_mul_ps(vfeps,H);
 791             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 792             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 793             velec            = _mm256_mul_ps(qq22,VV);
 794             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 795             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
 796
 797             /* Update potential sum for this i atom from the interaction with this j atom. */
 798             velecsum         = _mm256_add_ps(velecsum,velec);
 799
 800             fscal            = felec;
 801
 802             /* Calculate temporary vectorial force */
 803             tx               = _mm256_mul_ps(fscal,dx22);
 804             ty               = _mm256_mul_ps(fscal,dy22);
 805             tz               = _mm256_mul_ps(fscal,dz22);
 806
 807             /* Update vectorial force */
 808             fix2             = _mm256_add_ps(fix2,tx);
 809             fiy2             = _mm256_add_ps(fiy2,ty);
 810             fiz2             = _mm256_add_ps(fiz2,tz);
 811
 812             fjx2             = _mm256_add_ps(fjx2,tx);
 813             fjy2             = _mm256_add_ps(fjy2,ty);
 814             fjz2             = _mm256_add_ps(fjz2,tz);
 815
 816             fjptrA             = f+j_coord_offsetA;
 817             fjptrB             = f+j_coord_offsetB;
 818             fjptrC             = f+j_coord_offsetC;
 819             fjptrD             = f+j_coord_offsetD;
 820             fjptrE             = f+j_coord_offsetE;
 821             fjptrF             = f+j_coord_offsetF;
 822             fjptrG             = f+j_coord_offsetG;
 823             fjptrH             = f+j_coord_offsetH;
 824
 825             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 826                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 827
 828             /* Inner loop uses 417 flops */
 829         }
 830
 831         if(jidx<j_index_end)
 832         {
 833
 834             /* Get j neighbor index, and coordinate index */
 835             jnrlistA         = jjnr[jidx];
 836             jnrlistB         = jjnr[jidx+1];
 837             jnrlistC         = jjnr[jidx+2];
 838             jnrlistD         = jjnr[jidx+3];
 839             jnrlistE         = jjnr[jidx+4];
 840             jnrlistF         = jjnr[jidx+5];
 841             jnrlistG         = jjnr[jidx+6];
 842             jnrlistH         = jjnr[jidx+7];
 843             /* Sign of each element will be negative for non-real atoms.
 844              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 845              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 846              */
 847             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 848                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 849
 850             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 851             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 852             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 853             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 854             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 855             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 856             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 857             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 858             j_coord_offsetA  = DIM*jnrA;
 859             j_coord_offsetB  = DIM*jnrB;
 860             j_coord_offsetC  = DIM*jnrC;
 861             j_coord_offsetD  = DIM*jnrD;
 862             j_coord_offsetE  = DIM*jnrE;
 863             j_coord_offsetF  = DIM*jnrF;
 864             j_coord_offsetG  = DIM*jnrG;
 865             j_coord_offsetH  = DIM*jnrH;
 866
 867             /* load j atom coordinates */
 868             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 869                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 870                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 871                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 872                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 873
 874             /* Calculate displacement vector */
 875             dx00             = _mm256_sub_ps(ix0,jx0);
 876             dy00             = _mm256_sub_ps(iy0,jy0);
 877             dz00             = _mm256_sub_ps(iz0,jz0);
 878             dx01             = _mm256_sub_ps(ix0,jx1);
 879             dy01             = _mm256_sub_ps(iy0,jy1);
 880             dz01             = _mm256_sub_ps(iz0,jz1);
 881             dx02             = _mm256_sub_ps(ix0,jx2);
 882             dy02             = _mm256_sub_ps(iy0,jy2);
 883             dz02             = _mm256_sub_ps(iz0,jz2);
 884             dx10             = _mm256_sub_ps(ix1,jx0);
 885             dy10             = _mm256_sub_ps(iy1,jy0);
 886             dz10             = _mm256_sub_ps(iz1,jz0);
 887             dx11             = _mm256_sub_ps(ix1,jx1);
 888             dy11             = _mm256_sub_ps(iy1,jy1);
 889             dz11             = _mm256_sub_ps(iz1,jz1);
 890             dx12             = _mm256_sub_ps(ix1,jx2);
 891             dy12             = _mm256_sub_ps(iy1,jy2);
 892             dz12             = _mm256_sub_ps(iz1,jz2);
 893             dx20             = _mm256_sub_ps(ix2,jx0);
 894             dy20             = _mm256_sub_ps(iy2,jy0);
 895             dz20             = _mm256_sub_ps(iz2,jz0);
 896             dx21             = _mm256_sub_ps(ix2,jx1);
 897             dy21             = _mm256_sub_ps(iy2,jy1);
 898             dz21             = _mm256_sub_ps(iz2,jz1);
 899             dx22             = _mm256_sub_ps(ix2,jx2);
 900             dy22             = _mm256_sub_ps(iy2,jy2);
 901             dz22             = _mm256_sub_ps(iz2,jz2);
 902
 903             /* Calculate squared distance and things based on it */
 904             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 905             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 906             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 907             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 908             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 909             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 910             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 911             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 912             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 913
 914             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 915             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 916             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 917             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 918             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 919             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 920             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 921             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 922             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 923
 924             fjx0             = _mm256_setzero_ps();
 925             fjy0             = _mm256_setzero_ps();
 926             fjz0             = _mm256_setzero_ps();
 927             fjx1             = _mm256_setzero_ps();
 928             fjy1             = _mm256_setzero_ps();
 929             fjz1             = _mm256_setzero_ps();
 930             fjx2             = _mm256_setzero_ps();
 931             fjy2             = _mm256_setzero_ps();
 932             fjz2             = _mm256_setzero_ps();
 933
 934             /**************************
 935              * CALCULATE INTERACTIONS *
 936              **************************/
 937
 938             r00              = _mm256_mul_ps(rsq00,rinv00);
 939             r00              = _mm256_andnot_ps(dummy_mask,r00);
 940
 941             /* Calculate table index by multiplying r with table scale and truncate to integer */
 942             rt               = _mm256_mul_ps(r00,vftabscale);
 943             vfitab           = _mm256_cvttps_epi32(rt);
 944             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 945             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 946             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 947             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 948             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 949             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 950
 951             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 952             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 953                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 954             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 955                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 956             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 957                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 958             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 959                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 960             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 961             Heps             = _mm256_mul_ps(vfeps,H);
 962             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 963             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 964             velec            = _mm256_mul_ps(qq00,VV);
 965             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 966             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 967
 968             /* CUBIC SPLINE TABLE DISPERSION */
 969             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 970             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 971             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 972                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 973             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 974                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 975             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 976                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 977             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 978                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 979             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 980             Heps             = _mm256_mul_ps(vfeps,H);
 981             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 982             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 983             vvdw6            = _mm256_mul_ps(c6_00,VV);
 984             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 985             fvdw6            = _mm256_mul_ps(c6_00,FF);
 986
 987             /* CUBIC SPLINE TABLE REPULSION */
 988             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 989             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 990             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 991                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 992             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 993                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 994             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 995                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 996             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 997                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 998             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 999             Heps             = _mm256_mul_ps(vfeps,H);
1000             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1001             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1002             vvdw12           = _mm256_mul_ps(c12_00,VV);
1003             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1004             fvdw12           = _mm256_mul_ps(c12_00,FF);
1005             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
1006             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1007
1008             /* Update potential sum for this i atom from the interaction with this j atom. */
1009             velec            = _mm256_andnot_ps(dummy_mask,velec);
1010             velecsum         = _mm256_add_ps(velecsum,velec);
1011             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
1012             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
1013
1014             fscal            = _mm256_add_ps(felec,fvdw);
1015
1016             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1017
1018             /* Calculate temporary vectorial force */
1019             tx               = _mm256_mul_ps(fscal,dx00);
1020             ty               = _mm256_mul_ps(fscal,dy00);
1021             tz               = _mm256_mul_ps(fscal,dz00);
1022
1023             /* Update vectorial force */
1024             fix0             = _mm256_add_ps(fix0,tx);
1025             fiy0             = _mm256_add_ps(fiy0,ty);
1026             fiz0             = _mm256_add_ps(fiz0,tz);
1027
1028             fjx0             = _mm256_add_ps(fjx0,tx);
1029             fjy0             = _mm256_add_ps(fjy0,ty);
1030             fjz0             = _mm256_add_ps(fjz0,tz);
1031
1032             /**************************
1033              * CALCULATE INTERACTIONS *
1034              **************************/
1035
1036             r01              = _mm256_mul_ps(rsq01,rinv01);
1037             r01              = _mm256_andnot_ps(dummy_mask,r01);
1038
1039             /* Calculate table index by multiplying r with table scale and truncate to integer */
1040             rt               = _mm256_mul_ps(r01,vftabscale);
1041             vfitab           = _mm256_cvttps_epi32(rt);
1042             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1043             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1044             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1045             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1046             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1047             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1048
1049             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1050             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1051                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1052             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1053                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1054             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1055                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1056             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1057                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1058             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1059             Heps             = _mm256_mul_ps(vfeps,H);
1060             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1061             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1062             velec            = _mm256_mul_ps(qq01,VV);
1063             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1064             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
1065
1066             /* Update potential sum for this i atom from the interaction with this j atom. */
1067             velec            = _mm256_andnot_ps(dummy_mask,velec);
1068             velecsum         = _mm256_add_ps(velecsum,velec);
1069
1070             fscal            = felec;
1071
1072             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1073
1074             /* Calculate temporary vectorial force */
1075             tx               = _mm256_mul_ps(fscal,dx01);
1076             ty               = _mm256_mul_ps(fscal,dy01);
1077             tz               = _mm256_mul_ps(fscal,dz01);
1078
1079             /* Update vectorial force */
1080             fix0             = _mm256_add_ps(fix0,tx);
1081             fiy0             = _mm256_add_ps(fiy0,ty);
1082             fiz0             = _mm256_add_ps(fiz0,tz);
1083
1084             fjx1             = _mm256_add_ps(fjx1,tx);
1085             fjy1             = _mm256_add_ps(fjy1,ty);
1086             fjz1             = _mm256_add_ps(fjz1,tz);
1087
1088             /**************************
1089              * CALCULATE INTERACTIONS *
1090              **************************/
1091
1092             r02              = _mm256_mul_ps(rsq02,rinv02);
1093             r02              = _mm256_andnot_ps(dummy_mask,r02);
1094
1095             /* Calculate table index by multiplying r with table scale and truncate to integer */
1096             rt               = _mm256_mul_ps(r02,vftabscale);
1097             vfitab           = _mm256_cvttps_epi32(rt);
1098             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1099             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1100             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1101             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1102             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1103             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1104
1105             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1106             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1107                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1108             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1109                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1110             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1111                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1112             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1113                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1114             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1115             Heps             = _mm256_mul_ps(vfeps,H);
1116             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1117             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1118             velec            = _mm256_mul_ps(qq02,VV);
1119             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1120             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
1121
1122             /* Update potential sum for this i atom from the interaction with this j atom. */
1123             velec            = _mm256_andnot_ps(dummy_mask,velec);
1124             velecsum         = _mm256_add_ps(velecsum,velec);
1125
1126             fscal            = felec;
1127
1128             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1129
1130             /* Calculate temporary vectorial force */
1131             tx               = _mm256_mul_ps(fscal,dx02);
1132             ty               = _mm256_mul_ps(fscal,dy02);
1133             tz               = _mm256_mul_ps(fscal,dz02);
1134
1135             /* Update vectorial force */
1136             fix0             = _mm256_add_ps(fix0,tx);
1137             fiy0             = _mm256_add_ps(fiy0,ty);
1138             fiz0             = _mm256_add_ps(fiz0,tz);
1139
1140             fjx2             = _mm256_add_ps(fjx2,tx);
1141             fjy2             = _mm256_add_ps(fjy2,ty);
1142             fjz2             = _mm256_add_ps(fjz2,tz);
1143
1144             /**************************
1145              * CALCULATE INTERACTIONS *
1146              **************************/
1147
1148             r10              = _mm256_mul_ps(rsq10,rinv10);
1149             r10              = _mm256_andnot_ps(dummy_mask,r10);
1150
1151             /* Calculate table index by multiplying r with table scale and truncate to integer */
1152             rt               = _mm256_mul_ps(r10,vftabscale);
1153             vfitab           = _mm256_cvttps_epi32(rt);
1154             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1155             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1156             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1157             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1158             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1159             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1160
1161             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1162             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1163                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1164             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1165                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1166             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1167                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1168             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1169                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1170             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1171             Heps             = _mm256_mul_ps(vfeps,H);
1172             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1173             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1174             velec            = _mm256_mul_ps(qq10,VV);
1175             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1176             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1177
1178             /* Update potential sum for this i atom from the interaction with this j atom. */
1179             velec            = _mm256_andnot_ps(dummy_mask,velec);
1180             velecsum         = _mm256_add_ps(velecsum,velec);
1181
1182             fscal            = felec;
1183
1184             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1185
1186             /* Calculate temporary vectorial force */
1187             tx               = _mm256_mul_ps(fscal,dx10);
1188             ty               = _mm256_mul_ps(fscal,dy10);
1189             tz               = _mm256_mul_ps(fscal,dz10);
1190
1191             /* Update vectorial force */
1192             fix1             = _mm256_add_ps(fix1,tx);
1193             fiy1             = _mm256_add_ps(fiy1,ty);
1194             fiz1             = _mm256_add_ps(fiz1,tz);
1195
1196             fjx0             = _mm256_add_ps(fjx0,tx);
1197             fjy0             = _mm256_add_ps(fjy0,ty);
1198             fjz0             = _mm256_add_ps(fjz0,tz);
1199
1200             /**************************
1201              * CALCULATE INTERACTIONS *
1202              **************************/
1203
1204             r11              = _mm256_mul_ps(rsq11,rinv11);
1205             r11              = _mm256_andnot_ps(dummy_mask,r11);
1206
1207             /* Calculate table index by multiplying r with table scale and truncate to integer */
1208             rt               = _mm256_mul_ps(r11,vftabscale);
1209             vfitab           = _mm256_cvttps_epi32(rt);
1210             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1211             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1212             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1213             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1214             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1215             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1216
1217             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1218             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1219                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1220             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1221                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1222             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1223                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1224             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1225                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1226             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1227             Heps             = _mm256_mul_ps(vfeps,H);
1228             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1229             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1230             velec            = _mm256_mul_ps(qq11,VV);
1231             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1232             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
1233
1234             /* Update potential sum for this i atom from the interaction with this j atom. */
1235             velec            = _mm256_andnot_ps(dummy_mask,velec);
1236             velecsum         = _mm256_add_ps(velecsum,velec);
1237
1238             fscal            = felec;
1239
1240             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1241
1242             /* Calculate temporary vectorial force */
1243             tx               = _mm256_mul_ps(fscal,dx11);
1244             ty               = _mm256_mul_ps(fscal,dy11);
1245             tz               = _mm256_mul_ps(fscal,dz11);
1246
1247             /* Update vectorial force */
1248             fix1             = _mm256_add_ps(fix1,tx);
1249             fiy1             = _mm256_add_ps(fiy1,ty);
1250             fiz1             = _mm256_add_ps(fiz1,tz);
1251
1252             fjx1             = _mm256_add_ps(fjx1,tx);
1253             fjy1             = _mm256_add_ps(fjy1,ty);
1254             fjz1             = _mm256_add_ps(fjz1,tz);
1255
1256             /**************************
1257              * CALCULATE INTERACTIONS *
1258              **************************/
1259
1260             r12              = _mm256_mul_ps(rsq12,rinv12);
1261             r12              = _mm256_andnot_ps(dummy_mask,r12);
1262
1263             /* Calculate table index by multiplying r with table scale and truncate to integer */
1264             rt               = _mm256_mul_ps(r12,vftabscale);
1265             vfitab           = _mm256_cvttps_epi32(rt);
1266             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1267             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1268             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1269             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1270             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1271             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1272
1273             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1274             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1275                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1276             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1277                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1278             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1279                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1280             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1281                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1282             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1283             Heps             = _mm256_mul_ps(vfeps,H);
1284             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1285             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1286             velec            = _mm256_mul_ps(qq12,VV);
1287             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1288             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
1289
1290             /* Update potential sum for this i atom from the interaction with this j atom. */
1291             velec            = _mm256_andnot_ps(dummy_mask,velec);
1292             velecsum         = _mm256_add_ps(velecsum,velec);
1293
1294             fscal            = felec;
1295
1296             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1297
1298             /* Calculate temporary vectorial force */
1299             tx               = _mm256_mul_ps(fscal,dx12);
1300             ty               = _mm256_mul_ps(fscal,dy12);
1301             tz               = _mm256_mul_ps(fscal,dz12);
1302
1303             /* Update vectorial force */
1304             fix1             = _mm256_add_ps(fix1,tx);
1305             fiy1             = _mm256_add_ps(fiy1,ty);
1306             fiz1             = _mm256_add_ps(fiz1,tz);
1307
1308             fjx2             = _mm256_add_ps(fjx2,tx);
1309             fjy2             = _mm256_add_ps(fjy2,ty);
1310             fjz2             = _mm256_add_ps(fjz2,tz);
1311
1312             /**************************
1313              * CALCULATE INTERACTIONS *
1314              **************************/
1315
1316             r20              = _mm256_mul_ps(rsq20,rinv20);
1317             r20              = _mm256_andnot_ps(dummy_mask,r20);
1318
1319             /* Calculate table index by multiplying r with table scale and truncate to integer */
1320             rt               = _mm256_mul_ps(r20,vftabscale);
1321             vfitab           = _mm256_cvttps_epi32(rt);
1322             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1323             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1324             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1325             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1326             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1327             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1328
1329             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1330             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1331                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1332             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1333                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1334             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1335                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1336             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1337                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1338             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1339             Heps             = _mm256_mul_ps(vfeps,H);
1340             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1341             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1342             velec            = _mm256_mul_ps(qq20,VV);
1343             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1344             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1345
1346             /* Update potential sum for this i atom from the interaction with this j atom. */
1347             velec            = _mm256_andnot_ps(dummy_mask,velec);
1348             velecsum         = _mm256_add_ps(velecsum,velec);
1349
1350             fscal            = felec;
1351
1352             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1353
1354             /* Calculate temporary vectorial force */
1355             tx               = _mm256_mul_ps(fscal,dx20);
1356             ty               = _mm256_mul_ps(fscal,dy20);
1357             tz               = _mm256_mul_ps(fscal,dz20);
1358
1359             /* Update vectorial force */
1360             fix2             = _mm256_add_ps(fix2,tx);
1361             fiy2             = _mm256_add_ps(fiy2,ty);
1362             fiz2             = _mm256_add_ps(fiz2,tz);
1363
1364             fjx0             = _mm256_add_ps(fjx0,tx);
1365             fjy0             = _mm256_add_ps(fjy0,ty);
1366             fjz0             = _mm256_add_ps(fjz0,tz);
1367
1368             /**************************
1369              * CALCULATE INTERACTIONS *
1370              **************************/
1371
1372             r21              = _mm256_mul_ps(rsq21,rinv21);
1373             r21              = _mm256_andnot_ps(dummy_mask,r21);
1374
1375             /* Calculate table index by multiplying r with table scale and truncate to integer */
1376             rt               = _mm256_mul_ps(r21,vftabscale);
1377             vfitab           = _mm256_cvttps_epi32(rt);
1378             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1379             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1380             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1381             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1382             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1383             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1384
1385             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1386             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1388             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1389                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1390             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1391                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1392             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1393                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1394             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1395             Heps             = _mm256_mul_ps(vfeps,H);
1396             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1397             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1398             velec            = _mm256_mul_ps(qq21,VV);
1399             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1400             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
1401
1402             /* Update potential sum for this i atom from the interaction with this j atom. */
1403             velec            = _mm256_andnot_ps(dummy_mask,velec);
1404             velecsum         = _mm256_add_ps(velecsum,velec);
1405
1406             fscal            = felec;
1407
1408             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1409
1410             /* Calculate temporary vectorial force */
1411             tx               = _mm256_mul_ps(fscal,dx21);
1412             ty               = _mm256_mul_ps(fscal,dy21);
1413             tz               = _mm256_mul_ps(fscal,dz21);
1414
1415             /* Update vectorial force */
1416             fix2             = _mm256_add_ps(fix2,tx);
1417             fiy2             = _mm256_add_ps(fiy2,ty);
1418             fiz2             = _mm256_add_ps(fiz2,tz);
1419
1420             fjx1             = _mm256_add_ps(fjx1,tx);
1421             fjy1             = _mm256_add_ps(fjy1,ty);
1422             fjz1             = _mm256_add_ps(fjz1,tz);
1423
1424             /**************************
1425              * CALCULATE INTERACTIONS *
1426              **************************/
1427
1428             r22              = _mm256_mul_ps(rsq22,rinv22);
1429             r22              = _mm256_andnot_ps(dummy_mask,r22);
1430
1431             /* Calculate table index by multiplying r with table scale and truncate to integer */
1432             rt               = _mm256_mul_ps(r22,vftabscale);
1433             vfitab           = _mm256_cvttps_epi32(rt);
1434             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1435             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1436             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1437             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1438             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1439             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1440
1441             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1442             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1443                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1444             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1445                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1446             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1447                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1448             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1449                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1450             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1451             Heps             = _mm256_mul_ps(vfeps,H);
1452             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1453             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1454             velec            = _mm256_mul_ps(qq22,VV);
1455             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1456             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
1457
1458             /* Update potential sum for this i atom from the interaction with this j atom. */
1459             velec            = _mm256_andnot_ps(dummy_mask,velec);
1460             velecsum         = _mm256_add_ps(velecsum,velec);
1461
1462             fscal            = felec;
1463
1464             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1465
1466             /* Calculate temporary vectorial force */
1467             tx               = _mm256_mul_ps(fscal,dx22);
1468             ty               = _mm256_mul_ps(fscal,dy22);
1469             tz               = _mm256_mul_ps(fscal,dz22);
1470
1471             /* Update vectorial force */
1472             fix2             = _mm256_add_ps(fix2,tx);
1473             fiy2             = _mm256_add_ps(fiy2,ty);
1474             fiz2             = _mm256_add_ps(fiz2,tz);
1475
1476             fjx2             = _mm256_add_ps(fjx2,tx);
1477             fjy2             = _mm256_add_ps(fjy2,ty);
1478             fjz2             = _mm256_add_ps(fjz2,tz);
1479
1480             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1481             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1482             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1483             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1484             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1485             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1486             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1487             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1488
1489             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1490                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1491
1492             /* Inner loop uses 426 flops */
1493         }
1494
1495         /* End of innermost loop */
1496
1497         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1498                                                  f+i_coord_offset,fshift+i_shift_offset);
1499
1500         ggid                        = gid[iidx];
1501         /* Update potential energies */
1502         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1503         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1504
1505         /* Increment number of inner iterations */
1506         inneriter                  += j_index_end - j_index_start;
1507
1508         /* Outer loop uses 20 flops */
1509     }
1510
1511     /* Increment number of outer iterations */
1512     outeriter        += nri;
1513
1514     /* Update outer/inner flops */
1515
1516     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*426);
1517 }
1518 /*
1519  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_F_avx_256_single
1520  * Electrostatics interaction: CubicSplineTable
1521  * VdW interaction:            CubicSplineTable
1522  * Geometry:                   Water3-Water3
1523  * Calculate force/pot:        Force
1524  */
1525 void
1526 nb_kernel_ElecCSTab_VdwCSTab_GeomW3W3_F_avx_256_single
1527                     (t_nblist                    * gmx_restrict       nlist,
1528                      rvec                        * gmx_restrict          xx,
1529                      rvec                        * gmx_restrict          ff,
1530                      t_forcerec                  * gmx_restrict          fr,
1531                      t_mdatoms                   * gmx_restrict     mdatoms,
1532                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1533                      t_nrnb                      * gmx_restrict        nrnb)
1534 {
1535     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1536      * just 0 for non-waters.
1537      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1538      * jnr indices corresponding to data put in the four positions in the SIMD register.
1539      */
1540     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1541     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1542     int              jnrA,jnrB,jnrC,jnrD;
1543     int              jnrE,jnrF,jnrG,jnrH;
1544     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1545     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1546     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1547     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1548     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1549     real             rcutoff_scalar;
1550     real             *shiftvec,*fshift,*x,*f;
1551     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1552     real             scratch[4*DIM];
1553     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1554     real *           vdwioffsetptr0;
1555     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1556     real *           vdwioffsetptr1;
1557     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1558     real *           vdwioffsetptr2;
1559     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1560     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1561     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1562     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1563     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1564     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1565     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1566     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1567     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1568     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1569     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1570     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1571     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1572     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1573     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1574     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1575     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1576     real             *charge;
1577     int              nvdwtype;
1578     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1579     int              *vdwtype;
1580     real             *vdwparam;
1581     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1582     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1583     __m256i          vfitab;
1584     __m128i          vfitab_lo,vfitab_hi;
1585     __m128i          ifour       = _mm_set1_epi32(4);
1586     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1587     real             *vftab;
1588     __m256           dummy_mask,cutoff_mask;
1589     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1590     __m256           one     = _mm256_set1_ps(1.0);
1591     __m256           two     = _mm256_set1_ps(2.0);
1592     x                = xx[0];
1593     f                = ff[0];
1594
1595     nri              = nlist->nri;
1596     iinr             = nlist->iinr;
1597     jindex           = nlist->jindex;
1598     jjnr             = nlist->jjnr;
1599     shiftidx         = nlist->shift;
1600     gid              = nlist->gid;
1601     shiftvec         = fr->shift_vec[0];
1602     fshift           = fr->fshift[0];
1603     facel            = _mm256_set1_ps(fr->epsfac);
1604     charge           = mdatoms->chargeA;
1605     nvdwtype         = fr->ntype;
1606     vdwparam         = fr->nbfp;
1607     vdwtype          = mdatoms->typeA;
1608
1609     vftab            = kernel_data->table_elec_vdw->data;
1610     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
1611
1612     /* Setup water-specific parameters */
1613     inr              = nlist->iinr[0];
1614     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1615     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1616     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1617     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1618
1619     jq0              = _mm256_set1_ps(charge[inr+0]);
1620     jq1              = _mm256_set1_ps(charge[inr+1]);
1621     jq2              = _mm256_set1_ps(charge[inr+2]);
1622     vdwjidx0A        = 2*vdwtype[inr+0];
1623     qq00             = _mm256_mul_ps(iq0,jq0);
1624     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1625     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1626     qq01             = _mm256_mul_ps(iq0,jq1);
1627     qq02             = _mm256_mul_ps(iq0,jq2);
1628     qq10             = _mm256_mul_ps(iq1,jq0);
1629     qq11             = _mm256_mul_ps(iq1,jq1);
1630     qq12             = _mm256_mul_ps(iq1,jq2);
1631     qq20             = _mm256_mul_ps(iq2,jq0);
1632     qq21             = _mm256_mul_ps(iq2,jq1);
1633     qq22             = _mm256_mul_ps(iq2,jq2);
1634
1635     /* Avoid stupid compiler warnings */
1636     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1637     j_coord_offsetA = 0;
1638     j_coord_offsetB = 0;
1639     j_coord_offsetC = 0;
1640     j_coord_offsetD = 0;
1641     j_coord_offsetE = 0;
1642     j_coord_offsetF = 0;
1643     j_coord_offsetG = 0;
1644     j_coord_offsetH = 0;
1645
1646     outeriter        = 0;
1647     inneriter        = 0;
1648
1649     for(iidx=0;iidx<4*DIM;iidx++)
1650     {
1651         scratch[iidx] = 0.0;
1652     }
1653
1654     /* Start outer loop over neighborlists */
1655     for(iidx=0; iidx<nri; iidx++)
1656     {
1657         /* Load shift vector for this list */
1658         i_shift_offset   = DIM*shiftidx[iidx];
1659
1660         /* Load limits for loop over neighbors */
1661         j_index_start    = jindex[iidx];
1662         j_index_end      = jindex[iidx+1];
1663
1664         /* Get outer coordinate index */
1665         inr              = iinr[iidx];
1666         i_coord_offset   = DIM*inr;
1667
1668         /* Load i particle coords and add shift vector */
1669         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1670                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1671
1672         fix0             = _mm256_setzero_ps();
1673         fiy0             = _mm256_setzero_ps();
1674         fiz0             = _mm256_setzero_ps();
1675         fix1             = _mm256_setzero_ps();
1676         fiy1             = _mm256_setzero_ps();
1677         fiz1             = _mm256_setzero_ps();
1678         fix2             = _mm256_setzero_ps();
1679         fiy2             = _mm256_setzero_ps();
1680         fiz2             = _mm256_setzero_ps();
1681
1682         /* Start inner kernel loop */
1683         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1684         {
1685
1686             /* Get j neighbor index, and coordinate index */
1687             jnrA             = jjnr[jidx];
1688             jnrB             = jjnr[jidx+1];
1689             jnrC             = jjnr[jidx+2];
1690             jnrD             = jjnr[jidx+3];
1691             jnrE             = jjnr[jidx+4];
1692             jnrF             = jjnr[jidx+5];
1693             jnrG             = jjnr[jidx+6];
1694             jnrH             = jjnr[jidx+7];
1695             j_coord_offsetA  = DIM*jnrA;
1696             j_coord_offsetB  = DIM*jnrB;
1697             j_coord_offsetC  = DIM*jnrC;
1698             j_coord_offsetD  = DIM*jnrD;
1699             j_coord_offsetE  = DIM*jnrE;
1700             j_coord_offsetF  = DIM*jnrF;
1701             j_coord_offsetG  = DIM*jnrG;
1702             j_coord_offsetH  = DIM*jnrH;
1703
1704             /* load j atom coordinates */
1705             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1706                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1707                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1708                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1709                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1710
1711             /* Calculate displacement vector */
1712             dx00             = _mm256_sub_ps(ix0,jx0);
1713             dy00             = _mm256_sub_ps(iy0,jy0);
1714             dz00             = _mm256_sub_ps(iz0,jz0);
1715             dx01             = _mm256_sub_ps(ix0,jx1);
1716             dy01             = _mm256_sub_ps(iy0,jy1);
1717             dz01             = _mm256_sub_ps(iz0,jz1);
1718             dx02             = _mm256_sub_ps(ix0,jx2);
1719             dy02             = _mm256_sub_ps(iy0,jy2);
1720             dz02             = _mm256_sub_ps(iz0,jz2);
1721             dx10             = _mm256_sub_ps(ix1,jx0);
1722             dy10             = _mm256_sub_ps(iy1,jy0);
1723             dz10             = _mm256_sub_ps(iz1,jz0);
1724             dx11             = _mm256_sub_ps(ix1,jx1);
1725             dy11             = _mm256_sub_ps(iy1,jy1);
1726             dz11             = _mm256_sub_ps(iz1,jz1);
1727             dx12             = _mm256_sub_ps(ix1,jx2);
1728             dy12             = _mm256_sub_ps(iy1,jy2);
1729             dz12             = _mm256_sub_ps(iz1,jz2);
1730             dx20             = _mm256_sub_ps(ix2,jx0);
1731             dy20             = _mm256_sub_ps(iy2,jy0);
1732             dz20             = _mm256_sub_ps(iz2,jz0);
1733             dx21             = _mm256_sub_ps(ix2,jx1);
1734             dy21             = _mm256_sub_ps(iy2,jy1);
1735             dz21             = _mm256_sub_ps(iz2,jz1);
1736             dx22             = _mm256_sub_ps(ix2,jx2);
1737             dy22             = _mm256_sub_ps(iy2,jy2);
1738             dz22             = _mm256_sub_ps(iz2,jz2);
1739
1740             /* Calculate squared distance and things based on it */
1741             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1742             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1743             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1744             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1745             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1746             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1747             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1748             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1749             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1750
1751             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1752             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1753             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1754             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1755             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1756             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1757             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1758             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1759             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1760
1761             fjx0             = _mm256_setzero_ps();
1762             fjy0             = _mm256_setzero_ps();
1763             fjz0             = _mm256_setzero_ps();
1764             fjx1             = _mm256_setzero_ps();
1765             fjy1             = _mm256_setzero_ps();
1766             fjz1             = _mm256_setzero_ps();
1767             fjx2             = _mm256_setzero_ps();
1768             fjy2             = _mm256_setzero_ps();
1769             fjz2             = _mm256_setzero_ps();
1770
1771             /**************************
1772              * CALCULATE INTERACTIONS *
1773              **************************/
1774
1775             r00              = _mm256_mul_ps(rsq00,rinv00);
1776
1777             /* Calculate table index by multiplying r with table scale and truncate to integer */
1778             rt               = _mm256_mul_ps(r00,vftabscale);
1779             vfitab           = _mm256_cvttps_epi32(rt);
1780             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1781             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1782             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1783             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1784             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1785             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1786
1787             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1788             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1789                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1790             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1791                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1792             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1793                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1794             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1795                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1796             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1797             Heps             = _mm256_mul_ps(vfeps,H);
1798             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1799             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1800             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1801
1802             /* CUBIC SPLINE TABLE DISPERSION */
1803             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1804             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1805             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1806                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1807             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1808                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1809             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1810                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1811             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1812                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1813             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1814             Heps             = _mm256_mul_ps(vfeps,H);
1815             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1816             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1817             fvdw6            = _mm256_mul_ps(c6_00,FF);
1818
1819             /* CUBIC SPLINE TABLE REPULSION */
1820             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1821             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1822             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1823                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1824             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1825                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1826             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1827                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1828             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1829                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1830             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1831             Heps             = _mm256_mul_ps(vfeps,H);
1832             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1833             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1834             fvdw12           = _mm256_mul_ps(c12_00,FF);
1835             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1836
1837             fscal            = _mm256_add_ps(felec,fvdw);
1838
1839             /* Calculate temporary vectorial force */
1840             tx               = _mm256_mul_ps(fscal,dx00);
1841             ty               = _mm256_mul_ps(fscal,dy00);
1842             tz               = _mm256_mul_ps(fscal,dz00);
1843
1844             /* Update vectorial force */
1845             fix0             = _mm256_add_ps(fix0,tx);
1846             fiy0             = _mm256_add_ps(fiy0,ty);
1847             fiz0             = _mm256_add_ps(fiz0,tz);
1848
1849             fjx0             = _mm256_add_ps(fjx0,tx);
1850             fjy0             = _mm256_add_ps(fjy0,ty);
1851             fjz0             = _mm256_add_ps(fjz0,tz);
1852
1853             /**************************
1854              * CALCULATE INTERACTIONS *
1855              **************************/
1856
1857             r01              = _mm256_mul_ps(rsq01,rinv01);
1858
1859             /* Calculate table index by multiplying r with table scale and truncate to integer */
1860             rt               = _mm256_mul_ps(r01,vftabscale);
1861             vfitab           = _mm256_cvttps_epi32(rt);
1862             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1863             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1864             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1865             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1866             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1867             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1868
1869             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1870             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1871                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1872             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1873                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1874             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1875                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1876             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1877                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1878             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1879             Heps             = _mm256_mul_ps(vfeps,H);
1880             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1881             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1882             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
1883
1884             fscal            = felec;
1885
1886             /* Calculate temporary vectorial force */
1887             tx               = _mm256_mul_ps(fscal,dx01);
1888             ty               = _mm256_mul_ps(fscal,dy01);
1889             tz               = _mm256_mul_ps(fscal,dz01);
1890
1891             /* Update vectorial force */
1892             fix0             = _mm256_add_ps(fix0,tx);
1893             fiy0             = _mm256_add_ps(fiy0,ty);
1894             fiz0             = _mm256_add_ps(fiz0,tz);
1895
1896             fjx1             = _mm256_add_ps(fjx1,tx);
1897             fjy1             = _mm256_add_ps(fjy1,ty);
1898             fjz1             = _mm256_add_ps(fjz1,tz);
1899
1900             /**************************
1901              * CALCULATE INTERACTIONS *
1902              **************************/
1903
1904             r02              = _mm256_mul_ps(rsq02,rinv02);
1905
1906             /* Calculate table index by multiplying r with table scale and truncate to integer */
1907             rt               = _mm256_mul_ps(r02,vftabscale);
1908             vfitab           = _mm256_cvttps_epi32(rt);
1909             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1910             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1911             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1912             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1913             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1914             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1915
1916             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1917             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1918                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1919             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1920                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1921             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1922                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1923             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1924                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1925             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1926             Heps             = _mm256_mul_ps(vfeps,H);
1927             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1928             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1929             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
1930
1931             fscal            = felec;
1932
1933             /* Calculate temporary vectorial force */
1934             tx               = _mm256_mul_ps(fscal,dx02);
1935             ty               = _mm256_mul_ps(fscal,dy02);
1936             tz               = _mm256_mul_ps(fscal,dz02);
1937
1938             /* Update vectorial force */
1939             fix0             = _mm256_add_ps(fix0,tx);
1940             fiy0             = _mm256_add_ps(fiy0,ty);
1941             fiz0             = _mm256_add_ps(fiz0,tz);
1942
1943             fjx2             = _mm256_add_ps(fjx2,tx);
1944             fjy2             = _mm256_add_ps(fjy2,ty);
1945             fjz2             = _mm256_add_ps(fjz2,tz);
1946
1947             /**************************
1948              * CALCULATE INTERACTIONS *
1949              **************************/
1950
1951             r10              = _mm256_mul_ps(rsq10,rinv10);
1952
1953             /* Calculate table index by multiplying r with table scale and truncate to integer */
1954             rt               = _mm256_mul_ps(r10,vftabscale);
1955             vfitab           = _mm256_cvttps_epi32(rt);
1956             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1957             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1958             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1959             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1960             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1961             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1962
1963             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1964             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1965                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1966             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1967                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1968             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1969                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1970             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1971                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1972             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1973             Heps             = _mm256_mul_ps(vfeps,H);
1974             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1975             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1976             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1977
1978             fscal            = felec;
1979
1980             /* Calculate temporary vectorial force */
1981             tx               = _mm256_mul_ps(fscal,dx10);
1982             ty               = _mm256_mul_ps(fscal,dy10);
1983             tz               = _mm256_mul_ps(fscal,dz10);
1984
1985             /* Update vectorial force */
1986             fix1             = _mm256_add_ps(fix1,tx);
1987             fiy1             = _mm256_add_ps(fiy1,ty);
1988             fiz1             = _mm256_add_ps(fiz1,tz);
1989
1990             fjx0             = _mm256_add_ps(fjx0,tx);
1991             fjy0             = _mm256_add_ps(fjy0,ty);
1992             fjz0             = _mm256_add_ps(fjz0,tz);
1993
1994             /**************************
1995              * CALCULATE INTERACTIONS *
1996              **************************/
1997
1998             r11              = _mm256_mul_ps(rsq11,rinv11);
1999
2000             /* Calculate table index by multiplying r with table scale and truncate to integer */
2001             rt               = _mm256_mul_ps(r11,vftabscale);
2002             vfitab           = _mm256_cvttps_epi32(rt);
2003             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2004             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2005             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2006             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2007             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2008             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2009
2010             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2011             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2012                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2013             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2014                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2015             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2016                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2017             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2018                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2019             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2020             Heps             = _mm256_mul_ps(vfeps,H);
2021             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2022             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2023             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
2024
2025             fscal            = felec;
2026
2027             /* Calculate temporary vectorial force */
2028             tx               = _mm256_mul_ps(fscal,dx11);
2029             ty               = _mm256_mul_ps(fscal,dy11);
2030             tz               = _mm256_mul_ps(fscal,dz11);
2031
2032             /* Update vectorial force */
2033             fix1             = _mm256_add_ps(fix1,tx);
2034             fiy1             = _mm256_add_ps(fiy1,ty);
2035             fiz1             = _mm256_add_ps(fiz1,tz);
2036
2037             fjx1             = _mm256_add_ps(fjx1,tx);
2038             fjy1             = _mm256_add_ps(fjy1,ty);
2039             fjz1             = _mm256_add_ps(fjz1,tz);
2040
2041             /**************************
2042              * CALCULATE INTERACTIONS *
2043              **************************/
2044
2045             r12              = _mm256_mul_ps(rsq12,rinv12);
2046
2047             /* Calculate table index by multiplying r with table scale and truncate to integer */
2048             rt               = _mm256_mul_ps(r12,vftabscale);
2049             vfitab           = _mm256_cvttps_epi32(rt);
2050             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2051             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2052             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2053             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2054             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2055             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2056
2057             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2058             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2059                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2060             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2061                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2062             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2063                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2064             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2065                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2066             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2067             Heps             = _mm256_mul_ps(vfeps,H);
2068             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2069             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2070             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
2071
2072             fscal            = felec;
2073
2074             /* Calculate temporary vectorial force */
2075             tx               = _mm256_mul_ps(fscal,dx12);
2076             ty               = _mm256_mul_ps(fscal,dy12);
2077             tz               = _mm256_mul_ps(fscal,dz12);
2078
2079             /* Update vectorial force */
2080             fix1             = _mm256_add_ps(fix1,tx);
2081             fiy1             = _mm256_add_ps(fiy1,ty);
2082             fiz1             = _mm256_add_ps(fiz1,tz);
2083
2084             fjx2             = _mm256_add_ps(fjx2,tx);
2085             fjy2             = _mm256_add_ps(fjy2,ty);
2086             fjz2             = _mm256_add_ps(fjz2,tz);
2087
2088             /**************************
2089              * CALCULATE INTERACTIONS *
2090              **************************/
2091
2092             r20              = _mm256_mul_ps(rsq20,rinv20);
2093
2094             /* Calculate table index by multiplying r with table scale and truncate to integer */
2095             rt               = _mm256_mul_ps(r20,vftabscale);
2096             vfitab           = _mm256_cvttps_epi32(rt);
2097             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2098             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2099             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2100             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2101             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2102             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2103
2104             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2105             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2106                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2107             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2108                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2109             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2110                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2111             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2112                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2113             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2114             Heps             = _mm256_mul_ps(vfeps,H);
2115             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2116             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2117             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
2118
2119             fscal            = felec;
2120
2121             /* Calculate temporary vectorial force */
2122             tx               = _mm256_mul_ps(fscal,dx20);
2123             ty               = _mm256_mul_ps(fscal,dy20);
2124             tz               = _mm256_mul_ps(fscal,dz20);
2125
2126             /* Update vectorial force */
2127             fix2             = _mm256_add_ps(fix2,tx);
2128             fiy2             = _mm256_add_ps(fiy2,ty);
2129             fiz2             = _mm256_add_ps(fiz2,tz);
2130
2131             fjx0             = _mm256_add_ps(fjx0,tx);
2132             fjy0             = _mm256_add_ps(fjy0,ty);
2133             fjz0             = _mm256_add_ps(fjz0,tz);
2134
2135             /**************************
2136              * CALCULATE INTERACTIONS *
2137              **************************/
2138
2139             r21              = _mm256_mul_ps(rsq21,rinv21);
2140
2141             /* Calculate table index by multiplying r with table scale and truncate to integer */
2142             rt               = _mm256_mul_ps(r21,vftabscale);
2143             vfitab           = _mm256_cvttps_epi32(rt);
2144             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2145             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2146             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2147             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2148             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2149             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2150
2151             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2152             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2153                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2154             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2155                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2156             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2157                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2158             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2159                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2160             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2161             Heps             = _mm256_mul_ps(vfeps,H);
2162             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2163             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2164             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
2165
2166             fscal            = felec;
2167
2168             /* Calculate temporary vectorial force */
2169             tx               = _mm256_mul_ps(fscal,dx21);
2170             ty               = _mm256_mul_ps(fscal,dy21);
2171             tz               = _mm256_mul_ps(fscal,dz21);
2172
2173             /* Update vectorial force */
2174             fix2             = _mm256_add_ps(fix2,tx);
2175             fiy2             = _mm256_add_ps(fiy2,ty);
2176             fiz2             = _mm256_add_ps(fiz2,tz);
2177
2178             fjx1             = _mm256_add_ps(fjx1,tx);
2179             fjy1             = _mm256_add_ps(fjy1,ty);
2180             fjz1             = _mm256_add_ps(fjz1,tz);
2181
2182             /**************************
2183              * CALCULATE INTERACTIONS *
2184              **************************/
2185
2186             r22              = _mm256_mul_ps(rsq22,rinv22);
2187
2188             /* Calculate table index by multiplying r with table scale and truncate to integer */
2189             rt               = _mm256_mul_ps(r22,vftabscale);
2190             vfitab           = _mm256_cvttps_epi32(rt);
2191             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2192             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2193             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2194             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2195             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2196             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2197
2198             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2199             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2200                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2201             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2202                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2203             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2204                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2205             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2206                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2207             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2208             Heps             = _mm256_mul_ps(vfeps,H);
2209             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2210             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2211             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
2212
2213             fscal            = felec;
2214
2215             /* Calculate temporary vectorial force */
2216             tx               = _mm256_mul_ps(fscal,dx22);
2217             ty               = _mm256_mul_ps(fscal,dy22);
2218             tz               = _mm256_mul_ps(fscal,dz22);
2219
2220             /* Update vectorial force */
2221             fix2             = _mm256_add_ps(fix2,tx);
2222             fiy2             = _mm256_add_ps(fiy2,ty);
2223             fiz2             = _mm256_add_ps(fiz2,tz);
2224
2225             fjx2             = _mm256_add_ps(fjx2,tx);
2226             fjy2             = _mm256_add_ps(fjy2,ty);
2227             fjz2             = _mm256_add_ps(fjz2,tz);
2228
2229             fjptrA             = f+j_coord_offsetA;
2230             fjptrB             = f+j_coord_offsetB;
2231             fjptrC             = f+j_coord_offsetC;
2232             fjptrD             = f+j_coord_offsetD;
2233             fjptrE             = f+j_coord_offsetE;
2234             fjptrF             = f+j_coord_offsetF;
2235             fjptrG             = f+j_coord_offsetG;
2236             fjptrH             = f+j_coord_offsetH;
2237
2238             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2239                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2240
2241             /* Inner loop uses 373 flops */
2242         }
2243
2244         if(jidx<j_index_end)
2245         {
2246
2247             /* Get j neighbor index, and coordinate index */
2248             jnrlistA         = jjnr[jidx];
2249             jnrlistB         = jjnr[jidx+1];
2250             jnrlistC         = jjnr[jidx+2];
2251             jnrlistD         = jjnr[jidx+3];
2252             jnrlistE         = jjnr[jidx+4];
2253             jnrlistF         = jjnr[jidx+5];
2254             jnrlistG         = jjnr[jidx+6];
2255             jnrlistH         = jjnr[jidx+7];
2256             /* Sign of each element will be negative for non-real atoms.
2257              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2258              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2259              */
2260             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2261                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2262
2263             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2264             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2265             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2266             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2267             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2268             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2269             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2270             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2271             j_coord_offsetA  = DIM*jnrA;
2272             j_coord_offsetB  = DIM*jnrB;
2273             j_coord_offsetC  = DIM*jnrC;
2274             j_coord_offsetD  = DIM*jnrD;
2275             j_coord_offsetE  = DIM*jnrE;
2276             j_coord_offsetF  = DIM*jnrF;
2277             j_coord_offsetG  = DIM*jnrG;
2278             j_coord_offsetH  = DIM*jnrH;
2279
2280             /* load j atom coordinates */
2281             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2282                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2283                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2284                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2285                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2286
2287             /* Calculate displacement vector */
2288             dx00             = _mm256_sub_ps(ix0,jx0);
2289             dy00             = _mm256_sub_ps(iy0,jy0);
2290             dz00             = _mm256_sub_ps(iz0,jz0);
2291             dx01             = _mm256_sub_ps(ix0,jx1);
2292             dy01             = _mm256_sub_ps(iy0,jy1);
2293             dz01             = _mm256_sub_ps(iz0,jz1);
2294             dx02             = _mm256_sub_ps(ix0,jx2);
2295             dy02             = _mm256_sub_ps(iy0,jy2);
2296             dz02             = _mm256_sub_ps(iz0,jz2);
2297             dx10             = _mm256_sub_ps(ix1,jx0);
2298             dy10             = _mm256_sub_ps(iy1,jy0);
2299             dz10             = _mm256_sub_ps(iz1,jz0);
2300             dx11             = _mm256_sub_ps(ix1,jx1);
2301             dy11             = _mm256_sub_ps(iy1,jy1);
2302             dz11             = _mm256_sub_ps(iz1,jz1);
2303             dx12             = _mm256_sub_ps(ix1,jx2);
2304             dy12             = _mm256_sub_ps(iy1,jy2);
2305             dz12             = _mm256_sub_ps(iz1,jz2);
2306             dx20             = _mm256_sub_ps(ix2,jx0);
2307             dy20             = _mm256_sub_ps(iy2,jy0);
2308             dz20             = _mm256_sub_ps(iz2,jz0);
2309             dx21             = _mm256_sub_ps(ix2,jx1);
2310             dy21             = _mm256_sub_ps(iy2,jy1);
2311             dz21             = _mm256_sub_ps(iz2,jz1);
2312             dx22             = _mm256_sub_ps(ix2,jx2);
2313             dy22             = _mm256_sub_ps(iy2,jy2);
2314             dz22             = _mm256_sub_ps(iz2,jz2);
2315
2316             /* Calculate squared distance and things based on it */
2317             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2318             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
2319             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
2320             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
2321             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2322             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2323             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
2324             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2325             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2326
2327             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2328             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
2329             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
2330             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
2331             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2332             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2333             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
2334             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2335             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2336
2337             fjx0             = _mm256_setzero_ps();
2338             fjy0             = _mm256_setzero_ps();
2339             fjz0             = _mm256_setzero_ps();
2340             fjx1             = _mm256_setzero_ps();
2341             fjy1             = _mm256_setzero_ps();
2342             fjz1             = _mm256_setzero_ps();
2343             fjx2             = _mm256_setzero_ps();
2344             fjy2             = _mm256_setzero_ps();
2345             fjz2             = _mm256_setzero_ps();
2346
2347             /**************************
2348              * CALCULATE INTERACTIONS *
2349              **************************/
2350
2351             r00              = _mm256_mul_ps(rsq00,rinv00);
2352             r00              = _mm256_andnot_ps(dummy_mask,r00);
2353
2354             /* Calculate table index by multiplying r with table scale and truncate to integer */
2355             rt               = _mm256_mul_ps(r00,vftabscale);
2356             vfitab           = _mm256_cvttps_epi32(rt);
2357             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2358             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2359             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2360             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2361             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2362             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2363
2364             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2365             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2366                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2367             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2368                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2369             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2370                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2371             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2372                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2373             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2374             Heps             = _mm256_mul_ps(vfeps,H);
2375             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2376             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2377             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
2378
2379             /* CUBIC SPLINE TABLE DISPERSION */
2380             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2381             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2382             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2383                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2384             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2385                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2386             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2388             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2389                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2390             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2391             Heps             = _mm256_mul_ps(vfeps,H);
2392             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2393             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2394             fvdw6            = _mm256_mul_ps(c6_00,FF);
2395
2396             /* CUBIC SPLINE TABLE REPULSION */
2397             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2398             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2399             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2400                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2401             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2402                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2403             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2404                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2405             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2406                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2407             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2408             Heps             = _mm256_mul_ps(vfeps,H);
2409             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2410             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2411             fvdw12           = _mm256_mul_ps(c12_00,FF);
2412             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2413
2414             fscal            = _mm256_add_ps(felec,fvdw);
2415
2416             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2417
2418             /* Calculate temporary vectorial force */
2419             tx               = _mm256_mul_ps(fscal,dx00);
2420             ty               = _mm256_mul_ps(fscal,dy00);
2421             tz               = _mm256_mul_ps(fscal,dz00);
2422
2423             /* Update vectorial force */
2424             fix0             = _mm256_add_ps(fix0,tx);
2425             fiy0             = _mm256_add_ps(fiy0,ty);
2426             fiz0             = _mm256_add_ps(fiz0,tz);
2427
2428             fjx0             = _mm256_add_ps(fjx0,tx);
2429             fjy0             = _mm256_add_ps(fjy0,ty);
2430             fjz0             = _mm256_add_ps(fjz0,tz);
2431
2432             /**************************
2433              * CALCULATE INTERACTIONS *
2434              **************************/
2435
2436             r01              = _mm256_mul_ps(rsq01,rinv01);
2437             r01              = _mm256_andnot_ps(dummy_mask,r01);
2438
2439             /* Calculate table index by multiplying r with table scale and truncate to integer */
2440             rt               = _mm256_mul_ps(r01,vftabscale);
2441             vfitab           = _mm256_cvttps_epi32(rt);
2442             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2443             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2444             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2445             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2446             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2447             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2448
2449             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2450             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2451                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2452             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2453                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2454             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2455                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2456             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2457                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2458             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2459             Heps             = _mm256_mul_ps(vfeps,H);
2460             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2461             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2462             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
2463
2464             fscal            = felec;
2465
2466             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2467
2468             /* Calculate temporary vectorial force */
2469             tx               = _mm256_mul_ps(fscal,dx01);
2470             ty               = _mm256_mul_ps(fscal,dy01);
2471             tz               = _mm256_mul_ps(fscal,dz01);
2472
2473             /* Update vectorial force */
2474             fix0             = _mm256_add_ps(fix0,tx);
2475             fiy0             = _mm256_add_ps(fiy0,ty);
2476             fiz0             = _mm256_add_ps(fiz0,tz);
2477
2478             fjx1             = _mm256_add_ps(fjx1,tx);
2479             fjy1             = _mm256_add_ps(fjy1,ty);
2480             fjz1             = _mm256_add_ps(fjz1,tz);
2481
2482             /**************************
2483              * CALCULATE INTERACTIONS *
2484              **************************/
2485
2486             r02              = _mm256_mul_ps(rsq02,rinv02);
2487             r02              = _mm256_andnot_ps(dummy_mask,r02);
2488
2489             /* Calculate table index by multiplying r with table scale and truncate to integer */
2490             rt               = _mm256_mul_ps(r02,vftabscale);
2491             vfitab           = _mm256_cvttps_epi32(rt);
2492             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2493             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2494             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2495             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2496             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2497             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2498
2499             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2500             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2501                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2502             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2503                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2504             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2505                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2506             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2507                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2508             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2509             Heps             = _mm256_mul_ps(vfeps,H);
2510             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2511             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2512             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
2513
2514             fscal            = felec;
2515
2516             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2517
2518             /* Calculate temporary vectorial force */
2519             tx               = _mm256_mul_ps(fscal,dx02);
2520             ty               = _mm256_mul_ps(fscal,dy02);
2521             tz               = _mm256_mul_ps(fscal,dz02);
2522
2523             /* Update vectorial force */
2524             fix0             = _mm256_add_ps(fix0,tx);
2525             fiy0             = _mm256_add_ps(fiy0,ty);
2526             fiz0             = _mm256_add_ps(fiz0,tz);
2527
2528             fjx2             = _mm256_add_ps(fjx2,tx);
2529             fjy2             = _mm256_add_ps(fjy2,ty);
2530             fjz2             = _mm256_add_ps(fjz2,tz);
2531
2532             /**************************
2533              * CALCULATE INTERACTIONS *
2534              **************************/
2535
2536             r10              = _mm256_mul_ps(rsq10,rinv10);
2537             r10              = _mm256_andnot_ps(dummy_mask,r10);
2538
2539             /* Calculate table index by multiplying r with table scale and truncate to integer */
2540             rt               = _mm256_mul_ps(r10,vftabscale);
2541             vfitab           = _mm256_cvttps_epi32(rt);
2542             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2543             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2544             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2545             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2546             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2547             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2548
2549             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2550             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2551                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2552             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2553                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2554             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2555                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2556             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2557                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2558             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2559             Heps             = _mm256_mul_ps(vfeps,H);
2560             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2561             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2562             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
2563
2564             fscal            = felec;
2565
2566             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2567
2568             /* Calculate temporary vectorial force */
2569             tx               = _mm256_mul_ps(fscal,dx10);
2570             ty               = _mm256_mul_ps(fscal,dy10);
2571             tz               = _mm256_mul_ps(fscal,dz10);
2572
2573             /* Update vectorial force */
2574             fix1             = _mm256_add_ps(fix1,tx);
2575             fiy1             = _mm256_add_ps(fiy1,ty);
2576             fiz1             = _mm256_add_ps(fiz1,tz);
2577
2578             fjx0             = _mm256_add_ps(fjx0,tx);
2579             fjy0             = _mm256_add_ps(fjy0,ty);
2580             fjz0             = _mm256_add_ps(fjz0,tz);
2581
2582             /**************************
2583              * CALCULATE INTERACTIONS *
2584              **************************/
2585
2586             r11              = _mm256_mul_ps(rsq11,rinv11);
2587             r11              = _mm256_andnot_ps(dummy_mask,r11);
2588
2589             /* Calculate table index by multiplying r with table scale and truncate to integer */
2590             rt               = _mm256_mul_ps(r11,vftabscale);
2591             vfitab           = _mm256_cvttps_epi32(rt);
2592             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2593             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2594             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2595             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2596             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2597             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2598
2599             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2600             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2601                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2602             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2603                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2604             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2605                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2606             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2607                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2608             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2609             Heps             = _mm256_mul_ps(vfeps,H);
2610             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2611             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2612             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
2613
2614             fscal            = felec;
2615
2616             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2617
2618             /* Calculate temporary vectorial force */
2619             tx               = _mm256_mul_ps(fscal,dx11);
2620             ty               = _mm256_mul_ps(fscal,dy11);
2621             tz               = _mm256_mul_ps(fscal,dz11);
2622
2623             /* Update vectorial force */
2624             fix1             = _mm256_add_ps(fix1,tx);
2625             fiy1             = _mm256_add_ps(fiy1,ty);
2626             fiz1             = _mm256_add_ps(fiz1,tz);
2627
2628             fjx1             = _mm256_add_ps(fjx1,tx);
2629             fjy1             = _mm256_add_ps(fjy1,ty);
2630             fjz1             = _mm256_add_ps(fjz1,tz);
2631
2632             /**************************
2633              * CALCULATE INTERACTIONS *
2634              **************************/
2635
2636             r12              = _mm256_mul_ps(rsq12,rinv12);
2637             r12              = _mm256_andnot_ps(dummy_mask,r12);
2638
2639             /* Calculate table index by multiplying r with table scale and truncate to integer */
2640             rt               = _mm256_mul_ps(r12,vftabscale);
2641             vfitab           = _mm256_cvttps_epi32(rt);
2642             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2643             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2644             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2645             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2646             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2647             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2648
2649             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2650             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2651                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2652             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2653                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2654             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2655                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2656             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2657                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2658             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2659             Heps             = _mm256_mul_ps(vfeps,H);
2660             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2661             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2662             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
2663
2664             fscal            = felec;
2665
2666             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2667
2668             /* Calculate temporary vectorial force */
2669             tx               = _mm256_mul_ps(fscal,dx12);
2670             ty               = _mm256_mul_ps(fscal,dy12);
2671             tz               = _mm256_mul_ps(fscal,dz12);
2672
2673             /* Update vectorial force */
2674             fix1             = _mm256_add_ps(fix1,tx);
2675             fiy1             = _mm256_add_ps(fiy1,ty);
2676             fiz1             = _mm256_add_ps(fiz1,tz);
2677
2678             fjx2             = _mm256_add_ps(fjx2,tx);
2679             fjy2             = _mm256_add_ps(fjy2,ty);
2680             fjz2             = _mm256_add_ps(fjz2,tz);
2681
2682             /**************************
2683              * CALCULATE INTERACTIONS *
2684              **************************/
2685
2686             r20              = _mm256_mul_ps(rsq20,rinv20);
2687             r20              = _mm256_andnot_ps(dummy_mask,r20);
2688
2689             /* Calculate table index by multiplying r with table scale and truncate to integer */
2690             rt               = _mm256_mul_ps(r20,vftabscale);
2691             vfitab           = _mm256_cvttps_epi32(rt);
2692             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2693             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2694             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2695             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2696             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2697             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2698
2699             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2700             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2701                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2702             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2703                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2704             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2705                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2706             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2707                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2708             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2709             Heps             = _mm256_mul_ps(vfeps,H);
2710             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2711             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2712             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
2713
2714             fscal            = felec;
2715
2716             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2717
2718             /* Calculate temporary vectorial force */
2719             tx               = _mm256_mul_ps(fscal,dx20);
2720             ty               = _mm256_mul_ps(fscal,dy20);
2721             tz               = _mm256_mul_ps(fscal,dz20);
2722
2723             /* Update vectorial force */
2724             fix2             = _mm256_add_ps(fix2,tx);
2725             fiy2             = _mm256_add_ps(fiy2,ty);
2726             fiz2             = _mm256_add_ps(fiz2,tz);
2727
2728             fjx0             = _mm256_add_ps(fjx0,tx);
2729             fjy0             = _mm256_add_ps(fjy0,ty);
2730             fjz0             = _mm256_add_ps(fjz0,tz);
2731
2732             /**************************
2733              * CALCULATE INTERACTIONS *
2734              **************************/
2735
2736             r21              = _mm256_mul_ps(rsq21,rinv21);
2737             r21              = _mm256_andnot_ps(dummy_mask,r21);
2738
2739             /* Calculate table index by multiplying r with table scale and truncate to integer */
2740             rt               = _mm256_mul_ps(r21,vftabscale);
2741             vfitab           = _mm256_cvttps_epi32(rt);
2742             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2743             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2744             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2745             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2746             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2747             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2748
2749             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2750             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2751                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2752             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2753                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2754             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2755                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2756             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2757                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2758             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2759             Heps             = _mm256_mul_ps(vfeps,H);
2760             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2761             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2762             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
2763
2764             fscal            = felec;
2765
2766             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2767
2768             /* Calculate temporary vectorial force */
2769             tx               = _mm256_mul_ps(fscal,dx21);
2770             ty               = _mm256_mul_ps(fscal,dy21);
2771             tz               = _mm256_mul_ps(fscal,dz21);
2772
2773             /* Update vectorial force */
2774             fix2             = _mm256_add_ps(fix2,tx);
2775             fiy2             = _mm256_add_ps(fiy2,ty);
2776             fiz2             = _mm256_add_ps(fiz2,tz);
2777
2778             fjx1             = _mm256_add_ps(fjx1,tx);
2779             fjy1             = _mm256_add_ps(fjy1,ty);
2780             fjz1             = _mm256_add_ps(fjz1,tz);
2781
2782             /**************************
2783              * CALCULATE INTERACTIONS *
2784              **************************/
2785
2786             r22              = _mm256_mul_ps(rsq22,rinv22);
2787             r22              = _mm256_andnot_ps(dummy_mask,r22);
2788
2789             /* Calculate table index by multiplying r with table scale and truncate to integer */
2790             rt               = _mm256_mul_ps(r22,vftabscale);
2791             vfitab           = _mm256_cvttps_epi32(rt);
2792             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2793             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2794             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2795             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2796             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2797             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2798
2799             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2800             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2801                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2802             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2803                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2804             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2805                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2806             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2807                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2808             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2809             Heps             = _mm256_mul_ps(vfeps,H);
2810             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2811             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2812             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
2813
2814             fscal            = felec;
2815
2816             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2817
2818             /* Calculate temporary vectorial force */
2819             tx               = _mm256_mul_ps(fscal,dx22);
2820             ty               = _mm256_mul_ps(fscal,dy22);
2821             tz               = _mm256_mul_ps(fscal,dz22);
2822
2823             /* Update vectorial force */
2824             fix2             = _mm256_add_ps(fix2,tx);
2825             fiy2             = _mm256_add_ps(fiy2,ty);
2826             fiz2             = _mm256_add_ps(fiz2,tz);
2827
2828             fjx2             = _mm256_add_ps(fjx2,tx);
2829             fjy2             = _mm256_add_ps(fjy2,ty);
2830             fjz2             = _mm256_add_ps(fjz2,tz);
2831
2832             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2833             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2834             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2835             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2836             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2837             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2838             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2839             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2840
2841             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2842                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2843
2844             /* Inner loop uses 382 flops */
2845         }
2846
2847         /* End of innermost loop */
2848
2849         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2850                                                  f+i_coord_offset,fshift+i_shift_offset);
2851
2852         /* Increment number of inner iterations */
2853         inneriter                  += j_index_end - j_index_start;
2854
2855         /* Outer loop uses 18 flops */
2856     }
2857
2858     /* Increment number of outer iterations */
2859     outeriter        += nri;
2860
2861     /* Update outer/inner flops */
2862
2863     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*382);
2864 }