src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_VF_avx_256_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     real *           vdwioffsetptr3;
  95     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  96     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  97     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  98     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  99     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 100     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
 101     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 102     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
 103     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 104     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 105     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 106     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 107     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 108     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 109     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 110     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 111     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 112     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 113     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 114     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 115     real             *charge;
 116     int              nvdwtype;
 117     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 118     int              *vdwtype;
 119     real             *vdwparam;
 120     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 121     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 122     __m256i          vfitab;
 123     __m128i          vfitab_lo,vfitab_hi;
 124     __m128i          ifour       = _mm_set1_epi32(4);
 125     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 126     real             *vftab;
 127     __m256           dummy_mask,cutoff_mask;
 128     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 129     __m256           one     = _mm256_set1_ps(1.0);
 130     __m256           two     = _mm256_set1_ps(2.0);
 131     x                = xx[0];
 132     f                = ff[0];
 133
 134     nri              = nlist->nri;
 135     iinr             = nlist->iinr;
 136     jindex           = nlist->jindex;
 137     jjnr             = nlist->jjnr;
 138     shiftidx         = nlist->shift;
 139     gid              = nlist->gid;
 140     shiftvec         = fr->shift_vec[0];
 141     fshift           = fr->fshift[0];
 142     facel            = _mm256_set1_ps(fr->epsfac);
 143     charge           = mdatoms->chargeA;
 144     nvdwtype         = fr->ntype;
 145     vdwparam         = fr->nbfp;
 146     vdwtype          = mdatoms->typeA;
 147
 148     vftab            = kernel_data->table_elec_vdw->data;
 149     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 150
 151     /* Setup water-specific parameters */
 152     inr              = nlist->iinr[0];
 153     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 154     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 155     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 156     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 157
 158     jq1              = _mm256_set1_ps(charge[inr+1]);
 159     jq2              = _mm256_set1_ps(charge[inr+2]);
 160     jq3              = _mm256_set1_ps(charge[inr+3]);
 161     vdwjidx0A        = 2*vdwtype[inr+0];
 162     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
 163     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
 164     qq11             = _mm256_mul_ps(iq1,jq1);
 165     qq12             = _mm256_mul_ps(iq1,jq2);
 166     qq13             = _mm256_mul_ps(iq1,jq3);
 167     qq21             = _mm256_mul_ps(iq2,jq1);
 168     qq22             = _mm256_mul_ps(iq2,jq2);
 169     qq23             = _mm256_mul_ps(iq2,jq3);
 170     qq31             = _mm256_mul_ps(iq3,jq1);
 171     qq32             = _mm256_mul_ps(iq3,jq2);
 172     qq33             = _mm256_mul_ps(iq3,jq3);
 173
 174     /* Avoid stupid compiler warnings */
 175     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 176     j_coord_offsetA = 0;
 177     j_coord_offsetB = 0;
 178     j_coord_offsetC = 0;
 179     j_coord_offsetD = 0;
 180     j_coord_offsetE = 0;
 181     j_coord_offsetF = 0;
 182     j_coord_offsetG = 0;
 183     j_coord_offsetH = 0;
 184
 185     outeriter        = 0;
 186     inneriter        = 0;
 187
 188     for(iidx=0;iidx<4*DIM;iidx++)
 189     {
 190         scratch[iidx] = 0.0;
 191     }
 192
 193     /* Start outer loop over neighborlists */
 194     for(iidx=0; iidx<nri; iidx++)
 195     {
 196         /* Load shift vector for this list */
 197         i_shift_offset   = DIM*shiftidx[iidx];
 198
 199         /* Load limits for loop over neighbors */
 200         j_index_start    = jindex[iidx];
 201         j_index_end      = jindex[iidx+1];
 202
 203         /* Get outer coordinate index */
 204         inr              = iinr[iidx];
 205         i_coord_offset   = DIM*inr;
 206
 207         /* Load i particle coords and add shift vector */
 208         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 209                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 210
 211         fix0             = _mm256_setzero_ps();
 212         fiy0             = _mm256_setzero_ps();
 213         fiz0             = _mm256_setzero_ps();
 214         fix1             = _mm256_setzero_ps();
 215         fiy1             = _mm256_setzero_ps();
 216         fiz1             = _mm256_setzero_ps();
 217         fix2             = _mm256_setzero_ps();
 218         fiy2             = _mm256_setzero_ps();
 219         fiz2             = _mm256_setzero_ps();
 220         fix3             = _mm256_setzero_ps();
 221         fiy3             = _mm256_setzero_ps();
 222         fiz3             = _mm256_setzero_ps();
 223
 224         /* Reset potential sums */
 225         velecsum         = _mm256_setzero_ps();
 226         vvdwsum          = _mm256_setzero_ps();
 227
 228         /* Start inner kernel loop */
 229         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 230         {
 231
 232             /* Get j neighbor index, and coordinate index */
 233             jnrA             = jjnr[jidx];
 234             jnrB             = jjnr[jidx+1];
 235             jnrC             = jjnr[jidx+2];
 236             jnrD             = jjnr[jidx+3];
 237             jnrE             = jjnr[jidx+4];
 238             jnrF             = jjnr[jidx+5];
 239             jnrG             = jjnr[jidx+6];
 240             jnrH             = jjnr[jidx+7];
 241             j_coord_offsetA  = DIM*jnrA;
 242             j_coord_offsetB  = DIM*jnrB;
 243             j_coord_offsetC  = DIM*jnrC;
 244             j_coord_offsetD  = DIM*jnrD;
 245             j_coord_offsetE  = DIM*jnrE;
 246             j_coord_offsetF  = DIM*jnrF;
 247             j_coord_offsetG  = DIM*jnrG;
 248             j_coord_offsetH  = DIM*jnrH;
 249
 250             /* load j atom coordinates */
 251             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 252                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 253                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 254                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 255                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 256                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _mm256_sub_ps(ix0,jx0);
 260             dy00             = _mm256_sub_ps(iy0,jy0);
 261             dz00             = _mm256_sub_ps(iz0,jz0);
 262             dx11             = _mm256_sub_ps(ix1,jx1);
 263             dy11             = _mm256_sub_ps(iy1,jy1);
 264             dz11             = _mm256_sub_ps(iz1,jz1);
 265             dx12             = _mm256_sub_ps(ix1,jx2);
 266             dy12             = _mm256_sub_ps(iy1,jy2);
 267             dz12             = _mm256_sub_ps(iz1,jz2);
 268             dx13             = _mm256_sub_ps(ix1,jx3);
 269             dy13             = _mm256_sub_ps(iy1,jy3);
 270             dz13             = _mm256_sub_ps(iz1,jz3);
 271             dx21             = _mm256_sub_ps(ix2,jx1);
 272             dy21             = _mm256_sub_ps(iy2,jy1);
 273             dz21             = _mm256_sub_ps(iz2,jz1);
 274             dx22             = _mm256_sub_ps(ix2,jx2);
 275             dy22             = _mm256_sub_ps(iy2,jy2);
 276             dz22             = _mm256_sub_ps(iz2,jz2);
 277             dx23             = _mm256_sub_ps(ix2,jx3);
 278             dy23             = _mm256_sub_ps(iy2,jy3);
 279             dz23             = _mm256_sub_ps(iz2,jz3);
 280             dx31             = _mm256_sub_ps(ix3,jx1);
 281             dy31             = _mm256_sub_ps(iy3,jy1);
 282             dz31             = _mm256_sub_ps(iz3,jz1);
 283             dx32             = _mm256_sub_ps(ix3,jx2);
 284             dy32             = _mm256_sub_ps(iy3,jy2);
 285             dz32             = _mm256_sub_ps(iz3,jz2);
 286             dx33             = _mm256_sub_ps(ix3,jx3);
 287             dy33             = _mm256_sub_ps(iy3,jy3);
 288             dz33             = _mm256_sub_ps(iz3,jz3);
 289
 290             /* Calculate squared distance and things based on it */
 291             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 292             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 293             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 294             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 295             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 296             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 297             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 298             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 299             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 300             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 301
 302             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 303             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 304             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 305             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 306             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 307             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 308             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 309             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 310             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 311             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 312
 313             fjx0             = _mm256_setzero_ps();
 314             fjy0             = _mm256_setzero_ps();
 315             fjz0             = _mm256_setzero_ps();
 316             fjx1             = _mm256_setzero_ps();
 317             fjy1             = _mm256_setzero_ps();
 318             fjz1             = _mm256_setzero_ps();
 319             fjx2             = _mm256_setzero_ps();
 320             fjy2             = _mm256_setzero_ps();
 321             fjz2             = _mm256_setzero_ps();
 322             fjx3             = _mm256_setzero_ps();
 323             fjy3             = _mm256_setzero_ps();
 324             fjz3             = _mm256_setzero_ps();
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             r00              = _mm256_mul_ps(rsq00,rinv00);
 331
 332             /* Calculate table index by multiplying r with table scale and truncate to integer */
 333             rt               = _mm256_mul_ps(r00,vftabscale);
 334             vfitab           = _mm256_cvttps_epi32(rt);
 335             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 336             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 337             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 338             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 339             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 340             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 341
 342             /* CUBIC SPLINE TABLE DISPERSION */
 343             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 344             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 345             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 346                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 347             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 348                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 349             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 350                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 351             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 352                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 353             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 354             Heps             = _mm256_mul_ps(vfeps,H);
 355             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 356             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 357             vvdw6            = _mm256_mul_ps(c6_00,VV);
 358             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 359             fvdw6            = _mm256_mul_ps(c6_00,FF);
 360
 361             /* CUBIC SPLINE TABLE REPULSION */
 362             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 363             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 364             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 365                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 366             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 367                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 368             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 369                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 370             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 371                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 372             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 373             Heps             = _mm256_mul_ps(vfeps,H);
 374             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 375             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 376             vvdw12           = _mm256_mul_ps(c12_00,VV);
 377             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 378             fvdw12           = _mm256_mul_ps(c12_00,FF);
 379             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 380             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 384
 385             fscal            = fvdw;
 386
 387             /* Calculate temporary vectorial force */
 388             tx               = _mm256_mul_ps(fscal,dx00);
 389             ty               = _mm256_mul_ps(fscal,dy00);
 390             tz               = _mm256_mul_ps(fscal,dz00);
 391
 392             /* Update vectorial force */
 393             fix0             = _mm256_add_ps(fix0,tx);
 394             fiy0             = _mm256_add_ps(fiy0,ty);
 395             fiz0             = _mm256_add_ps(fiz0,tz);
 396
 397             fjx0             = _mm256_add_ps(fjx0,tx);
 398             fjy0             = _mm256_add_ps(fjy0,ty);
 399             fjz0             = _mm256_add_ps(fjz0,tz);
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             r11              = _mm256_mul_ps(rsq11,rinv11);
 406
 407             /* Calculate table index by multiplying r with table scale and truncate to integer */
 408             rt               = _mm256_mul_ps(r11,vftabscale);
 409             vfitab           = _mm256_cvttps_epi32(rt);
 410             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 411             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 412             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 413             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 414             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 415             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 416
 417             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 418             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 419                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 420             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 421                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 422             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 423                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 424             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 425                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 426             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 427             Heps             = _mm256_mul_ps(vfeps,H);
 428             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 429             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 430             velec            = _mm256_mul_ps(qq11,VV);
 431             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 432             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velecsum         = _mm256_add_ps(velecsum,velec);
 436
 437             fscal            = felec;
 438
 439             /* Calculate temporary vectorial force */
 440             tx               = _mm256_mul_ps(fscal,dx11);
 441             ty               = _mm256_mul_ps(fscal,dy11);
 442             tz               = _mm256_mul_ps(fscal,dz11);
 443
 444             /* Update vectorial force */
 445             fix1             = _mm256_add_ps(fix1,tx);
 446             fiy1             = _mm256_add_ps(fiy1,ty);
 447             fiz1             = _mm256_add_ps(fiz1,tz);
 448
 449             fjx1             = _mm256_add_ps(fjx1,tx);
 450             fjy1             = _mm256_add_ps(fjy1,ty);
 451             fjz1             = _mm256_add_ps(fjz1,tz);
 452
 453             /**************************
 454              * CALCULATE INTERACTIONS *
 455              **************************/
 456
 457             r12              = _mm256_mul_ps(rsq12,rinv12);
 458
 459             /* Calculate table index by multiplying r with table scale and truncate to integer */
 460             rt               = _mm256_mul_ps(r12,vftabscale);
 461             vfitab           = _mm256_cvttps_epi32(rt);
 462             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 463             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 464             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 465             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 466             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 467             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 468
 469             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 470             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 471                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 472             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 473                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 474             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 475                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 476             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 477                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 478             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 479             Heps             = _mm256_mul_ps(vfeps,H);
 480             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 481             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 482             velec            = _mm256_mul_ps(qq12,VV);
 483             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 484             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             velecsum         = _mm256_add_ps(velecsum,velec);
 488
 489             fscal            = felec;
 490
 491             /* Calculate temporary vectorial force */
 492             tx               = _mm256_mul_ps(fscal,dx12);
 493             ty               = _mm256_mul_ps(fscal,dy12);
 494             tz               = _mm256_mul_ps(fscal,dz12);
 495
 496             /* Update vectorial force */
 497             fix1             = _mm256_add_ps(fix1,tx);
 498             fiy1             = _mm256_add_ps(fiy1,ty);
 499             fiz1             = _mm256_add_ps(fiz1,tz);
 500
 501             fjx2             = _mm256_add_ps(fjx2,tx);
 502             fjy2             = _mm256_add_ps(fjy2,ty);
 503             fjz2             = _mm256_add_ps(fjz2,tz);
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             r13              = _mm256_mul_ps(rsq13,rinv13);
 510
 511             /* Calculate table index by multiplying r with table scale and truncate to integer */
 512             rt               = _mm256_mul_ps(r13,vftabscale);
 513             vfitab           = _mm256_cvttps_epi32(rt);
 514             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 515             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 516             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 517             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 518             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 519             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 520
 521             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 522             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 523                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 524             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 525                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 526             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 527                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 528             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 529                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 530             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 531             Heps             = _mm256_mul_ps(vfeps,H);
 532             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 533             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 534             velec            = _mm256_mul_ps(qq13,VV);
 535             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 536             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq13,FF),_mm256_mul_ps(vftabscale,rinv13)));
 537
 538             /* Update potential sum for this i atom from the interaction with this j atom. */
 539             velecsum         = _mm256_add_ps(velecsum,velec);
 540
 541             fscal            = felec;
 542
 543             /* Calculate temporary vectorial force */
 544             tx               = _mm256_mul_ps(fscal,dx13);
 545             ty               = _mm256_mul_ps(fscal,dy13);
 546             tz               = _mm256_mul_ps(fscal,dz13);
 547
 548             /* Update vectorial force */
 549             fix1             = _mm256_add_ps(fix1,tx);
 550             fiy1             = _mm256_add_ps(fiy1,ty);
 551             fiz1             = _mm256_add_ps(fiz1,tz);
 552
 553             fjx3             = _mm256_add_ps(fjx3,tx);
 554             fjy3             = _mm256_add_ps(fjy3,ty);
 555             fjz3             = _mm256_add_ps(fjz3,tz);
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             r21              = _mm256_mul_ps(rsq21,rinv21);
 562
 563             /* Calculate table index by multiplying r with table scale and truncate to integer */
 564             rt               = _mm256_mul_ps(r21,vftabscale);
 565             vfitab           = _mm256_cvttps_epi32(rt);
 566             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 567             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 568             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 569             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 570             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 571             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 572
 573             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 574             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 575                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 576             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 577                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 578             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 579                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 580             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 581                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 582             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 583             Heps             = _mm256_mul_ps(vfeps,H);
 584             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 585             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 586             velec            = _mm256_mul_ps(qq21,VV);
 587             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 588             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             velecsum         = _mm256_add_ps(velecsum,velec);
 592
 593             fscal            = felec;
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm256_mul_ps(fscal,dx21);
 597             ty               = _mm256_mul_ps(fscal,dy21);
 598             tz               = _mm256_mul_ps(fscal,dz21);
 599
 600             /* Update vectorial force */
 601             fix2             = _mm256_add_ps(fix2,tx);
 602             fiy2             = _mm256_add_ps(fiy2,ty);
 603             fiz2             = _mm256_add_ps(fiz2,tz);
 604
 605             fjx1             = _mm256_add_ps(fjx1,tx);
 606             fjy1             = _mm256_add_ps(fjy1,ty);
 607             fjz1             = _mm256_add_ps(fjz1,tz);
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             r22              = _mm256_mul_ps(rsq22,rinv22);
 614
 615             /* Calculate table index by multiplying r with table scale and truncate to integer */
 616             rt               = _mm256_mul_ps(r22,vftabscale);
 617             vfitab           = _mm256_cvttps_epi32(rt);
 618             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 619             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 620             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 621             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 622             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 623             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 624
 625             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 626             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 627                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 628             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 629                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 630             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 631                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 632             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 633                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 634             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 635             Heps             = _mm256_mul_ps(vfeps,H);
 636             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 637             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 638             velec            = _mm256_mul_ps(qq22,VV);
 639             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 640             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velecsum         = _mm256_add_ps(velecsum,velec);
 644
 645             fscal            = felec;
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm256_mul_ps(fscal,dx22);
 649             ty               = _mm256_mul_ps(fscal,dy22);
 650             tz               = _mm256_mul_ps(fscal,dz22);
 651
 652             /* Update vectorial force */
 653             fix2             = _mm256_add_ps(fix2,tx);
 654             fiy2             = _mm256_add_ps(fiy2,ty);
 655             fiz2             = _mm256_add_ps(fiz2,tz);
 656
 657             fjx2             = _mm256_add_ps(fjx2,tx);
 658             fjy2             = _mm256_add_ps(fjy2,ty);
 659             fjz2             = _mm256_add_ps(fjz2,tz);
 660
 661             /**************************
 662              * CALCULATE INTERACTIONS *
 663              **************************/
 664
 665             r23              = _mm256_mul_ps(rsq23,rinv23);
 666
 667             /* Calculate table index by multiplying r with table scale and truncate to integer */
 668             rt               = _mm256_mul_ps(r23,vftabscale);
 669             vfitab           = _mm256_cvttps_epi32(rt);
 670             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 671             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 672             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 673             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 674             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 675             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 676
 677             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 678             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 679                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 680             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 681                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 682             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 683                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 684             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 685                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 686             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 687             Heps             = _mm256_mul_ps(vfeps,H);
 688             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 689             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 690             velec            = _mm256_mul_ps(qq23,VV);
 691             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 692             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq23,FF),_mm256_mul_ps(vftabscale,rinv23)));
 693
 694             /* Update potential sum for this i atom from the interaction with this j atom. */
 695             velecsum         = _mm256_add_ps(velecsum,velec);
 696
 697             fscal            = felec;
 698
 699             /* Calculate temporary vectorial force */
 700             tx               = _mm256_mul_ps(fscal,dx23);
 701             ty               = _mm256_mul_ps(fscal,dy23);
 702             tz               = _mm256_mul_ps(fscal,dz23);
 703
 704             /* Update vectorial force */
 705             fix2             = _mm256_add_ps(fix2,tx);
 706             fiy2             = _mm256_add_ps(fiy2,ty);
 707             fiz2             = _mm256_add_ps(fiz2,tz);
 708
 709             fjx3             = _mm256_add_ps(fjx3,tx);
 710             fjy3             = _mm256_add_ps(fjy3,ty);
 711             fjz3             = _mm256_add_ps(fjz3,tz);
 712
 713             /**************************
 714              * CALCULATE INTERACTIONS *
 715              **************************/
 716
 717             r31              = _mm256_mul_ps(rsq31,rinv31);
 718
 719             /* Calculate table index by multiplying r with table scale and truncate to integer */
 720             rt               = _mm256_mul_ps(r31,vftabscale);
 721             vfitab           = _mm256_cvttps_epi32(rt);
 722             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 723             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 724             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 725             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 726             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 727             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 728
 729             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 730             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 731                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 732             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 733                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 734             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 735                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 736             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 737                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 738             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 739             Heps             = _mm256_mul_ps(vfeps,H);
 740             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 741             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 742             velec            = _mm256_mul_ps(qq31,VV);
 743             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 744             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq31,FF),_mm256_mul_ps(vftabscale,rinv31)));
 745
 746             /* Update potential sum for this i atom from the interaction with this j atom. */
 747             velecsum         = _mm256_add_ps(velecsum,velec);
 748
 749             fscal            = felec;
 750
 751             /* Calculate temporary vectorial force */
 752             tx               = _mm256_mul_ps(fscal,dx31);
 753             ty               = _mm256_mul_ps(fscal,dy31);
 754             tz               = _mm256_mul_ps(fscal,dz31);
 755
 756             /* Update vectorial force */
 757             fix3             = _mm256_add_ps(fix3,tx);
 758             fiy3             = _mm256_add_ps(fiy3,ty);
 759             fiz3             = _mm256_add_ps(fiz3,tz);
 760
 761             fjx1             = _mm256_add_ps(fjx1,tx);
 762             fjy1             = _mm256_add_ps(fjy1,ty);
 763             fjz1             = _mm256_add_ps(fjz1,tz);
 764
 765             /**************************
 766              * CALCULATE INTERACTIONS *
 767              **************************/
 768
 769             r32              = _mm256_mul_ps(rsq32,rinv32);
 770
 771             /* Calculate table index by multiplying r with table scale and truncate to integer */
 772             rt               = _mm256_mul_ps(r32,vftabscale);
 773             vfitab           = _mm256_cvttps_epi32(rt);
 774             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 775             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 776             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 777             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 778             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 779             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 780
 781             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 782             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 783                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 784             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 785                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 786             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 787                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 788             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 789                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 790             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 791             Heps             = _mm256_mul_ps(vfeps,H);
 792             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 793             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 794             velec            = _mm256_mul_ps(qq32,VV);
 795             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 796             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq32,FF),_mm256_mul_ps(vftabscale,rinv32)));
 797
 798             /* Update potential sum for this i atom from the interaction with this j atom. */
 799             velecsum         = _mm256_add_ps(velecsum,velec);
 800
 801             fscal            = felec;
 802
 803             /* Calculate temporary vectorial force */
 804             tx               = _mm256_mul_ps(fscal,dx32);
 805             ty               = _mm256_mul_ps(fscal,dy32);
 806             tz               = _mm256_mul_ps(fscal,dz32);
 807
 808             /* Update vectorial force */
 809             fix3             = _mm256_add_ps(fix3,tx);
 810             fiy3             = _mm256_add_ps(fiy3,ty);
 811             fiz3             = _mm256_add_ps(fiz3,tz);
 812
 813             fjx2             = _mm256_add_ps(fjx2,tx);
 814             fjy2             = _mm256_add_ps(fjy2,ty);
 815             fjz2             = _mm256_add_ps(fjz2,tz);
 816
 817             /**************************
 818              * CALCULATE INTERACTIONS *
 819              **************************/
 820
 821             r33              = _mm256_mul_ps(rsq33,rinv33);
 822
 823             /* Calculate table index by multiplying r with table scale and truncate to integer */
 824             rt               = _mm256_mul_ps(r33,vftabscale);
 825             vfitab           = _mm256_cvttps_epi32(rt);
 826             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 827             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 828             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 829             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 830             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 831             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 832
 833             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 834             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 835                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 836             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 837                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 838             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 839                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 840             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 841                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 842             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 843             Heps             = _mm256_mul_ps(vfeps,H);
 844             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 845             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 846             velec            = _mm256_mul_ps(qq33,VV);
 847             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 848             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq33,FF),_mm256_mul_ps(vftabscale,rinv33)));
 849
 850             /* Update potential sum for this i atom from the interaction with this j atom. */
 851             velecsum         = _mm256_add_ps(velecsum,velec);
 852
 853             fscal            = felec;
 854
 855             /* Calculate temporary vectorial force */
 856             tx               = _mm256_mul_ps(fscal,dx33);
 857             ty               = _mm256_mul_ps(fscal,dy33);
 858             tz               = _mm256_mul_ps(fscal,dz33);
 859
 860             /* Update vectorial force */
 861             fix3             = _mm256_add_ps(fix3,tx);
 862             fiy3             = _mm256_add_ps(fiy3,ty);
 863             fiz3             = _mm256_add_ps(fiz3,tz);
 864
 865             fjx3             = _mm256_add_ps(fjx3,tx);
 866             fjy3             = _mm256_add_ps(fjy3,ty);
 867             fjz3             = _mm256_add_ps(fjz3,tz);
 868
 869             fjptrA             = f+j_coord_offsetA;
 870             fjptrB             = f+j_coord_offsetB;
 871             fjptrC             = f+j_coord_offsetC;
 872             fjptrD             = f+j_coord_offsetD;
 873             fjptrE             = f+j_coord_offsetE;
 874             fjptrF             = f+j_coord_offsetF;
 875             fjptrG             = f+j_coord_offsetG;
 876             fjptrH             = f+j_coord_offsetH;
 877
 878             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 879                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 880                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 881
 882             /* Inner loop uses 446 flops */
 883         }
 884
 885         if(jidx<j_index_end)
 886         {
 887
 888             /* Get j neighbor index, and coordinate index */
 889             jnrlistA         = jjnr[jidx];
 890             jnrlistB         = jjnr[jidx+1];
 891             jnrlistC         = jjnr[jidx+2];
 892             jnrlistD         = jjnr[jidx+3];
 893             jnrlistE         = jjnr[jidx+4];
 894             jnrlistF         = jjnr[jidx+5];
 895             jnrlistG         = jjnr[jidx+6];
 896             jnrlistH         = jjnr[jidx+7];
 897             /* Sign of each element will be negative for non-real atoms.
 898              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 899              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 900              */
 901             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 902                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 903
 904             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 905             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 906             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 907             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 908             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 909             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 910             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 911             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 912             j_coord_offsetA  = DIM*jnrA;
 913             j_coord_offsetB  = DIM*jnrB;
 914             j_coord_offsetC  = DIM*jnrC;
 915             j_coord_offsetD  = DIM*jnrD;
 916             j_coord_offsetE  = DIM*jnrE;
 917             j_coord_offsetF  = DIM*jnrF;
 918             j_coord_offsetG  = DIM*jnrG;
 919             j_coord_offsetH  = DIM*jnrH;
 920
 921             /* load j atom coordinates */
 922             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 923                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 924                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 925                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 926                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 927                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 928
 929             /* Calculate displacement vector */
 930             dx00             = _mm256_sub_ps(ix0,jx0);
 931             dy00             = _mm256_sub_ps(iy0,jy0);
 932             dz00             = _mm256_sub_ps(iz0,jz0);
 933             dx11             = _mm256_sub_ps(ix1,jx1);
 934             dy11             = _mm256_sub_ps(iy1,jy1);
 935             dz11             = _mm256_sub_ps(iz1,jz1);
 936             dx12             = _mm256_sub_ps(ix1,jx2);
 937             dy12             = _mm256_sub_ps(iy1,jy2);
 938             dz12             = _mm256_sub_ps(iz1,jz2);
 939             dx13             = _mm256_sub_ps(ix1,jx3);
 940             dy13             = _mm256_sub_ps(iy1,jy3);
 941             dz13             = _mm256_sub_ps(iz1,jz3);
 942             dx21             = _mm256_sub_ps(ix2,jx1);
 943             dy21             = _mm256_sub_ps(iy2,jy1);
 944             dz21             = _mm256_sub_ps(iz2,jz1);
 945             dx22             = _mm256_sub_ps(ix2,jx2);
 946             dy22             = _mm256_sub_ps(iy2,jy2);
 947             dz22             = _mm256_sub_ps(iz2,jz2);
 948             dx23             = _mm256_sub_ps(ix2,jx3);
 949             dy23             = _mm256_sub_ps(iy2,jy3);
 950             dz23             = _mm256_sub_ps(iz2,jz3);
 951             dx31             = _mm256_sub_ps(ix3,jx1);
 952             dy31             = _mm256_sub_ps(iy3,jy1);
 953             dz31             = _mm256_sub_ps(iz3,jz1);
 954             dx32             = _mm256_sub_ps(ix3,jx2);
 955             dy32             = _mm256_sub_ps(iy3,jy2);
 956             dz32             = _mm256_sub_ps(iz3,jz2);
 957             dx33             = _mm256_sub_ps(ix3,jx3);
 958             dy33             = _mm256_sub_ps(iy3,jy3);
 959             dz33             = _mm256_sub_ps(iz3,jz3);
 960
 961             /* Calculate squared distance and things based on it */
 962             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 963             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 964             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 965             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 966             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 967             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 968             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 969             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 970             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 971             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 972
 973             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 974             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 975             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 976             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 977             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 978             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 979             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 980             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 981             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 982             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 983
 984             fjx0             = _mm256_setzero_ps();
 985             fjy0             = _mm256_setzero_ps();
 986             fjz0             = _mm256_setzero_ps();
 987             fjx1             = _mm256_setzero_ps();
 988             fjy1             = _mm256_setzero_ps();
 989             fjz1             = _mm256_setzero_ps();
 990             fjx2             = _mm256_setzero_ps();
 991             fjy2             = _mm256_setzero_ps();
 992             fjz2             = _mm256_setzero_ps();
 993             fjx3             = _mm256_setzero_ps();
 994             fjy3             = _mm256_setzero_ps();
 995             fjz3             = _mm256_setzero_ps();
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             r00              = _mm256_mul_ps(rsq00,rinv00);
1002             r00              = _mm256_andnot_ps(dummy_mask,r00);
1003
1004             /* Calculate table index by multiplying r with table scale and truncate to integer */
1005             rt               = _mm256_mul_ps(r00,vftabscale);
1006             vfitab           = _mm256_cvttps_epi32(rt);
1007             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1008             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1009             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1010             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1011             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1012             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1013
1014             /* CUBIC SPLINE TABLE DISPERSION */
1015             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1016             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1017             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1018                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1019             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1020                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1021             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1022                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1023             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1024                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1025             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1026             Heps             = _mm256_mul_ps(vfeps,H);
1027             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1028             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1029             vvdw6            = _mm256_mul_ps(c6_00,VV);
1030             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1031             fvdw6            = _mm256_mul_ps(c6_00,FF);
1032
1033             /* CUBIC SPLINE TABLE REPULSION */
1034             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1035             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1036             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1037                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1038             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1039                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1040             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1041                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1042             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1043                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1044             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1045             Heps             = _mm256_mul_ps(vfeps,H);
1046             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1047             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1048             vvdw12           = _mm256_mul_ps(c12_00,VV);
1049             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1050             fvdw12           = _mm256_mul_ps(c12_00,FF);
1051             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
1052             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1053
1054             /* Update potential sum for this i atom from the interaction with this j atom. */
1055             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
1056             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
1057
1058             fscal            = fvdw;
1059
1060             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1061
1062             /* Calculate temporary vectorial force */
1063             tx               = _mm256_mul_ps(fscal,dx00);
1064             ty               = _mm256_mul_ps(fscal,dy00);
1065             tz               = _mm256_mul_ps(fscal,dz00);
1066
1067             /* Update vectorial force */
1068             fix0             = _mm256_add_ps(fix0,tx);
1069             fiy0             = _mm256_add_ps(fiy0,ty);
1070             fiz0             = _mm256_add_ps(fiz0,tz);
1071
1072             fjx0             = _mm256_add_ps(fjx0,tx);
1073             fjy0             = _mm256_add_ps(fjy0,ty);
1074             fjz0             = _mm256_add_ps(fjz0,tz);
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             r11              = _mm256_mul_ps(rsq11,rinv11);
1081             r11              = _mm256_andnot_ps(dummy_mask,r11);
1082
1083             /* Calculate table index by multiplying r with table scale and truncate to integer */
1084             rt               = _mm256_mul_ps(r11,vftabscale);
1085             vfitab           = _mm256_cvttps_epi32(rt);
1086             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1087             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1088             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1089             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1090             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1091             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1092
1093             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1094             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1095                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1096             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1097                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1098             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1099                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1100             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1101                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1102             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1103             Heps             = _mm256_mul_ps(vfeps,H);
1104             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1105             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1106             velec            = _mm256_mul_ps(qq11,VV);
1107             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1108             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
1109
1110             /* Update potential sum for this i atom from the interaction with this j atom. */
1111             velec            = _mm256_andnot_ps(dummy_mask,velec);
1112             velecsum         = _mm256_add_ps(velecsum,velec);
1113
1114             fscal            = felec;
1115
1116             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm256_mul_ps(fscal,dx11);
1120             ty               = _mm256_mul_ps(fscal,dy11);
1121             tz               = _mm256_mul_ps(fscal,dz11);
1122
1123             /* Update vectorial force */
1124             fix1             = _mm256_add_ps(fix1,tx);
1125             fiy1             = _mm256_add_ps(fiy1,ty);
1126             fiz1             = _mm256_add_ps(fiz1,tz);
1127
1128             fjx1             = _mm256_add_ps(fjx1,tx);
1129             fjy1             = _mm256_add_ps(fjy1,ty);
1130             fjz1             = _mm256_add_ps(fjz1,tz);
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             r12              = _mm256_mul_ps(rsq12,rinv12);
1137             r12              = _mm256_andnot_ps(dummy_mask,r12);
1138
1139             /* Calculate table index by multiplying r with table scale and truncate to integer */
1140             rt               = _mm256_mul_ps(r12,vftabscale);
1141             vfitab           = _mm256_cvttps_epi32(rt);
1142             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1143             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1144             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1145             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1146             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1147             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1148
1149             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1150             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1151                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1152             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1153                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1154             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1155                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1156             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1157                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1158             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1159             Heps             = _mm256_mul_ps(vfeps,H);
1160             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1161             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1162             velec            = _mm256_mul_ps(qq12,VV);
1163             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1164             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
1165
1166             /* Update potential sum for this i atom from the interaction with this j atom. */
1167             velec            = _mm256_andnot_ps(dummy_mask,velec);
1168             velecsum         = _mm256_add_ps(velecsum,velec);
1169
1170             fscal            = felec;
1171
1172             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1173
1174             /* Calculate temporary vectorial force */
1175             tx               = _mm256_mul_ps(fscal,dx12);
1176             ty               = _mm256_mul_ps(fscal,dy12);
1177             tz               = _mm256_mul_ps(fscal,dz12);
1178
1179             /* Update vectorial force */
1180             fix1             = _mm256_add_ps(fix1,tx);
1181             fiy1             = _mm256_add_ps(fiy1,ty);
1182             fiz1             = _mm256_add_ps(fiz1,tz);
1183
1184             fjx2             = _mm256_add_ps(fjx2,tx);
1185             fjy2             = _mm256_add_ps(fjy2,ty);
1186             fjz2             = _mm256_add_ps(fjz2,tz);
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             r13              = _mm256_mul_ps(rsq13,rinv13);
1193             r13              = _mm256_andnot_ps(dummy_mask,r13);
1194
1195             /* Calculate table index by multiplying r with table scale and truncate to integer */
1196             rt               = _mm256_mul_ps(r13,vftabscale);
1197             vfitab           = _mm256_cvttps_epi32(rt);
1198             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1199             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1200             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1201             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1202             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1203             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1204
1205             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1206             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1207                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1208             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1209                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1210             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1211                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1212             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1213                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1214             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1215             Heps             = _mm256_mul_ps(vfeps,H);
1216             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1217             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1218             velec            = _mm256_mul_ps(qq13,VV);
1219             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1220             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq13,FF),_mm256_mul_ps(vftabscale,rinv13)));
1221
1222             /* Update potential sum for this i atom from the interaction with this j atom. */
1223             velec            = _mm256_andnot_ps(dummy_mask,velec);
1224             velecsum         = _mm256_add_ps(velecsum,velec);
1225
1226             fscal            = felec;
1227
1228             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1229
1230             /* Calculate temporary vectorial force */
1231             tx               = _mm256_mul_ps(fscal,dx13);
1232             ty               = _mm256_mul_ps(fscal,dy13);
1233             tz               = _mm256_mul_ps(fscal,dz13);
1234
1235             /* Update vectorial force */
1236             fix1             = _mm256_add_ps(fix1,tx);
1237             fiy1             = _mm256_add_ps(fiy1,ty);
1238             fiz1             = _mm256_add_ps(fiz1,tz);
1239
1240             fjx3             = _mm256_add_ps(fjx3,tx);
1241             fjy3             = _mm256_add_ps(fjy3,ty);
1242             fjz3             = _mm256_add_ps(fjz3,tz);
1243
1244             /**************************
1245              * CALCULATE INTERACTIONS *
1246              **************************/
1247
1248             r21              = _mm256_mul_ps(rsq21,rinv21);
1249             r21              = _mm256_andnot_ps(dummy_mask,r21);
1250
1251             /* Calculate table index by multiplying r with table scale and truncate to integer */
1252             rt               = _mm256_mul_ps(r21,vftabscale);
1253             vfitab           = _mm256_cvttps_epi32(rt);
1254             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1255             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1256             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1257             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1258             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1259             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1260
1261             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1262             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1263                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1264             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1265                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1266             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1267                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1268             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1269                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1270             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1271             Heps             = _mm256_mul_ps(vfeps,H);
1272             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1273             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1274             velec            = _mm256_mul_ps(qq21,VV);
1275             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1276             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
1277
1278             /* Update potential sum for this i atom from the interaction with this j atom. */
1279             velec            = _mm256_andnot_ps(dummy_mask,velec);
1280             velecsum         = _mm256_add_ps(velecsum,velec);
1281
1282             fscal            = felec;
1283
1284             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1285
1286             /* Calculate temporary vectorial force */
1287             tx               = _mm256_mul_ps(fscal,dx21);
1288             ty               = _mm256_mul_ps(fscal,dy21);
1289             tz               = _mm256_mul_ps(fscal,dz21);
1290
1291             /* Update vectorial force */
1292             fix2             = _mm256_add_ps(fix2,tx);
1293             fiy2             = _mm256_add_ps(fiy2,ty);
1294             fiz2             = _mm256_add_ps(fiz2,tz);
1295
1296             fjx1             = _mm256_add_ps(fjx1,tx);
1297             fjy1             = _mm256_add_ps(fjy1,ty);
1298             fjz1             = _mm256_add_ps(fjz1,tz);
1299
1300             /**************************
1301              * CALCULATE INTERACTIONS *
1302              **************************/
1303
1304             r22              = _mm256_mul_ps(rsq22,rinv22);
1305             r22              = _mm256_andnot_ps(dummy_mask,r22);
1306
1307             /* Calculate table index by multiplying r with table scale and truncate to integer */
1308             rt               = _mm256_mul_ps(r22,vftabscale);
1309             vfitab           = _mm256_cvttps_epi32(rt);
1310             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1311             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1312             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1313             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1314             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1315             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1316
1317             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1318             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1319                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1320             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1321                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1322             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1323                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1324             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1326             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1327             Heps             = _mm256_mul_ps(vfeps,H);
1328             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1329             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1330             velec            = _mm256_mul_ps(qq22,VV);
1331             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1332             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
1333
1334             /* Update potential sum for this i atom from the interaction with this j atom. */
1335             velec            = _mm256_andnot_ps(dummy_mask,velec);
1336             velecsum         = _mm256_add_ps(velecsum,velec);
1337
1338             fscal            = felec;
1339
1340             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1341
1342             /* Calculate temporary vectorial force */
1343             tx               = _mm256_mul_ps(fscal,dx22);
1344             ty               = _mm256_mul_ps(fscal,dy22);
1345             tz               = _mm256_mul_ps(fscal,dz22);
1346
1347             /* Update vectorial force */
1348             fix2             = _mm256_add_ps(fix2,tx);
1349             fiy2             = _mm256_add_ps(fiy2,ty);
1350             fiz2             = _mm256_add_ps(fiz2,tz);
1351
1352             fjx2             = _mm256_add_ps(fjx2,tx);
1353             fjy2             = _mm256_add_ps(fjy2,ty);
1354             fjz2             = _mm256_add_ps(fjz2,tz);
1355
1356             /**************************
1357              * CALCULATE INTERACTIONS *
1358              **************************/
1359
1360             r23              = _mm256_mul_ps(rsq23,rinv23);
1361             r23              = _mm256_andnot_ps(dummy_mask,r23);
1362
1363             /* Calculate table index by multiplying r with table scale and truncate to integer */
1364             rt               = _mm256_mul_ps(r23,vftabscale);
1365             vfitab           = _mm256_cvttps_epi32(rt);
1366             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1367             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1368             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1369             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1370             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1371             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1372
1373             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1374             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1375                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1376             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1377                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1378             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1379                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1380             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1381                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1382             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1383             Heps             = _mm256_mul_ps(vfeps,H);
1384             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1385             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1386             velec            = _mm256_mul_ps(qq23,VV);
1387             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1388             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq23,FF),_mm256_mul_ps(vftabscale,rinv23)));
1389
1390             /* Update potential sum for this i atom from the interaction with this j atom. */
1391             velec            = _mm256_andnot_ps(dummy_mask,velec);
1392             velecsum         = _mm256_add_ps(velecsum,velec);
1393
1394             fscal            = felec;
1395
1396             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1397
1398             /* Calculate temporary vectorial force */
1399             tx               = _mm256_mul_ps(fscal,dx23);
1400             ty               = _mm256_mul_ps(fscal,dy23);
1401             tz               = _mm256_mul_ps(fscal,dz23);
1402
1403             /* Update vectorial force */
1404             fix2             = _mm256_add_ps(fix2,tx);
1405             fiy2             = _mm256_add_ps(fiy2,ty);
1406             fiz2             = _mm256_add_ps(fiz2,tz);
1407
1408             fjx3             = _mm256_add_ps(fjx3,tx);
1409             fjy3             = _mm256_add_ps(fjy3,ty);
1410             fjz3             = _mm256_add_ps(fjz3,tz);
1411
1412             /**************************
1413              * CALCULATE INTERACTIONS *
1414              **************************/
1415
1416             r31              = _mm256_mul_ps(rsq31,rinv31);
1417             r31              = _mm256_andnot_ps(dummy_mask,r31);
1418
1419             /* Calculate table index by multiplying r with table scale and truncate to integer */
1420             rt               = _mm256_mul_ps(r31,vftabscale);
1421             vfitab           = _mm256_cvttps_epi32(rt);
1422             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1423             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1424             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1425             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1426             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1427             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1428
1429             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1430             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1431                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1432             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1433                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1434             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1435                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1436             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1437                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1438             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1439             Heps             = _mm256_mul_ps(vfeps,H);
1440             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1441             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1442             velec            = _mm256_mul_ps(qq31,VV);
1443             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1444             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq31,FF),_mm256_mul_ps(vftabscale,rinv31)));
1445
1446             /* Update potential sum for this i atom from the interaction with this j atom. */
1447             velec            = _mm256_andnot_ps(dummy_mask,velec);
1448             velecsum         = _mm256_add_ps(velecsum,velec);
1449
1450             fscal            = felec;
1451
1452             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1453
1454             /* Calculate temporary vectorial force */
1455             tx               = _mm256_mul_ps(fscal,dx31);
1456             ty               = _mm256_mul_ps(fscal,dy31);
1457             tz               = _mm256_mul_ps(fscal,dz31);
1458
1459             /* Update vectorial force */
1460             fix3             = _mm256_add_ps(fix3,tx);
1461             fiy3             = _mm256_add_ps(fiy3,ty);
1462             fiz3             = _mm256_add_ps(fiz3,tz);
1463
1464             fjx1             = _mm256_add_ps(fjx1,tx);
1465             fjy1             = _mm256_add_ps(fjy1,ty);
1466             fjz1             = _mm256_add_ps(fjz1,tz);
1467
1468             /**************************
1469              * CALCULATE INTERACTIONS *
1470              **************************/
1471
1472             r32              = _mm256_mul_ps(rsq32,rinv32);
1473             r32              = _mm256_andnot_ps(dummy_mask,r32);
1474
1475             /* Calculate table index by multiplying r with table scale and truncate to integer */
1476             rt               = _mm256_mul_ps(r32,vftabscale);
1477             vfitab           = _mm256_cvttps_epi32(rt);
1478             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1479             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1480             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1481             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1482             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1483             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1484
1485             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1486             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1487                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1488             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1489                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1490             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1491                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1492             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1493                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1494             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1495             Heps             = _mm256_mul_ps(vfeps,H);
1496             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1497             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1498             velec            = _mm256_mul_ps(qq32,VV);
1499             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1500             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq32,FF),_mm256_mul_ps(vftabscale,rinv32)));
1501
1502             /* Update potential sum for this i atom from the interaction with this j atom. */
1503             velec            = _mm256_andnot_ps(dummy_mask,velec);
1504             velecsum         = _mm256_add_ps(velecsum,velec);
1505
1506             fscal            = felec;
1507
1508             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1509
1510             /* Calculate temporary vectorial force */
1511             tx               = _mm256_mul_ps(fscal,dx32);
1512             ty               = _mm256_mul_ps(fscal,dy32);
1513             tz               = _mm256_mul_ps(fscal,dz32);
1514
1515             /* Update vectorial force */
1516             fix3             = _mm256_add_ps(fix3,tx);
1517             fiy3             = _mm256_add_ps(fiy3,ty);
1518             fiz3             = _mm256_add_ps(fiz3,tz);
1519
1520             fjx2             = _mm256_add_ps(fjx2,tx);
1521             fjy2             = _mm256_add_ps(fjy2,ty);
1522             fjz2             = _mm256_add_ps(fjz2,tz);
1523
1524             /**************************
1525              * CALCULATE INTERACTIONS *
1526              **************************/
1527
1528             r33              = _mm256_mul_ps(rsq33,rinv33);
1529             r33              = _mm256_andnot_ps(dummy_mask,r33);
1530
1531             /* Calculate table index by multiplying r with table scale and truncate to integer */
1532             rt               = _mm256_mul_ps(r33,vftabscale);
1533             vfitab           = _mm256_cvttps_epi32(rt);
1534             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1535             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1536             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1537             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1538             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1539             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1540
1541             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1542             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1543                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1544             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1545                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1546             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1547                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1548             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1549                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1550             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1551             Heps             = _mm256_mul_ps(vfeps,H);
1552             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1553             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1554             velec            = _mm256_mul_ps(qq33,VV);
1555             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1556             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq33,FF),_mm256_mul_ps(vftabscale,rinv33)));
1557
1558             /* Update potential sum for this i atom from the interaction with this j atom. */
1559             velec            = _mm256_andnot_ps(dummy_mask,velec);
1560             velecsum         = _mm256_add_ps(velecsum,velec);
1561
1562             fscal            = felec;
1563
1564             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1565
1566             /* Calculate temporary vectorial force */
1567             tx               = _mm256_mul_ps(fscal,dx33);
1568             ty               = _mm256_mul_ps(fscal,dy33);
1569             tz               = _mm256_mul_ps(fscal,dz33);
1570
1571             /* Update vectorial force */
1572             fix3             = _mm256_add_ps(fix3,tx);
1573             fiy3             = _mm256_add_ps(fiy3,ty);
1574             fiz3             = _mm256_add_ps(fiz3,tz);
1575
1576             fjx3             = _mm256_add_ps(fjx3,tx);
1577             fjy3             = _mm256_add_ps(fjy3,ty);
1578             fjz3             = _mm256_add_ps(fjz3,tz);
1579
1580             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1581             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1582             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1583             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1584             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1585             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1586             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1587             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1588
1589             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1590                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1591                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1592
1593             /* Inner loop uses 456 flops */
1594         }
1595
1596         /* End of innermost loop */
1597
1598         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1599                                                  f+i_coord_offset,fshift+i_shift_offset);
1600
1601         ggid                        = gid[iidx];
1602         /* Update potential energies */
1603         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1604         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1605
1606         /* Increment number of inner iterations */
1607         inneriter                  += j_index_end - j_index_start;
1608
1609         /* Outer loop uses 26 flops */
1610     }
1611
1612     /* Increment number of outer iterations */
1613     outeriter        += nri;
1614
1615     /* Update outer/inner flops */
1616
1617     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*456);
1618 }
1619 /*
1620  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_F_avx_256_single
1621  * Electrostatics interaction: CubicSplineTable
1622  * VdW interaction:            CubicSplineTable
1623  * Geometry:                   Water4-Water4
1624  * Calculate force/pot:        Force
1625  */
1626 void
1627 nb_kernel_ElecCSTab_VdwCSTab_GeomW4W4_F_avx_256_single
1628                     (t_nblist                    * gmx_restrict       nlist,
1629                      rvec                        * gmx_restrict          xx,
1630                      rvec                        * gmx_restrict          ff,
1631                      t_forcerec                  * gmx_restrict          fr,
1632                      t_mdatoms                   * gmx_restrict     mdatoms,
1633                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1634                      t_nrnb                      * gmx_restrict        nrnb)
1635 {
1636     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1637      * just 0 for non-waters.
1638      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1639      * jnr indices corresponding to data put in the four positions in the SIMD register.
1640      */
1641     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1642     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1643     int              jnrA,jnrB,jnrC,jnrD;
1644     int              jnrE,jnrF,jnrG,jnrH;
1645     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1646     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1647     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1648     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1649     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1650     real             rcutoff_scalar;
1651     real             *shiftvec,*fshift,*x,*f;
1652     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1653     real             scratch[4*DIM];
1654     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1655     real *           vdwioffsetptr0;
1656     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1657     real *           vdwioffsetptr1;
1658     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1659     real *           vdwioffsetptr2;
1660     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1661     real *           vdwioffsetptr3;
1662     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1663     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1664     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1665     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1666     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1667     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1668     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1669     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1670     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1671     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1672     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1673     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1674     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1675     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1676     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1677     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1678     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1679     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1680     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1681     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1682     real             *charge;
1683     int              nvdwtype;
1684     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1685     int              *vdwtype;
1686     real             *vdwparam;
1687     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
1688     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
1689     __m256i          vfitab;
1690     __m128i          vfitab_lo,vfitab_hi;
1691     __m128i          ifour       = _mm_set1_epi32(4);
1692     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1693     real             *vftab;
1694     __m256           dummy_mask,cutoff_mask;
1695     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1696     __m256           one     = _mm256_set1_ps(1.0);
1697     __m256           two     = _mm256_set1_ps(2.0);
1698     x                = xx[0];
1699     f                = ff[0];
1700
1701     nri              = nlist->nri;
1702     iinr             = nlist->iinr;
1703     jindex           = nlist->jindex;
1704     jjnr             = nlist->jjnr;
1705     shiftidx         = nlist->shift;
1706     gid              = nlist->gid;
1707     shiftvec         = fr->shift_vec[0];
1708     fshift           = fr->fshift[0];
1709     facel            = _mm256_set1_ps(fr->epsfac);
1710     charge           = mdatoms->chargeA;
1711     nvdwtype         = fr->ntype;
1712     vdwparam         = fr->nbfp;
1713     vdwtype          = mdatoms->typeA;
1714
1715     vftab            = kernel_data->table_elec_vdw->data;
1716     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
1717
1718     /* Setup water-specific parameters */
1719     inr              = nlist->iinr[0];
1720     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1721     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1722     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1723     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1724
1725     jq1              = _mm256_set1_ps(charge[inr+1]);
1726     jq2              = _mm256_set1_ps(charge[inr+2]);
1727     jq3              = _mm256_set1_ps(charge[inr+3]);
1728     vdwjidx0A        = 2*vdwtype[inr+0];
1729     c6_00            = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A]);
1730     c12_00           = _mm256_set1_ps(vdwioffsetptr0[vdwjidx0A+1]);
1731     qq11             = _mm256_mul_ps(iq1,jq1);
1732     qq12             = _mm256_mul_ps(iq1,jq2);
1733     qq13             = _mm256_mul_ps(iq1,jq3);
1734     qq21             = _mm256_mul_ps(iq2,jq1);
1735     qq22             = _mm256_mul_ps(iq2,jq2);
1736     qq23             = _mm256_mul_ps(iq2,jq3);
1737     qq31             = _mm256_mul_ps(iq3,jq1);
1738     qq32             = _mm256_mul_ps(iq3,jq2);
1739     qq33             = _mm256_mul_ps(iq3,jq3);
1740
1741     /* Avoid stupid compiler warnings */
1742     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1743     j_coord_offsetA = 0;
1744     j_coord_offsetB = 0;
1745     j_coord_offsetC = 0;
1746     j_coord_offsetD = 0;
1747     j_coord_offsetE = 0;
1748     j_coord_offsetF = 0;
1749     j_coord_offsetG = 0;
1750     j_coord_offsetH = 0;
1751
1752     outeriter        = 0;
1753     inneriter        = 0;
1754
1755     for(iidx=0;iidx<4*DIM;iidx++)
1756     {
1757         scratch[iidx] = 0.0;
1758     }
1759
1760     /* Start outer loop over neighborlists */
1761     for(iidx=0; iidx<nri; iidx++)
1762     {
1763         /* Load shift vector for this list */
1764         i_shift_offset   = DIM*shiftidx[iidx];
1765
1766         /* Load limits for loop over neighbors */
1767         j_index_start    = jindex[iidx];
1768         j_index_end      = jindex[iidx+1];
1769
1770         /* Get outer coordinate index */
1771         inr              = iinr[iidx];
1772         i_coord_offset   = DIM*inr;
1773
1774         /* Load i particle coords and add shift vector */
1775         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1776                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1777
1778         fix0             = _mm256_setzero_ps();
1779         fiy0             = _mm256_setzero_ps();
1780         fiz0             = _mm256_setzero_ps();
1781         fix1             = _mm256_setzero_ps();
1782         fiy1             = _mm256_setzero_ps();
1783         fiz1             = _mm256_setzero_ps();
1784         fix2             = _mm256_setzero_ps();
1785         fiy2             = _mm256_setzero_ps();
1786         fiz2             = _mm256_setzero_ps();
1787         fix3             = _mm256_setzero_ps();
1788         fiy3             = _mm256_setzero_ps();
1789         fiz3             = _mm256_setzero_ps();
1790
1791         /* Start inner kernel loop */
1792         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1793         {
1794
1795             /* Get j neighbor index, and coordinate index */
1796             jnrA             = jjnr[jidx];
1797             jnrB             = jjnr[jidx+1];
1798             jnrC             = jjnr[jidx+2];
1799             jnrD             = jjnr[jidx+3];
1800             jnrE             = jjnr[jidx+4];
1801             jnrF             = jjnr[jidx+5];
1802             jnrG             = jjnr[jidx+6];
1803             jnrH             = jjnr[jidx+7];
1804             j_coord_offsetA  = DIM*jnrA;
1805             j_coord_offsetB  = DIM*jnrB;
1806             j_coord_offsetC  = DIM*jnrC;
1807             j_coord_offsetD  = DIM*jnrD;
1808             j_coord_offsetE  = DIM*jnrE;
1809             j_coord_offsetF  = DIM*jnrF;
1810             j_coord_offsetG  = DIM*jnrG;
1811             j_coord_offsetH  = DIM*jnrH;
1812
1813             /* load j atom coordinates */
1814             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1815                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1816                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1817                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1818                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1819                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
1820
1821             /* Calculate displacement vector */
1822             dx00             = _mm256_sub_ps(ix0,jx0);
1823             dy00             = _mm256_sub_ps(iy0,jy0);
1824             dz00             = _mm256_sub_ps(iz0,jz0);
1825             dx11             = _mm256_sub_ps(ix1,jx1);
1826             dy11             = _mm256_sub_ps(iy1,jy1);
1827             dz11             = _mm256_sub_ps(iz1,jz1);
1828             dx12             = _mm256_sub_ps(ix1,jx2);
1829             dy12             = _mm256_sub_ps(iy1,jy2);
1830             dz12             = _mm256_sub_ps(iz1,jz2);
1831             dx13             = _mm256_sub_ps(ix1,jx3);
1832             dy13             = _mm256_sub_ps(iy1,jy3);
1833             dz13             = _mm256_sub_ps(iz1,jz3);
1834             dx21             = _mm256_sub_ps(ix2,jx1);
1835             dy21             = _mm256_sub_ps(iy2,jy1);
1836             dz21             = _mm256_sub_ps(iz2,jz1);
1837             dx22             = _mm256_sub_ps(ix2,jx2);
1838             dy22             = _mm256_sub_ps(iy2,jy2);
1839             dz22             = _mm256_sub_ps(iz2,jz2);
1840             dx23             = _mm256_sub_ps(ix2,jx3);
1841             dy23             = _mm256_sub_ps(iy2,jy3);
1842             dz23             = _mm256_sub_ps(iz2,jz3);
1843             dx31             = _mm256_sub_ps(ix3,jx1);
1844             dy31             = _mm256_sub_ps(iy3,jy1);
1845             dz31             = _mm256_sub_ps(iz3,jz1);
1846             dx32             = _mm256_sub_ps(ix3,jx2);
1847             dy32             = _mm256_sub_ps(iy3,jy2);
1848             dz32             = _mm256_sub_ps(iz3,jz2);
1849             dx33             = _mm256_sub_ps(ix3,jx3);
1850             dy33             = _mm256_sub_ps(iy3,jy3);
1851             dz33             = _mm256_sub_ps(iz3,jz3);
1852
1853             /* Calculate squared distance and things based on it */
1854             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1855             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1856             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1857             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1858             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1859             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1860             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1861             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1862             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1863             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1864
1865             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1866             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1867             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1868             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1869             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1870             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1871             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1872             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1873             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1874             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1875
1876             fjx0             = _mm256_setzero_ps();
1877             fjy0             = _mm256_setzero_ps();
1878             fjz0             = _mm256_setzero_ps();
1879             fjx1             = _mm256_setzero_ps();
1880             fjy1             = _mm256_setzero_ps();
1881             fjz1             = _mm256_setzero_ps();
1882             fjx2             = _mm256_setzero_ps();
1883             fjy2             = _mm256_setzero_ps();
1884             fjz2             = _mm256_setzero_ps();
1885             fjx3             = _mm256_setzero_ps();
1886             fjy3             = _mm256_setzero_ps();
1887             fjz3             = _mm256_setzero_ps();
1888
1889             /**************************
1890              * CALCULATE INTERACTIONS *
1891              **************************/
1892
1893             r00              = _mm256_mul_ps(rsq00,rinv00);
1894
1895             /* Calculate table index by multiplying r with table scale and truncate to integer */
1896             rt               = _mm256_mul_ps(r00,vftabscale);
1897             vfitab           = _mm256_cvttps_epi32(rt);
1898             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1899             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1900             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1901             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1902             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1903             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1904
1905             /* CUBIC SPLINE TABLE DISPERSION */
1906             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1907             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1908             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1909                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1910             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1911                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1912             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1913                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1914             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1915                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1916             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1917             Heps             = _mm256_mul_ps(vfeps,H);
1918             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1919             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1920             fvdw6            = _mm256_mul_ps(c6_00,FF);
1921
1922             /* CUBIC SPLINE TABLE REPULSION */
1923             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1924             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1925             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1926                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1927             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1928                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1929             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1930                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1931             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1932                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1933             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1934             Heps             = _mm256_mul_ps(vfeps,H);
1935             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1936             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1937             fvdw12           = _mm256_mul_ps(c12_00,FF);
1938             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1939
1940             fscal            = fvdw;
1941
1942             /* Calculate temporary vectorial force */
1943             tx               = _mm256_mul_ps(fscal,dx00);
1944             ty               = _mm256_mul_ps(fscal,dy00);
1945             tz               = _mm256_mul_ps(fscal,dz00);
1946
1947             /* Update vectorial force */
1948             fix0             = _mm256_add_ps(fix0,tx);
1949             fiy0             = _mm256_add_ps(fiy0,ty);
1950             fiz0             = _mm256_add_ps(fiz0,tz);
1951
1952             fjx0             = _mm256_add_ps(fjx0,tx);
1953             fjy0             = _mm256_add_ps(fjy0,ty);
1954             fjz0             = _mm256_add_ps(fjz0,tz);
1955
1956             /**************************
1957              * CALCULATE INTERACTIONS *
1958              **************************/
1959
1960             r11              = _mm256_mul_ps(rsq11,rinv11);
1961
1962             /* Calculate table index by multiplying r with table scale and truncate to integer */
1963             rt               = _mm256_mul_ps(r11,vftabscale);
1964             vfitab           = _mm256_cvttps_epi32(rt);
1965             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1966             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1967             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1968             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1969             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1970             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1971
1972             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1973             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1974                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1975             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1976                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1977             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1978                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1979             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1980                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1981             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1982             Heps             = _mm256_mul_ps(vfeps,H);
1983             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1984             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1985             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
1986
1987             fscal            = felec;
1988
1989             /* Calculate temporary vectorial force */
1990             tx               = _mm256_mul_ps(fscal,dx11);
1991             ty               = _mm256_mul_ps(fscal,dy11);
1992             tz               = _mm256_mul_ps(fscal,dz11);
1993
1994             /* Update vectorial force */
1995             fix1             = _mm256_add_ps(fix1,tx);
1996             fiy1             = _mm256_add_ps(fiy1,ty);
1997             fiz1             = _mm256_add_ps(fiz1,tz);
1998
1999             fjx1             = _mm256_add_ps(fjx1,tx);
2000             fjy1             = _mm256_add_ps(fjy1,ty);
2001             fjz1             = _mm256_add_ps(fjz1,tz);
2002
2003             /**************************
2004              * CALCULATE INTERACTIONS *
2005              **************************/
2006
2007             r12              = _mm256_mul_ps(rsq12,rinv12);
2008
2009             /* Calculate table index by multiplying r with table scale and truncate to integer */
2010             rt               = _mm256_mul_ps(r12,vftabscale);
2011             vfitab           = _mm256_cvttps_epi32(rt);
2012             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2013             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2014             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2015             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2016             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2017             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2018
2019             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2020             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2021                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2022             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2023                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2024             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2025                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2026             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2027                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2028             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2029             Heps             = _mm256_mul_ps(vfeps,H);
2030             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2031             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2032             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
2033
2034             fscal            = felec;
2035
2036             /* Calculate temporary vectorial force */
2037             tx               = _mm256_mul_ps(fscal,dx12);
2038             ty               = _mm256_mul_ps(fscal,dy12);
2039             tz               = _mm256_mul_ps(fscal,dz12);
2040
2041             /* Update vectorial force */
2042             fix1             = _mm256_add_ps(fix1,tx);
2043             fiy1             = _mm256_add_ps(fiy1,ty);
2044             fiz1             = _mm256_add_ps(fiz1,tz);
2045
2046             fjx2             = _mm256_add_ps(fjx2,tx);
2047             fjy2             = _mm256_add_ps(fjy2,ty);
2048             fjz2             = _mm256_add_ps(fjz2,tz);
2049
2050             /**************************
2051              * CALCULATE INTERACTIONS *
2052              **************************/
2053
2054             r13              = _mm256_mul_ps(rsq13,rinv13);
2055
2056             /* Calculate table index by multiplying r with table scale and truncate to integer */
2057             rt               = _mm256_mul_ps(r13,vftabscale);
2058             vfitab           = _mm256_cvttps_epi32(rt);
2059             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2060             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2061             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2062             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2063             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2064             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2065
2066             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2067             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2068                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2069             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2070                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2071             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2072                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2073             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2074                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2075             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2076             Heps             = _mm256_mul_ps(vfeps,H);
2077             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2078             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2079             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq13,FF),_mm256_mul_ps(vftabscale,rinv13)));
2080
2081             fscal            = felec;
2082
2083             /* Calculate temporary vectorial force */
2084             tx               = _mm256_mul_ps(fscal,dx13);
2085             ty               = _mm256_mul_ps(fscal,dy13);
2086             tz               = _mm256_mul_ps(fscal,dz13);
2087
2088             /* Update vectorial force */
2089             fix1             = _mm256_add_ps(fix1,tx);
2090             fiy1             = _mm256_add_ps(fiy1,ty);
2091             fiz1             = _mm256_add_ps(fiz1,tz);
2092
2093             fjx3             = _mm256_add_ps(fjx3,tx);
2094             fjy3             = _mm256_add_ps(fjy3,ty);
2095             fjz3             = _mm256_add_ps(fjz3,tz);
2096
2097             /**************************
2098              * CALCULATE INTERACTIONS *
2099              **************************/
2100
2101             r21              = _mm256_mul_ps(rsq21,rinv21);
2102
2103             /* Calculate table index by multiplying r with table scale and truncate to integer */
2104             rt               = _mm256_mul_ps(r21,vftabscale);
2105             vfitab           = _mm256_cvttps_epi32(rt);
2106             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2107             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2108             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2109             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2110             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2111             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2112
2113             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2114             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2115                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2116             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2117                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2118             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2119                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2120             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2121                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2122             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2123             Heps             = _mm256_mul_ps(vfeps,H);
2124             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2125             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2126             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
2127
2128             fscal            = felec;
2129
2130             /* Calculate temporary vectorial force */
2131             tx               = _mm256_mul_ps(fscal,dx21);
2132             ty               = _mm256_mul_ps(fscal,dy21);
2133             tz               = _mm256_mul_ps(fscal,dz21);
2134
2135             /* Update vectorial force */
2136             fix2             = _mm256_add_ps(fix2,tx);
2137             fiy2             = _mm256_add_ps(fiy2,ty);
2138             fiz2             = _mm256_add_ps(fiz2,tz);
2139
2140             fjx1             = _mm256_add_ps(fjx1,tx);
2141             fjy1             = _mm256_add_ps(fjy1,ty);
2142             fjz1             = _mm256_add_ps(fjz1,tz);
2143
2144             /**************************
2145              * CALCULATE INTERACTIONS *
2146              **************************/
2147
2148             r22              = _mm256_mul_ps(rsq22,rinv22);
2149
2150             /* Calculate table index by multiplying r with table scale and truncate to integer */
2151             rt               = _mm256_mul_ps(r22,vftabscale);
2152             vfitab           = _mm256_cvttps_epi32(rt);
2153             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2154             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2155             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2156             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2157             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2158             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2159
2160             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2161             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2162                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2163             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2164                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2165             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2166                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2167             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2168                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2169             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2170             Heps             = _mm256_mul_ps(vfeps,H);
2171             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2172             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2173             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
2174
2175             fscal            = felec;
2176
2177             /* Calculate temporary vectorial force */
2178             tx               = _mm256_mul_ps(fscal,dx22);
2179             ty               = _mm256_mul_ps(fscal,dy22);
2180             tz               = _mm256_mul_ps(fscal,dz22);
2181
2182             /* Update vectorial force */
2183             fix2             = _mm256_add_ps(fix2,tx);
2184             fiy2             = _mm256_add_ps(fiy2,ty);
2185             fiz2             = _mm256_add_ps(fiz2,tz);
2186
2187             fjx2             = _mm256_add_ps(fjx2,tx);
2188             fjy2             = _mm256_add_ps(fjy2,ty);
2189             fjz2             = _mm256_add_ps(fjz2,tz);
2190
2191             /**************************
2192              * CALCULATE INTERACTIONS *
2193              **************************/
2194
2195             r23              = _mm256_mul_ps(rsq23,rinv23);
2196
2197             /* Calculate table index by multiplying r with table scale and truncate to integer */
2198             rt               = _mm256_mul_ps(r23,vftabscale);
2199             vfitab           = _mm256_cvttps_epi32(rt);
2200             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2201             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2202             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2203             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2204             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2205             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2206
2207             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2208             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2209                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2210             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2211                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2212             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2213                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2214             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2215                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2216             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2217             Heps             = _mm256_mul_ps(vfeps,H);
2218             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2219             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2220             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq23,FF),_mm256_mul_ps(vftabscale,rinv23)));
2221
2222             fscal            = felec;
2223
2224             /* Calculate temporary vectorial force */
2225             tx               = _mm256_mul_ps(fscal,dx23);
2226             ty               = _mm256_mul_ps(fscal,dy23);
2227             tz               = _mm256_mul_ps(fscal,dz23);
2228
2229             /* Update vectorial force */
2230             fix2             = _mm256_add_ps(fix2,tx);
2231             fiy2             = _mm256_add_ps(fiy2,ty);
2232             fiz2             = _mm256_add_ps(fiz2,tz);
2233
2234             fjx3             = _mm256_add_ps(fjx3,tx);
2235             fjy3             = _mm256_add_ps(fjy3,ty);
2236             fjz3             = _mm256_add_ps(fjz3,tz);
2237
2238             /**************************
2239              * CALCULATE INTERACTIONS *
2240              **************************/
2241
2242             r31              = _mm256_mul_ps(rsq31,rinv31);
2243
2244             /* Calculate table index by multiplying r with table scale and truncate to integer */
2245             rt               = _mm256_mul_ps(r31,vftabscale);
2246             vfitab           = _mm256_cvttps_epi32(rt);
2247             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2248             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2249             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2250             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2251             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2252             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2253
2254             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2255             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2256                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2257             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2258                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2259             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2260                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2261             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2262                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2263             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2264             Heps             = _mm256_mul_ps(vfeps,H);
2265             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2266             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2267             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq31,FF),_mm256_mul_ps(vftabscale,rinv31)));
2268
2269             fscal            = felec;
2270
2271             /* Calculate temporary vectorial force */
2272             tx               = _mm256_mul_ps(fscal,dx31);
2273             ty               = _mm256_mul_ps(fscal,dy31);
2274             tz               = _mm256_mul_ps(fscal,dz31);
2275
2276             /* Update vectorial force */
2277             fix3             = _mm256_add_ps(fix3,tx);
2278             fiy3             = _mm256_add_ps(fiy3,ty);
2279             fiz3             = _mm256_add_ps(fiz3,tz);
2280
2281             fjx1             = _mm256_add_ps(fjx1,tx);
2282             fjy1             = _mm256_add_ps(fjy1,ty);
2283             fjz1             = _mm256_add_ps(fjz1,tz);
2284
2285             /**************************
2286              * CALCULATE INTERACTIONS *
2287              **************************/
2288
2289             r32              = _mm256_mul_ps(rsq32,rinv32);
2290
2291             /* Calculate table index by multiplying r with table scale and truncate to integer */
2292             rt               = _mm256_mul_ps(r32,vftabscale);
2293             vfitab           = _mm256_cvttps_epi32(rt);
2294             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2295             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2296             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2297             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2298             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2299             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2300
2301             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2302             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2303                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2304             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2305                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2306             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2307                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2308             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2310             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2311             Heps             = _mm256_mul_ps(vfeps,H);
2312             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2313             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2314             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq32,FF),_mm256_mul_ps(vftabscale,rinv32)));
2315
2316             fscal            = felec;
2317
2318             /* Calculate temporary vectorial force */
2319             tx               = _mm256_mul_ps(fscal,dx32);
2320             ty               = _mm256_mul_ps(fscal,dy32);
2321             tz               = _mm256_mul_ps(fscal,dz32);
2322
2323             /* Update vectorial force */
2324             fix3             = _mm256_add_ps(fix3,tx);
2325             fiy3             = _mm256_add_ps(fiy3,ty);
2326             fiz3             = _mm256_add_ps(fiz3,tz);
2327
2328             fjx2             = _mm256_add_ps(fjx2,tx);
2329             fjy2             = _mm256_add_ps(fjy2,ty);
2330             fjz2             = _mm256_add_ps(fjz2,tz);
2331
2332             /**************************
2333              * CALCULATE INTERACTIONS *
2334              **************************/
2335
2336             r33              = _mm256_mul_ps(rsq33,rinv33);
2337
2338             /* Calculate table index by multiplying r with table scale and truncate to integer */
2339             rt               = _mm256_mul_ps(r33,vftabscale);
2340             vfitab           = _mm256_cvttps_epi32(rt);
2341             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2342             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2343             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2344             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2345             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2346             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2347
2348             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2349             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2350                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2351             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2352                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2353             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2354                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2355             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2356                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2357             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2358             Heps             = _mm256_mul_ps(vfeps,H);
2359             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2360             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2361             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq33,FF),_mm256_mul_ps(vftabscale,rinv33)));
2362
2363             fscal            = felec;
2364
2365             /* Calculate temporary vectorial force */
2366             tx               = _mm256_mul_ps(fscal,dx33);
2367             ty               = _mm256_mul_ps(fscal,dy33);
2368             tz               = _mm256_mul_ps(fscal,dz33);
2369
2370             /* Update vectorial force */
2371             fix3             = _mm256_add_ps(fix3,tx);
2372             fiy3             = _mm256_add_ps(fiy3,ty);
2373             fiz3             = _mm256_add_ps(fiz3,tz);
2374
2375             fjx3             = _mm256_add_ps(fjx3,tx);
2376             fjy3             = _mm256_add_ps(fjy3,ty);
2377             fjz3             = _mm256_add_ps(fjz3,tz);
2378
2379             fjptrA             = f+j_coord_offsetA;
2380             fjptrB             = f+j_coord_offsetB;
2381             fjptrC             = f+j_coord_offsetC;
2382             fjptrD             = f+j_coord_offsetD;
2383             fjptrE             = f+j_coord_offsetE;
2384             fjptrF             = f+j_coord_offsetF;
2385             fjptrG             = f+j_coord_offsetG;
2386             fjptrH             = f+j_coord_offsetH;
2387
2388             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2389                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2390                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2391
2392             /* Inner loop uses 402 flops */
2393         }
2394
2395         if(jidx<j_index_end)
2396         {
2397
2398             /* Get j neighbor index, and coordinate index */
2399             jnrlistA         = jjnr[jidx];
2400             jnrlistB         = jjnr[jidx+1];
2401             jnrlistC         = jjnr[jidx+2];
2402             jnrlistD         = jjnr[jidx+3];
2403             jnrlistE         = jjnr[jidx+4];
2404             jnrlistF         = jjnr[jidx+5];
2405             jnrlistG         = jjnr[jidx+6];
2406             jnrlistH         = jjnr[jidx+7];
2407             /* Sign of each element will be negative for non-real atoms.
2408              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2409              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2410              */
2411             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2412                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2413
2414             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2415             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2416             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2417             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2418             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2419             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2420             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2421             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2422             j_coord_offsetA  = DIM*jnrA;
2423             j_coord_offsetB  = DIM*jnrB;
2424             j_coord_offsetC  = DIM*jnrC;
2425             j_coord_offsetD  = DIM*jnrD;
2426             j_coord_offsetE  = DIM*jnrE;
2427             j_coord_offsetF  = DIM*jnrF;
2428             j_coord_offsetG  = DIM*jnrG;
2429             j_coord_offsetH  = DIM*jnrH;
2430
2431             /* load j atom coordinates */
2432             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2433                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2434                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2435                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2436                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2437                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
2438
2439             /* Calculate displacement vector */
2440             dx00             = _mm256_sub_ps(ix0,jx0);
2441             dy00             = _mm256_sub_ps(iy0,jy0);
2442             dz00             = _mm256_sub_ps(iz0,jz0);
2443             dx11             = _mm256_sub_ps(ix1,jx1);
2444             dy11             = _mm256_sub_ps(iy1,jy1);
2445             dz11             = _mm256_sub_ps(iz1,jz1);
2446             dx12             = _mm256_sub_ps(ix1,jx2);
2447             dy12             = _mm256_sub_ps(iy1,jy2);
2448             dz12             = _mm256_sub_ps(iz1,jz2);
2449             dx13             = _mm256_sub_ps(ix1,jx3);
2450             dy13             = _mm256_sub_ps(iy1,jy3);
2451             dz13             = _mm256_sub_ps(iz1,jz3);
2452             dx21             = _mm256_sub_ps(ix2,jx1);
2453             dy21             = _mm256_sub_ps(iy2,jy1);
2454             dz21             = _mm256_sub_ps(iz2,jz1);
2455             dx22             = _mm256_sub_ps(ix2,jx2);
2456             dy22             = _mm256_sub_ps(iy2,jy2);
2457             dz22             = _mm256_sub_ps(iz2,jz2);
2458             dx23             = _mm256_sub_ps(ix2,jx3);
2459             dy23             = _mm256_sub_ps(iy2,jy3);
2460             dz23             = _mm256_sub_ps(iz2,jz3);
2461             dx31             = _mm256_sub_ps(ix3,jx1);
2462             dy31             = _mm256_sub_ps(iy3,jy1);
2463             dz31             = _mm256_sub_ps(iz3,jz1);
2464             dx32             = _mm256_sub_ps(ix3,jx2);
2465             dy32             = _mm256_sub_ps(iy3,jy2);
2466             dz32             = _mm256_sub_ps(iz3,jz2);
2467             dx33             = _mm256_sub_ps(ix3,jx3);
2468             dy33             = _mm256_sub_ps(iy3,jy3);
2469             dz33             = _mm256_sub_ps(iz3,jz3);
2470
2471             /* Calculate squared distance and things based on it */
2472             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2473             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2474             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2475             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2476             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2477             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2478             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2479             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2480             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2481             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2482
2483             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2484             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2485             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2486             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
2487             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2488             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2489             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
2490             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
2491             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
2492             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
2493
2494             fjx0             = _mm256_setzero_ps();
2495             fjy0             = _mm256_setzero_ps();
2496             fjz0             = _mm256_setzero_ps();
2497             fjx1             = _mm256_setzero_ps();
2498             fjy1             = _mm256_setzero_ps();
2499             fjz1             = _mm256_setzero_ps();
2500             fjx2             = _mm256_setzero_ps();
2501             fjy2             = _mm256_setzero_ps();
2502             fjz2             = _mm256_setzero_ps();
2503             fjx3             = _mm256_setzero_ps();
2504             fjy3             = _mm256_setzero_ps();
2505             fjz3             = _mm256_setzero_ps();
2506
2507             /**************************
2508              * CALCULATE INTERACTIONS *
2509              **************************/
2510
2511             r00              = _mm256_mul_ps(rsq00,rinv00);
2512             r00              = _mm256_andnot_ps(dummy_mask,r00);
2513
2514             /* Calculate table index by multiplying r with table scale and truncate to integer */
2515             rt               = _mm256_mul_ps(r00,vftabscale);
2516             vfitab           = _mm256_cvttps_epi32(rt);
2517             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2518             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2519             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2520             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2521             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2522             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2523
2524             /* CUBIC SPLINE TABLE DISPERSION */
2525             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2526             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2527             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2528                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2529             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2530                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2531             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2532                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2533             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2534                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2535             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2536             Heps             = _mm256_mul_ps(vfeps,H);
2537             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2538             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2539             fvdw6            = _mm256_mul_ps(c6_00,FF);
2540
2541             /* CUBIC SPLINE TABLE REPULSION */
2542             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
2543             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
2544             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2545                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2546             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2547                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2548             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2549                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2550             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2551                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2552             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2553             Heps             = _mm256_mul_ps(vfeps,H);
2554             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2555             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2556             fvdw12           = _mm256_mul_ps(c12_00,FF);
2557             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
2558
2559             fscal            = fvdw;
2560
2561             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2562
2563             /* Calculate temporary vectorial force */
2564             tx               = _mm256_mul_ps(fscal,dx00);
2565             ty               = _mm256_mul_ps(fscal,dy00);
2566             tz               = _mm256_mul_ps(fscal,dz00);
2567
2568             /* Update vectorial force */
2569             fix0             = _mm256_add_ps(fix0,tx);
2570             fiy0             = _mm256_add_ps(fiy0,ty);
2571             fiz0             = _mm256_add_ps(fiz0,tz);
2572
2573             fjx0             = _mm256_add_ps(fjx0,tx);
2574             fjy0             = _mm256_add_ps(fjy0,ty);
2575             fjz0             = _mm256_add_ps(fjz0,tz);
2576
2577             /**************************
2578              * CALCULATE INTERACTIONS *
2579              **************************/
2580
2581             r11              = _mm256_mul_ps(rsq11,rinv11);
2582             r11              = _mm256_andnot_ps(dummy_mask,r11);
2583
2584             /* Calculate table index by multiplying r with table scale and truncate to integer */
2585             rt               = _mm256_mul_ps(r11,vftabscale);
2586             vfitab           = _mm256_cvttps_epi32(rt);
2587             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2588             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2589             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2590             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2591             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2592             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2593
2594             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2595             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2596                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2597             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2598                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2599             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2600                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2601             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2602                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2603             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2604             Heps             = _mm256_mul_ps(vfeps,H);
2605             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2606             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2607             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
2608
2609             fscal            = felec;
2610
2611             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2612
2613             /* Calculate temporary vectorial force */
2614             tx               = _mm256_mul_ps(fscal,dx11);
2615             ty               = _mm256_mul_ps(fscal,dy11);
2616             tz               = _mm256_mul_ps(fscal,dz11);
2617
2618             /* Update vectorial force */
2619             fix1             = _mm256_add_ps(fix1,tx);
2620             fiy1             = _mm256_add_ps(fiy1,ty);
2621             fiz1             = _mm256_add_ps(fiz1,tz);
2622
2623             fjx1             = _mm256_add_ps(fjx1,tx);
2624             fjy1             = _mm256_add_ps(fjy1,ty);
2625             fjz1             = _mm256_add_ps(fjz1,tz);
2626
2627             /**************************
2628              * CALCULATE INTERACTIONS *
2629              **************************/
2630
2631             r12              = _mm256_mul_ps(rsq12,rinv12);
2632             r12              = _mm256_andnot_ps(dummy_mask,r12);
2633
2634             /* Calculate table index by multiplying r with table scale and truncate to integer */
2635             rt               = _mm256_mul_ps(r12,vftabscale);
2636             vfitab           = _mm256_cvttps_epi32(rt);
2637             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2638             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2639             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2640             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2641             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2642             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2643
2644             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2645             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2646                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2647             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2648                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2649             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2650                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2651             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2652                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2653             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2654             Heps             = _mm256_mul_ps(vfeps,H);
2655             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2656             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2657             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
2658
2659             fscal            = felec;
2660
2661             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2662
2663             /* Calculate temporary vectorial force */
2664             tx               = _mm256_mul_ps(fscal,dx12);
2665             ty               = _mm256_mul_ps(fscal,dy12);
2666             tz               = _mm256_mul_ps(fscal,dz12);
2667
2668             /* Update vectorial force */
2669             fix1             = _mm256_add_ps(fix1,tx);
2670             fiy1             = _mm256_add_ps(fiy1,ty);
2671             fiz1             = _mm256_add_ps(fiz1,tz);
2672
2673             fjx2             = _mm256_add_ps(fjx2,tx);
2674             fjy2             = _mm256_add_ps(fjy2,ty);
2675             fjz2             = _mm256_add_ps(fjz2,tz);
2676
2677             /**************************
2678              * CALCULATE INTERACTIONS *
2679              **************************/
2680
2681             r13              = _mm256_mul_ps(rsq13,rinv13);
2682             r13              = _mm256_andnot_ps(dummy_mask,r13);
2683
2684             /* Calculate table index by multiplying r with table scale and truncate to integer */
2685             rt               = _mm256_mul_ps(r13,vftabscale);
2686             vfitab           = _mm256_cvttps_epi32(rt);
2687             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2688             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2689             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2690             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2691             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2692             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2693
2694             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2695             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2696                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2697             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2698                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2699             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2700                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2701             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2702                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2703             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2704             Heps             = _mm256_mul_ps(vfeps,H);
2705             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2706             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2707             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq13,FF),_mm256_mul_ps(vftabscale,rinv13)));
2708
2709             fscal            = felec;
2710
2711             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2712
2713             /* Calculate temporary vectorial force */
2714             tx               = _mm256_mul_ps(fscal,dx13);
2715             ty               = _mm256_mul_ps(fscal,dy13);
2716             tz               = _mm256_mul_ps(fscal,dz13);
2717
2718             /* Update vectorial force */
2719             fix1             = _mm256_add_ps(fix1,tx);
2720             fiy1             = _mm256_add_ps(fiy1,ty);
2721             fiz1             = _mm256_add_ps(fiz1,tz);
2722
2723             fjx3             = _mm256_add_ps(fjx3,tx);
2724             fjy3             = _mm256_add_ps(fjy3,ty);
2725             fjz3             = _mm256_add_ps(fjz3,tz);
2726
2727             /**************************
2728              * CALCULATE INTERACTIONS *
2729              **************************/
2730
2731             r21              = _mm256_mul_ps(rsq21,rinv21);
2732             r21              = _mm256_andnot_ps(dummy_mask,r21);
2733
2734             /* Calculate table index by multiplying r with table scale and truncate to integer */
2735             rt               = _mm256_mul_ps(r21,vftabscale);
2736             vfitab           = _mm256_cvttps_epi32(rt);
2737             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2738             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2739             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2740             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2741             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2742             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2743
2744             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2745             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2746                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2747             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2748                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2749             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2750                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2751             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2752                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2753             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2754             Heps             = _mm256_mul_ps(vfeps,H);
2755             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2756             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2757             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
2758
2759             fscal            = felec;
2760
2761             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2762
2763             /* Calculate temporary vectorial force */
2764             tx               = _mm256_mul_ps(fscal,dx21);
2765             ty               = _mm256_mul_ps(fscal,dy21);
2766             tz               = _mm256_mul_ps(fscal,dz21);
2767
2768             /* Update vectorial force */
2769             fix2             = _mm256_add_ps(fix2,tx);
2770             fiy2             = _mm256_add_ps(fiy2,ty);
2771             fiz2             = _mm256_add_ps(fiz2,tz);
2772
2773             fjx1             = _mm256_add_ps(fjx1,tx);
2774             fjy1             = _mm256_add_ps(fjy1,ty);
2775             fjz1             = _mm256_add_ps(fjz1,tz);
2776
2777             /**************************
2778              * CALCULATE INTERACTIONS *
2779              **************************/
2780
2781             r22              = _mm256_mul_ps(rsq22,rinv22);
2782             r22              = _mm256_andnot_ps(dummy_mask,r22);
2783
2784             /* Calculate table index by multiplying r with table scale and truncate to integer */
2785             rt               = _mm256_mul_ps(r22,vftabscale);
2786             vfitab           = _mm256_cvttps_epi32(rt);
2787             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2788             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2789             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2790             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2791             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2792             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2793
2794             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2795             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2796                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2797             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2798                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2799             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2800                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2801             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2802                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2803             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2804             Heps             = _mm256_mul_ps(vfeps,H);
2805             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2806             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2807             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
2808
2809             fscal            = felec;
2810
2811             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2812
2813             /* Calculate temporary vectorial force */
2814             tx               = _mm256_mul_ps(fscal,dx22);
2815             ty               = _mm256_mul_ps(fscal,dy22);
2816             tz               = _mm256_mul_ps(fscal,dz22);
2817
2818             /* Update vectorial force */
2819             fix2             = _mm256_add_ps(fix2,tx);
2820             fiy2             = _mm256_add_ps(fiy2,ty);
2821             fiz2             = _mm256_add_ps(fiz2,tz);
2822
2823             fjx2             = _mm256_add_ps(fjx2,tx);
2824             fjy2             = _mm256_add_ps(fjy2,ty);
2825             fjz2             = _mm256_add_ps(fjz2,tz);
2826
2827             /**************************
2828              * CALCULATE INTERACTIONS *
2829              **************************/
2830
2831             r23              = _mm256_mul_ps(rsq23,rinv23);
2832             r23              = _mm256_andnot_ps(dummy_mask,r23);
2833
2834             /* Calculate table index by multiplying r with table scale and truncate to integer */
2835             rt               = _mm256_mul_ps(r23,vftabscale);
2836             vfitab           = _mm256_cvttps_epi32(rt);
2837             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2838             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2839             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2840             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2841             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2842             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2843
2844             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2845             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2846                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2847             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2848                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2849             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2850                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2851             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2852                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2853             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2854             Heps             = _mm256_mul_ps(vfeps,H);
2855             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2856             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2857             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq23,FF),_mm256_mul_ps(vftabscale,rinv23)));
2858
2859             fscal            = felec;
2860
2861             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2862
2863             /* Calculate temporary vectorial force */
2864             tx               = _mm256_mul_ps(fscal,dx23);
2865             ty               = _mm256_mul_ps(fscal,dy23);
2866             tz               = _mm256_mul_ps(fscal,dz23);
2867
2868             /* Update vectorial force */
2869             fix2             = _mm256_add_ps(fix2,tx);
2870             fiy2             = _mm256_add_ps(fiy2,ty);
2871             fiz2             = _mm256_add_ps(fiz2,tz);
2872
2873             fjx3             = _mm256_add_ps(fjx3,tx);
2874             fjy3             = _mm256_add_ps(fjy3,ty);
2875             fjz3             = _mm256_add_ps(fjz3,tz);
2876
2877             /**************************
2878              * CALCULATE INTERACTIONS *
2879              **************************/
2880
2881             r31              = _mm256_mul_ps(rsq31,rinv31);
2882             r31              = _mm256_andnot_ps(dummy_mask,r31);
2883
2884             /* Calculate table index by multiplying r with table scale and truncate to integer */
2885             rt               = _mm256_mul_ps(r31,vftabscale);
2886             vfitab           = _mm256_cvttps_epi32(rt);
2887             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2888             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2889             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2890             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2891             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2892             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2893
2894             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2895             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2896                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2897             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2898                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2899             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2900                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2901             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2902                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2903             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2904             Heps             = _mm256_mul_ps(vfeps,H);
2905             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2906             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2907             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq31,FF),_mm256_mul_ps(vftabscale,rinv31)));
2908
2909             fscal            = felec;
2910
2911             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2912
2913             /* Calculate temporary vectorial force */
2914             tx               = _mm256_mul_ps(fscal,dx31);
2915             ty               = _mm256_mul_ps(fscal,dy31);
2916             tz               = _mm256_mul_ps(fscal,dz31);
2917
2918             /* Update vectorial force */
2919             fix3             = _mm256_add_ps(fix3,tx);
2920             fiy3             = _mm256_add_ps(fiy3,ty);
2921             fiz3             = _mm256_add_ps(fiz3,tz);
2922
2923             fjx1             = _mm256_add_ps(fjx1,tx);
2924             fjy1             = _mm256_add_ps(fjy1,ty);
2925             fjz1             = _mm256_add_ps(fjz1,tz);
2926
2927             /**************************
2928              * CALCULATE INTERACTIONS *
2929              **************************/
2930
2931             r32              = _mm256_mul_ps(rsq32,rinv32);
2932             r32              = _mm256_andnot_ps(dummy_mask,r32);
2933
2934             /* Calculate table index by multiplying r with table scale and truncate to integer */
2935             rt               = _mm256_mul_ps(r32,vftabscale);
2936             vfitab           = _mm256_cvttps_epi32(rt);
2937             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2938             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2939             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2940             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2941             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2942             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2943
2944             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2945             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2946                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2947             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2948                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2949             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2950                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2951             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2952                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2953             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2954             Heps             = _mm256_mul_ps(vfeps,H);
2955             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2956             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2957             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq32,FF),_mm256_mul_ps(vftabscale,rinv32)));
2958
2959             fscal            = felec;
2960
2961             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2962
2963             /* Calculate temporary vectorial force */
2964             tx               = _mm256_mul_ps(fscal,dx32);
2965             ty               = _mm256_mul_ps(fscal,dy32);
2966             tz               = _mm256_mul_ps(fscal,dz32);
2967
2968             /* Update vectorial force */
2969             fix3             = _mm256_add_ps(fix3,tx);
2970             fiy3             = _mm256_add_ps(fiy3,ty);
2971             fiz3             = _mm256_add_ps(fiz3,tz);
2972
2973             fjx2             = _mm256_add_ps(fjx2,tx);
2974             fjy2             = _mm256_add_ps(fjy2,ty);
2975             fjz2             = _mm256_add_ps(fjz2,tz);
2976
2977             /**************************
2978              * CALCULATE INTERACTIONS *
2979              **************************/
2980
2981             r33              = _mm256_mul_ps(rsq33,rinv33);
2982             r33              = _mm256_andnot_ps(dummy_mask,r33);
2983
2984             /* Calculate table index by multiplying r with table scale and truncate to integer */
2985             rt               = _mm256_mul_ps(r33,vftabscale);
2986             vfitab           = _mm256_cvttps_epi32(rt);
2987             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2988             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2989             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2990             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2991             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
2992             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
2993
2994             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2995             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2996                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2997             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2998                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2999             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
3000                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
3001             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
3002                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
3003             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
3004             Heps             = _mm256_mul_ps(vfeps,H);
3005             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
3006             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
3007             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq33,FF),_mm256_mul_ps(vftabscale,rinv33)));
3008
3009             fscal            = felec;
3010
3011             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
3012
3013             /* Calculate temporary vectorial force */
3014             tx               = _mm256_mul_ps(fscal,dx33);
3015             ty               = _mm256_mul_ps(fscal,dy33);
3016             tz               = _mm256_mul_ps(fscal,dz33);
3017
3018             /* Update vectorial force */
3019             fix3             = _mm256_add_ps(fix3,tx);
3020             fiy3             = _mm256_add_ps(fiy3,ty);
3021             fiz3             = _mm256_add_ps(fiz3,tz);
3022
3023             fjx3             = _mm256_add_ps(fjx3,tx);
3024             fjy3             = _mm256_add_ps(fjy3,ty);
3025             fjz3             = _mm256_add_ps(fjz3,tz);
3026
3027             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
3028             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
3029             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
3030             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
3031             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
3032             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
3033             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
3034             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
3035
3036             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
3037                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
3038                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
3039
3040             /* Inner loop uses 412 flops */
3041         }
3042
3043         /* End of innermost loop */
3044
3045         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
3046                                                  f+i_coord_offset,fshift+i_shift_offset);
3047
3048         /* Increment number of inner iterations */
3049         inneriter                  += j_index_end - j_index_start;
3050
3051         /* Outer loop uses 24 flops */
3052     }
3053
3054     /* Increment number of outer iterations */
3055     outeriter        += nri;
3056
3057     /* Update outer/inner flops */
3058
3059     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*412);
3060 }