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