src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwNone_GeomW4W4_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4W4_VF_avx_256_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            None
  56  * Geometry:                   Water4-Water4
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           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              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  95     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  96     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  97     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  98     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
  99     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
 100     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 101     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 102     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 103     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 104     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 105     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 106     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 107     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 108     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 109     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     __m256i          vfitab;
 112     __m128i          vfitab_lo,vfitab_hi;
 113     __m128i          ifour       = _mm_set1_epi32(4);
 114     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 115     real             *vftab;
 116     __m256           dummy_mask,cutoff_mask;
 117     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 118     __m256           one     = _mm256_set1_ps(1.0);
 119     __m256           two     = _mm256_set1_ps(2.0);
 120     x                = xx[0];
 121     f                = ff[0];
 122
 123     nri              = nlist->nri;
 124     iinr             = nlist->iinr;
 125     jindex           = nlist->jindex;
 126     jjnr             = nlist->jjnr;
 127     shiftidx         = nlist->shift;
 128     gid              = nlist->gid;
 129     shiftvec         = fr->shift_vec[0];
 130     fshift           = fr->fshift[0];
 131     facel            = _mm256_set1_ps(fr->epsfac);
 132     charge           = mdatoms->chargeA;
 133
 134     vftab            = kernel_data->table_elec->data;
 135     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 140     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 141     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 142
 143     jq1              = _mm256_set1_ps(charge[inr+1]);
 144     jq2              = _mm256_set1_ps(charge[inr+2]);
 145     jq3              = _mm256_set1_ps(charge[inr+3]);
 146     qq11             = _mm256_mul_ps(iq1,jq1);
 147     qq12             = _mm256_mul_ps(iq1,jq2);
 148     qq13             = _mm256_mul_ps(iq1,jq3);
 149     qq21             = _mm256_mul_ps(iq2,jq1);
 150     qq22             = _mm256_mul_ps(iq2,jq2);
 151     qq23             = _mm256_mul_ps(iq2,jq3);
 152     qq31             = _mm256_mul_ps(iq3,jq1);
 153     qq32             = _mm256_mul_ps(iq3,jq2);
 154     qq33             = _mm256_mul_ps(iq3,jq3);
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160     j_coord_offsetC = 0;
 161     j_coord_offsetD = 0;
 162     j_coord_offsetE = 0;
 163     j_coord_offsetF = 0;
 164     j_coord_offsetG = 0;
 165     j_coord_offsetH = 0;
 166
 167     outeriter        = 0;
 168     inneriter        = 0;
 169
 170     for(iidx=0;iidx<4*DIM;iidx++)
 171     {
 172         scratch[iidx] = 0.0;
 173     }
 174
 175     /* Start outer loop over neighborlists */
 176     for(iidx=0; iidx<nri; iidx++)
 177     {
 178         /* Load shift vector for this list */
 179         i_shift_offset   = DIM*shiftidx[iidx];
 180
 181         /* Load limits for loop over neighbors */
 182         j_index_start    = jindex[iidx];
 183         j_index_end      = jindex[iidx+1];
 184
 185         /* Get outer coordinate index */
 186         inr              = iinr[iidx];
 187         i_coord_offset   = DIM*inr;
 188
 189         /* Load i particle coords and add shift vector */
 190         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 191                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 192
 193         fix1             = _mm256_setzero_ps();
 194         fiy1             = _mm256_setzero_ps();
 195         fiz1             = _mm256_setzero_ps();
 196         fix2             = _mm256_setzero_ps();
 197         fiy2             = _mm256_setzero_ps();
 198         fiz2             = _mm256_setzero_ps();
 199         fix3             = _mm256_setzero_ps();
 200         fiy3             = _mm256_setzero_ps();
 201         fiz3             = _mm256_setzero_ps();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm256_setzero_ps();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             jnrE             = jjnr[jidx+4];
 216             jnrF             = jjnr[jidx+5];
 217             jnrG             = jjnr[jidx+6];
 218             jnrH             = jjnr[jidx+7];
 219             j_coord_offsetA  = DIM*jnrA;
 220             j_coord_offsetB  = DIM*jnrB;
 221             j_coord_offsetC  = DIM*jnrC;
 222             j_coord_offsetD  = DIM*jnrD;
 223             j_coord_offsetE  = DIM*jnrE;
 224             j_coord_offsetF  = DIM*jnrF;
 225             j_coord_offsetG  = DIM*jnrG;
 226             j_coord_offsetH  = DIM*jnrH;
 227
 228             /* load j atom coordinates */
 229             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 230                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 231                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 232                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 233                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 234
 235             /* Calculate displacement vector */
 236             dx11             = _mm256_sub_ps(ix1,jx1);
 237             dy11             = _mm256_sub_ps(iy1,jy1);
 238             dz11             = _mm256_sub_ps(iz1,jz1);
 239             dx12             = _mm256_sub_ps(ix1,jx2);
 240             dy12             = _mm256_sub_ps(iy1,jy2);
 241             dz12             = _mm256_sub_ps(iz1,jz2);
 242             dx13             = _mm256_sub_ps(ix1,jx3);
 243             dy13             = _mm256_sub_ps(iy1,jy3);
 244             dz13             = _mm256_sub_ps(iz1,jz3);
 245             dx21             = _mm256_sub_ps(ix2,jx1);
 246             dy21             = _mm256_sub_ps(iy2,jy1);
 247             dz21             = _mm256_sub_ps(iz2,jz1);
 248             dx22             = _mm256_sub_ps(ix2,jx2);
 249             dy22             = _mm256_sub_ps(iy2,jy2);
 250             dz22             = _mm256_sub_ps(iz2,jz2);
 251             dx23             = _mm256_sub_ps(ix2,jx3);
 252             dy23             = _mm256_sub_ps(iy2,jy3);
 253             dz23             = _mm256_sub_ps(iz2,jz3);
 254             dx31             = _mm256_sub_ps(ix3,jx1);
 255             dy31             = _mm256_sub_ps(iy3,jy1);
 256             dz31             = _mm256_sub_ps(iz3,jz1);
 257             dx32             = _mm256_sub_ps(ix3,jx2);
 258             dy32             = _mm256_sub_ps(iy3,jy2);
 259             dz32             = _mm256_sub_ps(iz3,jz2);
 260             dx33             = _mm256_sub_ps(ix3,jx3);
 261             dy33             = _mm256_sub_ps(iy3,jy3);
 262             dz33             = _mm256_sub_ps(iz3,jz3);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 266             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 267             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 268             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 269             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 270             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 271             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 272             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 273             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 274
 275             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 276             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 277             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 278             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 279             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 280             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 281             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 282             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 283             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 284
 285             fjx1             = _mm256_setzero_ps();
 286             fjy1             = _mm256_setzero_ps();
 287             fjz1             = _mm256_setzero_ps();
 288             fjx2             = _mm256_setzero_ps();
 289             fjy2             = _mm256_setzero_ps();
 290             fjz2             = _mm256_setzero_ps();
 291             fjx3             = _mm256_setzero_ps();
 292             fjy3             = _mm256_setzero_ps();
 293             fjz3             = _mm256_setzero_ps();
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             r11              = _mm256_mul_ps(rsq11,rinv11);
 300
 301             /* Calculate table index by multiplying r with table scale and truncate to integer */
 302             rt               = _mm256_mul_ps(r11,vftabscale);
 303             vfitab           = _mm256_cvttps_epi32(rt);
 304             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 305             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 306             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 307             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 308             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 309             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 310
 311             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 312             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 314             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 315                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 316             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 317                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 318             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 319                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 320             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 321             Heps             = _mm256_mul_ps(vfeps,H);
 322             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 323             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 324             velec            = _mm256_mul_ps(qq11,VV);
 325             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 326             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm256_add_ps(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm256_mul_ps(fscal,dx11);
 335             ty               = _mm256_mul_ps(fscal,dy11);
 336             tz               = _mm256_mul_ps(fscal,dz11);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm256_add_ps(fix1,tx);
 340             fiy1             = _mm256_add_ps(fiy1,ty);
 341             fiz1             = _mm256_add_ps(fiz1,tz);
 342
 343             fjx1             = _mm256_add_ps(fjx1,tx);
 344             fjy1             = _mm256_add_ps(fjy1,ty);
 345             fjz1             = _mm256_add_ps(fjz1,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r12              = _mm256_mul_ps(rsq12,rinv12);
 352
 353             /* Calculate table index by multiplying r with table scale and truncate to integer */
 354             rt               = _mm256_mul_ps(r12,vftabscale);
 355             vfitab           = _mm256_cvttps_epi32(rt);
 356             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 357             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 358             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 359             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 360             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 361             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 362
 363             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 364             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 365                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 366             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 367                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 368             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 369                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 370             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 371                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 372             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 373             Heps             = _mm256_mul_ps(vfeps,H);
 374             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 375             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 376             velec            = _mm256_mul_ps(qq12,VV);
 377             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 378             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velecsum         = _mm256_add_ps(velecsum,velec);
 382
 383             fscal            = felec;
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm256_mul_ps(fscal,dx12);
 387             ty               = _mm256_mul_ps(fscal,dy12);
 388             tz               = _mm256_mul_ps(fscal,dz12);
 389
 390             /* Update vectorial force */
 391             fix1             = _mm256_add_ps(fix1,tx);
 392             fiy1             = _mm256_add_ps(fiy1,ty);
 393             fiz1             = _mm256_add_ps(fiz1,tz);
 394
 395             fjx2             = _mm256_add_ps(fjx2,tx);
 396             fjy2             = _mm256_add_ps(fjy2,ty);
 397             fjz2             = _mm256_add_ps(fjz2,tz);
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             r13              = _mm256_mul_ps(rsq13,rinv13);
 404
 405             /* Calculate table index by multiplying r with table scale and truncate to integer */
 406             rt               = _mm256_mul_ps(r13,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(vfitab_lo,2);
 413             vfitab_hi        = _mm_slli_epi32(vfitab_hi,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(qq13,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(qq13,FF),_mm256_mul_ps(vftabscale,rinv13)));
 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,dx13);
 439             ty               = _mm256_mul_ps(fscal,dy13);
 440             tz               = _mm256_mul_ps(fscal,dz13);
 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             fjx3             = _mm256_add_ps(fjx3,tx);
 448             fjy3             = _mm256_add_ps(fjy3,ty);
 449             fjz3             = _mm256_add_ps(fjz3,tz);
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             r21              = _mm256_mul_ps(rsq21,rinv21);
 456
 457             /* Calculate table index by multiplying r with table scale and truncate to integer */
 458             rt               = _mm256_mul_ps(r21,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(vfitab_lo,2);
 465             vfitab_hi        = _mm_slli_epi32(vfitab_hi,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(qq21,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(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
 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,dx21);
 491             ty               = _mm256_mul_ps(fscal,dy21);
 492             tz               = _mm256_mul_ps(fscal,dz21);
 493
 494             /* Update vectorial force */
 495             fix2             = _mm256_add_ps(fix2,tx);
 496             fiy2             = _mm256_add_ps(fiy2,ty);
 497             fiz2             = _mm256_add_ps(fiz2,tz);
 498
 499             fjx1             = _mm256_add_ps(fjx1,tx);
 500             fjy1             = _mm256_add_ps(fjy1,ty);
 501             fjz1             = _mm256_add_ps(fjz1,tz);
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             r22              = _mm256_mul_ps(rsq22,rinv22);
 508
 509             /* Calculate table index by multiplying r with table scale and truncate to integer */
 510             rt               = _mm256_mul_ps(r22,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(vfitab_lo,2);
 517             vfitab_hi        = _mm_slli_epi32(vfitab_hi,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(qq22,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(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
 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,dx22);
 543             ty               = _mm256_mul_ps(fscal,dy22);
 544             tz               = _mm256_mul_ps(fscal,dz22);
 545
 546             /* Update vectorial force */
 547             fix2             = _mm256_add_ps(fix2,tx);
 548             fiy2             = _mm256_add_ps(fiy2,ty);
 549             fiz2             = _mm256_add_ps(fiz2,tz);
 550
 551             fjx2             = _mm256_add_ps(fjx2,tx);
 552             fjy2             = _mm256_add_ps(fjy2,ty);
 553             fjz2             = _mm256_add_ps(fjz2,tz);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             r23              = _mm256_mul_ps(rsq23,rinv23);
 560
 561             /* Calculate table index by multiplying r with table scale and truncate to integer */
 562             rt               = _mm256_mul_ps(r23,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(vfitab_lo,2);
 569             vfitab_hi        = _mm_slli_epi32(vfitab_hi,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(qq23,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(qq23,FF),_mm256_mul_ps(vftabscale,rinv23)));
 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,dx23);
 595             ty               = _mm256_mul_ps(fscal,dy23);
 596             tz               = _mm256_mul_ps(fscal,dz23);
 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             fjx3             = _mm256_add_ps(fjx3,tx);
 604             fjy3             = _mm256_add_ps(fjy3,ty);
 605             fjz3             = _mm256_add_ps(fjz3,tz);
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             r31              = _mm256_mul_ps(rsq31,rinv31);
 612
 613             /* Calculate table index by multiplying r with table scale and truncate to integer */
 614             rt               = _mm256_mul_ps(r31,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(vfitab_lo,2);
 621             vfitab_hi        = _mm_slli_epi32(vfitab_hi,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(qq31,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(qq31,FF),_mm256_mul_ps(vftabscale,rinv31)));
 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,dx31);
 647             ty               = _mm256_mul_ps(fscal,dy31);
 648             tz               = _mm256_mul_ps(fscal,dz31);
 649
 650             /* Update vectorial force */
 651             fix3             = _mm256_add_ps(fix3,tx);
 652             fiy3             = _mm256_add_ps(fiy3,ty);
 653             fiz3             = _mm256_add_ps(fiz3,tz);
 654
 655             fjx1             = _mm256_add_ps(fjx1,tx);
 656             fjy1             = _mm256_add_ps(fjy1,ty);
 657             fjz1             = _mm256_add_ps(fjz1,tz);
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             r32              = _mm256_mul_ps(rsq32,rinv32);
 664
 665             /* Calculate table index by multiplying r with table scale and truncate to integer */
 666             rt               = _mm256_mul_ps(r32,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(vfitab_lo,2);
 673             vfitab_hi        = _mm_slli_epi32(vfitab_hi,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(qq32,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(qq32,FF),_mm256_mul_ps(vftabscale,rinv32)));
 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,dx32);
 699             ty               = _mm256_mul_ps(fscal,dy32);
 700             tz               = _mm256_mul_ps(fscal,dz32);
 701
 702             /* Update vectorial force */
 703             fix3             = _mm256_add_ps(fix3,tx);
 704             fiy3             = _mm256_add_ps(fiy3,ty);
 705             fiz3             = _mm256_add_ps(fiz3,tz);
 706
 707             fjx2             = _mm256_add_ps(fjx2,tx);
 708             fjy2             = _mm256_add_ps(fjy2,ty);
 709             fjz2             = _mm256_add_ps(fjz2,tz);
 710
 711             /**************************
 712              * CALCULATE INTERACTIONS *
 713              **************************/
 714
 715             r33              = _mm256_mul_ps(rsq33,rinv33);
 716
 717             /* Calculate table index by multiplying r with table scale and truncate to integer */
 718             rt               = _mm256_mul_ps(r33,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(vfitab_lo,2);
 725             vfitab_hi        = _mm_slli_epi32(vfitab_hi,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(qq33,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(qq33,FF),_mm256_mul_ps(vftabscale,rinv33)));
 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,dx33);
 751             ty               = _mm256_mul_ps(fscal,dy33);
 752             tz               = _mm256_mul_ps(fscal,dz33);
 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             fjx3             = _mm256_add_ps(fjx3,tx);
 760             fjy3             = _mm256_add_ps(fjy3,ty);
 761             fjz3             = _mm256_add_ps(fjz3,tz);
 762
 763             fjptrA             = f+j_coord_offsetA;
 764             fjptrB             = f+j_coord_offsetB;
 765             fjptrC             = f+j_coord_offsetC;
 766             fjptrD             = f+j_coord_offsetD;
 767             fjptrE             = f+j_coord_offsetE;
 768             fjptrF             = f+j_coord_offsetF;
 769             fjptrG             = f+j_coord_offsetG;
 770             fjptrH             = f+j_coord_offsetH;
 771
 772             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 773                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
 774                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 775
 776             /* Inner loop uses 387 flops */
 777         }
 778
 779         if(jidx<j_index_end)
 780         {
 781
 782             /* Get j neighbor index, and coordinate index */
 783             jnrlistA         = jjnr[jidx];
 784             jnrlistB         = jjnr[jidx+1];
 785             jnrlistC         = jjnr[jidx+2];
 786             jnrlistD         = jjnr[jidx+3];
 787             jnrlistE         = jjnr[jidx+4];
 788             jnrlistF         = jjnr[jidx+5];
 789             jnrlistG         = jjnr[jidx+6];
 790             jnrlistH         = jjnr[jidx+7];
 791             /* Sign of each element will be negative for non-real atoms.
 792              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 793              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 794              */
 795             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 796                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 797
 798             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 799             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 800             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 801             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 802             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 803             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 804             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 805             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 806             j_coord_offsetA  = DIM*jnrA;
 807             j_coord_offsetB  = DIM*jnrB;
 808             j_coord_offsetC  = DIM*jnrC;
 809             j_coord_offsetD  = DIM*jnrD;
 810             j_coord_offsetE  = DIM*jnrE;
 811             j_coord_offsetF  = DIM*jnrF;
 812             j_coord_offsetG  = DIM*jnrG;
 813             j_coord_offsetH  = DIM*jnrH;
 814
 815             /* load j atom coordinates */
 816             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 817                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 818                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 819                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 820                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 821
 822             /* Calculate displacement vector */
 823             dx11             = _mm256_sub_ps(ix1,jx1);
 824             dy11             = _mm256_sub_ps(iy1,jy1);
 825             dz11             = _mm256_sub_ps(iz1,jz1);
 826             dx12             = _mm256_sub_ps(ix1,jx2);
 827             dy12             = _mm256_sub_ps(iy1,jy2);
 828             dz12             = _mm256_sub_ps(iz1,jz2);
 829             dx13             = _mm256_sub_ps(ix1,jx3);
 830             dy13             = _mm256_sub_ps(iy1,jy3);
 831             dz13             = _mm256_sub_ps(iz1,jz3);
 832             dx21             = _mm256_sub_ps(ix2,jx1);
 833             dy21             = _mm256_sub_ps(iy2,jy1);
 834             dz21             = _mm256_sub_ps(iz2,jz1);
 835             dx22             = _mm256_sub_ps(ix2,jx2);
 836             dy22             = _mm256_sub_ps(iy2,jy2);
 837             dz22             = _mm256_sub_ps(iz2,jz2);
 838             dx23             = _mm256_sub_ps(ix2,jx3);
 839             dy23             = _mm256_sub_ps(iy2,jy3);
 840             dz23             = _mm256_sub_ps(iz2,jz3);
 841             dx31             = _mm256_sub_ps(ix3,jx1);
 842             dy31             = _mm256_sub_ps(iy3,jy1);
 843             dz31             = _mm256_sub_ps(iz3,jz1);
 844             dx32             = _mm256_sub_ps(ix3,jx2);
 845             dy32             = _mm256_sub_ps(iy3,jy2);
 846             dz32             = _mm256_sub_ps(iz3,jz2);
 847             dx33             = _mm256_sub_ps(ix3,jx3);
 848             dy33             = _mm256_sub_ps(iy3,jy3);
 849             dz33             = _mm256_sub_ps(iz3,jz3);
 850
 851             /* Calculate squared distance and things based on it */
 852             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 853             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 854             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
 855             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 856             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 857             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
 858             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
 859             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
 860             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
 861
 862             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 863             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 864             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
 865             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 866             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 867             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
 868             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
 869             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
 870             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
 871
 872             fjx1             = _mm256_setzero_ps();
 873             fjy1             = _mm256_setzero_ps();
 874             fjz1             = _mm256_setzero_ps();
 875             fjx2             = _mm256_setzero_ps();
 876             fjy2             = _mm256_setzero_ps();
 877             fjz2             = _mm256_setzero_ps();
 878             fjx3             = _mm256_setzero_ps();
 879             fjy3             = _mm256_setzero_ps();
 880             fjz3             = _mm256_setzero_ps();
 881
 882             /**************************
 883              * CALCULATE INTERACTIONS *
 884              **************************/
 885
 886             r11              = _mm256_mul_ps(rsq11,rinv11);
 887             r11              = _mm256_andnot_ps(dummy_mask,r11);
 888
 889             /* Calculate table index by multiplying r with table scale and truncate to integer */
 890             rt               = _mm256_mul_ps(r11,vftabscale);
 891             vfitab           = _mm256_cvttps_epi32(rt);
 892             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 893             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 894             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 895             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 896             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 897             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 898
 899             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 900             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 901                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 902             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 903                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 904             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 905                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 906             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 907                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 908             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 909             Heps             = _mm256_mul_ps(vfeps,H);
 910             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 911             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 912             velec            = _mm256_mul_ps(qq11,VV);
 913             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 914             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
 915
 916             /* Update potential sum for this i atom from the interaction with this j atom. */
 917             velec            = _mm256_andnot_ps(dummy_mask,velec);
 918             velecsum         = _mm256_add_ps(velecsum,velec);
 919
 920             fscal            = felec;
 921
 922             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 923
 924             /* Calculate temporary vectorial force */
 925             tx               = _mm256_mul_ps(fscal,dx11);
 926             ty               = _mm256_mul_ps(fscal,dy11);
 927             tz               = _mm256_mul_ps(fscal,dz11);
 928
 929             /* Update vectorial force */
 930             fix1             = _mm256_add_ps(fix1,tx);
 931             fiy1             = _mm256_add_ps(fiy1,ty);
 932             fiz1             = _mm256_add_ps(fiz1,tz);
 933
 934             fjx1             = _mm256_add_ps(fjx1,tx);
 935             fjy1             = _mm256_add_ps(fjy1,ty);
 936             fjz1             = _mm256_add_ps(fjz1,tz);
 937
 938             /**************************
 939              * CALCULATE INTERACTIONS *
 940              **************************/
 941
 942             r12              = _mm256_mul_ps(rsq12,rinv12);
 943             r12              = _mm256_andnot_ps(dummy_mask,r12);
 944
 945             /* Calculate table index by multiplying r with table scale and truncate to integer */
 946             rt               = _mm256_mul_ps(r12,vftabscale);
 947             vfitab           = _mm256_cvttps_epi32(rt);
 948             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 949             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 950             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 951             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 952             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 953             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 954
 955             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 956             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 957                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 958             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 959                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 960             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 961                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 962             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 963                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 964             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 965             Heps             = _mm256_mul_ps(vfeps,H);
 966             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 967             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 968             velec            = _mm256_mul_ps(qq12,VV);
 969             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 970             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
 971
 972             /* Update potential sum for this i atom from the interaction with this j atom. */
 973             velec            = _mm256_andnot_ps(dummy_mask,velec);
 974             velecsum         = _mm256_add_ps(velecsum,velec);
 975
 976             fscal            = felec;
 977
 978             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm256_mul_ps(fscal,dx12);
 982             ty               = _mm256_mul_ps(fscal,dy12);
 983             tz               = _mm256_mul_ps(fscal,dz12);
 984
 985             /* Update vectorial force */
 986             fix1             = _mm256_add_ps(fix1,tx);
 987             fiy1             = _mm256_add_ps(fiy1,ty);
 988             fiz1             = _mm256_add_ps(fiz1,tz);
 989
 990             fjx2             = _mm256_add_ps(fjx2,tx);
 991             fjy2             = _mm256_add_ps(fjy2,ty);
 992             fjz2             = _mm256_add_ps(fjz2,tz);
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             r13              = _mm256_mul_ps(rsq13,rinv13);
 999             r13              = _mm256_andnot_ps(dummy_mask,r13);
1000
1001             /* Calculate table index by multiplying r with table scale and truncate to integer */
1002             rt               = _mm256_mul_ps(r13,vftabscale);
1003             vfitab           = _mm256_cvttps_epi32(rt);
1004             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1005             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1006             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1007             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1008             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1009             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1010
1011             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1012             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1013                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1014             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1015                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1016             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1017                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1018             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1019                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1020             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1021             Heps             = _mm256_mul_ps(vfeps,H);
1022             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1023             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1024             velec            = _mm256_mul_ps(qq13,VV);
1025             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1026             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq13,FF),_mm256_mul_ps(vftabscale,rinv13)));
1027
1028             /* Update potential sum for this i atom from the interaction with this j atom. */
1029             velec            = _mm256_andnot_ps(dummy_mask,velec);
1030             velecsum         = _mm256_add_ps(velecsum,velec);
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm256_mul_ps(fscal,dx13);
1038             ty               = _mm256_mul_ps(fscal,dy13);
1039             tz               = _mm256_mul_ps(fscal,dz13);
1040
1041             /* Update vectorial force */
1042             fix1             = _mm256_add_ps(fix1,tx);
1043             fiy1             = _mm256_add_ps(fiy1,ty);
1044             fiz1             = _mm256_add_ps(fiz1,tz);
1045
1046             fjx3             = _mm256_add_ps(fjx3,tx);
1047             fjy3             = _mm256_add_ps(fjy3,ty);
1048             fjz3             = _mm256_add_ps(fjz3,tz);
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             r21              = _mm256_mul_ps(rsq21,rinv21);
1055             r21              = _mm256_andnot_ps(dummy_mask,r21);
1056
1057             /* Calculate table index by multiplying r with table scale and truncate to integer */
1058             rt               = _mm256_mul_ps(r21,vftabscale);
1059             vfitab           = _mm256_cvttps_epi32(rt);
1060             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1061             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1062             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1063             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1064             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1065             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1066
1067             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1068             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1069                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1070             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1071                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1072             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1073                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1074             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1075                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1076             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1077             Heps             = _mm256_mul_ps(vfeps,H);
1078             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1079             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1080             velec            = _mm256_mul_ps(qq21,VV);
1081             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1082             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
1083
1084             /* Update potential sum for this i atom from the interaction with this j atom. */
1085             velec            = _mm256_andnot_ps(dummy_mask,velec);
1086             velecsum         = _mm256_add_ps(velecsum,velec);
1087
1088             fscal            = felec;
1089
1090             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm256_mul_ps(fscal,dx21);
1094             ty               = _mm256_mul_ps(fscal,dy21);
1095             tz               = _mm256_mul_ps(fscal,dz21);
1096
1097             /* Update vectorial force */
1098             fix2             = _mm256_add_ps(fix2,tx);
1099             fiy2             = _mm256_add_ps(fiy2,ty);
1100             fiz2             = _mm256_add_ps(fiz2,tz);
1101
1102             fjx1             = _mm256_add_ps(fjx1,tx);
1103             fjy1             = _mm256_add_ps(fjy1,ty);
1104             fjz1             = _mm256_add_ps(fjz1,tz);
1105
1106             /**************************
1107              * CALCULATE INTERACTIONS *
1108              **************************/
1109
1110             r22              = _mm256_mul_ps(rsq22,rinv22);
1111             r22              = _mm256_andnot_ps(dummy_mask,r22);
1112
1113             /* Calculate table index by multiplying r with table scale and truncate to integer */
1114             rt               = _mm256_mul_ps(r22,vftabscale);
1115             vfitab           = _mm256_cvttps_epi32(rt);
1116             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1117             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1118             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1119             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1120             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1121             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1122
1123             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1124             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1125                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1126             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1127                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1128             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1129                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1130             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1131                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1132             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1133             Heps             = _mm256_mul_ps(vfeps,H);
1134             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1135             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1136             velec            = _mm256_mul_ps(qq22,VV);
1137             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1138             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
1139
1140             /* Update potential sum for this i atom from the interaction with this j atom. */
1141             velec            = _mm256_andnot_ps(dummy_mask,velec);
1142             velecsum         = _mm256_add_ps(velecsum,velec);
1143
1144             fscal            = felec;
1145
1146             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1147
1148             /* Calculate temporary vectorial force */
1149             tx               = _mm256_mul_ps(fscal,dx22);
1150             ty               = _mm256_mul_ps(fscal,dy22);
1151             tz               = _mm256_mul_ps(fscal,dz22);
1152
1153             /* Update vectorial force */
1154             fix2             = _mm256_add_ps(fix2,tx);
1155             fiy2             = _mm256_add_ps(fiy2,ty);
1156             fiz2             = _mm256_add_ps(fiz2,tz);
1157
1158             fjx2             = _mm256_add_ps(fjx2,tx);
1159             fjy2             = _mm256_add_ps(fjy2,ty);
1160             fjz2             = _mm256_add_ps(fjz2,tz);
1161
1162             /**************************
1163              * CALCULATE INTERACTIONS *
1164              **************************/
1165
1166             r23              = _mm256_mul_ps(rsq23,rinv23);
1167             r23              = _mm256_andnot_ps(dummy_mask,r23);
1168
1169             /* Calculate table index by multiplying r with table scale and truncate to integer */
1170             rt               = _mm256_mul_ps(r23,vftabscale);
1171             vfitab           = _mm256_cvttps_epi32(rt);
1172             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1173             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1174             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1175             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1176             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1177             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1178
1179             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1180             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1181                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1182             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1183                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1184             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1185                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1186             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1187                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1188             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1189             Heps             = _mm256_mul_ps(vfeps,H);
1190             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1191             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1192             velec            = _mm256_mul_ps(qq23,VV);
1193             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1194             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq23,FF),_mm256_mul_ps(vftabscale,rinv23)));
1195
1196             /* Update potential sum for this i atom from the interaction with this j atom. */
1197             velec            = _mm256_andnot_ps(dummy_mask,velec);
1198             velecsum         = _mm256_add_ps(velecsum,velec);
1199
1200             fscal            = felec;
1201
1202             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm256_mul_ps(fscal,dx23);
1206             ty               = _mm256_mul_ps(fscal,dy23);
1207             tz               = _mm256_mul_ps(fscal,dz23);
1208
1209             /* Update vectorial force */
1210             fix2             = _mm256_add_ps(fix2,tx);
1211             fiy2             = _mm256_add_ps(fiy2,ty);
1212             fiz2             = _mm256_add_ps(fiz2,tz);
1213
1214             fjx3             = _mm256_add_ps(fjx3,tx);
1215             fjy3             = _mm256_add_ps(fjy3,ty);
1216             fjz3             = _mm256_add_ps(fjz3,tz);
1217
1218             /**************************
1219              * CALCULATE INTERACTIONS *
1220              **************************/
1221
1222             r31              = _mm256_mul_ps(rsq31,rinv31);
1223             r31              = _mm256_andnot_ps(dummy_mask,r31);
1224
1225             /* Calculate table index by multiplying r with table scale and truncate to integer */
1226             rt               = _mm256_mul_ps(r31,vftabscale);
1227             vfitab           = _mm256_cvttps_epi32(rt);
1228             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1229             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1230             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1231             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1232             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1233             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1234
1235             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1236             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1237                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1238             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1239                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1240             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1241                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1242             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1243                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1244             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1245             Heps             = _mm256_mul_ps(vfeps,H);
1246             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1247             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1248             velec            = _mm256_mul_ps(qq31,VV);
1249             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1250             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq31,FF),_mm256_mul_ps(vftabscale,rinv31)));
1251
1252             /* Update potential sum for this i atom from the interaction with this j atom. */
1253             velec            = _mm256_andnot_ps(dummy_mask,velec);
1254             velecsum         = _mm256_add_ps(velecsum,velec);
1255
1256             fscal            = felec;
1257
1258             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm256_mul_ps(fscal,dx31);
1262             ty               = _mm256_mul_ps(fscal,dy31);
1263             tz               = _mm256_mul_ps(fscal,dz31);
1264
1265             /* Update vectorial force */
1266             fix3             = _mm256_add_ps(fix3,tx);
1267             fiy3             = _mm256_add_ps(fiy3,ty);
1268             fiz3             = _mm256_add_ps(fiz3,tz);
1269
1270             fjx1             = _mm256_add_ps(fjx1,tx);
1271             fjy1             = _mm256_add_ps(fjy1,ty);
1272             fjz1             = _mm256_add_ps(fjz1,tz);
1273
1274             /**************************
1275              * CALCULATE INTERACTIONS *
1276              **************************/
1277
1278             r32              = _mm256_mul_ps(rsq32,rinv32);
1279             r32              = _mm256_andnot_ps(dummy_mask,r32);
1280
1281             /* Calculate table index by multiplying r with table scale and truncate to integer */
1282             rt               = _mm256_mul_ps(r32,vftabscale);
1283             vfitab           = _mm256_cvttps_epi32(rt);
1284             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1285             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1286             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1287             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1288             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1289             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1290
1291             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1292             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1294             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1296             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1297                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1298             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1299                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1300             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1301             Heps             = _mm256_mul_ps(vfeps,H);
1302             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1303             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1304             velec            = _mm256_mul_ps(qq32,VV);
1305             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1306             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq32,FF),_mm256_mul_ps(vftabscale,rinv32)));
1307
1308             /* Update potential sum for this i atom from the interaction with this j atom. */
1309             velec            = _mm256_andnot_ps(dummy_mask,velec);
1310             velecsum         = _mm256_add_ps(velecsum,velec);
1311
1312             fscal            = felec;
1313
1314             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1315
1316             /* Calculate temporary vectorial force */
1317             tx               = _mm256_mul_ps(fscal,dx32);
1318             ty               = _mm256_mul_ps(fscal,dy32);
1319             tz               = _mm256_mul_ps(fscal,dz32);
1320
1321             /* Update vectorial force */
1322             fix3             = _mm256_add_ps(fix3,tx);
1323             fiy3             = _mm256_add_ps(fiy3,ty);
1324             fiz3             = _mm256_add_ps(fiz3,tz);
1325
1326             fjx2             = _mm256_add_ps(fjx2,tx);
1327             fjy2             = _mm256_add_ps(fjy2,ty);
1328             fjz2             = _mm256_add_ps(fjz2,tz);
1329
1330             /**************************
1331              * CALCULATE INTERACTIONS *
1332              **************************/
1333
1334             r33              = _mm256_mul_ps(rsq33,rinv33);
1335             r33              = _mm256_andnot_ps(dummy_mask,r33);
1336
1337             /* Calculate table index by multiplying r with table scale and truncate to integer */
1338             rt               = _mm256_mul_ps(r33,vftabscale);
1339             vfitab           = _mm256_cvttps_epi32(rt);
1340             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1341             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1342             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1343             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1344             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1345             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1346
1347             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1348             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1349                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1350             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1351                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1352             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1353                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1354             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1355                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1356             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1357             Heps             = _mm256_mul_ps(vfeps,H);
1358             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1359             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1360             velec            = _mm256_mul_ps(qq33,VV);
1361             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1362             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq33,FF),_mm256_mul_ps(vftabscale,rinv33)));
1363
1364             /* Update potential sum for this i atom from the interaction with this j atom. */
1365             velec            = _mm256_andnot_ps(dummy_mask,velec);
1366             velecsum         = _mm256_add_ps(velecsum,velec);
1367
1368             fscal            = felec;
1369
1370             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1371
1372             /* Calculate temporary vectorial force */
1373             tx               = _mm256_mul_ps(fscal,dx33);
1374             ty               = _mm256_mul_ps(fscal,dy33);
1375             tz               = _mm256_mul_ps(fscal,dz33);
1376
1377             /* Update vectorial force */
1378             fix3             = _mm256_add_ps(fix3,tx);
1379             fiy3             = _mm256_add_ps(fiy3,ty);
1380             fiz3             = _mm256_add_ps(fiz3,tz);
1381
1382             fjx3             = _mm256_add_ps(fjx3,tx);
1383             fjy3             = _mm256_add_ps(fjy3,ty);
1384             fjz3             = _mm256_add_ps(fjz3,tz);
1385
1386             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1387             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1388             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1389             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1390             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1391             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1392             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1393             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1394
1395             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1396                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1397                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1398
1399             /* Inner loop uses 396 flops */
1400         }
1401
1402         /* End of innermost loop */
1403
1404         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1405                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1406
1407         ggid                        = gid[iidx];
1408         /* Update potential energies */
1409         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1410
1411         /* Increment number of inner iterations */
1412         inneriter                  += j_index_end - j_index_start;
1413
1414         /* Outer loop uses 19 flops */
1415     }
1416
1417     /* Increment number of outer iterations */
1418     outeriter        += nri;
1419
1420     /* Update outer/inner flops */
1421
1422     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_VF,outeriter*19 + inneriter*396);
1423 }
1424 /*
1425  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_avx_256_single
1426  * Electrostatics interaction: CubicSplineTable
1427  * VdW interaction:            None
1428  * Geometry:                   Water4-Water4
1429  * Calculate force/pot:        Force
1430  */
1431 void
1432 nb_kernel_ElecCSTab_VdwNone_GeomW4W4_F_avx_256_single
1433                     (t_nblist                    * gmx_restrict       nlist,
1434                      rvec                        * gmx_restrict          xx,
1435                      rvec                        * gmx_restrict          ff,
1436                      t_forcerec                  * gmx_restrict          fr,
1437                      t_mdatoms                   * gmx_restrict     mdatoms,
1438                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1439                      t_nrnb                      * gmx_restrict        nrnb)
1440 {
1441     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1442      * just 0 for non-waters.
1443      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1444      * jnr indices corresponding to data put in the four positions in the SIMD register.
1445      */
1446     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1447     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1448     int              jnrA,jnrB,jnrC,jnrD;
1449     int              jnrE,jnrF,jnrG,jnrH;
1450     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1451     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1452     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1453     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1454     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1455     real             rcutoff_scalar;
1456     real             *shiftvec,*fshift,*x,*f;
1457     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1458     real             scratch[4*DIM];
1459     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1460     real *           vdwioffsetptr1;
1461     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1462     real *           vdwioffsetptr2;
1463     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1464     real *           vdwioffsetptr3;
1465     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1466     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
1467     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1468     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
1469     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1470     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D,vdwjidx3E,vdwjidx3F,vdwjidx3G,vdwjidx3H;
1471     __m256           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1472     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1473     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1474     __m256           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1475     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1476     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1477     __m256           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1478     __m256           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1479     __m256           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1480     __m256           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1481     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1482     real             *charge;
1483     __m256i          vfitab;
1484     __m128i          vfitab_lo,vfitab_hi;
1485     __m128i          ifour       = _mm_set1_epi32(4);
1486     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1487     real             *vftab;
1488     __m256           dummy_mask,cutoff_mask;
1489     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1490     __m256           one     = _mm256_set1_ps(1.0);
1491     __m256           two     = _mm256_set1_ps(2.0);
1492     x                = xx[0];
1493     f                = ff[0];
1494
1495     nri              = nlist->nri;
1496     iinr             = nlist->iinr;
1497     jindex           = nlist->jindex;
1498     jjnr             = nlist->jjnr;
1499     shiftidx         = nlist->shift;
1500     gid              = nlist->gid;
1501     shiftvec         = fr->shift_vec[0];
1502     fshift           = fr->fshift[0];
1503     facel            = _mm256_set1_ps(fr->epsfac);
1504     charge           = mdatoms->chargeA;
1505
1506     vftab            = kernel_data->table_elec->data;
1507     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
1508
1509     /* Setup water-specific parameters */
1510     inr              = nlist->iinr[0];
1511     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1512     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1513     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1514
1515     jq1              = _mm256_set1_ps(charge[inr+1]);
1516     jq2              = _mm256_set1_ps(charge[inr+2]);
1517     jq3              = _mm256_set1_ps(charge[inr+3]);
1518     qq11             = _mm256_mul_ps(iq1,jq1);
1519     qq12             = _mm256_mul_ps(iq1,jq2);
1520     qq13             = _mm256_mul_ps(iq1,jq3);
1521     qq21             = _mm256_mul_ps(iq2,jq1);
1522     qq22             = _mm256_mul_ps(iq2,jq2);
1523     qq23             = _mm256_mul_ps(iq2,jq3);
1524     qq31             = _mm256_mul_ps(iq3,jq1);
1525     qq32             = _mm256_mul_ps(iq3,jq2);
1526     qq33             = _mm256_mul_ps(iq3,jq3);
1527
1528     /* Avoid stupid compiler warnings */
1529     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1530     j_coord_offsetA = 0;
1531     j_coord_offsetB = 0;
1532     j_coord_offsetC = 0;
1533     j_coord_offsetD = 0;
1534     j_coord_offsetE = 0;
1535     j_coord_offsetF = 0;
1536     j_coord_offsetG = 0;
1537     j_coord_offsetH = 0;
1538
1539     outeriter        = 0;
1540     inneriter        = 0;
1541
1542     for(iidx=0;iidx<4*DIM;iidx++)
1543     {
1544         scratch[iidx] = 0.0;
1545     }
1546
1547     /* Start outer loop over neighborlists */
1548     for(iidx=0; iidx<nri; iidx++)
1549     {
1550         /* Load shift vector for this list */
1551         i_shift_offset   = DIM*shiftidx[iidx];
1552
1553         /* Load limits for loop over neighbors */
1554         j_index_start    = jindex[iidx];
1555         j_index_end      = jindex[iidx+1];
1556
1557         /* Get outer coordinate index */
1558         inr              = iinr[iidx];
1559         i_coord_offset   = DIM*inr;
1560
1561         /* Load i particle coords and add shift vector */
1562         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
1563                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1564
1565         fix1             = _mm256_setzero_ps();
1566         fiy1             = _mm256_setzero_ps();
1567         fiz1             = _mm256_setzero_ps();
1568         fix2             = _mm256_setzero_ps();
1569         fiy2             = _mm256_setzero_ps();
1570         fiz2             = _mm256_setzero_ps();
1571         fix3             = _mm256_setzero_ps();
1572         fiy3             = _mm256_setzero_ps();
1573         fiz3             = _mm256_setzero_ps();
1574
1575         /* Start inner kernel loop */
1576         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1577         {
1578
1579             /* Get j neighbor index, and coordinate index */
1580             jnrA             = jjnr[jidx];
1581             jnrB             = jjnr[jidx+1];
1582             jnrC             = jjnr[jidx+2];
1583             jnrD             = jjnr[jidx+3];
1584             jnrE             = jjnr[jidx+4];
1585             jnrF             = jjnr[jidx+5];
1586             jnrG             = jjnr[jidx+6];
1587             jnrH             = jjnr[jidx+7];
1588             j_coord_offsetA  = DIM*jnrA;
1589             j_coord_offsetB  = DIM*jnrB;
1590             j_coord_offsetC  = DIM*jnrC;
1591             j_coord_offsetD  = DIM*jnrD;
1592             j_coord_offsetE  = DIM*jnrE;
1593             j_coord_offsetF  = DIM*jnrF;
1594             j_coord_offsetG  = DIM*jnrG;
1595             j_coord_offsetH  = DIM*jnrH;
1596
1597             /* load j atom coordinates */
1598             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
1599                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
1600                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
1601                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
1602                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
1603
1604             /* Calculate displacement vector */
1605             dx11             = _mm256_sub_ps(ix1,jx1);
1606             dy11             = _mm256_sub_ps(iy1,jy1);
1607             dz11             = _mm256_sub_ps(iz1,jz1);
1608             dx12             = _mm256_sub_ps(ix1,jx2);
1609             dy12             = _mm256_sub_ps(iy1,jy2);
1610             dz12             = _mm256_sub_ps(iz1,jz2);
1611             dx13             = _mm256_sub_ps(ix1,jx3);
1612             dy13             = _mm256_sub_ps(iy1,jy3);
1613             dz13             = _mm256_sub_ps(iz1,jz3);
1614             dx21             = _mm256_sub_ps(ix2,jx1);
1615             dy21             = _mm256_sub_ps(iy2,jy1);
1616             dz21             = _mm256_sub_ps(iz2,jz1);
1617             dx22             = _mm256_sub_ps(ix2,jx2);
1618             dy22             = _mm256_sub_ps(iy2,jy2);
1619             dz22             = _mm256_sub_ps(iz2,jz2);
1620             dx23             = _mm256_sub_ps(ix2,jx3);
1621             dy23             = _mm256_sub_ps(iy2,jy3);
1622             dz23             = _mm256_sub_ps(iz2,jz3);
1623             dx31             = _mm256_sub_ps(ix3,jx1);
1624             dy31             = _mm256_sub_ps(iy3,jy1);
1625             dz31             = _mm256_sub_ps(iz3,jz1);
1626             dx32             = _mm256_sub_ps(ix3,jx2);
1627             dy32             = _mm256_sub_ps(iy3,jy2);
1628             dz32             = _mm256_sub_ps(iz3,jz2);
1629             dx33             = _mm256_sub_ps(ix3,jx3);
1630             dy33             = _mm256_sub_ps(iy3,jy3);
1631             dz33             = _mm256_sub_ps(iz3,jz3);
1632
1633             /* Calculate squared distance and things based on it */
1634             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1635             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1636             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
1637             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1638             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1639             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
1640             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
1641             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
1642             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
1643
1644             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1645             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1646             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
1647             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1648             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1649             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
1650             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
1651             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
1652             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
1653
1654             fjx1             = _mm256_setzero_ps();
1655             fjy1             = _mm256_setzero_ps();
1656             fjz1             = _mm256_setzero_ps();
1657             fjx2             = _mm256_setzero_ps();
1658             fjy2             = _mm256_setzero_ps();
1659             fjz2             = _mm256_setzero_ps();
1660             fjx3             = _mm256_setzero_ps();
1661             fjy3             = _mm256_setzero_ps();
1662             fjz3             = _mm256_setzero_ps();
1663
1664             /**************************
1665              * CALCULATE INTERACTIONS *
1666              **************************/
1667
1668             r11              = _mm256_mul_ps(rsq11,rinv11);
1669
1670             /* Calculate table index by multiplying r with table scale and truncate to integer */
1671             rt               = _mm256_mul_ps(r11,vftabscale);
1672             vfitab           = _mm256_cvttps_epi32(rt);
1673             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1674             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1675             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1676             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1677             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1678             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1679
1680             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1681             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1682                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1683             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1684                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1685             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1686                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1687             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1688                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1689             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1690             Heps             = _mm256_mul_ps(vfeps,H);
1691             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1692             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1693             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
1694
1695             fscal            = felec;
1696
1697             /* Calculate temporary vectorial force */
1698             tx               = _mm256_mul_ps(fscal,dx11);
1699             ty               = _mm256_mul_ps(fscal,dy11);
1700             tz               = _mm256_mul_ps(fscal,dz11);
1701
1702             /* Update vectorial force */
1703             fix1             = _mm256_add_ps(fix1,tx);
1704             fiy1             = _mm256_add_ps(fiy1,ty);
1705             fiz1             = _mm256_add_ps(fiz1,tz);
1706
1707             fjx1             = _mm256_add_ps(fjx1,tx);
1708             fjy1             = _mm256_add_ps(fjy1,ty);
1709             fjz1             = _mm256_add_ps(fjz1,tz);
1710
1711             /**************************
1712              * CALCULATE INTERACTIONS *
1713              **************************/
1714
1715             r12              = _mm256_mul_ps(rsq12,rinv12);
1716
1717             /* Calculate table index by multiplying r with table scale and truncate to integer */
1718             rt               = _mm256_mul_ps(r12,vftabscale);
1719             vfitab           = _mm256_cvttps_epi32(rt);
1720             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1721             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1722             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1723             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1724             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1725             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1726
1727             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1728             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1729                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1730             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1731                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1732             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1733                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1734             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1735                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1736             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1737             Heps             = _mm256_mul_ps(vfeps,H);
1738             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1739             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1740             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
1741
1742             fscal            = felec;
1743
1744             /* Calculate temporary vectorial force */
1745             tx               = _mm256_mul_ps(fscal,dx12);
1746             ty               = _mm256_mul_ps(fscal,dy12);
1747             tz               = _mm256_mul_ps(fscal,dz12);
1748
1749             /* Update vectorial force */
1750             fix1             = _mm256_add_ps(fix1,tx);
1751             fiy1             = _mm256_add_ps(fiy1,ty);
1752             fiz1             = _mm256_add_ps(fiz1,tz);
1753
1754             fjx2             = _mm256_add_ps(fjx2,tx);
1755             fjy2             = _mm256_add_ps(fjy2,ty);
1756             fjz2             = _mm256_add_ps(fjz2,tz);
1757
1758             /**************************
1759              * CALCULATE INTERACTIONS *
1760              **************************/
1761
1762             r13              = _mm256_mul_ps(rsq13,rinv13);
1763
1764             /* Calculate table index by multiplying r with table scale and truncate to integer */
1765             rt               = _mm256_mul_ps(r13,vftabscale);
1766             vfitab           = _mm256_cvttps_epi32(rt);
1767             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1768             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1769             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1770             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1771             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1772             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1773
1774             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1775             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1776                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1777             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1778                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1779             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1780                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1781             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1782                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1783             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1784             Heps             = _mm256_mul_ps(vfeps,H);
1785             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1786             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1787             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq13,FF),_mm256_mul_ps(vftabscale,rinv13)));
1788
1789             fscal            = felec;
1790
1791             /* Calculate temporary vectorial force */
1792             tx               = _mm256_mul_ps(fscal,dx13);
1793             ty               = _mm256_mul_ps(fscal,dy13);
1794             tz               = _mm256_mul_ps(fscal,dz13);
1795
1796             /* Update vectorial force */
1797             fix1             = _mm256_add_ps(fix1,tx);
1798             fiy1             = _mm256_add_ps(fiy1,ty);
1799             fiz1             = _mm256_add_ps(fiz1,tz);
1800
1801             fjx3             = _mm256_add_ps(fjx3,tx);
1802             fjy3             = _mm256_add_ps(fjy3,ty);
1803             fjz3             = _mm256_add_ps(fjz3,tz);
1804
1805             /**************************
1806              * CALCULATE INTERACTIONS *
1807              **************************/
1808
1809             r21              = _mm256_mul_ps(rsq21,rinv21);
1810
1811             /* Calculate table index by multiplying r with table scale and truncate to integer */
1812             rt               = _mm256_mul_ps(r21,vftabscale);
1813             vfitab           = _mm256_cvttps_epi32(rt);
1814             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1815             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1816             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1817             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1818             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1819             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1820
1821             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1822             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1823                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1824             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1825                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1826             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1827                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1828             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1829                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1830             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1831             Heps             = _mm256_mul_ps(vfeps,H);
1832             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1833             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1834             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
1835
1836             fscal            = felec;
1837
1838             /* Calculate temporary vectorial force */
1839             tx               = _mm256_mul_ps(fscal,dx21);
1840             ty               = _mm256_mul_ps(fscal,dy21);
1841             tz               = _mm256_mul_ps(fscal,dz21);
1842
1843             /* Update vectorial force */
1844             fix2             = _mm256_add_ps(fix2,tx);
1845             fiy2             = _mm256_add_ps(fiy2,ty);
1846             fiz2             = _mm256_add_ps(fiz2,tz);
1847
1848             fjx1             = _mm256_add_ps(fjx1,tx);
1849             fjy1             = _mm256_add_ps(fjy1,ty);
1850             fjz1             = _mm256_add_ps(fjz1,tz);
1851
1852             /**************************
1853              * CALCULATE INTERACTIONS *
1854              **************************/
1855
1856             r22              = _mm256_mul_ps(rsq22,rinv22);
1857
1858             /* Calculate table index by multiplying r with table scale and truncate to integer */
1859             rt               = _mm256_mul_ps(r22,vftabscale);
1860             vfitab           = _mm256_cvttps_epi32(rt);
1861             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1862             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1863             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1864             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1865             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1866             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1867
1868             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1869             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1870                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1871             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1872                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1873             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1874                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1875             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1876                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1877             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1878             Heps             = _mm256_mul_ps(vfeps,H);
1879             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1880             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1881             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
1882
1883             fscal            = felec;
1884
1885             /* Calculate temporary vectorial force */
1886             tx               = _mm256_mul_ps(fscal,dx22);
1887             ty               = _mm256_mul_ps(fscal,dy22);
1888             tz               = _mm256_mul_ps(fscal,dz22);
1889
1890             /* Update vectorial force */
1891             fix2             = _mm256_add_ps(fix2,tx);
1892             fiy2             = _mm256_add_ps(fiy2,ty);
1893             fiz2             = _mm256_add_ps(fiz2,tz);
1894
1895             fjx2             = _mm256_add_ps(fjx2,tx);
1896             fjy2             = _mm256_add_ps(fjy2,ty);
1897             fjz2             = _mm256_add_ps(fjz2,tz);
1898
1899             /**************************
1900              * CALCULATE INTERACTIONS *
1901              **************************/
1902
1903             r23              = _mm256_mul_ps(rsq23,rinv23);
1904
1905             /* Calculate table index by multiplying r with table scale and truncate to integer */
1906             rt               = _mm256_mul_ps(r23,vftabscale);
1907             vfitab           = _mm256_cvttps_epi32(rt);
1908             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1909             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1910             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1911             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1912             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1913             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1914
1915             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1916             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1917                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1918             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1919                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1920             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1921                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1922             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1923                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1924             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1925             Heps             = _mm256_mul_ps(vfeps,H);
1926             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1927             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1928             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq23,FF),_mm256_mul_ps(vftabscale,rinv23)));
1929
1930             fscal            = felec;
1931
1932             /* Calculate temporary vectorial force */
1933             tx               = _mm256_mul_ps(fscal,dx23);
1934             ty               = _mm256_mul_ps(fscal,dy23);
1935             tz               = _mm256_mul_ps(fscal,dz23);
1936
1937             /* Update vectorial force */
1938             fix2             = _mm256_add_ps(fix2,tx);
1939             fiy2             = _mm256_add_ps(fiy2,ty);
1940             fiz2             = _mm256_add_ps(fiz2,tz);
1941
1942             fjx3             = _mm256_add_ps(fjx3,tx);
1943             fjy3             = _mm256_add_ps(fjy3,ty);
1944             fjz3             = _mm256_add_ps(fjz3,tz);
1945
1946             /**************************
1947              * CALCULATE INTERACTIONS *
1948              **************************/
1949
1950             r31              = _mm256_mul_ps(rsq31,rinv31);
1951
1952             /* Calculate table index by multiplying r with table scale and truncate to integer */
1953             rt               = _mm256_mul_ps(r31,vftabscale);
1954             vfitab           = _mm256_cvttps_epi32(rt);
1955             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1956             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1957             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1958             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1959             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1960             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1961
1962             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1963             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1964                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1965             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1966                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1967             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1968                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1969             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1970                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1971             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1972             Heps             = _mm256_mul_ps(vfeps,H);
1973             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1974             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1975             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq31,FF),_mm256_mul_ps(vftabscale,rinv31)));
1976
1977             fscal            = felec;
1978
1979             /* Calculate temporary vectorial force */
1980             tx               = _mm256_mul_ps(fscal,dx31);
1981             ty               = _mm256_mul_ps(fscal,dy31);
1982             tz               = _mm256_mul_ps(fscal,dz31);
1983
1984             /* Update vectorial force */
1985             fix3             = _mm256_add_ps(fix3,tx);
1986             fiy3             = _mm256_add_ps(fiy3,ty);
1987             fiz3             = _mm256_add_ps(fiz3,tz);
1988
1989             fjx1             = _mm256_add_ps(fjx1,tx);
1990             fjy1             = _mm256_add_ps(fjy1,ty);
1991             fjz1             = _mm256_add_ps(fjz1,tz);
1992
1993             /**************************
1994              * CALCULATE INTERACTIONS *
1995              **************************/
1996
1997             r32              = _mm256_mul_ps(rsq32,rinv32);
1998
1999             /* Calculate table index by multiplying r with table scale and truncate to integer */
2000             rt               = _mm256_mul_ps(r32,vftabscale);
2001             vfitab           = _mm256_cvttps_epi32(rt);
2002             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2003             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2004             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2005             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2006             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2007             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2008
2009             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2010             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2011                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2012             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2013                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2014             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2015                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2016             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2017                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2018             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2019             Heps             = _mm256_mul_ps(vfeps,H);
2020             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2021             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2022             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq32,FF),_mm256_mul_ps(vftabscale,rinv32)));
2023
2024             fscal            = felec;
2025
2026             /* Calculate temporary vectorial force */
2027             tx               = _mm256_mul_ps(fscal,dx32);
2028             ty               = _mm256_mul_ps(fscal,dy32);
2029             tz               = _mm256_mul_ps(fscal,dz32);
2030
2031             /* Update vectorial force */
2032             fix3             = _mm256_add_ps(fix3,tx);
2033             fiy3             = _mm256_add_ps(fiy3,ty);
2034             fiz3             = _mm256_add_ps(fiz3,tz);
2035
2036             fjx2             = _mm256_add_ps(fjx2,tx);
2037             fjy2             = _mm256_add_ps(fjy2,ty);
2038             fjz2             = _mm256_add_ps(fjz2,tz);
2039
2040             /**************************
2041              * CALCULATE INTERACTIONS *
2042              **************************/
2043
2044             r33              = _mm256_mul_ps(rsq33,rinv33);
2045
2046             /* Calculate table index by multiplying r with table scale and truncate to integer */
2047             rt               = _mm256_mul_ps(r33,vftabscale);
2048             vfitab           = _mm256_cvttps_epi32(rt);
2049             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2050             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2051             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2052             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2053             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2054             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2055
2056             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2057             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2058                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2059             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2060                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2061             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2062                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2063             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2064                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2065             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2066             Heps             = _mm256_mul_ps(vfeps,H);
2067             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2068             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2069             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq33,FF),_mm256_mul_ps(vftabscale,rinv33)));
2070
2071             fscal            = felec;
2072
2073             /* Calculate temporary vectorial force */
2074             tx               = _mm256_mul_ps(fscal,dx33);
2075             ty               = _mm256_mul_ps(fscal,dy33);
2076             tz               = _mm256_mul_ps(fscal,dz33);
2077
2078             /* Update vectorial force */
2079             fix3             = _mm256_add_ps(fix3,tx);
2080             fiy3             = _mm256_add_ps(fiy3,ty);
2081             fiz3             = _mm256_add_ps(fiz3,tz);
2082
2083             fjx3             = _mm256_add_ps(fjx3,tx);
2084             fjy3             = _mm256_add_ps(fjy3,ty);
2085             fjz3             = _mm256_add_ps(fjz3,tz);
2086
2087             fjptrA             = f+j_coord_offsetA;
2088             fjptrB             = f+j_coord_offsetB;
2089             fjptrC             = f+j_coord_offsetC;
2090             fjptrD             = f+j_coord_offsetD;
2091             fjptrE             = f+j_coord_offsetE;
2092             fjptrF             = f+j_coord_offsetF;
2093             fjptrG             = f+j_coord_offsetG;
2094             fjptrH             = f+j_coord_offsetH;
2095
2096             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2097                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
2098                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2099
2100             /* Inner loop uses 351 flops */
2101         }
2102
2103         if(jidx<j_index_end)
2104         {
2105
2106             /* Get j neighbor index, and coordinate index */
2107             jnrlistA         = jjnr[jidx];
2108             jnrlistB         = jjnr[jidx+1];
2109             jnrlistC         = jjnr[jidx+2];
2110             jnrlistD         = jjnr[jidx+3];
2111             jnrlistE         = jjnr[jidx+4];
2112             jnrlistF         = jjnr[jidx+5];
2113             jnrlistG         = jjnr[jidx+6];
2114             jnrlistH         = jjnr[jidx+7];
2115             /* Sign of each element will be negative for non-real atoms.
2116              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2117              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2118              */
2119             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2120                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2121
2122             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2123             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2124             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2125             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2126             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2127             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2128             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2129             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2130             j_coord_offsetA  = DIM*jnrA;
2131             j_coord_offsetB  = DIM*jnrB;
2132             j_coord_offsetC  = DIM*jnrC;
2133             j_coord_offsetD  = DIM*jnrD;
2134             j_coord_offsetE  = DIM*jnrE;
2135             j_coord_offsetF  = DIM*jnrF;
2136             j_coord_offsetG  = DIM*jnrG;
2137             j_coord_offsetH  = DIM*jnrH;
2138
2139             /* load j atom coordinates */
2140             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
2141                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
2142                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
2143                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
2144                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
2145
2146             /* Calculate displacement vector */
2147             dx11             = _mm256_sub_ps(ix1,jx1);
2148             dy11             = _mm256_sub_ps(iy1,jy1);
2149             dz11             = _mm256_sub_ps(iz1,jz1);
2150             dx12             = _mm256_sub_ps(ix1,jx2);
2151             dy12             = _mm256_sub_ps(iy1,jy2);
2152             dz12             = _mm256_sub_ps(iz1,jz2);
2153             dx13             = _mm256_sub_ps(ix1,jx3);
2154             dy13             = _mm256_sub_ps(iy1,jy3);
2155             dz13             = _mm256_sub_ps(iz1,jz3);
2156             dx21             = _mm256_sub_ps(ix2,jx1);
2157             dy21             = _mm256_sub_ps(iy2,jy1);
2158             dz21             = _mm256_sub_ps(iz2,jz1);
2159             dx22             = _mm256_sub_ps(ix2,jx2);
2160             dy22             = _mm256_sub_ps(iy2,jy2);
2161             dz22             = _mm256_sub_ps(iz2,jz2);
2162             dx23             = _mm256_sub_ps(ix2,jx3);
2163             dy23             = _mm256_sub_ps(iy2,jy3);
2164             dz23             = _mm256_sub_ps(iz2,jz3);
2165             dx31             = _mm256_sub_ps(ix3,jx1);
2166             dy31             = _mm256_sub_ps(iy3,jy1);
2167             dz31             = _mm256_sub_ps(iz3,jz1);
2168             dx32             = _mm256_sub_ps(ix3,jx2);
2169             dy32             = _mm256_sub_ps(iy3,jy2);
2170             dz32             = _mm256_sub_ps(iz3,jz2);
2171             dx33             = _mm256_sub_ps(ix3,jx3);
2172             dy33             = _mm256_sub_ps(iy3,jy3);
2173             dz33             = _mm256_sub_ps(iz3,jz3);
2174
2175             /* Calculate squared distance and things based on it */
2176             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2177             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2178             rsq13            = gmx_mm256_calc_rsq_ps(dx13,dy13,dz13);
2179             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2180             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2181             rsq23            = gmx_mm256_calc_rsq_ps(dx23,dy23,dz23);
2182             rsq31            = gmx_mm256_calc_rsq_ps(dx31,dy31,dz31);
2183             rsq32            = gmx_mm256_calc_rsq_ps(dx32,dy32,dz32);
2184             rsq33            = gmx_mm256_calc_rsq_ps(dx33,dy33,dz33);
2185
2186             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2187             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2188             rinv13           = gmx_mm256_invsqrt_ps(rsq13);
2189             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2190             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2191             rinv23           = gmx_mm256_invsqrt_ps(rsq23);
2192             rinv31           = gmx_mm256_invsqrt_ps(rsq31);
2193             rinv32           = gmx_mm256_invsqrt_ps(rsq32);
2194             rinv33           = gmx_mm256_invsqrt_ps(rsq33);
2195
2196             fjx1             = _mm256_setzero_ps();
2197             fjy1             = _mm256_setzero_ps();
2198             fjz1             = _mm256_setzero_ps();
2199             fjx2             = _mm256_setzero_ps();
2200             fjy2             = _mm256_setzero_ps();
2201             fjz2             = _mm256_setzero_ps();
2202             fjx3             = _mm256_setzero_ps();
2203             fjy3             = _mm256_setzero_ps();
2204             fjz3             = _mm256_setzero_ps();
2205
2206             /**************************
2207              * CALCULATE INTERACTIONS *
2208              **************************/
2209
2210             r11              = _mm256_mul_ps(rsq11,rinv11);
2211             r11              = _mm256_andnot_ps(dummy_mask,r11);
2212
2213             /* Calculate table index by multiplying r with table scale and truncate to integer */
2214             rt               = _mm256_mul_ps(r11,vftabscale);
2215             vfitab           = _mm256_cvttps_epi32(rt);
2216             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2217             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2218             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2219             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2220             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2221             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2222
2223             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2224             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2225                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2226             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2227                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2228             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2229                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2230             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2231                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2232             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2233             Heps             = _mm256_mul_ps(vfeps,H);
2234             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2235             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2236             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
2237
2238             fscal            = felec;
2239
2240             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2241
2242             /* Calculate temporary vectorial force */
2243             tx               = _mm256_mul_ps(fscal,dx11);
2244             ty               = _mm256_mul_ps(fscal,dy11);
2245             tz               = _mm256_mul_ps(fscal,dz11);
2246
2247             /* Update vectorial force */
2248             fix1             = _mm256_add_ps(fix1,tx);
2249             fiy1             = _mm256_add_ps(fiy1,ty);
2250             fiz1             = _mm256_add_ps(fiz1,tz);
2251
2252             fjx1             = _mm256_add_ps(fjx1,tx);
2253             fjy1             = _mm256_add_ps(fjy1,ty);
2254             fjz1             = _mm256_add_ps(fjz1,tz);
2255
2256             /**************************
2257              * CALCULATE INTERACTIONS *
2258              **************************/
2259
2260             r12              = _mm256_mul_ps(rsq12,rinv12);
2261             r12              = _mm256_andnot_ps(dummy_mask,r12);
2262
2263             /* Calculate table index by multiplying r with table scale and truncate to integer */
2264             rt               = _mm256_mul_ps(r12,vftabscale);
2265             vfitab           = _mm256_cvttps_epi32(rt);
2266             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2267             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2268             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2269             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2270             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2271             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2272
2273             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2274             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2275                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2276             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2277                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2278             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2279                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2280             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2281                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2282             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2283             Heps             = _mm256_mul_ps(vfeps,H);
2284             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2285             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2286             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
2287
2288             fscal            = felec;
2289
2290             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2291
2292             /* Calculate temporary vectorial force */
2293             tx               = _mm256_mul_ps(fscal,dx12);
2294             ty               = _mm256_mul_ps(fscal,dy12);
2295             tz               = _mm256_mul_ps(fscal,dz12);
2296
2297             /* Update vectorial force */
2298             fix1             = _mm256_add_ps(fix1,tx);
2299             fiy1             = _mm256_add_ps(fiy1,ty);
2300             fiz1             = _mm256_add_ps(fiz1,tz);
2301
2302             fjx2             = _mm256_add_ps(fjx2,tx);
2303             fjy2             = _mm256_add_ps(fjy2,ty);
2304             fjz2             = _mm256_add_ps(fjz2,tz);
2305
2306             /**************************
2307              * CALCULATE INTERACTIONS *
2308              **************************/
2309
2310             r13              = _mm256_mul_ps(rsq13,rinv13);
2311             r13              = _mm256_andnot_ps(dummy_mask,r13);
2312
2313             /* Calculate table index by multiplying r with table scale and truncate to integer */
2314             rt               = _mm256_mul_ps(r13,vftabscale);
2315             vfitab           = _mm256_cvttps_epi32(rt);
2316             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2317             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2318             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2319             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2320             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2321             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2322
2323             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2324             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2325                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2326             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2327                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2328             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2329                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2330             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2331                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2332             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2333             Heps             = _mm256_mul_ps(vfeps,H);
2334             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2335             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2336             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq13,FF),_mm256_mul_ps(vftabscale,rinv13)));
2337
2338             fscal            = felec;
2339
2340             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2341
2342             /* Calculate temporary vectorial force */
2343             tx               = _mm256_mul_ps(fscal,dx13);
2344             ty               = _mm256_mul_ps(fscal,dy13);
2345             tz               = _mm256_mul_ps(fscal,dz13);
2346
2347             /* Update vectorial force */
2348             fix1             = _mm256_add_ps(fix1,tx);
2349             fiy1             = _mm256_add_ps(fiy1,ty);
2350             fiz1             = _mm256_add_ps(fiz1,tz);
2351
2352             fjx3             = _mm256_add_ps(fjx3,tx);
2353             fjy3             = _mm256_add_ps(fjy3,ty);
2354             fjz3             = _mm256_add_ps(fjz3,tz);
2355
2356             /**************************
2357              * CALCULATE INTERACTIONS *
2358              **************************/
2359
2360             r21              = _mm256_mul_ps(rsq21,rinv21);
2361             r21              = _mm256_andnot_ps(dummy_mask,r21);
2362
2363             /* Calculate table index by multiplying r with table scale and truncate to integer */
2364             rt               = _mm256_mul_ps(r21,vftabscale);
2365             vfitab           = _mm256_cvttps_epi32(rt);
2366             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2367             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2368             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2369             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2370             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2371             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2372
2373             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2374             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2375                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2376             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2377                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2378             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2379                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2380             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2381                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2382             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2383             Heps             = _mm256_mul_ps(vfeps,H);
2384             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2385             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2386             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
2387
2388             fscal            = felec;
2389
2390             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2391
2392             /* Calculate temporary vectorial force */
2393             tx               = _mm256_mul_ps(fscal,dx21);
2394             ty               = _mm256_mul_ps(fscal,dy21);
2395             tz               = _mm256_mul_ps(fscal,dz21);
2396
2397             /* Update vectorial force */
2398             fix2             = _mm256_add_ps(fix2,tx);
2399             fiy2             = _mm256_add_ps(fiy2,ty);
2400             fiz2             = _mm256_add_ps(fiz2,tz);
2401
2402             fjx1             = _mm256_add_ps(fjx1,tx);
2403             fjy1             = _mm256_add_ps(fjy1,ty);
2404             fjz1             = _mm256_add_ps(fjz1,tz);
2405
2406             /**************************
2407              * CALCULATE INTERACTIONS *
2408              **************************/
2409
2410             r22              = _mm256_mul_ps(rsq22,rinv22);
2411             r22              = _mm256_andnot_ps(dummy_mask,r22);
2412
2413             /* Calculate table index by multiplying r with table scale and truncate to integer */
2414             rt               = _mm256_mul_ps(r22,vftabscale);
2415             vfitab           = _mm256_cvttps_epi32(rt);
2416             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2417             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2418             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2419             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2420             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2421             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2422
2423             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2424             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2425                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2426             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2427                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2428             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2429                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2430             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2431                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2432             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2433             Heps             = _mm256_mul_ps(vfeps,H);
2434             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2435             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2436             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
2437
2438             fscal            = felec;
2439
2440             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2441
2442             /* Calculate temporary vectorial force */
2443             tx               = _mm256_mul_ps(fscal,dx22);
2444             ty               = _mm256_mul_ps(fscal,dy22);
2445             tz               = _mm256_mul_ps(fscal,dz22);
2446
2447             /* Update vectorial force */
2448             fix2             = _mm256_add_ps(fix2,tx);
2449             fiy2             = _mm256_add_ps(fiy2,ty);
2450             fiz2             = _mm256_add_ps(fiz2,tz);
2451
2452             fjx2             = _mm256_add_ps(fjx2,tx);
2453             fjy2             = _mm256_add_ps(fjy2,ty);
2454             fjz2             = _mm256_add_ps(fjz2,tz);
2455
2456             /**************************
2457              * CALCULATE INTERACTIONS *
2458              **************************/
2459
2460             r23              = _mm256_mul_ps(rsq23,rinv23);
2461             r23              = _mm256_andnot_ps(dummy_mask,r23);
2462
2463             /* Calculate table index by multiplying r with table scale and truncate to integer */
2464             rt               = _mm256_mul_ps(r23,vftabscale);
2465             vfitab           = _mm256_cvttps_epi32(rt);
2466             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2467             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2468             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2469             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2470             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2471             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2472
2473             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2474             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2475                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2476             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2477                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2478             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2479                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2480             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2481                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2482             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2483             Heps             = _mm256_mul_ps(vfeps,H);
2484             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2485             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2486             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq23,FF),_mm256_mul_ps(vftabscale,rinv23)));
2487
2488             fscal            = felec;
2489
2490             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2491
2492             /* Calculate temporary vectorial force */
2493             tx               = _mm256_mul_ps(fscal,dx23);
2494             ty               = _mm256_mul_ps(fscal,dy23);
2495             tz               = _mm256_mul_ps(fscal,dz23);
2496
2497             /* Update vectorial force */
2498             fix2             = _mm256_add_ps(fix2,tx);
2499             fiy2             = _mm256_add_ps(fiy2,ty);
2500             fiz2             = _mm256_add_ps(fiz2,tz);
2501
2502             fjx3             = _mm256_add_ps(fjx3,tx);
2503             fjy3             = _mm256_add_ps(fjy3,ty);
2504             fjz3             = _mm256_add_ps(fjz3,tz);
2505
2506             /**************************
2507              * CALCULATE INTERACTIONS *
2508              **************************/
2509
2510             r31              = _mm256_mul_ps(rsq31,rinv31);
2511             r31              = _mm256_andnot_ps(dummy_mask,r31);
2512
2513             /* Calculate table index by multiplying r with table scale and truncate to integer */
2514             rt               = _mm256_mul_ps(r31,vftabscale);
2515             vfitab           = _mm256_cvttps_epi32(rt);
2516             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2517             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2518             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2519             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2520             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2521             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2522
2523             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2524             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2525                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2526             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2527                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2528             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2529                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2530             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2531                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2532             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2533             Heps             = _mm256_mul_ps(vfeps,H);
2534             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2535             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2536             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq31,FF),_mm256_mul_ps(vftabscale,rinv31)));
2537
2538             fscal            = felec;
2539
2540             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2541
2542             /* Calculate temporary vectorial force */
2543             tx               = _mm256_mul_ps(fscal,dx31);
2544             ty               = _mm256_mul_ps(fscal,dy31);
2545             tz               = _mm256_mul_ps(fscal,dz31);
2546
2547             /* Update vectorial force */
2548             fix3             = _mm256_add_ps(fix3,tx);
2549             fiy3             = _mm256_add_ps(fiy3,ty);
2550             fiz3             = _mm256_add_ps(fiz3,tz);
2551
2552             fjx1             = _mm256_add_ps(fjx1,tx);
2553             fjy1             = _mm256_add_ps(fjy1,ty);
2554             fjz1             = _mm256_add_ps(fjz1,tz);
2555
2556             /**************************
2557              * CALCULATE INTERACTIONS *
2558              **************************/
2559
2560             r32              = _mm256_mul_ps(rsq32,rinv32);
2561             r32              = _mm256_andnot_ps(dummy_mask,r32);
2562
2563             /* Calculate table index by multiplying r with table scale and truncate to integer */
2564             rt               = _mm256_mul_ps(r32,vftabscale);
2565             vfitab           = _mm256_cvttps_epi32(rt);
2566             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2567             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2568             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2569             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2570             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2571             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2572
2573             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2574             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2575                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2576             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2577                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2578             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2579                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2580             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2581                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2582             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2583             Heps             = _mm256_mul_ps(vfeps,H);
2584             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2585             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2586             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq32,FF),_mm256_mul_ps(vftabscale,rinv32)));
2587
2588             fscal            = felec;
2589
2590             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2591
2592             /* Calculate temporary vectorial force */
2593             tx               = _mm256_mul_ps(fscal,dx32);
2594             ty               = _mm256_mul_ps(fscal,dy32);
2595             tz               = _mm256_mul_ps(fscal,dz32);
2596
2597             /* Update vectorial force */
2598             fix3             = _mm256_add_ps(fix3,tx);
2599             fiy3             = _mm256_add_ps(fiy3,ty);
2600             fiz3             = _mm256_add_ps(fiz3,tz);
2601
2602             fjx2             = _mm256_add_ps(fjx2,tx);
2603             fjy2             = _mm256_add_ps(fjy2,ty);
2604             fjz2             = _mm256_add_ps(fjz2,tz);
2605
2606             /**************************
2607              * CALCULATE INTERACTIONS *
2608              **************************/
2609
2610             r33              = _mm256_mul_ps(rsq33,rinv33);
2611             r33              = _mm256_andnot_ps(dummy_mask,r33);
2612
2613             /* Calculate table index by multiplying r with table scale and truncate to integer */
2614             rt               = _mm256_mul_ps(r33,vftabscale);
2615             vfitab           = _mm256_cvttps_epi32(rt);
2616             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2617             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2618             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2619             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2620             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2621             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2622
2623             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2624             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2625                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2626             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2627                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2628             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2629                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2630             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2631                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2632             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2633             Heps             = _mm256_mul_ps(vfeps,H);
2634             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2635             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2636             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq33,FF),_mm256_mul_ps(vftabscale,rinv33)));
2637
2638             fscal            = felec;
2639
2640             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2641
2642             /* Calculate temporary vectorial force */
2643             tx               = _mm256_mul_ps(fscal,dx33);
2644             ty               = _mm256_mul_ps(fscal,dy33);
2645             tz               = _mm256_mul_ps(fscal,dz33);
2646
2647             /* Update vectorial force */
2648             fix3             = _mm256_add_ps(fix3,tx);
2649             fiy3             = _mm256_add_ps(fiy3,ty);
2650             fiz3             = _mm256_add_ps(fiz3,tz);
2651
2652             fjx3             = _mm256_add_ps(fjx3,tx);
2653             fjy3             = _mm256_add_ps(fjy3,ty);
2654             fjz3             = _mm256_add_ps(fjz3,tz);
2655
2656             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2657             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2658             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2659             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2660             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2661             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2662             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2663             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2664
2665             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
2666                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
2667                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2668
2669             /* Inner loop uses 360 flops */
2670         }
2671
2672         /* End of innermost loop */
2673
2674         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2675                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
2676
2677         /* Increment number of inner iterations */
2678         inneriter                  += j_index_end - j_index_start;
2679
2680         /* Outer loop uses 18 flops */
2681     }
2682
2683     /* Increment number of outer iterations */
2684     outeriter        += nri;
2685
2686     /* Update outer/inner flops */
2687
2688     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4W4_F,outeriter*18 + inneriter*360);
2689 }