src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwNone_GeomW3W3_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "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_GeomW3W3_VF_avx_256_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Water3
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwNone_GeomW3W3_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D,vdwjidx1E,vdwjidx1F,vdwjidx1G,vdwjidx1H;
  97     __m256           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  98     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D,vdwjidx2E,vdwjidx2F,vdwjidx2G,vdwjidx2H;
  99     __m256           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 100     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 102     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 103     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 104     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 105     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 106     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 107     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 108     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 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     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 140     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 141     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 142
 143     jq0              = _mm256_set1_ps(charge[inr+0]);
 144     jq1              = _mm256_set1_ps(charge[inr+1]);
 145     jq2              = _mm256_set1_ps(charge[inr+2]);
 146     qq00             = _mm256_mul_ps(iq0,jq0);
 147     qq01             = _mm256_mul_ps(iq0,jq1);
 148     qq02             = _mm256_mul_ps(iq0,jq2);
 149     qq10             = _mm256_mul_ps(iq1,jq0);
 150     qq11             = _mm256_mul_ps(iq1,jq1);
 151     qq12             = _mm256_mul_ps(iq1,jq2);
 152     qq20             = _mm256_mul_ps(iq2,jq0);
 153     qq21             = _mm256_mul_ps(iq2,jq1);
 154     qq22             = _mm256_mul_ps(iq2,jq2);
 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,
 191                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 192
 193         fix0             = _mm256_setzero_ps();
 194         fiy0             = _mm256_setzero_ps();
 195         fiz0             = _mm256_setzero_ps();
 196         fix1             = _mm256_setzero_ps();
 197         fiy1             = _mm256_setzero_ps();
 198         fiz1             = _mm256_setzero_ps();
 199         fix2             = _mm256_setzero_ps();
 200         fiy2             = _mm256_setzero_ps();
 201         fiz2             = _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,x+j_coord_offsetB,
 230                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 231                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 232                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 233                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 234
 235             /* Calculate displacement vector */
 236             dx00             = _mm256_sub_ps(ix0,jx0);
 237             dy00             = _mm256_sub_ps(iy0,jy0);
 238             dz00             = _mm256_sub_ps(iz0,jz0);
 239             dx01             = _mm256_sub_ps(ix0,jx1);
 240             dy01             = _mm256_sub_ps(iy0,jy1);
 241             dz01             = _mm256_sub_ps(iz0,jz1);
 242             dx02             = _mm256_sub_ps(ix0,jx2);
 243             dy02             = _mm256_sub_ps(iy0,jy2);
 244             dz02             = _mm256_sub_ps(iz0,jz2);
 245             dx10             = _mm256_sub_ps(ix1,jx0);
 246             dy10             = _mm256_sub_ps(iy1,jy0);
 247             dz10             = _mm256_sub_ps(iz1,jz0);
 248             dx11             = _mm256_sub_ps(ix1,jx1);
 249             dy11             = _mm256_sub_ps(iy1,jy1);
 250             dz11             = _mm256_sub_ps(iz1,jz1);
 251             dx12             = _mm256_sub_ps(ix1,jx2);
 252             dy12             = _mm256_sub_ps(iy1,jy2);
 253             dz12             = _mm256_sub_ps(iz1,jz2);
 254             dx20             = _mm256_sub_ps(ix2,jx0);
 255             dy20             = _mm256_sub_ps(iy2,jy0);
 256             dz20             = _mm256_sub_ps(iz2,jz0);
 257             dx21             = _mm256_sub_ps(ix2,jx1);
 258             dy21             = _mm256_sub_ps(iy2,jy1);
 259             dz21             = _mm256_sub_ps(iz2,jz1);
 260             dx22             = _mm256_sub_ps(ix2,jx2);
 261             dy22             = _mm256_sub_ps(iy2,jy2);
 262             dz22             = _mm256_sub_ps(iz2,jz2);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 266             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 267             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 268             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 269             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 270             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 271             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 272             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 273             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 274
 275             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 276             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 277             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 278             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 279             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 280             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 281             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 282             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 283             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 284
 285             fjx0             = _mm256_setzero_ps();
 286             fjy0             = _mm256_setzero_ps();
 287             fjz0             = _mm256_setzero_ps();
 288             fjx1             = _mm256_setzero_ps();
 289             fjy1             = _mm256_setzero_ps();
 290             fjz1             = _mm256_setzero_ps();
 291             fjx2             = _mm256_setzero_ps();
 292             fjy2             = _mm256_setzero_ps();
 293             fjz2             = _mm256_setzero_ps();
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             r00              = _mm256_mul_ps(rsq00,rinv00);
 300
 301             /* Calculate table index by multiplying r with table scale and truncate to integer */
 302             rt               = _mm256_mul_ps(r00,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(qq00,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(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 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,dx00);
 335             ty               = _mm256_mul_ps(fscal,dy00);
 336             tz               = _mm256_mul_ps(fscal,dz00);
 337
 338             /* Update vectorial force */
 339             fix0             = _mm256_add_ps(fix0,tx);
 340             fiy0             = _mm256_add_ps(fiy0,ty);
 341             fiz0             = _mm256_add_ps(fiz0,tz);
 342
 343             fjx0             = _mm256_add_ps(fjx0,tx);
 344             fjy0             = _mm256_add_ps(fjy0,ty);
 345             fjz0             = _mm256_add_ps(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r01              = _mm256_mul_ps(rsq01,rinv01);
 352
 353             /* Calculate table index by multiplying r with table scale and truncate to integer */
 354             rt               = _mm256_mul_ps(r01,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(qq01,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(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
 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,dx01);
 387             ty               = _mm256_mul_ps(fscal,dy01);
 388             tz               = _mm256_mul_ps(fscal,dz01);
 389
 390             /* Update vectorial force */
 391             fix0             = _mm256_add_ps(fix0,tx);
 392             fiy0             = _mm256_add_ps(fiy0,ty);
 393             fiz0             = _mm256_add_ps(fiz0,tz);
 394
 395             fjx1             = _mm256_add_ps(fjx1,tx);
 396             fjy1             = _mm256_add_ps(fjy1,ty);
 397             fjz1             = _mm256_add_ps(fjz1,tz);
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             r02              = _mm256_mul_ps(rsq02,rinv02);
 404
 405             /* Calculate table index by multiplying r with table scale and truncate to integer */
 406             rt               = _mm256_mul_ps(r02,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(qq02,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(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
 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,dx02);
 439             ty               = _mm256_mul_ps(fscal,dy02);
 440             tz               = _mm256_mul_ps(fscal,dz02);
 441
 442             /* Update vectorial force */
 443             fix0             = _mm256_add_ps(fix0,tx);
 444             fiy0             = _mm256_add_ps(fiy0,ty);
 445             fiz0             = _mm256_add_ps(fiz0,tz);
 446
 447             fjx2             = _mm256_add_ps(fjx2,tx);
 448             fjy2             = _mm256_add_ps(fjy2,ty);
 449             fjz2             = _mm256_add_ps(fjz2,tz);
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             r10              = _mm256_mul_ps(rsq10,rinv10);
 456
 457             /* Calculate table index by multiplying r with table scale and truncate to integer */
 458             rt               = _mm256_mul_ps(r10,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(qq10,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(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 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,dx10);
 491             ty               = _mm256_mul_ps(fscal,dy10);
 492             tz               = _mm256_mul_ps(fscal,dz10);
 493
 494             /* Update vectorial force */
 495             fix1             = _mm256_add_ps(fix1,tx);
 496             fiy1             = _mm256_add_ps(fiy1,ty);
 497             fiz1             = _mm256_add_ps(fiz1,tz);
 498
 499             fjx0             = _mm256_add_ps(fjx0,tx);
 500             fjy0             = _mm256_add_ps(fjy0,ty);
 501             fjz0             = _mm256_add_ps(fjz0,tz);
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             r11              = _mm256_mul_ps(rsq11,rinv11);
 508
 509             /* Calculate table index by multiplying r with table scale and truncate to integer */
 510             rt               = _mm256_mul_ps(r11,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(qq11,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(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
 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,dx11);
 543             ty               = _mm256_mul_ps(fscal,dy11);
 544             tz               = _mm256_mul_ps(fscal,dz11);
 545
 546             /* Update vectorial force */
 547             fix1             = _mm256_add_ps(fix1,tx);
 548             fiy1             = _mm256_add_ps(fiy1,ty);
 549             fiz1             = _mm256_add_ps(fiz1,tz);
 550
 551             fjx1             = _mm256_add_ps(fjx1,tx);
 552             fjy1             = _mm256_add_ps(fjy1,ty);
 553             fjz1             = _mm256_add_ps(fjz1,tz);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             r12              = _mm256_mul_ps(rsq12,rinv12);
 560
 561             /* Calculate table index by multiplying r with table scale and truncate to integer */
 562             rt               = _mm256_mul_ps(r12,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(qq12,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(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
 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,dx12);
 595             ty               = _mm256_mul_ps(fscal,dy12);
 596             tz               = _mm256_mul_ps(fscal,dz12);
 597
 598             /* Update vectorial force */
 599             fix1             = _mm256_add_ps(fix1,tx);
 600             fiy1             = _mm256_add_ps(fiy1,ty);
 601             fiz1             = _mm256_add_ps(fiz1,tz);
 602
 603             fjx2             = _mm256_add_ps(fjx2,tx);
 604             fjy2             = _mm256_add_ps(fjy2,ty);
 605             fjz2             = _mm256_add_ps(fjz2,tz);
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             r20              = _mm256_mul_ps(rsq20,rinv20);
 612
 613             /* Calculate table index by multiplying r with table scale and truncate to integer */
 614             rt               = _mm256_mul_ps(r20,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(qq20,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(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 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,dx20);
 647             ty               = _mm256_mul_ps(fscal,dy20);
 648             tz               = _mm256_mul_ps(fscal,dz20);
 649
 650             /* Update vectorial force */
 651             fix2             = _mm256_add_ps(fix2,tx);
 652             fiy2             = _mm256_add_ps(fiy2,ty);
 653             fiz2             = _mm256_add_ps(fiz2,tz);
 654
 655             fjx0             = _mm256_add_ps(fjx0,tx);
 656             fjy0             = _mm256_add_ps(fjy0,ty);
 657             fjz0             = _mm256_add_ps(fjz0,tz);
 658
 659             /**************************
 660              * CALCULATE INTERACTIONS *
 661              **************************/
 662
 663             r21              = _mm256_mul_ps(rsq21,rinv21);
 664
 665             /* Calculate table index by multiplying r with table scale and truncate to integer */
 666             rt               = _mm256_mul_ps(r21,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(qq21,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(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
 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,dx21);
 699             ty               = _mm256_mul_ps(fscal,dy21);
 700             tz               = _mm256_mul_ps(fscal,dz21);
 701
 702             /* Update vectorial force */
 703             fix2             = _mm256_add_ps(fix2,tx);
 704             fiy2             = _mm256_add_ps(fiy2,ty);
 705             fiz2             = _mm256_add_ps(fiz2,tz);
 706
 707             fjx1             = _mm256_add_ps(fjx1,tx);
 708             fjy1             = _mm256_add_ps(fjy1,ty);
 709             fjz1             = _mm256_add_ps(fjz1,tz);
 710
 711             /**************************
 712              * CALCULATE INTERACTIONS *
 713              **************************/
 714
 715             r22              = _mm256_mul_ps(rsq22,rinv22);
 716
 717             /* Calculate table index by multiplying r with table scale and truncate to integer */
 718             rt               = _mm256_mul_ps(r22,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(qq22,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(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
 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,dx22);
 751             ty               = _mm256_mul_ps(fscal,dy22);
 752             tz               = _mm256_mul_ps(fscal,dz22);
 753
 754             /* Update vectorial force */
 755             fix2             = _mm256_add_ps(fix2,tx);
 756             fiy2             = _mm256_add_ps(fiy2,ty);
 757             fiz2             = _mm256_add_ps(fiz2,tz);
 758
 759             fjx2             = _mm256_add_ps(fjx2,tx);
 760             fjy2             = _mm256_add_ps(fjy2,ty);
 761             fjz2             = _mm256_add_ps(fjz2,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,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 773                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 774
 775             /* Inner loop uses 387 flops */
 776         }
 777
 778         if(jidx<j_index_end)
 779         {
 780
 781             /* Get j neighbor index, and coordinate index */
 782             jnrlistA         = jjnr[jidx];
 783             jnrlistB         = jjnr[jidx+1];
 784             jnrlistC         = jjnr[jidx+2];
 785             jnrlistD         = jjnr[jidx+3];
 786             jnrlistE         = jjnr[jidx+4];
 787             jnrlistF         = jjnr[jidx+5];
 788             jnrlistG         = jjnr[jidx+6];
 789             jnrlistH         = jjnr[jidx+7];
 790             /* Sign of each element will be negative for non-real atoms.
 791              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 792              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 793              */
 794             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 795                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 796
 797             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 798             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 799             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 800             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 801             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 802             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 803             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 804             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 805             j_coord_offsetA  = DIM*jnrA;
 806             j_coord_offsetB  = DIM*jnrB;
 807             j_coord_offsetC  = DIM*jnrC;
 808             j_coord_offsetD  = DIM*jnrD;
 809             j_coord_offsetE  = DIM*jnrE;
 810             j_coord_offsetF  = DIM*jnrF;
 811             j_coord_offsetG  = DIM*jnrG;
 812             j_coord_offsetH  = DIM*jnrH;
 813
 814             /* load j atom coordinates */
 815             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 816                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 817                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 818                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 819                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 820
 821             /* Calculate displacement vector */
 822             dx00             = _mm256_sub_ps(ix0,jx0);
 823             dy00             = _mm256_sub_ps(iy0,jy0);
 824             dz00             = _mm256_sub_ps(iz0,jz0);
 825             dx01             = _mm256_sub_ps(ix0,jx1);
 826             dy01             = _mm256_sub_ps(iy0,jy1);
 827             dz01             = _mm256_sub_ps(iz0,jz1);
 828             dx02             = _mm256_sub_ps(ix0,jx2);
 829             dy02             = _mm256_sub_ps(iy0,jy2);
 830             dz02             = _mm256_sub_ps(iz0,jz2);
 831             dx10             = _mm256_sub_ps(ix1,jx0);
 832             dy10             = _mm256_sub_ps(iy1,jy0);
 833             dz10             = _mm256_sub_ps(iz1,jz0);
 834             dx11             = _mm256_sub_ps(ix1,jx1);
 835             dy11             = _mm256_sub_ps(iy1,jy1);
 836             dz11             = _mm256_sub_ps(iz1,jz1);
 837             dx12             = _mm256_sub_ps(ix1,jx2);
 838             dy12             = _mm256_sub_ps(iy1,jy2);
 839             dz12             = _mm256_sub_ps(iz1,jz2);
 840             dx20             = _mm256_sub_ps(ix2,jx0);
 841             dy20             = _mm256_sub_ps(iy2,jy0);
 842             dz20             = _mm256_sub_ps(iz2,jz0);
 843             dx21             = _mm256_sub_ps(ix2,jx1);
 844             dy21             = _mm256_sub_ps(iy2,jy1);
 845             dz21             = _mm256_sub_ps(iz2,jz1);
 846             dx22             = _mm256_sub_ps(ix2,jx2);
 847             dy22             = _mm256_sub_ps(iy2,jy2);
 848             dz22             = _mm256_sub_ps(iz2,jz2);
 849
 850             /* Calculate squared distance and things based on it */
 851             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 852             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
 853             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
 854             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 855             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
 856             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
 857             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 858             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
 859             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
 860
 861             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 862             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
 863             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
 864             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 865             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
 866             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
 867             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 868             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
 869             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
 870
 871             fjx0             = _mm256_setzero_ps();
 872             fjy0             = _mm256_setzero_ps();
 873             fjz0             = _mm256_setzero_ps();
 874             fjx1             = _mm256_setzero_ps();
 875             fjy1             = _mm256_setzero_ps();
 876             fjz1             = _mm256_setzero_ps();
 877             fjx2             = _mm256_setzero_ps();
 878             fjy2             = _mm256_setzero_ps();
 879             fjz2             = _mm256_setzero_ps();
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             r00              = _mm256_mul_ps(rsq00,rinv00);
 886             r00              = _mm256_andnot_ps(dummy_mask,r00);
 887
 888             /* Calculate table index by multiplying r with table scale and truncate to integer */
 889             rt               = _mm256_mul_ps(r00,vftabscale);
 890             vfitab           = _mm256_cvttps_epi32(rt);
 891             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 892             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 893             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 894             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 895             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 896             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 897
 898             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 899             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 900                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 901             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 902                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 903             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 904                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 905             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 906                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 907             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 908             Heps             = _mm256_mul_ps(vfeps,H);
 909             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 910             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 911             velec            = _mm256_mul_ps(qq00,VV);
 912             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 913             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 914
 915             /* Update potential sum for this i atom from the interaction with this j atom. */
 916             velec            = _mm256_andnot_ps(dummy_mask,velec);
 917             velecsum         = _mm256_add_ps(velecsum,velec);
 918
 919             fscal            = felec;
 920
 921             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 922
 923             /* Calculate temporary vectorial force */
 924             tx               = _mm256_mul_ps(fscal,dx00);
 925             ty               = _mm256_mul_ps(fscal,dy00);
 926             tz               = _mm256_mul_ps(fscal,dz00);
 927
 928             /* Update vectorial force */
 929             fix0             = _mm256_add_ps(fix0,tx);
 930             fiy0             = _mm256_add_ps(fiy0,ty);
 931             fiz0             = _mm256_add_ps(fiz0,tz);
 932
 933             fjx0             = _mm256_add_ps(fjx0,tx);
 934             fjy0             = _mm256_add_ps(fjy0,ty);
 935             fjz0             = _mm256_add_ps(fjz0,tz);
 936
 937             /**************************
 938              * CALCULATE INTERACTIONS *
 939              **************************/
 940
 941             r01              = _mm256_mul_ps(rsq01,rinv01);
 942             r01              = _mm256_andnot_ps(dummy_mask,r01);
 943
 944             /* Calculate table index by multiplying r with table scale and truncate to integer */
 945             rt               = _mm256_mul_ps(r01,vftabscale);
 946             vfitab           = _mm256_cvttps_epi32(rt);
 947             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 948             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 949             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 950             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 951             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 952             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 953
 954             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 955             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 956                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 957             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 958                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 959             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 960                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 961             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 962                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 963             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 964             Heps             = _mm256_mul_ps(vfeps,H);
 965             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 966             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 967             velec            = _mm256_mul_ps(qq01,VV);
 968             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 969             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
 970
 971             /* Update potential sum for this i atom from the interaction with this j atom. */
 972             velec            = _mm256_andnot_ps(dummy_mask,velec);
 973             velecsum         = _mm256_add_ps(velecsum,velec);
 974
 975             fscal            = felec;
 976
 977             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 978
 979             /* Calculate temporary vectorial force */
 980             tx               = _mm256_mul_ps(fscal,dx01);
 981             ty               = _mm256_mul_ps(fscal,dy01);
 982             tz               = _mm256_mul_ps(fscal,dz01);
 983
 984             /* Update vectorial force */
 985             fix0             = _mm256_add_ps(fix0,tx);
 986             fiy0             = _mm256_add_ps(fiy0,ty);
 987             fiz0             = _mm256_add_ps(fiz0,tz);
 988
 989             fjx1             = _mm256_add_ps(fjx1,tx);
 990             fjy1             = _mm256_add_ps(fjy1,ty);
 991             fjz1             = _mm256_add_ps(fjz1,tz);
 992
 993             /**************************
 994              * CALCULATE INTERACTIONS *
 995              **************************/
 996
 997             r02              = _mm256_mul_ps(rsq02,rinv02);
 998             r02              = _mm256_andnot_ps(dummy_mask,r02);
 999
1000             /* Calculate table index by multiplying r with table scale and truncate to integer */
1001             rt               = _mm256_mul_ps(r02,vftabscale);
1002             vfitab           = _mm256_cvttps_epi32(rt);
1003             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1004             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1005             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1006             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1007             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1008             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1009
1010             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1011             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1012                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1013             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1014                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1015             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1016                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1017             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1018                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1019             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1020             Heps             = _mm256_mul_ps(vfeps,H);
1021             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1022             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1023             velec            = _mm256_mul_ps(qq02,VV);
1024             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1025             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
1026
1027             /* Update potential sum for this i atom from the interaction with this j atom. */
1028             velec            = _mm256_andnot_ps(dummy_mask,velec);
1029             velecsum         = _mm256_add_ps(velecsum,velec);
1030
1031             fscal            = felec;
1032
1033             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm256_mul_ps(fscal,dx02);
1037             ty               = _mm256_mul_ps(fscal,dy02);
1038             tz               = _mm256_mul_ps(fscal,dz02);
1039
1040             /* Update vectorial force */
1041             fix0             = _mm256_add_ps(fix0,tx);
1042             fiy0             = _mm256_add_ps(fiy0,ty);
1043             fiz0             = _mm256_add_ps(fiz0,tz);
1044
1045             fjx2             = _mm256_add_ps(fjx2,tx);
1046             fjy2             = _mm256_add_ps(fjy2,ty);
1047             fjz2             = _mm256_add_ps(fjz2,tz);
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             r10              = _mm256_mul_ps(rsq10,rinv10);
1054             r10              = _mm256_andnot_ps(dummy_mask,r10);
1055
1056             /* Calculate table index by multiplying r with table scale and truncate to integer */
1057             rt               = _mm256_mul_ps(r10,vftabscale);
1058             vfitab           = _mm256_cvttps_epi32(rt);
1059             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1060             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1061             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1062             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1063             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1064             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1065
1066             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1067             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1068                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1069             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1070                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1071             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1072                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1073             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1074                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1075             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1076             Heps             = _mm256_mul_ps(vfeps,H);
1077             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1078             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1079             velec            = _mm256_mul_ps(qq10,VV);
1080             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1081             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1082
1083             /* Update potential sum for this i atom from the interaction with this j atom. */
1084             velec            = _mm256_andnot_ps(dummy_mask,velec);
1085             velecsum         = _mm256_add_ps(velecsum,velec);
1086
1087             fscal            = felec;
1088
1089             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1090
1091             /* Calculate temporary vectorial force */
1092             tx               = _mm256_mul_ps(fscal,dx10);
1093             ty               = _mm256_mul_ps(fscal,dy10);
1094             tz               = _mm256_mul_ps(fscal,dz10);
1095
1096             /* Update vectorial force */
1097             fix1             = _mm256_add_ps(fix1,tx);
1098             fiy1             = _mm256_add_ps(fiy1,ty);
1099             fiz1             = _mm256_add_ps(fiz1,tz);
1100
1101             fjx0             = _mm256_add_ps(fjx0,tx);
1102             fjy0             = _mm256_add_ps(fjy0,ty);
1103             fjz0             = _mm256_add_ps(fjz0,tz);
1104
1105             /**************************
1106              * CALCULATE INTERACTIONS *
1107              **************************/
1108
1109             r11              = _mm256_mul_ps(rsq11,rinv11);
1110             r11              = _mm256_andnot_ps(dummy_mask,r11);
1111
1112             /* Calculate table index by multiplying r with table scale and truncate to integer */
1113             rt               = _mm256_mul_ps(r11,vftabscale);
1114             vfitab           = _mm256_cvttps_epi32(rt);
1115             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1116             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1117             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1118             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1119             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1120             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1121
1122             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1123             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1124                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1125             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1126                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1127             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1128                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1129             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1130                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1131             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1132             Heps             = _mm256_mul_ps(vfeps,H);
1133             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1134             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1135             velec            = _mm256_mul_ps(qq11,VV);
1136             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1137             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
1138
1139             /* Update potential sum for this i atom from the interaction with this j atom. */
1140             velec            = _mm256_andnot_ps(dummy_mask,velec);
1141             velecsum         = _mm256_add_ps(velecsum,velec);
1142
1143             fscal            = felec;
1144
1145             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1146
1147             /* Calculate temporary vectorial force */
1148             tx               = _mm256_mul_ps(fscal,dx11);
1149             ty               = _mm256_mul_ps(fscal,dy11);
1150             tz               = _mm256_mul_ps(fscal,dz11);
1151
1152             /* Update vectorial force */
1153             fix1             = _mm256_add_ps(fix1,tx);
1154             fiy1             = _mm256_add_ps(fiy1,ty);
1155             fiz1             = _mm256_add_ps(fiz1,tz);
1156
1157             fjx1             = _mm256_add_ps(fjx1,tx);
1158             fjy1             = _mm256_add_ps(fjy1,ty);
1159             fjz1             = _mm256_add_ps(fjz1,tz);
1160
1161             /**************************
1162              * CALCULATE INTERACTIONS *
1163              **************************/
1164
1165             r12              = _mm256_mul_ps(rsq12,rinv12);
1166             r12              = _mm256_andnot_ps(dummy_mask,r12);
1167
1168             /* Calculate table index by multiplying r with table scale and truncate to integer */
1169             rt               = _mm256_mul_ps(r12,vftabscale);
1170             vfitab           = _mm256_cvttps_epi32(rt);
1171             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1172             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1173             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1174             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1175             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1176             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1177
1178             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1179             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1180                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1181             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1182                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1183             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1184                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1185             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1186                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1187             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1188             Heps             = _mm256_mul_ps(vfeps,H);
1189             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1190             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1191             velec            = _mm256_mul_ps(qq12,VV);
1192             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1193             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
1194
1195             /* Update potential sum for this i atom from the interaction with this j atom. */
1196             velec            = _mm256_andnot_ps(dummy_mask,velec);
1197             velecsum         = _mm256_add_ps(velecsum,velec);
1198
1199             fscal            = felec;
1200
1201             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1202
1203             /* Calculate temporary vectorial force */
1204             tx               = _mm256_mul_ps(fscal,dx12);
1205             ty               = _mm256_mul_ps(fscal,dy12);
1206             tz               = _mm256_mul_ps(fscal,dz12);
1207
1208             /* Update vectorial force */
1209             fix1             = _mm256_add_ps(fix1,tx);
1210             fiy1             = _mm256_add_ps(fiy1,ty);
1211             fiz1             = _mm256_add_ps(fiz1,tz);
1212
1213             fjx2             = _mm256_add_ps(fjx2,tx);
1214             fjy2             = _mm256_add_ps(fjy2,ty);
1215             fjz2             = _mm256_add_ps(fjz2,tz);
1216
1217             /**************************
1218              * CALCULATE INTERACTIONS *
1219              **************************/
1220
1221             r20              = _mm256_mul_ps(rsq20,rinv20);
1222             r20              = _mm256_andnot_ps(dummy_mask,r20);
1223
1224             /* Calculate table index by multiplying r with table scale and truncate to integer */
1225             rt               = _mm256_mul_ps(r20,vftabscale);
1226             vfitab           = _mm256_cvttps_epi32(rt);
1227             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1228             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1229             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1230             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1231             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1232             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1233
1234             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1235             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1236                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1237             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1238                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1239             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1240                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1241             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1242                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1243             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1244             Heps             = _mm256_mul_ps(vfeps,H);
1245             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1246             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1247             velec            = _mm256_mul_ps(qq20,VV);
1248             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1249             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1250
1251             /* Update potential sum for this i atom from the interaction with this j atom. */
1252             velec            = _mm256_andnot_ps(dummy_mask,velec);
1253             velecsum         = _mm256_add_ps(velecsum,velec);
1254
1255             fscal            = felec;
1256
1257             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1258
1259             /* Calculate temporary vectorial force */
1260             tx               = _mm256_mul_ps(fscal,dx20);
1261             ty               = _mm256_mul_ps(fscal,dy20);
1262             tz               = _mm256_mul_ps(fscal,dz20);
1263
1264             /* Update vectorial force */
1265             fix2             = _mm256_add_ps(fix2,tx);
1266             fiy2             = _mm256_add_ps(fiy2,ty);
1267             fiz2             = _mm256_add_ps(fiz2,tz);
1268
1269             fjx0             = _mm256_add_ps(fjx0,tx);
1270             fjy0             = _mm256_add_ps(fjy0,ty);
1271             fjz0             = _mm256_add_ps(fjz0,tz);
1272
1273             /**************************
1274              * CALCULATE INTERACTIONS *
1275              **************************/
1276
1277             r21              = _mm256_mul_ps(rsq21,rinv21);
1278             r21              = _mm256_andnot_ps(dummy_mask,r21);
1279
1280             /* Calculate table index by multiplying r with table scale and truncate to integer */
1281             rt               = _mm256_mul_ps(r21,vftabscale);
1282             vfitab           = _mm256_cvttps_epi32(rt);
1283             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1284             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1285             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1286             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1287             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1288             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1289
1290             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1291             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1293             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1295             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1297             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1298                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1299             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1300             Heps             = _mm256_mul_ps(vfeps,H);
1301             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1302             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1303             velec            = _mm256_mul_ps(qq21,VV);
1304             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1305             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
1306
1307             /* Update potential sum for this i atom from the interaction with this j atom. */
1308             velec            = _mm256_andnot_ps(dummy_mask,velec);
1309             velecsum         = _mm256_add_ps(velecsum,velec);
1310
1311             fscal            = felec;
1312
1313             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1314
1315             /* Calculate temporary vectorial force */
1316             tx               = _mm256_mul_ps(fscal,dx21);
1317             ty               = _mm256_mul_ps(fscal,dy21);
1318             tz               = _mm256_mul_ps(fscal,dz21);
1319
1320             /* Update vectorial force */
1321             fix2             = _mm256_add_ps(fix2,tx);
1322             fiy2             = _mm256_add_ps(fiy2,ty);
1323             fiz2             = _mm256_add_ps(fiz2,tz);
1324
1325             fjx1             = _mm256_add_ps(fjx1,tx);
1326             fjy1             = _mm256_add_ps(fjy1,ty);
1327             fjz1             = _mm256_add_ps(fjz1,tz);
1328
1329             /**************************
1330              * CALCULATE INTERACTIONS *
1331              **************************/
1332
1333             r22              = _mm256_mul_ps(rsq22,rinv22);
1334             r22              = _mm256_andnot_ps(dummy_mask,r22);
1335
1336             /* Calculate table index by multiplying r with table scale and truncate to integer */
1337             rt               = _mm256_mul_ps(r22,vftabscale);
1338             vfitab           = _mm256_cvttps_epi32(rt);
1339             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1340             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1341             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1342             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1343             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1344             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1345
1346             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1347             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1348                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1349             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1350                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1351             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1352                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1353             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1354                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1355             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1356             Heps             = _mm256_mul_ps(vfeps,H);
1357             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1358             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
1359             velec            = _mm256_mul_ps(qq22,VV);
1360             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1361             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
1362
1363             /* Update potential sum for this i atom from the interaction with this j atom. */
1364             velec            = _mm256_andnot_ps(dummy_mask,velec);
1365             velecsum         = _mm256_add_ps(velecsum,velec);
1366
1367             fscal            = felec;
1368
1369             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1370
1371             /* Calculate temporary vectorial force */
1372             tx               = _mm256_mul_ps(fscal,dx22);
1373             ty               = _mm256_mul_ps(fscal,dy22);
1374             tz               = _mm256_mul_ps(fscal,dz22);
1375
1376             /* Update vectorial force */
1377             fix2             = _mm256_add_ps(fix2,tx);
1378             fiy2             = _mm256_add_ps(fiy2,ty);
1379             fiz2             = _mm256_add_ps(fiz2,tz);
1380
1381             fjx2             = _mm256_add_ps(fjx2,tx);
1382             fjy2             = _mm256_add_ps(fjy2,ty);
1383             fjz2             = _mm256_add_ps(fjz2,tz);
1384
1385             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1386             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1387             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1388             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1389             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1390             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1391             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1392             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1393
1394             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1395                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1396
1397             /* Inner loop uses 396 flops */
1398         }
1399
1400         /* End of innermost loop */
1401
1402         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1403                                                  f+i_coord_offset,fshift+i_shift_offset);
1404
1405         ggid                        = gid[iidx];
1406         /* Update potential energies */
1407         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1408
1409         /* Increment number of inner iterations */
1410         inneriter                  += j_index_end - j_index_start;
1411
1412         /* Outer loop uses 19 flops */
1413     }
1414
1415     /* Increment number of outer iterations */
1416     outeriter        += nri;
1417
1418     /* Update outer/inner flops */
1419
1420     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*19 + inneriter*396);
1421 }
1422 /*
1423  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3W3_F_avx_256_single
1424  * Electrostatics interaction: CubicSplineTable
1425  * VdW interaction:            None
1426  * Geometry:                   Water3-Water3
1427  * Calculate force/pot:        Force
1428  */
1429 void
1430 nb_kernel_ElecCSTab_VdwNone_GeomW3W3_F_avx_256_single
1431                     (t_nblist                    * gmx_restrict       nlist,
1432                      rvec                        * gmx_restrict          xx,
1433                      rvec                        * gmx_restrict          ff,
1434                      t_forcerec                  * gmx_restrict          fr,
1435                      t_mdatoms                   * gmx_restrict     mdatoms,
1436                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1437                      t_nrnb                      * gmx_restrict        nrnb)
1438 {
1439     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1440      * just 0 for non-waters.
1441      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
1442      * jnr indices corresponding to data put in the four positions in the SIMD register.
1443      */
1444     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1445     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1446     int              jnrA,jnrB,jnrC,jnrD;
1447     int              jnrE,jnrF,jnrG,jnrH;
1448     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1449     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
1450     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1451     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
1452     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1453     real             rcutoff_scalar;
1454     real             *shiftvec,*fshift,*x,*f;
1455     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
1456     real             scratch[4*DIM];
1457     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1458     real *           vdwioffsetptr0;
1459     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
1465     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
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     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1471     __m256           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1472     __m256           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1473     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1474     __m256           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1475     __m256           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1476     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1477     __m256           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1478     __m256           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1479     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
1480     real             *charge;
1481     __m256i          vfitab;
1482     __m128i          vfitab_lo,vfitab_hi;
1483     __m128i          ifour       = _mm_set1_epi32(4);
1484     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1485     real             *vftab;
1486     __m256           dummy_mask,cutoff_mask;
1487     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
1488     __m256           one     = _mm256_set1_ps(1.0);
1489     __m256           two     = _mm256_set1_ps(2.0);
1490     x                = xx[0];
1491     f                = ff[0];
1492
1493     nri              = nlist->nri;
1494     iinr             = nlist->iinr;
1495     jindex           = nlist->jindex;
1496     jjnr             = nlist->jjnr;
1497     shiftidx         = nlist->shift;
1498     gid              = nlist->gid;
1499     shiftvec         = fr->shift_vec[0];
1500     fshift           = fr->fshift[0];
1501     facel            = _mm256_set1_ps(fr->epsfac);
1502     charge           = mdatoms->chargeA;
1503
1504     vftab            = kernel_data->table_elec->data;
1505     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
1506
1507     /* Setup water-specific parameters */
1508     inr              = nlist->iinr[0];
1509     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
1510     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1511     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1512
1513     jq0              = _mm256_set1_ps(charge[inr+0]);
1514     jq1              = _mm256_set1_ps(charge[inr+1]);
1515     jq2              = _mm256_set1_ps(charge[inr+2]);
1516     qq00             = _mm256_mul_ps(iq0,jq0);
1517     qq01             = _mm256_mul_ps(iq0,jq1);
1518     qq02             = _mm256_mul_ps(iq0,jq2);
1519     qq10             = _mm256_mul_ps(iq1,jq0);
1520     qq11             = _mm256_mul_ps(iq1,jq1);
1521     qq12             = _mm256_mul_ps(iq1,jq2);
1522     qq20             = _mm256_mul_ps(iq2,jq0);
1523     qq21             = _mm256_mul_ps(iq2,jq1);
1524     qq22             = _mm256_mul_ps(iq2,jq2);
1525
1526     /* Avoid stupid compiler warnings */
1527     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1528     j_coord_offsetA = 0;
1529     j_coord_offsetB = 0;
1530     j_coord_offsetC = 0;
1531     j_coord_offsetD = 0;
1532     j_coord_offsetE = 0;
1533     j_coord_offsetF = 0;
1534     j_coord_offsetG = 0;
1535     j_coord_offsetH = 0;
1536
1537     outeriter        = 0;
1538     inneriter        = 0;
1539
1540     for(iidx=0;iidx<4*DIM;iidx++)
1541     {
1542         scratch[iidx] = 0.0;
1543     }
1544
1545     /* Start outer loop over neighborlists */
1546     for(iidx=0; iidx<nri; iidx++)
1547     {
1548         /* Load shift vector for this list */
1549         i_shift_offset   = DIM*shiftidx[iidx];
1550
1551         /* Load limits for loop over neighbors */
1552         j_index_start    = jindex[iidx];
1553         j_index_end      = jindex[iidx+1];
1554
1555         /* Get outer coordinate index */
1556         inr              = iinr[iidx];
1557         i_coord_offset   = DIM*inr;
1558
1559         /* Load i particle coords and add shift vector */
1560         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1561                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1562
1563         fix0             = _mm256_setzero_ps();
1564         fiy0             = _mm256_setzero_ps();
1565         fiz0             = _mm256_setzero_ps();
1566         fix1             = _mm256_setzero_ps();
1567         fiy1             = _mm256_setzero_ps();
1568         fiz1             = _mm256_setzero_ps();
1569         fix2             = _mm256_setzero_ps();
1570         fiy2             = _mm256_setzero_ps();
1571         fiz2             = _mm256_setzero_ps();
1572
1573         /* Start inner kernel loop */
1574         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1575         {
1576
1577             /* Get j neighbor index, and coordinate index */
1578             jnrA             = jjnr[jidx];
1579             jnrB             = jjnr[jidx+1];
1580             jnrC             = jjnr[jidx+2];
1581             jnrD             = jjnr[jidx+3];
1582             jnrE             = jjnr[jidx+4];
1583             jnrF             = jjnr[jidx+5];
1584             jnrG             = jjnr[jidx+6];
1585             jnrH             = jjnr[jidx+7];
1586             j_coord_offsetA  = DIM*jnrA;
1587             j_coord_offsetB  = DIM*jnrB;
1588             j_coord_offsetC  = DIM*jnrC;
1589             j_coord_offsetD  = DIM*jnrD;
1590             j_coord_offsetE  = DIM*jnrE;
1591             j_coord_offsetF  = DIM*jnrF;
1592             j_coord_offsetG  = DIM*jnrG;
1593             j_coord_offsetH  = DIM*jnrH;
1594
1595             /* load j atom coordinates */
1596             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1597                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1598                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1599                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1600                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1601
1602             /* Calculate displacement vector */
1603             dx00             = _mm256_sub_ps(ix0,jx0);
1604             dy00             = _mm256_sub_ps(iy0,jy0);
1605             dz00             = _mm256_sub_ps(iz0,jz0);
1606             dx01             = _mm256_sub_ps(ix0,jx1);
1607             dy01             = _mm256_sub_ps(iy0,jy1);
1608             dz01             = _mm256_sub_ps(iz0,jz1);
1609             dx02             = _mm256_sub_ps(ix0,jx2);
1610             dy02             = _mm256_sub_ps(iy0,jy2);
1611             dz02             = _mm256_sub_ps(iz0,jz2);
1612             dx10             = _mm256_sub_ps(ix1,jx0);
1613             dy10             = _mm256_sub_ps(iy1,jy0);
1614             dz10             = _mm256_sub_ps(iz1,jz0);
1615             dx11             = _mm256_sub_ps(ix1,jx1);
1616             dy11             = _mm256_sub_ps(iy1,jy1);
1617             dz11             = _mm256_sub_ps(iz1,jz1);
1618             dx12             = _mm256_sub_ps(ix1,jx2);
1619             dy12             = _mm256_sub_ps(iy1,jy2);
1620             dz12             = _mm256_sub_ps(iz1,jz2);
1621             dx20             = _mm256_sub_ps(ix2,jx0);
1622             dy20             = _mm256_sub_ps(iy2,jy0);
1623             dz20             = _mm256_sub_ps(iz2,jz0);
1624             dx21             = _mm256_sub_ps(ix2,jx1);
1625             dy21             = _mm256_sub_ps(iy2,jy1);
1626             dz21             = _mm256_sub_ps(iz2,jz1);
1627             dx22             = _mm256_sub_ps(ix2,jx2);
1628             dy22             = _mm256_sub_ps(iy2,jy2);
1629             dz22             = _mm256_sub_ps(iz2,jz2);
1630
1631             /* Calculate squared distance and things based on it */
1632             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1633             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
1634             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
1635             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1636             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
1637             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
1638             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1639             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
1640             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
1641
1642             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1643             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
1644             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
1645             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1646             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
1647             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
1648             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1649             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
1650             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
1651
1652             fjx0             = _mm256_setzero_ps();
1653             fjy0             = _mm256_setzero_ps();
1654             fjz0             = _mm256_setzero_ps();
1655             fjx1             = _mm256_setzero_ps();
1656             fjy1             = _mm256_setzero_ps();
1657             fjz1             = _mm256_setzero_ps();
1658             fjx2             = _mm256_setzero_ps();
1659             fjy2             = _mm256_setzero_ps();
1660             fjz2             = _mm256_setzero_ps();
1661
1662             /**************************
1663              * CALCULATE INTERACTIONS *
1664              **************************/
1665
1666             r00              = _mm256_mul_ps(rsq00,rinv00);
1667
1668             /* Calculate table index by multiplying r with table scale and truncate to integer */
1669             rt               = _mm256_mul_ps(r00,vftabscale);
1670             vfitab           = _mm256_cvttps_epi32(rt);
1671             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1672             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1673             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1674             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1675             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1676             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1677
1678             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1679             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1680                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1681             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1682                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1683             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1684                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1685             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1686                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1687             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1688             Heps             = _mm256_mul_ps(vfeps,H);
1689             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1690             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1691             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1692
1693             fscal            = felec;
1694
1695             /* Calculate temporary vectorial force */
1696             tx               = _mm256_mul_ps(fscal,dx00);
1697             ty               = _mm256_mul_ps(fscal,dy00);
1698             tz               = _mm256_mul_ps(fscal,dz00);
1699
1700             /* Update vectorial force */
1701             fix0             = _mm256_add_ps(fix0,tx);
1702             fiy0             = _mm256_add_ps(fiy0,ty);
1703             fiz0             = _mm256_add_ps(fiz0,tz);
1704
1705             fjx0             = _mm256_add_ps(fjx0,tx);
1706             fjy0             = _mm256_add_ps(fjy0,ty);
1707             fjz0             = _mm256_add_ps(fjz0,tz);
1708
1709             /**************************
1710              * CALCULATE INTERACTIONS *
1711              **************************/
1712
1713             r01              = _mm256_mul_ps(rsq01,rinv01);
1714
1715             /* Calculate table index by multiplying r with table scale and truncate to integer */
1716             rt               = _mm256_mul_ps(r01,vftabscale);
1717             vfitab           = _mm256_cvttps_epi32(rt);
1718             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1719             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1720             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1721             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1722             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1723             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1724
1725             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1726             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1727                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1728             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1729                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1730             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1731                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1732             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1733                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1734             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1735             Heps             = _mm256_mul_ps(vfeps,H);
1736             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1737             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1738             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
1739
1740             fscal            = felec;
1741
1742             /* Calculate temporary vectorial force */
1743             tx               = _mm256_mul_ps(fscal,dx01);
1744             ty               = _mm256_mul_ps(fscal,dy01);
1745             tz               = _mm256_mul_ps(fscal,dz01);
1746
1747             /* Update vectorial force */
1748             fix0             = _mm256_add_ps(fix0,tx);
1749             fiy0             = _mm256_add_ps(fiy0,ty);
1750             fiz0             = _mm256_add_ps(fiz0,tz);
1751
1752             fjx1             = _mm256_add_ps(fjx1,tx);
1753             fjy1             = _mm256_add_ps(fjy1,ty);
1754             fjz1             = _mm256_add_ps(fjz1,tz);
1755
1756             /**************************
1757              * CALCULATE INTERACTIONS *
1758              **************************/
1759
1760             r02              = _mm256_mul_ps(rsq02,rinv02);
1761
1762             /* Calculate table index by multiplying r with table scale and truncate to integer */
1763             rt               = _mm256_mul_ps(r02,vftabscale);
1764             vfitab           = _mm256_cvttps_epi32(rt);
1765             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1766             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1767             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1768             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1769             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1770             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1771
1772             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1773             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1774                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1775             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1776                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1777             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1778                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1779             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1780                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1781             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1782             Heps             = _mm256_mul_ps(vfeps,H);
1783             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1784             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1785             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
1786
1787             fscal            = felec;
1788
1789             /* Calculate temporary vectorial force */
1790             tx               = _mm256_mul_ps(fscal,dx02);
1791             ty               = _mm256_mul_ps(fscal,dy02);
1792             tz               = _mm256_mul_ps(fscal,dz02);
1793
1794             /* Update vectorial force */
1795             fix0             = _mm256_add_ps(fix0,tx);
1796             fiy0             = _mm256_add_ps(fiy0,ty);
1797             fiz0             = _mm256_add_ps(fiz0,tz);
1798
1799             fjx2             = _mm256_add_ps(fjx2,tx);
1800             fjy2             = _mm256_add_ps(fjy2,ty);
1801             fjz2             = _mm256_add_ps(fjz2,tz);
1802
1803             /**************************
1804              * CALCULATE INTERACTIONS *
1805              **************************/
1806
1807             r10              = _mm256_mul_ps(rsq10,rinv10);
1808
1809             /* Calculate table index by multiplying r with table scale and truncate to integer */
1810             rt               = _mm256_mul_ps(r10,vftabscale);
1811             vfitab           = _mm256_cvttps_epi32(rt);
1812             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1813             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1814             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1815             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1816             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1817             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1818
1819             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1820             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1821                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1822             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1823                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1824             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1825                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1826             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1827                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1828             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1829             Heps             = _mm256_mul_ps(vfeps,H);
1830             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1831             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1832             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1833
1834             fscal            = felec;
1835
1836             /* Calculate temporary vectorial force */
1837             tx               = _mm256_mul_ps(fscal,dx10);
1838             ty               = _mm256_mul_ps(fscal,dy10);
1839             tz               = _mm256_mul_ps(fscal,dz10);
1840
1841             /* Update vectorial force */
1842             fix1             = _mm256_add_ps(fix1,tx);
1843             fiy1             = _mm256_add_ps(fiy1,ty);
1844             fiz1             = _mm256_add_ps(fiz1,tz);
1845
1846             fjx0             = _mm256_add_ps(fjx0,tx);
1847             fjy0             = _mm256_add_ps(fjy0,ty);
1848             fjz0             = _mm256_add_ps(fjz0,tz);
1849
1850             /**************************
1851              * CALCULATE INTERACTIONS *
1852              **************************/
1853
1854             r11              = _mm256_mul_ps(rsq11,rinv11);
1855
1856             /* Calculate table index by multiplying r with table scale and truncate to integer */
1857             rt               = _mm256_mul_ps(r11,vftabscale);
1858             vfitab           = _mm256_cvttps_epi32(rt);
1859             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1860             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1861             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1862             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1863             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1864             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1865
1866             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1867             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1868                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1869             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1870                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1871             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1872                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1873             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1874                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1875             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1876             Heps             = _mm256_mul_ps(vfeps,H);
1877             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1878             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1879             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
1880
1881             fscal            = felec;
1882
1883             /* Calculate temporary vectorial force */
1884             tx               = _mm256_mul_ps(fscal,dx11);
1885             ty               = _mm256_mul_ps(fscal,dy11);
1886             tz               = _mm256_mul_ps(fscal,dz11);
1887
1888             /* Update vectorial force */
1889             fix1             = _mm256_add_ps(fix1,tx);
1890             fiy1             = _mm256_add_ps(fiy1,ty);
1891             fiz1             = _mm256_add_ps(fiz1,tz);
1892
1893             fjx1             = _mm256_add_ps(fjx1,tx);
1894             fjy1             = _mm256_add_ps(fjy1,ty);
1895             fjz1             = _mm256_add_ps(fjz1,tz);
1896
1897             /**************************
1898              * CALCULATE INTERACTIONS *
1899              **************************/
1900
1901             r12              = _mm256_mul_ps(rsq12,rinv12);
1902
1903             /* Calculate table index by multiplying r with table scale and truncate to integer */
1904             rt               = _mm256_mul_ps(r12,vftabscale);
1905             vfitab           = _mm256_cvttps_epi32(rt);
1906             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1907             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1908             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1909             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1910             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1911             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1912
1913             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1914             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1915                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1916             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1917                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1918             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1919                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1920             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1921                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1922             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1923             Heps             = _mm256_mul_ps(vfeps,H);
1924             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1925             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1926             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
1927
1928             fscal            = felec;
1929
1930             /* Calculate temporary vectorial force */
1931             tx               = _mm256_mul_ps(fscal,dx12);
1932             ty               = _mm256_mul_ps(fscal,dy12);
1933             tz               = _mm256_mul_ps(fscal,dz12);
1934
1935             /* Update vectorial force */
1936             fix1             = _mm256_add_ps(fix1,tx);
1937             fiy1             = _mm256_add_ps(fiy1,ty);
1938             fiz1             = _mm256_add_ps(fiz1,tz);
1939
1940             fjx2             = _mm256_add_ps(fjx2,tx);
1941             fjy2             = _mm256_add_ps(fjy2,ty);
1942             fjz2             = _mm256_add_ps(fjz2,tz);
1943
1944             /**************************
1945              * CALCULATE INTERACTIONS *
1946              **************************/
1947
1948             r20              = _mm256_mul_ps(rsq20,rinv20);
1949
1950             /* Calculate table index by multiplying r with table scale and truncate to integer */
1951             rt               = _mm256_mul_ps(r20,vftabscale);
1952             vfitab           = _mm256_cvttps_epi32(rt);
1953             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1954             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1955             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1956             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1957             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
1958             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
1959
1960             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1961             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1962                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1963             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1964                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1965             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1966                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1967             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1968                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1969             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1970             Heps             = _mm256_mul_ps(vfeps,H);
1971             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1972             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1973             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1974
1975             fscal            = felec;
1976
1977             /* Calculate temporary vectorial force */
1978             tx               = _mm256_mul_ps(fscal,dx20);
1979             ty               = _mm256_mul_ps(fscal,dy20);
1980             tz               = _mm256_mul_ps(fscal,dz20);
1981
1982             /* Update vectorial force */
1983             fix2             = _mm256_add_ps(fix2,tx);
1984             fiy2             = _mm256_add_ps(fiy2,ty);
1985             fiz2             = _mm256_add_ps(fiz2,tz);
1986
1987             fjx0             = _mm256_add_ps(fjx0,tx);
1988             fjy0             = _mm256_add_ps(fjy0,ty);
1989             fjz0             = _mm256_add_ps(fjz0,tz);
1990
1991             /**************************
1992              * CALCULATE INTERACTIONS *
1993              **************************/
1994
1995             r21              = _mm256_mul_ps(rsq21,rinv21);
1996
1997             /* Calculate table index by multiplying r with table scale and truncate to integer */
1998             rt               = _mm256_mul_ps(r21,vftabscale);
1999             vfitab           = _mm256_cvttps_epi32(rt);
2000             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2001             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2002             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2003             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2004             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2005             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2006
2007             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2008             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2009                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2010             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2011                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2012             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2013                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2014             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2015                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2016             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2017             Heps             = _mm256_mul_ps(vfeps,H);
2018             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2019             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2020             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
2021
2022             fscal            = felec;
2023
2024             /* Calculate temporary vectorial force */
2025             tx               = _mm256_mul_ps(fscal,dx21);
2026             ty               = _mm256_mul_ps(fscal,dy21);
2027             tz               = _mm256_mul_ps(fscal,dz21);
2028
2029             /* Update vectorial force */
2030             fix2             = _mm256_add_ps(fix2,tx);
2031             fiy2             = _mm256_add_ps(fiy2,ty);
2032             fiz2             = _mm256_add_ps(fiz2,tz);
2033
2034             fjx1             = _mm256_add_ps(fjx1,tx);
2035             fjy1             = _mm256_add_ps(fjy1,ty);
2036             fjz1             = _mm256_add_ps(fjz1,tz);
2037
2038             /**************************
2039              * CALCULATE INTERACTIONS *
2040              **************************/
2041
2042             r22              = _mm256_mul_ps(rsq22,rinv22);
2043
2044             /* Calculate table index by multiplying r with table scale and truncate to integer */
2045             rt               = _mm256_mul_ps(r22,vftabscale);
2046             vfitab           = _mm256_cvttps_epi32(rt);
2047             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2048             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2049             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2050             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2051             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2052             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2053
2054             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2055             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2056                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2057             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2058                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2059             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2060                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2061             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2062                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2063             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2064             Heps             = _mm256_mul_ps(vfeps,H);
2065             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2066             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2067             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
2068
2069             fscal            = felec;
2070
2071             /* Calculate temporary vectorial force */
2072             tx               = _mm256_mul_ps(fscal,dx22);
2073             ty               = _mm256_mul_ps(fscal,dy22);
2074             tz               = _mm256_mul_ps(fscal,dz22);
2075
2076             /* Update vectorial force */
2077             fix2             = _mm256_add_ps(fix2,tx);
2078             fiy2             = _mm256_add_ps(fiy2,ty);
2079             fiz2             = _mm256_add_ps(fiz2,tz);
2080
2081             fjx2             = _mm256_add_ps(fjx2,tx);
2082             fjy2             = _mm256_add_ps(fjy2,ty);
2083             fjz2             = _mm256_add_ps(fjz2,tz);
2084
2085             fjptrA             = f+j_coord_offsetA;
2086             fjptrB             = f+j_coord_offsetB;
2087             fjptrC             = f+j_coord_offsetC;
2088             fjptrD             = f+j_coord_offsetD;
2089             fjptrE             = f+j_coord_offsetE;
2090             fjptrF             = f+j_coord_offsetF;
2091             fjptrG             = f+j_coord_offsetG;
2092             fjptrH             = f+j_coord_offsetH;
2093
2094             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2095                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2096
2097             /* Inner loop uses 351 flops */
2098         }
2099
2100         if(jidx<j_index_end)
2101         {
2102
2103             /* Get j neighbor index, and coordinate index */
2104             jnrlistA         = jjnr[jidx];
2105             jnrlistB         = jjnr[jidx+1];
2106             jnrlistC         = jjnr[jidx+2];
2107             jnrlistD         = jjnr[jidx+3];
2108             jnrlistE         = jjnr[jidx+4];
2109             jnrlistF         = jjnr[jidx+5];
2110             jnrlistG         = jjnr[jidx+6];
2111             jnrlistH         = jjnr[jidx+7];
2112             /* Sign of each element will be negative for non-real atoms.
2113              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2114              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2115              */
2116             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
2117                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
2118
2119             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2120             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2121             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2122             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2123             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
2124             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
2125             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
2126             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
2127             j_coord_offsetA  = DIM*jnrA;
2128             j_coord_offsetB  = DIM*jnrB;
2129             j_coord_offsetC  = DIM*jnrC;
2130             j_coord_offsetD  = DIM*jnrD;
2131             j_coord_offsetE  = DIM*jnrE;
2132             j_coord_offsetF  = DIM*jnrF;
2133             j_coord_offsetG  = DIM*jnrG;
2134             j_coord_offsetH  = DIM*jnrH;
2135
2136             /* load j atom coordinates */
2137             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2138                                                  x+j_coord_offsetC,x+j_coord_offsetD,
2139                                                  x+j_coord_offsetE,x+j_coord_offsetF,
2140                                                  x+j_coord_offsetG,x+j_coord_offsetH,
2141                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2142
2143             /* Calculate displacement vector */
2144             dx00             = _mm256_sub_ps(ix0,jx0);
2145             dy00             = _mm256_sub_ps(iy0,jy0);
2146             dz00             = _mm256_sub_ps(iz0,jz0);
2147             dx01             = _mm256_sub_ps(ix0,jx1);
2148             dy01             = _mm256_sub_ps(iy0,jy1);
2149             dz01             = _mm256_sub_ps(iz0,jz1);
2150             dx02             = _mm256_sub_ps(ix0,jx2);
2151             dy02             = _mm256_sub_ps(iy0,jy2);
2152             dz02             = _mm256_sub_ps(iz0,jz2);
2153             dx10             = _mm256_sub_ps(ix1,jx0);
2154             dy10             = _mm256_sub_ps(iy1,jy0);
2155             dz10             = _mm256_sub_ps(iz1,jz0);
2156             dx11             = _mm256_sub_ps(ix1,jx1);
2157             dy11             = _mm256_sub_ps(iy1,jy1);
2158             dz11             = _mm256_sub_ps(iz1,jz1);
2159             dx12             = _mm256_sub_ps(ix1,jx2);
2160             dy12             = _mm256_sub_ps(iy1,jy2);
2161             dz12             = _mm256_sub_ps(iz1,jz2);
2162             dx20             = _mm256_sub_ps(ix2,jx0);
2163             dy20             = _mm256_sub_ps(iy2,jy0);
2164             dz20             = _mm256_sub_ps(iz2,jz0);
2165             dx21             = _mm256_sub_ps(ix2,jx1);
2166             dy21             = _mm256_sub_ps(iy2,jy1);
2167             dz21             = _mm256_sub_ps(iz2,jz1);
2168             dx22             = _mm256_sub_ps(ix2,jx2);
2169             dy22             = _mm256_sub_ps(iy2,jy2);
2170             dz22             = _mm256_sub_ps(iz2,jz2);
2171
2172             /* Calculate squared distance and things based on it */
2173             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
2174             rsq01            = gmx_mm256_calc_rsq_ps(dx01,dy01,dz01);
2175             rsq02            = gmx_mm256_calc_rsq_ps(dx02,dy02,dz02);
2176             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
2177             rsq11            = gmx_mm256_calc_rsq_ps(dx11,dy11,dz11);
2178             rsq12            = gmx_mm256_calc_rsq_ps(dx12,dy12,dz12);
2179             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
2180             rsq21            = gmx_mm256_calc_rsq_ps(dx21,dy21,dz21);
2181             rsq22            = gmx_mm256_calc_rsq_ps(dx22,dy22,dz22);
2182
2183             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
2184             rinv01           = gmx_mm256_invsqrt_ps(rsq01);
2185             rinv02           = gmx_mm256_invsqrt_ps(rsq02);
2186             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
2187             rinv11           = gmx_mm256_invsqrt_ps(rsq11);
2188             rinv12           = gmx_mm256_invsqrt_ps(rsq12);
2189             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
2190             rinv21           = gmx_mm256_invsqrt_ps(rsq21);
2191             rinv22           = gmx_mm256_invsqrt_ps(rsq22);
2192
2193             fjx0             = _mm256_setzero_ps();
2194             fjy0             = _mm256_setzero_ps();
2195             fjz0             = _mm256_setzero_ps();
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
2203             /**************************
2204              * CALCULATE INTERACTIONS *
2205              **************************/
2206
2207             r00              = _mm256_mul_ps(rsq00,rinv00);
2208             r00              = _mm256_andnot_ps(dummy_mask,r00);
2209
2210             /* Calculate table index by multiplying r with table scale and truncate to integer */
2211             rt               = _mm256_mul_ps(r00,vftabscale);
2212             vfitab           = _mm256_cvttps_epi32(rt);
2213             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2214             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2215             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2216             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2217             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2218             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2219
2220             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2221             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2222                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2223             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2224                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2225             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2226                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2227             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2228                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2229             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2230             Heps             = _mm256_mul_ps(vfeps,H);
2231             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2232             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2233             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
2234
2235             fscal            = felec;
2236
2237             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2238
2239             /* Calculate temporary vectorial force */
2240             tx               = _mm256_mul_ps(fscal,dx00);
2241             ty               = _mm256_mul_ps(fscal,dy00);
2242             tz               = _mm256_mul_ps(fscal,dz00);
2243
2244             /* Update vectorial force */
2245             fix0             = _mm256_add_ps(fix0,tx);
2246             fiy0             = _mm256_add_ps(fiy0,ty);
2247             fiz0             = _mm256_add_ps(fiz0,tz);
2248
2249             fjx0             = _mm256_add_ps(fjx0,tx);
2250             fjy0             = _mm256_add_ps(fjy0,ty);
2251             fjz0             = _mm256_add_ps(fjz0,tz);
2252
2253             /**************************
2254              * CALCULATE INTERACTIONS *
2255              **************************/
2256
2257             r01              = _mm256_mul_ps(rsq01,rinv01);
2258             r01              = _mm256_andnot_ps(dummy_mask,r01);
2259
2260             /* Calculate table index by multiplying r with table scale and truncate to integer */
2261             rt               = _mm256_mul_ps(r01,vftabscale);
2262             vfitab           = _mm256_cvttps_epi32(rt);
2263             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2264             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2265             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2266             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2267             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2268             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2269
2270             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2271             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2272                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2273             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2274                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2275             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2276                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2277             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2278                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2279             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2280             Heps             = _mm256_mul_ps(vfeps,H);
2281             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2282             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2283             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq01,FF),_mm256_mul_ps(vftabscale,rinv01)));
2284
2285             fscal            = felec;
2286
2287             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2288
2289             /* Calculate temporary vectorial force */
2290             tx               = _mm256_mul_ps(fscal,dx01);
2291             ty               = _mm256_mul_ps(fscal,dy01);
2292             tz               = _mm256_mul_ps(fscal,dz01);
2293
2294             /* Update vectorial force */
2295             fix0             = _mm256_add_ps(fix0,tx);
2296             fiy0             = _mm256_add_ps(fiy0,ty);
2297             fiz0             = _mm256_add_ps(fiz0,tz);
2298
2299             fjx1             = _mm256_add_ps(fjx1,tx);
2300             fjy1             = _mm256_add_ps(fjy1,ty);
2301             fjz1             = _mm256_add_ps(fjz1,tz);
2302
2303             /**************************
2304              * CALCULATE INTERACTIONS *
2305              **************************/
2306
2307             r02              = _mm256_mul_ps(rsq02,rinv02);
2308             r02              = _mm256_andnot_ps(dummy_mask,r02);
2309
2310             /* Calculate table index by multiplying r with table scale and truncate to integer */
2311             rt               = _mm256_mul_ps(r02,vftabscale);
2312             vfitab           = _mm256_cvttps_epi32(rt);
2313             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2314             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2315             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2316             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2317             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2318             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2319
2320             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2321             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2322                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2323             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2324                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2325             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2327             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2329             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2330             Heps             = _mm256_mul_ps(vfeps,H);
2331             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2332             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2333             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq02,FF),_mm256_mul_ps(vftabscale,rinv02)));
2334
2335             fscal            = felec;
2336
2337             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2338
2339             /* Calculate temporary vectorial force */
2340             tx               = _mm256_mul_ps(fscal,dx02);
2341             ty               = _mm256_mul_ps(fscal,dy02);
2342             tz               = _mm256_mul_ps(fscal,dz02);
2343
2344             /* Update vectorial force */
2345             fix0             = _mm256_add_ps(fix0,tx);
2346             fiy0             = _mm256_add_ps(fiy0,ty);
2347             fiz0             = _mm256_add_ps(fiz0,tz);
2348
2349             fjx2             = _mm256_add_ps(fjx2,tx);
2350             fjy2             = _mm256_add_ps(fjy2,ty);
2351             fjz2             = _mm256_add_ps(fjz2,tz);
2352
2353             /**************************
2354              * CALCULATE INTERACTIONS *
2355              **************************/
2356
2357             r10              = _mm256_mul_ps(rsq10,rinv10);
2358             r10              = _mm256_andnot_ps(dummy_mask,r10);
2359
2360             /* Calculate table index by multiplying r with table scale and truncate to integer */
2361             rt               = _mm256_mul_ps(r10,vftabscale);
2362             vfitab           = _mm256_cvttps_epi32(rt);
2363             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2364             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2365             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2366             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2367             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2368             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2369
2370             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2371             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2372                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2373             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2374                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2375             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2376                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2377             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2378                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2379             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2380             Heps             = _mm256_mul_ps(vfeps,H);
2381             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2382             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2383             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
2384
2385             fscal            = felec;
2386
2387             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2388
2389             /* Calculate temporary vectorial force */
2390             tx               = _mm256_mul_ps(fscal,dx10);
2391             ty               = _mm256_mul_ps(fscal,dy10);
2392             tz               = _mm256_mul_ps(fscal,dz10);
2393
2394             /* Update vectorial force */
2395             fix1             = _mm256_add_ps(fix1,tx);
2396             fiy1             = _mm256_add_ps(fiy1,ty);
2397             fiz1             = _mm256_add_ps(fiz1,tz);
2398
2399             fjx0             = _mm256_add_ps(fjx0,tx);
2400             fjy0             = _mm256_add_ps(fjy0,ty);
2401             fjz0             = _mm256_add_ps(fjz0,tz);
2402
2403             /**************************
2404              * CALCULATE INTERACTIONS *
2405              **************************/
2406
2407             r11              = _mm256_mul_ps(rsq11,rinv11);
2408             r11              = _mm256_andnot_ps(dummy_mask,r11);
2409
2410             /* Calculate table index by multiplying r with table scale and truncate to integer */
2411             rt               = _mm256_mul_ps(r11,vftabscale);
2412             vfitab           = _mm256_cvttps_epi32(rt);
2413             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2414             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2415             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2416             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2417             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2418             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2419
2420             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2421             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2422                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2423             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2424                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2425             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2426                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2427             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2428                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2429             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2430             Heps             = _mm256_mul_ps(vfeps,H);
2431             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2432             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2433             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq11,FF),_mm256_mul_ps(vftabscale,rinv11)));
2434
2435             fscal            = felec;
2436
2437             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2438
2439             /* Calculate temporary vectorial force */
2440             tx               = _mm256_mul_ps(fscal,dx11);
2441             ty               = _mm256_mul_ps(fscal,dy11);
2442             tz               = _mm256_mul_ps(fscal,dz11);
2443
2444             /* Update vectorial force */
2445             fix1             = _mm256_add_ps(fix1,tx);
2446             fiy1             = _mm256_add_ps(fiy1,ty);
2447             fiz1             = _mm256_add_ps(fiz1,tz);
2448
2449             fjx1             = _mm256_add_ps(fjx1,tx);
2450             fjy1             = _mm256_add_ps(fjy1,ty);
2451             fjz1             = _mm256_add_ps(fjz1,tz);
2452
2453             /**************************
2454              * CALCULATE INTERACTIONS *
2455              **************************/
2456
2457             r12              = _mm256_mul_ps(rsq12,rinv12);
2458             r12              = _mm256_andnot_ps(dummy_mask,r12);
2459
2460             /* Calculate table index by multiplying r with table scale and truncate to integer */
2461             rt               = _mm256_mul_ps(r12,vftabscale);
2462             vfitab           = _mm256_cvttps_epi32(rt);
2463             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2464             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2465             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2466             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2467             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2468             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2469
2470             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2471             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2472                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2473             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2474                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2475             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2476                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2477             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2478                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2479             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2480             Heps             = _mm256_mul_ps(vfeps,H);
2481             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2482             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2483             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq12,FF),_mm256_mul_ps(vftabscale,rinv12)));
2484
2485             fscal            = felec;
2486
2487             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2488
2489             /* Calculate temporary vectorial force */
2490             tx               = _mm256_mul_ps(fscal,dx12);
2491             ty               = _mm256_mul_ps(fscal,dy12);
2492             tz               = _mm256_mul_ps(fscal,dz12);
2493
2494             /* Update vectorial force */
2495             fix1             = _mm256_add_ps(fix1,tx);
2496             fiy1             = _mm256_add_ps(fiy1,ty);
2497             fiz1             = _mm256_add_ps(fiz1,tz);
2498
2499             fjx2             = _mm256_add_ps(fjx2,tx);
2500             fjy2             = _mm256_add_ps(fjy2,ty);
2501             fjz2             = _mm256_add_ps(fjz2,tz);
2502
2503             /**************************
2504              * CALCULATE INTERACTIONS *
2505              **************************/
2506
2507             r20              = _mm256_mul_ps(rsq20,rinv20);
2508             r20              = _mm256_andnot_ps(dummy_mask,r20);
2509
2510             /* Calculate table index by multiplying r with table scale and truncate to integer */
2511             rt               = _mm256_mul_ps(r20,vftabscale);
2512             vfitab           = _mm256_cvttps_epi32(rt);
2513             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2514             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2515             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2516             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2517             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2518             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2519
2520             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2521             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2522                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2523             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2524                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2525             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2526                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2527             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2528                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2529             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2530             Heps             = _mm256_mul_ps(vfeps,H);
2531             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2532             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2533             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
2534
2535             fscal            = felec;
2536
2537             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2538
2539             /* Calculate temporary vectorial force */
2540             tx               = _mm256_mul_ps(fscal,dx20);
2541             ty               = _mm256_mul_ps(fscal,dy20);
2542             tz               = _mm256_mul_ps(fscal,dz20);
2543
2544             /* Update vectorial force */
2545             fix2             = _mm256_add_ps(fix2,tx);
2546             fiy2             = _mm256_add_ps(fiy2,ty);
2547             fiz2             = _mm256_add_ps(fiz2,tz);
2548
2549             fjx0             = _mm256_add_ps(fjx0,tx);
2550             fjy0             = _mm256_add_ps(fjy0,ty);
2551             fjz0             = _mm256_add_ps(fjz0,tz);
2552
2553             /**************************
2554              * CALCULATE INTERACTIONS *
2555              **************************/
2556
2557             r21              = _mm256_mul_ps(rsq21,rinv21);
2558             r21              = _mm256_andnot_ps(dummy_mask,r21);
2559
2560             /* Calculate table index by multiplying r with table scale and truncate to integer */
2561             rt               = _mm256_mul_ps(r21,vftabscale);
2562             vfitab           = _mm256_cvttps_epi32(rt);
2563             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2564             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2565             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2566             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2567             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2568             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2569
2570             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2571             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2572                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2573             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2574                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2575             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2576                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2577             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2578                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2579             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2580             Heps             = _mm256_mul_ps(vfeps,H);
2581             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2582             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2583             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq21,FF),_mm256_mul_ps(vftabscale,rinv21)));
2584
2585             fscal            = felec;
2586
2587             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2588
2589             /* Calculate temporary vectorial force */
2590             tx               = _mm256_mul_ps(fscal,dx21);
2591             ty               = _mm256_mul_ps(fscal,dy21);
2592             tz               = _mm256_mul_ps(fscal,dz21);
2593
2594             /* Update vectorial force */
2595             fix2             = _mm256_add_ps(fix2,tx);
2596             fiy2             = _mm256_add_ps(fiy2,ty);
2597             fiz2             = _mm256_add_ps(fiz2,tz);
2598
2599             fjx1             = _mm256_add_ps(fjx1,tx);
2600             fjy1             = _mm256_add_ps(fjy1,ty);
2601             fjz1             = _mm256_add_ps(fjz1,tz);
2602
2603             /**************************
2604              * CALCULATE INTERACTIONS *
2605              **************************/
2606
2607             r22              = _mm256_mul_ps(rsq22,rinv22);
2608             r22              = _mm256_andnot_ps(dummy_mask,r22);
2609
2610             /* Calculate table index by multiplying r with table scale and truncate to integer */
2611             rt               = _mm256_mul_ps(r22,vftabscale);
2612             vfitab           = _mm256_cvttps_epi32(rt);
2613             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
2614             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
2615             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
2616             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
2617             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
2618             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
2619
2620             /* CUBIC SPLINE TABLE ELECTROSTATICS */
2621             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
2622                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
2623             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
2624                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
2625             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
2626                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
2627             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
2628                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
2629             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
2630             Heps             = _mm256_mul_ps(vfeps,H);
2631             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
2632             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
2633             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq22,FF),_mm256_mul_ps(vftabscale,rinv22)));
2634
2635             fscal            = felec;
2636
2637             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
2638
2639             /* Calculate temporary vectorial force */
2640             tx               = _mm256_mul_ps(fscal,dx22);
2641             ty               = _mm256_mul_ps(fscal,dy22);
2642             tz               = _mm256_mul_ps(fscal,dz22);
2643
2644             /* Update vectorial force */
2645             fix2             = _mm256_add_ps(fix2,tx);
2646             fiy2             = _mm256_add_ps(fiy2,ty);
2647             fiz2             = _mm256_add_ps(fiz2,tz);
2648
2649             fjx2             = _mm256_add_ps(fjx2,tx);
2650             fjy2             = _mm256_add_ps(fjy2,ty);
2651             fjz2             = _mm256_add_ps(fjz2,tz);
2652
2653             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2654             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2655             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2656             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2657             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
2658             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
2659             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
2660             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
2661
2662             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
2663                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2664
2665             /* Inner loop uses 360 flops */
2666         }
2667
2668         /* End of innermost loop */
2669
2670         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2671                                                  f+i_coord_offset,fshift+i_shift_offset);
2672
2673         /* Increment number of inner iterations */
2674         inneriter                  += j_index_end - j_index_start;
2675
2676         /* Outer loop uses 18 flops */
2677     }
2678
2679     /* Increment number of outer iterations */
2680     outeriter        += nri;
2681
2682     /* Update outer/inner flops */
2683
2684     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*18 + inneriter*360);
2685 }