src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  93     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 104     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 105     __m256i          vfitab;
 106     __m128i          vfitab_lo,vfitab_hi;
 107     __m128i          ifour       = _mm_set1_epi32(4);
 108     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 109     real             *vftab;
 110     __m256           dummy_mask,cutoff_mask;
 111     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 112     __m256           one     = _mm256_set1_ps(1.0);
 113     __m256           two     = _mm256_set1_ps(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_ps(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     vftab            = kernel_data->table_elec_vdw->data;
 132     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 137     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 138     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 139     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147     j_coord_offsetE = 0;
 148     j_coord_offsetF = 0;
 149     j_coord_offsetG = 0;
 150     j_coord_offsetH = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm256_setzero_ps();
 179         fiy0             = _mm256_setzero_ps();
 180         fiz0             = _mm256_setzero_ps();
 181         fix1             = _mm256_setzero_ps();
 182         fiy1             = _mm256_setzero_ps();
 183         fiz1             = _mm256_setzero_ps();
 184         fix2             = _mm256_setzero_ps();
 185         fiy2             = _mm256_setzero_ps();
 186         fiz2             = _mm256_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm256_setzero_ps();
 190         vvdwsum          = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225             dx10             = _mm256_sub_ps(ix1,jx0);
 226             dy10             = _mm256_sub_ps(iy1,jy0);
 227             dz10             = _mm256_sub_ps(iz1,jz0);
 228             dx20             = _mm256_sub_ps(ix2,jx0);
 229             dy20             = _mm256_sub_ps(iy2,jy0);
 230             dz20             = _mm256_sub_ps(iz2,jz0);
 231
 232             /* Calculate squared distance and things based on it */
 233             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 234             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 235             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 236
 237             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 238             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 239             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 243                                                                  charge+jnrC+0,charge+jnrD+0,
 244                                                                  charge+jnrE+0,charge+jnrF+0,
 245                                                                  charge+jnrG+0,charge+jnrH+0);
 246             vdwjidx0A        = 2*vdwtype[jnrA+0];
 247             vdwjidx0B        = 2*vdwtype[jnrB+0];
 248             vdwjidx0C        = 2*vdwtype[jnrC+0];
 249             vdwjidx0D        = 2*vdwtype[jnrD+0];
 250             vdwjidx0E        = 2*vdwtype[jnrE+0];
 251             vdwjidx0F        = 2*vdwtype[jnrF+0];
 252             vdwjidx0G        = 2*vdwtype[jnrG+0];
 253             vdwjidx0H        = 2*vdwtype[jnrH+0];
 254
 255             fjx0             = _mm256_setzero_ps();
 256             fjy0             = _mm256_setzero_ps();
 257             fjz0             = _mm256_setzero_ps();
 258
 259             /**************************
 260              * CALCULATE INTERACTIONS *
 261              **************************/
 262
 263             r00              = _mm256_mul_ps(rsq00,rinv00);
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             qq00             = _mm256_mul_ps(iq0,jq0);
 267             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 268                                             vdwioffsetptr0+vdwjidx0B,
 269                                             vdwioffsetptr0+vdwjidx0C,
 270                                             vdwioffsetptr0+vdwjidx0D,
 271                                             vdwioffsetptr0+vdwjidx0E,
 272                                             vdwioffsetptr0+vdwjidx0F,
 273                                             vdwioffsetptr0+vdwjidx0G,
 274                                             vdwioffsetptr0+vdwjidx0H,
 275                                             &c6_00,&c12_00);
 276
 277             /* Calculate table index by multiplying r with table scale and truncate to integer */
 278             rt               = _mm256_mul_ps(r00,vftabscale);
 279             vfitab           = _mm256_cvttps_epi32(rt);
 280             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 281             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 282             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 283             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 284             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 285             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 286
 287             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 288             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 289                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 290             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 291                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 292             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 293                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 294             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 295                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 296             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 297             Heps             = _mm256_mul_ps(vfeps,H);
 298             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 299             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 300             velec            = _mm256_mul_ps(qq00,VV);
 301             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 302             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 303
 304             /* CUBIC SPLINE TABLE DISPERSION */
 305             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 306             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 307             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 308                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 309             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 311             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 313             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 314                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 315             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 316             Heps             = _mm256_mul_ps(vfeps,H);
 317             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 318             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 319             vvdw6            = _mm256_mul_ps(c6_00,VV);
 320             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 321             fvdw6            = _mm256_mul_ps(c6_00,FF);
 322
 323             /* CUBIC SPLINE TABLE REPULSION */
 324             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 325             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 326             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 327                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 328             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 329                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 330             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 331                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 332             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 333                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 334             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 335             Heps             = _mm256_mul_ps(vfeps,H);
 336             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 337             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 338             vvdw12           = _mm256_mul_ps(c12_00,VV);
 339             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 340             fvdw12           = _mm256_mul_ps(c12_00,FF);
 341             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 342             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm256_add_ps(velecsum,velec);
 346             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 347
 348             fscal            = _mm256_add_ps(felec,fvdw);
 349
 350             /* Calculate temporary vectorial force */
 351             tx               = _mm256_mul_ps(fscal,dx00);
 352             ty               = _mm256_mul_ps(fscal,dy00);
 353             tz               = _mm256_mul_ps(fscal,dz00);
 354
 355             /* Update vectorial force */
 356             fix0             = _mm256_add_ps(fix0,tx);
 357             fiy0             = _mm256_add_ps(fiy0,ty);
 358             fiz0             = _mm256_add_ps(fiz0,tz);
 359
 360             fjx0             = _mm256_add_ps(fjx0,tx);
 361             fjy0             = _mm256_add_ps(fjy0,ty);
 362             fjz0             = _mm256_add_ps(fjz0,tz);
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             r10              = _mm256_mul_ps(rsq10,rinv10);
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq10             = _mm256_mul_ps(iq1,jq0);
 372
 373             /* Calculate table index by multiplying r with table scale and truncate to integer */
 374             rt               = _mm256_mul_ps(r10,vftabscale);
 375             vfitab           = _mm256_cvttps_epi32(rt);
 376             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 377             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 378             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 379             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 380             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 381             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 382
 383             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 384             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 385                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 386             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 387                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 388             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 389                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 390             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 391                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 392             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 393             Heps             = _mm256_mul_ps(vfeps,H);
 394             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 395             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 396             velec            = _mm256_mul_ps(qq10,VV);
 397             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 398             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velecsum         = _mm256_add_ps(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405             /* Calculate temporary vectorial force */
 406             tx               = _mm256_mul_ps(fscal,dx10);
 407             ty               = _mm256_mul_ps(fscal,dy10);
 408             tz               = _mm256_mul_ps(fscal,dz10);
 409
 410             /* Update vectorial force */
 411             fix1             = _mm256_add_ps(fix1,tx);
 412             fiy1             = _mm256_add_ps(fiy1,ty);
 413             fiz1             = _mm256_add_ps(fiz1,tz);
 414
 415             fjx0             = _mm256_add_ps(fjx0,tx);
 416             fjy0             = _mm256_add_ps(fjy0,ty);
 417             fjz0             = _mm256_add_ps(fjz0,tz);
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             r20              = _mm256_mul_ps(rsq20,rinv20);
 424
 425             /* Compute parameters for interactions between i and j atoms */
 426             qq20             = _mm256_mul_ps(iq2,jq0);
 427
 428             /* Calculate table index by multiplying r with table scale and truncate to integer */
 429             rt               = _mm256_mul_ps(r20,vftabscale);
 430             vfitab           = _mm256_cvttps_epi32(rt);
 431             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 432             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 433             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 434             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 435             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 436             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 437
 438             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 439             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 440                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 441             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 442                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 443             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 444                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 445             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 446                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 447             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 448             Heps             = _mm256_mul_ps(vfeps,H);
 449             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 450             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 451             velec            = _mm256_mul_ps(qq20,VV);
 452             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 453             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 454
 455             /* Update potential sum for this i atom from the interaction with this j atom. */
 456             velecsum         = _mm256_add_ps(velecsum,velec);
 457
 458             fscal            = felec;
 459
 460             /* Calculate temporary vectorial force */
 461             tx               = _mm256_mul_ps(fscal,dx20);
 462             ty               = _mm256_mul_ps(fscal,dy20);
 463             tz               = _mm256_mul_ps(fscal,dz20);
 464
 465             /* Update vectorial force */
 466             fix2             = _mm256_add_ps(fix2,tx);
 467             fiy2             = _mm256_add_ps(fiy2,ty);
 468             fiz2             = _mm256_add_ps(fiz2,tz);
 469
 470             fjx0             = _mm256_add_ps(fjx0,tx);
 471             fjy0             = _mm256_add_ps(fjy0,ty);
 472             fjz0             = _mm256_add_ps(fjz0,tz);
 473
 474             fjptrA             = f+j_coord_offsetA;
 475             fjptrB             = f+j_coord_offsetB;
 476             fjptrC             = f+j_coord_offsetC;
 477             fjptrD             = f+j_coord_offsetD;
 478             fjptrE             = f+j_coord_offsetE;
 479             fjptrF             = f+j_coord_offsetF;
 480             fjptrG             = f+j_coord_offsetG;
 481             fjptrH             = f+j_coord_offsetH;
 482
 483             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 484
 485             /* Inner loop uses 162 flops */
 486         }
 487
 488         if(jidx<j_index_end)
 489         {
 490
 491             /* Get j neighbor index, and coordinate index */
 492             jnrlistA         = jjnr[jidx];
 493             jnrlistB         = jjnr[jidx+1];
 494             jnrlistC         = jjnr[jidx+2];
 495             jnrlistD         = jjnr[jidx+3];
 496             jnrlistE         = jjnr[jidx+4];
 497             jnrlistF         = jjnr[jidx+5];
 498             jnrlistG         = jjnr[jidx+6];
 499             jnrlistH         = jjnr[jidx+7];
 500             /* Sign of each element will be negative for non-real atoms.
 501              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 502              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 503              */
 504             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 505                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 506
 507             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 508             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 509             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 510             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 511             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 512             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 513             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 514             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 515             j_coord_offsetA  = DIM*jnrA;
 516             j_coord_offsetB  = DIM*jnrB;
 517             j_coord_offsetC  = DIM*jnrC;
 518             j_coord_offsetD  = DIM*jnrD;
 519             j_coord_offsetE  = DIM*jnrE;
 520             j_coord_offsetF  = DIM*jnrF;
 521             j_coord_offsetG  = DIM*jnrG;
 522             j_coord_offsetH  = DIM*jnrH;
 523
 524             /* load j atom coordinates */
 525             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 526                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 527                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 528                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 529                                                  &jx0,&jy0,&jz0);
 530
 531             /* Calculate displacement vector */
 532             dx00             = _mm256_sub_ps(ix0,jx0);
 533             dy00             = _mm256_sub_ps(iy0,jy0);
 534             dz00             = _mm256_sub_ps(iz0,jz0);
 535             dx10             = _mm256_sub_ps(ix1,jx0);
 536             dy10             = _mm256_sub_ps(iy1,jy0);
 537             dz10             = _mm256_sub_ps(iz1,jz0);
 538             dx20             = _mm256_sub_ps(ix2,jx0);
 539             dy20             = _mm256_sub_ps(iy2,jy0);
 540             dz20             = _mm256_sub_ps(iz2,jz0);
 541
 542             /* Calculate squared distance and things based on it */
 543             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 544             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 545             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 546
 547             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 548             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 549             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 550
 551             /* Load parameters for j particles */
 552             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 553                                                                  charge+jnrC+0,charge+jnrD+0,
 554                                                                  charge+jnrE+0,charge+jnrF+0,
 555                                                                  charge+jnrG+0,charge+jnrH+0);
 556             vdwjidx0A        = 2*vdwtype[jnrA+0];
 557             vdwjidx0B        = 2*vdwtype[jnrB+0];
 558             vdwjidx0C        = 2*vdwtype[jnrC+0];
 559             vdwjidx0D        = 2*vdwtype[jnrD+0];
 560             vdwjidx0E        = 2*vdwtype[jnrE+0];
 561             vdwjidx0F        = 2*vdwtype[jnrF+0];
 562             vdwjidx0G        = 2*vdwtype[jnrG+0];
 563             vdwjidx0H        = 2*vdwtype[jnrH+0];
 564
 565             fjx0             = _mm256_setzero_ps();
 566             fjy0             = _mm256_setzero_ps();
 567             fjz0             = _mm256_setzero_ps();
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             r00              = _mm256_mul_ps(rsq00,rinv00);
 574             r00              = _mm256_andnot_ps(dummy_mask,r00);
 575
 576             /* Compute parameters for interactions between i and j atoms */
 577             qq00             = _mm256_mul_ps(iq0,jq0);
 578             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 579                                             vdwioffsetptr0+vdwjidx0B,
 580                                             vdwioffsetptr0+vdwjidx0C,
 581                                             vdwioffsetptr0+vdwjidx0D,
 582                                             vdwioffsetptr0+vdwjidx0E,
 583                                             vdwioffsetptr0+vdwjidx0F,
 584                                             vdwioffsetptr0+vdwjidx0G,
 585                                             vdwioffsetptr0+vdwjidx0H,
 586                                             &c6_00,&c12_00);
 587
 588             /* Calculate table index by multiplying r with table scale and truncate to integer */
 589             rt               = _mm256_mul_ps(r00,vftabscale);
 590             vfitab           = _mm256_cvttps_epi32(rt);
 591             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 592             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 593             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 594             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 595             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 596             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 597
 598             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 599             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 600                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 601             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 602                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 603             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 604                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 605             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 606                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 607             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 608             Heps             = _mm256_mul_ps(vfeps,H);
 609             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 610             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 611             velec            = _mm256_mul_ps(qq00,VV);
 612             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 613             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 614
 615             /* CUBIC SPLINE TABLE DISPERSION */
 616             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 617             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 618             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 619                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 620             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 621                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 622             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 623                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 624             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 625                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 626             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 627             Heps             = _mm256_mul_ps(vfeps,H);
 628             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 629             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 630             vvdw6            = _mm256_mul_ps(c6_00,VV);
 631             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 632             fvdw6            = _mm256_mul_ps(c6_00,FF);
 633
 634             /* CUBIC SPLINE TABLE REPULSION */
 635             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 636             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 637             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 638                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 639             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 640                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 641             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 642                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 643             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 644                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 645             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 646             Heps             = _mm256_mul_ps(vfeps,H);
 647             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 648             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 649             vvdw12           = _mm256_mul_ps(c12_00,VV);
 650             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 651             fvdw12           = _mm256_mul_ps(c12_00,FF);
 652             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 653             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velec            = _mm256_andnot_ps(dummy_mask,velec);
 657             velecsum         = _mm256_add_ps(velecsum,velec);
 658             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 659             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 660
 661             fscal            = _mm256_add_ps(felec,fvdw);
 662
 663             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 664
 665             /* Calculate temporary vectorial force */
 666             tx               = _mm256_mul_ps(fscal,dx00);
 667             ty               = _mm256_mul_ps(fscal,dy00);
 668             tz               = _mm256_mul_ps(fscal,dz00);
 669
 670             /* Update vectorial force */
 671             fix0             = _mm256_add_ps(fix0,tx);
 672             fiy0             = _mm256_add_ps(fiy0,ty);
 673             fiz0             = _mm256_add_ps(fiz0,tz);
 674
 675             fjx0             = _mm256_add_ps(fjx0,tx);
 676             fjy0             = _mm256_add_ps(fjy0,ty);
 677             fjz0             = _mm256_add_ps(fjz0,tz);
 678
 679             /**************************
 680              * CALCULATE INTERACTIONS *
 681              **************************/
 682
 683             r10              = _mm256_mul_ps(rsq10,rinv10);
 684             r10              = _mm256_andnot_ps(dummy_mask,r10);
 685
 686             /* Compute parameters for interactions between i and j atoms */
 687             qq10             = _mm256_mul_ps(iq1,jq0);
 688
 689             /* Calculate table index by multiplying r with table scale and truncate to integer */
 690             rt               = _mm256_mul_ps(r10,vftabscale);
 691             vfitab           = _mm256_cvttps_epi32(rt);
 692             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 693             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 694             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 695             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 696             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 697             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 698
 699             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 700             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 701                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 702             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 703                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 704             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 705                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 706             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 707                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 708             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 709             Heps             = _mm256_mul_ps(vfeps,H);
 710             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 711             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 712             velec            = _mm256_mul_ps(qq10,VV);
 713             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 714             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 715
 716             /* Update potential sum for this i atom from the interaction with this j atom. */
 717             velec            = _mm256_andnot_ps(dummy_mask,velec);
 718             velecsum         = _mm256_add_ps(velecsum,velec);
 719
 720             fscal            = felec;
 721
 722             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 723
 724             /* Calculate temporary vectorial force */
 725             tx               = _mm256_mul_ps(fscal,dx10);
 726             ty               = _mm256_mul_ps(fscal,dy10);
 727             tz               = _mm256_mul_ps(fscal,dz10);
 728
 729             /* Update vectorial force */
 730             fix1             = _mm256_add_ps(fix1,tx);
 731             fiy1             = _mm256_add_ps(fiy1,ty);
 732             fiz1             = _mm256_add_ps(fiz1,tz);
 733
 734             fjx0             = _mm256_add_ps(fjx0,tx);
 735             fjy0             = _mm256_add_ps(fjy0,ty);
 736             fjz0             = _mm256_add_ps(fjz0,tz);
 737
 738             /**************************
 739              * CALCULATE INTERACTIONS *
 740              **************************/
 741
 742             r20              = _mm256_mul_ps(rsq20,rinv20);
 743             r20              = _mm256_andnot_ps(dummy_mask,r20);
 744
 745             /* Compute parameters for interactions between i and j atoms */
 746             qq20             = _mm256_mul_ps(iq2,jq0);
 747
 748             /* Calculate table index by multiplying r with table scale and truncate to integer */
 749             rt               = _mm256_mul_ps(r20,vftabscale);
 750             vfitab           = _mm256_cvttps_epi32(rt);
 751             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 752             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 753             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 754             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 755             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 756             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 757
 758             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 759             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 760                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 761             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 762                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 763             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 764                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 765             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 766                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 767             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 768             Heps             = _mm256_mul_ps(vfeps,H);
 769             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 770             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 771             velec            = _mm256_mul_ps(qq20,VV);
 772             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 773             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 774
 775             /* Update potential sum for this i atom from the interaction with this j atom. */
 776             velec            = _mm256_andnot_ps(dummy_mask,velec);
 777             velecsum         = _mm256_add_ps(velecsum,velec);
 778
 779             fscal            = felec;
 780
 781             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 782
 783             /* Calculate temporary vectorial force */
 784             tx               = _mm256_mul_ps(fscal,dx20);
 785             ty               = _mm256_mul_ps(fscal,dy20);
 786             tz               = _mm256_mul_ps(fscal,dz20);
 787
 788             /* Update vectorial force */
 789             fix2             = _mm256_add_ps(fix2,tx);
 790             fiy2             = _mm256_add_ps(fiy2,ty);
 791             fiz2             = _mm256_add_ps(fiz2,tz);
 792
 793             fjx0             = _mm256_add_ps(fjx0,tx);
 794             fjy0             = _mm256_add_ps(fjy0,ty);
 795             fjz0             = _mm256_add_ps(fjz0,tz);
 796
 797             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 798             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 799             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 800             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 801             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 802             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 803             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 804             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 805
 806             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 807
 808             /* Inner loop uses 165 flops */
 809         }
 810
 811         /* End of innermost loop */
 812
 813         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 814                                                  f+i_coord_offset,fshift+i_shift_offset);
 815
 816         ggid                        = gid[iidx];
 817         /* Update potential energies */
 818         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 819         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 820
 821         /* Increment number of inner iterations */
 822         inneriter                  += j_index_end - j_index_start;
 823
 824         /* Outer loop uses 20 flops */
 825     }
 826
 827     /* Increment number of outer iterations */
 828     outeriter        += nri;
 829
 830     /* Update outer/inner flops */
 831
 832     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*165);
 833 }
 834 /*
 835  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_single
 836  * Electrostatics interaction: CubicSplineTable
 837  * VdW interaction:            CubicSplineTable
 838  * Geometry:                   Water3-Particle
 839  * Calculate force/pot:        Force
 840  */
 841 void
 842 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_single
 843                     (t_nblist                    * gmx_restrict       nlist,
 844                      rvec                        * gmx_restrict          xx,
 845                      rvec                        * gmx_restrict          ff,
 846                      t_forcerec                  * gmx_restrict          fr,
 847                      t_mdatoms                   * gmx_restrict     mdatoms,
 848                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 849                      t_nrnb                      * gmx_restrict        nrnb)
 850 {
 851     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 852      * just 0 for non-waters.
 853      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 854      * jnr indices corresponding to data put in the four positions in the SIMD register.
 855      */
 856     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 857     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 858     int              jnrA,jnrB,jnrC,jnrD;
 859     int              jnrE,jnrF,jnrG,jnrH;
 860     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 861     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 862     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 863     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 864     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 865     real             rcutoff_scalar;
 866     real             *shiftvec,*fshift,*x,*f;
 867     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 868     real             scratch[4*DIM];
 869     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 870     real *           vdwioffsetptr0;
 871     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 872     real *           vdwioffsetptr1;
 873     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 874     real *           vdwioffsetptr2;
 875     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 876     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 877     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 878     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 879     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 880     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 881     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 882     real             *charge;
 883     int              nvdwtype;
 884     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 885     int              *vdwtype;
 886     real             *vdwparam;
 887     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 888     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 889     __m256i          vfitab;
 890     __m128i          vfitab_lo,vfitab_hi;
 891     __m128i          ifour       = _mm_set1_epi32(4);
 892     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 893     real             *vftab;
 894     __m256           dummy_mask,cutoff_mask;
 895     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 896     __m256           one     = _mm256_set1_ps(1.0);
 897     __m256           two     = _mm256_set1_ps(2.0);
 898     x                = xx[0];
 899     f                = ff[0];
 900
 901     nri              = nlist->nri;
 902     iinr             = nlist->iinr;
 903     jindex           = nlist->jindex;
 904     jjnr             = nlist->jjnr;
 905     shiftidx         = nlist->shift;
 906     gid              = nlist->gid;
 907     shiftvec         = fr->shift_vec[0];
 908     fshift           = fr->fshift[0];
 909     facel            = _mm256_set1_ps(fr->epsfac);
 910     charge           = mdatoms->chargeA;
 911     nvdwtype         = fr->ntype;
 912     vdwparam         = fr->nbfp;
 913     vdwtype          = mdatoms->typeA;
 914
 915     vftab            = kernel_data->table_elec_vdw->data;
 916     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 917
 918     /* Setup water-specific parameters */
 919     inr              = nlist->iinr[0];
 920     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 921     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 922     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 923     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 924
 925     /* Avoid stupid compiler warnings */
 926     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 927     j_coord_offsetA = 0;
 928     j_coord_offsetB = 0;
 929     j_coord_offsetC = 0;
 930     j_coord_offsetD = 0;
 931     j_coord_offsetE = 0;
 932     j_coord_offsetF = 0;
 933     j_coord_offsetG = 0;
 934     j_coord_offsetH = 0;
 935
 936     outeriter        = 0;
 937     inneriter        = 0;
 938
 939     for(iidx=0;iidx<4*DIM;iidx++)
 940     {
 941         scratch[iidx] = 0.0;
 942     }
 943
 944     /* Start outer loop over neighborlists */
 945     for(iidx=0; iidx<nri; iidx++)
 946     {
 947         /* Load shift vector for this list */
 948         i_shift_offset   = DIM*shiftidx[iidx];
 949
 950         /* Load limits for loop over neighbors */
 951         j_index_start    = jindex[iidx];
 952         j_index_end      = jindex[iidx+1];
 953
 954         /* Get outer coordinate index */
 955         inr              = iinr[iidx];
 956         i_coord_offset   = DIM*inr;
 957
 958         /* Load i particle coords and add shift vector */
 959         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 960                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 961
 962         fix0             = _mm256_setzero_ps();
 963         fiy0             = _mm256_setzero_ps();
 964         fiz0             = _mm256_setzero_ps();
 965         fix1             = _mm256_setzero_ps();
 966         fiy1             = _mm256_setzero_ps();
 967         fiz1             = _mm256_setzero_ps();
 968         fix2             = _mm256_setzero_ps();
 969         fiy2             = _mm256_setzero_ps();
 970         fiz2             = _mm256_setzero_ps();
 971
 972         /* Start inner kernel loop */
 973         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 974         {
 975
 976             /* Get j neighbor index, and coordinate index */
 977             jnrA             = jjnr[jidx];
 978             jnrB             = jjnr[jidx+1];
 979             jnrC             = jjnr[jidx+2];
 980             jnrD             = jjnr[jidx+3];
 981             jnrE             = jjnr[jidx+4];
 982             jnrF             = jjnr[jidx+5];
 983             jnrG             = jjnr[jidx+6];
 984             jnrH             = jjnr[jidx+7];
 985             j_coord_offsetA  = DIM*jnrA;
 986             j_coord_offsetB  = DIM*jnrB;
 987             j_coord_offsetC  = DIM*jnrC;
 988             j_coord_offsetD  = DIM*jnrD;
 989             j_coord_offsetE  = DIM*jnrE;
 990             j_coord_offsetF  = DIM*jnrF;
 991             j_coord_offsetG  = DIM*jnrG;
 992             j_coord_offsetH  = DIM*jnrH;
 993
 994             /* load j atom coordinates */
 995             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 996                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 997                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 998                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 999                                                  &jx0,&jy0,&jz0);
1000
1001             /* Calculate displacement vector */
1002             dx00             = _mm256_sub_ps(ix0,jx0);
1003             dy00             = _mm256_sub_ps(iy0,jy0);
1004             dz00             = _mm256_sub_ps(iz0,jz0);
1005             dx10             = _mm256_sub_ps(ix1,jx0);
1006             dy10             = _mm256_sub_ps(iy1,jy0);
1007             dz10             = _mm256_sub_ps(iz1,jz0);
1008             dx20             = _mm256_sub_ps(ix2,jx0);
1009             dy20             = _mm256_sub_ps(iy2,jy0);
1010             dz20             = _mm256_sub_ps(iz2,jz0);
1011
1012             /* Calculate squared distance and things based on it */
1013             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1014             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1015             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1016
1017             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1018             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1019             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1020
1021             /* Load parameters for j particles */
1022             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1023                                                                  charge+jnrC+0,charge+jnrD+0,
1024                                                                  charge+jnrE+0,charge+jnrF+0,
1025                                                                  charge+jnrG+0,charge+jnrH+0);
1026             vdwjidx0A        = 2*vdwtype[jnrA+0];
1027             vdwjidx0B        = 2*vdwtype[jnrB+0];
1028             vdwjidx0C        = 2*vdwtype[jnrC+0];
1029             vdwjidx0D        = 2*vdwtype[jnrD+0];
1030             vdwjidx0E        = 2*vdwtype[jnrE+0];
1031             vdwjidx0F        = 2*vdwtype[jnrF+0];
1032             vdwjidx0G        = 2*vdwtype[jnrG+0];
1033             vdwjidx0H        = 2*vdwtype[jnrH+0];
1034
1035             fjx0             = _mm256_setzero_ps();
1036             fjy0             = _mm256_setzero_ps();
1037             fjz0             = _mm256_setzero_ps();
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             r00              = _mm256_mul_ps(rsq00,rinv00);
1044
1045             /* Compute parameters for interactions between i and j atoms */
1046             qq00             = _mm256_mul_ps(iq0,jq0);
1047             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1048                                             vdwioffsetptr0+vdwjidx0B,
1049                                             vdwioffsetptr0+vdwjidx0C,
1050                                             vdwioffsetptr0+vdwjidx0D,
1051                                             vdwioffsetptr0+vdwjidx0E,
1052                                             vdwioffsetptr0+vdwjidx0F,
1053                                             vdwioffsetptr0+vdwjidx0G,
1054                                             vdwioffsetptr0+vdwjidx0H,
1055                                             &c6_00,&c12_00);
1056
1057             /* Calculate table index by multiplying r with table scale and truncate to integer */
1058             rt               = _mm256_mul_ps(r00,vftabscale);
1059             vfitab           = _mm256_cvttps_epi32(rt);
1060             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1061             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1062             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1063             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1064             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1065             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1066
1067             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1068             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1069                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1070             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1071                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1072             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1073                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1074             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1075                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1076             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1077             Heps             = _mm256_mul_ps(vfeps,H);
1078             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1079             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1080             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1081
1082             /* CUBIC SPLINE TABLE DISPERSION */
1083             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1084             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1085             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1086                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1087             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1088                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1089             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1090                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1091             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1092                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1093             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1094             Heps             = _mm256_mul_ps(vfeps,H);
1095             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1096             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1097             fvdw6            = _mm256_mul_ps(c6_00,FF);
1098
1099             /* CUBIC SPLINE TABLE REPULSION */
1100             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1101             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1102             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1103                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1104             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1105                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1106             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1107                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1108             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1109                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1110             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1111             Heps             = _mm256_mul_ps(vfeps,H);
1112             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1113             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1114             fvdw12           = _mm256_mul_ps(c12_00,FF);
1115             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1116
1117             fscal            = _mm256_add_ps(felec,fvdw);
1118
1119             /* Calculate temporary vectorial force */
1120             tx               = _mm256_mul_ps(fscal,dx00);
1121             ty               = _mm256_mul_ps(fscal,dy00);
1122             tz               = _mm256_mul_ps(fscal,dz00);
1123
1124             /* Update vectorial force */
1125             fix0             = _mm256_add_ps(fix0,tx);
1126             fiy0             = _mm256_add_ps(fiy0,ty);
1127             fiz0             = _mm256_add_ps(fiz0,tz);
1128
1129             fjx0             = _mm256_add_ps(fjx0,tx);
1130             fjy0             = _mm256_add_ps(fjy0,ty);
1131             fjz0             = _mm256_add_ps(fjz0,tz);
1132
1133             /**************************
1134              * CALCULATE INTERACTIONS *
1135              **************************/
1136
1137             r10              = _mm256_mul_ps(rsq10,rinv10);
1138
1139             /* Compute parameters for interactions between i and j atoms */
1140             qq10             = _mm256_mul_ps(iq1,jq0);
1141
1142             /* Calculate table index by multiplying r with table scale and truncate to integer */
1143             rt               = _mm256_mul_ps(r10,vftabscale);
1144             vfitab           = _mm256_cvttps_epi32(rt);
1145             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1146             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1147             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1148             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1149             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1150             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1151
1152             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1153             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1154                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1155             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1156                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1157             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1158                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1159             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1160                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1161             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1162             Heps             = _mm256_mul_ps(vfeps,H);
1163             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1164             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1165             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1166
1167             fscal            = felec;
1168
1169             /* Calculate temporary vectorial force */
1170             tx               = _mm256_mul_ps(fscal,dx10);
1171             ty               = _mm256_mul_ps(fscal,dy10);
1172             tz               = _mm256_mul_ps(fscal,dz10);
1173
1174             /* Update vectorial force */
1175             fix1             = _mm256_add_ps(fix1,tx);
1176             fiy1             = _mm256_add_ps(fiy1,ty);
1177             fiz1             = _mm256_add_ps(fiz1,tz);
1178
1179             fjx0             = _mm256_add_ps(fjx0,tx);
1180             fjy0             = _mm256_add_ps(fjy0,ty);
1181             fjz0             = _mm256_add_ps(fjz0,tz);
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             r20              = _mm256_mul_ps(rsq20,rinv20);
1188
1189             /* Compute parameters for interactions between i and j atoms */
1190             qq20             = _mm256_mul_ps(iq2,jq0);
1191
1192             /* Calculate table index by multiplying r with table scale and truncate to integer */
1193             rt               = _mm256_mul_ps(r20,vftabscale);
1194             vfitab           = _mm256_cvttps_epi32(rt);
1195             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1196             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1197             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1198             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1199             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1200             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1201
1202             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1203             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1204                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1205             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1206                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1207             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1208                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1209             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1210                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1211             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1212             Heps             = _mm256_mul_ps(vfeps,H);
1213             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1214             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1215             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1216
1217             fscal            = felec;
1218
1219             /* Calculate temporary vectorial force */
1220             tx               = _mm256_mul_ps(fscal,dx20);
1221             ty               = _mm256_mul_ps(fscal,dy20);
1222             tz               = _mm256_mul_ps(fscal,dz20);
1223
1224             /* Update vectorial force */
1225             fix2             = _mm256_add_ps(fix2,tx);
1226             fiy2             = _mm256_add_ps(fiy2,ty);
1227             fiz2             = _mm256_add_ps(fiz2,tz);
1228
1229             fjx0             = _mm256_add_ps(fjx0,tx);
1230             fjy0             = _mm256_add_ps(fjy0,ty);
1231             fjz0             = _mm256_add_ps(fjz0,tz);
1232
1233             fjptrA             = f+j_coord_offsetA;
1234             fjptrB             = f+j_coord_offsetB;
1235             fjptrC             = f+j_coord_offsetC;
1236             fjptrD             = f+j_coord_offsetD;
1237             fjptrE             = f+j_coord_offsetE;
1238             fjptrF             = f+j_coord_offsetF;
1239             fjptrG             = f+j_coord_offsetG;
1240             fjptrH             = f+j_coord_offsetH;
1241
1242             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1243
1244             /* Inner loop uses 142 flops */
1245         }
1246
1247         if(jidx<j_index_end)
1248         {
1249
1250             /* Get j neighbor index, and coordinate index */
1251             jnrlistA         = jjnr[jidx];
1252             jnrlistB         = jjnr[jidx+1];
1253             jnrlistC         = jjnr[jidx+2];
1254             jnrlistD         = jjnr[jidx+3];
1255             jnrlistE         = jjnr[jidx+4];
1256             jnrlistF         = jjnr[jidx+5];
1257             jnrlistG         = jjnr[jidx+6];
1258             jnrlistH         = jjnr[jidx+7];
1259             /* Sign of each element will be negative for non-real atoms.
1260              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1261              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1262              */
1263             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1264                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1265
1266             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1267             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1268             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1269             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1270             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1271             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1272             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1273             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1274             j_coord_offsetA  = DIM*jnrA;
1275             j_coord_offsetB  = DIM*jnrB;
1276             j_coord_offsetC  = DIM*jnrC;
1277             j_coord_offsetD  = DIM*jnrD;
1278             j_coord_offsetE  = DIM*jnrE;
1279             j_coord_offsetF  = DIM*jnrF;
1280             j_coord_offsetG  = DIM*jnrG;
1281             j_coord_offsetH  = DIM*jnrH;
1282
1283             /* load j atom coordinates */
1284             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1285                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1286                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1287                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1288                                                  &jx0,&jy0,&jz0);
1289
1290             /* Calculate displacement vector */
1291             dx00             = _mm256_sub_ps(ix0,jx0);
1292             dy00             = _mm256_sub_ps(iy0,jy0);
1293             dz00             = _mm256_sub_ps(iz0,jz0);
1294             dx10             = _mm256_sub_ps(ix1,jx0);
1295             dy10             = _mm256_sub_ps(iy1,jy0);
1296             dz10             = _mm256_sub_ps(iz1,jz0);
1297             dx20             = _mm256_sub_ps(ix2,jx0);
1298             dy20             = _mm256_sub_ps(iy2,jy0);
1299             dz20             = _mm256_sub_ps(iz2,jz0);
1300
1301             /* Calculate squared distance and things based on it */
1302             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1303             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1304             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1305
1306             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1307             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1308             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1309
1310             /* Load parameters for j particles */
1311             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1312                                                                  charge+jnrC+0,charge+jnrD+0,
1313                                                                  charge+jnrE+0,charge+jnrF+0,
1314                                                                  charge+jnrG+0,charge+jnrH+0);
1315             vdwjidx0A        = 2*vdwtype[jnrA+0];
1316             vdwjidx0B        = 2*vdwtype[jnrB+0];
1317             vdwjidx0C        = 2*vdwtype[jnrC+0];
1318             vdwjidx0D        = 2*vdwtype[jnrD+0];
1319             vdwjidx0E        = 2*vdwtype[jnrE+0];
1320             vdwjidx0F        = 2*vdwtype[jnrF+0];
1321             vdwjidx0G        = 2*vdwtype[jnrG+0];
1322             vdwjidx0H        = 2*vdwtype[jnrH+0];
1323
1324             fjx0             = _mm256_setzero_ps();
1325             fjy0             = _mm256_setzero_ps();
1326             fjz0             = _mm256_setzero_ps();
1327
1328             /**************************
1329              * CALCULATE INTERACTIONS *
1330              **************************/
1331
1332             r00              = _mm256_mul_ps(rsq00,rinv00);
1333             r00              = _mm256_andnot_ps(dummy_mask,r00);
1334
1335             /* Compute parameters for interactions between i and j atoms */
1336             qq00             = _mm256_mul_ps(iq0,jq0);
1337             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1338                                             vdwioffsetptr0+vdwjidx0B,
1339                                             vdwioffsetptr0+vdwjidx0C,
1340                                             vdwioffsetptr0+vdwjidx0D,
1341                                             vdwioffsetptr0+vdwjidx0E,
1342                                             vdwioffsetptr0+vdwjidx0F,
1343                                             vdwioffsetptr0+vdwjidx0G,
1344                                             vdwioffsetptr0+vdwjidx0H,
1345                                             &c6_00,&c12_00);
1346
1347             /* Calculate table index by multiplying r with table scale and truncate to integer */
1348             rt               = _mm256_mul_ps(r00,vftabscale);
1349             vfitab           = _mm256_cvttps_epi32(rt);
1350             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1351             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1352             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1353             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1354             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1355             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1356
1357             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1358             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1359                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1360             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1361                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1362             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1363                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1364             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1365                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1366             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1367             Heps             = _mm256_mul_ps(vfeps,H);
1368             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1369             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1370             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1371
1372             /* CUBIC SPLINE TABLE DISPERSION */
1373             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1374             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1375             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1376                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1377             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1378                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1379             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1380                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1381             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1382                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1383             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1384             Heps             = _mm256_mul_ps(vfeps,H);
1385             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1386             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1387             fvdw6            = _mm256_mul_ps(c6_00,FF);
1388
1389             /* CUBIC SPLINE TABLE REPULSION */
1390             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1391             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1392             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1393                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1394             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1395                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1396             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1397                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1398             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1399                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1400             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1401             Heps             = _mm256_mul_ps(vfeps,H);
1402             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1403             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1404             fvdw12           = _mm256_mul_ps(c12_00,FF);
1405             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1406
1407             fscal            = _mm256_add_ps(felec,fvdw);
1408
1409             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1410
1411             /* Calculate temporary vectorial force */
1412             tx               = _mm256_mul_ps(fscal,dx00);
1413             ty               = _mm256_mul_ps(fscal,dy00);
1414             tz               = _mm256_mul_ps(fscal,dz00);
1415
1416             /* Update vectorial force */
1417             fix0             = _mm256_add_ps(fix0,tx);
1418             fiy0             = _mm256_add_ps(fiy0,ty);
1419             fiz0             = _mm256_add_ps(fiz0,tz);
1420
1421             fjx0             = _mm256_add_ps(fjx0,tx);
1422             fjy0             = _mm256_add_ps(fjy0,ty);
1423             fjz0             = _mm256_add_ps(fjz0,tz);
1424
1425             /**************************
1426              * CALCULATE INTERACTIONS *
1427              **************************/
1428
1429             r10              = _mm256_mul_ps(rsq10,rinv10);
1430             r10              = _mm256_andnot_ps(dummy_mask,r10);
1431
1432             /* Compute parameters for interactions between i and j atoms */
1433             qq10             = _mm256_mul_ps(iq1,jq0);
1434
1435             /* Calculate table index by multiplying r with table scale and truncate to integer */
1436             rt               = _mm256_mul_ps(r10,vftabscale);
1437             vfitab           = _mm256_cvttps_epi32(rt);
1438             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1439             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1440             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1441             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1442             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1443             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1444
1445             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1446             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1447                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1448             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1449                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1450             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1451                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1452             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1453                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1454             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1455             Heps             = _mm256_mul_ps(vfeps,H);
1456             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1457             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1458             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1459
1460             fscal            = felec;
1461
1462             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1463
1464             /* Calculate temporary vectorial force */
1465             tx               = _mm256_mul_ps(fscal,dx10);
1466             ty               = _mm256_mul_ps(fscal,dy10);
1467             tz               = _mm256_mul_ps(fscal,dz10);
1468
1469             /* Update vectorial force */
1470             fix1             = _mm256_add_ps(fix1,tx);
1471             fiy1             = _mm256_add_ps(fiy1,ty);
1472             fiz1             = _mm256_add_ps(fiz1,tz);
1473
1474             fjx0             = _mm256_add_ps(fjx0,tx);
1475             fjy0             = _mm256_add_ps(fjy0,ty);
1476             fjz0             = _mm256_add_ps(fjz0,tz);
1477
1478             /**************************
1479              * CALCULATE INTERACTIONS *
1480              **************************/
1481
1482             r20              = _mm256_mul_ps(rsq20,rinv20);
1483             r20              = _mm256_andnot_ps(dummy_mask,r20);
1484
1485             /* Compute parameters for interactions between i and j atoms */
1486             qq20             = _mm256_mul_ps(iq2,jq0);
1487
1488             /* Calculate table index by multiplying r with table scale and truncate to integer */
1489             rt               = _mm256_mul_ps(r20,vftabscale);
1490             vfitab           = _mm256_cvttps_epi32(rt);
1491             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1492             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1493             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1494             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1495             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1496             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1497
1498             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1499             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1500                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1501             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1502                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1503             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1504                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1505             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1506                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1507             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1508             Heps             = _mm256_mul_ps(vfeps,H);
1509             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1510             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1511             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1512
1513             fscal            = felec;
1514
1515             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1516
1517             /* Calculate temporary vectorial force */
1518             tx               = _mm256_mul_ps(fscal,dx20);
1519             ty               = _mm256_mul_ps(fscal,dy20);
1520             tz               = _mm256_mul_ps(fscal,dz20);
1521
1522             /* Update vectorial force */
1523             fix2             = _mm256_add_ps(fix2,tx);
1524             fiy2             = _mm256_add_ps(fiy2,ty);
1525             fiz2             = _mm256_add_ps(fiz2,tz);
1526
1527             fjx0             = _mm256_add_ps(fjx0,tx);
1528             fjy0             = _mm256_add_ps(fjy0,ty);
1529             fjz0             = _mm256_add_ps(fjz0,tz);
1530
1531             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1532             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1533             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1534             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1535             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1536             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1537             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1538             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1539
1540             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1541
1542             /* Inner loop uses 145 flops */
1543         }
1544
1545         /* End of innermost loop */
1546
1547         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1548                                                  f+i_coord_offset,fshift+i_shift_offset);
1549
1550         /* Increment number of inner iterations */
1551         inneriter                  += j_index_end - j_index_start;
1552
1553         /* Outer loop uses 18 flops */
1554     }
1555
1556     /* Increment number of outer iterations */
1557     outeriter        += nri;
1558
1559     /* Update outer/inner flops */
1560
1561     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*145);
1562 }