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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 103     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 104     __m256i          vfitab;
 105     __m128i          vfitab_lo,vfitab_hi;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m256           dummy_mask,cutoff_mask;
 110     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 111     __m256           one     = _mm256_set1_ps(1.0);
 112     __m256           two     = _mm256_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_ps(fr->ic->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     vftab            = kernel_data->table_elec_vdw->data;
 131     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 136     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 137     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 138     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146     j_coord_offsetE = 0;
 147     j_coord_offsetF = 0;
 148     j_coord_offsetG = 0;
 149     j_coord_offsetH = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 176
 177         fix0             = _mm256_setzero_ps();
 178         fiy0             = _mm256_setzero_ps();
 179         fiz0             = _mm256_setzero_ps();
 180         fix1             = _mm256_setzero_ps();
 181         fiy1             = _mm256_setzero_ps();
 182         fiz1             = _mm256_setzero_ps();
 183         fix2             = _mm256_setzero_ps();
 184         fiy2             = _mm256_setzero_ps();
 185         fiz2             = _mm256_setzero_ps();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm256_setzero_ps();
 189         vvdwsum          = _mm256_setzero_ps();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             jnrC             = jjnr[jidx+2];
 199             jnrD             = jjnr[jidx+3];
 200             jnrE             = jjnr[jidx+4];
 201             jnrF             = jjnr[jidx+5];
 202             jnrG             = jjnr[jidx+6];
 203             jnrH             = jjnr[jidx+7];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208             j_coord_offsetE  = DIM*jnrE;
 209             j_coord_offsetF  = DIM*jnrF;
 210             j_coord_offsetG  = DIM*jnrG;
 211             j_coord_offsetH  = DIM*jnrH;
 212
 213             /* load j atom coordinates */
 214             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 215                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 216                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 217                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 218                                                  &jx0,&jy0,&jz0);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm256_sub_ps(ix0,jx0);
 222             dy00             = _mm256_sub_ps(iy0,jy0);
 223             dz00             = _mm256_sub_ps(iz0,jz0);
 224             dx10             = _mm256_sub_ps(ix1,jx0);
 225             dy10             = _mm256_sub_ps(iy1,jy0);
 226             dz10             = _mm256_sub_ps(iz1,jz0);
 227             dx20             = _mm256_sub_ps(ix2,jx0);
 228             dy20             = _mm256_sub_ps(iy2,jy0);
 229             dz20             = _mm256_sub_ps(iz2,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 233             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 234             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 235
 236             rinv00           = avx256_invsqrt_f(rsq00);
 237             rinv10           = avx256_invsqrt_f(rsq10);
 238             rinv20           = avx256_invsqrt_f(rsq20);
 239
 240             /* Load parameters for j particles */
 241             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 242                                                                  charge+jnrC+0,charge+jnrD+0,
 243                                                                  charge+jnrE+0,charge+jnrF+0,
 244                                                                  charge+jnrG+0,charge+jnrH+0);
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246             vdwjidx0B        = 2*vdwtype[jnrB+0];
 247             vdwjidx0C        = 2*vdwtype[jnrC+0];
 248             vdwjidx0D        = 2*vdwtype[jnrD+0];
 249             vdwjidx0E        = 2*vdwtype[jnrE+0];
 250             vdwjidx0F        = 2*vdwtype[jnrF+0];
 251             vdwjidx0G        = 2*vdwtype[jnrG+0];
 252             vdwjidx0H        = 2*vdwtype[jnrH+0];
 253
 254             fjx0             = _mm256_setzero_ps();
 255             fjy0             = _mm256_setzero_ps();
 256             fjz0             = _mm256_setzero_ps();
 257
 258             /**************************
 259              * CALCULATE INTERACTIONS *
 260              **************************/
 261
 262             r00              = _mm256_mul_ps(rsq00,rinv00);
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             qq00             = _mm256_mul_ps(iq0,jq0);
 266             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 267                                             vdwioffsetptr0+vdwjidx0B,
 268                                             vdwioffsetptr0+vdwjidx0C,
 269                                             vdwioffsetptr0+vdwjidx0D,
 270                                             vdwioffsetptr0+vdwjidx0E,
 271                                             vdwioffsetptr0+vdwjidx0F,
 272                                             vdwioffsetptr0+vdwjidx0G,
 273                                             vdwioffsetptr0+vdwjidx0H,
 274                                             &c6_00,&c12_00);
 275
 276             /* Calculate table index by multiplying r with table scale and truncate to integer */
 277             rt               = _mm256_mul_ps(r00,vftabscale);
 278             vfitab           = _mm256_cvttps_epi32(rt);
 279             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 280             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 281             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 282             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 283             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 284             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 285
 286             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 287             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 288                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 289             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 290                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 291             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 292                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 293             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 295             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 296             Heps             = _mm256_mul_ps(vfeps,H);
 297             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 298             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 299             velec            = _mm256_mul_ps(qq00,VV);
 300             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 301             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 302
 303             /* CUBIC SPLINE TABLE DISPERSION */
 304             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 305             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 306             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 307                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 308             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 309                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 310             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 311                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 312             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 314             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 315             Heps             = _mm256_mul_ps(vfeps,H);
 316             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 317             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 318             vvdw6            = _mm256_mul_ps(c6_00,VV);
 319             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 320             fvdw6            = _mm256_mul_ps(c6_00,FF);
 321
 322             /* CUBIC SPLINE TABLE REPULSION */
 323             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 324             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 325             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 326                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 327             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 328                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 329             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 330                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 331             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 332                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 333             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 334             Heps             = _mm256_mul_ps(vfeps,H);
 335             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 336             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 337             vvdw12           = _mm256_mul_ps(c12_00,VV);
 338             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 339             fvdw12           = _mm256_mul_ps(c12_00,FF);
 340             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 341             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm256_add_ps(velecsum,velec);
 345             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 346
 347             fscal            = _mm256_add_ps(felec,fvdw);
 348
 349             /* Calculate temporary vectorial force */
 350             tx               = _mm256_mul_ps(fscal,dx00);
 351             ty               = _mm256_mul_ps(fscal,dy00);
 352             tz               = _mm256_mul_ps(fscal,dz00);
 353
 354             /* Update vectorial force */
 355             fix0             = _mm256_add_ps(fix0,tx);
 356             fiy0             = _mm256_add_ps(fiy0,ty);
 357             fiz0             = _mm256_add_ps(fiz0,tz);
 358
 359             fjx0             = _mm256_add_ps(fjx0,tx);
 360             fjy0             = _mm256_add_ps(fjy0,ty);
 361             fjz0             = _mm256_add_ps(fjz0,tz);
 362
 363             /**************************
 364              * CALCULATE INTERACTIONS *
 365              **************************/
 366
 367             r10              = _mm256_mul_ps(rsq10,rinv10);
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq10             = _mm256_mul_ps(iq1,jq0);
 371
 372             /* Calculate table index by multiplying r with table scale and truncate to integer */
 373             rt               = _mm256_mul_ps(r10,vftabscale);
 374             vfitab           = _mm256_cvttps_epi32(rt);
 375             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 376             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 377             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 378             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 379             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 380             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 381
 382             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 383             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 384                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 385             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 386                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 387             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 388                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 389             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 390                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 391             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 392             Heps             = _mm256_mul_ps(vfeps,H);
 393             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 394             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 395             velec            = _mm256_mul_ps(qq10,VV);
 396             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 397             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 398
 399             /* Update potential sum for this i atom from the interaction with this j atom. */
 400             velecsum         = _mm256_add_ps(velecsum,velec);
 401
 402             fscal            = felec;
 403
 404             /* Calculate temporary vectorial force */
 405             tx               = _mm256_mul_ps(fscal,dx10);
 406             ty               = _mm256_mul_ps(fscal,dy10);
 407             tz               = _mm256_mul_ps(fscal,dz10);
 408
 409             /* Update vectorial force */
 410             fix1             = _mm256_add_ps(fix1,tx);
 411             fiy1             = _mm256_add_ps(fiy1,ty);
 412             fiz1             = _mm256_add_ps(fiz1,tz);
 413
 414             fjx0             = _mm256_add_ps(fjx0,tx);
 415             fjy0             = _mm256_add_ps(fjy0,ty);
 416             fjz0             = _mm256_add_ps(fjz0,tz);
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             r20              = _mm256_mul_ps(rsq20,rinv20);
 423
 424             /* Compute parameters for interactions between i and j atoms */
 425             qq20             = _mm256_mul_ps(iq2,jq0);
 426
 427             /* Calculate table index by multiplying r with table scale and truncate to integer */
 428             rt               = _mm256_mul_ps(r20,vftabscale);
 429             vfitab           = _mm256_cvttps_epi32(rt);
 430             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 431             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 432             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 433             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 434             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 435             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 436
 437             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 438             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 439                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 440             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 441                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 442             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 443                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 444             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 445                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 446             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 447             Heps             = _mm256_mul_ps(vfeps,H);
 448             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 449             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 450             velec            = _mm256_mul_ps(qq20,VV);
 451             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 452             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             velecsum         = _mm256_add_ps(velecsum,velec);
 456
 457             fscal            = felec;
 458
 459             /* Calculate temporary vectorial force */
 460             tx               = _mm256_mul_ps(fscal,dx20);
 461             ty               = _mm256_mul_ps(fscal,dy20);
 462             tz               = _mm256_mul_ps(fscal,dz20);
 463
 464             /* Update vectorial force */
 465             fix2             = _mm256_add_ps(fix2,tx);
 466             fiy2             = _mm256_add_ps(fiy2,ty);
 467             fiz2             = _mm256_add_ps(fiz2,tz);
 468
 469             fjx0             = _mm256_add_ps(fjx0,tx);
 470             fjy0             = _mm256_add_ps(fjy0,ty);
 471             fjz0             = _mm256_add_ps(fjz0,tz);
 472
 473             fjptrA             = f+j_coord_offsetA;
 474             fjptrB             = f+j_coord_offsetB;
 475             fjptrC             = f+j_coord_offsetC;
 476             fjptrD             = f+j_coord_offsetD;
 477             fjptrE             = f+j_coord_offsetE;
 478             fjptrF             = f+j_coord_offsetF;
 479             fjptrG             = f+j_coord_offsetG;
 480             fjptrH             = f+j_coord_offsetH;
 481
 482             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 483
 484             /* Inner loop uses 162 flops */
 485         }
 486
 487         if(jidx<j_index_end)
 488         {
 489
 490             /* Get j neighbor index, and coordinate index */
 491             jnrlistA         = jjnr[jidx];
 492             jnrlistB         = jjnr[jidx+1];
 493             jnrlistC         = jjnr[jidx+2];
 494             jnrlistD         = jjnr[jidx+3];
 495             jnrlistE         = jjnr[jidx+4];
 496             jnrlistF         = jjnr[jidx+5];
 497             jnrlistG         = jjnr[jidx+6];
 498             jnrlistH         = jjnr[jidx+7];
 499             /* Sign of each element will be negative for non-real atoms.
 500              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 501              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 502              */
 503             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 504                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 505
 506             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 507             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 508             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 509             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 510             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 511             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 512             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 513             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 514             j_coord_offsetA  = DIM*jnrA;
 515             j_coord_offsetB  = DIM*jnrB;
 516             j_coord_offsetC  = DIM*jnrC;
 517             j_coord_offsetD  = DIM*jnrD;
 518             j_coord_offsetE  = DIM*jnrE;
 519             j_coord_offsetF  = DIM*jnrF;
 520             j_coord_offsetG  = DIM*jnrG;
 521             j_coord_offsetH  = DIM*jnrH;
 522
 523             /* load j atom coordinates */
 524             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 525                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 526                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 527                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 528                                                  &jx0,&jy0,&jz0);
 529
 530             /* Calculate displacement vector */
 531             dx00             = _mm256_sub_ps(ix0,jx0);
 532             dy00             = _mm256_sub_ps(iy0,jy0);
 533             dz00             = _mm256_sub_ps(iz0,jz0);
 534             dx10             = _mm256_sub_ps(ix1,jx0);
 535             dy10             = _mm256_sub_ps(iy1,jy0);
 536             dz10             = _mm256_sub_ps(iz1,jz0);
 537             dx20             = _mm256_sub_ps(ix2,jx0);
 538             dy20             = _mm256_sub_ps(iy2,jy0);
 539             dz20             = _mm256_sub_ps(iz2,jz0);
 540
 541             /* Calculate squared distance and things based on it */
 542             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 543             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 544             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 545
 546             rinv00           = avx256_invsqrt_f(rsq00);
 547             rinv10           = avx256_invsqrt_f(rsq10);
 548             rinv20           = avx256_invsqrt_f(rsq20);
 549
 550             /* Load parameters for j particles */
 551             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 552                                                                  charge+jnrC+0,charge+jnrD+0,
 553                                                                  charge+jnrE+0,charge+jnrF+0,
 554                                                                  charge+jnrG+0,charge+jnrH+0);
 555             vdwjidx0A        = 2*vdwtype[jnrA+0];
 556             vdwjidx0B        = 2*vdwtype[jnrB+0];
 557             vdwjidx0C        = 2*vdwtype[jnrC+0];
 558             vdwjidx0D        = 2*vdwtype[jnrD+0];
 559             vdwjidx0E        = 2*vdwtype[jnrE+0];
 560             vdwjidx0F        = 2*vdwtype[jnrF+0];
 561             vdwjidx0G        = 2*vdwtype[jnrG+0];
 562             vdwjidx0H        = 2*vdwtype[jnrH+0];
 563
 564             fjx0             = _mm256_setzero_ps();
 565             fjy0             = _mm256_setzero_ps();
 566             fjz0             = _mm256_setzero_ps();
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             r00              = _mm256_mul_ps(rsq00,rinv00);
 573             r00              = _mm256_andnot_ps(dummy_mask,r00);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq00             = _mm256_mul_ps(iq0,jq0);
 577             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 578                                             vdwioffsetptr0+vdwjidx0B,
 579                                             vdwioffsetptr0+vdwjidx0C,
 580                                             vdwioffsetptr0+vdwjidx0D,
 581                                             vdwioffsetptr0+vdwjidx0E,
 582                                             vdwioffsetptr0+vdwjidx0F,
 583                                             vdwioffsetptr0+vdwjidx0G,
 584                                             vdwioffsetptr0+vdwjidx0H,
 585                                             &c6_00,&c12_00);
 586
 587             /* Calculate table index by multiplying r with table scale and truncate to integer */
 588             rt               = _mm256_mul_ps(r00,vftabscale);
 589             vfitab           = _mm256_cvttps_epi32(rt);
 590             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 591             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 592             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 593             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 594             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 595             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 596
 597             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 598             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 599                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 600             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 601                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 602             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 603                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 604             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 605                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 606             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 607             Heps             = _mm256_mul_ps(vfeps,H);
 608             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 609             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 610             velec            = _mm256_mul_ps(qq00,VV);
 611             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 612             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 613
 614             /* CUBIC SPLINE TABLE DISPERSION */
 615             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 616             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 617             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 618                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 619             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 620                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 621             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 622                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 623             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 624                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 625             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 626             Heps             = _mm256_mul_ps(vfeps,H);
 627             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 628             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 629             vvdw6            = _mm256_mul_ps(c6_00,VV);
 630             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 631             fvdw6            = _mm256_mul_ps(c6_00,FF);
 632
 633             /* CUBIC SPLINE TABLE REPULSION */
 634             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 635             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 636             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 637                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 638             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 639                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 640             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 641                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 642             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 643                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 644             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 645             Heps             = _mm256_mul_ps(vfeps,H);
 646             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 647             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 648             vvdw12           = _mm256_mul_ps(c12_00,VV);
 649             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 650             fvdw12           = _mm256_mul_ps(c12_00,FF);
 651             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 652             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 653
 654             /* Update potential sum for this i atom from the interaction with this j atom. */
 655             velec            = _mm256_andnot_ps(dummy_mask,velec);
 656             velecsum         = _mm256_add_ps(velecsum,velec);
 657             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 658             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 659
 660             fscal            = _mm256_add_ps(felec,fvdw);
 661
 662             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm256_mul_ps(fscal,dx00);
 666             ty               = _mm256_mul_ps(fscal,dy00);
 667             tz               = _mm256_mul_ps(fscal,dz00);
 668
 669             /* Update vectorial force */
 670             fix0             = _mm256_add_ps(fix0,tx);
 671             fiy0             = _mm256_add_ps(fiy0,ty);
 672             fiz0             = _mm256_add_ps(fiz0,tz);
 673
 674             fjx0             = _mm256_add_ps(fjx0,tx);
 675             fjy0             = _mm256_add_ps(fjy0,ty);
 676             fjz0             = _mm256_add_ps(fjz0,tz);
 677
 678             /**************************
 679              * CALCULATE INTERACTIONS *
 680              **************************/
 681
 682             r10              = _mm256_mul_ps(rsq10,rinv10);
 683             r10              = _mm256_andnot_ps(dummy_mask,r10);
 684
 685             /* Compute parameters for interactions between i and j atoms */
 686             qq10             = _mm256_mul_ps(iq1,jq0);
 687
 688             /* Calculate table index by multiplying r with table scale and truncate to integer */
 689             rt               = _mm256_mul_ps(r10,vftabscale);
 690             vfitab           = _mm256_cvttps_epi32(rt);
 691             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 692             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 693             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 694             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 695             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 696             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 697
 698             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 699             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 700                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 701             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 702                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 703             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 704                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 705             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 706                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 707             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 708             Heps             = _mm256_mul_ps(vfeps,H);
 709             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 710             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 711             velec            = _mm256_mul_ps(qq10,VV);
 712             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 713             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 714
 715             /* Update potential sum for this i atom from the interaction with this j atom. */
 716             velec            = _mm256_andnot_ps(dummy_mask,velec);
 717             velecsum         = _mm256_add_ps(velecsum,velec);
 718
 719             fscal            = felec;
 720
 721             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 722
 723             /* Calculate temporary vectorial force */
 724             tx               = _mm256_mul_ps(fscal,dx10);
 725             ty               = _mm256_mul_ps(fscal,dy10);
 726             tz               = _mm256_mul_ps(fscal,dz10);
 727
 728             /* Update vectorial force */
 729             fix1             = _mm256_add_ps(fix1,tx);
 730             fiy1             = _mm256_add_ps(fiy1,ty);
 731             fiz1             = _mm256_add_ps(fiz1,tz);
 732
 733             fjx0             = _mm256_add_ps(fjx0,tx);
 734             fjy0             = _mm256_add_ps(fjy0,ty);
 735             fjz0             = _mm256_add_ps(fjz0,tz);
 736
 737             /**************************
 738              * CALCULATE INTERACTIONS *
 739              **************************/
 740
 741             r20              = _mm256_mul_ps(rsq20,rinv20);
 742             r20              = _mm256_andnot_ps(dummy_mask,r20);
 743
 744             /* Compute parameters for interactions between i and j atoms */
 745             qq20             = _mm256_mul_ps(iq2,jq0);
 746
 747             /* Calculate table index by multiplying r with table scale and truncate to integer */
 748             rt               = _mm256_mul_ps(r20,vftabscale);
 749             vfitab           = _mm256_cvttps_epi32(rt);
 750             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 751             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 752             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 753             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 754             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 755             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 756
 757             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 758             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 759                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 760             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 761                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 762             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 763                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 764             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 765                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 766             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 767             Heps             = _mm256_mul_ps(vfeps,H);
 768             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 769             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 770             velec            = _mm256_mul_ps(qq20,VV);
 771             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 772             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 773
 774             /* Update potential sum for this i atom from the interaction with this j atom. */
 775             velec            = _mm256_andnot_ps(dummy_mask,velec);
 776             velecsum         = _mm256_add_ps(velecsum,velec);
 777
 778             fscal            = felec;
 779
 780             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 781
 782             /* Calculate temporary vectorial force */
 783             tx               = _mm256_mul_ps(fscal,dx20);
 784             ty               = _mm256_mul_ps(fscal,dy20);
 785             tz               = _mm256_mul_ps(fscal,dz20);
 786
 787             /* Update vectorial force */
 788             fix2             = _mm256_add_ps(fix2,tx);
 789             fiy2             = _mm256_add_ps(fiy2,ty);
 790             fiz2             = _mm256_add_ps(fiz2,tz);
 791
 792             fjx0             = _mm256_add_ps(fjx0,tx);
 793             fjy0             = _mm256_add_ps(fjy0,ty);
 794             fjz0             = _mm256_add_ps(fjz0,tz);
 795
 796             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 797             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 798             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 799             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 800             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 801             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 802             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 803             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 804
 805             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 806
 807             /* Inner loop uses 165 flops */
 808         }
 809
 810         /* End of innermost loop */
 811
 812         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 813                                                  f+i_coord_offset,fshift+i_shift_offset);
 814
 815         ggid                        = gid[iidx];
 816         /* Update potential energies */
 817         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 818         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 819
 820         /* Increment number of inner iterations */
 821         inneriter                  += j_index_end - j_index_start;
 822
 823         /* Outer loop uses 20 flops */
 824     }
 825
 826     /* Increment number of outer iterations */
 827     outeriter        += nri;
 828
 829     /* Update outer/inner flops */
 830
 831     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*165);
 832 }
 833 /*
 834  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_single
 835  * Electrostatics interaction: CubicSplineTable
 836  * VdW interaction:            CubicSplineTable
 837  * Geometry:                   Water3-Particle
 838  * Calculate force/pot:        Force
 839  */
 840 void
 841 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_single
 842                     (t_nblist                    * gmx_restrict       nlist,
 843                      rvec                        * gmx_restrict          xx,
 844                      rvec                        * gmx_restrict          ff,
 845                      struct t_forcerec           * gmx_restrict          fr,
 846                      t_mdatoms                   * gmx_restrict     mdatoms,
 847                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 848                      t_nrnb                      * gmx_restrict        nrnb)
 849 {
 850     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 851      * just 0 for non-waters.
 852      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 853      * jnr indices corresponding to data put in the four positions in the SIMD register.
 854      */
 855     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 856     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 857     int              jnrA,jnrB,jnrC,jnrD;
 858     int              jnrE,jnrF,jnrG,jnrH;
 859     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 860     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 861     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 862     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 863     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 864     real             rcutoff_scalar;
 865     real             *shiftvec,*fshift,*x,*f;
 866     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 867     real             scratch[4*DIM];
 868     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 869     real *           vdwioffsetptr0;
 870     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 871     real *           vdwioffsetptr1;
 872     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 873     real *           vdwioffsetptr2;
 874     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 875     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 876     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 877     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 878     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 879     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 880     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 881     real             *charge;
 882     int              nvdwtype;
 883     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 884     int              *vdwtype;
 885     real             *vdwparam;
 886     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 887     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 888     __m256i          vfitab;
 889     __m128i          vfitab_lo,vfitab_hi;
 890     __m128i          ifour       = _mm_set1_epi32(4);
 891     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 892     real             *vftab;
 893     __m256           dummy_mask,cutoff_mask;
 894     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 895     __m256           one     = _mm256_set1_ps(1.0);
 896     __m256           two     = _mm256_set1_ps(2.0);
 897     x                = xx[0];
 898     f                = ff[0];
 899
 900     nri              = nlist->nri;
 901     iinr             = nlist->iinr;
 902     jindex           = nlist->jindex;
 903     jjnr             = nlist->jjnr;
 904     shiftidx         = nlist->shift;
 905     gid              = nlist->gid;
 906     shiftvec         = fr->shift_vec[0];
 907     fshift           = fr->fshift[0];
 908     facel            = _mm256_set1_ps(fr->ic->epsfac);
 909     charge           = mdatoms->chargeA;
 910     nvdwtype         = fr->ntype;
 911     vdwparam         = fr->nbfp;
 912     vdwtype          = mdatoms->typeA;
 913
 914     vftab            = kernel_data->table_elec_vdw->data;
 915     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 916
 917     /* Setup water-specific parameters */
 918     inr              = nlist->iinr[0];
 919     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 920     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 921     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 922     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 923
 924     /* Avoid stupid compiler warnings */
 925     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 926     j_coord_offsetA = 0;
 927     j_coord_offsetB = 0;
 928     j_coord_offsetC = 0;
 929     j_coord_offsetD = 0;
 930     j_coord_offsetE = 0;
 931     j_coord_offsetF = 0;
 932     j_coord_offsetG = 0;
 933     j_coord_offsetH = 0;
 934
 935     outeriter        = 0;
 936     inneriter        = 0;
 937
 938     for(iidx=0;iidx<4*DIM;iidx++)
 939     {
 940         scratch[iidx] = 0.0;
 941     }
 942
 943     /* Start outer loop over neighborlists */
 944     for(iidx=0; iidx<nri; iidx++)
 945     {
 946         /* Load shift vector for this list */
 947         i_shift_offset   = DIM*shiftidx[iidx];
 948
 949         /* Load limits for loop over neighbors */
 950         j_index_start    = jindex[iidx];
 951         j_index_end      = jindex[iidx+1];
 952
 953         /* Get outer coordinate index */
 954         inr              = iinr[iidx];
 955         i_coord_offset   = DIM*inr;
 956
 957         /* Load i particle coords and add shift vector */
 958         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 959                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 960
 961         fix0             = _mm256_setzero_ps();
 962         fiy0             = _mm256_setzero_ps();
 963         fiz0             = _mm256_setzero_ps();
 964         fix1             = _mm256_setzero_ps();
 965         fiy1             = _mm256_setzero_ps();
 966         fiz1             = _mm256_setzero_ps();
 967         fix2             = _mm256_setzero_ps();
 968         fiy2             = _mm256_setzero_ps();
 969         fiz2             = _mm256_setzero_ps();
 970
 971         /* Start inner kernel loop */
 972         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 973         {
 974
 975             /* Get j neighbor index, and coordinate index */
 976             jnrA             = jjnr[jidx];
 977             jnrB             = jjnr[jidx+1];
 978             jnrC             = jjnr[jidx+2];
 979             jnrD             = jjnr[jidx+3];
 980             jnrE             = jjnr[jidx+4];
 981             jnrF             = jjnr[jidx+5];
 982             jnrG             = jjnr[jidx+6];
 983             jnrH             = jjnr[jidx+7];
 984             j_coord_offsetA  = DIM*jnrA;
 985             j_coord_offsetB  = DIM*jnrB;
 986             j_coord_offsetC  = DIM*jnrC;
 987             j_coord_offsetD  = DIM*jnrD;
 988             j_coord_offsetE  = DIM*jnrE;
 989             j_coord_offsetF  = DIM*jnrF;
 990             j_coord_offsetG  = DIM*jnrG;
 991             j_coord_offsetH  = DIM*jnrH;
 992
 993             /* load j atom coordinates */
 994             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 995                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 996                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 997                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 998                                                  &jx0,&jy0,&jz0);
 999
1000             /* Calculate displacement vector */
1001             dx00             = _mm256_sub_ps(ix0,jx0);
1002             dy00             = _mm256_sub_ps(iy0,jy0);
1003             dz00             = _mm256_sub_ps(iz0,jz0);
1004             dx10             = _mm256_sub_ps(ix1,jx0);
1005             dy10             = _mm256_sub_ps(iy1,jy0);
1006             dz10             = _mm256_sub_ps(iz1,jz0);
1007             dx20             = _mm256_sub_ps(ix2,jx0);
1008             dy20             = _mm256_sub_ps(iy2,jy0);
1009             dz20             = _mm256_sub_ps(iz2,jz0);
1010
1011             /* Calculate squared distance and things based on it */
1012             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1013             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1014             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1015
1016             rinv00           = avx256_invsqrt_f(rsq00);
1017             rinv10           = avx256_invsqrt_f(rsq10);
1018             rinv20           = avx256_invsqrt_f(rsq20);
1019
1020             /* Load parameters for j particles */
1021             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1022                                                                  charge+jnrC+0,charge+jnrD+0,
1023                                                                  charge+jnrE+0,charge+jnrF+0,
1024                                                                  charge+jnrG+0,charge+jnrH+0);
1025             vdwjidx0A        = 2*vdwtype[jnrA+0];
1026             vdwjidx0B        = 2*vdwtype[jnrB+0];
1027             vdwjidx0C        = 2*vdwtype[jnrC+0];
1028             vdwjidx0D        = 2*vdwtype[jnrD+0];
1029             vdwjidx0E        = 2*vdwtype[jnrE+0];
1030             vdwjidx0F        = 2*vdwtype[jnrF+0];
1031             vdwjidx0G        = 2*vdwtype[jnrG+0];
1032             vdwjidx0H        = 2*vdwtype[jnrH+0];
1033
1034             fjx0             = _mm256_setzero_ps();
1035             fjy0             = _mm256_setzero_ps();
1036             fjz0             = _mm256_setzero_ps();
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             r00              = _mm256_mul_ps(rsq00,rinv00);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq00             = _mm256_mul_ps(iq0,jq0);
1046             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1047                                             vdwioffsetptr0+vdwjidx0B,
1048                                             vdwioffsetptr0+vdwjidx0C,
1049                                             vdwioffsetptr0+vdwjidx0D,
1050                                             vdwioffsetptr0+vdwjidx0E,
1051                                             vdwioffsetptr0+vdwjidx0F,
1052                                             vdwioffsetptr0+vdwjidx0G,
1053                                             vdwioffsetptr0+vdwjidx0H,
1054                                             &c6_00,&c12_00);
1055
1056             /* Calculate table index by multiplying r with table scale and truncate to integer */
1057             rt               = _mm256_mul_ps(r00,vftabscale);
1058             vfitab           = _mm256_cvttps_epi32(rt);
1059             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1060             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1061             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1062             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1063             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1064             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1065
1066             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1067             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1068                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1069             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1070                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1071             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1072                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1073             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1074                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1075             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1076             Heps             = _mm256_mul_ps(vfeps,H);
1077             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1078             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1079             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1080
1081             /* CUBIC SPLINE TABLE DISPERSION */
1082             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1083             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1084             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1085                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1086             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1087                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1088             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1089                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1090             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1091                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1092             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1093             Heps             = _mm256_mul_ps(vfeps,H);
1094             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1095             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1096             fvdw6            = _mm256_mul_ps(c6_00,FF);
1097
1098             /* CUBIC SPLINE TABLE REPULSION */
1099             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1100             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1101             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1102                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1103             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1104                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1105             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1106                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1107             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1108                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1109             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1110             Heps             = _mm256_mul_ps(vfeps,H);
1111             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1112             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1113             fvdw12           = _mm256_mul_ps(c12_00,FF);
1114             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1115
1116             fscal            = _mm256_add_ps(felec,fvdw);
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm256_mul_ps(fscal,dx00);
1120             ty               = _mm256_mul_ps(fscal,dy00);
1121             tz               = _mm256_mul_ps(fscal,dz00);
1122
1123             /* Update vectorial force */
1124             fix0             = _mm256_add_ps(fix0,tx);
1125             fiy0             = _mm256_add_ps(fiy0,ty);
1126             fiz0             = _mm256_add_ps(fiz0,tz);
1127
1128             fjx0             = _mm256_add_ps(fjx0,tx);
1129             fjy0             = _mm256_add_ps(fjy0,ty);
1130             fjz0             = _mm256_add_ps(fjz0,tz);
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             r10              = _mm256_mul_ps(rsq10,rinv10);
1137
1138             /* Compute parameters for interactions between i and j atoms */
1139             qq10             = _mm256_mul_ps(iq1,jq0);
1140
1141             /* Calculate table index by multiplying r with table scale and truncate to integer */
1142             rt               = _mm256_mul_ps(r10,vftabscale);
1143             vfitab           = _mm256_cvttps_epi32(rt);
1144             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1145             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1146             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1147             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1148             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1149             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1150
1151             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1152             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1153                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1154             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1155                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1156             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1157                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1158             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1159                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1160             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1161             Heps             = _mm256_mul_ps(vfeps,H);
1162             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1163             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1164             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1165
1166             fscal            = felec;
1167
1168             /* Calculate temporary vectorial force */
1169             tx               = _mm256_mul_ps(fscal,dx10);
1170             ty               = _mm256_mul_ps(fscal,dy10);
1171             tz               = _mm256_mul_ps(fscal,dz10);
1172
1173             /* Update vectorial force */
1174             fix1             = _mm256_add_ps(fix1,tx);
1175             fiy1             = _mm256_add_ps(fiy1,ty);
1176             fiz1             = _mm256_add_ps(fiz1,tz);
1177
1178             fjx0             = _mm256_add_ps(fjx0,tx);
1179             fjy0             = _mm256_add_ps(fjy0,ty);
1180             fjz0             = _mm256_add_ps(fjz0,tz);
1181
1182             /**************************
1183              * CALCULATE INTERACTIONS *
1184              **************************/
1185
1186             r20              = _mm256_mul_ps(rsq20,rinv20);
1187
1188             /* Compute parameters for interactions between i and j atoms */
1189             qq20             = _mm256_mul_ps(iq2,jq0);
1190
1191             /* Calculate table index by multiplying r with table scale and truncate to integer */
1192             rt               = _mm256_mul_ps(r20,vftabscale);
1193             vfitab           = _mm256_cvttps_epi32(rt);
1194             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1195             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1196             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1197             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1198             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1199             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1200
1201             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1202             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1203                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1204             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1205                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1206             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1207                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1208             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1209                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1210             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1211             Heps             = _mm256_mul_ps(vfeps,H);
1212             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1213             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1214             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1215
1216             fscal            = felec;
1217
1218             /* Calculate temporary vectorial force */
1219             tx               = _mm256_mul_ps(fscal,dx20);
1220             ty               = _mm256_mul_ps(fscal,dy20);
1221             tz               = _mm256_mul_ps(fscal,dz20);
1222
1223             /* Update vectorial force */
1224             fix2             = _mm256_add_ps(fix2,tx);
1225             fiy2             = _mm256_add_ps(fiy2,ty);
1226             fiz2             = _mm256_add_ps(fiz2,tz);
1227
1228             fjx0             = _mm256_add_ps(fjx0,tx);
1229             fjy0             = _mm256_add_ps(fjy0,ty);
1230             fjz0             = _mm256_add_ps(fjz0,tz);
1231
1232             fjptrA             = f+j_coord_offsetA;
1233             fjptrB             = f+j_coord_offsetB;
1234             fjptrC             = f+j_coord_offsetC;
1235             fjptrD             = f+j_coord_offsetD;
1236             fjptrE             = f+j_coord_offsetE;
1237             fjptrF             = f+j_coord_offsetF;
1238             fjptrG             = f+j_coord_offsetG;
1239             fjptrH             = f+j_coord_offsetH;
1240
1241             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1242
1243             /* Inner loop uses 142 flops */
1244         }
1245
1246         if(jidx<j_index_end)
1247         {
1248
1249             /* Get j neighbor index, and coordinate index */
1250             jnrlistA         = jjnr[jidx];
1251             jnrlistB         = jjnr[jidx+1];
1252             jnrlistC         = jjnr[jidx+2];
1253             jnrlistD         = jjnr[jidx+3];
1254             jnrlistE         = jjnr[jidx+4];
1255             jnrlistF         = jjnr[jidx+5];
1256             jnrlistG         = jjnr[jidx+6];
1257             jnrlistH         = jjnr[jidx+7];
1258             /* Sign of each element will be negative for non-real atoms.
1259              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1260              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1261              */
1262             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1263                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1264
1265             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1266             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1267             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1268             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1269             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1270             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1271             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1272             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1273             j_coord_offsetA  = DIM*jnrA;
1274             j_coord_offsetB  = DIM*jnrB;
1275             j_coord_offsetC  = DIM*jnrC;
1276             j_coord_offsetD  = DIM*jnrD;
1277             j_coord_offsetE  = DIM*jnrE;
1278             j_coord_offsetF  = DIM*jnrF;
1279             j_coord_offsetG  = DIM*jnrG;
1280             j_coord_offsetH  = DIM*jnrH;
1281
1282             /* load j atom coordinates */
1283             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1284                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1285                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1286                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1287                                                  &jx0,&jy0,&jz0);
1288
1289             /* Calculate displacement vector */
1290             dx00             = _mm256_sub_ps(ix0,jx0);
1291             dy00             = _mm256_sub_ps(iy0,jy0);
1292             dz00             = _mm256_sub_ps(iz0,jz0);
1293             dx10             = _mm256_sub_ps(ix1,jx0);
1294             dy10             = _mm256_sub_ps(iy1,jy0);
1295             dz10             = _mm256_sub_ps(iz1,jz0);
1296             dx20             = _mm256_sub_ps(ix2,jx0);
1297             dy20             = _mm256_sub_ps(iy2,jy0);
1298             dz20             = _mm256_sub_ps(iz2,jz0);
1299
1300             /* Calculate squared distance and things based on it */
1301             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1302             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1303             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1304
1305             rinv00           = avx256_invsqrt_f(rsq00);
1306             rinv10           = avx256_invsqrt_f(rsq10);
1307             rinv20           = avx256_invsqrt_f(rsq20);
1308
1309             /* Load parameters for j particles */
1310             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1311                                                                  charge+jnrC+0,charge+jnrD+0,
1312                                                                  charge+jnrE+0,charge+jnrF+0,
1313                                                                  charge+jnrG+0,charge+jnrH+0);
1314             vdwjidx0A        = 2*vdwtype[jnrA+0];
1315             vdwjidx0B        = 2*vdwtype[jnrB+0];
1316             vdwjidx0C        = 2*vdwtype[jnrC+0];
1317             vdwjidx0D        = 2*vdwtype[jnrD+0];
1318             vdwjidx0E        = 2*vdwtype[jnrE+0];
1319             vdwjidx0F        = 2*vdwtype[jnrF+0];
1320             vdwjidx0G        = 2*vdwtype[jnrG+0];
1321             vdwjidx0H        = 2*vdwtype[jnrH+0];
1322
1323             fjx0             = _mm256_setzero_ps();
1324             fjy0             = _mm256_setzero_ps();
1325             fjz0             = _mm256_setzero_ps();
1326
1327             /**************************
1328              * CALCULATE INTERACTIONS *
1329              **************************/
1330
1331             r00              = _mm256_mul_ps(rsq00,rinv00);
1332             r00              = _mm256_andnot_ps(dummy_mask,r00);
1333
1334             /* Compute parameters for interactions between i and j atoms */
1335             qq00             = _mm256_mul_ps(iq0,jq0);
1336             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1337                                             vdwioffsetptr0+vdwjidx0B,
1338                                             vdwioffsetptr0+vdwjidx0C,
1339                                             vdwioffsetptr0+vdwjidx0D,
1340                                             vdwioffsetptr0+vdwjidx0E,
1341                                             vdwioffsetptr0+vdwjidx0F,
1342                                             vdwioffsetptr0+vdwjidx0G,
1343                                             vdwioffsetptr0+vdwjidx0H,
1344                                             &c6_00,&c12_00);
1345
1346             /* Calculate table index by multiplying r with table scale and truncate to integer */
1347             rt               = _mm256_mul_ps(r00,vftabscale);
1348             vfitab           = _mm256_cvttps_epi32(rt);
1349             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1350             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1351             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1352             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1353             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1354             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1355
1356             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1357             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1358                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1359             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1360                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1361             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1362                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1363             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1364                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1365             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1366             Heps             = _mm256_mul_ps(vfeps,H);
1367             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1368             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1369             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1370
1371             /* CUBIC SPLINE TABLE DISPERSION */
1372             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1373             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1374             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1375                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1376             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1377                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1378             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1379                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1380             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1381                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1382             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1383             Heps             = _mm256_mul_ps(vfeps,H);
1384             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1385             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1386             fvdw6            = _mm256_mul_ps(c6_00,FF);
1387
1388             /* CUBIC SPLINE TABLE REPULSION */
1389             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1390             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1391             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1392                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1393             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1394                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1395             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1396                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1397             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1398                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1399             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1400             Heps             = _mm256_mul_ps(vfeps,H);
1401             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1402             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1403             fvdw12           = _mm256_mul_ps(c12_00,FF);
1404             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1405
1406             fscal            = _mm256_add_ps(felec,fvdw);
1407
1408             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1409
1410             /* Calculate temporary vectorial force */
1411             tx               = _mm256_mul_ps(fscal,dx00);
1412             ty               = _mm256_mul_ps(fscal,dy00);
1413             tz               = _mm256_mul_ps(fscal,dz00);
1414
1415             /* Update vectorial force */
1416             fix0             = _mm256_add_ps(fix0,tx);
1417             fiy0             = _mm256_add_ps(fiy0,ty);
1418             fiz0             = _mm256_add_ps(fiz0,tz);
1419
1420             fjx0             = _mm256_add_ps(fjx0,tx);
1421             fjy0             = _mm256_add_ps(fjy0,ty);
1422             fjz0             = _mm256_add_ps(fjz0,tz);
1423
1424             /**************************
1425              * CALCULATE INTERACTIONS *
1426              **************************/
1427
1428             r10              = _mm256_mul_ps(rsq10,rinv10);
1429             r10              = _mm256_andnot_ps(dummy_mask,r10);
1430
1431             /* Compute parameters for interactions between i and j atoms */
1432             qq10             = _mm256_mul_ps(iq1,jq0);
1433
1434             /* Calculate table index by multiplying r with table scale and truncate to integer */
1435             rt               = _mm256_mul_ps(r10,vftabscale);
1436             vfitab           = _mm256_cvttps_epi32(rt);
1437             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1438             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1439             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1440             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1441             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1442             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1443
1444             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1445             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1446                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1447             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1448                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1449             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1450                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1451             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1452                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1453             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1454             Heps             = _mm256_mul_ps(vfeps,H);
1455             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1456             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1457             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1458
1459             fscal            = felec;
1460
1461             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1462
1463             /* Calculate temporary vectorial force */
1464             tx               = _mm256_mul_ps(fscal,dx10);
1465             ty               = _mm256_mul_ps(fscal,dy10);
1466             tz               = _mm256_mul_ps(fscal,dz10);
1467
1468             /* Update vectorial force */
1469             fix1             = _mm256_add_ps(fix1,tx);
1470             fiy1             = _mm256_add_ps(fiy1,ty);
1471             fiz1             = _mm256_add_ps(fiz1,tz);
1472
1473             fjx0             = _mm256_add_ps(fjx0,tx);
1474             fjy0             = _mm256_add_ps(fjy0,ty);
1475             fjz0             = _mm256_add_ps(fjz0,tz);
1476
1477             /**************************
1478              * CALCULATE INTERACTIONS *
1479              **************************/
1480
1481             r20              = _mm256_mul_ps(rsq20,rinv20);
1482             r20              = _mm256_andnot_ps(dummy_mask,r20);
1483
1484             /* Compute parameters for interactions between i and j atoms */
1485             qq20             = _mm256_mul_ps(iq2,jq0);
1486
1487             /* Calculate table index by multiplying r with table scale and truncate to integer */
1488             rt               = _mm256_mul_ps(r20,vftabscale);
1489             vfitab           = _mm256_cvttps_epi32(rt);
1490             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1491             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1492             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1493             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1494             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1495             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1496
1497             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1498             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1499                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1500             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1501                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1502             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1503                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1504             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1505                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1506             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1507             Heps             = _mm256_mul_ps(vfeps,H);
1508             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1509             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1510             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1511
1512             fscal            = felec;
1513
1514             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1515
1516             /* Calculate temporary vectorial force */
1517             tx               = _mm256_mul_ps(fscal,dx20);
1518             ty               = _mm256_mul_ps(fscal,dy20);
1519             tz               = _mm256_mul_ps(fscal,dz20);
1520
1521             /* Update vectorial force */
1522             fix2             = _mm256_add_ps(fix2,tx);
1523             fiy2             = _mm256_add_ps(fiy2,ty);
1524             fiz2             = _mm256_add_ps(fiz2,tz);
1525
1526             fjx0             = _mm256_add_ps(fjx0,tx);
1527             fjy0             = _mm256_add_ps(fjy0,ty);
1528             fjz0             = _mm256_add_ps(fjz0,tz);
1529
1530             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1531             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1532             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1533             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1534             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1535             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1536             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1537             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1538
1539             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1540
1541             /* Inner loop uses 145 flops */
1542         }
1543
1544         /* End of innermost loop */
1545
1546         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1547                                                  f+i_coord_offset,fshift+i_shift_offset);
1548
1549         /* Increment number of inner iterations */
1550         inneriter                  += j_index_end - j_index_start;
1551
1552         /* Outer loop uses 18 flops */
1553     }
1554
1555     /* Increment number of outer iterations */
1556     outeriter        += nri;
1557
1558     /* Update outer/inner flops */
1559
1560     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*145);
1561 }