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

   1 /*
   2  * Note: this file was generated by the Gromacs avx_256_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_256_single.h"
  34 #include "kernelutil_x86_avx_256_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrE,jnrF,jnrG,jnrH;
  62     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  63     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  64     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  65     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  66     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  67     real             rcutoff_scalar;
  68     real             *shiftvec,*fshift,*x,*f;
  69     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  70     real             scratch[4*DIM];
  71     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  72     real *           vdwioffsetptr0;
  73     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  74     real *           vdwioffsetptr1;
  75     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  76     real *           vdwioffsetptr2;
  77     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  78     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  79     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  80     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  81     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  82     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  83     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  90     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  91     __m256i          vfitab;
  92     __m128i          vfitab_lo,vfitab_hi;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  96     __m256           dummy_mask,cutoff_mask;
  97     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  98     __m256           one     = _mm256_set1_ps(1.0);
  99     __m256           two     = _mm256_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm256_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     vftab            = kernel_data->table_elec_vdw->data;
 118     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 119
 120     /* Setup water-specific parameters */
 121     inr              = nlist->iinr[0];
 122     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 123     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 124     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 125     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133     j_coord_offsetE = 0;
 134     j_coord_offsetF = 0;
 135     j_coord_offsetG = 0;
 136     j_coord_offsetH = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 163
 164         fix0             = _mm256_setzero_ps();
 165         fiy0             = _mm256_setzero_ps();
 166         fiz0             = _mm256_setzero_ps();
 167         fix1             = _mm256_setzero_ps();
 168         fiy1             = _mm256_setzero_ps();
 169         fiz1             = _mm256_setzero_ps();
 170         fix2             = _mm256_setzero_ps();
 171         fiy2             = _mm256_setzero_ps();
 172         fiz2             = _mm256_setzero_ps();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm256_setzero_ps();
 176         vvdwsum          = _mm256_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             jnrE             = jjnr[jidx+4];
 188             jnrF             = jjnr[jidx+5];
 189             jnrG             = jjnr[jidx+6];
 190             jnrH             = jjnr[jidx+7];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195             j_coord_offsetE  = DIM*jnrE;
 196             j_coord_offsetF  = DIM*jnrF;
 197             j_coord_offsetG  = DIM*jnrG;
 198             j_coord_offsetH  = DIM*jnrH;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 204                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_ps(ix0,jx0);
 209             dy00             = _mm256_sub_ps(iy0,jy0);
 210             dz00             = _mm256_sub_ps(iz0,jz0);
 211             dx10             = _mm256_sub_ps(ix1,jx0);
 212             dy10             = _mm256_sub_ps(iy1,jy0);
 213             dz10             = _mm256_sub_ps(iz1,jz0);
 214             dx20             = _mm256_sub_ps(ix2,jx0);
 215             dy20             = _mm256_sub_ps(iy2,jy0);
 216             dz20             = _mm256_sub_ps(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 220             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 221             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 222
 223             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 224             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 229                                                                  charge+jnrC+0,charge+jnrD+0,
 230                                                                  charge+jnrE+0,charge+jnrF+0,
 231                                                                  charge+jnrG+0,charge+jnrH+0);
 232             vdwjidx0A        = 2*vdwtype[jnrA+0];
 233             vdwjidx0B        = 2*vdwtype[jnrB+0];
 234             vdwjidx0C        = 2*vdwtype[jnrC+0];
 235             vdwjidx0D        = 2*vdwtype[jnrD+0];
 236             vdwjidx0E        = 2*vdwtype[jnrE+0];
 237             vdwjidx0F        = 2*vdwtype[jnrF+0];
 238             vdwjidx0G        = 2*vdwtype[jnrG+0];
 239             vdwjidx0H        = 2*vdwtype[jnrH+0];
 240
 241             fjx0             = _mm256_setzero_ps();
 242             fjy0             = _mm256_setzero_ps();
 243             fjz0             = _mm256_setzero_ps();
 244
 245             /**************************
 246              * CALCULATE INTERACTIONS *
 247              **************************/
 248
 249             r00              = _mm256_mul_ps(rsq00,rinv00);
 250
 251             /* Compute parameters for interactions between i and j atoms */
 252             qq00             = _mm256_mul_ps(iq0,jq0);
 253             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 254                                             vdwioffsetptr0+vdwjidx0B,
 255                                             vdwioffsetptr0+vdwjidx0C,
 256                                             vdwioffsetptr0+vdwjidx0D,
 257                                             vdwioffsetptr0+vdwjidx0E,
 258                                             vdwioffsetptr0+vdwjidx0F,
 259                                             vdwioffsetptr0+vdwjidx0G,
 260                                             vdwioffsetptr0+vdwjidx0H,
 261                                             &c6_00,&c12_00);
 262
 263             /* Calculate table index by multiplying r with table scale and truncate to integer */
 264             rt               = _mm256_mul_ps(r00,vftabscale);
 265             vfitab           = _mm256_cvttps_epi32(rt);
 266             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 267             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 268             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 269             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 270             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 271             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 272
 273             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 274             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 275                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 276             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 277                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 278             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 279                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 280             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 281                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 282             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 283             Heps             = _mm256_mul_ps(vfeps,H);
 284             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 285             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 286             velec            = _mm256_mul_ps(qq00,VV);
 287             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 288             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 289
 290             /* CUBIC SPLINE TABLE DISPERSION */
 291             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 292             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 293             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 294                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 295             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 296                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 297             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 298                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 299             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 300                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 301             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 302             Heps             = _mm256_mul_ps(vfeps,H);
 303             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 304             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 305             vvdw6            = _mm256_mul_ps(c6_00,VV);
 306             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 307             fvdw6            = _mm256_mul_ps(c6_00,FF);
 308
 309             /* CUBIC SPLINE TABLE REPULSION */
 310             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 311             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 312             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 313                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 314             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 315                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 316             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 317                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 318             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 319                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 320             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 321             Heps             = _mm256_mul_ps(vfeps,H);
 322             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 323             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 324             vvdw12           = _mm256_mul_ps(c12_00,VV);
 325             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 326             fvdw12           = _mm256_mul_ps(c12_00,FF);
 327             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 328             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm256_add_ps(velecsum,velec);
 332             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 333
 334             fscal            = _mm256_add_ps(felec,fvdw);
 335
 336             /* Calculate temporary vectorial force */
 337             tx               = _mm256_mul_ps(fscal,dx00);
 338             ty               = _mm256_mul_ps(fscal,dy00);
 339             tz               = _mm256_mul_ps(fscal,dz00);
 340
 341             /* Update vectorial force */
 342             fix0             = _mm256_add_ps(fix0,tx);
 343             fiy0             = _mm256_add_ps(fiy0,ty);
 344             fiz0             = _mm256_add_ps(fiz0,tz);
 345
 346             fjx0             = _mm256_add_ps(fjx0,tx);
 347             fjy0             = _mm256_add_ps(fjy0,ty);
 348             fjz0             = _mm256_add_ps(fjz0,tz);
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             r10              = _mm256_mul_ps(rsq10,rinv10);
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq10             = _mm256_mul_ps(iq1,jq0);
 358
 359             /* Calculate table index by multiplying r with table scale and truncate to integer */
 360             rt               = _mm256_mul_ps(r10,vftabscale);
 361             vfitab           = _mm256_cvttps_epi32(rt);
 362             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 363             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 364             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 365             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 366             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 367             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 368
 369             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 370             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 371                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 372             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 373                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 374             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 375                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 376             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 377                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 378             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 379             Heps             = _mm256_mul_ps(vfeps,H);
 380             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 381             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 382             velec            = _mm256_mul_ps(qq10,VV);
 383             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 384             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velecsum         = _mm256_add_ps(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             /* Calculate temporary vectorial force */
 392             tx               = _mm256_mul_ps(fscal,dx10);
 393             ty               = _mm256_mul_ps(fscal,dy10);
 394             tz               = _mm256_mul_ps(fscal,dz10);
 395
 396             /* Update vectorial force */
 397             fix1             = _mm256_add_ps(fix1,tx);
 398             fiy1             = _mm256_add_ps(fiy1,ty);
 399             fiz1             = _mm256_add_ps(fiz1,tz);
 400
 401             fjx0             = _mm256_add_ps(fjx0,tx);
 402             fjy0             = _mm256_add_ps(fjy0,ty);
 403             fjz0             = _mm256_add_ps(fjz0,tz);
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             r20              = _mm256_mul_ps(rsq20,rinv20);
 410
 411             /* Compute parameters for interactions between i and j atoms */
 412             qq20             = _mm256_mul_ps(iq2,jq0);
 413
 414             /* Calculate table index by multiplying r with table scale and truncate to integer */
 415             rt               = _mm256_mul_ps(r20,vftabscale);
 416             vfitab           = _mm256_cvttps_epi32(rt);
 417             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 418             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 419             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 420             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 421             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 422             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 423
 424             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 425             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 426                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 427             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 428                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 429             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 430                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 431             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 432                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 433             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 434             Heps             = _mm256_mul_ps(vfeps,H);
 435             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 436             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 437             velec            = _mm256_mul_ps(qq20,VV);
 438             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 439             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 440
 441             /* Update potential sum for this i atom from the interaction with this j atom. */
 442             velecsum         = _mm256_add_ps(velecsum,velec);
 443
 444             fscal            = felec;
 445
 446             /* Calculate temporary vectorial force */
 447             tx               = _mm256_mul_ps(fscal,dx20);
 448             ty               = _mm256_mul_ps(fscal,dy20);
 449             tz               = _mm256_mul_ps(fscal,dz20);
 450
 451             /* Update vectorial force */
 452             fix2             = _mm256_add_ps(fix2,tx);
 453             fiy2             = _mm256_add_ps(fiy2,ty);
 454             fiz2             = _mm256_add_ps(fiz2,tz);
 455
 456             fjx0             = _mm256_add_ps(fjx0,tx);
 457             fjy0             = _mm256_add_ps(fjy0,ty);
 458             fjz0             = _mm256_add_ps(fjz0,tz);
 459
 460             fjptrA             = f+j_coord_offsetA;
 461             fjptrB             = f+j_coord_offsetB;
 462             fjptrC             = f+j_coord_offsetC;
 463             fjptrD             = f+j_coord_offsetD;
 464             fjptrE             = f+j_coord_offsetE;
 465             fjptrF             = f+j_coord_offsetF;
 466             fjptrG             = f+j_coord_offsetG;
 467             fjptrH             = f+j_coord_offsetH;
 468
 469             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 470
 471             /* Inner loop uses 162 flops */
 472         }
 473
 474         if(jidx<j_index_end)
 475         {
 476
 477             /* Get j neighbor index, and coordinate index */
 478             jnrlistA         = jjnr[jidx];
 479             jnrlistB         = jjnr[jidx+1];
 480             jnrlistC         = jjnr[jidx+2];
 481             jnrlistD         = jjnr[jidx+3];
 482             jnrlistE         = jjnr[jidx+4];
 483             jnrlistF         = jjnr[jidx+5];
 484             jnrlistG         = jjnr[jidx+6];
 485             jnrlistH         = jjnr[jidx+7];
 486             /* Sign of each element will be negative for non-real atoms.
 487              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 488              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 489              */
 490             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 491                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 492
 493             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 494             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 495             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 496             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 497             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 498             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 499             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 500             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 501             j_coord_offsetA  = DIM*jnrA;
 502             j_coord_offsetB  = DIM*jnrB;
 503             j_coord_offsetC  = DIM*jnrC;
 504             j_coord_offsetD  = DIM*jnrD;
 505             j_coord_offsetE  = DIM*jnrE;
 506             j_coord_offsetF  = DIM*jnrF;
 507             j_coord_offsetG  = DIM*jnrG;
 508             j_coord_offsetH  = DIM*jnrH;
 509
 510             /* load j atom coordinates */
 511             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 512                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 513                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 514                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 515                                                  &jx0,&jy0,&jz0);
 516
 517             /* Calculate displacement vector */
 518             dx00             = _mm256_sub_ps(ix0,jx0);
 519             dy00             = _mm256_sub_ps(iy0,jy0);
 520             dz00             = _mm256_sub_ps(iz0,jz0);
 521             dx10             = _mm256_sub_ps(ix1,jx0);
 522             dy10             = _mm256_sub_ps(iy1,jy0);
 523             dz10             = _mm256_sub_ps(iz1,jz0);
 524             dx20             = _mm256_sub_ps(ix2,jx0);
 525             dy20             = _mm256_sub_ps(iy2,jy0);
 526             dz20             = _mm256_sub_ps(iz2,jz0);
 527
 528             /* Calculate squared distance and things based on it */
 529             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 530             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 531             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 532
 533             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 534             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 535             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 536
 537             /* Load parameters for j particles */
 538             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 539                                                                  charge+jnrC+0,charge+jnrD+0,
 540                                                                  charge+jnrE+0,charge+jnrF+0,
 541                                                                  charge+jnrG+0,charge+jnrH+0);
 542             vdwjidx0A        = 2*vdwtype[jnrA+0];
 543             vdwjidx0B        = 2*vdwtype[jnrB+0];
 544             vdwjidx0C        = 2*vdwtype[jnrC+0];
 545             vdwjidx0D        = 2*vdwtype[jnrD+0];
 546             vdwjidx0E        = 2*vdwtype[jnrE+0];
 547             vdwjidx0F        = 2*vdwtype[jnrF+0];
 548             vdwjidx0G        = 2*vdwtype[jnrG+0];
 549             vdwjidx0H        = 2*vdwtype[jnrH+0];
 550
 551             fjx0             = _mm256_setzero_ps();
 552             fjy0             = _mm256_setzero_ps();
 553             fjz0             = _mm256_setzero_ps();
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             r00              = _mm256_mul_ps(rsq00,rinv00);
 560             r00              = _mm256_andnot_ps(dummy_mask,r00);
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq00             = _mm256_mul_ps(iq0,jq0);
 564             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 565                                             vdwioffsetptr0+vdwjidx0B,
 566                                             vdwioffsetptr0+vdwjidx0C,
 567                                             vdwioffsetptr0+vdwjidx0D,
 568                                             vdwioffsetptr0+vdwjidx0E,
 569                                             vdwioffsetptr0+vdwjidx0F,
 570                                             vdwioffsetptr0+vdwjidx0G,
 571                                             vdwioffsetptr0+vdwjidx0H,
 572                                             &c6_00,&c12_00);
 573
 574             /* Calculate table index by multiplying r with table scale and truncate to integer */
 575             rt               = _mm256_mul_ps(r00,vftabscale);
 576             vfitab           = _mm256_cvttps_epi32(rt);
 577             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 578             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 579             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 580             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 581             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 582             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 583
 584             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 585             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 586                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 587             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 588                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 589             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 590                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 591             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 592                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 593             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 594             Heps             = _mm256_mul_ps(vfeps,H);
 595             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 596             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 597             velec            = _mm256_mul_ps(qq00,VV);
 598             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 599             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
 600
 601             /* CUBIC SPLINE TABLE DISPERSION */
 602             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 603             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 604             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 605                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 606             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 607                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 608             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 609                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 610             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 611                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 612             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 613             Heps             = _mm256_mul_ps(vfeps,H);
 614             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 615             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 616             vvdw6            = _mm256_mul_ps(c6_00,VV);
 617             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 618             fvdw6            = _mm256_mul_ps(c6_00,FF);
 619
 620             /* CUBIC SPLINE TABLE REPULSION */
 621             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 622             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 623             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 624                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 625             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 626                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 627             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 628                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 629             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 630                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 631             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 632             Heps             = _mm256_mul_ps(vfeps,H);
 633             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 634             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 635             vvdw12           = _mm256_mul_ps(c12_00,VV);
 636             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 637             fvdw12           = _mm256_mul_ps(c12_00,FF);
 638             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 639             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 640
 641             /* Update potential sum for this i atom from the interaction with this j atom. */
 642             velec            = _mm256_andnot_ps(dummy_mask,velec);
 643             velecsum         = _mm256_add_ps(velecsum,velec);
 644             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 645             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 646
 647             fscal            = _mm256_add_ps(felec,fvdw);
 648
 649             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 650
 651             /* Calculate temporary vectorial force */
 652             tx               = _mm256_mul_ps(fscal,dx00);
 653             ty               = _mm256_mul_ps(fscal,dy00);
 654             tz               = _mm256_mul_ps(fscal,dz00);
 655
 656             /* Update vectorial force */
 657             fix0             = _mm256_add_ps(fix0,tx);
 658             fiy0             = _mm256_add_ps(fiy0,ty);
 659             fiz0             = _mm256_add_ps(fiz0,tz);
 660
 661             fjx0             = _mm256_add_ps(fjx0,tx);
 662             fjy0             = _mm256_add_ps(fjy0,ty);
 663             fjz0             = _mm256_add_ps(fjz0,tz);
 664
 665             /**************************
 666              * CALCULATE INTERACTIONS *
 667              **************************/
 668
 669             r10              = _mm256_mul_ps(rsq10,rinv10);
 670             r10              = _mm256_andnot_ps(dummy_mask,r10);
 671
 672             /* Compute parameters for interactions between i and j atoms */
 673             qq10             = _mm256_mul_ps(iq1,jq0);
 674
 675             /* Calculate table index by multiplying r with table scale and truncate to integer */
 676             rt               = _mm256_mul_ps(r10,vftabscale);
 677             vfitab           = _mm256_cvttps_epi32(rt);
 678             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 679             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 680             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 681             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 682             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 683             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 684
 685             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 686             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 687                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 688             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 689                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 690             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 691                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 692             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 693                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 694             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 695             Heps             = _mm256_mul_ps(vfeps,H);
 696             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 697             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 698             velec            = _mm256_mul_ps(qq10,VV);
 699             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 700             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
 701
 702             /* Update potential sum for this i atom from the interaction with this j atom. */
 703             velec            = _mm256_andnot_ps(dummy_mask,velec);
 704             velecsum         = _mm256_add_ps(velecsum,velec);
 705
 706             fscal            = felec;
 707
 708             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 709
 710             /* Calculate temporary vectorial force */
 711             tx               = _mm256_mul_ps(fscal,dx10);
 712             ty               = _mm256_mul_ps(fscal,dy10);
 713             tz               = _mm256_mul_ps(fscal,dz10);
 714
 715             /* Update vectorial force */
 716             fix1             = _mm256_add_ps(fix1,tx);
 717             fiy1             = _mm256_add_ps(fiy1,ty);
 718             fiz1             = _mm256_add_ps(fiz1,tz);
 719
 720             fjx0             = _mm256_add_ps(fjx0,tx);
 721             fjy0             = _mm256_add_ps(fjy0,ty);
 722             fjz0             = _mm256_add_ps(fjz0,tz);
 723
 724             /**************************
 725              * CALCULATE INTERACTIONS *
 726              **************************/
 727
 728             r20              = _mm256_mul_ps(rsq20,rinv20);
 729             r20              = _mm256_andnot_ps(dummy_mask,r20);
 730
 731             /* Compute parameters for interactions between i and j atoms */
 732             qq20             = _mm256_mul_ps(iq2,jq0);
 733
 734             /* Calculate table index by multiplying r with table scale and truncate to integer */
 735             rt               = _mm256_mul_ps(r20,vftabscale);
 736             vfitab           = _mm256_cvttps_epi32(rt);
 737             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 738             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 739             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 740             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 741             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 742             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 743
 744             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 745             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 746                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 747             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 748                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 749             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 750                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 751             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 752                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 753             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 754             Heps             = _mm256_mul_ps(vfeps,H);
 755             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 756             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 757             velec            = _mm256_mul_ps(qq20,VV);
 758             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 759             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
 760
 761             /* Update potential sum for this i atom from the interaction with this j atom. */
 762             velec            = _mm256_andnot_ps(dummy_mask,velec);
 763             velecsum         = _mm256_add_ps(velecsum,velec);
 764
 765             fscal            = felec;
 766
 767             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 768
 769             /* Calculate temporary vectorial force */
 770             tx               = _mm256_mul_ps(fscal,dx20);
 771             ty               = _mm256_mul_ps(fscal,dy20);
 772             tz               = _mm256_mul_ps(fscal,dz20);
 773
 774             /* Update vectorial force */
 775             fix2             = _mm256_add_ps(fix2,tx);
 776             fiy2             = _mm256_add_ps(fiy2,ty);
 777             fiz2             = _mm256_add_ps(fiz2,tz);
 778
 779             fjx0             = _mm256_add_ps(fjx0,tx);
 780             fjy0             = _mm256_add_ps(fjy0,ty);
 781             fjz0             = _mm256_add_ps(fjz0,tz);
 782
 783             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 784             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 785             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 786             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 787             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 788             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 789             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 790             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 791
 792             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 793
 794             /* Inner loop uses 165 flops */
 795         }
 796
 797         /* End of innermost loop */
 798
 799         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 800                                                  f+i_coord_offset,fshift+i_shift_offset);
 801
 802         ggid                        = gid[iidx];
 803         /* Update potential energies */
 804         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 805         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 806
 807         /* Increment number of inner iterations */
 808         inneriter                  += j_index_end - j_index_start;
 809
 810         /* Outer loop uses 20 flops */
 811     }
 812
 813     /* Increment number of outer iterations */
 814     outeriter        += nri;
 815
 816     /* Update outer/inner flops */
 817
 818     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*165);
 819 }
 820 /*
 821  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_single
 822  * Electrostatics interaction: CubicSplineTable
 823  * VdW interaction:            CubicSplineTable
 824  * Geometry:                   Water3-Particle
 825  * Calculate force/pot:        Force
 826  */
 827 void
 828 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_single
 829                     (t_nblist * gmx_restrict                nlist,
 830                      rvec * gmx_restrict                    xx,
 831                      rvec * gmx_restrict                    ff,
 832                      t_forcerec * gmx_restrict              fr,
 833                      t_mdatoms * gmx_restrict               mdatoms,
 834                      nb_kernel_data_t * gmx_restrict        kernel_data,
 835                      t_nrnb * gmx_restrict                  nrnb)
 836 {
 837     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 838      * just 0 for non-waters.
 839      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 840      * jnr indices corresponding to data put in the four positions in the SIMD register.
 841      */
 842     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 843     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 844     int              jnrA,jnrB,jnrC,jnrD;
 845     int              jnrE,jnrF,jnrG,jnrH;
 846     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 847     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 848     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 849     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 850     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 851     real             rcutoff_scalar;
 852     real             *shiftvec,*fshift,*x,*f;
 853     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 854     real             scratch[4*DIM];
 855     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 856     real *           vdwioffsetptr0;
 857     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 858     real *           vdwioffsetptr1;
 859     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 860     real *           vdwioffsetptr2;
 861     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 862     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 863     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 864     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 865     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 866     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 867     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 868     real             *charge;
 869     int              nvdwtype;
 870     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 871     int              *vdwtype;
 872     real             *vdwparam;
 873     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 874     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 875     __m256i          vfitab;
 876     __m128i          vfitab_lo,vfitab_hi;
 877     __m128i          ifour       = _mm_set1_epi32(4);
 878     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 879     real             *vftab;
 880     __m256           dummy_mask,cutoff_mask;
 881     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 882     __m256           one     = _mm256_set1_ps(1.0);
 883     __m256           two     = _mm256_set1_ps(2.0);
 884     x                = xx[0];
 885     f                = ff[0];
 886
 887     nri              = nlist->nri;
 888     iinr             = nlist->iinr;
 889     jindex           = nlist->jindex;
 890     jjnr             = nlist->jjnr;
 891     shiftidx         = nlist->shift;
 892     gid              = nlist->gid;
 893     shiftvec         = fr->shift_vec[0];
 894     fshift           = fr->fshift[0];
 895     facel            = _mm256_set1_ps(fr->epsfac);
 896     charge           = mdatoms->chargeA;
 897     nvdwtype         = fr->ntype;
 898     vdwparam         = fr->nbfp;
 899     vdwtype          = mdatoms->typeA;
 900
 901     vftab            = kernel_data->table_elec_vdw->data;
 902     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 903
 904     /* Setup water-specific parameters */
 905     inr              = nlist->iinr[0];
 906     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 907     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 908     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 909     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 910
 911     /* Avoid stupid compiler warnings */
 912     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 913     j_coord_offsetA = 0;
 914     j_coord_offsetB = 0;
 915     j_coord_offsetC = 0;
 916     j_coord_offsetD = 0;
 917     j_coord_offsetE = 0;
 918     j_coord_offsetF = 0;
 919     j_coord_offsetG = 0;
 920     j_coord_offsetH = 0;
 921
 922     outeriter        = 0;
 923     inneriter        = 0;
 924
 925     for(iidx=0;iidx<4*DIM;iidx++)
 926     {
 927         scratch[iidx] = 0.0;
 928     }
 929
 930     /* Start outer loop over neighborlists */
 931     for(iidx=0; iidx<nri; iidx++)
 932     {
 933         /* Load shift vector for this list */
 934         i_shift_offset   = DIM*shiftidx[iidx];
 935
 936         /* Load limits for loop over neighbors */
 937         j_index_start    = jindex[iidx];
 938         j_index_end      = jindex[iidx+1];
 939
 940         /* Get outer coordinate index */
 941         inr              = iinr[iidx];
 942         i_coord_offset   = DIM*inr;
 943
 944         /* Load i particle coords and add shift vector */
 945         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 946                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 947
 948         fix0             = _mm256_setzero_ps();
 949         fiy0             = _mm256_setzero_ps();
 950         fiz0             = _mm256_setzero_ps();
 951         fix1             = _mm256_setzero_ps();
 952         fiy1             = _mm256_setzero_ps();
 953         fiz1             = _mm256_setzero_ps();
 954         fix2             = _mm256_setzero_ps();
 955         fiy2             = _mm256_setzero_ps();
 956         fiz2             = _mm256_setzero_ps();
 957
 958         /* Start inner kernel loop */
 959         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 960         {
 961
 962             /* Get j neighbor index, and coordinate index */
 963             jnrA             = jjnr[jidx];
 964             jnrB             = jjnr[jidx+1];
 965             jnrC             = jjnr[jidx+2];
 966             jnrD             = jjnr[jidx+3];
 967             jnrE             = jjnr[jidx+4];
 968             jnrF             = jjnr[jidx+5];
 969             jnrG             = jjnr[jidx+6];
 970             jnrH             = jjnr[jidx+7];
 971             j_coord_offsetA  = DIM*jnrA;
 972             j_coord_offsetB  = DIM*jnrB;
 973             j_coord_offsetC  = DIM*jnrC;
 974             j_coord_offsetD  = DIM*jnrD;
 975             j_coord_offsetE  = DIM*jnrE;
 976             j_coord_offsetF  = DIM*jnrF;
 977             j_coord_offsetG  = DIM*jnrG;
 978             j_coord_offsetH  = DIM*jnrH;
 979
 980             /* load j atom coordinates */
 981             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 982                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 983                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 984                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 985                                                  &jx0,&jy0,&jz0);
 986
 987             /* Calculate displacement vector */
 988             dx00             = _mm256_sub_ps(ix0,jx0);
 989             dy00             = _mm256_sub_ps(iy0,jy0);
 990             dz00             = _mm256_sub_ps(iz0,jz0);
 991             dx10             = _mm256_sub_ps(ix1,jx0);
 992             dy10             = _mm256_sub_ps(iy1,jy0);
 993             dz10             = _mm256_sub_ps(iz1,jz0);
 994             dx20             = _mm256_sub_ps(ix2,jx0);
 995             dy20             = _mm256_sub_ps(iy2,jy0);
 996             dz20             = _mm256_sub_ps(iz2,jz0);
 997
 998             /* Calculate squared distance and things based on it */
 999             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1000             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1001             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1002
1003             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1004             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1005             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1006
1007             /* Load parameters for j particles */
1008             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1009                                                                  charge+jnrC+0,charge+jnrD+0,
1010                                                                  charge+jnrE+0,charge+jnrF+0,
1011                                                                  charge+jnrG+0,charge+jnrH+0);
1012             vdwjidx0A        = 2*vdwtype[jnrA+0];
1013             vdwjidx0B        = 2*vdwtype[jnrB+0];
1014             vdwjidx0C        = 2*vdwtype[jnrC+0];
1015             vdwjidx0D        = 2*vdwtype[jnrD+0];
1016             vdwjidx0E        = 2*vdwtype[jnrE+0];
1017             vdwjidx0F        = 2*vdwtype[jnrF+0];
1018             vdwjidx0G        = 2*vdwtype[jnrG+0];
1019             vdwjidx0H        = 2*vdwtype[jnrH+0];
1020
1021             fjx0             = _mm256_setzero_ps();
1022             fjy0             = _mm256_setzero_ps();
1023             fjz0             = _mm256_setzero_ps();
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             r00              = _mm256_mul_ps(rsq00,rinv00);
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq00             = _mm256_mul_ps(iq0,jq0);
1033             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1034                                             vdwioffsetptr0+vdwjidx0B,
1035                                             vdwioffsetptr0+vdwjidx0C,
1036                                             vdwioffsetptr0+vdwjidx0D,
1037                                             vdwioffsetptr0+vdwjidx0E,
1038                                             vdwioffsetptr0+vdwjidx0F,
1039                                             vdwioffsetptr0+vdwjidx0G,
1040                                             vdwioffsetptr0+vdwjidx0H,
1041                                             &c6_00,&c12_00);
1042
1043             /* Calculate table index by multiplying r with table scale and truncate to integer */
1044             rt               = _mm256_mul_ps(r00,vftabscale);
1045             vfitab           = _mm256_cvttps_epi32(rt);
1046             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1047             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1048             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1049             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1050             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1051             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1052
1053             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1054             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1055                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1056             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1057                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1058             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1059                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1060             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1061                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1062             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1063             Heps             = _mm256_mul_ps(vfeps,H);
1064             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1065             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1066             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1067
1068             /* CUBIC SPLINE TABLE DISPERSION */
1069             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1070             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1071             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1072                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1073             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1074                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1075             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1076                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1077             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1078                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1079             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1080             Heps             = _mm256_mul_ps(vfeps,H);
1081             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1082             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1083             fvdw6            = _mm256_mul_ps(c6_00,FF);
1084
1085             /* CUBIC SPLINE TABLE REPULSION */
1086             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1087             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1088             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1089                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1090             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1091                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1092             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1093                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1094             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1095                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1096             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1097             Heps             = _mm256_mul_ps(vfeps,H);
1098             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1099             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1100             fvdw12           = _mm256_mul_ps(c12_00,FF);
1101             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1102
1103             fscal            = _mm256_add_ps(felec,fvdw);
1104
1105             /* Calculate temporary vectorial force */
1106             tx               = _mm256_mul_ps(fscal,dx00);
1107             ty               = _mm256_mul_ps(fscal,dy00);
1108             tz               = _mm256_mul_ps(fscal,dz00);
1109
1110             /* Update vectorial force */
1111             fix0             = _mm256_add_ps(fix0,tx);
1112             fiy0             = _mm256_add_ps(fiy0,ty);
1113             fiz0             = _mm256_add_ps(fiz0,tz);
1114
1115             fjx0             = _mm256_add_ps(fjx0,tx);
1116             fjy0             = _mm256_add_ps(fjy0,ty);
1117             fjz0             = _mm256_add_ps(fjz0,tz);
1118
1119             /**************************
1120              * CALCULATE INTERACTIONS *
1121              **************************/
1122
1123             r10              = _mm256_mul_ps(rsq10,rinv10);
1124
1125             /* Compute parameters for interactions between i and j atoms */
1126             qq10             = _mm256_mul_ps(iq1,jq0);
1127
1128             /* Calculate table index by multiplying r with table scale and truncate to integer */
1129             rt               = _mm256_mul_ps(r10,vftabscale);
1130             vfitab           = _mm256_cvttps_epi32(rt);
1131             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1132             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1133             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1134             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1135             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1136             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1137
1138             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1139             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1140                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1141             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1142                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1143             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1144                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1145             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1146                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1147             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1148             Heps             = _mm256_mul_ps(vfeps,H);
1149             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1150             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1151             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1152
1153             fscal            = felec;
1154
1155             /* Calculate temporary vectorial force */
1156             tx               = _mm256_mul_ps(fscal,dx10);
1157             ty               = _mm256_mul_ps(fscal,dy10);
1158             tz               = _mm256_mul_ps(fscal,dz10);
1159
1160             /* Update vectorial force */
1161             fix1             = _mm256_add_ps(fix1,tx);
1162             fiy1             = _mm256_add_ps(fiy1,ty);
1163             fiz1             = _mm256_add_ps(fiz1,tz);
1164
1165             fjx0             = _mm256_add_ps(fjx0,tx);
1166             fjy0             = _mm256_add_ps(fjy0,ty);
1167             fjz0             = _mm256_add_ps(fjz0,tz);
1168
1169             /**************************
1170              * CALCULATE INTERACTIONS *
1171              **************************/
1172
1173             r20              = _mm256_mul_ps(rsq20,rinv20);
1174
1175             /* Compute parameters for interactions between i and j atoms */
1176             qq20             = _mm256_mul_ps(iq2,jq0);
1177
1178             /* Calculate table index by multiplying r with table scale and truncate to integer */
1179             rt               = _mm256_mul_ps(r20,vftabscale);
1180             vfitab           = _mm256_cvttps_epi32(rt);
1181             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1182             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1183             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1184             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1185             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1186             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1187
1188             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1189             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1190                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1191             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1192                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1193             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1194                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1195             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1196                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1197             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1198             Heps             = _mm256_mul_ps(vfeps,H);
1199             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1200             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1201             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1202
1203             fscal            = felec;
1204
1205             /* Calculate temporary vectorial force */
1206             tx               = _mm256_mul_ps(fscal,dx20);
1207             ty               = _mm256_mul_ps(fscal,dy20);
1208             tz               = _mm256_mul_ps(fscal,dz20);
1209
1210             /* Update vectorial force */
1211             fix2             = _mm256_add_ps(fix2,tx);
1212             fiy2             = _mm256_add_ps(fiy2,ty);
1213             fiz2             = _mm256_add_ps(fiz2,tz);
1214
1215             fjx0             = _mm256_add_ps(fjx0,tx);
1216             fjy0             = _mm256_add_ps(fjy0,ty);
1217             fjz0             = _mm256_add_ps(fjz0,tz);
1218
1219             fjptrA             = f+j_coord_offsetA;
1220             fjptrB             = f+j_coord_offsetB;
1221             fjptrC             = f+j_coord_offsetC;
1222             fjptrD             = f+j_coord_offsetD;
1223             fjptrE             = f+j_coord_offsetE;
1224             fjptrF             = f+j_coord_offsetF;
1225             fjptrG             = f+j_coord_offsetG;
1226             fjptrH             = f+j_coord_offsetH;
1227
1228             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1229
1230             /* Inner loop uses 142 flops */
1231         }
1232
1233         if(jidx<j_index_end)
1234         {
1235
1236             /* Get j neighbor index, and coordinate index */
1237             jnrlistA         = jjnr[jidx];
1238             jnrlistB         = jjnr[jidx+1];
1239             jnrlistC         = jjnr[jidx+2];
1240             jnrlistD         = jjnr[jidx+3];
1241             jnrlistE         = jjnr[jidx+4];
1242             jnrlistF         = jjnr[jidx+5];
1243             jnrlistG         = jjnr[jidx+6];
1244             jnrlistH         = jjnr[jidx+7];
1245             /* Sign of each element will be negative for non-real atoms.
1246              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1247              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1248              */
1249             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1250                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1251
1252             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1253             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1254             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1255             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1256             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1257             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1258             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1259             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1260             j_coord_offsetA  = DIM*jnrA;
1261             j_coord_offsetB  = DIM*jnrB;
1262             j_coord_offsetC  = DIM*jnrC;
1263             j_coord_offsetD  = DIM*jnrD;
1264             j_coord_offsetE  = DIM*jnrE;
1265             j_coord_offsetF  = DIM*jnrF;
1266             j_coord_offsetG  = DIM*jnrG;
1267             j_coord_offsetH  = DIM*jnrH;
1268
1269             /* load j atom coordinates */
1270             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1271                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1272                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1273                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1274                                                  &jx0,&jy0,&jz0);
1275
1276             /* Calculate displacement vector */
1277             dx00             = _mm256_sub_ps(ix0,jx0);
1278             dy00             = _mm256_sub_ps(iy0,jy0);
1279             dz00             = _mm256_sub_ps(iz0,jz0);
1280             dx10             = _mm256_sub_ps(ix1,jx0);
1281             dy10             = _mm256_sub_ps(iy1,jy0);
1282             dz10             = _mm256_sub_ps(iz1,jz0);
1283             dx20             = _mm256_sub_ps(ix2,jx0);
1284             dy20             = _mm256_sub_ps(iy2,jy0);
1285             dz20             = _mm256_sub_ps(iz2,jz0);
1286
1287             /* Calculate squared distance and things based on it */
1288             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1289             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1290             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1291
1292             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1293             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1294             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1295
1296             /* Load parameters for j particles */
1297             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1298                                                                  charge+jnrC+0,charge+jnrD+0,
1299                                                                  charge+jnrE+0,charge+jnrF+0,
1300                                                                  charge+jnrG+0,charge+jnrH+0);
1301             vdwjidx0A        = 2*vdwtype[jnrA+0];
1302             vdwjidx0B        = 2*vdwtype[jnrB+0];
1303             vdwjidx0C        = 2*vdwtype[jnrC+0];
1304             vdwjidx0D        = 2*vdwtype[jnrD+0];
1305             vdwjidx0E        = 2*vdwtype[jnrE+0];
1306             vdwjidx0F        = 2*vdwtype[jnrF+0];
1307             vdwjidx0G        = 2*vdwtype[jnrG+0];
1308             vdwjidx0H        = 2*vdwtype[jnrH+0];
1309
1310             fjx0             = _mm256_setzero_ps();
1311             fjy0             = _mm256_setzero_ps();
1312             fjz0             = _mm256_setzero_ps();
1313
1314             /**************************
1315              * CALCULATE INTERACTIONS *
1316              **************************/
1317
1318             r00              = _mm256_mul_ps(rsq00,rinv00);
1319             r00              = _mm256_andnot_ps(dummy_mask,r00);
1320
1321             /* Compute parameters for interactions between i and j atoms */
1322             qq00             = _mm256_mul_ps(iq0,jq0);
1323             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1324                                             vdwioffsetptr0+vdwjidx0B,
1325                                             vdwioffsetptr0+vdwjidx0C,
1326                                             vdwioffsetptr0+vdwjidx0D,
1327                                             vdwioffsetptr0+vdwjidx0E,
1328                                             vdwioffsetptr0+vdwjidx0F,
1329                                             vdwioffsetptr0+vdwjidx0G,
1330                                             vdwioffsetptr0+vdwjidx0H,
1331                                             &c6_00,&c12_00);
1332
1333             /* Calculate table index by multiplying r with table scale and truncate to integer */
1334             rt               = _mm256_mul_ps(r00,vftabscale);
1335             vfitab           = _mm256_cvttps_epi32(rt);
1336             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1337             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1338             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1339             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1340             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1341             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1342
1343             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1344             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1345                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1346             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1347                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1348             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1349                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1350             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1351                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1352             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1353             Heps             = _mm256_mul_ps(vfeps,H);
1354             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1355             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1356             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq00,FF),_mm256_mul_ps(vftabscale,rinv00)));
1357
1358             /* CUBIC SPLINE TABLE DISPERSION */
1359             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1360             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1361             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1362                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1363             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1364                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1365             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1366                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1367             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1368                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1369             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1370             Heps             = _mm256_mul_ps(vfeps,H);
1371             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1372             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1373             fvdw6            = _mm256_mul_ps(c6_00,FF);
1374
1375             /* CUBIC SPLINE TABLE REPULSION */
1376             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1377             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1378             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1379                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1380             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1381                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1382             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1383                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1384             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1385                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1386             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1387             Heps             = _mm256_mul_ps(vfeps,H);
1388             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1389             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1390             fvdw12           = _mm256_mul_ps(c12_00,FF);
1391             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1392
1393             fscal            = _mm256_add_ps(felec,fvdw);
1394
1395             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1396
1397             /* Calculate temporary vectorial force */
1398             tx               = _mm256_mul_ps(fscal,dx00);
1399             ty               = _mm256_mul_ps(fscal,dy00);
1400             tz               = _mm256_mul_ps(fscal,dz00);
1401
1402             /* Update vectorial force */
1403             fix0             = _mm256_add_ps(fix0,tx);
1404             fiy0             = _mm256_add_ps(fiy0,ty);
1405             fiz0             = _mm256_add_ps(fiz0,tz);
1406
1407             fjx0             = _mm256_add_ps(fjx0,tx);
1408             fjy0             = _mm256_add_ps(fjy0,ty);
1409             fjz0             = _mm256_add_ps(fjz0,tz);
1410
1411             /**************************
1412              * CALCULATE INTERACTIONS *
1413              **************************/
1414
1415             r10              = _mm256_mul_ps(rsq10,rinv10);
1416             r10              = _mm256_andnot_ps(dummy_mask,r10);
1417
1418             /* Compute parameters for interactions between i and j atoms */
1419             qq10             = _mm256_mul_ps(iq1,jq0);
1420
1421             /* Calculate table index by multiplying r with table scale and truncate to integer */
1422             rt               = _mm256_mul_ps(r10,vftabscale);
1423             vfitab           = _mm256_cvttps_epi32(rt);
1424             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1425             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1426             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1427             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1428             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1429             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1430
1431             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1432             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1433                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1434             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1435                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1436             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1437                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1438             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1439                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1440             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1441             Heps             = _mm256_mul_ps(vfeps,H);
1442             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1443             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1444             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq10,FF),_mm256_mul_ps(vftabscale,rinv10)));
1445
1446             fscal            = felec;
1447
1448             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1449
1450             /* Calculate temporary vectorial force */
1451             tx               = _mm256_mul_ps(fscal,dx10);
1452             ty               = _mm256_mul_ps(fscal,dy10);
1453             tz               = _mm256_mul_ps(fscal,dz10);
1454
1455             /* Update vectorial force */
1456             fix1             = _mm256_add_ps(fix1,tx);
1457             fiy1             = _mm256_add_ps(fiy1,ty);
1458             fiz1             = _mm256_add_ps(fiz1,tz);
1459
1460             fjx0             = _mm256_add_ps(fjx0,tx);
1461             fjy0             = _mm256_add_ps(fjy0,ty);
1462             fjz0             = _mm256_add_ps(fjz0,tz);
1463
1464             /**************************
1465              * CALCULATE INTERACTIONS *
1466              **************************/
1467
1468             r20              = _mm256_mul_ps(rsq20,rinv20);
1469             r20              = _mm256_andnot_ps(dummy_mask,r20);
1470
1471             /* Compute parameters for interactions between i and j atoms */
1472             qq20             = _mm256_mul_ps(iq2,jq0);
1473
1474             /* Calculate table index by multiplying r with table scale and truncate to integer */
1475             rt               = _mm256_mul_ps(r20,vftabscale);
1476             vfitab           = _mm256_cvttps_epi32(rt);
1477             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1478             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1479             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1480             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1481             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
1482             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
1483
1484             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1485             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1486                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1487             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1488                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1489             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1490                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1491             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1492                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1493             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1494             Heps             = _mm256_mul_ps(vfeps,H);
1495             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1496             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1497             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq20,FF),_mm256_mul_ps(vftabscale,rinv20)));
1498
1499             fscal            = felec;
1500
1501             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1502
1503             /* Calculate temporary vectorial force */
1504             tx               = _mm256_mul_ps(fscal,dx20);
1505             ty               = _mm256_mul_ps(fscal,dy20);
1506             tz               = _mm256_mul_ps(fscal,dz20);
1507
1508             /* Update vectorial force */
1509             fix2             = _mm256_add_ps(fix2,tx);
1510             fiy2             = _mm256_add_ps(fiy2,ty);
1511             fiz2             = _mm256_add_ps(fiz2,tz);
1512
1513             fjx0             = _mm256_add_ps(fjx0,tx);
1514             fjy0             = _mm256_add_ps(fjy0,ty);
1515             fjz0             = _mm256_add_ps(fjz0,tz);
1516
1517             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1518             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1519             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1520             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1521             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1522             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1523             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1524             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1525
1526             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1527
1528             /* Inner loop uses 145 flops */
1529         }
1530
1531         /* End of innermost loop */
1532
1533         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1534                                                  f+i_coord_offset,fshift+i_shift_offset);
1535
1536         /* Increment number of inner iterations */
1537         inneriter                  += j_index_end - j_index_start;
1538
1539         /* Outer loop uses 18 flops */
1540     }
1541
1542     /* Increment number of outer iterations */
1543     outeriter        += nri;
1544
1545     /* Update outer/inner flops */
1546
1547     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*145);
1548 }