src/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecGB_VdwCSTab_GeomP1P1_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_ElecGB_VdwCSTab_GeomP1P1_VF_avx_256_single
  38  * Electrostatics interaction: GeneralizedBorn
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  75     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m256i          gbitab;
  80     __m128i          gbitab_lo,gbitab_hi;
  81     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  82     __m256           minushalf = _mm256_set1_ps(-0.5);
  83     real             *invsqrta,*dvda,*gbtab;
  84     int              nvdwtype;
  85     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  86     int              *vdwtype;
  87     real             *vdwparam;
  88     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  89     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  90     __m256i          vfitab;
  91     __m128i          vfitab_lo,vfitab_hi;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m256           dummy_mask,cutoff_mask;
  96     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
  97     __m256           one     = _mm256_set1_ps(1.0);
  98     __m256           two     = _mm256_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm256_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 118
 119     invsqrta         = fr->invsqrta;
 120     dvda             = fr->dvda;
 121     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 122     gbtab            = fr->gbtab.data;
 123     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 124
 125     /* Avoid stupid compiler warnings */
 126     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 127     j_coord_offsetA = 0;
 128     j_coord_offsetB = 0;
 129     j_coord_offsetC = 0;
 130     j_coord_offsetD = 0;
 131     j_coord_offsetE = 0;
 132     j_coord_offsetF = 0;
 133     j_coord_offsetG = 0;
 134     j_coord_offsetH = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 160
 161         fix0             = _mm256_setzero_ps();
 162         fiy0             = _mm256_setzero_ps();
 163         fiz0             = _mm256_setzero_ps();
 164
 165         /* Load parameters for i particles */
 166         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 167         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 168         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm256_setzero_ps();
 172         vgbsum           = _mm256_setzero_ps();
 173         vvdwsum          = _mm256_setzero_ps();
 174         dvdasum          = _mm256_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             jnrE             = jjnr[jidx+4];
 186             jnrF             = jjnr[jidx+5];
 187             jnrG             = jjnr[jidx+6];
 188             jnrH             = jjnr[jidx+7];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193             j_coord_offsetE  = DIM*jnrE;
 194             j_coord_offsetF  = DIM*jnrF;
 195             j_coord_offsetG  = DIM*jnrG;
 196             j_coord_offsetH  = DIM*jnrH;
 197
 198             /* load j atom coordinates */
 199             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 201                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 202                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 203                                                  &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm256_sub_ps(ix0,jx0);
 207             dy00             = _mm256_sub_ps(iy0,jy0);
 208             dz00             = _mm256_sub_ps(iz0,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 212
 213             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 214
 215             /* Load parameters for j particles */
 216             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 217                                                                  charge+jnrC+0,charge+jnrD+0,
 218                                                                  charge+jnrE+0,charge+jnrF+0,
 219                                                                  charge+jnrG+0,charge+jnrH+0);
 220             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 221                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 222                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 223                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228             vdwjidx0E        = 2*vdwtype[jnrE+0];
 229             vdwjidx0F        = 2*vdwtype[jnrF+0];
 230             vdwjidx0G        = 2*vdwtype[jnrG+0];
 231             vdwjidx0H        = 2*vdwtype[jnrH+0];
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             r00              = _mm256_mul_ps(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm256_mul_ps(iq0,jq0);
 241             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 242                                             vdwioffsetptr0+vdwjidx0B,
 243                                             vdwioffsetptr0+vdwjidx0C,
 244                                             vdwioffsetptr0+vdwjidx0D,
 245                                             vdwioffsetptr0+vdwjidx0E,
 246                                             vdwioffsetptr0+vdwjidx0F,
 247                                             vdwioffsetptr0+vdwjidx0G,
 248                                             vdwioffsetptr0+vdwjidx0H,
 249                                             &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm256_mul_ps(r00,vftabscale);
 253             vfitab           = _mm256_cvttps_epi32(rt);
 254             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 255             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 256             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 257             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 258             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 259             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 260
 261             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 262             isaprod          = _mm256_mul_ps(isai0,isaj0);
 263             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 264             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 265
 266             /* Calculate generalized born table index - this is a separate table from the normal one,
 267              * but we use the same procedure by multiplying r with scale and truncating to integer.
 268              */
 269             rt               = _mm256_mul_ps(r00,gbscale);
 270             gbitab           = _mm256_cvttps_epi32(rt);
 271             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 272             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 273             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 274             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 275             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 276             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 277             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 278                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 279             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 280                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 281             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 282                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 283             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 284                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 285             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 286             Heps             = _mm256_mul_ps(gbeps,H);
 287             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 288             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 289             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 290
 291             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 292             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 293             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 294             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 295             fjptrA           = dvda+jnrA;
 296             fjptrB           = dvda+jnrB;
 297             fjptrC           = dvda+jnrC;
 298             fjptrD           = dvda+jnrD;
 299             fjptrE           = dvda+jnrE;
 300             fjptrF           = dvda+jnrF;
 301             fjptrG           = dvda+jnrG;
 302             fjptrH           = dvda+jnrH;
 303             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 304                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 305             velec            = _mm256_mul_ps(qq00,rinv00);
 306             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 307
 308             /* CUBIC SPLINE TABLE DISPERSION */
 309             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 310                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 311             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 312                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 313             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 314                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 315             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 316                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 317             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 318             Heps             = _mm256_mul_ps(vfeps,H);
 319             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 320             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 321             vvdw6            = _mm256_mul_ps(c6_00,VV);
 322             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 323             fvdw6            = _mm256_mul_ps(c6_00,FF);
 324
 325             /* CUBIC SPLINE TABLE REPULSION */
 326             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 327             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 328             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 329                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 330             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 331                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 332             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 333                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 334             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 335                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 336             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 337             Heps             = _mm256_mul_ps(vfeps,H);
 338             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 339             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 340             vvdw12           = _mm256_mul_ps(c12_00,VV);
 341             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 342             fvdw12           = _mm256_mul_ps(c12_00,FF);
 343             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 344             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm256_add_ps(velecsum,velec);
 348             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 349             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 350
 351             fscal            = _mm256_add_ps(felec,fvdw);
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm256_mul_ps(fscal,dx00);
 355             ty               = _mm256_mul_ps(fscal,dy00);
 356             tz               = _mm256_mul_ps(fscal,dz00);
 357
 358             /* Update vectorial force */
 359             fix0             = _mm256_add_ps(fix0,tx);
 360             fiy0             = _mm256_add_ps(fiy0,ty);
 361             fiz0             = _mm256_add_ps(fiz0,tz);
 362
 363             fjptrA             = f+j_coord_offsetA;
 364             fjptrB             = f+j_coord_offsetB;
 365             fjptrC             = f+j_coord_offsetC;
 366             fjptrD             = f+j_coord_offsetD;
 367             fjptrE             = f+j_coord_offsetE;
 368             fjptrF             = f+j_coord_offsetF;
 369             fjptrG             = f+j_coord_offsetG;
 370             fjptrH             = f+j_coord_offsetH;
 371             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 372
 373             /* Inner loop uses 91 flops */
 374         }
 375
 376         if(jidx<j_index_end)
 377         {
 378
 379             /* Get j neighbor index, and coordinate index */
 380             jnrlistA         = jjnr[jidx];
 381             jnrlistB         = jjnr[jidx+1];
 382             jnrlistC         = jjnr[jidx+2];
 383             jnrlistD         = jjnr[jidx+3];
 384             jnrlistE         = jjnr[jidx+4];
 385             jnrlistF         = jjnr[jidx+5];
 386             jnrlistG         = jjnr[jidx+6];
 387             jnrlistH         = jjnr[jidx+7];
 388             /* Sign of each element will be negative for non-real atoms.
 389              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 390              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 391              */
 392             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 393                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 394
 395             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 396             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 397             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 398             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 399             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 400             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 401             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 402             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 403             j_coord_offsetA  = DIM*jnrA;
 404             j_coord_offsetB  = DIM*jnrB;
 405             j_coord_offsetC  = DIM*jnrC;
 406             j_coord_offsetD  = DIM*jnrD;
 407             j_coord_offsetE  = DIM*jnrE;
 408             j_coord_offsetF  = DIM*jnrF;
 409             j_coord_offsetG  = DIM*jnrG;
 410             j_coord_offsetH  = DIM*jnrH;
 411
 412             /* load j atom coordinates */
 413             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 414                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 415                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 416                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 417                                                  &jx0,&jy0,&jz0);
 418
 419             /* Calculate displacement vector */
 420             dx00             = _mm256_sub_ps(ix0,jx0);
 421             dy00             = _mm256_sub_ps(iy0,jy0);
 422             dz00             = _mm256_sub_ps(iz0,jz0);
 423
 424             /* Calculate squared distance and things based on it */
 425             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 426
 427             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 428
 429             /* Load parameters for j particles */
 430             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 431                                                                  charge+jnrC+0,charge+jnrD+0,
 432                                                                  charge+jnrE+0,charge+jnrF+0,
 433                                                                  charge+jnrG+0,charge+jnrH+0);
 434             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 435                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 436                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 437                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 438             vdwjidx0A        = 2*vdwtype[jnrA+0];
 439             vdwjidx0B        = 2*vdwtype[jnrB+0];
 440             vdwjidx0C        = 2*vdwtype[jnrC+0];
 441             vdwjidx0D        = 2*vdwtype[jnrD+0];
 442             vdwjidx0E        = 2*vdwtype[jnrE+0];
 443             vdwjidx0F        = 2*vdwtype[jnrF+0];
 444             vdwjidx0G        = 2*vdwtype[jnrG+0];
 445             vdwjidx0H        = 2*vdwtype[jnrH+0];
 446
 447             /**************************
 448              * CALCULATE INTERACTIONS *
 449              **************************/
 450
 451             r00              = _mm256_mul_ps(rsq00,rinv00);
 452             r00              = _mm256_andnot_ps(dummy_mask,r00);
 453
 454             /* Compute parameters for interactions between i and j atoms */
 455             qq00             = _mm256_mul_ps(iq0,jq0);
 456             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 457                                             vdwioffsetptr0+vdwjidx0B,
 458                                             vdwioffsetptr0+vdwjidx0C,
 459                                             vdwioffsetptr0+vdwjidx0D,
 460                                             vdwioffsetptr0+vdwjidx0E,
 461                                             vdwioffsetptr0+vdwjidx0F,
 462                                             vdwioffsetptr0+vdwjidx0G,
 463                                             vdwioffsetptr0+vdwjidx0H,
 464                                             &c6_00,&c12_00);
 465
 466             /* Calculate table index by multiplying r with table scale and truncate to integer */
 467             rt               = _mm256_mul_ps(r00,vftabscale);
 468             vfitab           = _mm256_cvttps_epi32(rt);
 469             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 470             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 471             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 472             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 473             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 474             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 475
 476             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 477             isaprod          = _mm256_mul_ps(isai0,isaj0);
 478             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 479             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 480
 481             /* Calculate generalized born table index - this is a separate table from the normal one,
 482              * but we use the same procedure by multiplying r with scale and truncating to integer.
 483              */
 484             rt               = _mm256_mul_ps(r00,gbscale);
 485             gbitab           = _mm256_cvttps_epi32(rt);
 486             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 487             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 488             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 489             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 490             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 491             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 492             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 493                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 494             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 495                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 496             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 497                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 498             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 499                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 500             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 501             Heps             = _mm256_mul_ps(gbeps,H);
 502             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 503             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 504             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 505
 506             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 507             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 508             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 509             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 510             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 511             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 512             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 513             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 514             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 515             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 516             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 517             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 518             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 519             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 520                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 521             velec            = _mm256_mul_ps(qq00,rinv00);
 522             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 523
 524             /* CUBIC SPLINE TABLE DISPERSION */
 525             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 526                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 527             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 528                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 529             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 530                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 531             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 532                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 533             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 534             Heps             = _mm256_mul_ps(vfeps,H);
 535             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 536             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 537             vvdw6            = _mm256_mul_ps(c6_00,VV);
 538             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 539             fvdw6            = _mm256_mul_ps(c6_00,FF);
 540
 541             /* CUBIC SPLINE TABLE REPULSION */
 542             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 543             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 544             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 545                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 546             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 547                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 548             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 549                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 550             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 551                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 552             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 553             Heps             = _mm256_mul_ps(vfeps,H);
 554             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 555             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 556             vvdw12           = _mm256_mul_ps(c12_00,VV);
 557             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 558             fvdw12           = _mm256_mul_ps(c12_00,FF);
 559             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 560             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             velec            = _mm256_andnot_ps(dummy_mask,velec);
 564             velecsum         = _mm256_add_ps(velecsum,velec);
 565             vgb              = _mm256_andnot_ps(dummy_mask,vgb);
 566             vgbsum           = _mm256_add_ps(vgbsum,vgb);
 567             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 568             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 569
 570             fscal            = _mm256_add_ps(felec,fvdw);
 571
 572             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 573
 574             /* Calculate temporary vectorial force */
 575             tx               = _mm256_mul_ps(fscal,dx00);
 576             ty               = _mm256_mul_ps(fscal,dy00);
 577             tz               = _mm256_mul_ps(fscal,dz00);
 578
 579             /* Update vectorial force */
 580             fix0             = _mm256_add_ps(fix0,tx);
 581             fiy0             = _mm256_add_ps(fiy0,ty);
 582             fiz0             = _mm256_add_ps(fiz0,tz);
 583
 584             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 585             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 586             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 587             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 588             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 589             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 590             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 591             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 592             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 593
 594             /* Inner loop uses 92 flops */
 595         }
 596
 597         /* End of innermost loop */
 598
 599         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 600                                                  f+i_coord_offset,fshift+i_shift_offset);
 601
 602         ggid                        = gid[iidx];
 603         /* Update potential energies */
 604         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 605         gmx_mm256_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
 606         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 607         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
 608         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
 609
 610         /* Increment number of inner iterations */
 611         inneriter                  += j_index_end - j_index_start;
 612
 613         /* Outer loop uses 10 flops */
 614     }
 615
 616     /* Increment number of outer iterations */
 617     outeriter        += nri;
 618
 619     /* Update outer/inner flops */
 620
 621     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*92);
 622 }
 623 /*
 624  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_avx_256_single
 625  * Electrostatics interaction: GeneralizedBorn
 626  * VdW interaction:            CubicSplineTable
 627  * Geometry:                   Particle-Particle
 628  * Calculate force/pot:        Force
 629  */
 630 void
 631 nb_kernel_ElecGB_VdwCSTab_GeomP1P1_F_avx_256_single
 632                     (t_nblist * gmx_restrict                nlist,
 633                      rvec * gmx_restrict                    xx,
 634                      rvec * gmx_restrict                    ff,
 635                      t_forcerec * gmx_restrict              fr,
 636                      t_mdatoms * gmx_restrict               mdatoms,
 637                      nb_kernel_data_t * gmx_restrict        kernel_data,
 638                      t_nrnb * gmx_restrict                  nrnb)
 639 {
 640     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 641      * just 0 for non-waters.
 642      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 643      * jnr indices corresponding to data put in the four positions in the SIMD register.
 644      */
 645     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 646     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 647     int              jnrA,jnrB,jnrC,jnrD;
 648     int              jnrE,jnrF,jnrG,jnrH;
 649     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 650     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 651     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 652     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 653     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 654     real             rcutoff_scalar;
 655     real             *shiftvec,*fshift,*x,*f;
 656     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 657     real             scratch[4*DIM];
 658     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 659     real *           vdwioffsetptr0;
 660     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 661     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 662     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 663     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 664     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 665     real             *charge;
 666     __m256i          gbitab;
 667     __m128i          gbitab_lo,gbitab_hi;
 668     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 669     __m256           minushalf = _mm256_set1_ps(-0.5);
 670     real             *invsqrta,*dvda,*gbtab;
 671     int              nvdwtype;
 672     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 673     int              *vdwtype;
 674     real             *vdwparam;
 675     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 676     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 677     __m256i          vfitab;
 678     __m128i          vfitab_lo,vfitab_hi;
 679     __m128i          ifour       = _mm_set1_epi32(4);
 680     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 681     real             *vftab;
 682     __m256           dummy_mask,cutoff_mask;
 683     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 684     __m256           one     = _mm256_set1_ps(1.0);
 685     __m256           two     = _mm256_set1_ps(2.0);
 686     x                = xx[0];
 687     f                = ff[0];
 688
 689     nri              = nlist->nri;
 690     iinr             = nlist->iinr;
 691     jindex           = nlist->jindex;
 692     jjnr             = nlist->jjnr;
 693     shiftidx         = nlist->shift;
 694     gid              = nlist->gid;
 695     shiftvec         = fr->shift_vec[0];
 696     fshift           = fr->fshift[0];
 697     facel            = _mm256_set1_ps(fr->epsfac);
 698     charge           = mdatoms->chargeA;
 699     nvdwtype         = fr->ntype;
 700     vdwparam         = fr->nbfp;
 701     vdwtype          = mdatoms->typeA;
 702
 703     vftab            = kernel_data->table_vdw->data;
 704     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 705
 706     invsqrta         = fr->invsqrta;
 707     dvda             = fr->dvda;
 708     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 709     gbtab            = fr->gbtab.data;
 710     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 711
 712     /* Avoid stupid compiler warnings */
 713     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 714     j_coord_offsetA = 0;
 715     j_coord_offsetB = 0;
 716     j_coord_offsetC = 0;
 717     j_coord_offsetD = 0;
 718     j_coord_offsetE = 0;
 719     j_coord_offsetF = 0;
 720     j_coord_offsetG = 0;
 721     j_coord_offsetH = 0;
 722
 723     outeriter        = 0;
 724     inneriter        = 0;
 725
 726     for(iidx=0;iidx<4*DIM;iidx++)
 727     {
 728         scratch[iidx] = 0.0;
 729     }
 730
 731     /* Start outer loop over neighborlists */
 732     for(iidx=0; iidx<nri; iidx++)
 733     {
 734         /* Load shift vector for this list */
 735         i_shift_offset   = DIM*shiftidx[iidx];
 736
 737         /* Load limits for loop over neighbors */
 738         j_index_start    = jindex[iidx];
 739         j_index_end      = jindex[iidx+1];
 740
 741         /* Get outer coordinate index */
 742         inr              = iinr[iidx];
 743         i_coord_offset   = DIM*inr;
 744
 745         /* Load i particle coords and add shift vector */
 746         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 747
 748         fix0             = _mm256_setzero_ps();
 749         fiy0             = _mm256_setzero_ps();
 750         fiz0             = _mm256_setzero_ps();
 751
 752         /* Load parameters for i particles */
 753         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 754         isai0            = _mm256_set1_ps(invsqrta[inr+0]);
 755         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 756
 757         dvdasum          = _mm256_setzero_ps();
 758
 759         /* Start inner kernel loop */
 760         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 761         {
 762
 763             /* Get j neighbor index, and coordinate index */
 764             jnrA             = jjnr[jidx];
 765             jnrB             = jjnr[jidx+1];
 766             jnrC             = jjnr[jidx+2];
 767             jnrD             = jjnr[jidx+3];
 768             jnrE             = jjnr[jidx+4];
 769             jnrF             = jjnr[jidx+5];
 770             jnrG             = jjnr[jidx+6];
 771             jnrH             = jjnr[jidx+7];
 772             j_coord_offsetA  = DIM*jnrA;
 773             j_coord_offsetB  = DIM*jnrB;
 774             j_coord_offsetC  = DIM*jnrC;
 775             j_coord_offsetD  = DIM*jnrD;
 776             j_coord_offsetE  = DIM*jnrE;
 777             j_coord_offsetF  = DIM*jnrF;
 778             j_coord_offsetG  = DIM*jnrG;
 779             j_coord_offsetH  = DIM*jnrH;
 780
 781             /* load j atom coordinates */
 782             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 783                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 784                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 785                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 786                                                  &jx0,&jy0,&jz0);
 787
 788             /* Calculate displacement vector */
 789             dx00             = _mm256_sub_ps(ix0,jx0);
 790             dy00             = _mm256_sub_ps(iy0,jy0);
 791             dz00             = _mm256_sub_ps(iz0,jz0);
 792
 793             /* Calculate squared distance and things based on it */
 794             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 795
 796             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 797
 798             /* Load parameters for j particles */
 799             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 800                                                                  charge+jnrC+0,charge+jnrD+0,
 801                                                                  charge+jnrE+0,charge+jnrF+0,
 802                                                                  charge+jnrG+0,charge+jnrH+0);
 803             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
 804                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
 805                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
 806                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
 807             vdwjidx0A        = 2*vdwtype[jnrA+0];
 808             vdwjidx0B        = 2*vdwtype[jnrB+0];
 809             vdwjidx0C        = 2*vdwtype[jnrC+0];
 810             vdwjidx0D        = 2*vdwtype[jnrD+0];
 811             vdwjidx0E        = 2*vdwtype[jnrE+0];
 812             vdwjidx0F        = 2*vdwtype[jnrF+0];
 813             vdwjidx0G        = 2*vdwtype[jnrG+0];
 814             vdwjidx0H        = 2*vdwtype[jnrH+0];
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             r00              = _mm256_mul_ps(rsq00,rinv00);
 821
 822             /* Compute parameters for interactions between i and j atoms */
 823             qq00             = _mm256_mul_ps(iq0,jq0);
 824             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 825                                             vdwioffsetptr0+vdwjidx0B,
 826                                             vdwioffsetptr0+vdwjidx0C,
 827                                             vdwioffsetptr0+vdwjidx0D,
 828                                             vdwioffsetptr0+vdwjidx0E,
 829                                             vdwioffsetptr0+vdwjidx0F,
 830                                             vdwioffsetptr0+vdwjidx0G,
 831                                             vdwioffsetptr0+vdwjidx0H,
 832                                             &c6_00,&c12_00);
 833
 834             /* Calculate table index by multiplying r with table scale and truncate to integer */
 835             rt               = _mm256_mul_ps(r00,vftabscale);
 836             vfitab           = _mm256_cvttps_epi32(rt);
 837             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 838             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 839             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 840             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 841             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 842             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 843
 844             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 845             isaprod          = _mm256_mul_ps(isai0,isaj0);
 846             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
 847             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 848
 849             /* Calculate generalized born table index - this is a separate table from the normal one,
 850              * but we use the same procedure by multiplying r with scale and truncating to integer.
 851              */
 852             rt               = _mm256_mul_ps(r00,gbscale);
 853             gbitab           = _mm256_cvttps_epi32(rt);
 854             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 855             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 856             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 857             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 858             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 859             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 860             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 861                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 862             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 863                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 864             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 865                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 866             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 867                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 868             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 869             Heps             = _mm256_mul_ps(gbeps,H);
 870             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 871             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 872             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 873
 874             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 875             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 876             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
 877             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 878             fjptrA           = dvda+jnrA;
 879             fjptrB           = dvda+jnrB;
 880             fjptrC           = dvda+jnrC;
 881             fjptrD           = dvda+jnrD;
 882             fjptrE           = dvda+jnrE;
 883             fjptrF           = dvda+jnrF;
 884             fjptrG           = dvda+jnrG;
 885             fjptrH           = dvda+jnrH;
 886             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 887                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
 888             velec            = _mm256_mul_ps(qq00,rinv00);
 889             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
 890
 891             /* CUBIC SPLINE TABLE DISPERSION */
 892             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 893                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 894             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 895                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 896             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 897                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 898             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 899                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 900             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 901             Heps             = _mm256_mul_ps(vfeps,H);
 902             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 903             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 904             fvdw6            = _mm256_mul_ps(c6_00,FF);
 905
 906             /* CUBIC SPLINE TABLE REPULSION */
 907             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 908             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 909             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 910                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 911             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 912                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 913             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 914                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 915             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 916                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 917             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 918             Heps             = _mm256_mul_ps(vfeps,H);
 919             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 920             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 921             fvdw12           = _mm256_mul_ps(c12_00,FF);
 922             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
 923
 924             fscal            = _mm256_add_ps(felec,fvdw);
 925
 926             /* Calculate temporary vectorial force */
 927             tx               = _mm256_mul_ps(fscal,dx00);
 928             ty               = _mm256_mul_ps(fscal,dy00);
 929             tz               = _mm256_mul_ps(fscal,dz00);
 930
 931             /* Update vectorial force */
 932             fix0             = _mm256_add_ps(fix0,tx);
 933             fiy0             = _mm256_add_ps(fiy0,ty);
 934             fiz0             = _mm256_add_ps(fiz0,tz);
 935
 936             fjptrA             = f+j_coord_offsetA;
 937             fjptrB             = f+j_coord_offsetB;
 938             fjptrC             = f+j_coord_offsetC;
 939             fjptrD             = f+j_coord_offsetD;
 940             fjptrE             = f+j_coord_offsetE;
 941             fjptrF             = f+j_coord_offsetF;
 942             fjptrG             = f+j_coord_offsetG;
 943             fjptrH             = f+j_coord_offsetH;
 944             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 945
 946             /* Inner loop uses 81 flops */
 947         }
 948
 949         if(jidx<j_index_end)
 950         {
 951
 952             /* Get j neighbor index, and coordinate index */
 953             jnrlistA         = jjnr[jidx];
 954             jnrlistB         = jjnr[jidx+1];
 955             jnrlistC         = jjnr[jidx+2];
 956             jnrlistD         = jjnr[jidx+3];
 957             jnrlistE         = jjnr[jidx+4];
 958             jnrlistF         = jjnr[jidx+5];
 959             jnrlistG         = jjnr[jidx+6];
 960             jnrlistH         = jjnr[jidx+7];
 961             /* Sign of each element will be negative for non-real atoms.
 962              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 963              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 964              */
 965             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 966                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 967
 968             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 969             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 970             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 971             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 972             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 973             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 974             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 975             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 976             j_coord_offsetA  = DIM*jnrA;
 977             j_coord_offsetB  = DIM*jnrB;
 978             j_coord_offsetC  = DIM*jnrC;
 979             j_coord_offsetD  = DIM*jnrD;
 980             j_coord_offsetE  = DIM*jnrE;
 981             j_coord_offsetF  = DIM*jnrF;
 982             j_coord_offsetG  = DIM*jnrG;
 983             j_coord_offsetH  = DIM*jnrH;
 984
 985             /* load j atom coordinates */
 986             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 987                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 988                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 989                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 990                                                  &jx0,&jy0,&jz0);
 991
 992             /* Calculate displacement vector */
 993             dx00             = _mm256_sub_ps(ix0,jx0);
 994             dy00             = _mm256_sub_ps(iy0,jy0);
 995             dz00             = _mm256_sub_ps(iz0,jz0);
 996
 997             /* Calculate squared distance and things based on it */
 998             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 999
1000             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1001
1002             /* Load parameters for j particles */
1003             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1004                                                                  charge+jnrC+0,charge+jnrD+0,
1005                                                                  charge+jnrE+0,charge+jnrF+0,
1006                                                                  charge+jnrG+0,charge+jnrH+0);
1007             isaj0            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+0,invsqrta+jnrB+0,
1008                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0,
1009                                                                  invsqrta+jnrE+0,invsqrta+jnrF+0,
1010                                                                  invsqrta+jnrG+0,invsqrta+jnrH+0);
1011             vdwjidx0A        = 2*vdwtype[jnrA+0];
1012             vdwjidx0B        = 2*vdwtype[jnrB+0];
1013             vdwjidx0C        = 2*vdwtype[jnrC+0];
1014             vdwjidx0D        = 2*vdwtype[jnrD+0];
1015             vdwjidx0E        = 2*vdwtype[jnrE+0];
1016             vdwjidx0F        = 2*vdwtype[jnrF+0];
1017             vdwjidx0G        = 2*vdwtype[jnrG+0];
1018             vdwjidx0H        = 2*vdwtype[jnrH+0];
1019
1020             /**************************
1021              * CALCULATE INTERACTIONS *
1022              **************************/
1023
1024             r00              = _mm256_mul_ps(rsq00,rinv00);
1025             r00              = _mm256_andnot_ps(dummy_mask,r00);
1026
1027             /* Compute parameters for interactions between i and j atoms */
1028             qq00             = _mm256_mul_ps(iq0,jq0);
1029             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1030                                             vdwioffsetptr0+vdwjidx0B,
1031                                             vdwioffsetptr0+vdwjidx0C,
1032                                             vdwioffsetptr0+vdwjidx0D,
1033                                             vdwioffsetptr0+vdwjidx0E,
1034                                             vdwioffsetptr0+vdwjidx0F,
1035                                             vdwioffsetptr0+vdwjidx0G,
1036                                             vdwioffsetptr0+vdwjidx0H,
1037                                             &c6_00,&c12_00);
1038
1039             /* Calculate table index by multiplying r with table scale and truncate to integer */
1040             rt               = _mm256_mul_ps(r00,vftabscale);
1041             vfitab           = _mm256_cvttps_epi32(rt);
1042             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1043             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1044             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
1045             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
1046             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
1047             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
1048
1049             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
1050             isaprod          = _mm256_mul_ps(isai0,isaj0);
1051             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq00,_mm256_mul_ps(isaprod,gbinvepsdiff)));
1052             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
1053
1054             /* Calculate generalized born table index - this is a separate table from the normal one,
1055              * but we use the same procedure by multiplying r with scale and truncating to integer.
1056              */
1057             rt               = _mm256_mul_ps(r00,gbscale);
1058             gbitab           = _mm256_cvttps_epi32(rt);
1059             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
1060             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
1061             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
1062             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
1063             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
1064             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
1065             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
1066                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
1067             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
1068                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
1069             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
1070                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
1071             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
1072                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
1073             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1074             Heps             = _mm256_mul_ps(gbeps,H);
1075             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
1076             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
1077             vgb              = _mm256_mul_ps(gbqqfactor,VV);
1078
1079             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1080             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
1081             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r00)));
1082             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
1083             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
1084             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
1085             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
1086             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
1087             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
1088             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
1089             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
1090             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
1091             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
1092             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1093                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj0,isaj0)));
1094             velec            = _mm256_mul_ps(qq00,rinv00);
1095             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv00),fgb),rinv00);
1096
1097             /* CUBIC SPLINE TABLE DISPERSION */
1098             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1099                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1100             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1101                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1102             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1103                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1104             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1105                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1106             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1107             Heps             = _mm256_mul_ps(vfeps,H);
1108             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1109             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1110             fvdw6            = _mm256_mul_ps(c6_00,FF);
1111
1112             /* CUBIC SPLINE TABLE REPULSION */
1113             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
1114             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
1115             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
1116                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
1117             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
1118                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
1119             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
1120                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
1121             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
1122                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
1123             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
1124             Heps             = _mm256_mul_ps(vfeps,H);
1125             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
1126             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
1127             fvdw12           = _mm256_mul_ps(c12_00,FF);
1128             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv00)));
1129
1130             fscal            = _mm256_add_ps(felec,fvdw);
1131
1132             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1133
1134             /* Calculate temporary vectorial force */
1135             tx               = _mm256_mul_ps(fscal,dx00);
1136             ty               = _mm256_mul_ps(fscal,dy00);
1137             tz               = _mm256_mul_ps(fscal,dz00);
1138
1139             /* Update vectorial force */
1140             fix0             = _mm256_add_ps(fix0,tx);
1141             fiy0             = _mm256_add_ps(fiy0,ty);
1142             fiz0             = _mm256_add_ps(fiz0,tz);
1143
1144             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1145             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1146             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1147             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1148             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1149             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1150             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1151             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1152             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1153
1154             /* Inner loop uses 82 flops */
1155         }
1156
1157         /* End of innermost loop */
1158
1159         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1160                                                  f+i_coord_offset,fshift+i_shift_offset);
1161
1162         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai0,isai0));
1163         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
1164
1165         /* Increment number of inner iterations */
1166         inneriter                  += j_index_end - j_index_start;
1167
1168         /* Outer loop uses 7 flops */
1169     }
1170
1171     /* Increment number of outer iterations */
1172     outeriter        += nri;
1173
1174     /* Update outer/inner flops */
1175
1176     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*82);
1177 }