src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecGB_VdwLJ_GeomP1P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012, by the GROMACS development team, led by
   5  * David van der Spoel, Berk Hess, Erik Lindahl, and including many
   6  * others, as listed in the AUTHORS file in the top-level source
   7  * directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: GeneralizedBorn
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecGB_VdwLJ_GeomP1P1_VF_avx_256_double
  61                     (t_nblist * gmx_restrict                nlist,
  62                      rvec * gmx_restrict                    xx,
  63                      rvec * gmx_restrict                    ff,
  64                      t_forcerec * gmx_restrict              fr,
  65                      t_mdatoms * gmx_restrict               mdatoms,
  66                      nb_kernel_data_t * gmx_restrict        kernel_data,
  67                      t_nrnb * gmx_restrict                  nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     __m128i          gbitab;
  94     __m256d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
  95     __m256d          minushalf = _mm256_set1_pd(-0.5);
  96     real             *invsqrta,*dvda,*gbtab;
  97     int              nvdwtype;
  98     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 102     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 103     __m128i          vfitab;
 104     __m128i          ifour       = _mm_set1_epi32(4);
 105     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 106     real             *vftab;
 107     __m256d          dummy_mask,cutoff_mask;
 108     __m128           tmpmask0,tmpmask1;
 109     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 110     __m256d          one     = _mm256_set1_pd(1.0);
 111     __m256d          two     = _mm256_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     invsqrta         = fr->invsqrta;
 130     dvda             = fr->dvda;
 131     gbtabscale       = _mm256_set1_pd(fr->gbtab.scale);
 132     gbtab            = fr->gbtab.data;
 133     gbinvepsdiff     = _mm256_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 166
 167         fix0             = _mm256_setzero_pd();
 168         fiy0             = _mm256_setzero_pd();
 169         fiz0             = _mm256_setzero_pd();
 170
 171         /* Load parameters for i particles */
 172         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 173         isai0            = _mm256_set1_pd(invsqrta[inr+0]);
 174         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm256_setzero_pd();
 178         vgbsum           = _mm256_setzero_pd();
 179         vvdwsum          = _mm256_setzero_pd();
 180         dvdasum          = _mm256_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             jnrC             = jjnr[jidx+2];
 190             jnrD             = jjnr[jidx+3];
 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
 196             /* load j atom coordinates */
 197             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 199                                                  &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm256_sub_pd(ix0,jx0);
 203             dy00             = _mm256_sub_pd(iy0,jy0);
 204             dz00             = _mm256_sub_pd(iz0,jz0);
 205
 206             /* Calculate squared distance and things based on it */
 207             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 208
 209             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 210
 211             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 212
 213             /* Load parameters for j particles */
 214             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 215                                                                  charge+jnrC+0,charge+jnrD+0);
 216             isaj0            = gmx_mm256_load_4real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0,
 217                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0);
 218             vdwjidx0A        = 2*vdwtype[jnrA+0];
 219             vdwjidx0B        = 2*vdwtype[jnrB+0];
 220             vdwjidx0C        = 2*vdwtype[jnrC+0];
 221             vdwjidx0D        = 2*vdwtype[jnrD+0];
 222
 223             /**************************
 224              * CALCULATE INTERACTIONS *
 225              **************************/
 226
 227             r00              = _mm256_mul_pd(rsq00,rinv00);
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq00             = _mm256_mul_pd(iq0,jq0);
 231             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 232                                             vdwioffsetptr0+vdwjidx0B,
 233                                             vdwioffsetptr0+vdwjidx0C,
 234                                             vdwioffsetptr0+vdwjidx0D,
 235                                             &c6_00,&c12_00);
 236
 237             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 238             isaprod          = _mm256_mul_pd(isai0,isaj0);
 239             gbqqfactor       = _mm256_xor_pd(signbit,_mm256_mul_pd(qq00,_mm256_mul_pd(isaprod,gbinvepsdiff)));
 240             gbscale          = _mm256_mul_pd(isaprod,gbtabscale);
 241
 242             /* Calculate generalized born table index - this is a separate table from the normal one,
 243              * but we use the same procedure by multiplying r with scale and truncating to integer.
 244              */
 245             rt               = _mm256_mul_pd(r00,gbscale);
 246             gbitab           = _mm256_cvttpd_epi32(rt);
 247             gbeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 248             gbitab           = _mm_slli_epi32(gbitab,2);
 249             Y                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
 250             F                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,1) );
 251             G                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,2) );
 252             H                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,3) );
 253             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 254             Heps             = _mm256_mul_pd(gbeps,H);
 255             Fp               = _mm256_add_pd(F,_mm256_mul_pd(gbeps,_mm256_add_pd(G,Heps)));
 256             VV               = _mm256_add_pd(Y,_mm256_mul_pd(gbeps,Fp));
 257             vgb              = _mm256_mul_pd(gbqqfactor,VV);
 258
 259             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(gbeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 260             fgb              = _mm256_mul_pd(gbqqfactor,_mm256_mul_pd(FF,gbscale));
 261             dvdatmp          = _mm256_mul_pd(minushalf,_mm256_add_pd(vgb,_mm256_mul_pd(fgb,r00)));
 262             dvdasum          = _mm256_add_pd(dvdasum,dvdatmp);
 263             fjptrA           = dvda+jnrA;
 264             fjptrB           = dvda+jnrB;
 265             fjptrC           = dvda+jnrC;
 266             fjptrD           = dvda+jnrD;
 267             gmx_mm256_increment_4real_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 268                                                  _mm256_mul_pd(dvdatmp,_mm256_mul_pd(isaj0,isaj0)));
 269             velec            = _mm256_mul_pd(qq00,rinv00);
 270             felec            = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(velec,rinv00),fgb),rinv00);
 271
 272             /* LENNARD-JONES DISPERSION/REPULSION */
 273
 274             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 275             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 276             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 277             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 278             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm256_add_pd(velecsum,velec);
 282             vgbsum           = _mm256_add_pd(vgbsum,vgb);
 283             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 284
 285             fscal            = _mm256_add_pd(felec,fvdw);
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm256_mul_pd(fscal,dx00);
 289             ty               = _mm256_mul_pd(fscal,dy00);
 290             tz               = _mm256_mul_pd(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm256_add_pd(fix0,tx);
 294             fiy0             = _mm256_add_pd(fiy0,ty);
 295             fiz0             = _mm256_add_pd(fiz0,tz);
 296
 297             fjptrA             = f+j_coord_offsetA;
 298             fjptrB             = f+j_coord_offsetB;
 299             fjptrC             = f+j_coord_offsetC;
 300             fjptrD             = f+j_coord_offsetD;
 301             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 302
 303             /* Inner loop uses 70 flops */
 304         }
 305
 306         if(jidx<j_index_end)
 307         {
 308
 309             /* Get j neighbor index, and coordinate index */
 310             jnrlistA         = jjnr[jidx];
 311             jnrlistB         = jjnr[jidx+1];
 312             jnrlistC         = jjnr[jidx+2];
 313             jnrlistD         = jjnr[jidx+3];
 314             /* Sign of each element will be negative for non-real atoms.
 315              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 316              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 317              */
 318             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 319
 320             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 321             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 322             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 323
 324             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 325             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 326             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 327             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 328             j_coord_offsetA  = DIM*jnrA;
 329             j_coord_offsetB  = DIM*jnrB;
 330             j_coord_offsetC  = DIM*jnrC;
 331             j_coord_offsetD  = DIM*jnrD;
 332
 333             /* load j atom coordinates */
 334             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 335                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 336                                                  &jx0,&jy0,&jz0);
 337
 338             /* Calculate displacement vector */
 339             dx00             = _mm256_sub_pd(ix0,jx0);
 340             dy00             = _mm256_sub_pd(iy0,jy0);
 341             dz00             = _mm256_sub_pd(iz0,jz0);
 342
 343             /* Calculate squared distance and things based on it */
 344             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 345
 346             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 347
 348             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 349
 350             /* Load parameters for j particles */
 351             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 352                                                                  charge+jnrC+0,charge+jnrD+0);
 353             isaj0            = gmx_mm256_load_4real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0,
 354                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0);
 355             vdwjidx0A        = 2*vdwtype[jnrA+0];
 356             vdwjidx0B        = 2*vdwtype[jnrB+0];
 357             vdwjidx0C        = 2*vdwtype[jnrC+0];
 358             vdwjidx0D        = 2*vdwtype[jnrD+0];
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             r00              = _mm256_mul_pd(rsq00,rinv00);
 365             r00              = _mm256_andnot_pd(dummy_mask,r00);
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq00             = _mm256_mul_pd(iq0,jq0);
 369             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 370                                             vdwioffsetptr0+vdwjidx0B,
 371                                             vdwioffsetptr0+vdwjidx0C,
 372                                             vdwioffsetptr0+vdwjidx0D,
 373                                             &c6_00,&c12_00);
 374
 375             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 376             isaprod          = _mm256_mul_pd(isai0,isaj0);
 377             gbqqfactor       = _mm256_xor_pd(signbit,_mm256_mul_pd(qq00,_mm256_mul_pd(isaprod,gbinvepsdiff)));
 378             gbscale          = _mm256_mul_pd(isaprod,gbtabscale);
 379
 380             /* Calculate generalized born table index - this is a separate table from the normal one,
 381              * but we use the same procedure by multiplying r with scale and truncating to integer.
 382              */
 383             rt               = _mm256_mul_pd(r00,gbscale);
 384             gbitab           = _mm256_cvttpd_epi32(rt);
 385             gbeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 386             gbitab           = _mm_slli_epi32(gbitab,2);
 387             Y                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
 388             F                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,1) );
 389             G                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,2) );
 390             H                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,3) );
 391             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 392             Heps             = _mm256_mul_pd(gbeps,H);
 393             Fp               = _mm256_add_pd(F,_mm256_mul_pd(gbeps,_mm256_add_pd(G,Heps)));
 394             VV               = _mm256_add_pd(Y,_mm256_mul_pd(gbeps,Fp));
 395             vgb              = _mm256_mul_pd(gbqqfactor,VV);
 396
 397             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(gbeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 398             fgb              = _mm256_mul_pd(gbqqfactor,_mm256_mul_pd(FF,gbscale));
 399             dvdatmp          = _mm256_mul_pd(minushalf,_mm256_add_pd(vgb,_mm256_mul_pd(fgb,r00)));
 400             dvdasum          = _mm256_add_pd(dvdasum,dvdatmp);
 401             /* 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. */
 402             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 403             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 404             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 405             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 406             gmx_mm256_increment_4real_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 407                                                  _mm256_mul_pd(dvdatmp,_mm256_mul_pd(isaj0,isaj0)));
 408             velec            = _mm256_mul_pd(qq00,rinv00);
 409             felec            = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(velec,rinv00),fgb),rinv00);
 410
 411             /* LENNARD-JONES DISPERSION/REPULSION */
 412
 413             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 414             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 415             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 416             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 417             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velec            = _mm256_andnot_pd(dummy_mask,velec);
 421             velecsum         = _mm256_add_pd(velecsum,velec);
 422             vgb              = _mm256_andnot_pd(dummy_mask,vgb);
 423             vgbsum           = _mm256_add_pd(vgbsum,vgb);
 424             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 425             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 426
 427             fscal            = _mm256_add_pd(felec,fvdw);
 428
 429             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 430
 431             /* Calculate temporary vectorial force */
 432             tx               = _mm256_mul_pd(fscal,dx00);
 433             ty               = _mm256_mul_pd(fscal,dy00);
 434             tz               = _mm256_mul_pd(fscal,dz00);
 435
 436             /* Update vectorial force */
 437             fix0             = _mm256_add_pd(fix0,tx);
 438             fiy0             = _mm256_add_pd(fiy0,ty);
 439             fiz0             = _mm256_add_pd(fiz0,tz);
 440
 441             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 442             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 443             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 444             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 445             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 446
 447             /* Inner loop uses 71 flops */
 448         }
 449
 450         /* End of innermost loop */
 451
 452         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 453                                                  f+i_coord_offset,fshift+i_shift_offset);
 454
 455         ggid                        = gid[iidx];
 456         /* Update potential energies */
 457         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 458         gmx_mm256_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 459         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 460         dvdasum = _mm256_mul_pd(dvdasum, _mm256_mul_pd(isai0,isai0));
 461         gmx_mm256_update_1pot_pd(dvdasum,dvda+inr);
 462
 463         /* Increment number of inner iterations */
 464         inneriter                  += j_index_end - j_index_start;
 465
 466         /* Outer loop uses 10 flops */
 467     }
 468
 469     /* Increment number of outer iterations */
 470     outeriter        += nri;
 471
 472     /* Update outer/inner flops */
 473
 474     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*10 + inneriter*71);
 475 }
 476 /*
 477  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_avx_256_double
 478  * Electrostatics interaction: GeneralizedBorn
 479  * VdW interaction:            LennardJones
 480  * Geometry:                   Particle-Particle
 481  * Calculate force/pot:        Force
 482  */
 483 void
 484 nb_kernel_ElecGB_VdwLJ_GeomP1P1_F_avx_256_double
 485                     (t_nblist * gmx_restrict                nlist,
 486                      rvec * gmx_restrict                    xx,
 487                      rvec * gmx_restrict                    ff,
 488                      t_forcerec * gmx_restrict              fr,
 489                      t_mdatoms * gmx_restrict               mdatoms,
 490                      nb_kernel_data_t * gmx_restrict        kernel_data,
 491                      t_nrnb * gmx_restrict                  nrnb)
 492 {
 493     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 494      * just 0 for non-waters.
 495      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 496      * jnr indices corresponding to data put in the four positions in the SIMD register.
 497      */
 498     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 499     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 500     int              jnrA,jnrB,jnrC,jnrD;
 501     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 502     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 503     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 504     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 505     real             rcutoff_scalar;
 506     real             *shiftvec,*fshift,*x,*f;
 507     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 508     real             scratch[4*DIM];
 509     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 510     real *           vdwioffsetptr0;
 511     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 512     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 513     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 514     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 515     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 516     real             *charge;
 517     __m128i          gbitab;
 518     __m256d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 519     __m256d          minushalf = _mm256_set1_pd(-0.5);
 520     real             *invsqrta,*dvda,*gbtab;
 521     int              nvdwtype;
 522     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 523     int              *vdwtype;
 524     real             *vdwparam;
 525     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 526     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 527     __m128i          vfitab;
 528     __m128i          ifour       = _mm_set1_epi32(4);
 529     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 530     real             *vftab;
 531     __m256d          dummy_mask,cutoff_mask;
 532     __m128           tmpmask0,tmpmask1;
 533     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 534     __m256d          one     = _mm256_set1_pd(1.0);
 535     __m256d          two     = _mm256_set1_pd(2.0);
 536     x                = xx[0];
 537     f                = ff[0];
 538
 539     nri              = nlist->nri;
 540     iinr             = nlist->iinr;
 541     jindex           = nlist->jindex;
 542     jjnr             = nlist->jjnr;
 543     shiftidx         = nlist->shift;
 544     gid              = nlist->gid;
 545     shiftvec         = fr->shift_vec[0];
 546     fshift           = fr->fshift[0];
 547     facel            = _mm256_set1_pd(fr->epsfac);
 548     charge           = mdatoms->chargeA;
 549     nvdwtype         = fr->ntype;
 550     vdwparam         = fr->nbfp;
 551     vdwtype          = mdatoms->typeA;
 552
 553     invsqrta         = fr->invsqrta;
 554     dvda             = fr->dvda;
 555     gbtabscale       = _mm256_set1_pd(fr->gbtab.scale);
 556     gbtab            = fr->gbtab.data;
 557     gbinvepsdiff     = _mm256_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 558
 559     /* Avoid stupid compiler warnings */
 560     jnrA = jnrB = jnrC = jnrD = 0;
 561     j_coord_offsetA = 0;
 562     j_coord_offsetB = 0;
 563     j_coord_offsetC = 0;
 564     j_coord_offsetD = 0;
 565
 566     outeriter        = 0;
 567     inneriter        = 0;
 568
 569     for(iidx=0;iidx<4*DIM;iidx++)
 570     {
 571         scratch[iidx] = 0.0;
 572     }
 573
 574     /* Start outer loop over neighborlists */
 575     for(iidx=0; iidx<nri; iidx++)
 576     {
 577         /* Load shift vector for this list */
 578         i_shift_offset   = DIM*shiftidx[iidx];
 579
 580         /* Load limits for loop over neighbors */
 581         j_index_start    = jindex[iidx];
 582         j_index_end      = jindex[iidx+1];
 583
 584         /* Get outer coordinate index */
 585         inr              = iinr[iidx];
 586         i_coord_offset   = DIM*inr;
 587
 588         /* Load i particle coords and add shift vector */
 589         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 590
 591         fix0             = _mm256_setzero_pd();
 592         fiy0             = _mm256_setzero_pd();
 593         fiz0             = _mm256_setzero_pd();
 594
 595         /* Load parameters for i particles */
 596         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 597         isai0            = _mm256_set1_pd(invsqrta[inr+0]);
 598         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 599
 600         dvdasum          = _mm256_setzero_pd();
 601
 602         /* Start inner kernel loop */
 603         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 604         {
 605
 606             /* Get j neighbor index, and coordinate index */
 607             jnrA             = jjnr[jidx];
 608             jnrB             = jjnr[jidx+1];
 609             jnrC             = jjnr[jidx+2];
 610             jnrD             = jjnr[jidx+3];
 611             j_coord_offsetA  = DIM*jnrA;
 612             j_coord_offsetB  = DIM*jnrB;
 613             j_coord_offsetC  = DIM*jnrC;
 614             j_coord_offsetD  = DIM*jnrD;
 615
 616             /* load j atom coordinates */
 617             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 618                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 619                                                  &jx0,&jy0,&jz0);
 620
 621             /* Calculate displacement vector */
 622             dx00             = _mm256_sub_pd(ix0,jx0);
 623             dy00             = _mm256_sub_pd(iy0,jy0);
 624             dz00             = _mm256_sub_pd(iz0,jz0);
 625
 626             /* Calculate squared distance and things based on it */
 627             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 628
 629             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 630
 631             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 632
 633             /* Load parameters for j particles */
 634             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 635                                                                  charge+jnrC+0,charge+jnrD+0);
 636             isaj0            = gmx_mm256_load_4real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0,
 637                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0);
 638             vdwjidx0A        = 2*vdwtype[jnrA+0];
 639             vdwjidx0B        = 2*vdwtype[jnrB+0];
 640             vdwjidx0C        = 2*vdwtype[jnrC+0];
 641             vdwjidx0D        = 2*vdwtype[jnrD+0];
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             r00              = _mm256_mul_pd(rsq00,rinv00);
 648
 649             /* Compute parameters for interactions between i and j atoms */
 650             qq00             = _mm256_mul_pd(iq0,jq0);
 651             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 652                                             vdwioffsetptr0+vdwjidx0B,
 653                                             vdwioffsetptr0+vdwjidx0C,
 654                                             vdwioffsetptr0+vdwjidx0D,
 655                                             &c6_00,&c12_00);
 656
 657             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 658             isaprod          = _mm256_mul_pd(isai0,isaj0);
 659             gbqqfactor       = _mm256_xor_pd(signbit,_mm256_mul_pd(qq00,_mm256_mul_pd(isaprod,gbinvepsdiff)));
 660             gbscale          = _mm256_mul_pd(isaprod,gbtabscale);
 661
 662             /* Calculate generalized born table index - this is a separate table from the normal one,
 663              * but we use the same procedure by multiplying r with scale and truncating to integer.
 664              */
 665             rt               = _mm256_mul_pd(r00,gbscale);
 666             gbitab           = _mm256_cvttpd_epi32(rt);
 667             gbeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 668             gbitab           = _mm_slli_epi32(gbitab,2);
 669             Y                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
 670             F                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,1) );
 671             G                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,2) );
 672             H                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,3) );
 673             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 674             Heps             = _mm256_mul_pd(gbeps,H);
 675             Fp               = _mm256_add_pd(F,_mm256_mul_pd(gbeps,_mm256_add_pd(G,Heps)));
 676             VV               = _mm256_add_pd(Y,_mm256_mul_pd(gbeps,Fp));
 677             vgb              = _mm256_mul_pd(gbqqfactor,VV);
 678
 679             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(gbeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 680             fgb              = _mm256_mul_pd(gbqqfactor,_mm256_mul_pd(FF,gbscale));
 681             dvdatmp          = _mm256_mul_pd(minushalf,_mm256_add_pd(vgb,_mm256_mul_pd(fgb,r00)));
 682             dvdasum          = _mm256_add_pd(dvdasum,dvdatmp);
 683             fjptrA           = dvda+jnrA;
 684             fjptrB           = dvda+jnrB;
 685             fjptrC           = dvda+jnrC;
 686             fjptrD           = dvda+jnrD;
 687             gmx_mm256_increment_4real_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 688                                                  _mm256_mul_pd(dvdatmp,_mm256_mul_pd(isaj0,isaj0)));
 689             velec            = _mm256_mul_pd(qq00,rinv00);
 690             felec            = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(velec,rinv00),fgb),rinv00);
 691
 692             /* LENNARD-JONES DISPERSION/REPULSION */
 693
 694             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 695             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 696
 697             fscal            = _mm256_add_pd(felec,fvdw);
 698
 699             /* Calculate temporary vectorial force */
 700             tx               = _mm256_mul_pd(fscal,dx00);
 701             ty               = _mm256_mul_pd(fscal,dy00);
 702             tz               = _mm256_mul_pd(fscal,dz00);
 703
 704             /* Update vectorial force */
 705             fix0             = _mm256_add_pd(fix0,tx);
 706             fiy0             = _mm256_add_pd(fiy0,ty);
 707             fiz0             = _mm256_add_pd(fiz0,tz);
 708
 709             fjptrA             = f+j_coord_offsetA;
 710             fjptrB             = f+j_coord_offsetB;
 711             fjptrC             = f+j_coord_offsetC;
 712             fjptrD             = f+j_coord_offsetD;
 713             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 714
 715             /* Inner loop uses 63 flops */
 716         }
 717
 718         if(jidx<j_index_end)
 719         {
 720
 721             /* Get j neighbor index, and coordinate index */
 722             jnrlistA         = jjnr[jidx];
 723             jnrlistB         = jjnr[jidx+1];
 724             jnrlistC         = jjnr[jidx+2];
 725             jnrlistD         = jjnr[jidx+3];
 726             /* Sign of each element will be negative for non-real atoms.
 727              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 728              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 729              */
 730             tmpmask0 = gmx_mm_castsi128_pd(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 731
 732             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 733             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 734             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 735
 736             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 737             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 738             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 739             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 740             j_coord_offsetA  = DIM*jnrA;
 741             j_coord_offsetB  = DIM*jnrB;
 742             j_coord_offsetC  = DIM*jnrC;
 743             j_coord_offsetD  = DIM*jnrD;
 744
 745             /* load j atom coordinates */
 746             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 747                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 748                                                  &jx0,&jy0,&jz0);
 749
 750             /* Calculate displacement vector */
 751             dx00             = _mm256_sub_pd(ix0,jx0);
 752             dy00             = _mm256_sub_pd(iy0,jy0);
 753             dz00             = _mm256_sub_pd(iz0,jz0);
 754
 755             /* Calculate squared distance and things based on it */
 756             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 757
 758             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 759
 760             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 761
 762             /* Load parameters for j particles */
 763             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 764                                                                  charge+jnrC+0,charge+jnrD+0);
 765             isaj0            = gmx_mm256_load_4real_swizzle_pd(invsqrta+jnrA+0,invsqrta+jnrB+0,
 766                                                                  invsqrta+jnrC+0,invsqrta+jnrD+0);
 767             vdwjidx0A        = 2*vdwtype[jnrA+0];
 768             vdwjidx0B        = 2*vdwtype[jnrB+0];
 769             vdwjidx0C        = 2*vdwtype[jnrC+0];
 770             vdwjidx0D        = 2*vdwtype[jnrD+0];
 771
 772             /**************************
 773              * CALCULATE INTERACTIONS *
 774              **************************/
 775
 776             r00              = _mm256_mul_pd(rsq00,rinv00);
 777             r00              = _mm256_andnot_pd(dummy_mask,r00);
 778
 779             /* Compute parameters for interactions between i and j atoms */
 780             qq00             = _mm256_mul_pd(iq0,jq0);
 781             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 782                                             vdwioffsetptr0+vdwjidx0B,
 783                                             vdwioffsetptr0+vdwjidx0C,
 784                                             vdwioffsetptr0+vdwjidx0D,
 785                                             &c6_00,&c12_00);
 786
 787             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 788             isaprod          = _mm256_mul_pd(isai0,isaj0);
 789             gbqqfactor       = _mm256_xor_pd(signbit,_mm256_mul_pd(qq00,_mm256_mul_pd(isaprod,gbinvepsdiff)));
 790             gbscale          = _mm256_mul_pd(isaprod,gbtabscale);
 791
 792             /* Calculate generalized born table index - this is a separate table from the normal one,
 793              * but we use the same procedure by multiplying r with scale and truncating to integer.
 794              */
 795             rt               = _mm256_mul_pd(r00,gbscale);
 796             gbitab           = _mm256_cvttpd_epi32(rt);
 797             gbeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 798             gbitab           = _mm_slli_epi32(gbitab,2);
 799             Y                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
 800             F                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,1) );
 801             G                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,2) );
 802             H                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,3) );
 803             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 804             Heps             = _mm256_mul_pd(gbeps,H);
 805             Fp               = _mm256_add_pd(F,_mm256_mul_pd(gbeps,_mm256_add_pd(G,Heps)));
 806             VV               = _mm256_add_pd(Y,_mm256_mul_pd(gbeps,Fp));
 807             vgb              = _mm256_mul_pd(gbqqfactor,VV);
 808
 809             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(gbeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 810             fgb              = _mm256_mul_pd(gbqqfactor,_mm256_mul_pd(FF,gbscale));
 811             dvdatmp          = _mm256_mul_pd(minushalf,_mm256_add_pd(vgb,_mm256_mul_pd(fgb,r00)));
 812             dvdasum          = _mm256_add_pd(dvdasum,dvdatmp);
 813             /* 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. */
 814             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 815             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 816             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 817             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 818             gmx_mm256_increment_4real_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 819                                                  _mm256_mul_pd(dvdatmp,_mm256_mul_pd(isaj0,isaj0)));
 820             velec            = _mm256_mul_pd(qq00,rinv00);
 821             felec            = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(velec,rinv00),fgb),rinv00);
 822
 823             /* LENNARD-JONES DISPERSION/REPULSION */
 824
 825             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 826             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 827
 828             fscal            = _mm256_add_pd(felec,fvdw);
 829
 830             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 831
 832             /* Calculate temporary vectorial force */
 833             tx               = _mm256_mul_pd(fscal,dx00);
 834             ty               = _mm256_mul_pd(fscal,dy00);
 835             tz               = _mm256_mul_pd(fscal,dz00);
 836
 837             /* Update vectorial force */
 838             fix0             = _mm256_add_pd(fix0,tx);
 839             fiy0             = _mm256_add_pd(fiy0,ty);
 840             fiz0             = _mm256_add_pd(fiz0,tz);
 841
 842             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 843             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 844             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 845             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 846             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 847
 848             /* Inner loop uses 64 flops */
 849         }
 850
 851         /* End of innermost loop */
 852
 853         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 854                                                  f+i_coord_offset,fshift+i_shift_offset);
 855
 856         dvdasum = _mm256_mul_pd(dvdasum, _mm256_mul_pd(isai0,isai0));
 857         gmx_mm256_update_1pot_pd(dvdasum,dvda+inr);
 858
 859         /* Increment number of inner iterations */
 860         inneriter                  += j_index_end - j_index_start;
 861
 862         /* Outer loop uses 7 flops */
 863     }
 864
 865     /* Increment number of outer iterations */
 866     outeriter        += nri;
 867
 868     /* Update outer/inner flops */
 869
 870     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 871 }