src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwCSTab_GeomP1P1_avx_256_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_avx_256_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  86     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  89     real             *charge;
  90     int              nvdwtype;
  91     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  92     int              *vdwtype;
  93     real             *vdwparam;
  94     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  95     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  96     __m128i          vfitab;
  97     __m128i          ifour       = _mm_set1_epi32(4);
  98     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  99     real             *vftab;
 100     __m128i          ewitab;
 101     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 102     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 103     real             *ewtab;
 104     __m256d          dummy_mask,cutoff_mask;
 105     __m128           tmpmask0,tmpmask1;
 106     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 107     __m256d          one     = _mm256_set1_pd(1.0);
 108     __m256d          two     = _mm256_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm256_set1_pd(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_vdw->data;
 127     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 128
 129     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 130     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 131     beta2            = _mm256_mul_pd(beta,beta);
 132     beta3            = _mm256_mul_pd(beta,beta2);
 133
 134     ewtab            = fr->ic->tabq_coul_FDV0;
 135     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 136     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 169
 170         fix0             = _mm256_setzero_pd();
 171         fiy0             = _mm256_setzero_pd();
 172         fiz0             = _mm256_setzero_pd();
 173
 174         /* Load parameters for i particles */
 175         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 176         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm256_setzero_pd();
 180         vvdwsum          = _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           = avx256_invsqrt_d(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             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218             vdwjidx0C        = 2*vdwtype[jnrC+0];
 219             vdwjidx0D        = 2*vdwtype[jnrD+0];
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm256_mul_pd(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm256_mul_pd(iq0,jq0);
 229             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 230                                             vdwioffsetptr0+vdwjidx0B,
 231                                             vdwioffsetptr0+vdwjidx0C,
 232                                             vdwioffsetptr0+vdwjidx0D,
 233                                             &c6_00,&c12_00);
 234
 235             /* Calculate table index by multiplying r with table scale and truncate to integer */
 236             rt               = _mm256_mul_pd(r00,vftabscale);
 237             vfitab           = _mm256_cvttpd_epi32(rt);
 238             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 239             vfitab           = _mm_slli_epi32(vfitab,3);
 240
 241             /* EWALD ELECTROSTATICS */
 242
 243             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 244             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 245             ewitab           = _mm256_cvttpd_epi32(ewrt);
 246             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 247             ewitab           = _mm_slli_epi32(ewitab,2);
 248             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 249             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 250             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 251             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 252             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 253             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 254             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 255             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 256             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 257
 258             /* CUBIC SPLINE TABLE DISPERSION */
 259             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 260             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 261             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 262             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 263             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 264             Heps             = _mm256_mul_pd(vfeps,H);
 265             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 266             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 267             vvdw6            = _mm256_mul_pd(c6_00,VV);
 268             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 269             fvdw6            = _mm256_mul_pd(c6_00,FF);
 270
 271             /* CUBIC SPLINE TABLE REPULSION */
 272             vfitab           = _mm_add_epi32(vfitab,ifour);
 273             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 274             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 275             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 276             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 277             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 278             Heps             = _mm256_mul_pd(vfeps,H);
 279             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 280             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 281             vvdw12           = _mm256_mul_pd(c12_00,VV);
 282             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 283             fvdw12           = _mm256_mul_pd(c12_00,FF);
 284             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 285             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             velecsum         = _mm256_add_pd(velecsum,velec);
 289             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 290
 291             fscal            = _mm256_add_pd(felec,fvdw);
 292
 293             /* Calculate temporary vectorial force */
 294             tx               = _mm256_mul_pd(fscal,dx00);
 295             ty               = _mm256_mul_pd(fscal,dy00);
 296             tz               = _mm256_mul_pd(fscal,dz00);
 297
 298             /* Update vectorial force */
 299             fix0             = _mm256_add_pd(fix0,tx);
 300             fiy0             = _mm256_add_pd(fiy0,ty);
 301             fiz0             = _mm256_add_pd(fiz0,tz);
 302
 303             fjptrA             = f+j_coord_offsetA;
 304             fjptrB             = f+j_coord_offsetB;
 305             fjptrC             = f+j_coord_offsetC;
 306             fjptrD             = f+j_coord_offsetD;
 307             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 308
 309             /* Inner loop uses 75 flops */
 310         }
 311
 312         if(jidx<j_index_end)
 313         {
 314
 315             /* Get j neighbor index, and coordinate index */
 316             jnrlistA         = jjnr[jidx];
 317             jnrlistB         = jjnr[jidx+1];
 318             jnrlistC         = jjnr[jidx+2];
 319             jnrlistD         = jjnr[jidx+3];
 320             /* Sign of each element will be negative for non-real atoms.
 321              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 322              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 323              */
 324             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 325
 326             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 327             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 328             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 329
 330             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 331             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 332             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 333             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 334             j_coord_offsetA  = DIM*jnrA;
 335             j_coord_offsetB  = DIM*jnrB;
 336             j_coord_offsetC  = DIM*jnrC;
 337             j_coord_offsetD  = DIM*jnrD;
 338
 339             /* load j atom coordinates */
 340             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 341                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 342                                                  &jx0,&jy0,&jz0);
 343
 344             /* Calculate displacement vector */
 345             dx00             = _mm256_sub_pd(ix0,jx0);
 346             dy00             = _mm256_sub_pd(iy0,jy0);
 347             dz00             = _mm256_sub_pd(iz0,jz0);
 348
 349             /* Calculate squared distance and things based on it */
 350             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 351
 352             rinv00           = avx256_invsqrt_d(rsq00);
 353
 354             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 355
 356             /* Load parameters for j particles */
 357             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 358                                                                  charge+jnrC+0,charge+jnrD+0);
 359             vdwjidx0A        = 2*vdwtype[jnrA+0];
 360             vdwjidx0B        = 2*vdwtype[jnrB+0];
 361             vdwjidx0C        = 2*vdwtype[jnrC+0];
 362             vdwjidx0D        = 2*vdwtype[jnrD+0];
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             r00              = _mm256_mul_pd(rsq00,rinv00);
 369             r00              = _mm256_andnot_pd(dummy_mask,r00);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq00             = _mm256_mul_pd(iq0,jq0);
 373             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 374                                             vdwioffsetptr0+vdwjidx0B,
 375                                             vdwioffsetptr0+vdwjidx0C,
 376                                             vdwioffsetptr0+vdwjidx0D,
 377                                             &c6_00,&c12_00);
 378
 379             /* Calculate table index by multiplying r with table scale and truncate to integer */
 380             rt               = _mm256_mul_pd(r00,vftabscale);
 381             vfitab           = _mm256_cvttpd_epi32(rt);
 382             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 383             vfitab           = _mm_slli_epi32(vfitab,3);
 384
 385             /* EWALD ELECTROSTATICS */
 386
 387             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 388             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 389             ewitab           = _mm256_cvttpd_epi32(ewrt);
 390             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 391             ewitab           = _mm_slli_epi32(ewitab,2);
 392             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 393             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 394             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 395             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 396             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 397             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 398             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 399             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 400             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 401
 402             /* CUBIC SPLINE TABLE DISPERSION */
 403             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 404             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 405             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 406             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 407             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 408             Heps             = _mm256_mul_pd(vfeps,H);
 409             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 410             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 411             vvdw6            = _mm256_mul_pd(c6_00,VV);
 412             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 413             fvdw6            = _mm256_mul_pd(c6_00,FF);
 414
 415             /* CUBIC SPLINE TABLE REPULSION */
 416             vfitab           = _mm_add_epi32(vfitab,ifour);
 417             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 418             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 419             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 420             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 421             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 422             Heps             = _mm256_mul_pd(vfeps,H);
 423             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 424             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 425             vvdw12           = _mm256_mul_pd(c12_00,VV);
 426             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 427             fvdw12           = _mm256_mul_pd(c12_00,FF);
 428             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 429             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velec            = _mm256_andnot_pd(dummy_mask,velec);
 433             velecsum         = _mm256_add_pd(velecsum,velec);
 434             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 435             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 436
 437             fscal            = _mm256_add_pd(felec,fvdw);
 438
 439             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 440
 441             /* Calculate temporary vectorial force */
 442             tx               = _mm256_mul_pd(fscal,dx00);
 443             ty               = _mm256_mul_pd(fscal,dy00);
 444             tz               = _mm256_mul_pd(fscal,dz00);
 445
 446             /* Update vectorial force */
 447             fix0             = _mm256_add_pd(fix0,tx);
 448             fiy0             = _mm256_add_pd(fiy0,ty);
 449             fiz0             = _mm256_add_pd(fiz0,tz);
 450
 451             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 452             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 453             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 454             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 455             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 456
 457             /* Inner loop uses 76 flops */
 458         }
 459
 460         /* End of innermost loop */
 461
 462         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 463                                                  f+i_coord_offset,fshift+i_shift_offset);
 464
 465         ggid                        = gid[iidx];
 466         /* Update potential energies */
 467         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 468         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 469
 470         /* Increment number of inner iterations */
 471         inneriter                  += j_index_end - j_index_start;
 472
 473         /* Outer loop uses 9 flops */
 474     }
 475
 476     /* Increment number of outer iterations */
 477     outeriter        += nri;
 478
 479     /* Update outer/inner flops */
 480
 481     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*76);
 482 }
 483 /*
 484  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_256_double
 485  * Electrostatics interaction: Ewald
 486  * VdW interaction:            CubicSplineTable
 487  * Geometry:                   Particle-Particle
 488  * Calculate force/pot:        Force
 489  */
 490 void
 491 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_256_double
 492                     (t_nblist                    * gmx_restrict       nlist,
 493                      rvec                        * gmx_restrict          xx,
 494                      rvec                        * gmx_restrict          ff,
 495                      struct t_forcerec           * gmx_restrict          fr,
 496                      t_mdatoms                   * gmx_restrict     mdatoms,
 497                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 498                      t_nrnb                      * gmx_restrict        nrnb)
 499 {
 500     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 501      * just 0 for non-waters.
 502      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 503      * jnr indices corresponding to data put in the four positions in the SIMD register.
 504      */
 505     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 506     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 507     int              jnrA,jnrB,jnrC,jnrD;
 508     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 509     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 510     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 511     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 512     real             rcutoff_scalar;
 513     real             *shiftvec,*fshift,*x,*f;
 514     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 515     real             scratch[4*DIM];
 516     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 517     real *           vdwioffsetptr0;
 518     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 519     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 520     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 521     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 522     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 523     real             *charge;
 524     int              nvdwtype;
 525     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 526     int              *vdwtype;
 527     real             *vdwparam;
 528     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 529     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 530     __m128i          vfitab;
 531     __m128i          ifour       = _mm_set1_epi32(4);
 532     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 533     real             *vftab;
 534     __m128i          ewitab;
 535     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 536     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 537     real             *ewtab;
 538     __m256d          dummy_mask,cutoff_mask;
 539     __m128           tmpmask0,tmpmask1;
 540     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 541     __m256d          one     = _mm256_set1_pd(1.0);
 542     __m256d          two     = _mm256_set1_pd(2.0);
 543     x                = xx[0];
 544     f                = ff[0];
 545
 546     nri              = nlist->nri;
 547     iinr             = nlist->iinr;
 548     jindex           = nlist->jindex;
 549     jjnr             = nlist->jjnr;
 550     shiftidx         = nlist->shift;
 551     gid              = nlist->gid;
 552     shiftvec         = fr->shift_vec[0];
 553     fshift           = fr->fshift[0];
 554     facel            = _mm256_set1_pd(fr->ic->epsfac);
 555     charge           = mdatoms->chargeA;
 556     nvdwtype         = fr->ntype;
 557     vdwparam         = fr->nbfp;
 558     vdwtype          = mdatoms->typeA;
 559
 560     vftab            = kernel_data->table_vdw->data;
 561     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 562
 563     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 564     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 565     beta2            = _mm256_mul_pd(beta,beta);
 566     beta3            = _mm256_mul_pd(beta,beta2);
 567
 568     ewtab            = fr->ic->tabq_coul_F;
 569     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 570     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 571
 572     /* Avoid stupid compiler warnings */
 573     jnrA = jnrB = jnrC = jnrD = 0;
 574     j_coord_offsetA = 0;
 575     j_coord_offsetB = 0;
 576     j_coord_offsetC = 0;
 577     j_coord_offsetD = 0;
 578
 579     outeriter        = 0;
 580     inneriter        = 0;
 581
 582     for(iidx=0;iidx<4*DIM;iidx++)
 583     {
 584         scratch[iidx] = 0.0;
 585     }
 586
 587     /* Start outer loop over neighborlists */
 588     for(iidx=0; iidx<nri; iidx++)
 589     {
 590         /* Load shift vector for this list */
 591         i_shift_offset   = DIM*shiftidx[iidx];
 592
 593         /* Load limits for loop over neighbors */
 594         j_index_start    = jindex[iidx];
 595         j_index_end      = jindex[iidx+1];
 596
 597         /* Get outer coordinate index */
 598         inr              = iinr[iidx];
 599         i_coord_offset   = DIM*inr;
 600
 601         /* Load i particle coords and add shift vector */
 602         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 603
 604         fix0             = _mm256_setzero_pd();
 605         fiy0             = _mm256_setzero_pd();
 606         fiz0             = _mm256_setzero_pd();
 607
 608         /* Load parameters for i particles */
 609         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 610         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 611
 612         /* Start inner kernel loop */
 613         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 614         {
 615
 616             /* Get j neighbor index, and coordinate index */
 617             jnrA             = jjnr[jidx];
 618             jnrB             = jjnr[jidx+1];
 619             jnrC             = jjnr[jidx+2];
 620             jnrD             = jjnr[jidx+3];
 621             j_coord_offsetA  = DIM*jnrA;
 622             j_coord_offsetB  = DIM*jnrB;
 623             j_coord_offsetC  = DIM*jnrC;
 624             j_coord_offsetD  = DIM*jnrD;
 625
 626             /* load j atom coordinates */
 627             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 628                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 629                                                  &jx0,&jy0,&jz0);
 630
 631             /* Calculate displacement vector */
 632             dx00             = _mm256_sub_pd(ix0,jx0);
 633             dy00             = _mm256_sub_pd(iy0,jy0);
 634             dz00             = _mm256_sub_pd(iz0,jz0);
 635
 636             /* Calculate squared distance and things based on it */
 637             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 638
 639             rinv00           = avx256_invsqrt_d(rsq00);
 640
 641             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 642
 643             /* Load parameters for j particles */
 644             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 645                                                                  charge+jnrC+0,charge+jnrD+0);
 646             vdwjidx0A        = 2*vdwtype[jnrA+0];
 647             vdwjidx0B        = 2*vdwtype[jnrB+0];
 648             vdwjidx0C        = 2*vdwtype[jnrC+0];
 649             vdwjidx0D        = 2*vdwtype[jnrD+0];
 650
 651             /**************************
 652              * CALCULATE INTERACTIONS *
 653              **************************/
 654
 655             r00              = _mm256_mul_pd(rsq00,rinv00);
 656
 657             /* Compute parameters for interactions between i and j atoms */
 658             qq00             = _mm256_mul_pd(iq0,jq0);
 659             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 660                                             vdwioffsetptr0+vdwjidx0B,
 661                                             vdwioffsetptr0+vdwjidx0C,
 662                                             vdwioffsetptr0+vdwjidx0D,
 663                                             &c6_00,&c12_00);
 664
 665             /* Calculate table index by multiplying r with table scale and truncate to integer */
 666             rt               = _mm256_mul_pd(r00,vftabscale);
 667             vfitab           = _mm256_cvttpd_epi32(rt);
 668             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 669             vfitab           = _mm_slli_epi32(vfitab,3);
 670
 671             /* EWALD ELECTROSTATICS */
 672
 673             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 674             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 675             ewitab           = _mm256_cvttpd_epi32(ewrt);
 676             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 677             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 678                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 679                                             &ewtabF,&ewtabFn);
 680             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 681             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 682
 683             /* CUBIC SPLINE TABLE DISPERSION */
 684             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 685             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 686             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 687             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 688             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 689             Heps             = _mm256_mul_pd(vfeps,H);
 690             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 691             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 692             fvdw6            = _mm256_mul_pd(c6_00,FF);
 693
 694             /* CUBIC SPLINE TABLE REPULSION */
 695             vfitab           = _mm_add_epi32(vfitab,ifour);
 696             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 697             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 698             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 699             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 700             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 701             Heps             = _mm256_mul_pd(vfeps,H);
 702             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 703             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 704             fvdw12           = _mm256_mul_pd(c12_00,FF);
 705             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 706
 707             fscal            = _mm256_add_pd(felec,fvdw);
 708
 709             /* Calculate temporary vectorial force */
 710             tx               = _mm256_mul_pd(fscal,dx00);
 711             ty               = _mm256_mul_pd(fscal,dy00);
 712             tz               = _mm256_mul_pd(fscal,dz00);
 713
 714             /* Update vectorial force */
 715             fix0             = _mm256_add_pd(fix0,tx);
 716             fiy0             = _mm256_add_pd(fiy0,ty);
 717             fiz0             = _mm256_add_pd(fiz0,tz);
 718
 719             fjptrA             = f+j_coord_offsetA;
 720             fjptrB             = f+j_coord_offsetB;
 721             fjptrC             = f+j_coord_offsetC;
 722             fjptrD             = f+j_coord_offsetD;
 723             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 724
 725             /* Inner loop uses 62 flops */
 726         }
 727
 728         if(jidx<j_index_end)
 729         {
 730
 731             /* Get j neighbor index, and coordinate index */
 732             jnrlistA         = jjnr[jidx];
 733             jnrlistB         = jjnr[jidx+1];
 734             jnrlistC         = jjnr[jidx+2];
 735             jnrlistD         = jjnr[jidx+3];
 736             /* Sign of each element will be negative for non-real atoms.
 737              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 738              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 739              */
 740             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 741
 742             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 743             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 744             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 745
 746             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 747             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 748             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 749             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 750             j_coord_offsetA  = DIM*jnrA;
 751             j_coord_offsetB  = DIM*jnrB;
 752             j_coord_offsetC  = DIM*jnrC;
 753             j_coord_offsetD  = DIM*jnrD;
 754
 755             /* load j atom coordinates */
 756             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 757                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 758                                                  &jx0,&jy0,&jz0);
 759
 760             /* Calculate displacement vector */
 761             dx00             = _mm256_sub_pd(ix0,jx0);
 762             dy00             = _mm256_sub_pd(iy0,jy0);
 763             dz00             = _mm256_sub_pd(iz0,jz0);
 764
 765             /* Calculate squared distance and things based on it */
 766             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 767
 768             rinv00           = avx256_invsqrt_d(rsq00);
 769
 770             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 771
 772             /* Load parameters for j particles */
 773             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 774                                                                  charge+jnrC+0,charge+jnrD+0);
 775             vdwjidx0A        = 2*vdwtype[jnrA+0];
 776             vdwjidx0B        = 2*vdwtype[jnrB+0];
 777             vdwjidx0C        = 2*vdwtype[jnrC+0];
 778             vdwjidx0D        = 2*vdwtype[jnrD+0];
 779
 780             /**************************
 781              * CALCULATE INTERACTIONS *
 782              **************************/
 783
 784             r00              = _mm256_mul_pd(rsq00,rinv00);
 785             r00              = _mm256_andnot_pd(dummy_mask,r00);
 786
 787             /* Compute parameters for interactions between i and j atoms */
 788             qq00             = _mm256_mul_pd(iq0,jq0);
 789             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 790                                             vdwioffsetptr0+vdwjidx0B,
 791                                             vdwioffsetptr0+vdwjidx0C,
 792                                             vdwioffsetptr0+vdwjidx0D,
 793                                             &c6_00,&c12_00);
 794
 795             /* Calculate table index by multiplying r with table scale and truncate to integer */
 796             rt               = _mm256_mul_pd(r00,vftabscale);
 797             vfitab           = _mm256_cvttpd_epi32(rt);
 798             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 799             vfitab           = _mm_slli_epi32(vfitab,3);
 800
 801             /* EWALD ELECTROSTATICS */
 802
 803             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 804             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 805             ewitab           = _mm256_cvttpd_epi32(ewrt);
 806             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 807             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 808                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 809                                             &ewtabF,&ewtabFn);
 810             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 811             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 812
 813             /* CUBIC SPLINE TABLE DISPERSION */
 814             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 815             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 816             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 817             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 818             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 819             Heps             = _mm256_mul_pd(vfeps,H);
 820             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 821             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 822             fvdw6            = _mm256_mul_pd(c6_00,FF);
 823
 824             /* CUBIC SPLINE TABLE REPULSION */
 825             vfitab           = _mm_add_epi32(vfitab,ifour);
 826             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 827             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 828             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 829             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 830             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 831             Heps             = _mm256_mul_pd(vfeps,H);
 832             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 833             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 834             fvdw12           = _mm256_mul_pd(c12_00,FF);
 835             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 836
 837             fscal            = _mm256_add_pd(felec,fvdw);
 838
 839             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 840
 841             /* Calculate temporary vectorial force */
 842             tx               = _mm256_mul_pd(fscal,dx00);
 843             ty               = _mm256_mul_pd(fscal,dy00);
 844             tz               = _mm256_mul_pd(fscal,dz00);
 845
 846             /* Update vectorial force */
 847             fix0             = _mm256_add_pd(fix0,tx);
 848             fiy0             = _mm256_add_pd(fiy0,ty);
 849             fiz0             = _mm256_add_pd(fiz0,tz);
 850
 851             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 852             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 853             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 854             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 855             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 856
 857             /* Inner loop uses 63 flops */
 858         }
 859
 860         /* End of innermost loop */
 861
 862         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 863                                                  f+i_coord_offset,fshift+i_shift_offset);
 864
 865         /* Increment number of inner iterations */
 866         inneriter                  += j_index_end - j_index_start;
 867
 868         /* Outer loop uses 7 flops */
 869     }
 870
 871     /* Increment number of outer iterations */
 872     outeriter        += nri;
 873
 874     /* Update outer/inner flops */
 875
 876     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*63);
 877 }