src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_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_unused * 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     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 107     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 108     __m128i          vfitab;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256d          dummy_mask,cutoff_mask;
 113     __m128           tmpmask0,tmpmask1;
 114     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 115     __m256d          one     = _mm256_set1_pd(1.0);
 116     __m256d          two     = _mm256_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     vftab            = kernel_data->table_elec->data;
 135     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 140     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 141     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 176
 177         fix0             = _mm256_setzero_pd();
 178         fiy0             = _mm256_setzero_pd();
 179         fiz0             = _mm256_setzero_pd();
 180         fix1             = _mm256_setzero_pd();
 181         fiy1             = _mm256_setzero_pd();
 182         fiz1             = _mm256_setzero_pd();
 183         fix2             = _mm256_setzero_pd();
 184         fiy2             = _mm256_setzero_pd();
 185         fiz2             = _mm256_setzero_pd();
 186         fix3             = _mm256_setzero_pd();
 187         fiy3             = _mm256_setzero_pd();
 188         fiz3             = _mm256_setzero_pd();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_pd();
 192         vvdwsum          = _mm256_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_pd(ix0,jx0);
 215             dy00             = _mm256_sub_pd(iy0,jy0);
 216             dz00             = _mm256_sub_pd(iz0,jz0);
 217             dx10             = _mm256_sub_pd(ix1,jx0);
 218             dy10             = _mm256_sub_pd(iy1,jy0);
 219             dz10             = _mm256_sub_pd(iz1,jz0);
 220             dx20             = _mm256_sub_pd(ix2,jx0);
 221             dy20             = _mm256_sub_pd(iy2,jy0);
 222             dz20             = _mm256_sub_pd(iz2,jz0);
 223             dx30             = _mm256_sub_pd(ix3,jx0);
 224             dy30             = _mm256_sub_pd(iy3,jy0);
 225             dz30             = _mm256_sub_pd(iz3,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 229             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 230             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 231             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 232
 233             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 234             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 235             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 236
 237             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 241                                                                  charge+jnrC+0,charge+jnrD+0);
 242             vdwjidx0A        = 2*vdwtype[jnrA+0];
 243             vdwjidx0B        = 2*vdwtype[jnrB+0];
 244             vdwjidx0C        = 2*vdwtype[jnrC+0];
 245             vdwjidx0D        = 2*vdwtype[jnrD+0];
 246
 247             fjx0             = _mm256_setzero_pd();
 248             fjy0             = _mm256_setzero_pd();
 249             fjz0             = _mm256_setzero_pd();
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 257                                             vdwioffsetptr0+vdwjidx0B,
 258                                             vdwioffsetptr0+vdwjidx0C,
 259                                             vdwioffsetptr0+vdwjidx0D,
 260                                             &c6_00,&c12_00);
 261
 262             /* LENNARD-JONES DISPERSION/REPULSION */
 263
 264             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 265             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 266             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 267             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 268             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 269
 270             /* Update potential sum for this i atom from the interaction with this j atom. */
 271             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 272
 273             fscal            = fvdw;
 274
 275             /* Calculate temporary vectorial force */
 276             tx               = _mm256_mul_pd(fscal,dx00);
 277             ty               = _mm256_mul_pd(fscal,dy00);
 278             tz               = _mm256_mul_pd(fscal,dz00);
 279
 280             /* Update vectorial force */
 281             fix0             = _mm256_add_pd(fix0,tx);
 282             fiy0             = _mm256_add_pd(fiy0,ty);
 283             fiz0             = _mm256_add_pd(fiz0,tz);
 284
 285             fjx0             = _mm256_add_pd(fjx0,tx);
 286             fjy0             = _mm256_add_pd(fjy0,ty);
 287             fjz0             = _mm256_add_pd(fjz0,tz);
 288
 289             /**************************
 290              * CALCULATE INTERACTIONS *
 291              **************************/
 292
 293             r10              = _mm256_mul_pd(rsq10,rinv10);
 294
 295             /* Compute parameters for interactions between i and j atoms */
 296             qq10             = _mm256_mul_pd(iq1,jq0);
 297
 298             /* Calculate table index by multiplying r with table scale and truncate to integer */
 299             rt               = _mm256_mul_pd(r10,vftabscale);
 300             vfitab           = _mm256_cvttpd_epi32(rt);
 301             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 302             vfitab           = _mm_slli_epi32(vfitab,2);
 303
 304             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 305             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 306             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 307             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 308             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 309             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 310             Heps             = _mm256_mul_pd(vfeps,H);
 311             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 312             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 313             velec            = _mm256_mul_pd(qq10,VV);
 314             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 315             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 316
 317             /* Update potential sum for this i atom from the interaction with this j atom. */
 318             velecsum         = _mm256_add_pd(velecsum,velec);
 319
 320             fscal            = felec;
 321
 322             /* Calculate temporary vectorial force */
 323             tx               = _mm256_mul_pd(fscal,dx10);
 324             ty               = _mm256_mul_pd(fscal,dy10);
 325             tz               = _mm256_mul_pd(fscal,dz10);
 326
 327             /* Update vectorial force */
 328             fix1             = _mm256_add_pd(fix1,tx);
 329             fiy1             = _mm256_add_pd(fiy1,ty);
 330             fiz1             = _mm256_add_pd(fiz1,tz);
 331
 332             fjx0             = _mm256_add_pd(fjx0,tx);
 333             fjy0             = _mm256_add_pd(fjy0,ty);
 334             fjz0             = _mm256_add_pd(fjz0,tz);
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             r20              = _mm256_mul_pd(rsq20,rinv20);
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq20             = _mm256_mul_pd(iq2,jq0);
 344
 345             /* Calculate table index by multiplying r with table scale and truncate to integer */
 346             rt               = _mm256_mul_pd(r20,vftabscale);
 347             vfitab           = _mm256_cvttpd_epi32(rt);
 348             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 349             vfitab           = _mm_slli_epi32(vfitab,2);
 350
 351             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 352             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 353             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 354             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 355             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 356             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 357             Heps             = _mm256_mul_pd(vfeps,H);
 358             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 359             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 360             velec            = _mm256_mul_pd(qq20,VV);
 361             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 362             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velecsum         = _mm256_add_pd(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             /* Calculate temporary vectorial force */
 370             tx               = _mm256_mul_pd(fscal,dx20);
 371             ty               = _mm256_mul_pd(fscal,dy20);
 372             tz               = _mm256_mul_pd(fscal,dz20);
 373
 374             /* Update vectorial force */
 375             fix2             = _mm256_add_pd(fix2,tx);
 376             fiy2             = _mm256_add_pd(fiy2,ty);
 377             fiz2             = _mm256_add_pd(fiz2,tz);
 378
 379             fjx0             = _mm256_add_pd(fjx0,tx);
 380             fjy0             = _mm256_add_pd(fjy0,ty);
 381             fjz0             = _mm256_add_pd(fjz0,tz);
 382
 383             /**************************
 384              * CALCULATE INTERACTIONS *
 385              **************************/
 386
 387             r30              = _mm256_mul_pd(rsq30,rinv30);
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq30             = _mm256_mul_pd(iq3,jq0);
 391
 392             /* Calculate table index by multiplying r with table scale and truncate to integer */
 393             rt               = _mm256_mul_pd(r30,vftabscale);
 394             vfitab           = _mm256_cvttpd_epi32(rt);
 395             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 396             vfitab           = _mm_slli_epi32(vfitab,2);
 397
 398             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 399             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 400             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 401             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 402             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 403             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 404             Heps             = _mm256_mul_pd(vfeps,H);
 405             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 406             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 407             velec            = _mm256_mul_pd(qq30,VV);
 408             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 409             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velecsum         = _mm256_add_pd(velecsum,velec);
 413
 414             fscal            = felec;
 415
 416             /* Calculate temporary vectorial force */
 417             tx               = _mm256_mul_pd(fscal,dx30);
 418             ty               = _mm256_mul_pd(fscal,dy30);
 419             tz               = _mm256_mul_pd(fscal,dz30);
 420
 421             /* Update vectorial force */
 422             fix3             = _mm256_add_pd(fix3,tx);
 423             fiy3             = _mm256_add_pd(fiy3,ty);
 424             fiz3             = _mm256_add_pd(fiz3,tz);
 425
 426             fjx0             = _mm256_add_pd(fjx0,tx);
 427             fjy0             = _mm256_add_pd(fjy0,ty);
 428             fjz0             = _mm256_add_pd(fjz0,tz);
 429
 430             fjptrA             = f+j_coord_offsetA;
 431             fjptrB             = f+j_coord_offsetB;
 432             fjptrC             = f+j_coord_offsetC;
 433             fjptrD             = f+j_coord_offsetD;
 434
 435             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 436
 437             /* Inner loop uses 164 flops */
 438         }
 439
 440         if(jidx<j_index_end)
 441         {
 442
 443             /* Get j neighbor index, and coordinate index */
 444             jnrlistA         = jjnr[jidx];
 445             jnrlistB         = jjnr[jidx+1];
 446             jnrlistC         = jjnr[jidx+2];
 447             jnrlistD         = jjnr[jidx+3];
 448             /* Sign of each element will be negative for non-real atoms.
 449              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 450              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 451              */
 452             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 453
 454             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 455             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 456             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 457
 458             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 459             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 460             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 461             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 462             j_coord_offsetA  = DIM*jnrA;
 463             j_coord_offsetB  = DIM*jnrB;
 464             j_coord_offsetC  = DIM*jnrC;
 465             j_coord_offsetD  = DIM*jnrD;
 466
 467             /* load j atom coordinates */
 468             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 469                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 470                                                  &jx0,&jy0,&jz0);
 471
 472             /* Calculate displacement vector */
 473             dx00             = _mm256_sub_pd(ix0,jx0);
 474             dy00             = _mm256_sub_pd(iy0,jy0);
 475             dz00             = _mm256_sub_pd(iz0,jz0);
 476             dx10             = _mm256_sub_pd(ix1,jx0);
 477             dy10             = _mm256_sub_pd(iy1,jy0);
 478             dz10             = _mm256_sub_pd(iz1,jz0);
 479             dx20             = _mm256_sub_pd(ix2,jx0);
 480             dy20             = _mm256_sub_pd(iy2,jy0);
 481             dz20             = _mm256_sub_pd(iz2,jz0);
 482             dx30             = _mm256_sub_pd(ix3,jx0);
 483             dy30             = _mm256_sub_pd(iy3,jy0);
 484             dz30             = _mm256_sub_pd(iz3,jz0);
 485
 486             /* Calculate squared distance and things based on it */
 487             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 488             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 489             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 490             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 491
 492             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 493             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 494             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 495
 496             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 497
 498             /* Load parameters for j particles */
 499             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 500                                                                  charge+jnrC+0,charge+jnrD+0);
 501             vdwjidx0A        = 2*vdwtype[jnrA+0];
 502             vdwjidx0B        = 2*vdwtype[jnrB+0];
 503             vdwjidx0C        = 2*vdwtype[jnrC+0];
 504             vdwjidx0D        = 2*vdwtype[jnrD+0];
 505
 506             fjx0             = _mm256_setzero_pd();
 507             fjy0             = _mm256_setzero_pd();
 508             fjz0             = _mm256_setzero_pd();
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 516                                             vdwioffsetptr0+vdwjidx0B,
 517                                             vdwioffsetptr0+vdwjidx0C,
 518                                             vdwioffsetptr0+vdwjidx0D,
 519                                             &c6_00,&c12_00);
 520
 521             /* LENNARD-JONES DISPERSION/REPULSION */
 522
 523             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 524             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 525             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 526             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 527             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 531             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 532
 533             fscal            = fvdw;
 534
 535             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm256_mul_pd(fscal,dx00);
 539             ty               = _mm256_mul_pd(fscal,dy00);
 540             tz               = _mm256_mul_pd(fscal,dz00);
 541
 542             /* Update vectorial force */
 543             fix0             = _mm256_add_pd(fix0,tx);
 544             fiy0             = _mm256_add_pd(fiy0,ty);
 545             fiz0             = _mm256_add_pd(fiz0,tz);
 546
 547             fjx0             = _mm256_add_pd(fjx0,tx);
 548             fjy0             = _mm256_add_pd(fjy0,ty);
 549             fjz0             = _mm256_add_pd(fjz0,tz);
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             r10              = _mm256_mul_pd(rsq10,rinv10);
 556             r10              = _mm256_andnot_pd(dummy_mask,r10);
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             qq10             = _mm256_mul_pd(iq1,jq0);
 560
 561             /* Calculate table index by multiplying r with table scale and truncate to integer */
 562             rt               = _mm256_mul_pd(r10,vftabscale);
 563             vfitab           = _mm256_cvttpd_epi32(rt);
 564             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 565             vfitab           = _mm_slli_epi32(vfitab,2);
 566
 567             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 568             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 569             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 570             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 571             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 572             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 573             Heps             = _mm256_mul_pd(vfeps,H);
 574             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 575             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 576             velec            = _mm256_mul_pd(qq10,VV);
 577             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 578             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 579
 580             /* Update potential sum for this i atom from the interaction with this j atom. */
 581             velec            = _mm256_andnot_pd(dummy_mask,velec);
 582             velecsum         = _mm256_add_pd(velecsum,velec);
 583
 584             fscal            = felec;
 585
 586             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 587
 588             /* Calculate temporary vectorial force */
 589             tx               = _mm256_mul_pd(fscal,dx10);
 590             ty               = _mm256_mul_pd(fscal,dy10);
 591             tz               = _mm256_mul_pd(fscal,dz10);
 592
 593             /* Update vectorial force */
 594             fix1             = _mm256_add_pd(fix1,tx);
 595             fiy1             = _mm256_add_pd(fiy1,ty);
 596             fiz1             = _mm256_add_pd(fiz1,tz);
 597
 598             fjx0             = _mm256_add_pd(fjx0,tx);
 599             fjy0             = _mm256_add_pd(fjy0,ty);
 600             fjz0             = _mm256_add_pd(fjz0,tz);
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             r20              = _mm256_mul_pd(rsq20,rinv20);
 607             r20              = _mm256_andnot_pd(dummy_mask,r20);
 608
 609             /* Compute parameters for interactions between i and j atoms */
 610             qq20             = _mm256_mul_pd(iq2,jq0);
 611
 612             /* Calculate table index by multiplying r with table scale and truncate to integer */
 613             rt               = _mm256_mul_pd(r20,vftabscale);
 614             vfitab           = _mm256_cvttpd_epi32(rt);
 615             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 616             vfitab           = _mm_slli_epi32(vfitab,2);
 617
 618             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 619             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 620             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 621             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 622             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 623             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 624             Heps             = _mm256_mul_pd(vfeps,H);
 625             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 626             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 627             velec            = _mm256_mul_pd(qq20,VV);
 628             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 629             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 630
 631             /* Update potential sum for this i atom from the interaction with this j atom. */
 632             velec            = _mm256_andnot_pd(dummy_mask,velec);
 633             velecsum         = _mm256_add_pd(velecsum,velec);
 634
 635             fscal            = felec;
 636
 637             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 638
 639             /* Calculate temporary vectorial force */
 640             tx               = _mm256_mul_pd(fscal,dx20);
 641             ty               = _mm256_mul_pd(fscal,dy20);
 642             tz               = _mm256_mul_pd(fscal,dz20);
 643
 644             /* Update vectorial force */
 645             fix2             = _mm256_add_pd(fix2,tx);
 646             fiy2             = _mm256_add_pd(fiy2,ty);
 647             fiz2             = _mm256_add_pd(fiz2,tz);
 648
 649             fjx0             = _mm256_add_pd(fjx0,tx);
 650             fjy0             = _mm256_add_pd(fjy0,ty);
 651             fjz0             = _mm256_add_pd(fjz0,tz);
 652
 653             /**************************
 654              * CALCULATE INTERACTIONS *
 655              **************************/
 656
 657             r30              = _mm256_mul_pd(rsq30,rinv30);
 658             r30              = _mm256_andnot_pd(dummy_mask,r30);
 659
 660             /* Compute parameters for interactions between i and j atoms */
 661             qq30             = _mm256_mul_pd(iq3,jq0);
 662
 663             /* Calculate table index by multiplying r with table scale and truncate to integer */
 664             rt               = _mm256_mul_pd(r30,vftabscale);
 665             vfitab           = _mm256_cvttpd_epi32(rt);
 666             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 667             vfitab           = _mm_slli_epi32(vfitab,2);
 668
 669             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 670             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 671             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 672             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 673             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 674             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 675             Heps             = _mm256_mul_pd(vfeps,H);
 676             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 677             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 678             velec            = _mm256_mul_pd(qq30,VV);
 679             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 680             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 681
 682             /* Update potential sum for this i atom from the interaction with this j atom. */
 683             velec            = _mm256_andnot_pd(dummy_mask,velec);
 684             velecsum         = _mm256_add_pd(velecsum,velec);
 685
 686             fscal            = felec;
 687
 688             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 689
 690             /* Calculate temporary vectorial force */
 691             tx               = _mm256_mul_pd(fscal,dx30);
 692             ty               = _mm256_mul_pd(fscal,dy30);
 693             tz               = _mm256_mul_pd(fscal,dz30);
 694
 695             /* Update vectorial force */
 696             fix3             = _mm256_add_pd(fix3,tx);
 697             fiy3             = _mm256_add_pd(fiy3,ty);
 698             fiz3             = _mm256_add_pd(fiz3,tz);
 699
 700             fjx0             = _mm256_add_pd(fjx0,tx);
 701             fjy0             = _mm256_add_pd(fjy0,ty);
 702             fjz0             = _mm256_add_pd(fjz0,tz);
 703
 704             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 705             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 706             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 707             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 708
 709             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 710
 711             /* Inner loop uses 167 flops */
 712         }
 713
 714         /* End of innermost loop */
 715
 716         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 717                                                  f+i_coord_offset,fshift+i_shift_offset);
 718
 719         ggid                        = gid[iidx];
 720         /* Update potential energies */
 721         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 722         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 723
 724         /* Increment number of inner iterations */
 725         inneriter                  += j_index_end - j_index_start;
 726
 727         /* Outer loop uses 26 flops */
 728     }
 729
 730     /* Increment number of outer iterations */
 731     outeriter        += nri;
 732
 733     /* Update outer/inner flops */
 734
 735     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 736 }
 737 /*
 738  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_256_double
 739  * Electrostatics interaction: CubicSplineTable
 740  * VdW interaction:            LennardJones
 741  * Geometry:                   Water4-Particle
 742  * Calculate force/pot:        Force
 743  */
 744 void
 745 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_256_double
 746                     (t_nblist                    * gmx_restrict       nlist,
 747                      rvec                        * gmx_restrict          xx,
 748                      rvec                        * gmx_restrict          ff,
 749                      t_forcerec                  * gmx_restrict          fr,
 750                      t_mdatoms                   * gmx_restrict     mdatoms,
 751                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 752                      t_nrnb                      * gmx_restrict        nrnb)
 753 {
 754     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 755      * just 0 for non-waters.
 756      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 757      * jnr indices corresponding to data put in the four positions in the SIMD register.
 758      */
 759     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 760     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 761     int              jnrA,jnrB,jnrC,jnrD;
 762     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 763     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 764     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 765     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 766     real             rcutoff_scalar;
 767     real             *shiftvec,*fshift,*x,*f;
 768     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 769     real             scratch[4*DIM];
 770     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 771     real *           vdwioffsetptr0;
 772     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 773     real *           vdwioffsetptr1;
 774     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 775     real *           vdwioffsetptr2;
 776     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 777     real *           vdwioffsetptr3;
 778     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 779     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 780     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 781     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 782     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 783     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 784     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 785     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 786     real             *charge;
 787     int              nvdwtype;
 788     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 789     int              *vdwtype;
 790     real             *vdwparam;
 791     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 792     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 793     __m128i          vfitab;
 794     __m128i          ifour       = _mm_set1_epi32(4);
 795     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 796     real             *vftab;
 797     __m256d          dummy_mask,cutoff_mask;
 798     __m128           tmpmask0,tmpmask1;
 799     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 800     __m256d          one     = _mm256_set1_pd(1.0);
 801     __m256d          two     = _mm256_set1_pd(2.0);
 802     x                = xx[0];
 803     f                = ff[0];
 804
 805     nri              = nlist->nri;
 806     iinr             = nlist->iinr;
 807     jindex           = nlist->jindex;
 808     jjnr             = nlist->jjnr;
 809     shiftidx         = nlist->shift;
 810     gid              = nlist->gid;
 811     shiftvec         = fr->shift_vec[0];
 812     fshift           = fr->fshift[0];
 813     facel            = _mm256_set1_pd(fr->epsfac);
 814     charge           = mdatoms->chargeA;
 815     nvdwtype         = fr->ntype;
 816     vdwparam         = fr->nbfp;
 817     vdwtype          = mdatoms->typeA;
 818
 819     vftab            = kernel_data->table_elec->data;
 820     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 821
 822     /* Setup water-specific parameters */
 823     inr              = nlist->iinr[0];
 824     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 825     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 826     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 827     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 828
 829     /* Avoid stupid compiler warnings */
 830     jnrA = jnrB = jnrC = jnrD = 0;
 831     j_coord_offsetA = 0;
 832     j_coord_offsetB = 0;
 833     j_coord_offsetC = 0;
 834     j_coord_offsetD = 0;
 835
 836     outeriter        = 0;
 837     inneriter        = 0;
 838
 839     for(iidx=0;iidx<4*DIM;iidx++)
 840     {
 841         scratch[iidx] = 0.0;
 842     }
 843
 844     /* Start outer loop over neighborlists */
 845     for(iidx=0; iidx<nri; iidx++)
 846     {
 847         /* Load shift vector for this list */
 848         i_shift_offset   = DIM*shiftidx[iidx];
 849
 850         /* Load limits for loop over neighbors */
 851         j_index_start    = jindex[iidx];
 852         j_index_end      = jindex[iidx+1];
 853
 854         /* Get outer coordinate index */
 855         inr              = iinr[iidx];
 856         i_coord_offset   = DIM*inr;
 857
 858         /* Load i particle coords and add shift vector */
 859         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 860                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 861
 862         fix0             = _mm256_setzero_pd();
 863         fiy0             = _mm256_setzero_pd();
 864         fiz0             = _mm256_setzero_pd();
 865         fix1             = _mm256_setzero_pd();
 866         fiy1             = _mm256_setzero_pd();
 867         fiz1             = _mm256_setzero_pd();
 868         fix2             = _mm256_setzero_pd();
 869         fiy2             = _mm256_setzero_pd();
 870         fiz2             = _mm256_setzero_pd();
 871         fix3             = _mm256_setzero_pd();
 872         fiy3             = _mm256_setzero_pd();
 873         fiz3             = _mm256_setzero_pd();
 874
 875         /* Start inner kernel loop */
 876         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 877         {
 878
 879             /* Get j neighbor index, and coordinate index */
 880             jnrA             = jjnr[jidx];
 881             jnrB             = jjnr[jidx+1];
 882             jnrC             = jjnr[jidx+2];
 883             jnrD             = jjnr[jidx+3];
 884             j_coord_offsetA  = DIM*jnrA;
 885             j_coord_offsetB  = DIM*jnrB;
 886             j_coord_offsetC  = DIM*jnrC;
 887             j_coord_offsetD  = DIM*jnrD;
 888
 889             /* load j atom coordinates */
 890             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 891                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 892                                                  &jx0,&jy0,&jz0);
 893
 894             /* Calculate displacement vector */
 895             dx00             = _mm256_sub_pd(ix0,jx0);
 896             dy00             = _mm256_sub_pd(iy0,jy0);
 897             dz00             = _mm256_sub_pd(iz0,jz0);
 898             dx10             = _mm256_sub_pd(ix1,jx0);
 899             dy10             = _mm256_sub_pd(iy1,jy0);
 900             dz10             = _mm256_sub_pd(iz1,jz0);
 901             dx20             = _mm256_sub_pd(ix2,jx0);
 902             dy20             = _mm256_sub_pd(iy2,jy0);
 903             dz20             = _mm256_sub_pd(iz2,jz0);
 904             dx30             = _mm256_sub_pd(ix3,jx0);
 905             dy30             = _mm256_sub_pd(iy3,jy0);
 906             dz30             = _mm256_sub_pd(iz3,jz0);
 907
 908             /* Calculate squared distance and things based on it */
 909             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 910             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 911             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 912             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 913
 914             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 915             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 916             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 917
 918             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 919
 920             /* Load parameters for j particles */
 921             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 922                                                                  charge+jnrC+0,charge+jnrD+0);
 923             vdwjidx0A        = 2*vdwtype[jnrA+0];
 924             vdwjidx0B        = 2*vdwtype[jnrB+0];
 925             vdwjidx0C        = 2*vdwtype[jnrC+0];
 926             vdwjidx0D        = 2*vdwtype[jnrD+0];
 927
 928             fjx0             = _mm256_setzero_pd();
 929             fjy0             = _mm256_setzero_pd();
 930             fjz0             = _mm256_setzero_pd();
 931
 932             /**************************
 933              * CALCULATE INTERACTIONS *
 934              **************************/
 935
 936             /* Compute parameters for interactions between i and j atoms */
 937             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 938                                             vdwioffsetptr0+vdwjidx0B,
 939                                             vdwioffsetptr0+vdwjidx0C,
 940                                             vdwioffsetptr0+vdwjidx0D,
 941                                             &c6_00,&c12_00);
 942
 943             /* LENNARD-JONES DISPERSION/REPULSION */
 944
 945             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 946             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 947
 948             fscal            = fvdw;
 949
 950             /* Calculate temporary vectorial force */
 951             tx               = _mm256_mul_pd(fscal,dx00);
 952             ty               = _mm256_mul_pd(fscal,dy00);
 953             tz               = _mm256_mul_pd(fscal,dz00);
 954
 955             /* Update vectorial force */
 956             fix0             = _mm256_add_pd(fix0,tx);
 957             fiy0             = _mm256_add_pd(fiy0,ty);
 958             fiz0             = _mm256_add_pd(fiz0,tz);
 959
 960             fjx0             = _mm256_add_pd(fjx0,tx);
 961             fjy0             = _mm256_add_pd(fjy0,ty);
 962             fjz0             = _mm256_add_pd(fjz0,tz);
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             r10              = _mm256_mul_pd(rsq10,rinv10);
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq10             = _mm256_mul_pd(iq1,jq0);
 972
 973             /* Calculate table index by multiplying r with table scale and truncate to integer */
 974             rt               = _mm256_mul_pd(r10,vftabscale);
 975             vfitab           = _mm256_cvttpd_epi32(rt);
 976             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 977             vfitab           = _mm_slli_epi32(vfitab,2);
 978
 979             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 980             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 981             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 982             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 983             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 984             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 985             Heps             = _mm256_mul_pd(vfeps,H);
 986             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 987             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 988             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 989
 990             fscal            = felec;
 991
 992             /* Calculate temporary vectorial force */
 993             tx               = _mm256_mul_pd(fscal,dx10);
 994             ty               = _mm256_mul_pd(fscal,dy10);
 995             tz               = _mm256_mul_pd(fscal,dz10);
 996
 997             /* Update vectorial force */
 998             fix1             = _mm256_add_pd(fix1,tx);
 999             fiy1             = _mm256_add_pd(fiy1,ty);
1000             fiz1             = _mm256_add_pd(fiz1,tz);
1001
1002             fjx0             = _mm256_add_pd(fjx0,tx);
1003             fjy0             = _mm256_add_pd(fjy0,ty);
1004             fjz0             = _mm256_add_pd(fjz0,tz);
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             r20              = _mm256_mul_pd(rsq20,rinv20);
1011
1012             /* Compute parameters for interactions between i and j atoms */
1013             qq20             = _mm256_mul_pd(iq2,jq0);
1014
1015             /* Calculate table index by multiplying r with table scale and truncate to integer */
1016             rt               = _mm256_mul_pd(r20,vftabscale);
1017             vfitab           = _mm256_cvttpd_epi32(rt);
1018             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1019             vfitab           = _mm_slli_epi32(vfitab,2);
1020
1021             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1022             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1023             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1024             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1025             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1026             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1027             Heps             = _mm256_mul_pd(vfeps,H);
1028             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1029             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1030             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1031
1032             fscal            = felec;
1033
1034             /* Calculate temporary vectorial force */
1035             tx               = _mm256_mul_pd(fscal,dx20);
1036             ty               = _mm256_mul_pd(fscal,dy20);
1037             tz               = _mm256_mul_pd(fscal,dz20);
1038
1039             /* Update vectorial force */
1040             fix2             = _mm256_add_pd(fix2,tx);
1041             fiy2             = _mm256_add_pd(fiy2,ty);
1042             fiz2             = _mm256_add_pd(fiz2,tz);
1043
1044             fjx0             = _mm256_add_pd(fjx0,tx);
1045             fjy0             = _mm256_add_pd(fjy0,ty);
1046             fjz0             = _mm256_add_pd(fjz0,tz);
1047
1048             /**************************
1049              * CALCULATE INTERACTIONS *
1050              **************************/
1051
1052             r30              = _mm256_mul_pd(rsq30,rinv30);
1053
1054             /* Compute parameters for interactions between i and j atoms */
1055             qq30             = _mm256_mul_pd(iq3,jq0);
1056
1057             /* Calculate table index by multiplying r with table scale and truncate to integer */
1058             rt               = _mm256_mul_pd(r30,vftabscale);
1059             vfitab           = _mm256_cvttpd_epi32(rt);
1060             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1061             vfitab           = _mm_slli_epi32(vfitab,2);
1062
1063             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1064             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1065             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1066             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1067             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1068             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1069             Heps             = _mm256_mul_pd(vfeps,H);
1070             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1071             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1072             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
1073
1074             fscal            = felec;
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm256_mul_pd(fscal,dx30);
1078             ty               = _mm256_mul_pd(fscal,dy30);
1079             tz               = _mm256_mul_pd(fscal,dz30);
1080
1081             /* Update vectorial force */
1082             fix3             = _mm256_add_pd(fix3,tx);
1083             fiy3             = _mm256_add_pd(fiy3,ty);
1084             fiz3             = _mm256_add_pd(fiz3,tz);
1085
1086             fjx0             = _mm256_add_pd(fjx0,tx);
1087             fjy0             = _mm256_add_pd(fjy0,ty);
1088             fjz0             = _mm256_add_pd(fjz0,tz);
1089
1090             fjptrA             = f+j_coord_offsetA;
1091             fjptrB             = f+j_coord_offsetB;
1092             fjptrC             = f+j_coord_offsetC;
1093             fjptrD             = f+j_coord_offsetD;
1094
1095             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1096
1097             /* Inner loop uses 147 flops */
1098         }
1099
1100         if(jidx<j_index_end)
1101         {
1102
1103             /* Get j neighbor index, and coordinate index */
1104             jnrlistA         = jjnr[jidx];
1105             jnrlistB         = jjnr[jidx+1];
1106             jnrlistC         = jjnr[jidx+2];
1107             jnrlistD         = jjnr[jidx+3];
1108             /* Sign of each element will be negative for non-real atoms.
1109              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1110              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1111              */
1112             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1113
1114             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1115             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1116             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1117
1118             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1119             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1120             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1121             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1122             j_coord_offsetA  = DIM*jnrA;
1123             j_coord_offsetB  = DIM*jnrB;
1124             j_coord_offsetC  = DIM*jnrC;
1125             j_coord_offsetD  = DIM*jnrD;
1126
1127             /* load j atom coordinates */
1128             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1129                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1130                                                  &jx0,&jy0,&jz0);
1131
1132             /* Calculate displacement vector */
1133             dx00             = _mm256_sub_pd(ix0,jx0);
1134             dy00             = _mm256_sub_pd(iy0,jy0);
1135             dz00             = _mm256_sub_pd(iz0,jz0);
1136             dx10             = _mm256_sub_pd(ix1,jx0);
1137             dy10             = _mm256_sub_pd(iy1,jy0);
1138             dz10             = _mm256_sub_pd(iz1,jz0);
1139             dx20             = _mm256_sub_pd(ix2,jx0);
1140             dy20             = _mm256_sub_pd(iy2,jy0);
1141             dz20             = _mm256_sub_pd(iz2,jz0);
1142             dx30             = _mm256_sub_pd(ix3,jx0);
1143             dy30             = _mm256_sub_pd(iy3,jy0);
1144             dz30             = _mm256_sub_pd(iz3,jz0);
1145
1146             /* Calculate squared distance and things based on it */
1147             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1148             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1149             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1150             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1151
1152             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1153             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1154             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1155
1156             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1157
1158             /* Load parameters for j particles */
1159             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1160                                                                  charge+jnrC+0,charge+jnrD+0);
1161             vdwjidx0A        = 2*vdwtype[jnrA+0];
1162             vdwjidx0B        = 2*vdwtype[jnrB+0];
1163             vdwjidx0C        = 2*vdwtype[jnrC+0];
1164             vdwjidx0D        = 2*vdwtype[jnrD+0];
1165
1166             fjx0             = _mm256_setzero_pd();
1167             fjy0             = _mm256_setzero_pd();
1168             fjz0             = _mm256_setzero_pd();
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             /* Compute parameters for interactions between i and j atoms */
1175             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1176                                             vdwioffsetptr0+vdwjidx0B,
1177                                             vdwioffsetptr0+vdwjidx0C,
1178                                             vdwioffsetptr0+vdwjidx0D,
1179                                             &c6_00,&c12_00);
1180
1181             /* LENNARD-JONES DISPERSION/REPULSION */
1182
1183             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1184             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1185
1186             fscal            = fvdw;
1187
1188             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1189
1190             /* Calculate temporary vectorial force */
1191             tx               = _mm256_mul_pd(fscal,dx00);
1192             ty               = _mm256_mul_pd(fscal,dy00);
1193             tz               = _mm256_mul_pd(fscal,dz00);
1194
1195             /* Update vectorial force */
1196             fix0             = _mm256_add_pd(fix0,tx);
1197             fiy0             = _mm256_add_pd(fiy0,ty);
1198             fiz0             = _mm256_add_pd(fiz0,tz);
1199
1200             fjx0             = _mm256_add_pd(fjx0,tx);
1201             fjy0             = _mm256_add_pd(fjy0,ty);
1202             fjz0             = _mm256_add_pd(fjz0,tz);
1203
1204             /**************************
1205              * CALCULATE INTERACTIONS *
1206              **************************/
1207
1208             r10              = _mm256_mul_pd(rsq10,rinv10);
1209             r10              = _mm256_andnot_pd(dummy_mask,r10);
1210
1211             /* Compute parameters for interactions between i and j atoms */
1212             qq10             = _mm256_mul_pd(iq1,jq0);
1213
1214             /* Calculate table index by multiplying r with table scale and truncate to integer */
1215             rt               = _mm256_mul_pd(r10,vftabscale);
1216             vfitab           = _mm256_cvttpd_epi32(rt);
1217             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1218             vfitab           = _mm_slli_epi32(vfitab,2);
1219
1220             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1221             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1222             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1223             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1224             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1225             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1226             Heps             = _mm256_mul_pd(vfeps,H);
1227             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1228             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1229             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1230
1231             fscal            = felec;
1232
1233             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1234
1235             /* Calculate temporary vectorial force */
1236             tx               = _mm256_mul_pd(fscal,dx10);
1237             ty               = _mm256_mul_pd(fscal,dy10);
1238             tz               = _mm256_mul_pd(fscal,dz10);
1239
1240             /* Update vectorial force */
1241             fix1             = _mm256_add_pd(fix1,tx);
1242             fiy1             = _mm256_add_pd(fiy1,ty);
1243             fiz1             = _mm256_add_pd(fiz1,tz);
1244
1245             fjx0             = _mm256_add_pd(fjx0,tx);
1246             fjy0             = _mm256_add_pd(fjy0,ty);
1247             fjz0             = _mm256_add_pd(fjz0,tz);
1248
1249             /**************************
1250              * CALCULATE INTERACTIONS *
1251              **************************/
1252
1253             r20              = _mm256_mul_pd(rsq20,rinv20);
1254             r20              = _mm256_andnot_pd(dummy_mask,r20);
1255
1256             /* Compute parameters for interactions between i and j atoms */
1257             qq20             = _mm256_mul_pd(iq2,jq0);
1258
1259             /* Calculate table index by multiplying r with table scale and truncate to integer */
1260             rt               = _mm256_mul_pd(r20,vftabscale);
1261             vfitab           = _mm256_cvttpd_epi32(rt);
1262             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1263             vfitab           = _mm_slli_epi32(vfitab,2);
1264
1265             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1266             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1267             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1268             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1269             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1270             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1271             Heps             = _mm256_mul_pd(vfeps,H);
1272             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1273             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1274             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1275
1276             fscal            = felec;
1277
1278             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1279
1280             /* Calculate temporary vectorial force */
1281             tx               = _mm256_mul_pd(fscal,dx20);
1282             ty               = _mm256_mul_pd(fscal,dy20);
1283             tz               = _mm256_mul_pd(fscal,dz20);
1284
1285             /* Update vectorial force */
1286             fix2             = _mm256_add_pd(fix2,tx);
1287             fiy2             = _mm256_add_pd(fiy2,ty);
1288             fiz2             = _mm256_add_pd(fiz2,tz);
1289
1290             fjx0             = _mm256_add_pd(fjx0,tx);
1291             fjy0             = _mm256_add_pd(fjy0,ty);
1292             fjz0             = _mm256_add_pd(fjz0,tz);
1293
1294             /**************************
1295              * CALCULATE INTERACTIONS *
1296              **************************/
1297
1298             r30              = _mm256_mul_pd(rsq30,rinv30);
1299             r30              = _mm256_andnot_pd(dummy_mask,r30);
1300
1301             /* Compute parameters for interactions between i and j atoms */
1302             qq30             = _mm256_mul_pd(iq3,jq0);
1303
1304             /* Calculate table index by multiplying r with table scale and truncate to integer */
1305             rt               = _mm256_mul_pd(r30,vftabscale);
1306             vfitab           = _mm256_cvttpd_epi32(rt);
1307             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1308             vfitab           = _mm_slli_epi32(vfitab,2);
1309
1310             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1311             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1312             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1313             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1314             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1315             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1316             Heps             = _mm256_mul_pd(vfeps,H);
1317             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1318             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1319             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
1320
1321             fscal            = felec;
1322
1323             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1324
1325             /* Calculate temporary vectorial force */
1326             tx               = _mm256_mul_pd(fscal,dx30);
1327             ty               = _mm256_mul_pd(fscal,dy30);
1328             tz               = _mm256_mul_pd(fscal,dz30);
1329
1330             /* Update vectorial force */
1331             fix3             = _mm256_add_pd(fix3,tx);
1332             fiy3             = _mm256_add_pd(fiy3,ty);
1333             fiz3             = _mm256_add_pd(fiz3,tz);
1334
1335             fjx0             = _mm256_add_pd(fjx0,tx);
1336             fjy0             = _mm256_add_pd(fjy0,ty);
1337             fjz0             = _mm256_add_pd(fjz0,tz);
1338
1339             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1340             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1341             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1342             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1343
1344             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1345
1346             /* Inner loop uses 150 flops */
1347         }
1348
1349         /* End of innermost loop */
1350
1351         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1352                                                  f+i_coord_offset,fshift+i_shift_offset);
1353
1354         /* Increment number of inner iterations */
1355         inneriter                  += j_index_end - j_index_start;
1356
1357         /* Outer loop uses 24 flops */
1358     }
1359
1360     /* Increment number of outer iterations */
1361     outeriter        += nri;
1362
1363     /* Update outer/inner flops */
1364
1365     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1366 }