src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_GeomW3P1_VF_avx_256_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_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     real *           vdwioffsetptr1;
  86     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 101     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m128i          ewitab;
 107     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 109     real             *ewtab;
 110     __m256d          dummy_mask,cutoff_mask;
 111     __m128           tmpmask0,tmpmask1;
 112     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 113     __m256d          one     = _mm256_set1_pd(1.0);
 114     __m256d          two     = _mm256_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_pd(fr->ic->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 134
 135     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 136     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 137     beta2            = _mm256_mul_pd(beta,beta);
 138     beta3            = _mm256_mul_pd(beta,beta2);
 139
 140     ewtab            = fr->ic->tabq_coul_FDV0;
 141     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 142     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 143
 144     /* Setup water-specific parameters */
 145     inr              = nlist->iinr[0];
 146     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 147     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 148     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 149     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 150
 151     /* Avoid stupid compiler warnings */
 152     jnrA = jnrB = jnrC = jnrD = 0;
 153     j_coord_offsetA = 0;
 154     j_coord_offsetB = 0;
 155     j_coord_offsetC = 0;
 156     j_coord_offsetD = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     for(iidx=0;iidx<4*DIM;iidx++)
 162     {
 163         scratch[iidx] = 0.0;
 164     }
 165
 166     /* Start outer loop over neighborlists */
 167     for(iidx=0; iidx<nri; iidx++)
 168     {
 169         /* Load shift vector for this list */
 170         i_shift_offset   = DIM*shiftidx[iidx];
 171
 172         /* Load limits for loop over neighbors */
 173         j_index_start    = jindex[iidx];
 174         j_index_end      = jindex[iidx+1];
 175
 176         /* Get outer coordinate index */
 177         inr              = iinr[iidx];
 178         i_coord_offset   = DIM*inr;
 179
 180         /* Load i particle coords and add shift vector */
 181         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 182                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 183
 184         fix0             = _mm256_setzero_pd();
 185         fiy0             = _mm256_setzero_pd();
 186         fiz0             = _mm256_setzero_pd();
 187         fix1             = _mm256_setzero_pd();
 188         fiy1             = _mm256_setzero_pd();
 189         fiz1             = _mm256_setzero_pd();
 190         fix2             = _mm256_setzero_pd();
 191         fiy2             = _mm256_setzero_pd();
 192         fiz2             = _mm256_setzero_pd();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_pd();
 196         vvdwsum          = _mm256_setzero_pd();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                                  &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm256_sub_pd(ix0,jx0);
 219             dy00             = _mm256_sub_pd(iy0,jy0);
 220             dz00             = _mm256_sub_pd(iz0,jz0);
 221             dx10             = _mm256_sub_pd(ix1,jx0);
 222             dy10             = _mm256_sub_pd(iy1,jy0);
 223             dz10             = _mm256_sub_pd(iz1,jz0);
 224             dx20             = _mm256_sub_pd(ix2,jx0);
 225             dy20             = _mm256_sub_pd(iy2,jy0);
 226             dz20             = _mm256_sub_pd(iz2,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 230             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 231             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 232
 233             rinv00           = avx256_invsqrt_d(rsq00);
 234             rinv10           = avx256_invsqrt_d(rsq10);
 235             rinv20           = avx256_invsqrt_d(rsq20);
 236
 237             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 238             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 239             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 240
 241             /* Load parameters for j particles */
 242             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 243                                                                  charge+jnrC+0,charge+jnrD+0);
 244             vdwjidx0A        = 2*vdwtype[jnrA+0];
 245             vdwjidx0B        = 2*vdwtype[jnrB+0];
 246             vdwjidx0C        = 2*vdwtype[jnrC+0];
 247             vdwjidx0D        = 2*vdwtype[jnrD+0];
 248
 249             fjx0             = _mm256_setzero_pd();
 250             fjy0             = _mm256_setzero_pd();
 251             fjz0             = _mm256_setzero_pd();
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             r00              = _mm256_mul_pd(rsq00,rinv00);
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq00             = _mm256_mul_pd(iq0,jq0);
 261             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 262                                             vdwioffsetptr0+vdwjidx0B,
 263                                             vdwioffsetptr0+vdwjidx0C,
 264                                             vdwioffsetptr0+vdwjidx0D,
 265                                             &c6_00,&c12_00);
 266
 267             /* Calculate table index by multiplying r with table scale and truncate to integer */
 268             rt               = _mm256_mul_pd(r00,vftabscale);
 269             vfitab           = _mm256_cvttpd_epi32(rt);
 270             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 271             vfitab           = _mm_slli_epi32(vfitab,3);
 272
 273             /* EWALD ELECTROSTATICS */
 274
 275             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 276             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 277             ewitab           = _mm256_cvttpd_epi32(ewrt);
 278             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 279             ewitab           = _mm_slli_epi32(ewitab,2);
 280             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 281             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 282             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 283             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 284             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 285             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 286             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 287             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 288             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 289
 290             /* CUBIC SPLINE TABLE DISPERSION */
 291             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 292             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 293             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 294             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 295             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 296             Heps             = _mm256_mul_pd(vfeps,H);
 297             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 298             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 299             vvdw6            = _mm256_mul_pd(c6_00,VV);
 300             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 301             fvdw6            = _mm256_mul_pd(c6_00,FF);
 302
 303             /* CUBIC SPLINE TABLE REPULSION */
 304             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm256_mul_pd(c12_00,VV);
 314             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 315             fvdw12           = _mm256_mul_pd(c12_00,FF);
 316             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 317             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velecsum         = _mm256_add_pd(velecsum,velec);
 321             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 322
 323             fscal            = _mm256_add_pd(felec,fvdw);
 324
 325             /* Calculate temporary vectorial force */
 326             tx               = _mm256_mul_pd(fscal,dx00);
 327             ty               = _mm256_mul_pd(fscal,dy00);
 328             tz               = _mm256_mul_pd(fscal,dz00);
 329
 330             /* Update vectorial force */
 331             fix0             = _mm256_add_pd(fix0,tx);
 332             fiy0             = _mm256_add_pd(fiy0,ty);
 333             fiz0             = _mm256_add_pd(fiz0,tz);
 334
 335             fjx0             = _mm256_add_pd(fjx0,tx);
 336             fjy0             = _mm256_add_pd(fjy0,ty);
 337             fjz0             = _mm256_add_pd(fjz0,tz);
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             r10              = _mm256_mul_pd(rsq10,rinv10);
 344
 345             /* Compute parameters for interactions between i and j atoms */
 346             qq10             = _mm256_mul_pd(iq1,jq0);
 347
 348             /* EWALD ELECTROSTATICS */
 349
 350             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 351             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 352             ewitab           = _mm256_cvttpd_epi32(ewrt);
 353             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 354             ewitab           = _mm_slli_epi32(ewitab,2);
 355             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 356             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 357             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 358             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 359             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 360             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 361             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 362             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 363             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 364
 365             /* Update potential sum for this i atom from the interaction with this j atom. */
 366             velecsum         = _mm256_add_pd(velecsum,velec);
 367
 368             fscal            = felec;
 369
 370             /* Calculate temporary vectorial force */
 371             tx               = _mm256_mul_pd(fscal,dx10);
 372             ty               = _mm256_mul_pd(fscal,dy10);
 373             tz               = _mm256_mul_pd(fscal,dz10);
 374
 375             /* Update vectorial force */
 376             fix1             = _mm256_add_pd(fix1,tx);
 377             fiy1             = _mm256_add_pd(fiy1,ty);
 378             fiz1             = _mm256_add_pd(fiz1,tz);
 379
 380             fjx0             = _mm256_add_pd(fjx0,tx);
 381             fjy0             = _mm256_add_pd(fjy0,ty);
 382             fjz0             = _mm256_add_pd(fjz0,tz);
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             r20              = _mm256_mul_pd(rsq20,rinv20);
 389
 390             /* Compute parameters for interactions between i and j atoms */
 391             qq20             = _mm256_mul_pd(iq2,jq0);
 392
 393             /* EWALD ELECTROSTATICS */
 394
 395             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 396             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 397             ewitab           = _mm256_cvttpd_epi32(ewrt);
 398             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 399             ewitab           = _mm_slli_epi32(ewitab,2);
 400             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 401             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 402             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 403             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 404             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 405             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 406             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 407             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 408             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 409
 410             /* Update potential sum for this i atom from the interaction with this j atom. */
 411             velecsum         = _mm256_add_pd(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             /* Calculate temporary vectorial force */
 416             tx               = _mm256_mul_pd(fscal,dx20);
 417             ty               = _mm256_mul_pd(fscal,dy20);
 418             tz               = _mm256_mul_pd(fscal,dz20);
 419
 420             /* Update vectorial force */
 421             fix2             = _mm256_add_pd(fix2,tx);
 422             fiy2             = _mm256_add_pd(fiy2,ty);
 423             fiz2             = _mm256_add_pd(fiz2,tz);
 424
 425             fjx0             = _mm256_add_pd(fjx0,tx);
 426             fjy0             = _mm256_add_pd(fjy0,ty);
 427             fjz0             = _mm256_add_pd(fjz0,tz);
 428
 429             fjptrA             = f+j_coord_offsetA;
 430             fjptrB             = f+j_coord_offsetB;
 431             fjptrC             = f+j_coord_offsetC;
 432             fjptrD             = f+j_coord_offsetD;
 433
 434             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 435
 436             /* Inner loop uses 160 flops */
 437         }
 438
 439         if(jidx<j_index_end)
 440         {
 441
 442             /* Get j neighbor index, and coordinate index */
 443             jnrlistA         = jjnr[jidx];
 444             jnrlistB         = jjnr[jidx+1];
 445             jnrlistC         = jjnr[jidx+2];
 446             jnrlistD         = jjnr[jidx+3];
 447             /* Sign of each element will be negative for non-real atoms.
 448              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 449              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 450              */
 451             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 452
 453             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 454             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 455             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 456
 457             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 458             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 459             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 460             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 461             j_coord_offsetA  = DIM*jnrA;
 462             j_coord_offsetB  = DIM*jnrB;
 463             j_coord_offsetC  = DIM*jnrC;
 464             j_coord_offsetD  = DIM*jnrD;
 465
 466             /* load j atom coordinates */
 467             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 468                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 469                                                  &jx0,&jy0,&jz0);
 470
 471             /* Calculate displacement vector */
 472             dx00             = _mm256_sub_pd(ix0,jx0);
 473             dy00             = _mm256_sub_pd(iy0,jy0);
 474             dz00             = _mm256_sub_pd(iz0,jz0);
 475             dx10             = _mm256_sub_pd(ix1,jx0);
 476             dy10             = _mm256_sub_pd(iy1,jy0);
 477             dz10             = _mm256_sub_pd(iz1,jz0);
 478             dx20             = _mm256_sub_pd(ix2,jx0);
 479             dy20             = _mm256_sub_pd(iy2,jy0);
 480             dz20             = _mm256_sub_pd(iz2,jz0);
 481
 482             /* Calculate squared distance and things based on it */
 483             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 484             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 485             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 486
 487             rinv00           = avx256_invsqrt_d(rsq00);
 488             rinv10           = avx256_invsqrt_d(rsq10);
 489             rinv20           = avx256_invsqrt_d(rsq20);
 490
 491             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 492             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 493             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 494
 495             /* Load parameters for j particles */
 496             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 497                                                                  charge+jnrC+0,charge+jnrD+0);
 498             vdwjidx0A        = 2*vdwtype[jnrA+0];
 499             vdwjidx0B        = 2*vdwtype[jnrB+0];
 500             vdwjidx0C        = 2*vdwtype[jnrC+0];
 501             vdwjidx0D        = 2*vdwtype[jnrD+0];
 502
 503             fjx0             = _mm256_setzero_pd();
 504             fjy0             = _mm256_setzero_pd();
 505             fjz0             = _mm256_setzero_pd();
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             r00              = _mm256_mul_pd(rsq00,rinv00);
 512             r00              = _mm256_andnot_pd(dummy_mask,r00);
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq00             = _mm256_mul_pd(iq0,jq0);
 516             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 517                                             vdwioffsetptr0+vdwjidx0B,
 518                                             vdwioffsetptr0+vdwjidx0C,
 519                                             vdwioffsetptr0+vdwjidx0D,
 520                                             &c6_00,&c12_00);
 521
 522             /* Calculate table index by multiplying r with table scale and truncate to integer */
 523             rt               = _mm256_mul_pd(r00,vftabscale);
 524             vfitab           = _mm256_cvttpd_epi32(rt);
 525             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 526             vfitab           = _mm_slli_epi32(vfitab,3);
 527
 528             /* EWALD ELECTROSTATICS */
 529
 530             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 531             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 532             ewitab           = _mm256_cvttpd_epi32(ewrt);
 533             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 534             ewitab           = _mm_slli_epi32(ewitab,2);
 535             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 536             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 537             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 538             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 539             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 540             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 541             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 542             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 543             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 544
 545             /* CUBIC SPLINE TABLE DISPERSION */
 546             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 547             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 548             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 549             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 550             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 551             Heps             = _mm256_mul_pd(vfeps,H);
 552             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 553             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 554             vvdw6            = _mm256_mul_pd(c6_00,VV);
 555             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 556             fvdw6            = _mm256_mul_pd(c6_00,FF);
 557
 558             /* CUBIC SPLINE TABLE REPULSION */
 559             vfitab           = _mm_add_epi32(vfitab,ifour);
 560             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 561             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 562             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 563             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 564             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 565             Heps             = _mm256_mul_pd(vfeps,H);
 566             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 567             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 568             vvdw12           = _mm256_mul_pd(c12_00,VV);
 569             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 570             fvdw12           = _mm256_mul_pd(c12_00,FF);
 571             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 572             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 573
 574             /* Update potential sum for this i atom from the interaction with this j atom. */
 575             velec            = _mm256_andnot_pd(dummy_mask,velec);
 576             velecsum         = _mm256_add_pd(velecsum,velec);
 577             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 578             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 579
 580             fscal            = _mm256_add_pd(felec,fvdw);
 581
 582             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm256_mul_pd(fscal,dx00);
 586             ty               = _mm256_mul_pd(fscal,dy00);
 587             tz               = _mm256_mul_pd(fscal,dz00);
 588
 589             /* Update vectorial force */
 590             fix0             = _mm256_add_pd(fix0,tx);
 591             fiy0             = _mm256_add_pd(fiy0,ty);
 592             fiz0             = _mm256_add_pd(fiz0,tz);
 593
 594             fjx0             = _mm256_add_pd(fjx0,tx);
 595             fjy0             = _mm256_add_pd(fjy0,ty);
 596             fjz0             = _mm256_add_pd(fjz0,tz);
 597
 598             /**************************
 599              * CALCULATE INTERACTIONS *
 600              **************************/
 601
 602             r10              = _mm256_mul_pd(rsq10,rinv10);
 603             r10              = _mm256_andnot_pd(dummy_mask,r10);
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq10             = _mm256_mul_pd(iq1,jq0);
 607
 608             /* EWALD ELECTROSTATICS */
 609
 610             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 611             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 612             ewitab           = _mm256_cvttpd_epi32(ewrt);
 613             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 614             ewitab           = _mm_slli_epi32(ewitab,2);
 615             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 616             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 617             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 618             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 619             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 620             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 621             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 622             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 623             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 624
 625             /* Update potential sum for this i atom from the interaction with this j atom. */
 626             velec            = _mm256_andnot_pd(dummy_mask,velec);
 627             velecsum         = _mm256_add_pd(velecsum,velec);
 628
 629             fscal            = felec;
 630
 631             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 632
 633             /* Calculate temporary vectorial force */
 634             tx               = _mm256_mul_pd(fscal,dx10);
 635             ty               = _mm256_mul_pd(fscal,dy10);
 636             tz               = _mm256_mul_pd(fscal,dz10);
 637
 638             /* Update vectorial force */
 639             fix1             = _mm256_add_pd(fix1,tx);
 640             fiy1             = _mm256_add_pd(fiy1,ty);
 641             fiz1             = _mm256_add_pd(fiz1,tz);
 642
 643             fjx0             = _mm256_add_pd(fjx0,tx);
 644             fjy0             = _mm256_add_pd(fjy0,ty);
 645             fjz0             = _mm256_add_pd(fjz0,tz);
 646
 647             /**************************
 648              * CALCULATE INTERACTIONS *
 649              **************************/
 650
 651             r20              = _mm256_mul_pd(rsq20,rinv20);
 652             r20              = _mm256_andnot_pd(dummy_mask,r20);
 653
 654             /* Compute parameters for interactions between i and j atoms */
 655             qq20             = _mm256_mul_pd(iq2,jq0);
 656
 657             /* EWALD ELECTROSTATICS */
 658
 659             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 660             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 661             ewitab           = _mm256_cvttpd_epi32(ewrt);
 662             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 663             ewitab           = _mm_slli_epi32(ewitab,2);
 664             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 665             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 666             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 667             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 668             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 669             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 670             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 671             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 672             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 673
 674             /* Update potential sum for this i atom from the interaction with this j atom. */
 675             velec            = _mm256_andnot_pd(dummy_mask,velec);
 676             velecsum         = _mm256_add_pd(velecsum,velec);
 677
 678             fscal            = felec;
 679
 680             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 681
 682             /* Calculate temporary vectorial force */
 683             tx               = _mm256_mul_pd(fscal,dx20);
 684             ty               = _mm256_mul_pd(fscal,dy20);
 685             tz               = _mm256_mul_pd(fscal,dz20);
 686
 687             /* Update vectorial force */
 688             fix2             = _mm256_add_pd(fix2,tx);
 689             fiy2             = _mm256_add_pd(fiy2,ty);
 690             fiz2             = _mm256_add_pd(fiz2,tz);
 691
 692             fjx0             = _mm256_add_pd(fjx0,tx);
 693             fjy0             = _mm256_add_pd(fjy0,ty);
 694             fjz0             = _mm256_add_pd(fjz0,tz);
 695
 696             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 697             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 698             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 699             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 700
 701             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 702
 703             /* Inner loop uses 163 flops */
 704         }
 705
 706         /* End of innermost loop */
 707
 708         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 709                                                  f+i_coord_offset,fshift+i_shift_offset);
 710
 711         ggid                        = gid[iidx];
 712         /* Update potential energies */
 713         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 714         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 715
 716         /* Increment number of inner iterations */
 717         inneriter                  += j_index_end - j_index_start;
 718
 719         /* Outer loop uses 20 flops */
 720     }
 721
 722     /* Increment number of outer iterations */
 723     outeriter        += nri;
 724
 725     /* Update outer/inner flops */
 726
 727     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*163);
 728 }
 729 /*
 730  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_double
 731  * Electrostatics interaction: Ewald
 732  * VdW interaction:            CubicSplineTable
 733  * Geometry:                   Water3-Particle
 734  * Calculate force/pot:        Force
 735  */
 736 void
 737 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_double
 738                     (t_nblist                    * gmx_restrict       nlist,
 739                      rvec                        * gmx_restrict          xx,
 740                      rvec                        * gmx_restrict          ff,
 741                      struct t_forcerec           * gmx_restrict          fr,
 742                      t_mdatoms                   * gmx_restrict     mdatoms,
 743                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 744                      t_nrnb                      * gmx_restrict        nrnb)
 745 {
 746     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 747      * just 0 for non-waters.
 748      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 749      * jnr indices corresponding to data put in the four positions in the SIMD register.
 750      */
 751     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 752     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 753     int              jnrA,jnrB,jnrC,jnrD;
 754     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 755     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 756     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 757     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 758     real             rcutoff_scalar;
 759     real             *shiftvec,*fshift,*x,*f;
 760     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 761     real             scratch[4*DIM];
 762     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 763     real *           vdwioffsetptr0;
 764     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 765     real *           vdwioffsetptr1;
 766     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 767     real *           vdwioffsetptr2;
 768     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 769     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 770     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 771     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 772     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 773     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 774     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 775     real             *charge;
 776     int              nvdwtype;
 777     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 778     int              *vdwtype;
 779     real             *vdwparam;
 780     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 781     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 782     __m128i          vfitab;
 783     __m128i          ifour       = _mm_set1_epi32(4);
 784     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 785     real             *vftab;
 786     __m128i          ewitab;
 787     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 788     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 789     real             *ewtab;
 790     __m256d          dummy_mask,cutoff_mask;
 791     __m128           tmpmask0,tmpmask1;
 792     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 793     __m256d          one     = _mm256_set1_pd(1.0);
 794     __m256d          two     = _mm256_set1_pd(2.0);
 795     x                = xx[0];
 796     f                = ff[0];
 797
 798     nri              = nlist->nri;
 799     iinr             = nlist->iinr;
 800     jindex           = nlist->jindex;
 801     jjnr             = nlist->jjnr;
 802     shiftidx         = nlist->shift;
 803     gid              = nlist->gid;
 804     shiftvec         = fr->shift_vec[0];
 805     fshift           = fr->fshift[0];
 806     facel            = _mm256_set1_pd(fr->ic->epsfac);
 807     charge           = mdatoms->chargeA;
 808     nvdwtype         = fr->ntype;
 809     vdwparam         = fr->nbfp;
 810     vdwtype          = mdatoms->typeA;
 811
 812     vftab            = kernel_data->table_vdw->data;
 813     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 814
 815     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 816     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 817     beta2            = _mm256_mul_pd(beta,beta);
 818     beta3            = _mm256_mul_pd(beta,beta2);
 819
 820     ewtab            = fr->ic->tabq_coul_F;
 821     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 822     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 823
 824     /* Setup water-specific parameters */
 825     inr              = nlist->iinr[0];
 826     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 827     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 828     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 829     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 830
 831     /* Avoid stupid compiler warnings */
 832     jnrA = jnrB = jnrC = jnrD = 0;
 833     j_coord_offsetA = 0;
 834     j_coord_offsetB = 0;
 835     j_coord_offsetC = 0;
 836     j_coord_offsetD = 0;
 837
 838     outeriter        = 0;
 839     inneriter        = 0;
 840
 841     for(iidx=0;iidx<4*DIM;iidx++)
 842     {
 843         scratch[iidx] = 0.0;
 844     }
 845
 846     /* Start outer loop over neighborlists */
 847     for(iidx=0; iidx<nri; iidx++)
 848     {
 849         /* Load shift vector for this list */
 850         i_shift_offset   = DIM*shiftidx[iidx];
 851
 852         /* Load limits for loop over neighbors */
 853         j_index_start    = jindex[iidx];
 854         j_index_end      = jindex[iidx+1];
 855
 856         /* Get outer coordinate index */
 857         inr              = iinr[iidx];
 858         i_coord_offset   = DIM*inr;
 859
 860         /* Load i particle coords and add shift vector */
 861         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 862                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 863
 864         fix0             = _mm256_setzero_pd();
 865         fiy0             = _mm256_setzero_pd();
 866         fiz0             = _mm256_setzero_pd();
 867         fix1             = _mm256_setzero_pd();
 868         fiy1             = _mm256_setzero_pd();
 869         fiz1             = _mm256_setzero_pd();
 870         fix2             = _mm256_setzero_pd();
 871         fiy2             = _mm256_setzero_pd();
 872         fiz2             = _mm256_setzero_pd();
 873
 874         /* Start inner kernel loop */
 875         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 876         {
 877
 878             /* Get j neighbor index, and coordinate index */
 879             jnrA             = jjnr[jidx];
 880             jnrB             = jjnr[jidx+1];
 881             jnrC             = jjnr[jidx+2];
 882             jnrD             = jjnr[jidx+3];
 883             j_coord_offsetA  = DIM*jnrA;
 884             j_coord_offsetB  = DIM*jnrB;
 885             j_coord_offsetC  = DIM*jnrC;
 886             j_coord_offsetD  = DIM*jnrD;
 887
 888             /* load j atom coordinates */
 889             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 890                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 891                                                  &jx0,&jy0,&jz0);
 892
 893             /* Calculate displacement vector */
 894             dx00             = _mm256_sub_pd(ix0,jx0);
 895             dy00             = _mm256_sub_pd(iy0,jy0);
 896             dz00             = _mm256_sub_pd(iz0,jz0);
 897             dx10             = _mm256_sub_pd(ix1,jx0);
 898             dy10             = _mm256_sub_pd(iy1,jy0);
 899             dz10             = _mm256_sub_pd(iz1,jz0);
 900             dx20             = _mm256_sub_pd(ix2,jx0);
 901             dy20             = _mm256_sub_pd(iy2,jy0);
 902             dz20             = _mm256_sub_pd(iz2,jz0);
 903
 904             /* Calculate squared distance and things based on it */
 905             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 906             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 907             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 908
 909             rinv00           = avx256_invsqrt_d(rsq00);
 910             rinv10           = avx256_invsqrt_d(rsq10);
 911             rinv20           = avx256_invsqrt_d(rsq20);
 912
 913             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 914             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 915             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 916
 917             /* Load parameters for j particles */
 918             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 919                                                                  charge+jnrC+0,charge+jnrD+0);
 920             vdwjidx0A        = 2*vdwtype[jnrA+0];
 921             vdwjidx0B        = 2*vdwtype[jnrB+0];
 922             vdwjidx0C        = 2*vdwtype[jnrC+0];
 923             vdwjidx0D        = 2*vdwtype[jnrD+0];
 924
 925             fjx0             = _mm256_setzero_pd();
 926             fjy0             = _mm256_setzero_pd();
 927             fjz0             = _mm256_setzero_pd();
 928
 929             /**************************
 930              * CALCULATE INTERACTIONS *
 931              **************************/
 932
 933             r00              = _mm256_mul_pd(rsq00,rinv00);
 934
 935             /* Compute parameters for interactions between i and j atoms */
 936             qq00             = _mm256_mul_pd(iq0,jq0);
 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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 944             rt               = _mm256_mul_pd(r00,vftabscale);
 945             vfitab           = _mm256_cvttpd_epi32(rt);
 946             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 947             vfitab           = _mm_slli_epi32(vfitab,3);
 948
 949             /* EWALD ELECTROSTATICS */
 950
 951             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 952             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 953             ewitab           = _mm256_cvttpd_epi32(ewrt);
 954             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 955             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 956                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 957                                             &ewtabF,&ewtabFn);
 958             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 959             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 960
 961             /* CUBIC SPLINE TABLE DISPERSION */
 962             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 963             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 964             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 965             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 966             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 967             Heps             = _mm256_mul_pd(vfeps,H);
 968             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 969             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 970             fvdw6            = _mm256_mul_pd(c6_00,FF);
 971
 972             /* CUBIC SPLINE TABLE REPULSION */
 973             vfitab           = _mm_add_epi32(vfitab,ifour);
 974             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 975             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 976             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 977             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 978             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 979             Heps             = _mm256_mul_pd(vfeps,H);
 980             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 981             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 982             fvdw12           = _mm256_mul_pd(c12_00,FF);
 983             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 984
 985             fscal            = _mm256_add_pd(felec,fvdw);
 986
 987             /* Calculate temporary vectorial force */
 988             tx               = _mm256_mul_pd(fscal,dx00);
 989             ty               = _mm256_mul_pd(fscal,dy00);
 990             tz               = _mm256_mul_pd(fscal,dz00);
 991
 992             /* Update vectorial force */
 993             fix0             = _mm256_add_pd(fix0,tx);
 994             fiy0             = _mm256_add_pd(fiy0,ty);
 995             fiz0             = _mm256_add_pd(fiz0,tz);
 996
 997             fjx0             = _mm256_add_pd(fjx0,tx);
 998             fjy0             = _mm256_add_pd(fjy0,ty);
 999             fjz0             = _mm256_add_pd(fjz0,tz);
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             r10              = _mm256_mul_pd(rsq10,rinv10);
1006
1007             /* Compute parameters for interactions between i and j atoms */
1008             qq10             = _mm256_mul_pd(iq1,jq0);
1009
1010             /* EWALD ELECTROSTATICS */
1011
1012             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1013             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1014             ewitab           = _mm256_cvttpd_epi32(ewrt);
1015             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1016             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1017                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1018                                             &ewtabF,&ewtabFn);
1019             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1020             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1021
1022             fscal            = felec;
1023
1024             /* Calculate temporary vectorial force */
1025             tx               = _mm256_mul_pd(fscal,dx10);
1026             ty               = _mm256_mul_pd(fscal,dy10);
1027             tz               = _mm256_mul_pd(fscal,dz10);
1028
1029             /* Update vectorial force */
1030             fix1             = _mm256_add_pd(fix1,tx);
1031             fiy1             = _mm256_add_pd(fiy1,ty);
1032             fiz1             = _mm256_add_pd(fiz1,tz);
1033
1034             fjx0             = _mm256_add_pd(fjx0,tx);
1035             fjy0             = _mm256_add_pd(fjy0,ty);
1036             fjz0             = _mm256_add_pd(fjz0,tz);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             r20              = _mm256_mul_pd(rsq20,rinv20);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq20             = _mm256_mul_pd(iq2,jq0);
1046
1047             /* EWALD ELECTROSTATICS */
1048
1049             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1050             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1051             ewitab           = _mm256_cvttpd_epi32(ewrt);
1052             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1053             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1054                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1055                                             &ewtabF,&ewtabFn);
1056             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1057             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1058
1059             fscal            = felec;
1060
1061             /* Calculate temporary vectorial force */
1062             tx               = _mm256_mul_pd(fscal,dx20);
1063             ty               = _mm256_mul_pd(fscal,dy20);
1064             tz               = _mm256_mul_pd(fscal,dz20);
1065
1066             /* Update vectorial force */
1067             fix2             = _mm256_add_pd(fix2,tx);
1068             fiy2             = _mm256_add_pd(fiy2,ty);
1069             fiz2             = _mm256_add_pd(fiz2,tz);
1070
1071             fjx0             = _mm256_add_pd(fjx0,tx);
1072             fjy0             = _mm256_add_pd(fjy0,ty);
1073             fjz0             = _mm256_add_pd(fjz0,tz);
1074
1075             fjptrA             = f+j_coord_offsetA;
1076             fjptrB             = f+j_coord_offsetB;
1077             fjptrC             = f+j_coord_offsetC;
1078             fjptrD             = f+j_coord_offsetD;
1079
1080             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1081
1082             /* Inner loop uses 137 flops */
1083         }
1084
1085         if(jidx<j_index_end)
1086         {
1087
1088             /* Get j neighbor index, and coordinate index */
1089             jnrlistA         = jjnr[jidx];
1090             jnrlistB         = jjnr[jidx+1];
1091             jnrlistC         = jjnr[jidx+2];
1092             jnrlistD         = jjnr[jidx+3];
1093             /* Sign of each element will be negative for non-real atoms.
1094              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1095              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1096              */
1097             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1098
1099             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1100             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1101             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1102
1103             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1104             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1105             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1106             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1107             j_coord_offsetA  = DIM*jnrA;
1108             j_coord_offsetB  = DIM*jnrB;
1109             j_coord_offsetC  = DIM*jnrC;
1110             j_coord_offsetD  = DIM*jnrD;
1111
1112             /* load j atom coordinates */
1113             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1114                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1115                                                  &jx0,&jy0,&jz0);
1116
1117             /* Calculate displacement vector */
1118             dx00             = _mm256_sub_pd(ix0,jx0);
1119             dy00             = _mm256_sub_pd(iy0,jy0);
1120             dz00             = _mm256_sub_pd(iz0,jz0);
1121             dx10             = _mm256_sub_pd(ix1,jx0);
1122             dy10             = _mm256_sub_pd(iy1,jy0);
1123             dz10             = _mm256_sub_pd(iz1,jz0);
1124             dx20             = _mm256_sub_pd(ix2,jx0);
1125             dy20             = _mm256_sub_pd(iy2,jy0);
1126             dz20             = _mm256_sub_pd(iz2,jz0);
1127
1128             /* Calculate squared distance and things based on it */
1129             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1130             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1131             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1132
1133             rinv00           = avx256_invsqrt_d(rsq00);
1134             rinv10           = avx256_invsqrt_d(rsq10);
1135             rinv20           = avx256_invsqrt_d(rsq20);
1136
1137             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1138             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1139             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1140
1141             /* Load parameters for j particles */
1142             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1143                                                                  charge+jnrC+0,charge+jnrD+0);
1144             vdwjidx0A        = 2*vdwtype[jnrA+0];
1145             vdwjidx0B        = 2*vdwtype[jnrB+0];
1146             vdwjidx0C        = 2*vdwtype[jnrC+0];
1147             vdwjidx0D        = 2*vdwtype[jnrD+0];
1148
1149             fjx0             = _mm256_setzero_pd();
1150             fjy0             = _mm256_setzero_pd();
1151             fjz0             = _mm256_setzero_pd();
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             r00              = _mm256_mul_pd(rsq00,rinv00);
1158             r00              = _mm256_andnot_pd(dummy_mask,r00);
1159
1160             /* Compute parameters for interactions between i and j atoms */
1161             qq00             = _mm256_mul_pd(iq0,jq0);
1162             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1163                                             vdwioffsetptr0+vdwjidx0B,
1164                                             vdwioffsetptr0+vdwjidx0C,
1165                                             vdwioffsetptr0+vdwjidx0D,
1166                                             &c6_00,&c12_00);
1167
1168             /* Calculate table index by multiplying r with table scale and truncate to integer */
1169             rt               = _mm256_mul_pd(r00,vftabscale);
1170             vfitab           = _mm256_cvttpd_epi32(rt);
1171             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1172             vfitab           = _mm_slli_epi32(vfitab,3);
1173
1174             /* EWALD ELECTROSTATICS */
1175
1176             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1177             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1178             ewitab           = _mm256_cvttpd_epi32(ewrt);
1179             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1180             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1181                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1182                                             &ewtabF,&ewtabFn);
1183             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1184             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1185
1186             /* CUBIC SPLINE TABLE DISPERSION */
1187             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1188             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1189             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1190             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1191             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1192             Heps             = _mm256_mul_pd(vfeps,H);
1193             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1194             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1195             fvdw6            = _mm256_mul_pd(c6_00,FF);
1196
1197             /* CUBIC SPLINE TABLE REPULSION */
1198             vfitab           = _mm_add_epi32(vfitab,ifour);
1199             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1200             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1201             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1202             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1203             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1204             Heps             = _mm256_mul_pd(vfeps,H);
1205             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1206             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1207             fvdw12           = _mm256_mul_pd(c12_00,FF);
1208             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1209
1210             fscal            = _mm256_add_pd(felec,fvdw);
1211
1212             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1213
1214             /* Calculate temporary vectorial force */
1215             tx               = _mm256_mul_pd(fscal,dx00);
1216             ty               = _mm256_mul_pd(fscal,dy00);
1217             tz               = _mm256_mul_pd(fscal,dz00);
1218
1219             /* Update vectorial force */
1220             fix0             = _mm256_add_pd(fix0,tx);
1221             fiy0             = _mm256_add_pd(fiy0,ty);
1222             fiz0             = _mm256_add_pd(fiz0,tz);
1223
1224             fjx0             = _mm256_add_pd(fjx0,tx);
1225             fjy0             = _mm256_add_pd(fjy0,ty);
1226             fjz0             = _mm256_add_pd(fjz0,tz);
1227
1228             /**************************
1229              * CALCULATE INTERACTIONS *
1230              **************************/
1231
1232             r10              = _mm256_mul_pd(rsq10,rinv10);
1233             r10              = _mm256_andnot_pd(dummy_mask,r10);
1234
1235             /* Compute parameters for interactions between i and j atoms */
1236             qq10             = _mm256_mul_pd(iq1,jq0);
1237
1238             /* EWALD ELECTROSTATICS */
1239
1240             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1241             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1242             ewitab           = _mm256_cvttpd_epi32(ewrt);
1243             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1244             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1245                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1246                                             &ewtabF,&ewtabFn);
1247             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1248             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1249
1250             fscal            = felec;
1251
1252             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1253
1254             /* Calculate temporary vectorial force */
1255             tx               = _mm256_mul_pd(fscal,dx10);
1256             ty               = _mm256_mul_pd(fscal,dy10);
1257             tz               = _mm256_mul_pd(fscal,dz10);
1258
1259             /* Update vectorial force */
1260             fix1             = _mm256_add_pd(fix1,tx);
1261             fiy1             = _mm256_add_pd(fiy1,ty);
1262             fiz1             = _mm256_add_pd(fiz1,tz);
1263
1264             fjx0             = _mm256_add_pd(fjx0,tx);
1265             fjy0             = _mm256_add_pd(fjy0,ty);
1266             fjz0             = _mm256_add_pd(fjz0,tz);
1267
1268             /**************************
1269              * CALCULATE INTERACTIONS *
1270              **************************/
1271
1272             r20              = _mm256_mul_pd(rsq20,rinv20);
1273             r20              = _mm256_andnot_pd(dummy_mask,r20);
1274
1275             /* Compute parameters for interactions between i and j atoms */
1276             qq20             = _mm256_mul_pd(iq2,jq0);
1277
1278             /* EWALD ELECTROSTATICS */
1279
1280             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1281             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1282             ewitab           = _mm256_cvttpd_epi32(ewrt);
1283             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1284             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1285                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1286                                             &ewtabF,&ewtabFn);
1287             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1288             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1289
1290             fscal            = felec;
1291
1292             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1293
1294             /* Calculate temporary vectorial force */
1295             tx               = _mm256_mul_pd(fscal,dx20);
1296             ty               = _mm256_mul_pd(fscal,dy20);
1297             tz               = _mm256_mul_pd(fscal,dz20);
1298
1299             /* Update vectorial force */
1300             fix2             = _mm256_add_pd(fix2,tx);
1301             fiy2             = _mm256_add_pd(fiy2,ty);
1302             fiz2             = _mm256_add_pd(fiz2,tz);
1303
1304             fjx0             = _mm256_add_pd(fjx0,tx);
1305             fjy0             = _mm256_add_pd(fjy0,ty);
1306             fjz0             = _mm256_add_pd(fjz0,tz);
1307
1308             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1309             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1310             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1311             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1312
1313             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1314
1315             /* Inner loop uses 140 flops */
1316         }
1317
1318         /* End of innermost loop */
1319
1320         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1321                                                  f+i_coord_offset,fshift+i_shift_offset);
1322
1323         /* Increment number of inner iterations */
1324         inneriter                  += j_index_end - j_index_start;
1325
1326         /* Outer loop uses 18 flops */
1327     }
1328
1329     /* Increment number of outer iterations */
1330     outeriter        += nri;
1331
1332     /* Update outer/inner flops */
1333
1334     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*140);
1335 }