src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_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     real *           vdwgridioffsetptr0;
  88     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  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          velec,felec,velecsum,facel,crf,krf,krf2;
  93     real             *charge;
  94     int              nvdwtype;
  95     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  96     int              *vdwtype;
  97     real             *vdwparam;
  98     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  99     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 100     __m256d           c6grid_00;
 101     real             *vdwgridparam;
 102     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 103     __m256d           one_half  = _mm256_set1_pd(0.5);
 104     __m256d           minus_one = _mm256_set1_pd(-1.0);
 105     __m128i          ewitab;
 106     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 108     real             *ewtab;
 109     __m256d          dummy_mask,cutoff_mask;
 110     __m128           tmpmask0,tmpmask1;
 111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 112     __m256d          one     = _mm256_set1_pd(1.0);
 113     __m256d          two     = _mm256_set1_pd(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_pd(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130     vdwgridparam     = fr->ljpme_c6grid;
 131     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 132     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 133     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 145     rcutoff_scalar   = fr->rcoulomb;
 146     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 147     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 148
 149     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 150     rvdw             = _mm256_set1_pd(fr->rvdw);
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     for(iidx=0;iidx<4*DIM;iidx++)
 163     {
 164         scratch[iidx] = 0.0;
 165     }
 166
 167     /* Start outer loop over neighborlists */
 168     for(iidx=0; iidx<nri; iidx++)
 169     {
 170         /* Load shift vector for this list */
 171         i_shift_offset   = DIM*shiftidx[iidx];
 172
 173         /* Load limits for loop over neighbors */
 174         j_index_start    = jindex[iidx];
 175         j_index_end      = jindex[iidx+1];
 176
 177         /* Get outer coordinate index */
 178         inr              = iinr[iidx];
 179         i_coord_offset   = DIM*inr;
 180
 181         /* Load i particle coords and add shift vector */
 182         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 183
 184         fix0             = _mm256_setzero_pd();
 185         fiy0             = _mm256_setzero_pd();
 186         fiz0             = _mm256_setzero_pd();
 187
 188         /* Load parameters for i particles */
 189         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 190         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 191         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm256_setzero_pd();
 195         vvdwsum          = _mm256_setzero_pd();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             j_coord_offsetA  = DIM*jnrA;
 207             j_coord_offsetB  = DIM*jnrB;
 208             j_coord_offsetC  = DIM*jnrC;
 209             j_coord_offsetD  = DIM*jnrD;
 210
 211             /* load j atom coordinates */
 212             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 214                                                  &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm256_sub_pd(ix0,jx0);
 218             dy00             = _mm256_sub_pd(iy0,jy0);
 219             dz00             = _mm256_sub_pd(iz0,jz0);
 220
 221             /* Calculate squared distance and things based on it */
 222             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 223
 224             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 225
 226             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 230                                                                  charge+jnrC+0,charge+jnrD+0);
 231             vdwjidx0A        = 2*vdwtype[jnrA+0];
 232             vdwjidx0B        = 2*vdwtype[jnrB+0];
 233             vdwjidx0C        = 2*vdwtype[jnrC+0];
 234             vdwjidx0D        = 2*vdwtype[jnrD+0];
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 241             {
 242
 243             r00              = _mm256_mul_pd(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm256_mul_pd(iq0,jq0);
 247             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 248                                             vdwioffsetptr0+vdwjidx0B,
 249                                             vdwioffsetptr0+vdwjidx0C,
 250                                             vdwioffsetptr0+vdwjidx0D,
 251                                             &c6_00,&c12_00);
 252
 253             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 254                                                                   vdwgridioffsetptr0+vdwjidx0B,
 255                                                                   vdwgridioffsetptr0+vdwjidx0C,
 256                                                                   vdwgridioffsetptr0+vdwjidx0D);
 257
 258             /* EWALD ELECTROSTATICS */
 259
 260             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 261             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 262             ewitab           = _mm256_cvttpd_epi32(ewrt);
 263             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 264             ewitab           = _mm_slli_epi32(ewitab,2);
 265             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 266             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 267             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 268             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 269             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 270             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 271             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 272             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 273             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 274
 275             /* Analytical LJ-PME */
 276             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 277             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 278             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 279             exponent         = gmx_simd_exp_d(ewcljrsq);
 280             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 281             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 282             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 283             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 284             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 285             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 286                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_add_pd(_mm256_mul_pd(c6_00,sh_vdw_invrcut6),_mm256_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 287             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 288             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 289
 290             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velec            = _mm256_and_pd(velec,cutoff_mask);
 294             velecsum         = _mm256_add_pd(velecsum,velec);
 295             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 296             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 297
 298             fscal            = _mm256_add_pd(felec,fvdw);
 299
 300             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm256_mul_pd(fscal,dx00);
 304             ty               = _mm256_mul_pd(fscal,dy00);
 305             tz               = _mm256_mul_pd(fscal,dz00);
 306
 307             /* Update vectorial force */
 308             fix0             = _mm256_add_pd(fix0,tx);
 309             fiy0             = _mm256_add_pd(fiy0,ty);
 310             fiz0             = _mm256_add_pd(fiz0,tz);
 311
 312             fjptrA             = f+j_coord_offsetA;
 313             fjptrB             = f+j_coord_offsetB;
 314             fjptrC             = f+j_coord_offsetC;
 315             fjptrD             = f+j_coord_offsetD;
 316             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 317
 318             }
 319
 320             /* Inner loop uses 82 flops */
 321         }
 322
 323         if(jidx<j_index_end)
 324         {
 325
 326             /* Get j neighbor index, and coordinate index */
 327             jnrlistA         = jjnr[jidx];
 328             jnrlistB         = jjnr[jidx+1];
 329             jnrlistC         = jjnr[jidx+2];
 330             jnrlistD         = jjnr[jidx+3];
 331             /* Sign of each element will be negative for non-real atoms.
 332              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 333              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 334              */
 335             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 336
 337             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 338             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 339             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 340
 341             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 342             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 343             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 344             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 345             j_coord_offsetA  = DIM*jnrA;
 346             j_coord_offsetB  = DIM*jnrB;
 347             j_coord_offsetC  = DIM*jnrC;
 348             j_coord_offsetD  = DIM*jnrD;
 349
 350             /* load j atom coordinates */
 351             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 352                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 353                                                  &jx0,&jy0,&jz0);
 354
 355             /* Calculate displacement vector */
 356             dx00             = _mm256_sub_pd(ix0,jx0);
 357             dy00             = _mm256_sub_pd(iy0,jy0);
 358             dz00             = _mm256_sub_pd(iz0,jz0);
 359
 360             /* Calculate squared distance and things based on it */
 361             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 362
 363             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 364
 365             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 366
 367             /* Load parameters for j particles */
 368             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 369                                                                  charge+jnrC+0,charge+jnrD+0);
 370             vdwjidx0A        = 2*vdwtype[jnrA+0];
 371             vdwjidx0B        = 2*vdwtype[jnrB+0];
 372             vdwjidx0C        = 2*vdwtype[jnrC+0];
 373             vdwjidx0D        = 2*vdwtype[jnrD+0];
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 380             {
 381
 382             r00              = _mm256_mul_pd(rsq00,rinv00);
 383             r00              = _mm256_andnot_pd(dummy_mask,r00);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq00             = _mm256_mul_pd(iq0,jq0);
 387             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 388                                             vdwioffsetptr0+vdwjidx0B,
 389                                             vdwioffsetptr0+vdwjidx0C,
 390                                             vdwioffsetptr0+vdwjidx0D,
 391                                             &c6_00,&c12_00);
 392
 393             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 394                                                                   vdwgridioffsetptr0+vdwjidx0B,
 395                                                                   vdwgridioffsetptr0+vdwjidx0C,
 396                                                                   vdwgridioffsetptr0+vdwjidx0D);
 397
 398             /* EWALD ELECTROSTATICS */
 399
 400             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 401             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 402             ewitab           = _mm256_cvttpd_epi32(ewrt);
 403             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 404             ewitab           = _mm_slli_epi32(ewitab,2);
 405             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 406             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 407             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 408             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 409             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 410             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 411             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 412             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 413             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 414
 415             /* Analytical LJ-PME */
 416             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 417             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 418             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 419             exponent         = gmx_simd_exp_d(ewcljrsq);
 420             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 421             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 422             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 423             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 424             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 425             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 426                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_add_pd(_mm256_mul_pd(c6_00,sh_vdw_invrcut6),_mm256_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 427             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 428             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 429
 430             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 431
 432             /* Update potential sum for this i atom from the interaction with this j atom. */
 433             velec            = _mm256_and_pd(velec,cutoff_mask);
 434             velec            = _mm256_andnot_pd(dummy_mask,velec);
 435             velecsum         = _mm256_add_pd(velecsum,velec);
 436             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 437             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 438             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 439
 440             fscal            = _mm256_add_pd(felec,fvdw);
 441
 442             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 443
 444             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 445
 446             /* Calculate temporary vectorial force */
 447             tx               = _mm256_mul_pd(fscal,dx00);
 448             ty               = _mm256_mul_pd(fscal,dy00);
 449             tz               = _mm256_mul_pd(fscal,dz00);
 450
 451             /* Update vectorial force */
 452             fix0             = _mm256_add_pd(fix0,tx);
 453             fiy0             = _mm256_add_pd(fiy0,ty);
 454             fiz0             = _mm256_add_pd(fiz0,tz);
 455
 456             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 457             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 458             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 459             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 460             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 461
 462             }
 463
 464             /* Inner loop uses 83 flops */
 465         }
 466
 467         /* End of innermost loop */
 468
 469         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 470                                                  f+i_coord_offset,fshift+i_shift_offset);
 471
 472         ggid                        = gid[iidx];
 473         /* Update potential energies */
 474         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 475         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 476
 477         /* Increment number of inner iterations */
 478         inneriter                  += j_index_end - j_index_start;
 479
 480         /* Outer loop uses 9 flops */
 481     }
 482
 483     /* Increment number of outer iterations */
 484     outeriter        += nri;
 485
 486     /* Update outer/inner flops */
 487
 488     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*83);
 489 }
 490 /*
 491  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_double
 492  * Electrostatics interaction: Ewald
 493  * VdW interaction:            LJEwald
 494  * Geometry:                   Particle-Particle
 495  * Calculate force/pot:        Force
 496  */
 497 void
 498 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_double
 499                     (t_nblist                    * gmx_restrict       nlist,
 500                      rvec                        * gmx_restrict          xx,
 501                      rvec                        * gmx_restrict          ff,
 502                      t_forcerec                  * gmx_restrict          fr,
 503                      t_mdatoms                   * gmx_restrict     mdatoms,
 504                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 505                      t_nrnb                      * gmx_restrict        nrnb)
 506 {
 507     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 508      * just 0 for non-waters.
 509      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 510      * jnr indices corresponding to data put in the four positions in the SIMD register.
 511      */
 512     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 513     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 514     int              jnrA,jnrB,jnrC,jnrD;
 515     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 516     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 517     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 518     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 519     real             rcutoff_scalar;
 520     real             *shiftvec,*fshift,*x,*f;
 521     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 522     real             scratch[4*DIM];
 523     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 524     real *           vdwioffsetptr0;
 525     real *           vdwgridioffsetptr0;
 526     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 527     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 528     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 529     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 530     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 531     real             *charge;
 532     int              nvdwtype;
 533     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 534     int              *vdwtype;
 535     real             *vdwparam;
 536     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 537     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 538     __m256d           c6grid_00;
 539     real             *vdwgridparam;
 540     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 541     __m256d           one_half  = _mm256_set1_pd(0.5);
 542     __m256d           minus_one = _mm256_set1_pd(-1.0);
 543     __m128i          ewitab;
 544     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 545     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 546     real             *ewtab;
 547     __m256d          dummy_mask,cutoff_mask;
 548     __m128           tmpmask0,tmpmask1;
 549     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 550     __m256d          one     = _mm256_set1_pd(1.0);
 551     __m256d          two     = _mm256_set1_pd(2.0);
 552     x                = xx[0];
 553     f                = ff[0];
 554
 555     nri              = nlist->nri;
 556     iinr             = nlist->iinr;
 557     jindex           = nlist->jindex;
 558     jjnr             = nlist->jjnr;
 559     shiftidx         = nlist->shift;
 560     gid              = nlist->gid;
 561     shiftvec         = fr->shift_vec[0];
 562     fshift           = fr->fshift[0];
 563     facel            = _mm256_set1_pd(fr->epsfac);
 564     charge           = mdatoms->chargeA;
 565     nvdwtype         = fr->ntype;
 566     vdwparam         = fr->nbfp;
 567     vdwtype          = mdatoms->typeA;
 568     vdwgridparam     = fr->ljpme_c6grid;
 569     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 570     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 571     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 572
 573     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 574     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 575     beta2            = _mm256_mul_pd(beta,beta);
 576     beta3            = _mm256_mul_pd(beta,beta2);
 577
 578     ewtab            = fr->ic->tabq_coul_F;
 579     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 580     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 581
 582     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 583     rcutoff_scalar   = fr->rcoulomb;
 584     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 585     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 586
 587     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 588     rvdw             = _mm256_set1_pd(fr->rvdw);
 589
 590     /* Avoid stupid compiler warnings */
 591     jnrA = jnrB = jnrC = jnrD = 0;
 592     j_coord_offsetA = 0;
 593     j_coord_offsetB = 0;
 594     j_coord_offsetC = 0;
 595     j_coord_offsetD = 0;
 596
 597     outeriter        = 0;
 598     inneriter        = 0;
 599
 600     for(iidx=0;iidx<4*DIM;iidx++)
 601     {
 602         scratch[iidx] = 0.0;
 603     }
 604
 605     /* Start outer loop over neighborlists */
 606     for(iidx=0; iidx<nri; iidx++)
 607     {
 608         /* Load shift vector for this list */
 609         i_shift_offset   = DIM*shiftidx[iidx];
 610
 611         /* Load limits for loop over neighbors */
 612         j_index_start    = jindex[iidx];
 613         j_index_end      = jindex[iidx+1];
 614
 615         /* Get outer coordinate index */
 616         inr              = iinr[iidx];
 617         i_coord_offset   = DIM*inr;
 618
 619         /* Load i particle coords and add shift vector */
 620         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 621
 622         fix0             = _mm256_setzero_pd();
 623         fiy0             = _mm256_setzero_pd();
 624         fiz0             = _mm256_setzero_pd();
 625
 626         /* Load parameters for i particles */
 627         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 628         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 629         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 630
 631         /* Start inner kernel loop */
 632         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 633         {
 634
 635             /* Get j neighbor index, and coordinate index */
 636             jnrA             = jjnr[jidx];
 637             jnrB             = jjnr[jidx+1];
 638             jnrC             = jjnr[jidx+2];
 639             jnrD             = jjnr[jidx+3];
 640             j_coord_offsetA  = DIM*jnrA;
 641             j_coord_offsetB  = DIM*jnrB;
 642             j_coord_offsetC  = DIM*jnrC;
 643             j_coord_offsetD  = DIM*jnrD;
 644
 645             /* load j atom coordinates */
 646             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 647                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 648                                                  &jx0,&jy0,&jz0);
 649
 650             /* Calculate displacement vector */
 651             dx00             = _mm256_sub_pd(ix0,jx0);
 652             dy00             = _mm256_sub_pd(iy0,jy0);
 653             dz00             = _mm256_sub_pd(iz0,jz0);
 654
 655             /* Calculate squared distance and things based on it */
 656             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 657
 658             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 659
 660             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 661
 662             /* Load parameters for j particles */
 663             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 664                                                                  charge+jnrC+0,charge+jnrD+0);
 665             vdwjidx0A        = 2*vdwtype[jnrA+0];
 666             vdwjidx0B        = 2*vdwtype[jnrB+0];
 667             vdwjidx0C        = 2*vdwtype[jnrC+0];
 668             vdwjidx0D        = 2*vdwtype[jnrD+0];
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 675             {
 676
 677             r00              = _mm256_mul_pd(rsq00,rinv00);
 678
 679             /* Compute parameters for interactions between i and j atoms */
 680             qq00             = _mm256_mul_pd(iq0,jq0);
 681             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 682                                             vdwioffsetptr0+vdwjidx0B,
 683                                             vdwioffsetptr0+vdwjidx0C,
 684                                             vdwioffsetptr0+vdwjidx0D,
 685                                             &c6_00,&c12_00);
 686
 687             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 688                                                                   vdwgridioffsetptr0+vdwjidx0B,
 689                                                                   vdwgridioffsetptr0+vdwjidx0C,
 690                                                                   vdwgridioffsetptr0+vdwjidx0D);
 691
 692             /* EWALD ELECTROSTATICS */
 693
 694             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 695             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 696             ewitab           = _mm256_cvttpd_epi32(ewrt);
 697             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 698             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 699                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 700                                             &ewtabF,&ewtabFn);
 701             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 702             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 703
 704             /* Analytical LJ-PME */
 705             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 706             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 707             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 708             exponent         = gmx_simd_exp_d(ewcljrsq);
 709             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 710             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 711             /* f6A = 6 * C6grid * (1 - poly) */
 712             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 713             /* f6B = C6grid * exponent * beta^6 */
 714             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 715             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 716             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 717
 718             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 719
 720             fscal            = _mm256_add_pd(felec,fvdw);
 721
 722             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 723
 724             /* Calculate temporary vectorial force */
 725             tx               = _mm256_mul_pd(fscal,dx00);
 726             ty               = _mm256_mul_pd(fscal,dy00);
 727             tz               = _mm256_mul_pd(fscal,dz00);
 728
 729             /* Update vectorial force */
 730             fix0             = _mm256_add_pd(fix0,tx);
 731             fiy0             = _mm256_add_pd(fiy0,ty);
 732             fiz0             = _mm256_add_pd(fiz0,tz);
 733
 734             fjptrA             = f+j_coord_offsetA;
 735             fjptrB             = f+j_coord_offsetB;
 736             fjptrC             = f+j_coord_offsetC;
 737             fjptrD             = f+j_coord_offsetD;
 738             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 739
 740             }
 741
 742             /* Inner loop uses 62 flops */
 743         }
 744
 745         if(jidx<j_index_end)
 746         {
 747
 748             /* Get j neighbor index, and coordinate index */
 749             jnrlistA         = jjnr[jidx];
 750             jnrlistB         = jjnr[jidx+1];
 751             jnrlistC         = jjnr[jidx+2];
 752             jnrlistD         = jjnr[jidx+3];
 753             /* Sign of each element will be negative for non-real atoms.
 754              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 755              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 756              */
 757             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 758
 759             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 760             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 761             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 762
 763             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 764             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 765             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 766             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 767             j_coord_offsetA  = DIM*jnrA;
 768             j_coord_offsetB  = DIM*jnrB;
 769             j_coord_offsetC  = DIM*jnrC;
 770             j_coord_offsetD  = DIM*jnrD;
 771
 772             /* load j atom coordinates */
 773             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 774                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 775                                                  &jx0,&jy0,&jz0);
 776
 777             /* Calculate displacement vector */
 778             dx00             = _mm256_sub_pd(ix0,jx0);
 779             dy00             = _mm256_sub_pd(iy0,jy0);
 780             dz00             = _mm256_sub_pd(iz0,jz0);
 781
 782             /* Calculate squared distance and things based on it */
 783             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 784
 785             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 786
 787             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 788
 789             /* Load parameters for j particles */
 790             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 791                                                                  charge+jnrC+0,charge+jnrD+0);
 792             vdwjidx0A        = 2*vdwtype[jnrA+0];
 793             vdwjidx0B        = 2*vdwtype[jnrB+0];
 794             vdwjidx0C        = 2*vdwtype[jnrC+0];
 795             vdwjidx0D        = 2*vdwtype[jnrD+0];
 796
 797             /**************************
 798              * CALCULATE INTERACTIONS *
 799              **************************/
 800
 801             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 802             {
 803
 804             r00              = _mm256_mul_pd(rsq00,rinv00);
 805             r00              = _mm256_andnot_pd(dummy_mask,r00);
 806
 807             /* Compute parameters for interactions between i and j atoms */
 808             qq00             = _mm256_mul_pd(iq0,jq0);
 809             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 810                                             vdwioffsetptr0+vdwjidx0B,
 811                                             vdwioffsetptr0+vdwjidx0C,
 812                                             vdwioffsetptr0+vdwjidx0D,
 813                                             &c6_00,&c12_00);
 814
 815             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 816                                                                   vdwgridioffsetptr0+vdwjidx0B,
 817                                                                   vdwgridioffsetptr0+vdwjidx0C,
 818                                                                   vdwgridioffsetptr0+vdwjidx0D);
 819
 820             /* EWALD ELECTROSTATICS */
 821
 822             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 823             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 824             ewitab           = _mm256_cvttpd_epi32(ewrt);
 825             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 826             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 827                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 828                                             &ewtabF,&ewtabFn);
 829             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 830             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 831
 832             /* Analytical LJ-PME */
 833             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 834             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 835             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 836             exponent         = gmx_simd_exp_d(ewcljrsq);
 837             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 838             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 839             /* f6A = 6 * C6grid * (1 - poly) */
 840             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 841             /* f6B = C6grid * exponent * beta^6 */
 842             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 843             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 844             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
 845
 846             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 847
 848             fscal            = _mm256_add_pd(felec,fvdw);
 849
 850             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 851
 852             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 853
 854             /* Calculate temporary vectorial force */
 855             tx               = _mm256_mul_pd(fscal,dx00);
 856             ty               = _mm256_mul_pd(fscal,dy00);
 857             tz               = _mm256_mul_pd(fscal,dz00);
 858
 859             /* Update vectorial force */
 860             fix0             = _mm256_add_pd(fix0,tx);
 861             fiy0             = _mm256_add_pd(fiy0,ty);
 862             fiz0             = _mm256_add_pd(fiz0,tz);
 863
 864             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 865             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 866             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 867             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 868             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 869
 870             }
 871
 872             /* Inner loop uses 63 flops */
 873         }
 874
 875         /* End of innermost loop */
 876
 877         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 878                                                  f+i_coord_offset,fshift+i_shift_offset);
 879
 880         /* Increment number of inner iterations */
 881         inneriter                  += j_index_end - j_index_start;
 882
 883         /* Outer loop uses 7 flops */
 884     }
 885
 886     /* Increment number of outer iterations */
 887     outeriter        += nri;
 888
 889     /* Update outer/inner flops */
 890
 891     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*63);
 892 }