src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwLJEw_GeomP1P1_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_VdwLJEw_GeomP1P1_VF_avx_256_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     real *           vdwgridioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  87     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  90     real             *charge;
  91     int              nvdwtype;
  92     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  93     int              *vdwtype;
  94     real             *vdwparam;
  95     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  96     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  97     __m256d           c6grid_00;
  98     real             *vdwgridparam;
  99     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 100     __m256d           one_half  = _mm256_set1_pd(0.5);
 101     __m256d           minus_one = _mm256_set1_pd(-1.0);
 102     __m128i          ewitab;
 103     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 105     real             *ewtab;
 106     __m256d          dummy_mask,cutoff_mask;
 107     __m128           tmpmask0,tmpmask1;
 108     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 109     __m256d          one     = _mm256_set1_pd(1.0);
 110     __m256d          two     = _mm256_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_pd(fr->ic->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127     vdwgridparam     = fr->ljpme_c6grid;
 128     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 129     ewclj            = _mm256_set1_pd(fr->ic->ewaldcoeff_lj);
 130     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 131
 132     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 133     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 134     beta2            = _mm256_mul_pd(beta,beta);
 135     beta3            = _mm256_mul_pd(beta,beta2);
 136
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 172
 173         fix0             = _mm256_setzero_pd();
 174         fiy0             = _mm256_setzero_pd();
 175         fiz0             = _mm256_setzero_pd();
 176
 177         /* Load parameters for i particles */
 178         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 179         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 180         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm256_setzero_pd();
 184         vvdwsum          = _mm256_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 203                                                  &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm256_sub_pd(ix0,jx0);
 207             dy00             = _mm256_sub_pd(iy0,jy0);
 208             dz00             = _mm256_sub_pd(iz0,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 212
 213             rinv00           = avx256_invsqrt_d(rsq00);
 214
 215             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 219                                                                  charge+jnrC+0,charge+jnrD+0);
 220             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222             vdwjidx0C        = 2*vdwtype[jnrC+0];
 223             vdwjidx0D        = 2*vdwtype[jnrD+0];
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             r00              = _mm256_mul_pd(rsq00,rinv00);
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             qq00             = _mm256_mul_pd(iq0,jq0);
 233             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 234                                             vdwioffsetptr0+vdwjidx0B,
 235                                             vdwioffsetptr0+vdwjidx0C,
 236                                             vdwioffsetptr0+vdwjidx0D,
 237                                             &c6_00,&c12_00);
 238
 239             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 240                                                                   vdwgridioffsetptr0+vdwjidx0B,
 241                                                                   vdwgridioffsetptr0+vdwjidx0C,
 242                                                                   vdwgridioffsetptr0+vdwjidx0D);
 243
 244             /* EWALD ELECTROSTATICS */
 245
 246             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 247             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 248             ewitab           = _mm256_cvttpd_epi32(ewrt);
 249             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 250             ewitab           = _mm_slli_epi32(ewitab,2);
 251             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 252             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 253             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 254             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 255             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 256             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 257             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 258             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 259             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 260
 261             /* Analytical LJ-PME */
 262             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 263             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 264             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 265             exponent         = avx256_exp_d(ewcljrsq);
 266             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 267             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 268             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 269             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 270             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 271             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 272             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 273             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);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velecsum         = _mm256_add_pd(velecsum,velec);
 277             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 278
 279             fscal            = _mm256_add_pd(felec,fvdw);
 280
 281             /* Calculate temporary vectorial force */
 282             tx               = _mm256_mul_pd(fscal,dx00);
 283             ty               = _mm256_mul_pd(fscal,dy00);
 284             tz               = _mm256_mul_pd(fscal,dz00);
 285
 286             /* Update vectorial force */
 287             fix0             = _mm256_add_pd(fix0,tx);
 288             fiy0             = _mm256_add_pd(fiy0,ty);
 289             fiz0             = _mm256_add_pd(fiz0,tz);
 290
 291             fjptrA             = f+j_coord_offsetA;
 292             fjptrB             = f+j_coord_offsetB;
 293             fjptrC             = f+j_coord_offsetC;
 294             fjptrD             = f+j_coord_offsetD;
 295             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 296
 297             /* Inner loop uses 72 flops */
 298         }
 299
 300         if(jidx<j_index_end)
 301         {
 302
 303             /* Get j neighbor index, and coordinate index */
 304             jnrlistA         = jjnr[jidx];
 305             jnrlistB         = jjnr[jidx+1];
 306             jnrlistC         = jjnr[jidx+2];
 307             jnrlistD         = jjnr[jidx+3];
 308             /* Sign of each element will be negative for non-real atoms.
 309              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 310              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 311              */
 312             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 313
 314             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 315             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 316             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 317
 318             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 319             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 320             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 321             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 322             j_coord_offsetA  = DIM*jnrA;
 323             j_coord_offsetB  = DIM*jnrB;
 324             j_coord_offsetC  = DIM*jnrC;
 325             j_coord_offsetD  = DIM*jnrD;
 326
 327             /* load j atom coordinates */
 328             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 329                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 330                                                  &jx0,&jy0,&jz0);
 331
 332             /* Calculate displacement vector */
 333             dx00             = _mm256_sub_pd(ix0,jx0);
 334             dy00             = _mm256_sub_pd(iy0,jy0);
 335             dz00             = _mm256_sub_pd(iz0,jz0);
 336
 337             /* Calculate squared distance and things based on it */
 338             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 339
 340             rinv00           = avx256_invsqrt_d(rsq00);
 341
 342             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 343
 344             /* Load parameters for j particles */
 345             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 346                                                                  charge+jnrC+0,charge+jnrD+0);
 347             vdwjidx0A        = 2*vdwtype[jnrA+0];
 348             vdwjidx0B        = 2*vdwtype[jnrB+0];
 349             vdwjidx0C        = 2*vdwtype[jnrC+0];
 350             vdwjidx0D        = 2*vdwtype[jnrD+0];
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             r00              = _mm256_mul_pd(rsq00,rinv00);
 357             r00              = _mm256_andnot_pd(dummy_mask,r00);
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq00             = _mm256_mul_pd(iq0,jq0);
 361             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 362                                             vdwioffsetptr0+vdwjidx0B,
 363                                             vdwioffsetptr0+vdwjidx0C,
 364                                             vdwioffsetptr0+vdwjidx0D,
 365                                             &c6_00,&c12_00);
 366
 367             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 368                                                                   vdwgridioffsetptr0+vdwjidx0B,
 369                                                                   vdwgridioffsetptr0+vdwjidx0C,
 370                                                                   vdwgridioffsetptr0+vdwjidx0D);
 371
 372             /* EWALD ELECTROSTATICS */
 373
 374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 375             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 376             ewitab           = _mm256_cvttpd_epi32(ewrt);
 377             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 378             ewitab           = _mm_slli_epi32(ewitab,2);
 379             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 380             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 381             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 382             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 383             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 384             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 385             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 386             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 387             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 388
 389             /* Analytical LJ-PME */
 390             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 391             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 392             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 393             exponent         = avx256_exp_d(ewcljrsq);
 394             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 395             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 396             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 397             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 398             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 399             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 400             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 401             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);
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velec            = _mm256_andnot_pd(dummy_mask,velec);
 405             velecsum         = _mm256_add_pd(velecsum,velec);
 406             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 407             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 408
 409             fscal            = _mm256_add_pd(felec,fvdw);
 410
 411             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 412
 413             /* Calculate temporary vectorial force */
 414             tx               = _mm256_mul_pd(fscal,dx00);
 415             ty               = _mm256_mul_pd(fscal,dy00);
 416             tz               = _mm256_mul_pd(fscal,dz00);
 417
 418             /* Update vectorial force */
 419             fix0             = _mm256_add_pd(fix0,tx);
 420             fiy0             = _mm256_add_pd(fiy0,ty);
 421             fiz0             = _mm256_add_pd(fiz0,tz);
 422
 423             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 424             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 425             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 426             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 427             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 428
 429             /* Inner loop uses 73 flops */
 430         }
 431
 432         /* End of innermost loop */
 433
 434         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 435                                                  f+i_coord_offset,fshift+i_shift_offset);
 436
 437         ggid                        = gid[iidx];
 438         /* Update potential energies */
 439         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 440         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 441
 442         /* Increment number of inner iterations */
 443         inneriter                  += j_index_end - j_index_start;
 444
 445         /* Outer loop uses 9 flops */
 446     }
 447
 448     /* Increment number of outer iterations */
 449     outeriter        += nri;
 450
 451     /* Update outer/inner flops */
 452
 453     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*73);
 454 }
 455 /*
 456  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_256_double
 457  * Electrostatics interaction: Ewald
 458  * VdW interaction:            LJEwald
 459  * Geometry:                   Particle-Particle
 460  * Calculate force/pot:        Force
 461  */
 462 void
 463 nb_kernel_ElecEw_VdwLJEw_GeomP1P1_F_avx_256_double
 464                     (t_nblist                    * gmx_restrict       nlist,
 465                      rvec                        * gmx_restrict          xx,
 466                      rvec                        * gmx_restrict          ff,
 467                      struct t_forcerec           * gmx_restrict          fr,
 468                      t_mdatoms                   * gmx_restrict     mdatoms,
 469                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 470                      t_nrnb                      * gmx_restrict        nrnb)
 471 {
 472     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 473      * just 0 for non-waters.
 474      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 475      * jnr indices corresponding to data put in the four positions in the SIMD register.
 476      */
 477     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 478     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 479     int              jnrA,jnrB,jnrC,jnrD;
 480     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 481     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 482     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 483     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 484     real             rcutoff_scalar;
 485     real             *shiftvec,*fshift,*x,*f;
 486     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 487     real             scratch[4*DIM];
 488     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 489     real *           vdwioffsetptr0;
 490     real *           vdwgridioffsetptr0;
 491     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 492     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 493     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 494     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 495     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 496     real             *charge;
 497     int              nvdwtype;
 498     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 499     int              *vdwtype;
 500     real             *vdwparam;
 501     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 502     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 503     __m256d           c6grid_00;
 504     real             *vdwgridparam;
 505     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 506     __m256d           one_half  = _mm256_set1_pd(0.5);
 507     __m256d           minus_one = _mm256_set1_pd(-1.0);
 508     __m128i          ewitab;
 509     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 510     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 511     real             *ewtab;
 512     __m256d          dummy_mask,cutoff_mask;
 513     __m128           tmpmask0,tmpmask1;
 514     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 515     __m256d          one     = _mm256_set1_pd(1.0);
 516     __m256d          two     = _mm256_set1_pd(2.0);
 517     x                = xx[0];
 518     f                = ff[0];
 519
 520     nri              = nlist->nri;
 521     iinr             = nlist->iinr;
 522     jindex           = nlist->jindex;
 523     jjnr             = nlist->jjnr;
 524     shiftidx         = nlist->shift;
 525     gid              = nlist->gid;
 526     shiftvec         = fr->shift_vec[0];
 527     fshift           = fr->fshift[0];
 528     facel            = _mm256_set1_pd(fr->ic->epsfac);
 529     charge           = mdatoms->chargeA;
 530     nvdwtype         = fr->ntype;
 531     vdwparam         = fr->nbfp;
 532     vdwtype          = mdatoms->typeA;
 533     vdwgridparam     = fr->ljpme_c6grid;
 534     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 535     ewclj            = _mm256_set1_pd(fr->ic->ewaldcoeff_lj);
 536     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 537
 538     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 539     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 540     beta2            = _mm256_mul_pd(beta,beta);
 541     beta3            = _mm256_mul_pd(beta,beta2);
 542
 543     ewtab            = fr->ic->tabq_coul_F;
 544     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 545     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 546
 547     /* Avoid stupid compiler warnings */
 548     jnrA = jnrB = jnrC = jnrD = 0;
 549     j_coord_offsetA = 0;
 550     j_coord_offsetB = 0;
 551     j_coord_offsetC = 0;
 552     j_coord_offsetD = 0;
 553
 554     outeriter        = 0;
 555     inneriter        = 0;
 556
 557     for(iidx=0;iidx<4*DIM;iidx++)
 558     {
 559         scratch[iidx] = 0.0;
 560     }
 561
 562     /* Start outer loop over neighborlists */
 563     for(iidx=0; iidx<nri; iidx++)
 564     {
 565         /* Load shift vector for this list */
 566         i_shift_offset   = DIM*shiftidx[iidx];
 567
 568         /* Load limits for loop over neighbors */
 569         j_index_start    = jindex[iidx];
 570         j_index_end      = jindex[iidx+1];
 571
 572         /* Get outer coordinate index */
 573         inr              = iinr[iidx];
 574         i_coord_offset   = DIM*inr;
 575
 576         /* Load i particle coords and add shift vector */
 577         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 578
 579         fix0             = _mm256_setzero_pd();
 580         fiy0             = _mm256_setzero_pd();
 581         fiz0             = _mm256_setzero_pd();
 582
 583         /* Load parameters for i particles */
 584         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 585         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 586         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 587
 588         /* Start inner kernel loop */
 589         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 590         {
 591
 592             /* Get j neighbor index, and coordinate index */
 593             jnrA             = jjnr[jidx];
 594             jnrB             = jjnr[jidx+1];
 595             jnrC             = jjnr[jidx+2];
 596             jnrD             = jjnr[jidx+3];
 597             j_coord_offsetA  = DIM*jnrA;
 598             j_coord_offsetB  = DIM*jnrB;
 599             j_coord_offsetC  = DIM*jnrC;
 600             j_coord_offsetD  = DIM*jnrD;
 601
 602             /* load j atom coordinates */
 603             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 604                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 605                                                  &jx0,&jy0,&jz0);
 606
 607             /* Calculate displacement vector */
 608             dx00             = _mm256_sub_pd(ix0,jx0);
 609             dy00             = _mm256_sub_pd(iy0,jy0);
 610             dz00             = _mm256_sub_pd(iz0,jz0);
 611
 612             /* Calculate squared distance and things based on it */
 613             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 614
 615             rinv00           = avx256_invsqrt_d(rsq00);
 616
 617             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 618
 619             /* Load parameters for j particles */
 620             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 621                                                                  charge+jnrC+0,charge+jnrD+0);
 622             vdwjidx0A        = 2*vdwtype[jnrA+0];
 623             vdwjidx0B        = 2*vdwtype[jnrB+0];
 624             vdwjidx0C        = 2*vdwtype[jnrC+0];
 625             vdwjidx0D        = 2*vdwtype[jnrD+0];
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             r00              = _mm256_mul_pd(rsq00,rinv00);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq00             = _mm256_mul_pd(iq0,jq0);
 635             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 636                                             vdwioffsetptr0+vdwjidx0B,
 637                                             vdwioffsetptr0+vdwjidx0C,
 638                                             vdwioffsetptr0+vdwjidx0D,
 639                                             &c6_00,&c12_00);
 640
 641             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 642                                                                   vdwgridioffsetptr0+vdwjidx0B,
 643                                                                   vdwgridioffsetptr0+vdwjidx0C,
 644                                                                   vdwgridioffsetptr0+vdwjidx0D);
 645
 646             /* EWALD ELECTROSTATICS */
 647
 648             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 649             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 650             ewitab           = _mm256_cvttpd_epi32(ewrt);
 651             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 652             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 653                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 654                                             &ewtabF,&ewtabFn);
 655             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 656             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 657
 658             /* Analytical LJ-PME */
 659             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 660             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 661             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 662             exponent         = avx256_exp_d(ewcljrsq);
 663             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 664             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 665             /* f6A = 6 * C6grid * (1 - poly) */
 666             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 667             /* f6B = C6grid * exponent * beta^6 */
 668             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 669             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 670             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);
 671
 672             fscal            = _mm256_add_pd(felec,fvdw);
 673
 674             /* Calculate temporary vectorial force */
 675             tx               = _mm256_mul_pd(fscal,dx00);
 676             ty               = _mm256_mul_pd(fscal,dy00);
 677             tz               = _mm256_mul_pd(fscal,dz00);
 678
 679             /* Update vectorial force */
 680             fix0             = _mm256_add_pd(fix0,tx);
 681             fiy0             = _mm256_add_pd(fiy0,ty);
 682             fiz0             = _mm256_add_pd(fiz0,tz);
 683
 684             fjptrA             = f+j_coord_offsetA;
 685             fjptrB             = f+j_coord_offsetB;
 686             fjptrC             = f+j_coord_offsetC;
 687             fjptrD             = f+j_coord_offsetD;
 688             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 689
 690             /* Inner loop uses 59 flops */
 691         }
 692
 693         if(jidx<j_index_end)
 694         {
 695
 696             /* Get j neighbor index, and coordinate index */
 697             jnrlistA         = jjnr[jidx];
 698             jnrlistB         = jjnr[jidx+1];
 699             jnrlistC         = jjnr[jidx+2];
 700             jnrlistD         = jjnr[jidx+3];
 701             /* Sign of each element will be negative for non-real atoms.
 702              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 703              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 704              */
 705             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 706
 707             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 708             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 709             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 710
 711             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 712             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 713             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 714             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 715             j_coord_offsetA  = DIM*jnrA;
 716             j_coord_offsetB  = DIM*jnrB;
 717             j_coord_offsetC  = DIM*jnrC;
 718             j_coord_offsetD  = DIM*jnrD;
 719
 720             /* load j atom coordinates */
 721             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 722                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 723                                                  &jx0,&jy0,&jz0);
 724
 725             /* Calculate displacement vector */
 726             dx00             = _mm256_sub_pd(ix0,jx0);
 727             dy00             = _mm256_sub_pd(iy0,jy0);
 728             dz00             = _mm256_sub_pd(iz0,jz0);
 729
 730             /* Calculate squared distance and things based on it */
 731             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 732
 733             rinv00           = avx256_invsqrt_d(rsq00);
 734
 735             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 736
 737             /* Load parameters for j particles */
 738             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 739                                                                  charge+jnrC+0,charge+jnrD+0);
 740             vdwjidx0A        = 2*vdwtype[jnrA+0];
 741             vdwjidx0B        = 2*vdwtype[jnrB+0];
 742             vdwjidx0C        = 2*vdwtype[jnrC+0];
 743             vdwjidx0D        = 2*vdwtype[jnrD+0];
 744
 745             /**************************
 746              * CALCULATE INTERACTIONS *
 747              **************************/
 748
 749             r00              = _mm256_mul_pd(rsq00,rinv00);
 750             r00              = _mm256_andnot_pd(dummy_mask,r00);
 751
 752             /* Compute parameters for interactions between i and j atoms */
 753             qq00             = _mm256_mul_pd(iq0,jq0);
 754             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 755                                             vdwioffsetptr0+vdwjidx0B,
 756                                             vdwioffsetptr0+vdwjidx0C,
 757                                             vdwioffsetptr0+vdwjidx0D,
 758                                             &c6_00,&c12_00);
 759
 760             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 761                                                                   vdwgridioffsetptr0+vdwjidx0B,
 762                                                                   vdwgridioffsetptr0+vdwjidx0C,
 763                                                                   vdwgridioffsetptr0+vdwjidx0D);
 764
 765             /* EWALD ELECTROSTATICS */
 766
 767             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 768             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 769             ewitab           = _mm256_cvttpd_epi32(ewrt);
 770             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 771             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 772                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 773                                             &ewtabF,&ewtabFn);
 774             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 775             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 776
 777             /* Analytical LJ-PME */
 778             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 779             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 780             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 781             exponent         = avx256_exp_d(ewcljrsq);
 782             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 783             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 784             /* f6A = 6 * C6grid * (1 - poly) */
 785             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 786             /* f6B = C6grid * exponent * beta^6 */
 787             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 788             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 789             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);
 790
 791             fscal            = _mm256_add_pd(felec,fvdw);
 792
 793             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 794
 795             /* Calculate temporary vectorial force */
 796             tx               = _mm256_mul_pd(fscal,dx00);
 797             ty               = _mm256_mul_pd(fscal,dy00);
 798             tz               = _mm256_mul_pd(fscal,dz00);
 799
 800             /* Update vectorial force */
 801             fix0             = _mm256_add_pd(fix0,tx);
 802             fiy0             = _mm256_add_pd(fiy0,ty);
 803             fiz0             = _mm256_add_pd(fiz0,tz);
 804
 805             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 806             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 807             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 808             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 809             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 810
 811             /* Inner loop uses 60 flops */
 812         }
 813
 814         /* End of innermost loop */
 815
 816         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 817                                                  f+i_coord_offset,fshift+i_shift_offset);
 818
 819         /* Increment number of inner iterations */
 820         inneriter                  += j_index_end - j_index_start;
 821
 822         /* Outer loop uses 7 flops */
 823     }
 824
 825     /* Increment number of outer iterations */
 826     outeriter        += nri;
 827
 828     /* Update outer/inner flops */
 829
 830     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*60);
 831 }