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) 2012,2013,2014,2015,2017, 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_ElecEwSh_VdwLJEwSh_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_ElecEwSh_VdwLJEwSh_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 142     rcutoff_scalar   = fr->ic->rcoulomb;
 143     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 144     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 145
 146     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 147     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 148
 149     /* Avoid stupid compiler warnings */
 150     jnrA = jnrB = jnrC = jnrD = 0;
 151     j_coord_offsetA = 0;
 152     j_coord_offsetB = 0;
 153     j_coord_offsetC = 0;
 154     j_coord_offsetD = 0;
 155
 156     outeriter        = 0;
 157     inneriter        = 0;
 158
 159     for(iidx=0;iidx<4*DIM;iidx++)
 160     {
 161         scratch[iidx] = 0.0;
 162     }
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 180
 181         fix0             = _mm256_setzero_pd();
 182         fiy0             = _mm256_setzero_pd();
 183         fiz0             = _mm256_setzero_pd();
 184
 185         /* Load parameters for i particles */
 186         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 187         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 188         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_pd();
 192         vvdwsum          = _mm256_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_pd(ix0,jx0);
 215             dy00             = _mm256_sub_pd(iy0,jy0);
 216             dz00             = _mm256_sub_pd(iz0,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 220
 221             rinv00           = avx256_invsqrt_d(rsq00);
 222
 223             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 227                                                                  charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 238             {
 239
 240             r00              = _mm256_mul_pd(rsq00,rinv00);
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             qq00             = _mm256_mul_pd(iq0,jq0);
 244             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 245                                             vdwioffsetptr0+vdwjidx0B,
 246                                             vdwioffsetptr0+vdwjidx0C,
 247                                             vdwioffsetptr0+vdwjidx0D,
 248                                             &c6_00,&c12_00);
 249
 250             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 251                                                                   vdwgridioffsetptr0+vdwjidx0B,
 252                                                                   vdwgridioffsetptr0+vdwjidx0C,
 253                                                                   vdwgridioffsetptr0+vdwjidx0D);
 254
 255             /* EWALD ELECTROSTATICS */
 256
 257             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 258             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 259             ewitab           = _mm256_cvttpd_epi32(ewrt);
 260             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 261             ewitab           = _mm_slli_epi32(ewitab,2);
 262             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 263             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 264             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 265             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 266             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 267             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 268             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 269             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 270             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 271
 272             /* Analytical LJ-PME */
 273             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 274             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 275             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 276             exponent         = avx256_exp_d(ewcljrsq);
 277             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 278             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 279             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 280             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 281             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 282             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) ,
 283                                           _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));
 284             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 285             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);
 286
 287             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             velec            = _mm256_and_pd(velec,cutoff_mask);
 291             velecsum         = _mm256_add_pd(velecsum,velec);
 292             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 293             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 294
 295             fscal            = _mm256_add_pd(felec,fvdw);
 296
 297             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm256_mul_pd(fscal,dx00);
 301             ty               = _mm256_mul_pd(fscal,dy00);
 302             tz               = _mm256_mul_pd(fscal,dz00);
 303
 304             /* Update vectorial force */
 305             fix0             = _mm256_add_pd(fix0,tx);
 306             fiy0             = _mm256_add_pd(fiy0,ty);
 307             fiz0             = _mm256_add_pd(fiz0,tz);
 308
 309             fjptrA             = f+j_coord_offsetA;
 310             fjptrB             = f+j_coord_offsetB;
 311             fjptrC             = f+j_coord_offsetC;
 312             fjptrD             = f+j_coord_offsetD;
 313             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 314
 315             }
 316
 317             /* Inner loop uses 82 flops */
 318         }
 319
 320         if(jidx<j_index_end)
 321         {
 322
 323             /* Get j neighbor index, and coordinate index */
 324             jnrlistA         = jjnr[jidx];
 325             jnrlistB         = jjnr[jidx+1];
 326             jnrlistC         = jjnr[jidx+2];
 327             jnrlistD         = jjnr[jidx+3];
 328             /* Sign of each element will be negative for non-real atoms.
 329              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 330              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 331              */
 332             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 333
 334             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 335             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 336             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 337
 338             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 339             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 340             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 341             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 342             j_coord_offsetA  = DIM*jnrA;
 343             j_coord_offsetB  = DIM*jnrB;
 344             j_coord_offsetC  = DIM*jnrC;
 345             j_coord_offsetD  = DIM*jnrD;
 346
 347             /* load j atom coordinates */
 348             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 349                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 350                                                  &jx0,&jy0,&jz0);
 351
 352             /* Calculate displacement vector */
 353             dx00             = _mm256_sub_pd(ix0,jx0);
 354             dy00             = _mm256_sub_pd(iy0,jy0);
 355             dz00             = _mm256_sub_pd(iz0,jz0);
 356
 357             /* Calculate squared distance and things based on it */
 358             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 359
 360             rinv00           = avx256_invsqrt_d(rsq00);
 361
 362             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 363
 364             /* Load parameters for j particles */
 365             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 366                                                                  charge+jnrC+0,charge+jnrD+0);
 367             vdwjidx0A        = 2*vdwtype[jnrA+0];
 368             vdwjidx0B        = 2*vdwtype[jnrB+0];
 369             vdwjidx0C        = 2*vdwtype[jnrC+0];
 370             vdwjidx0D        = 2*vdwtype[jnrD+0];
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 377             {
 378
 379             r00              = _mm256_mul_pd(rsq00,rinv00);
 380             r00              = _mm256_andnot_pd(dummy_mask,r00);
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq00             = _mm256_mul_pd(iq0,jq0);
 384             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 385                                             vdwioffsetptr0+vdwjidx0B,
 386                                             vdwioffsetptr0+vdwjidx0C,
 387                                             vdwioffsetptr0+vdwjidx0D,
 388                                             &c6_00,&c12_00);
 389
 390             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 391                                                                   vdwgridioffsetptr0+vdwjidx0B,
 392                                                                   vdwgridioffsetptr0+vdwjidx0C,
 393                                                                   vdwgridioffsetptr0+vdwjidx0D);
 394
 395             /* EWALD ELECTROSTATICS */
 396
 397             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 398             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 399             ewitab           = _mm256_cvttpd_epi32(ewrt);
 400             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 401             ewitab           = _mm_slli_epi32(ewitab,2);
 402             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 403             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 404             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 405             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 406             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 407             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 408             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 409             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 410             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 411
 412             /* Analytical LJ-PME */
 413             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 414             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 415             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 416             exponent         = avx256_exp_d(ewcljrsq);
 417             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 418             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 419             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 420             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 421             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 422             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) ,
 423                                           _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));
 424             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 425             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);
 426
 427             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velec            = _mm256_and_pd(velec,cutoff_mask);
 431             velec            = _mm256_andnot_pd(dummy_mask,velec);
 432             velecsum         = _mm256_add_pd(velecsum,velec);
 433             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 434             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 435             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 436
 437             fscal            = _mm256_add_pd(felec,fvdw);
 438
 439             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 440
 441             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 442
 443             /* Calculate temporary vectorial force */
 444             tx               = _mm256_mul_pd(fscal,dx00);
 445             ty               = _mm256_mul_pd(fscal,dy00);
 446             tz               = _mm256_mul_pd(fscal,dz00);
 447
 448             /* Update vectorial force */
 449             fix0             = _mm256_add_pd(fix0,tx);
 450             fiy0             = _mm256_add_pd(fiy0,ty);
 451             fiz0             = _mm256_add_pd(fiz0,tz);
 452
 453             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 454             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 455             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 456             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 457             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 458
 459             }
 460
 461             /* Inner loop uses 83 flops */
 462         }
 463
 464         /* End of innermost loop */
 465
 466         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 467                                                  f+i_coord_offset,fshift+i_shift_offset);
 468
 469         ggid                        = gid[iidx];
 470         /* Update potential energies */
 471         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 472         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 473
 474         /* Increment number of inner iterations */
 475         inneriter                  += j_index_end - j_index_start;
 476
 477         /* Outer loop uses 9 flops */
 478     }
 479
 480     /* Increment number of outer iterations */
 481     outeriter        += nri;
 482
 483     /* Update outer/inner flops */
 484
 485     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*83);
 486 }
 487 /*
 488  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_double
 489  * Electrostatics interaction: Ewald
 490  * VdW interaction:            LJEwald
 491  * Geometry:                   Particle-Particle
 492  * Calculate force/pot:        Force
 493  */
 494 void
 495 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_double
 496                     (t_nblist                    * gmx_restrict       nlist,
 497                      rvec                        * gmx_restrict          xx,
 498                      rvec                        * gmx_restrict          ff,
 499                      struct t_forcerec           * gmx_restrict          fr,
 500                      t_mdatoms                   * gmx_restrict     mdatoms,
 501                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 502                      t_nrnb                      * gmx_restrict        nrnb)
 503 {
 504     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 505      * just 0 for non-waters.
 506      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 507      * jnr indices corresponding to data put in the four positions in the SIMD register.
 508      */
 509     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 510     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 511     int              jnrA,jnrB,jnrC,jnrD;
 512     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 513     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 514     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 515     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 516     real             rcutoff_scalar;
 517     real             *shiftvec,*fshift,*x,*f;
 518     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 519     real             scratch[4*DIM];
 520     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 521     real *           vdwioffsetptr0;
 522     real *           vdwgridioffsetptr0;
 523     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 524     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 525     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 526     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 527     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 528     real             *charge;
 529     int              nvdwtype;
 530     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 531     int              *vdwtype;
 532     real             *vdwparam;
 533     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 534     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 535     __m256d           c6grid_00;
 536     real             *vdwgridparam;
 537     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 538     __m256d           one_half  = _mm256_set1_pd(0.5);
 539     __m256d           minus_one = _mm256_set1_pd(-1.0);
 540     __m128i          ewitab;
 541     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 542     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 543     real             *ewtab;
 544     __m256d          dummy_mask,cutoff_mask;
 545     __m128           tmpmask0,tmpmask1;
 546     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 547     __m256d          one     = _mm256_set1_pd(1.0);
 548     __m256d          two     = _mm256_set1_pd(2.0);
 549     x                = xx[0];
 550     f                = ff[0];
 551
 552     nri              = nlist->nri;
 553     iinr             = nlist->iinr;
 554     jindex           = nlist->jindex;
 555     jjnr             = nlist->jjnr;
 556     shiftidx         = nlist->shift;
 557     gid              = nlist->gid;
 558     shiftvec         = fr->shift_vec[0];
 559     fshift           = fr->fshift[0];
 560     facel            = _mm256_set1_pd(fr->ic->epsfac);
 561     charge           = mdatoms->chargeA;
 562     nvdwtype         = fr->ntype;
 563     vdwparam         = fr->nbfp;
 564     vdwtype          = mdatoms->typeA;
 565     vdwgridparam     = fr->ljpme_c6grid;
 566     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 567     ewclj            = _mm256_set1_pd(fr->ic->ewaldcoeff_lj);
 568     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 569
 570     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 571     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 572     beta2            = _mm256_mul_pd(beta,beta);
 573     beta3            = _mm256_mul_pd(beta,beta2);
 574
 575     ewtab            = fr->ic->tabq_coul_F;
 576     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 577     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 578
 579     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 580     rcutoff_scalar   = fr->ic->rcoulomb;
 581     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 582     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 583
 584     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 585     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 586
 587     /* Avoid stupid compiler warnings */
 588     jnrA = jnrB = jnrC = jnrD = 0;
 589     j_coord_offsetA = 0;
 590     j_coord_offsetB = 0;
 591     j_coord_offsetC = 0;
 592     j_coord_offsetD = 0;
 593
 594     outeriter        = 0;
 595     inneriter        = 0;
 596
 597     for(iidx=0;iidx<4*DIM;iidx++)
 598     {
 599         scratch[iidx] = 0.0;
 600     }
 601
 602     /* Start outer loop over neighborlists */
 603     for(iidx=0; iidx<nri; iidx++)
 604     {
 605         /* Load shift vector for this list */
 606         i_shift_offset   = DIM*shiftidx[iidx];
 607
 608         /* Load limits for loop over neighbors */
 609         j_index_start    = jindex[iidx];
 610         j_index_end      = jindex[iidx+1];
 611
 612         /* Get outer coordinate index */
 613         inr              = iinr[iidx];
 614         i_coord_offset   = DIM*inr;
 615
 616         /* Load i particle coords and add shift vector */
 617         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 618
 619         fix0             = _mm256_setzero_pd();
 620         fiy0             = _mm256_setzero_pd();
 621         fiz0             = _mm256_setzero_pd();
 622
 623         /* Load parameters for i particles */
 624         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 625         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 626         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 627
 628         /* Start inner kernel loop */
 629         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 630         {
 631
 632             /* Get j neighbor index, and coordinate index */
 633             jnrA             = jjnr[jidx];
 634             jnrB             = jjnr[jidx+1];
 635             jnrC             = jjnr[jidx+2];
 636             jnrD             = jjnr[jidx+3];
 637             j_coord_offsetA  = DIM*jnrA;
 638             j_coord_offsetB  = DIM*jnrB;
 639             j_coord_offsetC  = DIM*jnrC;
 640             j_coord_offsetD  = DIM*jnrD;
 641
 642             /* load j atom coordinates */
 643             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 644                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 645                                                  &jx0,&jy0,&jz0);
 646
 647             /* Calculate displacement vector */
 648             dx00             = _mm256_sub_pd(ix0,jx0);
 649             dy00             = _mm256_sub_pd(iy0,jy0);
 650             dz00             = _mm256_sub_pd(iz0,jz0);
 651
 652             /* Calculate squared distance and things based on it */
 653             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 654
 655             rinv00           = avx256_invsqrt_d(rsq00);
 656
 657             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 658
 659             /* Load parameters for j particles */
 660             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 661                                                                  charge+jnrC+0,charge+jnrD+0);
 662             vdwjidx0A        = 2*vdwtype[jnrA+0];
 663             vdwjidx0B        = 2*vdwtype[jnrB+0];
 664             vdwjidx0C        = 2*vdwtype[jnrC+0];
 665             vdwjidx0D        = 2*vdwtype[jnrD+0];
 666
 667             /**************************
 668              * CALCULATE INTERACTIONS *
 669              **************************/
 670
 671             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 672             {
 673
 674             r00              = _mm256_mul_pd(rsq00,rinv00);
 675
 676             /* Compute parameters for interactions between i and j atoms */
 677             qq00             = _mm256_mul_pd(iq0,jq0);
 678             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 679                                             vdwioffsetptr0+vdwjidx0B,
 680                                             vdwioffsetptr0+vdwjidx0C,
 681                                             vdwioffsetptr0+vdwjidx0D,
 682                                             &c6_00,&c12_00);
 683
 684             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 685                                                                   vdwgridioffsetptr0+vdwjidx0B,
 686                                                                   vdwgridioffsetptr0+vdwjidx0C,
 687                                                                   vdwgridioffsetptr0+vdwjidx0D);
 688
 689             /* EWALD ELECTROSTATICS */
 690
 691             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 692             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 693             ewitab           = _mm256_cvttpd_epi32(ewrt);
 694             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 695             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 696                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 697                                             &ewtabF,&ewtabFn);
 698             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 699             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 700
 701             /* Analytical LJ-PME */
 702             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 703             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 704             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 705             exponent         = avx256_exp_d(ewcljrsq);
 706             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 707             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 708             /* f6A = 6 * C6grid * (1 - poly) */
 709             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 710             /* f6B = C6grid * exponent * beta^6 */
 711             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 712             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 713             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);
 714
 715             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 716
 717             fscal            = _mm256_add_pd(felec,fvdw);
 718
 719             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 720
 721             /* Calculate temporary vectorial force */
 722             tx               = _mm256_mul_pd(fscal,dx00);
 723             ty               = _mm256_mul_pd(fscal,dy00);
 724             tz               = _mm256_mul_pd(fscal,dz00);
 725
 726             /* Update vectorial force */
 727             fix0             = _mm256_add_pd(fix0,tx);
 728             fiy0             = _mm256_add_pd(fiy0,ty);
 729             fiz0             = _mm256_add_pd(fiz0,tz);
 730
 731             fjptrA             = f+j_coord_offsetA;
 732             fjptrB             = f+j_coord_offsetB;
 733             fjptrC             = f+j_coord_offsetC;
 734             fjptrD             = f+j_coord_offsetD;
 735             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 736
 737             }
 738
 739             /* Inner loop uses 62 flops */
 740         }
 741
 742         if(jidx<j_index_end)
 743         {
 744
 745             /* Get j neighbor index, and coordinate index */
 746             jnrlistA         = jjnr[jidx];
 747             jnrlistB         = jjnr[jidx+1];
 748             jnrlistC         = jjnr[jidx+2];
 749             jnrlistD         = jjnr[jidx+3];
 750             /* Sign of each element will be negative for non-real atoms.
 751              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 752              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 753              */
 754             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 755
 756             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 757             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 758             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 759
 760             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 761             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 762             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 763             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 764             j_coord_offsetA  = DIM*jnrA;
 765             j_coord_offsetB  = DIM*jnrB;
 766             j_coord_offsetC  = DIM*jnrC;
 767             j_coord_offsetD  = DIM*jnrD;
 768
 769             /* load j atom coordinates */
 770             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 771                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 772                                                  &jx0,&jy0,&jz0);
 773
 774             /* Calculate displacement vector */
 775             dx00             = _mm256_sub_pd(ix0,jx0);
 776             dy00             = _mm256_sub_pd(iy0,jy0);
 777             dz00             = _mm256_sub_pd(iz0,jz0);
 778
 779             /* Calculate squared distance and things based on it */
 780             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 781
 782             rinv00           = avx256_invsqrt_d(rsq00);
 783
 784             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 785
 786             /* Load parameters for j particles */
 787             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 788                                                                  charge+jnrC+0,charge+jnrD+0);
 789             vdwjidx0A        = 2*vdwtype[jnrA+0];
 790             vdwjidx0B        = 2*vdwtype[jnrB+0];
 791             vdwjidx0C        = 2*vdwtype[jnrC+0];
 792             vdwjidx0D        = 2*vdwtype[jnrD+0];
 793
 794             /**************************
 795              * CALCULATE INTERACTIONS *
 796              **************************/
 797
 798             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 799             {
 800
 801             r00              = _mm256_mul_pd(rsq00,rinv00);
 802             r00              = _mm256_andnot_pd(dummy_mask,r00);
 803
 804             /* Compute parameters for interactions between i and j atoms */
 805             qq00             = _mm256_mul_pd(iq0,jq0);
 806             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 807                                             vdwioffsetptr0+vdwjidx0B,
 808                                             vdwioffsetptr0+vdwjidx0C,
 809                                             vdwioffsetptr0+vdwjidx0D,
 810                                             &c6_00,&c12_00);
 811
 812             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 813                                                                   vdwgridioffsetptr0+vdwjidx0B,
 814                                                                   vdwgridioffsetptr0+vdwjidx0C,
 815                                                                   vdwgridioffsetptr0+vdwjidx0D);
 816
 817             /* EWALD ELECTROSTATICS */
 818
 819             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 820             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 821             ewitab           = _mm256_cvttpd_epi32(ewrt);
 822             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 823             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 824                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 825                                             &ewtabF,&ewtabFn);
 826             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 827             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 828
 829             /* Analytical LJ-PME */
 830             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 831             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 832             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 833             exponent         = avx256_exp_d(ewcljrsq);
 834             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 835             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 836             /* f6A = 6 * C6grid * (1 - poly) */
 837             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 838             /* f6B = C6grid * exponent * beta^6 */
 839             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 840             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 841             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);
 842
 843             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 844
 845             fscal            = _mm256_add_pd(felec,fvdw);
 846
 847             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 848
 849             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm256_mul_pd(fscal,dx00);
 853             ty               = _mm256_mul_pd(fscal,dy00);
 854             tz               = _mm256_mul_pd(fscal,dz00);
 855
 856             /* Update vectorial force */
 857             fix0             = _mm256_add_pd(fix0,tx);
 858             fiy0             = _mm256_add_pd(fiy0,ty);
 859             fiz0             = _mm256_add_pd(fiz0,tz);
 860
 861             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 862             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 863             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 864             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 865             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 866
 867             }
 868
 869             /* Inner loop uses 63 flops */
 870         }
 871
 872         /* End of innermost loop */
 873
 874         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 875                                                  f+i_coord_offset,fshift+i_shift_offset);
 876
 877         /* Increment number of inner iterations */
 878         inneriter                  += j_index_end - j_index_start;
 879
 880         /* Outer loop uses 7 flops */
 881     }
 882
 883     /* Increment number of outer iterations */
 884     outeriter        += nri;
 885
 886     /* Update outer/inner flops */
 887
 888     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*63);
 889 }