src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     real *           vdwgridioffsetptr0;
  86     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  88     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
  97     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
  98     __m256d           c6grid_00;
  99     real             *vdwgridparam;
 100     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 101     __m256d           one_half  = _mm256_set1_pd(0.5);
 102     __m256d           minus_one = _mm256_set1_pd(-1.0);
 103     __m128i          ewitab;
 104     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 105     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 106     real             *ewtab;
 107     __m256d          dummy_mask,cutoff_mask;
 108     __m128           tmpmask0,tmpmask1;
 109     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 110     __m256d          one     = _mm256_set1_pd(1.0);
 111     __m256d          two     = _mm256_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm256_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128     vdwgridparam     = fr->ljpme_c6grid;
 129     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 130     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 131     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 132
 133     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 134     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 135     beta2            = _mm256_mul_pd(beta,beta);
 136     beta3            = _mm256_mul_pd(beta,beta2);
 137
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 141
 142     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 143     rcutoff_scalar   = fr->rcoulomb;
 144     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 145     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 146
 147     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 148     rvdw             = _mm256_set1_pd(fr->rvdw);
 149
 150     /* Avoid stupid compiler warnings */
 151     jnrA = jnrB = jnrC = jnrD = 0;
 152     j_coord_offsetA = 0;
 153     j_coord_offsetB = 0;
 154     j_coord_offsetC = 0;
 155     j_coord_offsetD = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 181
 182         fix0             = _mm256_setzero_pd();
 183         fiy0             = _mm256_setzero_pd();
 184         fiz0             = _mm256_setzero_pd();
 185
 186         /* Load parameters for i particles */
 187         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 188         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 189         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm256_setzero_pd();
 193         vvdwsum          = _mm256_setzero_pd();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208
 209             /* load j atom coordinates */
 210             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 212                                                  &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm256_sub_pd(ix0,jx0);
 216             dy00             = _mm256_sub_pd(iy0,jy0);
 217             dz00             = _mm256_sub_pd(iz0,jz0);
 218
 219             /* Calculate squared distance and things based on it */
 220             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 221
 222             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 223
 224             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 228                                                                  charge+jnrC+0,charge+jnrD+0);
 229             vdwjidx0A        = 2*vdwtype[jnrA+0];
 230             vdwjidx0B        = 2*vdwtype[jnrB+0];
 231             vdwjidx0C        = 2*vdwtype[jnrC+0];
 232             vdwjidx0D        = 2*vdwtype[jnrD+0];
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 239             {
 240
 241             r00              = _mm256_mul_pd(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm256_mul_pd(iq0,jq0);
 245             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 246                                             vdwioffsetptr0+vdwjidx0B,
 247                                             vdwioffsetptr0+vdwjidx0C,
 248                                             vdwioffsetptr0+vdwjidx0D,
 249                                             &c6_00,&c12_00);
 250
 251             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 252                                                                   vdwgridioffsetptr0+vdwjidx0B,
 253                                                                   vdwgridioffsetptr0+vdwjidx0C,
 254                                                                   vdwgridioffsetptr0+vdwjidx0D);
 255
 256             /* EWALD ELECTROSTATICS */
 257
 258             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 259             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 260             ewitab           = _mm256_cvttpd_epi32(ewrt);
 261             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 262             ewitab           = _mm_slli_epi32(ewitab,2);
 263             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 264             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 265             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 266             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 267             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 268             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 269             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 270             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 271             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 272
 273             /* Analytical LJ-PME */
 274             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 275             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 276             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 277             exponent         = gmx_simd_exp_d(ewcljrsq);
 278             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 279             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 280             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 281             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 282             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 283             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) ,
 284                                           _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));
 285             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 286             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);
 287
 288             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             velec            = _mm256_and_pd(velec,cutoff_mask);
 292             velecsum         = _mm256_add_pd(velecsum,velec);
 293             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 294             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 295
 296             fscal            = _mm256_add_pd(felec,fvdw);
 297
 298             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm256_mul_pd(fscal,dx00);
 302             ty               = _mm256_mul_pd(fscal,dy00);
 303             tz               = _mm256_mul_pd(fscal,dz00);
 304
 305             /* Update vectorial force */
 306             fix0             = _mm256_add_pd(fix0,tx);
 307             fiy0             = _mm256_add_pd(fiy0,ty);
 308             fiz0             = _mm256_add_pd(fiz0,tz);
 309
 310             fjptrA             = f+j_coord_offsetA;
 311             fjptrB             = f+j_coord_offsetB;
 312             fjptrC             = f+j_coord_offsetC;
 313             fjptrD             = f+j_coord_offsetD;
 314             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 315
 316             }
 317
 318             /* Inner loop uses 82 flops */
 319         }
 320
 321         if(jidx<j_index_end)
 322         {
 323
 324             /* Get j neighbor index, and coordinate index */
 325             jnrlistA         = jjnr[jidx];
 326             jnrlistB         = jjnr[jidx+1];
 327             jnrlistC         = jjnr[jidx+2];
 328             jnrlistD         = jjnr[jidx+3];
 329             /* Sign of each element will be negative for non-real atoms.
 330              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 331              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 332              */
 333             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 334
 335             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 336             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 337             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 338
 339             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 340             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 341             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 342             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 343             j_coord_offsetA  = DIM*jnrA;
 344             j_coord_offsetB  = DIM*jnrB;
 345             j_coord_offsetC  = DIM*jnrC;
 346             j_coord_offsetD  = DIM*jnrD;
 347
 348             /* load j atom coordinates */
 349             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 350                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 351                                                  &jx0,&jy0,&jz0);
 352
 353             /* Calculate displacement vector */
 354             dx00             = _mm256_sub_pd(ix0,jx0);
 355             dy00             = _mm256_sub_pd(iy0,jy0);
 356             dz00             = _mm256_sub_pd(iz0,jz0);
 357
 358             /* Calculate squared distance and things based on it */
 359             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 360
 361             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 362
 363             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 364
 365             /* Load parameters for j particles */
 366             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 367                                                                  charge+jnrC+0,charge+jnrD+0);
 368             vdwjidx0A        = 2*vdwtype[jnrA+0];
 369             vdwjidx0B        = 2*vdwtype[jnrB+0];
 370             vdwjidx0C        = 2*vdwtype[jnrC+0];
 371             vdwjidx0D        = 2*vdwtype[jnrD+0];
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 378             {
 379
 380             r00              = _mm256_mul_pd(rsq00,rinv00);
 381             r00              = _mm256_andnot_pd(dummy_mask,r00);
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq00             = _mm256_mul_pd(iq0,jq0);
 385             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 386                                             vdwioffsetptr0+vdwjidx0B,
 387                                             vdwioffsetptr0+vdwjidx0C,
 388                                             vdwioffsetptr0+vdwjidx0D,
 389                                             &c6_00,&c12_00);
 390
 391             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 392                                                                   vdwgridioffsetptr0+vdwjidx0B,
 393                                                                   vdwgridioffsetptr0+vdwjidx0C,
 394                                                                   vdwgridioffsetptr0+vdwjidx0D);
 395
 396             /* EWALD ELECTROSTATICS */
 397
 398             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 399             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 400             ewitab           = _mm256_cvttpd_epi32(ewrt);
 401             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 402             ewitab           = _mm_slli_epi32(ewitab,2);
 403             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 404             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 405             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 406             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 407             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 408             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 409             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 410             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 411             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 412
 413             /* Analytical LJ-PME */
 414             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 415             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 416             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 417             exponent         = gmx_simd_exp_d(ewcljrsq);
 418             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 419             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 420             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 421             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 422             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 423             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) ,
 424                                           _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));
 425             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 426             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);
 427
 428             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velec            = _mm256_and_pd(velec,cutoff_mask);
 432             velec            = _mm256_andnot_pd(dummy_mask,velec);
 433             velecsum         = _mm256_add_pd(velecsum,velec);
 434             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 435             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 436             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 437
 438             fscal            = _mm256_add_pd(felec,fvdw);
 439
 440             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 441
 442             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 443
 444             /* Calculate temporary vectorial force */
 445             tx               = _mm256_mul_pd(fscal,dx00);
 446             ty               = _mm256_mul_pd(fscal,dy00);
 447             tz               = _mm256_mul_pd(fscal,dz00);
 448
 449             /* Update vectorial force */
 450             fix0             = _mm256_add_pd(fix0,tx);
 451             fiy0             = _mm256_add_pd(fiy0,ty);
 452             fiz0             = _mm256_add_pd(fiz0,tz);
 453
 454             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 455             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 456             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 457             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 458             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 459
 460             }
 461
 462             /* Inner loop uses 83 flops */
 463         }
 464
 465         /* End of innermost loop */
 466
 467         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 468                                                  f+i_coord_offset,fshift+i_shift_offset);
 469
 470         ggid                        = gid[iidx];
 471         /* Update potential energies */
 472         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 473         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 474
 475         /* Increment number of inner iterations */
 476         inneriter                  += j_index_end - j_index_start;
 477
 478         /* Outer loop uses 9 flops */
 479     }
 480
 481     /* Increment number of outer iterations */
 482     outeriter        += nri;
 483
 484     /* Update outer/inner flops */
 485
 486     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*83);
 487 }
 488 /*
 489  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_double
 490  * Electrostatics interaction: Ewald
 491  * VdW interaction:            LJEwald
 492  * Geometry:                   Particle-Particle
 493  * Calculate force/pot:        Force
 494  */
 495 void
 496 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_avx_256_double
 497                     (t_nblist                    * gmx_restrict       nlist,
 498                      rvec                        * gmx_restrict          xx,
 499                      rvec                        * gmx_restrict          ff,
 500                      t_forcerec                  * gmx_restrict          fr,
 501                      t_mdatoms                   * gmx_restrict     mdatoms,
 502                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 503                      t_nrnb                      * gmx_restrict        nrnb)
 504 {
 505     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 506      * just 0 for non-waters.
 507      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 508      * jnr indices corresponding to data put in the four positions in the SIMD register.
 509      */
 510     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 511     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 512     int              jnrA,jnrB,jnrC,jnrD;
 513     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 514     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 515     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 516     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 517     real             rcutoff_scalar;
 518     real             *shiftvec,*fshift,*x,*f;
 519     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 520     real             scratch[4*DIM];
 521     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 522     real *           vdwioffsetptr0;
 523     real *           vdwgridioffsetptr0;
 524     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 525     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 526     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 527     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 528     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 529     real             *charge;
 530     int              nvdwtype;
 531     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 532     int              *vdwtype;
 533     real             *vdwparam;
 534     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 535     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 536     __m256d           c6grid_00;
 537     real             *vdwgridparam;
 538     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 539     __m256d           one_half  = _mm256_set1_pd(0.5);
 540     __m256d           minus_one = _mm256_set1_pd(-1.0);
 541     __m128i          ewitab;
 542     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 543     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 544     real             *ewtab;
 545     __m256d          dummy_mask,cutoff_mask;
 546     __m128           tmpmask0,tmpmask1;
 547     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 548     __m256d          one     = _mm256_set1_pd(1.0);
 549     __m256d          two     = _mm256_set1_pd(2.0);
 550     x                = xx[0];
 551     f                = ff[0];
 552
 553     nri              = nlist->nri;
 554     iinr             = nlist->iinr;
 555     jindex           = nlist->jindex;
 556     jjnr             = nlist->jjnr;
 557     shiftidx         = nlist->shift;
 558     gid              = nlist->gid;
 559     shiftvec         = fr->shift_vec[0];
 560     fshift           = fr->fshift[0];
 561     facel            = _mm256_set1_pd(fr->epsfac);
 562     charge           = mdatoms->chargeA;
 563     nvdwtype         = fr->ntype;
 564     vdwparam         = fr->nbfp;
 565     vdwtype          = mdatoms->typeA;
 566     vdwgridparam     = fr->ljpme_c6grid;
 567     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 568     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 569     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 570
 571     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 572     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 573     beta2            = _mm256_mul_pd(beta,beta);
 574     beta3            = _mm256_mul_pd(beta,beta2);
 575
 576     ewtab            = fr->ic->tabq_coul_F;
 577     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 578     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 579
 580     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 581     rcutoff_scalar   = fr->rcoulomb;
 582     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 583     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 584
 585     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 586     rvdw             = _mm256_set1_pd(fr->rvdw);
 587
 588     /* Avoid stupid compiler warnings */
 589     jnrA = jnrB = jnrC = jnrD = 0;
 590     j_coord_offsetA = 0;
 591     j_coord_offsetB = 0;
 592     j_coord_offsetC = 0;
 593     j_coord_offsetD = 0;
 594
 595     outeriter        = 0;
 596     inneriter        = 0;
 597
 598     for(iidx=0;iidx<4*DIM;iidx++)
 599     {
 600         scratch[iidx] = 0.0;
 601     }
 602
 603     /* Start outer loop over neighborlists */
 604     for(iidx=0; iidx<nri; iidx++)
 605     {
 606         /* Load shift vector for this list */
 607         i_shift_offset   = DIM*shiftidx[iidx];
 608
 609         /* Load limits for loop over neighbors */
 610         j_index_start    = jindex[iidx];
 611         j_index_end      = jindex[iidx+1];
 612
 613         /* Get outer coordinate index */
 614         inr              = iinr[iidx];
 615         i_coord_offset   = DIM*inr;
 616
 617         /* Load i particle coords and add shift vector */
 618         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 619
 620         fix0             = _mm256_setzero_pd();
 621         fiy0             = _mm256_setzero_pd();
 622         fiz0             = _mm256_setzero_pd();
 623
 624         /* Load parameters for i particles */
 625         iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 626         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 627         vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 628
 629         /* Start inner kernel loop */
 630         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 631         {
 632
 633             /* Get j neighbor index, and coordinate index */
 634             jnrA             = jjnr[jidx];
 635             jnrB             = jjnr[jidx+1];
 636             jnrC             = jjnr[jidx+2];
 637             jnrD             = jjnr[jidx+3];
 638             j_coord_offsetA  = DIM*jnrA;
 639             j_coord_offsetB  = DIM*jnrB;
 640             j_coord_offsetC  = DIM*jnrC;
 641             j_coord_offsetD  = DIM*jnrD;
 642
 643             /* load j atom coordinates */
 644             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 645                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 646                                                  &jx0,&jy0,&jz0);
 647
 648             /* Calculate displacement vector */
 649             dx00             = _mm256_sub_pd(ix0,jx0);
 650             dy00             = _mm256_sub_pd(iy0,jy0);
 651             dz00             = _mm256_sub_pd(iz0,jz0);
 652
 653             /* Calculate squared distance and things based on it */
 654             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 655
 656             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 657
 658             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 659
 660             /* Load parameters for j particles */
 661             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 662                                                                  charge+jnrC+0,charge+jnrD+0);
 663             vdwjidx0A        = 2*vdwtype[jnrA+0];
 664             vdwjidx0B        = 2*vdwtype[jnrB+0];
 665             vdwjidx0C        = 2*vdwtype[jnrC+0];
 666             vdwjidx0D        = 2*vdwtype[jnrD+0];
 667
 668             /**************************
 669              * CALCULATE INTERACTIONS *
 670              **************************/
 671
 672             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 673             {
 674
 675             r00              = _mm256_mul_pd(rsq00,rinv00);
 676
 677             /* Compute parameters for interactions between i and j atoms */
 678             qq00             = _mm256_mul_pd(iq0,jq0);
 679             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 680                                             vdwioffsetptr0+vdwjidx0B,
 681                                             vdwioffsetptr0+vdwjidx0C,
 682                                             vdwioffsetptr0+vdwjidx0D,
 683                                             &c6_00,&c12_00);
 684
 685             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 686                                                                   vdwgridioffsetptr0+vdwjidx0B,
 687                                                                   vdwgridioffsetptr0+vdwjidx0C,
 688                                                                   vdwgridioffsetptr0+vdwjidx0D);
 689
 690             /* EWALD ELECTROSTATICS */
 691
 692             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 693             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 694             ewitab           = _mm256_cvttpd_epi32(ewrt);
 695             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 696             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 697                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 698                                             &ewtabF,&ewtabFn);
 699             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 700             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 701
 702             /* Analytical LJ-PME */
 703             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 704             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 705             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 706             exponent         = gmx_simd_exp_d(ewcljrsq);
 707             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 708             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 709             /* f6A = 6 * C6grid * (1 - poly) */
 710             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 711             /* f6B = C6grid * exponent * beta^6 */
 712             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 713             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 714             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);
 715
 716             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 717
 718             fscal            = _mm256_add_pd(felec,fvdw);
 719
 720             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 721
 722             /* Calculate temporary vectorial force */
 723             tx               = _mm256_mul_pd(fscal,dx00);
 724             ty               = _mm256_mul_pd(fscal,dy00);
 725             tz               = _mm256_mul_pd(fscal,dz00);
 726
 727             /* Update vectorial force */
 728             fix0             = _mm256_add_pd(fix0,tx);
 729             fiy0             = _mm256_add_pd(fiy0,ty);
 730             fiz0             = _mm256_add_pd(fiz0,tz);
 731
 732             fjptrA             = f+j_coord_offsetA;
 733             fjptrB             = f+j_coord_offsetB;
 734             fjptrC             = f+j_coord_offsetC;
 735             fjptrD             = f+j_coord_offsetD;
 736             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 737
 738             }
 739
 740             /* Inner loop uses 62 flops */
 741         }
 742
 743         if(jidx<j_index_end)
 744         {
 745
 746             /* Get j neighbor index, and coordinate index */
 747             jnrlistA         = jjnr[jidx];
 748             jnrlistB         = jjnr[jidx+1];
 749             jnrlistC         = jjnr[jidx+2];
 750             jnrlistD         = jjnr[jidx+3];
 751             /* Sign of each element will be negative for non-real atoms.
 752              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 753              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 754              */
 755             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 756
 757             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 758             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 759             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 760
 761             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 762             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 763             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 764             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 765             j_coord_offsetA  = DIM*jnrA;
 766             j_coord_offsetB  = DIM*jnrB;
 767             j_coord_offsetC  = DIM*jnrC;
 768             j_coord_offsetD  = DIM*jnrD;
 769
 770             /* load j atom coordinates */
 771             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 772                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 773                                                  &jx0,&jy0,&jz0);
 774
 775             /* Calculate displacement vector */
 776             dx00             = _mm256_sub_pd(ix0,jx0);
 777             dy00             = _mm256_sub_pd(iy0,jy0);
 778             dz00             = _mm256_sub_pd(iz0,jz0);
 779
 780             /* Calculate squared distance and things based on it */
 781             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 782
 783             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 784
 785             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 786
 787             /* Load parameters for j particles */
 788             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 789                                                                  charge+jnrC+0,charge+jnrD+0);
 790             vdwjidx0A        = 2*vdwtype[jnrA+0];
 791             vdwjidx0B        = 2*vdwtype[jnrB+0];
 792             vdwjidx0C        = 2*vdwtype[jnrC+0];
 793             vdwjidx0D        = 2*vdwtype[jnrD+0];
 794
 795             /**************************
 796              * CALCULATE INTERACTIONS *
 797              **************************/
 798
 799             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 800             {
 801
 802             r00              = _mm256_mul_pd(rsq00,rinv00);
 803             r00              = _mm256_andnot_pd(dummy_mask,r00);
 804
 805             /* Compute parameters for interactions between i and j atoms */
 806             qq00             = _mm256_mul_pd(iq0,jq0);
 807             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 808                                             vdwioffsetptr0+vdwjidx0B,
 809                                             vdwioffsetptr0+vdwjidx0C,
 810                                             vdwioffsetptr0+vdwjidx0D,
 811                                             &c6_00,&c12_00);
 812
 813             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 814                                                                   vdwgridioffsetptr0+vdwjidx0B,
 815                                                                   vdwgridioffsetptr0+vdwjidx0C,
 816                                                                   vdwgridioffsetptr0+vdwjidx0D);
 817
 818             /* EWALD ELECTROSTATICS */
 819
 820             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 821             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 822             ewitab           = _mm256_cvttpd_epi32(ewrt);
 823             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 824             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 825                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 826                                             &ewtabF,&ewtabFn);
 827             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 828             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 829
 830             /* Analytical LJ-PME */
 831             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 832             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 833             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 834             exponent         = gmx_simd_exp_d(ewcljrsq);
 835             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 836             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 837             /* f6A = 6 * C6grid * (1 - poly) */
 838             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 839             /* f6B = C6grid * exponent * beta^6 */
 840             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 841             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 842             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);
 843
 844             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 845
 846             fscal            = _mm256_add_pd(felec,fvdw);
 847
 848             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 849
 850             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm256_mul_pd(fscal,dx00);
 854             ty               = _mm256_mul_pd(fscal,dy00);
 855             tz               = _mm256_mul_pd(fscal,dz00);
 856
 857             /* Update vectorial force */
 858             fix0             = _mm256_add_pd(fix0,tx);
 859             fiy0             = _mm256_add_pd(fiy0,ty);
 860             fiz0             = _mm256_add_pd(fiz0,tz);
 861
 862             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 863             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 864             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 865             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 866             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 867
 868             }
 869
 870             /* Inner loop uses 63 flops */
 871         }
 872
 873         /* End of innermost loop */
 874
 875         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 876                                                  f+i_coord_offset,fshift+i_shift_offset);
 877
 878         /* Increment number of inner iterations */
 879         inneriter                  += j_index_end - j_index_start;
 880
 881         /* Outer loop uses 7 flops */
 882     }
 883
 884     /* Increment number of outer iterations */
 885     outeriter        += nri;
 886
 887     /* Update outer/inner flops */
 888
 889     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*63);
 890 }