src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_avx_256_single.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_single 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_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  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              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     real *           vdwgridioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     real *           vdwgridioffsetptr1;
  90     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  91     real *           vdwioffsetptr2;
  92     real *           vdwgridioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256           c6grid_00;
 108     __m256           c6grid_10;
 109     __m256           c6grid_20;
 110     real             *vdwgridparam;
 111     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 112     __m256           one_half  = _mm256_set1_ps(0.5);
 113     __m256           minus_one = _mm256_set1_ps(-1.0);
 114     __m256i          ewitab;
 115     __m128i          ewitab_lo,ewitab_hi;
 116     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 117     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 118     real             *ewtab;
 119     __m256           dummy_mask,cutoff_mask;
 120     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 121     __m256           one     = _mm256_set1_ps(1.0);
 122     __m256           two     = _mm256_set1_ps(2.0);
 123     x                = xx[0];
 124     f                = ff[0];
 125
 126     nri              = nlist->nri;
 127     iinr             = nlist->iinr;
 128     jindex           = nlist->jindex;
 129     jjnr             = nlist->jjnr;
 130     shiftidx         = nlist->shift;
 131     gid              = nlist->gid;
 132     shiftvec         = fr->shift_vec[0];
 133     fshift           = fr->fshift[0];
 134     facel            = _mm256_set1_ps(fr->ic->epsfac);
 135     charge           = mdatoms->chargeA;
 136     nvdwtype         = fr->ntype;
 137     vdwparam         = fr->nbfp;
 138     vdwtype          = mdatoms->typeA;
 139     vdwgridparam     = fr->ljpme_c6grid;
 140     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 141     ewclj            = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
 142     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 143
 144     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 145     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 146     beta2            = _mm256_mul_ps(beta,beta);
 147     beta3            = _mm256_mul_ps(beta,beta2);
 148
 149     ewtab            = fr->ic->tabq_coul_FDV0;
 150     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 151     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 152
 153     /* Setup water-specific parameters */
 154     inr              = nlist->iinr[0];
 155     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 156     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 157     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 158     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 159     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 160
 161     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 162     rcutoff_scalar   = fr->ic->rcoulomb;
 163     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 164     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 165
 166     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 167     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 168
 169     /* Avoid stupid compiler warnings */
 170     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 171     j_coord_offsetA = 0;
 172     j_coord_offsetB = 0;
 173     j_coord_offsetC = 0;
 174     j_coord_offsetD = 0;
 175     j_coord_offsetE = 0;
 176     j_coord_offsetF = 0;
 177     j_coord_offsetG = 0;
 178     j_coord_offsetH = 0;
 179
 180     outeriter        = 0;
 181     inneriter        = 0;
 182
 183     for(iidx=0;iidx<4*DIM;iidx++)
 184     {
 185         scratch[iidx] = 0.0;
 186     }
 187
 188     /* Start outer loop over neighborlists */
 189     for(iidx=0; iidx<nri; iidx++)
 190     {
 191         /* Load shift vector for this list */
 192         i_shift_offset   = DIM*shiftidx[iidx];
 193
 194         /* Load limits for loop over neighbors */
 195         j_index_start    = jindex[iidx];
 196         j_index_end      = jindex[iidx+1];
 197
 198         /* Get outer coordinate index */
 199         inr              = iinr[iidx];
 200         i_coord_offset   = DIM*inr;
 201
 202         /* Load i particle coords and add shift vector */
 203         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 204                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 205
 206         fix0             = _mm256_setzero_ps();
 207         fiy0             = _mm256_setzero_ps();
 208         fiz0             = _mm256_setzero_ps();
 209         fix1             = _mm256_setzero_ps();
 210         fiy1             = _mm256_setzero_ps();
 211         fiz1             = _mm256_setzero_ps();
 212         fix2             = _mm256_setzero_ps();
 213         fiy2             = _mm256_setzero_ps();
 214         fiz2             = _mm256_setzero_ps();
 215
 216         /* Reset potential sums */
 217         velecsum         = _mm256_setzero_ps();
 218         vvdwsum          = _mm256_setzero_ps();
 219
 220         /* Start inner kernel loop */
 221         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 222         {
 223
 224             /* Get j neighbor index, and coordinate index */
 225             jnrA             = jjnr[jidx];
 226             jnrB             = jjnr[jidx+1];
 227             jnrC             = jjnr[jidx+2];
 228             jnrD             = jjnr[jidx+3];
 229             jnrE             = jjnr[jidx+4];
 230             jnrF             = jjnr[jidx+5];
 231             jnrG             = jjnr[jidx+6];
 232             jnrH             = jjnr[jidx+7];
 233             j_coord_offsetA  = DIM*jnrA;
 234             j_coord_offsetB  = DIM*jnrB;
 235             j_coord_offsetC  = DIM*jnrC;
 236             j_coord_offsetD  = DIM*jnrD;
 237             j_coord_offsetE  = DIM*jnrE;
 238             j_coord_offsetF  = DIM*jnrF;
 239             j_coord_offsetG  = DIM*jnrG;
 240             j_coord_offsetH  = DIM*jnrH;
 241
 242             /* load j atom coordinates */
 243             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 244                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 245                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 246                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 247                                                  &jx0,&jy0,&jz0);
 248
 249             /* Calculate displacement vector */
 250             dx00             = _mm256_sub_ps(ix0,jx0);
 251             dy00             = _mm256_sub_ps(iy0,jy0);
 252             dz00             = _mm256_sub_ps(iz0,jz0);
 253             dx10             = _mm256_sub_ps(ix1,jx0);
 254             dy10             = _mm256_sub_ps(iy1,jy0);
 255             dz10             = _mm256_sub_ps(iz1,jz0);
 256             dx20             = _mm256_sub_ps(ix2,jx0);
 257             dy20             = _mm256_sub_ps(iy2,jy0);
 258             dz20             = _mm256_sub_ps(iz2,jz0);
 259
 260             /* Calculate squared distance and things based on it */
 261             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 262             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 263             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 264
 265             rinv00           = avx256_invsqrt_f(rsq00);
 266             rinv10           = avx256_invsqrt_f(rsq10);
 267             rinv20           = avx256_invsqrt_f(rsq20);
 268
 269             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 270             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 271             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 272
 273             /* Load parameters for j particles */
 274             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 275                                                                  charge+jnrC+0,charge+jnrD+0,
 276                                                                  charge+jnrE+0,charge+jnrF+0,
 277                                                                  charge+jnrG+0,charge+jnrH+0);
 278             vdwjidx0A        = 2*vdwtype[jnrA+0];
 279             vdwjidx0B        = 2*vdwtype[jnrB+0];
 280             vdwjidx0C        = 2*vdwtype[jnrC+0];
 281             vdwjidx0D        = 2*vdwtype[jnrD+0];
 282             vdwjidx0E        = 2*vdwtype[jnrE+0];
 283             vdwjidx0F        = 2*vdwtype[jnrF+0];
 284             vdwjidx0G        = 2*vdwtype[jnrG+0];
 285             vdwjidx0H        = 2*vdwtype[jnrH+0];
 286
 287             fjx0             = _mm256_setzero_ps();
 288             fjy0             = _mm256_setzero_ps();
 289             fjz0             = _mm256_setzero_ps();
 290
 291             /**************************
 292              * CALCULATE INTERACTIONS *
 293              **************************/
 294
 295             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 296             {
 297
 298             r00              = _mm256_mul_ps(rsq00,rinv00);
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq00             = _mm256_mul_ps(iq0,jq0);
 302             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 303                                             vdwioffsetptr0+vdwjidx0B,
 304                                             vdwioffsetptr0+vdwjidx0C,
 305                                             vdwioffsetptr0+vdwjidx0D,
 306                                             vdwioffsetptr0+vdwjidx0E,
 307                                             vdwioffsetptr0+vdwjidx0F,
 308                                             vdwioffsetptr0+vdwjidx0G,
 309                                             vdwioffsetptr0+vdwjidx0H,
 310                                             &c6_00,&c12_00);
 311
 312             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 313                                                                   vdwgridioffsetptr0+vdwjidx0B,
 314                                                                   vdwgridioffsetptr0+vdwjidx0C,
 315                                                                   vdwgridioffsetptr0+vdwjidx0D,
 316                                                                   vdwgridioffsetptr0+vdwjidx0E,
 317                                                                   vdwgridioffsetptr0+vdwjidx0F,
 318                                                                   vdwgridioffsetptr0+vdwjidx0G,
 319                                                                   vdwgridioffsetptr0+vdwjidx0H);
 320
 321             /* EWALD ELECTROSTATICS */
 322
 323             /* Analytical PME correction */
 324             zeta2            = _mm256_mul_ps(beta2,rsq00);
 325             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 326             pmecorrF         = avx256_pmecorrF_f(zeta2);
 327             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 328             felec            = _mm256_mul_ps(qq00,felec);
 329             pmecorrV         = avx256_pmecorrV_f(zeta2);
 330             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 331             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 332             velec            = _mm256_mul_ps(qq00,velec);
 333
 334             /* Analytical LJ-PME */
 335             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 336             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 337             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 338             exponent         = avx256_exp_f(ewcljrsq);
 339             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 340             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 341             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 342             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 343             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 344             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 345                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 346             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 347             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
 348
 349             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 350
 351             /* Update potential sum for this i atom from the interaction with this j atom. */
 352             velec            = _mm256_and_ps(velec,cutoff_mask);
 353             velecsum         = _mm256_add_ps(velecsum,velec);
 354             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 355             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 356
 357             fscal            = _mm256_add_ps(felec,fvdw);
 358
 359             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 360
 361             /* Calculate temporary vectorial force */
 362             tx               = _mm256_mul_ps(fscal,dx00);
 363             ty               = _mm256_mul_ps(fscal,dy00);
 364             tz               = _mm256_mul_ps(fscal,dz00);
 365
 366             /* Update vectorial force */
 367             fix0             = _mm256_add_ps(fix0,tx);
 368             fiy0             = _mm256_add_ps(fiy0,ty);
 369             fiz0             = _mm256_add_ps(fiz0,tz);
 370
 371             fjx0             = _mm256_add_ps(fjx0,tx);
 372             fjy0             = _mm256_add_ps(fjy0,ty);
 373             fjz0             = _mm256_add_ps(fjz0,tz);
 374
 375             }
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 382             {
 383
 384             r10              = _mm256_mul_ps(rsq10,rinv10);
 385
 386             /* Compute parameters for interactions between i and j atoms */
 387             qq10             = _mm256_mul_ps(iq1,jq0);
 388
 389             /* EWALD ELECTROSTATICS */
 390
 391             /* Analytical PME correction */
 392             zeta2            = _mm256_mul_ps(beta2,rsq10);
 393             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 394             pmecorrF         = avx256_pmecorrF_f(zeta2);
 395             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 396             felec            = _mm256_mul_ps(qq10,felec);
 397             pmecorrV         = avx256_pmecorrV_f(zeta2);
 398             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 399             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 400             velec            = _mm256_mul_ps(qq10,velec);
 401
 402             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm256_and_ps(velec,cutoff_mask);
 406             velecsum         = _mm256_add_ps(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 411
 412             /* Calculate temporary vectorial force */
 413             tx               = _mm256_mul_ps(fscal,dx10);
 414             ty               = _mm256_mul_ps(fscal,dy10);
 415             tz               = _mm256_mul_ps(fscal,dz10);
 416
 417             /* Update vectorial force */
 418             fix1             = _mm256_add_ps(fix1,tx);
 419             fiy1             = _mm256_add_ps(fiy1,ty);
 420             fiz1             = _mm256_add_ps(fiz1,tz);
 421
 422             fjx0             = _mm256_add_ps(fjx0,tx);
 423             fjy0             = _mm256_add_ps(fjy0,ty);
 424             fjz0             = _mm256_add_ps(fjz0,tz);
 425
 426             }
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 433             {
 434
 435             r20              = _mm256_mul_ps(rsq20,rinv20);
 436
 437             /* Compute parameters for interactions between i and j atoms */
 438             qq20             = _mm256_mul_ps(iq2,jq0);
 439
 440             /* EWALD ELECTROSTATICS */
 441
 442             /* Analytical PME correction */
 443             zeta2            = _mm256_mul_ps(beta2,rsq20);
 444             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 445             pmecorrF         = avx256_pmecorrF_f(zeta2);
 446             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 447             felec            = _mm256_mul_ps(qq20,felec);
 448             pmecorrV         = avx256_pmecorrV_f(zeta2);
 449             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 450             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 451             velec            = _mm256_mul_ps(qq20,velec);
 452
 453             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 454
 455             /* Update potential sum for this i atom from the interaction with this j atom. */
 456             velec            = _mm256_and_ps(velec,cutoff_mask);
 457             velecsum         = _mm256_add_ps(velecsum,velec);
 458
 459             fscal            = felec;
 460
 461             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 462
 463             /* Calculate temporary vectorial force */
 464             tx               = _mm256_mul_ps(fscal,dx20);
 465             ty               = _mm256_mul_ps(fscal,dy20);
 466             tz               = _mm256_mul_ps(fscal,dz20);
 467
 468             /* Update vectorial force */
 469             fix2             = _mm256_add_ps(fix2,tx);
 470             fiy2             = _mm256_add_ps(fiy2,ty);
 471             fiz2             = _mm256_add_ps(fiz2,tz);
 472
 473             fjx0             = _mm256_add_ps(fjx0,tx);
 474             fjy0             = _mm256_add_ps(fjy0,ty);
 475             fjz0             = _mm256_add_ps(fjz0,tz);
 476
 477             }
 478
 479             fjptrA             = f+j_coord_offsetA;
 480             fjptrB             = f+j_coord_offsetB;
 481             fjptrC             = f+j_coord_offsetC;
 482             fjptrD             = f+j_coord_offsetD;
 483             fjptrE             = f+j_coord_offsetE;
 484             fjptrF             = f+j_coord_offsetF;
 485             fjptrG             = f+j_coord_offsetG;
 486             fjptrH             = f+j_coord_offsetH;
 487
 488             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 489
 490             /* Inner loop uses 366 flops */
 491         }
 492
 493         if(jidx<j_index_end)
 494         {
 495
 496             /* Get j neighbor index, and coordinate index */
 497             jnrlistA         = jjnr[jidx];
 498             jnrlistB         = jjnr[jidx+1];
 499             jnrlistC         = jjnr[jidx+2];
 500             jnrlistD         = jjnr[jidx+3];
 501             jnrlistE         = jjnr[jidx+4];
 502             jnrlistF         = jjnr[jidx+5];
 503             jnrlistG         = jjnr[jidx+6];
 504             jnrlistH         = jjnr[jidx+7];
 505             /* Sign of each element will be negative for non-real atoms.
 506              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 507              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 508              */
 509             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 510                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 511
 512             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 513             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 514             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 515             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 516             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 517             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 518             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 519             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 520             j_coord_offsetA  = DIM*jnrA;
 521             j_coord_offsetB  = DIM*jnrB;
 522             j_coord_offsetC  = DIM*jnrC;
 523             j_coord_offsetD  = DIM*jnrD;
 524             j_coord_offsetE  = DIM*jnrE;
 525             j_coord_offsetF  = DIM*jnrF;
 526             j_coord_offsetG  = DIM*jnrG;
 527             j_coord_offsetH  = DIM*jnrH;
 528
 529             /* load j atom coordinates */
 530             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 531                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 532                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 533                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 534                                                  &jx0,&jy0,&jz0);
 535
 536             /* Calculate displacement vector */
 537             dx00             = _mm256_sub_ps(ix0,jx0);
 538             dy00             = _mm256_sub_ps(iy0,jy0);
 539             dz00             = _mm256_sub_ps(iz0,jz0);
 540             dx10             = _mm256_sub_ps(ix1,jx0);
 541             dy10             = _mm256_sub_ps(iy1,jy0);
 542             dz10             = _mm256_sub_ps(iz1,jz0);
 543             dx20             = _mm256_sub_ps(ix2,jx0);
 544             dy20             = _mm256_sub_ps(iy2,jy0);
 545             dz20             = _mm256_sub_ps(iz2,jz0);
 546
 547             /* Calculate squared distance and things based on it */
 548             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 549             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 550             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 551
 552             rinv00           = avx256_invsqrt_f(rsq00);
 553             rinv10           = avx256_invsqrt_f(rsq10);
 554             rinv20           = avx256_invsqrt_f(rsq20);
 555
 556             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 557             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 558             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 559
 560             /* Load parameters for j particles */
 561             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 562                                                                  charge+jnrC+0,charge+jnrD+0,
 563                                                                  charge+jnrE+0,charge+jnrF+0,
 564                                                                  charge+jnrG+0,charge+jnrH+0);
 565             vdwjidx0A        = 2*vdwtype[jnrA+0];
 566             vdwjidx0B        = 2*vdwtype[jnrB+0];
 567             vdwjidx0C        = 2*vdwtype[jnrC+0];
 568             vdwjidx0D        = 2*vdwtype[jnrD+0];
 569             vdwjidx0E        = 2*vdwtype[jnrE+0];
 570             vdwjidx0F        = 2*vdwtype[jnrF+0];
 571             vdwjidx0G        = 2*vdwtype[jnrG+0];
 572             vdwjidx0H        = 2*vdwtype[jnrH+0];
 573
 574             fjx0             = _mm256_setzero_ps();
 575             fjy0             = _mm256_setzero_ps();
 576             fjz0             = _mm256_setzero_ps();
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 583             {
 584
 585             r00              = _mm256_mul_ps(rsq00,rinv00);
 586             r00              = _mm256_andnot_ps(dummy_mask,r00);
 587
 588             /* Compute parameters for interactions between i and j atoms */
 589             qq00             = _mm256_mul_ps(iq0,jq0);
 590             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 591                                             vdwioffsetptr0+vdwjidx0B,
 592                                             vdwioffsetptr0+vdwjidx0C,
 593                                             vdwioffsetptr0+vdwjidx0D,
 594                                             vdwioffsetptr0+vdwjidx0E,
 595                                             vdwioffsetptr0+vdwjidx0F,
 596                                             vdwioffsetptr0+vdwjidx0G,
 597                                             vdwioffsetptr0+vdwjidx0H,
 598                                             &c6_00,&c12_00);
 599
 600             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 601                                                                   vdwgridioffsetptr0+vdwjidx0B,
 602                                                                   vdwgridioffsetptr0+vdwjidx0C,
 603                                                                   vdwgridioffsetptr0+vdwjidx0D,
 604                                                                   vdwgridioffsetptr0+vdwjidx0E,
 605                                                                   vdwgridioffsetptr0+vdwjidx0F,
 606                                                                   vdwgridioffsetptr0+vdwjidx0G,
 607                                                                   vdwgridioffsetptr0+vdwjidx0H);
 608
 609             /* EWALD ELECTROSTATICS */
 610
 611             /* Analytical PME correction */
 612             zeta2            = _mm256_mul_ps(beta2,rsq00);
 613             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 614             pmecorrF         = avx256_pmecorrF_f(zeta2);
 615             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 616             felec            = _mm256_mul_ps(qq00,felec);
 617             pmecorrV         = avx256_pmecorrV_f(zeta2);
 618             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 619             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 620             velec            = _mm256_mul_ps(qq00,velec);
 621
 622             /* Analytical LJ-PME */
 623             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 624             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 625             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 626             exponent         = avx256_exp_f(ewcljrsq);
 627             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 628             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 629             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 630             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 631             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 632             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 633                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_00,sh_vdw_invrcut6),_mm256_mul_ps(c6grid_00,sh_lj_ewald))),one_sixth));
 634             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 635             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq00);
 636
 637             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm256_and_ps(velec,cutoff_mask);
 641             velec            = _mm256_andnot_ps(dummy_mask,velec);
 642             velecsum         = _mm256_add_ps(velecsum,velec);
 643             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 644             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 645             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 646
 647             fscal            = _mm256_add_ps(felec,fvdw);
 648
 649             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 650
 651             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 652
 653             /* Calculate temporary vectorial force */
 654             tx               = _mm256_mul_ps(fscal,dx00);
 655             ty               = _mm256_mul_ps(fscal,dy00);
 656             tz               = _mm256_mul_ps(fscal,dz00);
 657
 658             /* Update vectorial force */
 659             fix0             = _mm256_add_ps(fix0,tx);
 660             fiy0             = _mm256_add_ps(fiy0,ty);
 661             fiz0             = _mm256_add_ps(fiz0,tz);
 662
 663             fjx0             = _mm256_add_ps(fjx0,tx);
 664             fjy0             = _mm256_add_ps(fjy0,ty);
 665             fjz0             = _mm256_add_ps(fjz0,tz);
 666
 667             }
 668
 669             /**************************
 670              * CALCULATE INTERACTIONS *
 671              **************************/
 672
 673             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 674             {
 675
 676             r10              = _mm256_mul_ps(rsq10,rinv10);
 677             r10              = _mm256_andnot_ps(dummy_mask,r10);
 678
 679             /* Compute parameters for interactions between i and j atoms */
 680             qq10             = _mm256_mul_ps(iq1,jq0);
 681
 682             /* EWALD ELECTROSTATICS */
 683
 684             /* Analytical PME correction */
 685             zeta2            = _mm256_mul_ps(beta2,rsq10);
 686             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 687             pmecorrF         = avx256_pmecorrF_f(zeta2);
 688             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 689             felec            = _mm256_mul_ps(qq10,felec);
 690             pmecorrV         = avx256_pmecorrV_f(zeta2);
 691             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 692             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 693             velec            = _mm256_mul_ps(qq10,velec);
 694
 695             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 696
 697             /* Update potential sum for this i atom from the interaction with this j atom. */
 698             velec            = _mm256_and_ps(velec,cutoff_mask);
 699             velec            = _mm256_andnot_ps(dummy_mask,velec);
 700             velecsum         = _mm256_add_ps(velecsum,velec);
 701
 702             fscal            = felec;
 703
 704             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 705
 706             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 707
 708             /* Calculate temporary vectorial force */
 709             tx               = _mm256_mul_ps(fscal,dx10);
 710             ty               = _mm256_mul_ps(fscal,dy10);
 711             tz               = _mm256_mul_ps(fscal,dz10);
 712
 713             /* Update vectorial force */
 714             fix1             = _mm256_add_ps(fix1,tx);
 715             fiy1             = _mm256_add_ps(fiy1,ty);
 716             fiz1             = _mm256_add_ps(fiz1,tz);
 717
 718             fjx0             = _mm256_add_ps(fjx0,tx);
 719             fjy0             = _mm256_add_ps(fjy0,ty);
 720             fjz0             = _mm256_add_ps(fjz0,tz);
 721
 722             }
 723
 724             /**************************
 725              * CALCULATE INTERACTIONS *
 726              **************************/
 727
 728             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 729             {
 730
 731             r20              = _mm256_mul_ps(rsq20,rinv20);
 732             r20              = _mm256_andnot_ps(dummy_mask,r20);
 733
 734             /* Compute parameters for interactions between i and j atoms */
 735             qq20             = _mm256_mul_ps(iq2,jq0);
 736
 737             /* EWALD ELECTROSTATICS */
 738
 739             /* Analytical PME correction */
 740             zeta2            = _mm256_mul_ps(beta2,rsq20);
 741             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 742             pmecorrF         = avx256_pmecorrF_f(zeta2);
 743             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 744             felec            = _mm256_mul_ps(qq20,felec);
 745             pmecorrV         = avx256_pmecorrV_f(zeta2);
 746             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 747             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 748             velec            = _mm256_mul_ps(qq20,velec);
 749
 750             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 751
 752             /* Update potential sum for this i atom from the interaction with this j atom. */
 753             velec            = _mm256_and_ps(velec,cutoff_mask);
 754             velec            = _mm256_andnot_ps(dummy_mask,velec);
 755             velecsum         = _mm256_add_ps(velecsum,velec);
 756
 757             fscal            = felec;
 758
 759             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 760
 761             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 762
 763             /* Calculate temporary vectorial force */
 764             tx               = _mm256_mul_ps(fscal,dx20);
 765             ty               = _mm256_mul_ps(fscal,dy20);
 766             tz               = _mm256_mul_ps(fscal,dz20);
 767
 768             /* Update vectorial force */
 769             fix2             = _mm256_add_ps(fix2,tx);
 770             fiy2             = _mm256_add_ps(fiy2,ty);
 771             fiz2             = _mm256_add_ps(fiz2,tz);
 772
 773             fjx0             = _mm256_add_ps(fjx0,tx);
 774             fjy0             = _mm256_add_ps(fjy0,ty);
 775             fjz0             = _mm256_add_ps(fjz0,tz);
 776
 777             }
 778
 779             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 780             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 781             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 782             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 783             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 784             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 785             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 786             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 787
 788             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 789
 790             /* Inner loop uses 369 flops */
 791         }
 792
 793         /* End of innermost loop */
 794
 795         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 796                                                  f+i_coord_offset,fshift+i_shift_offset);
 797
 798         ggid                        = gid[iidx];
 799         /* Update potential energies */
 800         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 801         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 802
 803         /* Increment number of inner iterations */
 804         inneriter                  += j_index_end - j_index_start;
 805
 806         /* Outer loop uses 20 flops */
 807     }
 808
 809     /* Increment number of outer iterations */
 810     outeriter        += nri;
 811
 812     /* Update outer/inner flops */
 813
 814     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*369);
 815 }
 816 /*
 817  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_single
 818  * Electrostatics interaction: Ewald
 819  * VdW interaction:            LJEwald
 820  * Geometry:                   Water3-Particle
 821  * Calculate force/pot:        Force
 822  */
 823 void
 824 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_single
 825                     (t_nblist                    * gmx_restrict       nlist,
 826                      rvec                        * gmx_restrict          xx,
 827                      rvec                        * gmx_restrict          ff,
 828                      struct t_forcerec           * gmx_restrict          fr,
 829                      t_mdatoms                   * gmx_restrict     mdatoms,
 830                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 831                      t_nrnb                      * gmx_restrict        nrnb)
 832 {
 833     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 834      * just 0 for non-waters.
 835      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 836      * jnr indices corresponding to data put in the four positions in the SIMD register.
 837      */
 838     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 839     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 840     int              jnrA,jnrB,jnrC,jnrD;
 841     int              jnrE,jnrF,jnrG,jnrH;
 842     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 843     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 844     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 845     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 846     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 847     real             rcutoff_scalar;
 848     real             *shiftvec,*fshift,*x,*f;
 849     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 850     real             scratch[4*DIM];
 851     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 852     real *           vdwioffsetptr0;
 853     real *           vdwgridioffsetptr0;
 854     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 855     real *           vdwioffsetptr1;
 856     real *           vdwgridioffsetptr1;
 857     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 858     real *           vdwioffsetptr2;
 859     real *           vdwgridioffsetptr2;
 860     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 861     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 862     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 863     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 864     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 865     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 866     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 867     real             *charge;
 868     int              nvdwtype;
 869     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 870     int              *vdwtype;
 871     real             *vdwparam;
 872     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 873     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 874     __m256           c6grid_00;
 875     __m256           c6grid_10;
 876     __m256           c6grid_20;
 877     real             *vdwgridparam;
 878     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 879     __m256           one_half  = _mm256_set1_ps(0.5);
 880     __m256           minus_one = _mm256_set1_ps(-1.0);
 881     __m256i          ewitab;
 882     __m128i          ewitab_lo,ewitab_hi;
 883     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 884     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 885     real             *ewtab;
 886     __m256           dummy_mask,cutoff_mask;
 887     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 888     __m256           one     = _mm256_set1_ps(1.0);
 889     __m256           two     = _mm256_set1_ps(2.0);
 890     x                = xx[0];
 891     f                = ff[0];
 892
 893     nri              = nlist->nri;
 894     iinr             = nlist->iinr;
 895     jindex           = nlist->jindex;
 896     jjnr             = nlist->jjnr;
 897     shiftidx         = nlist->shift;
 898     gid              = nlist->gid;
 899     shiftvec         = fr->shift_vec[0];
 900     fshift           = fr->fshift[0];
 901     facel            = _mm256_set1_ps(fr->ic->epsfac);
 902     charge           = mdatoms->chargeA;
 903     nvdwtype         = fr->ntype;
 904     vdwparam         = fr->nbfp;
 905     vdwtype          = mdatoms->typeA;
 906     vdwgridparam     = fr->ljpme_c6grid;
 907     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 908     ewclj            = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
 909     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 910
 911     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 912     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 913     beta2            = _mm256_mul_ps(beta,beta);
 914     beta3            = _mm256_mul_ps(beta,beta2);
 915
 916     ewtab            = fr->ic->tabq_coul_F;
 917     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 918     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 919
 920     /* Setup water-specific parameters */
 921     inr              = nlist->iinr[0];
 922     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 923     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 924     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 925     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 926     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 927
 928     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 929     rcutoff_scalar   = fr->ic->rcoulomb;
 930     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 931     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 932
 933     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 934     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 935
 936     /* Avoid stupid compiler warnings */
 937     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 938     j_coord_offsetA = 0;
 939     j_coord_offsetB = 0;
 940     j_coord_offsetC = 0;
 941     j_coord_offsetD = 0;
 942     j_coord_offsetE = 0;
 943     j_coord_offsetF = 0;
 944     j_coord_offsetG = 0;
 945     j_coord_offsetH = 0;
 946
 947     outeriter        = 0;
 948     inneriter        = 0;
 949
 950     for(iidx=0;iidx<4*DIM;iidx++)
 951     {
 952         scratch[iidx] = 0.0;
 953     }
 954
 955     /* Start outer loop over neighborlists */
 956     for(iidx=0; iidx<nri; iidx++)
 957     {
 958         /* Load shift vector for this list */
 959         i_shift_offset   = DIM*shiftidx[iidx];
 960
 961         /* Load limits for loop over neighbors */
 962         j_index_start    = jindex[iidx];
 963         j_index_end      = jindex[iidx+1];
 964
 965         /* Get outer coordinate index */
 966         inr              = iinr[iidx];
 967         i_coord_offset   = DIM*inr;
 968
 969         /* Load i particle coords and add shift vector */
 970         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 971                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 972
 973         fix0             = _mm256_setzero_ps();
 974         fiy0             = _mm256_setzero_ps();
 975         fiz0             = _mm256_setzero_ps();
 976         fix1             = _mm256_setzero_ps();
 977         fiy1             = _mm256_setzero_ps();
 978         fiz1             = _mm256_setzero_ps();
 979         fix2             = _mm256_setzero_ps();
 980         fiy2             = _mm256_setzero_ps();
 981         fiz2             = _mm256_setzero_ps();
 982
 983         /* Start inner kernel loop */
 984         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 985         {
 986
 987             /* Get j neighbor index, and coordinate index */
 988             jnrA             = jjnr[jidx];
 989             jnrB             = jjnr[jidx+1];
 990             jnrC             = jjnr[jidx+2];
 991             jnrD             = jjnr[jidx+3];
 992             jnrE             = jjnr[jidx+4];
 993             jnrF             = jjnr[jidx+5];
 994             jnrG             = jjnr[jidx+6];
 995             jnrH             = jjnr[jidx+7];
 996             j_coord_offsetA  = DIM*jnrA;
 997             j_coord_offsetB  = DIM*jnrB;
 998             j_coord_offsetC  = DIM*jnrC;
 999             j_coord_offsetD  = DIM*jnrD;
1000             j_coord_offsetE  = DIM*jnrE;
1001             j_coord_offsetF  = DIM*jnrF;
1002             j_coord_offsetG  = DIM*jnrG;
1003             j_coord_offsetH  = DIM*jnrH;
1004
1005             /* load j atom coordinates */
1006             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1007                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1008                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1009                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1010                                                  &jx0,&jy0,&jz0);
1011
1012             /* Calculate displacement vector */
1013             dx00             = _mm256_sub_ps(ix0,jx0);
1014             dy00             = _mm256_sub_ps(iy0,jy0);
1015             dz00             = _mm256_sub_ps(iz0,jz0);
1016             dx10             = _mm256_sub_ps(ix1,jx0);
1017             dy10             = _mm256_sub_ps(iy1,jy0);
1018             dz10             = _mm256_sub_ps(iz1,jz0);
1019             dx20             = _mm256_sub_ps(ix2,jx0);
1020             dy20             = _mm256_sub_ps(iy2,jy0);
1021             dz20             = _mm256_sub_ps(iz2,jz0);
1022
1023             /* Calculate squared distance and things based on it */
1024             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1025             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1026             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1027
1028             rinv00           = avx256_invsqrt_f(rsq00);
1029             rinv10           = avx256_invsqrt_f(rsq10);
1030             rinv20           = avx256_invsqrt_f(rsq20);
1031
1032             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1033             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1034             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1035
1036             /* Load parameters for j particles */
1037             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1038                                                                  charge+jnrC+0,charge+jnrD+0,
1039                                                                  charge+jnrE+0,charge+jnrF+0,
1040                                                                  charge+jnrG+0,charge+jnrH+0);
1041             vdwjidx0A        = 2*vdwtype[jnrA+0];
1042             vdwjidx0B        = 2*vdwtype[jnrB+0];
1043             vdwjidx0C        = 2*vdwtype[jnrC+0];
1044             vdwjidx0D        = 2*vdwtype[jnrD+0];
1045             vdwjidx0E        = 2*vdwtype[jnrE+0];
1046             vdwjidx0F        = 2*vdwtype[jnrF+0];
1047             vdwjidx0G        = 2*vdwtype[jnrG+0];
1048             vdwjidx0H        = 2*vdwtype[jnrH+0];
1049
1050             fjx0             = _mm256_setzero_ps();
1051             fjy0             = _mm256_setzero_ps();
1052             fjz0             = _mm256_setzero_ps();
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1059             {
1060
1061             r00              = _mm256_mul_ps(rsq00,rinv00);
1062
1063             /* Compute parameters for interactions between i and j atoms */
1064             qq00             = _mm256_mul_ps(iq0,jq0);
1065             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1066                                             vdwioffsetptr0+vdwjidx0B,
1067                                             vdwioffsetptr0+vdwjidx0C,
1068                                             vdwioffsetptr0+vdwjidx0D,
1069                                             vdwioffsetptr0+vdwjidx0E,
1070                                             vdwioffsetptr0+vdwjidx0F,
1071                                             vdwioffsetptr0+vdwjidx0G,
1072                                             vdwioffsetptr0+vdwjidx0H,
1073                                             &c6_00,&c12_00);
1074
1075             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1076                                                                   vdwgridioffsetptr0+vdwjidx0B,
1077                                                                   vdwgridioffsetptr0+vdwjidx0C,
1078                                                                   vdwgridioffsetptr0+vdwjidx0D,
1079                                                                   vdwgridioffsetptr0+vdwjidx0E,
1080                                                                   vdwgridioffsetptr0+vdwjidx0F,
1081                                                                   vdwgridioffsetptr0+vdwjidx0G,
1082                                                                   vdwgridioffsetptr0+vdwjidx0H);
1083
1084             /* EWALD ELECTROSTATICS */
1085
1086             /* Analytical PME correction */
1087             zeta2            = _mm256_mul_ps(beta2,rsq00);
1088             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1089             pmecorrF         = avx256_pmecorrF_f(zeta2);
1090             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1091             felec            = _mm256_mul_ps(qq00,felec);
1092
1093             /* Analytical LJ-PME */
1094             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1095             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1096             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1097             exponent         = avx256_exp_f(ewcljrsq);
1098             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1099             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1100             /* f6A = 6 * C6grid * (1 - poly) */
1101             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1102             /* f6B = C6grid * exponent * beta^6 */
1103             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1104             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1105             fvdw              = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1106
1107             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1108
1109             fscal            = _mm256_add_ps(felec,fvdw);
1110
1111             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1112
1113             /* Calculate temporary vectorial force */
1114             tx               = _mm256_mul_ps(fscal,dx00);
1115             ty               = _mm256_mul_ps(fscal,dy00);
1116             tz               = _mm256_mul_ps(fscal,dz00);
1117
1118             /* Update vectorial force */
1119             fix0             = _mm256_add_ps(fix0,tx);
1120             fiy0             = _mm256_add_ps(fiy0,ty);
1121             fiz0             = _mm256_add_ps(fiz0,tz);
1122
1123             fjx0             = _mm256_add_ps(fjx0,tx);
1124             fjy0             = _mm256_add_ps(fjy0,ty);
1125             fjz0             = _mm256_add_ps(fjz0,tz);
1126
1127             }
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1134             {
1135
1136             r10              = _mm256_mul_ps(rsq10,rinv10);
1137
1138             /* Compute parameters for interactions between i and j atoms */
1139             qq10             = _mm256_mul_ps(iq1,jq0);
1140
1141             /* EWALD ELECTROSTATICS */
1142
1143             /* Analytical PME correction */
1144             zeta2            = _mm256_mul_ps(beta2,rsq10);
1145             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1146             pmecorrF         = avx256_pmecorrF_f(zeta2);
1147             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1148             felec            = _mm256_mul_ps(qq10,felec);
1149
1150             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1151
1152             fscal            = felec;
1153
1154             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1155
1156             /* Calculate temporary vectorial force */
1157             tx               = _mm256_mul_ps(fscal,dx10);
1158             ty               = _mm256_mul_ps(fscal,dy10);
1159             tz               = _mm256_mul_ps(fscal,dz10);
1160
1161             /* Update vectorial force */
1162             fix1             = _mm256_add_ps(fix1,tx);
1163             fiy1             = _mm256_add_ps(fiy1,ty);
1164             fiz1             = _mm256_add_ps(fiz1,tz);
1165
1166             fjx0             = _mm256_add_ps(fjx0,tx);
1167             fjy0             = _mm256_add_ps(fjy0,ty);
1168             fjz0             = _mm256_add_ps(fjz0,tz);
1169
1170             }
1171
1172             /**************************
1173              * CALCULATE INTERACTIONS *
1174              **************************/
1175
1176             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1177             {
1178
1179             r20              = _mm256_mul_ps(rsq20,rinv20);
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             qq20             = _mm256_mul_ps(iq2,jq0);
1183
1184             /* EWALD ELECTROSTATICS */
1185
1186             /* Analytical PME correction */
1187             zeta2            = _mm256_mul_ps(beta2,rsq20);
1188             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1189             pmecorrF         = avx256_pmecorrF_f(zeta2);
1190             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1191             felec            = _mm256_mul_ps(qq20,felec);
1192
1193             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1194
1195             fscal            = felec;
1196
1197             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1198
1199             /* Calculate temporary vectorial force */
1200             tx               = _mm256_mul_ps(fscal,dx20);
1201             ty               = _mm256_mul_ps(fscal,dy20);
1202             tz               = _mm256_mul_ps(fscal,dz20);
1203
1204             /* Update vectorial force */
1205             fix2             = _mm256_add_ps(fix2,tx);
1206             fiy2             = _mm256_add_ps(fiy2,ty);
1207             fiz2             = _mm256_add_ps(fiz2,tz);
1208
1209             fjx0             = _mm256_add_ps(fjx0,tx);
1210             fjy0             = _mm256_add_ps(fjy0,ty);
1211             fjz0             = _mm256_add_ps(fjz0,tz);
1212
1213             }
1214
1215             fjptrA             = f+j_coord_offsetA;
1216             fjptrB             = f+j_coord_offsetB;
1217             fjptrC             = f+j_coord_offsetC;
1218             fjptrD             = f+j_coord_offsetD;
1219             fjptrE             = f+j_coord_offsetE;
1220             fjptrF             = f+j_coord_offsetF;
1221             fjptrG             = f+j_coord_offsetG;
1222             fjptrH             = f+j_coord_offsetH;
1223
1224             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1225
1226             /* Inner loop uses 203 flops */
1227         }
1228
1229         if(jidx<j_index_end)
1230         {
1231
1232             /* Get j neighbor index, and coordinate index */
1233             jnrlistA         = jjnr[jidx];
1234             jnrlistB         = jjnr[jidx+1];
1235             jnrlistC         = jjnr[jidx+2];
1236             jnrlistD         = jjnr[jidx+3];
1237             jnrlistE         = jjnr[jidx+4];
1238             jnrlistF         = jjnr[jidx+5];
1239             jnrlistG         = jjnr[jidx+6];
1240             jnrlistH         = jjnr[jidx+7];
1241             /* Sign of each element will be negative for non-real atoms.
1242              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1243              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1244              */
1245             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1246                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1247
1248             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1249             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1250             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1251             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1252             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1253             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1254             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1255             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1256             j_coord_offsetA  = DIM*jnrA;
1257             j_coord_offsetB  = DIM*jnrB;
1258             j_coord_offsetC  = DIM*jnrC;
1259             j_coord_offsetD  = DIM*jnrD;
1260             j_coord_offsetE  = DIM*jnrE;
1261             j_coord_offsetF  = DIM*jnrF;
1262             j_coord_offsetG  = DIM*jnrG;
1263             j_coord_offsetH  = DIM*jnrH;
1264
1265             /* load j atom coordinates */
1266             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1267                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1268                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1269                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1270                                                  &jx0,&jy0,&jz0);
1271
1272             /* Calculate displacement vector */
1273             dx00             = _mm256_sub_ps(ix0,jx0);
1274             dy00             = _mm256_sub_ps(iy0,jy0);
1275             dz00             = _mm256_sub_ps(iz0,jz0);
1276             dx10             = _mm256_sub_ps(ix1,jx0);
1277             dy10             = _mm256_sub_ps(iy1,jy0);
1278             dz10             = _mm256_sub_ps(iz1,jz0);
1279             dx20             = _mm256_sub_ps(ix2,jx0);
1280             dy20             = _mm256_sub_ps(iy2,jy0);
1281             dz20             = _mm256_sub_ps(iz2,jz0);
1282
1283             /* Calculate squared distance and things based on it */
1284             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1285             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1286             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1287
1288             rinv00           = avx256_invsqrt_f(rsq00);
1289             rinv10           = avx256_invsqrt_f(rsq10);
1290             rinv20           = avx256_invsqrt_f(rsq20);
1291
1292             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1293             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1294             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1295
1296             /* Load parameters for j particles */
1297             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1298                                                                  charge+jnrC+0,charge+jnrD+0,
1299                                                                  charge+jnrE+0,charge+jnrF+0,
1300                                                                  charge+jnrG+0,charge+jnrH+0);
1301             vdwjidx0A        = 2*vdwtype[jnrA+0];
1302             vdwjidx0B        = 2*vdwtype[jnrB+0];
1303             vdwjidx0C        = 2*vdwtype[jnrC+0];
1304             vdwjidx0D        = 2*vdwtype[jnrD+0];
1305             vdwjidx0E        = 2*vdwtype[jnrE+0];
1306             vdwjidx0F        = 2*vdwtype[jnrF+0];
1307             vdwjidx0G        = 2*vdwtype[jnrG+0];
1308             vdwjidx0H        = 2*vdwtype[jnrH+0];
1309
1310             fjx0             = _mm256_setzero_ps();
1311             fjy0             = _mm256_setzero_ps();
1312             fjz0             = _mm256_setzero_ps();
1313
1314             /**************************
1315              * CALCULATE INTERACTIONS *
1316              **************************/
1317
1318             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1319             {
1320
1321             r00              = _mm256_mul_ps(rsq00,rinv00);
1322             r00              = _mm256_andnot_ps(dummy_mask,r00);
1323
1324             /* Compute parameters for interactions between i and j atoms */
1325             qq00             = _mm256_mul_ps(iq0,jq0);
1326             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1327                                             vdwioffsetptr0+vdwjidx0B,
1328                                             vdwioffsetptr0+vdwjidx0C,
1329                                             vdwioffsetptr0+vdwjidx0D,
1330                                             vdwioffsetptr0+vdwjidx0E,
1331                                             vdwioffsetptr0+vdwjidx0F,
1332                                             vdwioffsetptr0+vdwjidx0G,
1333                                             vdwioffsetptr0+vdwjidx0H,
1334                                             &c6_00,&c12_00);
1335
1336             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1337                                                                   vdwgridioffsetptr0+vdwjidx0B,
1338                                                                   vdwgridioffsetptr0+vdwjidx0C,
1339                                                                   vdwgridioffsetptr0+vdwjidx0D,
1340                                                                   vdwgridioffsetptr0+vdwjidx0E,
1341                                                                   vdwgridioffsetptr0+vdwjidx0F,
1342                                                                   vdwgridioffsetptr0+vdwjidx0G,
1343                                                                   vdwgridioffsetptr0+vdwjidx0H);
1344
1345             /* EWALD ELECTROSTATICS */
1346
1347             /* Analytical PME correction */
1348             zeta2            = _mm256_mul_ps(beta2,rsq00);
1349             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1350             pmecorrF         = avx256_pmecorrF_f(zeta2);
1351             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1352             felec            = _mm256_mul_ps(qq00,felec);
1353
1354             /* Analytical LJ-PME */
1355             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1356             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1357             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1358             exponent         = avx256_exp_f(ewcljrsq);
1359             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1360             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1361             /* f6A = 6 * C6grid * (1 - poly) */
1362             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1363             /* f6B = C6grid * exponent * beta^6 */
1364             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1365             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1366             fvdw              = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),_mm256_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1367
1368             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1369
1370             fscal            = _mm256_add_ps(felec,fvdw);
1371
1372             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1373
1374             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1375
1376             /* Calculate temporary vectorial force */
1377             tx               = _mm256_mul_ps(fscal,dx00);
1378             ty               = _mm256_mul_ps(fscal,dy00);
1379             tz               = _mm256_mul_ps(fscal,dz00);
1380
1381             /* Update vectorial force */
1382             fix0             = _mm256_add_ps(fix0,tx);
1383             fiy0             = _mm256_add_ps(fiy0,ty);
1384             fiz0             = _mm256_add_ps(fiz0,tz);
1385
1386             fjx0             = _mm256_add_ps(fjx0,tx);
1387             fjy0             = _mm256_add_ps(fjy0,ty);
1388             fjz0             = _mm256_add_ps(fjz0,tz);
1389
1390             }
1391
1392             /**************************
1393              * CALCULATE INTERACTIONS *
1394              **************************/
1395
1396             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1397             {
1398
1399             r10              = _mm256_mul_ps(rsq10,rinv10);
1400             r10              = _mm256_andnot_ps(dummy_mask,r10);
1401
1402             /* Compute parameters for interactions between i and j atoms */
1403             qq10             = _mm256_mul_ps(iq1,jq0);
1404
1405             /* EWALD ELECTROSTATICS */
1406
1407             /* Analytical PME correction */
1408             zeta2            = _mm256_mul_ps(beta2,rsq10);
1409             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1410             pmecorrF         = avx256_pmecorrF_f(zeta2);
1411             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1412             felec            = _mm256_mul_ps(qq10,felec);
1413
1414             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1415
1416             fscal            = felec;
1417
1418             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1419
1420             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1421
1422             /* Calculate temporary vectorial force */
1423             tx               = _mm256_mul_ps(fscal,dx10);
1424             ty               = _mm256_mul_ps(fscal,dy10);
1425             tz               = _mm256_mul_ps(fscal,dz10);
1426
1427             /* Update vectorial force */
1428             fix1             = _mm256_add_ps(fix1,tx);
1429             fiy1             = _mm256_add_ps(fiy1,ty);
1430             fiz1             = _mm256_add_ps(fiz1,tz);
1431
1432             fjx0             = _mm256_add_ps(fjx0,tx);
1433             fjy0             = _mm256_add_ps(fjy0,ty);
1434             fjz0             = _mm256_add_ps(fjz0,tz);
1435
1436             }
1437
1438             /**************************
1439              * CALCULATE INTERACTIONS *
1440              **************************/
1441
1442             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1443             {
1444
1445             r20              = _mm256_mul_ps(rsq20,rinv20);
1446             r20              = _mm256_andnot_ps(dummy_mask,r20);
1447
1448             /* Compute parameters for interactions between i and j atoms */
1449             qq20             = _mm256_mul_ps(iq2,jq0);
1450
1451             /* EWALD ELECTROSTATICS */
1452
1453             /* Analytical PME correction */
1454             zeta2            = _mm256_mul_ps(beta2,rsq20);
1455             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1456             pmecorrF         = avx256_pmecorrF_f(zeta2);
1457             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1458             felec            = _mm256_mul_ps(qq20,felec);
1459
1460             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1461
1462             fscal            = felec;
1463
1464             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1465
1466             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1467
1468             /* Calculate temporary vectorial force */
1469             tx               = _mm256_mul_ps(fscal,dx20);
1470             ty               = _mm256_mul_ps(fscal,dy20);
1471             tz               = _mm256_mul_ps(fscal,dz20);
1472
1473             /* Update vectorial force */
1474             fix2             = _mm256_add_ps(fix2,tx);
1475             fiy2             = _mm256_add_ps(fiy2,ty);
1476             fiz2             = _mm256_add_ps(fiz2,tz);
1477
1478             fjx0             = _mm256_add_ps(fjx0,tx);
1479             fjy0             = _mm256_add_ps(fjy0,ty);
1480             fjz0             = _mm256_add_ps(fjz0,tz);
1481
1482             }
1483
1484             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1485             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1486             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1487             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1488             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1489             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1490             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1491             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1492
1493             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1494
1495             /* Inner loop uses 206 flops */
1496         }
1497
1498         /* End of innermost loop */
1499
1500         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1501                                                  f+i_coord_offset,fshift+i_shift_offset);
1502
1503         /* Increment number of inner iterations */
1504         inneriter                  += j_index_end - j_index_start;
1505
1506         /* Outer loop uses 18 flops */
1507     }
1508
1509     /* Increment number of outer iterations */
1510     outeriter        += nri;
1511
1512     /* Update outer/inner flops */
1513
1514     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*206);
1515 }