src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_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_GeomW4P1_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_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     real *           vdwioffsetptr3;
  95     real *           vdwgridioffsetptr3;
  96     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  97     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  98     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  99     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 100     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 101     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 102     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 103     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 104     real             *charge;
 105     int              nvdwtype;
 106     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 107     int              *vdwtype;
 108     real             *vdwparam;
 109     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 110     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 111     __m256           c6grid_00;
 112     __m256           c6grid_10;
 113     __m256           c6grid_20;
 114     __m256           c6grid_30;
 115     real             *vdwgridparam;
 116     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 117     __m256           one_half  = _mm256_set1_ps(0.5);
 118     __m256           minus_one = _mm256_set1_ps(-1.0);
 119     __m256i          ewitab;
 120     __m128i          ewitab_lo,ewitab_hi;
 121     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 122     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 123     real             *ewtab;
 124     __m256           dummy_mask,cutoff_mask;
 125     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 126     __m256           one     = _mm256_set1_ps(1.0);
 127     __m256           two     = _mm256_set1_ps(2.0);
 128     x                = xx[0];
 129     f                = ff[0];
 130
 131     nri              = nlist->nri;
 132     iinr             = nlist->iinr;
 133     jindex           = nlist->jindex;
 134     jjnr             = nlist->jjnr;
 135     shiftidx         = nlist->shift;
 136     gid              = nlist->gid;
 137     shiftvec         = fr->shift_vec[0];
 138     fshift           = fr->fshift[0];
 139     facel            = _mm256_set1_ps(fr->ic->epsfac);
 140     charge           = mdatoms->chargeA;
 141     nvdwtype         = fr->ntype;
 142     vdwparam         = fr->nbfp;
 143     vdwtype          = mdatoms->typeA;
 144     vdwgridparam     = fr->ljpme_c6grid;
 145     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 146     ewclj            = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
 147     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 148
 149     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 150     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 151     beta2            = _mm256_mul_ps(beta,beta);
 152     beta3            = _mm256_mul_ps(beta,beta2);
 153
 154     ewtab            = fr->ic->tabq_coul_FDV0;
 155     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 156     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 157
 158     /* Setup water-specific parameters */
 159     inr              = nlist->iinr[0];
 160     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 161     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 162     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
 163     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 164     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 165
 166     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 167     rcutoff_scalar   = fr->ic->rcoulomb;
 168     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 169     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 170
 171     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 172     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 173
 174     /* Avoid stupid compiler warnings */
 175     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 176     j_coord_offsetA = 0;
 177     j_coord_offsetB = 0;
 178     j_coord_offsetC = 0;
 179     j_coord_offsetD = 0;
 180     j_coord_offsetE = 0;
 181     j_coord_offsetF = 0;
 182     j_coord_offsetG = 0;
 183     j_coord_offsetH = 0;
 184
 185     outeriter        = 0;
 186     inneriter        = 0;
 187
 188     for(iidx=0;iidx<4*DIM;iidx++)
 189     {
 190         scratch[iidx] = 0.0;
 191     }
 192
 193     /* Start outer loop over neighborlists */
 194     for(iidx=0; iidx<nri; iidx++)
 195     {
 196         /* Load shift vector for this list */
 197         i_shift_offset   = DIM*shiftidx[iidx];
 198
 199         /* Load limits for loop over neighbors */
 200         j_index_start    = jindex[iidx];
 201         j_index_end      = jindex[iidx+1];
 202
 203         /* Get outer coordinate index */
 204         inr              = iinr[iidx];
 205         i_coord_offset   = DIM*inr;
 206
 207         /* Load i particle coords and add shift vector */
 208         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 209                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 210
 211         fix0             = _mm256_setzero_ps();
 212         fiy0             = _mm256_setzero_ps();
 213         fiz0             = _mm256_setzero_ps();
 214         fix1             = _mm256_setzero_ps();
 215         fiy1             = _mm256_setzero_ps();
 216         fiz1             = _mm256_setzero_ps();
 217         fix2             = _mm256_setzero_ps();
 218         fiy2             = _mm256_setzero_ps();
 219         fiz2             = _mm256_setzero_ps();
 220         fix3             = _mm256_setzero_ps();
 221         fiy3             = _mm256_setzero_ps();
 222         fiz3             = _mm256_setzero_ps();
 223
 224         /* Reset potential sums */
 225         velecsum         = _mm256_setzero_ps();
 226         vvdwsum          = _mm256_setzero_ps();
 227
 228         /* Start inner kernel loop */
 229         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 230         {
 231
 232             /* Get j neighbor index, and coordinate index */
 233             jnrA             = jjnr[jidx];
 234             jnrB             = jjnr[jidx+1];
 235             jnrC             = jjnr[jidx+2];
 236             jnrD             = jjnr[jidx+3];
 237             jnrE             = jjnr[jidx+4];
 238             jnrF             = jjnr[jidx+5];
 239             jnrG             = jjnr[jidx+6];
 240             jnrH             = jjnr[jidx+7];
 241             j_coord_offsetA  = DIM*jnrA;
 242             j_coord_offsetB  = DIM*jnrB;
 243             j_coord_offsetC  = DIM*jnrC;
 244             j_coord_offsetD  = DIM*jnrD;
 245             j_coord_offsetE  = DIM*jnrE;
 246             j_coord_offsetF  = DIM*jnrF;
 247             j_coord_offsetG  = DIM*jnrG;
 248             j_coord_offsetH  = DIM*jnrH;
 249
 250             /* load j atom coordinates */
 251             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 252                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 253                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 254                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 255                                                  &jx0,&jy0,&jz0);
 256
 257             /* Calculate displacement vector */
 258             dx00             = _mm256_sub_ps(ix0,jx0);
 259             dy00             = _mm256_sub_ps(iy0,jy0);
 260             dz00             = _mm256_sub_ps(iz0,jz0);
 261             dx10             = _mm256_sub_ps(ix1,jx0);
 262             dy10             = _mm256_sub_ps(iy1,jy0);
 263             dz10             = _mm256_sub_ps(iz1,jz0);
 264             dx20             = _mm256_sub_ps(ix2,jx0);
 265             dy20             = _mm256_sub_ps(iy2,jy0);
 266             dz20             = _mm256_sub_ps(iz2,jz0);
 267             dx30             = _mm256_sub_ps(ix3,jx0);
 268             dy30             = _mm256_sub_ps(iy3,jy0);
 269             dz30             = _mm256_sub_ps(iz3,jz0);
 270
 271             /* Calculate squared distance and things based on it */
 272             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 273             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 274             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 275             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 276
 277             rinv00           = avx256_invsqrt_f(rsq00);
 278             rinv10           = avx256_invsqrt_f(rsq10);
 279             rinv20           = avx256_invsqrt_f(rsq20);
 280             rinv30           = avx256_invsqrt_f(rsq30);
 281
 282             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 283             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 284             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 285             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 286
 287             /* Load parameters for j particles */
 288             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 289                                                                  charge+jnrC+0,charge+jnrD+0,
 290                                                                  charge+jnrE+0,charge+jnrF+0,
 291                                                                  charge+jnrG+0,charge+jnrH+0);
 292             vdwjidx0A        = 2*vdwtype[jnrA+0];
 293             vdwjidx0B        = 2*vdwtype[jnrB+0];
 294             vdwjidx0C        = 2*vdwtype[jnrC+0];
 295             vdwjidx0D        = 2*vdwtype[jnrD+0];
 296             vdwjidx0E        = 2*vdwtype[jnrE+0];
 297             vdwjidx0F        = 2*vdwtype[jnrF+0];
 298             vdwjidx0G        = 2*vdwtype[jnrG+0];
 299             vdwjidx0H        = 2*vdwtype[jnrH+0];
 300
 301             fjx0             = _mm256_setzero_ps();
 302             fjy0             = _mm256_setzero_ps();
 303             fjz0             = _mm256_setzero_ps();
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 310             {
 311
 312             r00              = _mm256_mul_ps(rsq00,rinv00);
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 316                                             vdwioffsetptr0+vdwjidx0B,
 317                                             vdwioffsetptr0+vdwjidx0C,
 318                                             vdwioffsetptr0+vdwjidx0D,
 319                                             vdwioffsetptr0+vdwjidx0E,
 320                                             vdwioffsetptr0+vdwjidx0F,
 321                                             vdwioffsetptr0+vdwjidx0G,
 322                                             vdwioffsetptr0+vdwjidx0H,
 323                                             &c6_00,&c12_00);
 324
 325             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 326                                                                   vdwgridioffsetptr0+vdwjidx0B,
 327                                                                   vdwgridioffsetptr0+vdwjidx0C,
 328                                                                   vdwgridioffsetptr0+vdwjidx0D,
 329                                                                   vdwgridioffsetptr0+vdwjidx0E,
 330                                                                   vdwgridioffsetptr0+vdwjidx0F,
 331                                                                   vdwgridioffsetptr0+vdwjidx0G,
 332                                                                   vdwgridioffsetptr0+vdwjidx0H);
 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             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 353             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 354
 355             fscal            = fvdw;
 356
 357             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm256_mul_ps(fscal,dx00);
 361             ty               = _mm256_mul_ps(fscal,dy00);
 362             tz               = _mm256_mul_ps(fscal,dz00);
 363
 364             /* Update vectorial force */
 365             fix0             = _mm256_add_ps(fix0,tx);
 366             fiy0             = _mm256_add_ps(fiy0,ty);
 367             fiz0             = _mm256_add_ps(fiz0,tz);
 368
 369             fjx0             = _mm256_add_ps(fjx0,tx);
 370             fjy0             = _mm256_add_ps(fjy0,ty);
 371             fjz0             = _mm256_add_ps(fjz0,tz);
 372
 373             }
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 380             {
 381
 382             r10              = _mm256_mul_ps(rsq10,rinv10);
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq10             = _mm256_mul_ps(iq1,jq0);
 386
 387             /* EWALD ELECTROSTATICS */
 388
 389             /* Analytical PME correction */
 390             zeta2            = _mm256_mul_ps(beta2,rsq10);
 391             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 392             pmecorrF         = avx256_pmecorrF_f(zeta2);
 393             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 394             felec            = _mm256_mul_ps(qq10,felec);
 395             pmecorrV         = avx256_pmecorrV_f(zeta2);
 396             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 397             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 398             velec            = _mm256_mul_ps(qq10,velec);
 399
 400             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 401
 402             /* Update potential sum for this i atom from the interaction with this j atom. */
 403             velec            = _mm256_and_ps(velec,cutoff_mask);
 404             velecsum         = _mm256_add_ps(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 409
 410             /* Calculate temporary vectorial force */
 411             tx               = _mm256_mul_ps(fscal,dx10);
 412             ty               = _mm256_mul_ps(fscal,dy10);
 413             tz               = _mm256_mul_ps(fscal,dz10);
 414
 415             /* Update vectorial force */
 416             fix1             = _mm256_add_ps(fix1,tx);
 417             fiy1             = _mm256_add_ps(fiy1,ty);
 418             fiz1             = _mm256_add_ps(fiz1,tz);
 419
 420             fjx0             = _mm256_add_ps(fjx0,tx);
 421             fjy0             = _mm256_add_ps(fjy0,ty);
 422             fjz0             = _mm256_add_ps(fjz0,tz);
 423
 424             }
 425
 426             /**************************
 427              * CALCULATE INTERACTIONS *
 428              **************************/
 429
 430             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 431             {
 432
 433             r20              = _mm256_mul_ps(rsq20,rinv20);
 434
 435             /* Compute parameters for interactions between i and j atoms */
 436             qq20             = _mm256_mul_ps(iq2,jq0);
 437
 438             /* EWALD ELECTROSTATICS */
 439
 440             /* Analytical PME correction */
 441             zeta2            = _mm256_mul_ps(beta2,rsq20);
 442             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 443             pmecorrF         = avx256_pmecorrF_f(zeta2);
 444             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 445             felec            = _mm256_mul_ps(qq20,felec);
 446             pmecorrV         = avx256_pmecorrV_f(zeta2);
 447             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 448             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 449             velec            = _mm256_mul_ps(qq20,velec);
 450
 451             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 452
 453             /* Update potential sum for this i atom from the interaction with this j atom. */
 454             velec            = _mm256_and_ps(velec,cutoff_mask);
 455             velecsum         = _mm256_add_ps(velecsum,velec);
 456
 457             fscal            = felec;
 458
 459             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 460
 461             /* Calculate temporary vectorial force */
 462             tx               = _mm256_mul_ps(fscal,dx20);
 463             ty               = _mm256_mul_ps(fscal,dy20);
 464             tz               = _mm256_mul_ps(fscal,dz20);
 465
 466             /* Update vectorial force */
 467             fix2             = _mm256_add_ps(fix2,tx);
 468             fiy2             = _mm256_add_ps(fiy2,ty);
 469             fiz2             = _mm256_add_ps(fiz2,tz);
 470
 471             fjx0             = _mm256_add_ps(fjx0,tx);
 472             fjy0             = _mm256_add_ps(fjy0,ty);
 473             fjz0             = _mm256_add_ps(fjz0,tz);
 474
 475             }
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 482             {
 483
 484             r30              = _mm256_mul_ps(rsq30,rinv30);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq30             = _mm256_mul_ps(iq3,jq0);
 488
 489             /* EWALD ELECTROSTATICS */
 490
 491             /* Analytical PME correction */
 492             zeta2            = _mm256_mul_ps(beta2,rsq30);
 493             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 494             pmecorrF         = avx256_pmecorrF_f(zeta2);
 495             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 496             felec            = _mm256_mul_ps(qq30,felec);
 497             pmecorrV         = avx256_pmecorrV_f(zeta2);
 498             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 499             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 500             velec            = _mm256_mul_ps(qq30,velec);
 501
 502             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velec            = _mm256_and_ps(velec,cutoff_mask);
 506             velecsum         = _mm256_add_ps(velecsum,velec);
 507
 508             fscal            = felec;
 509
 510             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 511
 512             /* Calculate temporary vectorial force */
 513             tx               = _mm256_mul_ps(fscal,dx30);
 514             ty               = _mm256_mul_ps(fscal,dy30);
 515             tz               = _mm256_mul_ps(fscal,dz30);
 516
 517             /* Update vectorial force */
 518             fix3             = _mm256_add_ps(fix3,tx);
 519             fiy3             = _mm256_add_ps(fiy3,ty);
 520             fiz3             = _mm256_add_ps(fiz3,tz);
 521
 522             fjx0             = _mm256_add_ps(fjx0,tx);
 523             fjy0             = _mm256_add_ps(fjy0,ty);
 524             fjz0             = _mm256_add_ps(fjz0,tz);
 525
 526             }
 527
 528             fjptrA             = f+j_coord_offsetA;
 529             fjptrB             = f+j_coord_offsetB;
 530             fjptrC             = f+j_coord_offsetC;
 531             fjptrD             = f+j_coord_offsetD;
 532             fjptrE             = f+j_coord_offsetE;
 533             fjptrF             = f+j_coord_offsetF;
 534             fjptrG             = f+j_coord_offsetG;
 535             fjptrH             = f+j_coord_offsetH;
 536
 537             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 538
 539             /* Inner loop uses 392 flops */
 540         }
 541
 542         if(jidx<j_index_end)
 543         {
 544
 545             /* Get j neighbor index, and coordinate index */
 546             jnrlistA         = jjnr[jidx];
 547             jnrlistB         = jjnr[jidx+1];
 548             jnrlistC         = jjnr[jidx+2];
 549             jnrlistD         = jjnr[jidx+3];
 550             jnrlistE         = jjnr[jidx+4];
 551             jnrlistF         = jjnr[jidx+5];
 552             jnrlistG         = jjnr[jidx+6];
 553             jnrlistH         = jjnr[jidx+7];
 554             /* Sign of each element will be negative for non-real atoms.
 555              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 556              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 557              */
 558             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 559                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 560
 561             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 562             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 563             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 564             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 565             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 566             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 567             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 568             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 569             j_coord_offsetA  = DIM*jnrA;
 570             j_coord_offsetB  = DIM*jnrB;
 571             j_coord_offsetC  = DIM*jnrC;
 572             j_coord_offsetD  = DIM*jnrD;
 573             j_coord_offsetE  = DIM*jnrE;
 574             j_coord_offsetF  = DIM*jnrF;
 575             j_coord_offsetG  = DIM*jnrG;
 576             j_coord_offsetH  = DIM*jnrH;
 577
 578             /* load j atom coordinates */
 579             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 580                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 581                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 582                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 583                                                  &jx0,&jy0,&jz0);
 584
 585             /* Calculate displacement vector */
 586             dx00             = _mm256_sub_ps(ix0,jx0);
 587             dy00             = _mm256_sub_ps(iy0,jy0);
 588             dz00             = _mm256_sub_ps(iz0,jz0);
 589             dx10             = _mm256_sub_ps(ix1,jx0);
 590             dy10             = _mm256_sub_ps(iy1,jy0);
 591             dz10             = _mm256_sub_ps(iz1,jz0);
 592             dx20             = _mm256_sub_ps(ix2,jx0);
 593             dy20             = _mm256_sub_ps(iy2,jy0);
 594             dz20             = _mm256_sub_ps(iz2,jz0);
 595             dx30             = _mm256_sub_ps(ix3,jx0);
 596             dy30             = _mm256_sub_ps(iy3,jy0);
 597             dz30             = _mm256_sub_ps(iz3,jz0);
 598
 599             /* Calculate squared distance and things based on it */
 600             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 601             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 602             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 603             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
 604
 605             rinv00           = avx256_invsqrt_f(rsq00);
 606             rinv10           = avx256_invsqrt_f(rsq10);
 607             rinv20           = avx256_invsqrt_f(rsq20);
 608             rinv30           = avx256_invsqrt_f(rsq30);
 609
 610             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 611             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 612             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 613             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
 614
 615             /* Load parameters for j particles */
 616             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 617                                                                  charge+jnrC+0,charge+jnrD+0,
 618                                                                  charge+jnrE+0,charge+jnrF+0,
 619                                                                  charge+jnrG+0,charge+jnrH+0);
 620             vdwjidx0A        = 2*vdwtype[jnrA+0];
 621             vdwjidx0B        = 2*vdwtype[jnrB+0];
 622             vdwjidx0C        = 2*vdwtype[jnrC+0];
 623             vdwjidx0D        = 2*vdwtype[jnrD+0];
 624             vdwjidx0E        = 2*vdwtype[jnrE+0];
 625             vdwjidx0F        = 2*vdwtype[jnrF+0];
 626             vdwjidx0G        = 2*vdwtype[jnrG+0];
 627             vdwjidx0H        = 2*vdwtype[jnrH+0];
 628
 629             fjx0             = _mm256_setzero_ps();
 630             fjy0             = _mm256_setzero_ps();
 631             fjz0             = _mm256_setzero_ps();
 632
 633             /**************************
 634              * CALCULATE INTERACTIONS *
 635              **************************/
 636
 637             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 638             {
 639
 640             r00              = _mm256_mul_ps(rsq00,rinv00);
 641             r00              = _mm256_andnot_ps(dummy_mask,r00);
 642
 643             /* Compute parameters for interactions between i and j atoms */
 644             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 645                                             vdwioffsetptr0+vdwjidx0B,
 646                                             vdwioffsetptr0+vdwjidx0C,
 647                                             vdwioffsetptr0+vdwjidx0D,
 648                                             vdwioffsetptr0+vdwjidx0E,
 649                                             vdwioffsetptr0+vdwjidx0F,
 650                                             vdwioffsetptr0+vdwjidx0G,
 651                                             vdwioffsetptr0+vdwjidx0H,
 652                                             &c6_00,&c12_00);
 653
 654             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 655                                                                   vdwgridioffsetptr0+vdwjidx0B,
 656                                                                   vdwgridioffsetptr0+vdwjidx0C,
 657                                                                   vdwgridioffsetptr0+vdwjidx0D,
 658                                                                   vdwgridioffsetptr0+vdwjidx0E,
 659                                                                   vdwgridioffsetptr0+vdwjidx0F,
 660                                                                   vdwgridioffsetptr0+vdwjidx0G,
 661                                                                   vdwgridioffsetptr0+vdwjidx0H);
 662
 663             /* Analytical LJ-PME */
 664             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 665             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 666             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 667             exponent         = avx256_exp_f(ewcljrsq);
 668             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 669             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 670             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 671             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 672             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 673             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) ,
 674                                           _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));
 675             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 676             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);
 677
 678             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 679
 680             /* Update potential sum for this i atom from the interaction with this j atom. */
 681             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 682             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 683             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 684
 685             fscal            = fvdw;
 686
 687             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 688
 689             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 690
 691             /* Calculate temporary vectorial force */
 692             tx               = _mm256_mul_ps(fscal,dx00);
 693             ty               = _mm256_mul_ps(fscal,dy00);
 694             tz               = _mm256_mul_ps(fscal,dz00);
 695
 696             /* Update vectorial force */
 697             fix0             = _mm256_add_ps(fix0,tx);
 698             fiy0             = _mm256_add_ps(fiy0,ty);
 699             fiz0             = _mm256_add_ps(fiz0,tz);
 700
 701             fjx0             = _mm256_add_ps(fjx0,tx);
 702             fjy0             = _mm256_add_ps(fjy0,ty);
 703             fjz0             = _mm256_add_ps(fjz0,tz);
 704
 705             }
 706
 707             /**************************
 708              * CALCULATE INTERACTIONS *
 709              **************************/
 710
 711             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 712             {
 713
 714             r10              = _mm256_mul_ps(rsq10,rinv10);
 715             r10              = _mm256_andnot_ps(dummy_mask,r10);
 716
 717             /* Compute parameters for interactions between i and j atoms */
 718             qq10             = _mm256_mul_ps(iq1,jq0);
 719
 720             /* EWALD ELECTROSTATICS */
 721
 722             /* Analytical PME correction */
 723             zeta2            = _mm256_mul_ps(beta2,rsq10);
 724             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 725             pmecorrF         = avx256_pmecorrF_f(zeta2);
 726             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 727             felec            = _mm256_mul_ps(qq10,felec);
 728             pmecorrV         = avx256_pmecorrV_f(zeta2);
 729             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 730             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 731             velec            = _mm256_mul_ps(qq10,velec);
 732
 733             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 734
 735             /* Update potential sum for this i atom from the interaction with this j atom. */
 736             velec            = _mm256_and_ps(velec,cutoff_mask);
 737             velec            = _mm256_andnot_ps(dummy_mask,velec);
 738             velecsum         = _mm256_add_ps(velecsum,velec);
 739
 740             fscal            = felec;
 741
 742             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 743
 744             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 745
 746             /* Calculate temporary vectorial force */
 747             tx               = _mm256_mul_ps(fscal,dx10);
 748             ty               = _mm256_mul_ps(fscal,dy10);
 749             tz               = _mm256_mul_ps(fscal,dz10);
 750
 751             /* Update vectorial force */
 752             fix1             = _mm256_add_ps(fix1,tx);
 753             fiy1             = _mm256_add_ps(fiy1,ty);
 754             fiz1             = _mm256_add_ps(fiz1,tz);
 755
 756             fjx0             = _mm256_add_ps(fjx0,tx);
 757             fjy0             = _mm256_add_ps(fjy0,ty);
 758             fjz0             = _mm256_add_ps(fjz0,tz);
 759
 760             }
 761
 762             /**************************
 763              * CALCULATE INTERACTIONS *
 764              **************************/
 765
 766             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 767             {
 768
 769             r20              = _mm256_mul_ps(rsq20,rinv20);
 770             r20              = _mm256_andnot_ps(dummy_mask,r20);
 771
 772             /* Compute parameters for interactions between i and j atoms */
 773             qq20             = _mm256_mul_ps(iq2,jq0);
 774
 775             /* EWALD ELECTROSTATICS */
 776
 777             /* Analytical PME correction */
 778             zeta2            = _mm256_mul_ps(beta2,rsq20);
 779             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 780             pmecorrF         = avx256_pmecorrF_f(zeta2);
 781             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 782             felec            = _mm256_mul_ps(qq20,felec);
 783             pmecorrV         = avx256_pmecorrV_f(zeta2);
 784             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 785             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 786             velec            = _mm256_mul_ps(qq20,velec);
 787
 788             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 789
 790             /* Update potential sum for this i atom from the interaction with this j atom. */
 791             velec            = _mm256_and_ps(velec,cutoff_mask);
 792             velec            = _mm256_andnot_ps(dummy_mask,velec);
 793             velecsum         = _mm256_add_ps(velecsum,velec);
 794
 795             fscal            = felec;
 796
 797             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 798
 799             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 800
 801             /* Calculate temporary vectorial force */
 802             tx               = _mm256_mul_ps(fscal,dx20);
 803             ty               = _mm256_mul_ps(fscal,dy20);
 804             tz               = _mm256_mul_ps(fscal,dz20);
 805
 806             /* Update vectorial force */
 807             fix2             = _mm256_add_ps(fix2,tx);
 808             fiy2             = _mm256_add_ps(fiy2,ty);
 809             fiz2             = _mm256_add_ps(fiz2,tz);
 810
 811             fjx0             = _mm256_add_ps(fjx0,tx);
 812             fjy0             = _mm256_add_ps(fjy0,ty);
 813             fjz0             = _mm256_add_ps(fjz0,tz);
 814
 815             }
 816
 817             /**************************
 818              * CALCULATE INTERACTIONS *
 819              **************************/
 820
 821             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 822             {
 823
 824             r30              = _mm256_mul_ps(rsq30,rinv30);
 825             r30              = _mm256_andnot_ps(dummy_mask,r30);
 826
 827             /* Compute parameters for interactions between i and j atoms */
 828             qq30             = _mm256_mul_ps(iq3,jq0);
 829
 830             /* EWALD ELECTROSTATICS */
 831
 832             /* Analytical PME correction */
 833             zeta2            = _mm256_mul_ps(beta2,rsq30);
 834             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
 835             pmecorrF         = avx256_pmecorrF_f(zeta2);
 836             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 837             felec            = _mm256_mul_ps(qq30,felec);
 838             pmecorrV         = avx256_pmecorrV_f(zeta2);
 839             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 840             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv30,sh_ewald),pmecorrV);
 841             velec            = _mm256_mul_ps(qq30,velec);
 842
 843             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
 844
 845             /* Update potential sum for this i atom from the interaction with this j atom. */
 846             velec            = _mm256_and_ps(velec,cutoff_mask);
 847             velec            = _mm256_andnot_ps(dummy_mask,velec);
 848             velecsum         = _mm256_add_ps(velecsum,velec);
 849
 850             fscal            = felec;
 851
 852             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 853
 854             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 855
 856             /* Calculate temporary vectorial force */
 857             tx               = _mm256_mul_ps(fscal,dx30);
 858             ty               = _mm256_mul_ps(fscal,dy30);
 859             tz               = _mm256_mul_ps(fscal,dz30);
 860
 861             /* Update vectorial force */
 862             fix3             = _mm256_add_ps(fix3,tx);
 863             fiy3             = _mm256_add_ps(fiy3,ty);
 864             fiz3             = _mm256_add_ps(fiz3,tz);
 865
 866             fjx0             = _mm256_add_ps(fjx0,tx);
 867             fjy0             = _mm256_add_ps(fjy0,ty);
 868             fjz0             = _mm256_add_ps(fjz0,tz);
 869
 870             }
 871
 872             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 873             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 874             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 875             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 876             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 877             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 878             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 879             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 880
 881             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 882
 883             /* Inner loop uses 396 flops */
 884         }
 885
 886         /* End of innermost loop */
 887
 888         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 889                                                  f+i_coord_offset,fshift+i_shift_offset);
 890
 891         ggid                        = gid[iidx];
 892         /* Update potential energies */
 893         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 894         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 895
 896         /* Increment number of inner iterations */
 897         inneriter                  += j_index_end - j_index_start;
 898
 899         /* Outer loop uses 26 flops */
 900     }
 901
 902     /* Increment number of outer iterations */
 903     outeriter        += nri;
 904
 905     /* Update outer/inner flops */
 906
 907     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*396);
 908 }
 909 /*
 910  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_256_single
 911  * Electrostatics interaction: Ewald
 912  * VdW interaction:            LJEwald
 913  * Geometry:                   Water4-Particle
 914  * Calculate force/pot:        Force
 915  */
 916 void
 917 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_256_single
 918                     (t_nblist                    * gmx_restrict       nlist,
 919                      rvec                        * gmx_restrict          xx,
 920                      rvec                        * gmx_restrict          ff,
 921                      struct t_forcerec           * gmx_restrict          fr,
 922                      t_mdatoms                   * gmx_restrict     mdatoms,
 923                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 924                      t_nrnb                      * gmx_restrict        nrnb)
 925 {
 926     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 927      * just 0 for non-waters.
 928      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 929      * jnr indices corresponding to data put in the four positions in the SIMD register.
 930      */
 931     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 932     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 933     int              jnrA,jnrB,jnrC,jnrD;
 934     int              jnrE,jnrF,jnrG,jnrH;
 935     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 936     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 937     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 938     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 939     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 940     real             rcutoff_scalar;
 941     real             *shiftvec,*fshift,*x,*f;
 942     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 943     real             scratch[4*DIM];
 944     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 945     real *           vdwioffsetptr0;
 946     real *           vdwgridioffsetptr0;
 947     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 948     real *           vdwioffsetptr1;
 949     real *           vdwgridioffsetptr1;
 950     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 951     real *           vdwioffsetptr2;
 952     real *           vdwgridioffsetptr2;
 953     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 954     real *           vdwioffsetptr3;
 955     real *           vdwgridioffsetptr3;
 956     __m256           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 957     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 958     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 959     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 960     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 961     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 962     __m256           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 963     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 964     real             *charge;
 965     int              nvdwtype;
 966     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 967     int              *vdwtype;
 968     real             *vdwparam;
 969     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 970     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 971     __m256           c6grid_00;
 972     __m256           c6grid_10;
 973     __m256           c6grid_20;
 974     __m256           c6grid_30;
 975     real             *vdwgridparam;
 976     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 977     __m256           one_half  = _mm256_set1_ps(0.5);
 978     __m256           minus_one = _mm256_set1_ps(-1.0);
 979     __m256i          ewitab;
 980     __m128i          ewitab_lo,ewitab_hi;
 981     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 982     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 983     real             *ewtab;
 984     __m256           dummy_mask,cutoff_mask;
 985     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 986     __m256           one     = _mm256_set1_ps(1.0);
 987     __m256           two     = _mm256_set1_ps(2.0);
 988     x                = xx[0];
 989     f                = ff[0];
 990
 991     nri              = nlist->nri;
 992     iinr             = nlist->iinr;
 993     jindex           = nlist->jindex;
 994     jjnr             = nlist->jjnr;
 995     shiftidx         = nlist->shift;
 996     gid              = nlist->gid;
 997     shiftvec         = fr->shift_vec[0];
 998     fshift           = fr->fshift[0];
 999     facel            = _mm256_set1_ps(fr->ic->epsfac);
1000     charge           = mdatoms->chargeA;
1001     nvdwtype         = fr->ntype;
1002     vdwparam         = fr->nbfp;
1003     vdwtype          = mdatoms->typeA;
1004     vdwgridparam     = fr->ljpme_c6grid;
1005     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
1006     ewclj            = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
1007     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
1008
1009     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
1010     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
1011     beta2            = _mm256_mul_ps(beta,beta);
1012     beta3            = _mm256_mul_ps(beta,beta2);
1013
1014     ewtab            = fr->ic->tabq_coul_F;
1015     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
1016     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
1017
1018     /* Setup water-specific parameters */
1019     inr              = nlist->iinr[0];
1020     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
1021     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
1022     iq3              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+3]));
1023     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
1024     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
1025
1026     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1027     rcutoff_scalar   = fr->ic->rcoulomb;
1028     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
1029     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
1030
1031     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
1032     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
1033
1034     /* Avoid stupid compiler warnings */
1035     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
1036     j_coord_offsetA = 0;
1037     j_coord_offsetB = 0;
1038     j_coord_offsetC = 0;
1039     j_coord_offsetD = 0;
1040     j_coord_offsetE = 0;
1041     j_coord_offsetF = 0;
1042     j_coord_offsetG = 0;
1043     j_coord_offsetH = 0;
1044
1045     outeriter        = 0;
1046     inneriter        = 0;
1047
1048     for(iidx=0;iidx<4*DIM;iidx++)
1049     {
1050         scratch[iidx] = 0.0;
1051     }
1052
1053     /* Start outer loop over neighborlists */
1054     for(iidx=0; iidx<nri; iidx++)
1055     {
1056         /* Load shift vector for this list */
1057         i_shift_offset   = DIM*shiftidx[iidx];
1058
1059         /* Load limits for loop over neighbors */
1060         j_index_start    = jindex[iidx];
1061         j_index_end      = jindex[iidx+1];
1062
1063         /* Get outer coordinate index */
1064         inr              = iinr[iidx];
1065         i_coord_offset   = DIM*inr;
1066
1067         /* Load i particle coords and add shift vector */
1068         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1069                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1070
1071         fix0             = _mm256_setzero_ps();
1072         fiy0             = _mm256_setzero_ps();
1073         fiz0             = _mm256_setzero_ps();
1074         fix1             = _mm256_setzero_ps();
1075         fiy1             = _mm256_setzero_ps();
1076         fiz1             = _mm256_setzero_ps();
1077         fix2             = _mm256_setzero_ps();
1078         fiy2             = _mm256_setzero_ps();
1079         fiz2             = _mm256_setzero_ps();
1080         fix3             = _mm256_setzero_ps();
1081         fiy3             = _mm256_setzero_ps();
1082         fiz3             = _mm256_setzero_ps();
1083
1084         /* Start inner kernel loop */
1085         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
1086         {
1087
1088             /* Get j neighbor index, and coordinate index */
1089             jnrA             = jjnr[jidx];
1090             jnrB             = jjnr[jidx+1];
1091             jnrC             = jjnr[jidx+2];
1092             jnrD             = jjnr[jidx+3];
1093             jnrE             = jjnr[jidx+4];
1094             jnrF             = jjnr[jidx+5];
1095             jnrG             = jjnr[jidx+6];
1096             jnrH             = jjnr[jidx+7];
1097             j_coord_offsetA  = DIM*jnrA;
1098             j_coord_offsetB  = DIM*jnrB;
1099             j_coord_offsetC  = DIM*jnrC;
1100             j_coord_offsetD  = DIM*jnrD;
1101             j_coord_offsetE  = DIM*jnrE;
1102             j_coord_offsetF  = DIM*jnrF;
1103             j_coord_offsetG  = DIM*jnrG;
1104             j_coord_offsetH  = DIM*jnrH;
1105
1106             /* load j atom coordinates */
1107             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1108                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1109                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1110                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1111                                                  &jx0,&jy0,&jz0);
1112
1113             /* Calculate displacement vector */
1114             dx00             = _mm256_sub_ps(ix0,jx0);
1115             dy00             = _mm256_sub_ps(iy0,jy0);
1116             dz00             = _mm256_sub_ps(iz0,jz0);
1117             dx10             = _mm256_sub_ps(ix1,jx0);
1118             dy10             = _mm256_sub_ps(iy1,jy0);
1119             dz10             = _mm256_sub_ps(iz1,jz0);
1120             dx20             = _mm256_sub_ps(ix2,jx0);
1121             dy20             = _mm256_sub_ps(iy2,jy0);
1122             dz20             = _mm256_sub_ps(iz2,jz0);
1123             dx30             = _mm256_sub_ps(ix3,jx0);
1124             dy30             = _mm256_sub_ps(iy3,jy0);
1125             dz30             = _mm256_sub_ps(iz3,jz0);
1126
1127             /* Calculate squared distance and things based on it */
1128             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1129             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1130             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1131             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1132
1133             rinv00           = avx256_invsqrt_f(rsq00);
1134             rinv10           = avx256_invsqrt_f(rsq10);
1135             rinv20           = avx256_invsqrt_f(rsq20);
1136             rinv30           = avx256_invsqrt_f(rsq30);
1137
1138             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1139             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1140             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1141             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1142
1143             /* Load parameters for j particles */
1144             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1145                                                                  charge+jnrC+0,charge+jnrD+0,
1146                                                                  charge+jnrE+0,charge+jnrF+0,
1147                                                                  charge+jnrG+0,charge+jnrH+0);
1148             vdwjidx0A        = 2*vdwtype[jnrA+0];
1149             vdwjidx0B        = 2*vdwtype[jnrB+0];
1150             vdwjidx0C        = 2*vdwtype[jnrC+0];
1151             vdwjidx0D        = 2*vdwtype[jnrD+0];
1152             vdwjidx0E        = 2*vdwtype[jnrE+0];
1153             vdwjidx0F        = 2*vdwtype[jnrF+0];
1154             vdwjidx0G        = 2*vdwtype[jnrG+0];
1155             vdwjidx0H        = 2*vdwtype[jnrH+0];
1156
1157             fjx0             = _mm256_setzero_ps();
1158             fjy0             = _mm256_setzero_ps();
1159             fjz0             = _mm256_setzero_ps();
1160
1161             /**************************
1162              * CALCULATE INTERACTIONS *
1163              **************************/
1164
1165             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1166             {
1167
1168             r00              = _mm256_mul_ps(rsq00,rinv00);
1169
1170             /* Compute parameters for interactions between i and j atoms */
1171             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1172                                             vdwioffsetptr0+vdwjidx0B,
1173                                             vdwioffsetptr0+vdwjidx0C,
1174                                             vdwioffsetptr0+vdwjidx0D,
1175                                             vdwioffsetptr0+vdwjidx0E,
1176                                             vdwioffsetptr0+vdwjidx0F,
1177                                             vdwioffsetptr0+vdwjidx0G,
1178                                             vdwioffsetptr0+vdwjidx0H,
1179                                             &c6_00,&c12_00);
1180
1181             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1182                                                                   vdwgridioffsetptr0+vdwjidx0B,
1183                                                                   vdwgridioffsetptr0+vdwjidx0C,
1184                                                                   vdwgridioffsetptr0+vdwjidx0D,
1185                                                                   vdwgridioffsetptr0+vdwjidx0E,
1186                                                                   vdwgridioffsetptr0+vdwjidx0F,
1187                                                                   vdwgridioffsetptr0+vdwjidx0G,
1188                                                                   vdwgridioffsetptr0+vdwjidx0H);
1189
1190             /* Analytical LJ-PME */
1191             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1192             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1193             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1194             exponent         = avx256_exp_f(ewcljrsq);
1195             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1196             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1197             /* f6A = 6 * C6grid * (1 - poly) */
1198             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1199             /* f6B = C6grid * exponent * beta^6 */
1200             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1201             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1202             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);
1203
1204             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1205
1206             fscal            = fvdw;
1207
1208             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1209
1210             /* Calculate temporary vectorial force */
1211             tx               = _mm256_mul_ps(fscal,dx00);
1212             ty               = _mm256_mul_ps(fscal,dy00);
1213             tz               = _mm256_mul_ps(fscal,dz00);
1214
1215             /* Update vectorial force */
1216             fix0             = _mm256_add_ps(fix0,tx);
1217             fiy0             = _mm256_add_ps(fiy0,ty);
1218             fiz0             = _mm256_add_ps(fiz0,tz);
1219
1220             fjx0             = _mm256_add_ps(fjx0,tx);
1221             fjy0             = _mm256_add_ps(fjy0,ty);
1222             fjz0             = _mm256_add_ps(fjz0,tz);
1223
1224             }
1225
1226             /**************************
1227              * CALCULATE INTERACTIONS *
1228              **************************/
1229
1230             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1231             {
1232
1233             r10              = _mm256_mul_ps(rsq10,rinv10);
1234
1235             /* Compute parameters for interactions between i and j atoms */
1236             qq10             = _mm256_mul_ps(iq1,jq0);
1237
1238             /* EWALD ELECTROSTATICS */
1239
1240             /* Analytical PME correction */
1241             zeta2            = _mm256_mul_ps(beta2,rsq10);
1242             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1243             pmecorrF         = avx256_pmecorrF_f(zeta2);
1244             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1245             felec            = _mm256_mul_ps(qq10,felec);
1246
1247             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1248
1249             fscal            = felec;
1250
1251             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1252
1253             /* Calculate temporary vectorial force */
1254             tx               = _mm256_mul_ps(fscal,dx10);
1255             ty               = _mm256_mul_ps(fscal,dy10);
1256             tz               = _mm256_mul_ps(fscal,dz10);
1257
1258             /* Update vectorial force */
1259             fix1             = _mm256_add_ps(fix1,tx);
1260             fiy1             = _mm256_add_ps(fiy1,ty);
1261             fiz1             = _mm256_add_ps(fiz1,tz);
1262
1263             fjx0             = _mm256_add_ps(fjx0,tx);
1264             fjy0             = _mm256_add_ps(fjy0,ty);
1265             fjz0             = _mm256_add_ps(fjz0,tz);
1266
1267             }
1268
1269             /**************************
1270              * CALCULATE INTERACTIONS *
1271              **************************/
1272
1273             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1274             {
1275
1276             r20              = _mm256_mul_ps(rsq20,rinv20);
1277
1278             /* Compute parameters for interactions between i and j atoms */
1279             qq20             = _mm256_mul_ps(iq2,jq0);
1280
1281             /* EWALD ELECTROSTATICS */
1282
1283             /* Analytical PME correction */
1284             zeta2            = _mm256_mul_ps(beta2,rsq20);
1285             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1286             pmecorrF         = avx256_pmecorrF_f(zeta2);
1287             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1288             felec            = _mm256_mul_ps(qq20,felec);
1289
1290             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1291
1292             fscal            = felec;
1293
1294             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1295
1296             /* Calculate temporary vectorial force */
1297             tx               = _mm256_mul_ps(fscal,dx20);
1298             ty               = _mm256_mul_ps(fscal,dy20);
1299             tz               = _mm256_mul_ps(fscal,dz20);
1300
1301             /* Update vectorial force */
1302             fix2             = _mm256_add_ps(fix2,tx);
1303             fiy2             = _mm256_add_ps(fiy2,ty);
1304             fiz2             = _mm256_add_ps(fiz2,tz);
1305
1306             fjx0             = _mm256_add_ps(fjx0,tx);
1307             fjy0             = _mm256_add_ps(fjy0,ty);
1308             fjz0             = _mm256_add_ps(fjz0,tz);
1309
1310             }
1311
1312             /**************************
1313              * CALCULATE INTERACTIONS *
1314              **************************/
1315
1316             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1317             {
1318
1319             r30              = _mm256_mul_ps(rsq30,rinv30);
1320
1321             /* Compute parameters for interactions between i and j atoms */
1322             qq30             = _mm256_mul_ps(iq3,jq0);
1323
1324             /* EWALD ELECTROSTATICS */
1325
1326             /* Analytical PME correction */
1327             zeta2            = _mm256_mul_ps(beta2,rsq30);
1328             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1329             pmecorrF         = avx256_pmecorrF_f(zeta2);
1330             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1331             felec            = _mm256_mul_ps(qq30,felec);
1332
1333             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1334
1335             fscal            = felec;
1336
1337             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1338
1339             /* Calculate temporary vectorial force */
1340             tx               = _mm256_mul_ps(fscal,dx30);
1341             ty               = _mm256_mul_ps(fscal,dy30);
1342             tz               = _mm256_mul_ps(fscal,dz30);
1343
1344             /* Update vectorial force */
1345             fix3             = _mm256_add_ps(fix3,tx);
1346             fiy3             = _mm256_add_ps(fiy3,ty);
1347             fiz3             = _mm256_add_ps(fiz3,tz);
1348
1349             fjx0             = _mm256_add_ps(fjx0,tx);
1350             fjy0             = _mm256_add_ps(fjy0,ty);
1351             fjz0             = _mm256_add_ps(fjz0,tz);
1352
1353             }
1354
1355             fjptrA             = f+j_coord_offsetA;
1356             fjptrB             = f+j_coord_offsetB;
1357             fjptrC             = f+j_coord_offsetC;
1358             fjptrD             = f+j_coord_offsetD;
1359             fjptrE             = f+j_coord_offsetE;
1360             fjptrF             = f+j_coord_offsetF;
1361             fjptrG             = f+j_coord_offsetG;
1362             fjptrH             = f+j_coord_offsetH;
1363
1364             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1365
1366             /* Inner loop uses 229 flops */
1367         }
1368
1369         if(jidx<j_index_end)
1370         {
1371
1372             /* Get j neighbor index, and coordinate index */
1373             jnrlistA         = jjnr[jidx];
1374             jnrlistB         = jjnr[jidx+1];
1375             jnrlistC         = jjnr[jidx+2];
1376             jnrlistD         = jjnr[jidx+3];
1377             jnrlistE         = jjnr[jidx+4];
1378             jnrlistF         = jjnr[jidx+5];
1379             jnrlistG         = jjnr[jidx+6];
1380             jnrlistH         = jjnr[jidx+7];
1381             /* Sign of each element will be negative for non-real atoms.
1382              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1383              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1384              */
1385             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1386                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1387
1388             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1389             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1390             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1391             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1392             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1393             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1394             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1395             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1396             j_coord_offsetA  = DIM*jnrA;
1397             j_coord_offsetB  = DIM*jnrB;
1398             j_coord_offsetC  = DIM*jnrC;
1399             j_coord_offsetD  = DIM*jnrD;
1400             j_coord_offsetE  = DIM*jnrE;
1401             j_coord_offsetF  = DIM*jnrF;
1402             j_coord_offsetG  = DIM*jnrG;
1403             j_coord_offsetH  = DIM*jnrH;
1404
1405             /* load j atom coordinates */
1406             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1407                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1408                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1409                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1410                                                  &jx0,&jy0,&jz0);
1411
1412             /* Calculate displacement vector */
1413             dx00             = _mm256_sub_ps(ix0,jx0);
1414             dy00             = _mm256_sub_ps(iy0,jy0);
1415             dz00             = _mm256_sub_ps(iz0,jz0);
1416             dx10             = _mm256_sub_ps(ix1,jx0);
1417             dy10             = _mm256_sub_ps(iy1,jy0);
1418             dz10             = _mm256_sub_ps(iz1,jz0);
1419             dx20             = _mm256_sub_ps(ix2,jx0);
1420             dy20             = _mm256_sub_ps(iy2,jy0);
1421             dz20             = _mm256_sub_ps(iz2,jz0);
1422             dx30             = _mm256_sub_ps(ix3,jx0);
1423             dy30             = _mm256_sub_ps(iy3,jy0);
1424             dz30             = _mm256_sub_ps(iz3,jz0);
1425
1426             /* Calculate squared distance and things based on it */
1427             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1428             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1429             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1430             rsq30            = gmx_mm256_calc_rsq_ps(dx30,dy30,dz30);
1431
1432             rinv00           = avx256_invsqrt_f(rsq00);
1433             rinv10           = avx256_invsqrt_f(rsq10);
1434             rinv20           = avx256_invsqrt_f(rsq20);
1435             rinv30           = avx256_invsqrt_f(rsq30);
1436
1437             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1438             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1439             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1440             rinvsq30         = _mm256_mul_ps(rinv30,rinv30);
1441
1442             /* Load parameters for j particles */
1443             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1444                                                                  charge+jnrC+0,charge+jnrD+0,
1445                                                                  charge+jnrE+0,charge+jnrF+0,
1446                                                                  charge+jnrG+0,charge+jnrH+0);
1447             vdwjidx0A        = 2*vdwtype[jnrA+0];
1448             vdwjidx0B        = 2*vdwtype[jnrB+0];
1449             vdwjidx0C        = 2*vdwtype[jnrC+0];
1450             vdwjidx0D        = 2*vdwtype[jnrD+0];
1451             vdwjidx0E        = 2*vdwtype[jnrE+0];
1452             vdwjidx0F        = 2*vdwtype[jnrF+0];
1453             vdwjidx0G        = 2*vdwtype[jnrG+0];
1454             vdwjidx0H        = 2*vdwtype[jnrH+0];
1455
1456             fjx0             = _mm256_setzero_ps();
1457             fjy0             = _mm256_setzero_ps();
1458             fjz0             = _mm256_setzero_ps();
1459
1460             /**************************
1461              * CALCULATE INTERACTIONS *
1462              **************************/
1463
1464             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1465             {
1466
1467             r00              = _mm256_mul_ps(rsq00,rinv00);
1468             r00              = _mm256_andnot_ps(dummy_mask,r00);
1469
1470             /* Compute parameters for interactions between i and j atoms */
1471             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1472                                             vdwioffsetptr0+vdwjidx0B,
1473                                             vdwioffsetptr0+vdwjidx0C,
1474                                             vdwioffsetptr0+vdwjidx0D,
1475                                             vdwioffsetptr0+vdwjidx0E,
1476                                             vdwioffsetptr0+vdwjidx0F,
1477                                             vdwioffsetptr0+vdwjidx0G,
1478                                             vdwioffsetptr0+vdwjidx0H,
1479                                             &c6_00,&c12_00);
1480
1481             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1482                                                                   vdwgridioffsetptr0+vdwjidx0B,
1483                                                                   vdwgridioffsetptr0+vdwjidx0C,
1484                                                                   vdwgridioffsetptr0+vdwjidx0D,
1485                                                                   vdwgridioffsetptr0+vdwjidx0E,
1486                                                                   vdwgridioffsetptr0+vdwjidx0F,
1487                                                                   vdwgridioffsetptr0+vdwjidx0G,
1488                                                                   vdwgridioffsetptr0+vdwjidx0H);
1489
1490             /* Analytical LJ-PME */
1491             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1492             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1493             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1494             exponent         = avx256_exp_f(ewcljrsq);
1495             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1496             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1497             /* f6A = 6 * C6grid * (1 - poly) */
1498             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1499             /* f6B = C6grid * exponent * beta^6 */
1500             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1501             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1502             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);
1503
1504             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1505
1506             fscal            = fvdw;
1507
1508             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1509
1510             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1511
1512             /* Calculate temporary vectorial force */
1513             tx               = _mm256_mul_ps(fscal,dx00);
1514             ty               = _mm256_mul_ps(fscal,dy00);
1515             tz               = _mm256_mul_ps(fscal,dz00);
1516
1517             /* Update vectorial force */
1518             fix0             = _mm256_add_ps(fix0,tx);
1519             fiy0             = _mm256_add_ps(fiy0,ty);
1520             fiz0             = _mm256_add_ps(fiz0,tz);
1521
1522             fjx0             = _mm256_add_ps(fjx0,tx);
1523             fjy0             = _mm256_add_ps(fjy0,ty);
1524             fjz0             = _mm256_add_ps(fjz0,tz);
1525
1526             }
1527
1528             /**************************
1529              * CALCULATE INTERACTIONS *
1530              **************************/
1531
1532             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1533             {
1534
1535             r10              = _mm256_mul_ps(rsq10,rinv10);
1536             r10              = _mm256_andnot_ps(dummy_mask,r10);
1537
1538             /* Compute parameters for interactions between i and j atoms */
1539             qq10             = _mm256_mul_ps(iq1,jq0);
1540
1541             /* EWALD ELECTROSTATICS */
1542
1543             /* Analytical PME correction */
1544             zeta2            = _mm256_mul_ps(beta2,rsq10);
1545             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1546             pmecorrF         = avx256_pmecorrF_f(zeta2);
1547             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1548             felec            = _mm256_mul_ps(qq10,felec);
1549
1550             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1551
1552             fscal            = felec;
1553
1554             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1555
1556             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1557
1558             /* Calculate temporary vectorial force */
1559             tx               = _mm256_mul_ps(fscal,dx10);
1560             ty               = _mm256_mul_ps(fscal,dy10);
1561             tz               = _mm256_mul_ps(fscal,dz10);
1562
1563             /* Update vectorial force */
1564             fix1             = _mm256_add_ps(fix1,tx);
1565             fiy1             = _mm256_add_ps(fiy1,ty);
1566             fiz1             = _mm256_add_ps(fiz1,tz);
1567
1568             fjx0             = _mm256_add_ps(fjx0,tx);
1569             fjy0             = _mm256_add_ps(fjy0,ty);
1570             fjz0             = _mm256_add_ps(fjz0,tz);
1571
1572             }
1573
1574             /**************************
1575              * CALCULATE INTERACTIONS *
1576              **************************/
1577
1578             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1579             {
1580
1581             r20              = _mm256_mul_ps(rsq20,rinv20);
1582             r20              = _mm256_andnot_ps(dummy_mask,r20);
1583
1584             /* Compute parameters for interactions between i and j atoms */
1585             qq20             = _mm256_mul_ps(iq2,jq0);
1586
1587             /* EWALD ELECTROSTATICS */
1588
1589             /* Analytical PME correction */
1590             zeta2            = _mm256_mul_ps(beta2,rsq20);
1591             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1592             pmecorrF         = avx256_pmecorrF_f(zeta2);
1593             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1594             felec            = _mm256_mul_ps(qq20,felec);
1595
1596             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1597
1598             fscal            = felec;
1599
1600             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1601
1602             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1603
1604             /* Calculate temporary vectorial force */
1605             tx               = _mm256_mul_ps(fscal,dx20);
1606             ty               = _mm256_mul_ps(fscal,dy20);
1607             tz               = _mm256_mul_ps(fscal,dz20);
1608
1609             /* Update vectorial force */
1610             fix2             = _mm256_add_ps(fix2,tx);
1611             fiy2             = _mm256_add_ps(fiy2,ty);
1612             fiz2             = _mm256_add_ps(fiz2,tz);
1613
1614             fjx0             = _mm256_add_ps(fjx0,tx);
1615             fjy0             = _mm256_add_ps(fjy0,ty);
1616             fjz0             = _mm256_add_ps(fjz0,tz);
1617
1618             }
1619
1620             /**************************
1621              * CALCULATE INTERACTIONS *
1622              **************************/
1623
1624             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1625             {
1626
1627             r30              = _mm256_mul_ps(rsq30,rinv30);
1628             r30              = _mm256_andnot_ps(dummy_mask,r30);
1629
1630             /* Compute parameters for interactions between i and j atoms */
1631             qq30             = _mm256_mul_ps(iq3,jq0);
1632
1633             /* EWALD ELECTROSTATICS */
1634
1635             /* Analytical PME correction */
1636             zeta2            = _mm256_mul_ps(beta2,rsq30);
1637             rinv3            = _mm256_mul_ps(rinvsq30,rinv30);
1638             pmecorrF         = avx256_pmecorrF_f(zeta2);
1639             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1640             felec            = _mm256_mul_ps(qq30,felec);
1641
1642             cutoff_mask      = _mm256_cmp_ps(rsq30,rcutoff2,_CMP_LT_OQ);
1643
1644             fscal            = felec;
1645
1646             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1647
1648             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1649
1650             /* Calculate temporary vectorial force */
1651             tx               = _mm256_mul_ps(fscal,dx30);
1652             ty               = _mm256_mul_ps(fscal,dy30);
1653             tz               = _mm256_mul_ps(fscal,dz30);
1654
1655             /* Update vectorial force */
1656             fix3             = _mm256_add_ps(fix3,tx);
1657             fiy3             = _mm256_add_ps(fiy3,ty);
1658             fiz3             = _mm256_add_ps(fiz3,tz);
1659
1660             fjx0             = _mm256_add_ps(fjx0,tx);
1661             fjy0             = _mm256_add_ps(fjy0,ty);
1662             fjz0             = _mm256_add_ps(fjz0,tz);
1663
1664             }
1665
1666             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1667             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1668             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1669             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1670             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1671             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1672             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1673             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1674
1675             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1676
1677             /* Inner loop uses 233 flops */
1678         }
1679
1680         /* End of innermost loop */
1681
1682         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1683                                                  f+i_coord_offset,fshift+i_shift_offset);
1684
1685         /* Increment number of inner iterations */
1686         inneriter                  += j_index_end - j_index_start;
1687
1688         /* Outer loop uses 24 flops */
1689     }
1690
1691     /* Increment number of outer iterations */
1692     outeriter        += nri;
1693
1694     /* Update outer/inner flops */
1695
1696     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*233);
1697 }