src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_avx_256_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_ElecEw_VdwLJEw_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_ElecEw_VdwLJEw_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     /* Avoid stupid compiler warnings */
 162     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 163     j_coord_offsetA = 0;
 164     j_coord_offsetB = 0;
 165     j_coord_offsetC = 0;
 166     j_coord_offsetD = 0;
 167     j_coord_offsetE = 0;
 168     j_coord_offsetF = 0;
 169     j_coord_offsetG = 0;
 170     j_coord_offsetH = 0;
 171
 172     outeriter        = 0;
 173     inneriter        = 0;
 174
 175     for(iidx=0;iidx<4*DIM;iidx++)
 176     {
 177         scratch[iidx] = 0.0;
 178     }
 179
 180     /* Start outer loop over neighborlists */
 181     for(iidx=0; iidx<nri; iidx++)
 182     {
 183         /* Load shift vector for this list */
 184         i_shift_offset   = DIM*shiftidx[iidx];
 185
 186         /* Load limits for loop over neighbors */
 187         j_index_start    = jindex[iidx];
 188         j_index_end      = jindex[iidx+1];
 189
 190         /* Get outer coordinate index */
 191         inr              = iinr[iidx];
 192         i_coord_offset   = DIM*inr;
 193
 194         /* Load i particle coords and add shift vector */
 195         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 196                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 197
 198         fix0             = _mm256_setzero_ps();
 199         fiy0             = _mm256_setzero_ps();
 200         fiz0             = _mm256_setzero_ps();
 201         fix1             = _mm256_setzero_ps();
 202         fiy1             = _mm256_setzero_ps();
 203         fiz1             = _mm256_setzero_ps();
 204         fix2             = _mm256_setzero_ps();
 205         fiy2             = _mm256_setzero_ps();
 206         fiz2             = _mm256_setzero_ps();
 207
 208         /* Reset potential sums */
 209         velecsum         = _mm256_setzero_ps();
 210         vvdwsum          = _mm256_setzero_ps();
 211
 212         /* Start inner kernel loop */
 213         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 214         {
 215
 216             /* Get j neighbor index, and coordinate index */
 217             jnrA             = jjnr[jidx];
 218             jnrB             = jjnr[jidx+1];
 219             jnrC             = jjnr[jidx+2];
 220             jnrD             = jjnr[jidx+3];
 221             jnrE             = jjnr[jidx+4];
 222             jnrF             = jjnr[jidx+5];
 223             jnrG             = jjnr[jidx+6];
 224             jnrH             = jjnr[jidx+7];
 225             j_coord_offsetA  = DIM*jnrA;
 226             j_coord_offsetB  = DIM*jnrB;
 227             j_coord_offsetC  = DIM*jnrC;
 228             j_coord_offsetD  = DIM*jnrD;
 229             j_coord_offsetE  = DIM*jnrE;
 230             j_coord_offsetF  = DIM*jnrF;
 231             j_coord_offsetG  = DIM*jnrG;
 232             j_coord_offsetH  = DIM*jnrH;
 233
 234             /* load j atom coordinates */
 235             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 236                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 237                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 238                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 239                                                  &jx0,&jy0,&jz0);
 240
 241             /* Calculate displacement vector */
 242             dx00             = _mm256_sub_ps(ix0,jx0);
 243             dy00             = _mm256_sub_ps(iy0,jy0);
 244             dz00             = _mm256_sub_ps(iz0,jz0);
 245             dx10             = _mm256_sub_ps(ix1,jx0);
 246             dy10             = _mm256_sub_ps(iy1,jy0);
 247             dz10             = _mm256_sub_ps(iz1,jz0);
 248             dx20             = _mm256_sub_ps(ix2,jx0);
 249             dy20             = _mm256_sub_ps(iy2,jy0);
 250             dz20             = _mm256_sub_ps(iz2,jz0);
 251
 252             /* Calculate squared distance and things based on it */
 253             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 254             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 255             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 256
 257             rinv00           = avx256_invsqrt_f(rsq00);
 258             rinv10           = avx256_invsqrt_f(rsq10);
 259             rinv20           = avx256_invsqrt_f(rsq20);
 260
 261             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 262             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 263             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 264
 265             /* Load parameters for j particles */
 266             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 267                                                                  charge+jnrC+0,charge+jnrD+0,
 268                                                                  charge+jnrE+0,charge+jnrF+0,
 269                                                                  charge+jnrG+0,charge+jnrH+0);
 270             vdwjidx0A        = 2*vdwtype[jnrA+0];
 271             vdwjidx0B        = 2*vdwtype[jnrB+0];
 272             vdwjidx0C        = 2*vdwtype[jnrC+0];
 273             vdwjidx0D        = 2*vdwtype[jnrD+0];
 274             vdwjidx0E        = 2*vdwtype[jnrE+0];
 275             vdwjidx0F        = 2*vdwtype[jnrF+0];
 276             vdwjidx0G        = 2*vdwtype[jnrG+0];
 277             vdwjidx0H        = 2*vdwtype[jnrH+0];
 278
 279             fjx0             = _mm256_setzero_ps();
 280             fjy0             = _mm256_setzero_ps();
 281             fjz0             = _mm256_setzero_ps();
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             r00              = _mm256_mul_ps(rsq00,rinv00);
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq00             = _mm256_mul_ps(iq0,jq0);
 291             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 292                                             vdwioffsetptr0+vdwjidx0B,
 293                                             vdwioffsetptr0+vdwjidx0C,
 294                                             vdwioffsetptr0+vdwjidx0D,
 295                                             vdwioffsetptr0+vdwjidx0E,
 296                                             vdwioffsetptr0+vdwjidx0F,
 297                                             vdwioffsetptr0+vdwjidx0G,
 298                                             vdwioffsetptr0+vdwjidx0H,
 299                                             &c6_00,&c12_00);
 300
 301             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 302                                                                   vdwgridioffsetptr0+vdwjidx0B,
 303                                                                   vdwgridioffsetptr0+vdwjidx0C,
 304                                                                   vdwgridioffsetptr0+vdwjidx0D,
 305                                                                   vdwgridioffsetptr0+vdwjidx0E,
 306                                                                   vdwgridioffsetptr0+vdwjidx0F,
 307                                                                   vdwgridioffsetptr0+vdwjidx0G,
 308                                                                   vdwgridioffsetptr0+vdwjidx0H);
 309
 310             /* EWALD ELECTROSTATICS */
 311
 312             /* Analytical PME correction */
 313             zeta2            = _mm256_mul_ps(beta2,rsq00);
 314             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 315             pmecorrF         = avx256_pmecorrF_f(zeta2);
 316             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 317             felec            = _mm256_mul_ps(qq00,felec);
 318             pmecorrV         = avx256_pmecorrV_f(zeta2);
 319             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 320             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 321             velec            = _mm256_mul_ps(qq00,velec);
 322
 323             /* Analytical LJ-PME */
 324             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 325             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 326             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 327             exponent         = avx256_exp_f(ewcljrsq);
 328             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 329             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 330             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 331             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 332             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 333             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 334             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 335             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);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velecsum         = _mm256_add_ps(velecsum,velec);
 339             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 340
 341             fscal            = _mm256_add_ps(felec,fvdw);
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm256_mul_ps(fscal,dx00);
 345             ty               = _mm256_mul_ps(fscal,dy00);
 346             tz               = _mm256_mul_ps(fscal,dz00);
 347
 348             /* Update vectorial force */
 349             fix0             = _mm256_add_ps(fix0,tx);
 350             fiy0             = _mm256_add_ps(fiy0,ty);
 351             fiz0             = _mm256_add_ps(fiz0,tz);
 352
 353             fjx0             = _mm256_add_ps(fjx0,tx);
 354             fjy0             = _mm256_add_ps(fjy0,ty);
 355             fjz0             = _mm256_add_ps(fjz0,tz);
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             r10              = _mm256_mul_ps(rsq10,rinv10);
 362
 363             /* Compute parameters for interactions between i and j atoms */
 364             qq10             = _mm256_mul_ps(iq1,jq0);
 365
 366             /* EWALD ELECTROSTATICS */
 367
 368             /* Analytical PME correction */
 369             zeta2            = _mm256_mul_ps(beta2,rsq10);
 370             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 371             pmecorrF         = avx256_pmecorrF_f(zeta2);
 372             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 373             felec            = _mm256_mul_ps(qq10,felec);
 374             pmecorrV         = avx256_pmecorrV_f(zeta2);
 375             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 376             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 377             velec            = _mm256_mul_ps(qq10,velec);
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm256_add_ps(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             /* Calculate temporary vectorial force */
 385             tx               = _mm256_mul_ps(fscal,dx10);
 386             ty               = _mm256_mul_ps(fscal,dy10);
 387             tz               = _mm256_mul_ps(fscal,dz10);
 388
 389             /* Update vectorial force */
 390             fix1             = _mm256_add_ps(fix1,tx);
 391             fiy1             = _mm256_add_ps(fiy1,ty);
 392             fiz1             = _mm256_add_ps(fiz1,tz);
 393
 394             fjx0             = _mm256_add_ps(fjx0,tx);
 395             fjy0             = _mm256_add_ps(fjy0,ty);
 396             fjz0             = _mm256_add_ps(fjz0,tz);
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             r20              = _mm256_mul_ps(rsq20,rinv20);
 403
 404             /* Compute parameters for interactions between i and j atoms */
 405             qq20             = _mm256_mul_ps(iq2,jq0);
 406
 407             /* EWALD ELECTROSTATICS */
 408
 409             /* Analytical PME correction */
 410             zeta2            = _mm256_mul_ps(beta2,rsq20);
 411             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 412             pmecorrF         = avx256_pmecorrF_f(zeta2);
 413             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 414             felec            = _mm256_mul_ps(qq20,felec);
 415             pmecorrV         = avx256_pmecorrV_f(zeta2);
 416             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 417             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 418             velec            = _mm256_mul_ps(qq20,velec);
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velecsum         = _mm256_add_ps(velecsum,velec);
 422
 423             fscal            = felec;
 424
 425             /* Calculate temporary vectorial force */
 426             tx               = _mm256_mul_ps(fscal,dx20);
 427             ty               = _mm256_mul_ps(fscal,dy20);
 428             tz               = _mm256_mul_ps(fscal,dz20);
 429
 430             /* Update vectorial force */
 431             fix2             = _mm256_add_ps(fix2,tx);
 432             fiy2             = _mm256_add_ps(fiy2,ty);
 433             fiz2             = _mm256_add_ps(fiz2,tz);
 434
 435             fjx0             = _mm256_add_ps(fjx0,tx);
 436             fjy0             = _mm256_add_ps(fjy0,ty);
 437             fjz0             = _mm256_add_ps(fjz0,tz);
 438
 439             fjptrA             = f+j_coord_offsetA;
 440             fjptrB             = f+j_coord_offsetB;
 441             fjptrC             = f+j_coord_offsetC;
 442             fjptrD             = f+j_coord_offsetD;
 443             fjptrE             = f+j_coord_offsetE;
 444             fjptrF             = f+j_coord_offsetF;
 445             fjptrG             = f+j_coord_offsetG;
 446             fjptrH             = f+j_coord_offsetH;
 447
 448             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 449
 450             /* Inner loop uses 283 flops */
 451         }
 452
 453         if(jidx<j_index_end)
 454         {
 455
 456             /* Get j neighbor index, and coordinate index */
 457             jnrlistA         = jjnr[jidx];
 458             jnrlistB         = jjnr[jidx+1];
 459             jnrlistC         = jjnr[jidx+2];
 460             jnrlistD         = jjnr[jidx+3];
 461             jnrlistE         = jjnr[jidx+4];
 462             jnrlistF         = jjnr[jidx+5];
 463             jnrlistG         = jjnr[jidx+6];
 464             jnrlistH         = jjnr[jidx+7];
 465             /* Sign of each element will be negative for non-real atoms.
 466              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 467              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 468              */
 469             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 470                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 471
 472             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 473             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 474             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 475             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 476             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 477             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 478             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 479             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 480             j_coord_offsetA  = DIM*jnrA;
 481             j_coord_offsetB  = DIM*jnrB;
 482             j_coord_offsetC  = DIM*jnrC;
 483             j_coord_offsetD  = DIM*jnrD;
 484             j_coord_offsetE  = DIM*jnrE;
 485             j_coord_offsetF  = DIM*jnrF;
 486             j_coord_offsetG  = DIM*jnrG;
 487             j_coord_offsetH  = DIM*jnrH;
 488
 489             /* load j atom coordinates */
 490             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 491                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 492                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 493                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 494                                                  &jx0,&jy0,&jz0);
 495
 496             /* Calculate displacement vector */
 497             dx00             = _mm256_sub_ps(ix0,jx0);
 498             dy00             = _mm256_sub_ps(iy0,jy0);
 499             dz00             = _mm256_sub_ps(iz0,jz0);
 500             dx10             = _mm256_sub_ps(ix1,jx0);
 501             dy10             = _mm256_sub_ps(iy1,jy0);
 502             dz10             = _mm256_sub_ps(iz1,jz0);
 503             dx20             = _mm256_sub_ps(ix2,jx0);
 504             dy20             = _mm256_sub_ps(iy2,jy0);
 505             dz20             = _mm256_sub_ps(iz2,jz0);
 506
 507             /* Calculate squared distance and things based on it */
 508             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 509             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 510             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 511
 512             rinv00           = avx256_invsqrt_f(rsq00);
 513             rinv10           = avx256_invsqrt_f(rsq10);
 514             rinv20           = avx256_invsqrt_f(rsq20);
 515
 516             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 517             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 518             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 519
 520             /* Load parameters for j particles */
 521             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 522                                                                  charge+jnrC+0,charge+jnrD+0,
 523                                                                  charge+jnrE+0,charge+jnrF+0,
 524                                                                  charge+jnrG+0,charge+jnrH+0);
 525             vdwjidx0A        = 2*vdwtype[jnrA+0];
 526             vdwjidx0B        = 2*vdwtype[jnrB+0];
 527             vdwjidx0C        = 2*vdwtype[jnrC+0];
 528             vdwjidx0D        = 2*vdwtype[jnrD+0];
 529             vdwjidx0E        = 2*vdwtype[jnrE+0];
 530             vdwjidx0F        = 2*vdwtype[jnrF+0];
 531             vdwjidx0G        = 2*vdwtype[jnrG+0];
 532             vdwjidx0H        = 2*vdwtype[jnrH+0];
 533
 534             fjx0             = _mm256_setzero_ps();
 535             fjy0             = _mm256_setzero_ps();
 536             fjz0             = _mm256_setzero_ps();
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             r00              = _mm256_mul_ps(rsq00,rinv00);
 543             r00              = _mm256_andnot_ps(dummy_mask,r00);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq00             = _mm256_mul_ps(iq0,jq0);
 547             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 548                                             vdwioffsetptr0+vdwjidx0B,
 549                                             vdwioffsetptr0+vdwjidx0C,
 550                                             vdwioffsetptr0+vdwjidx0D,
 551                                             vdwioffsetptr0+vdwjidx0E,
 552                                             vdwioffsetptr0+vdwjidx0F,
 553                                             vdwioffsetptr0+vdwjidx0G,
 554                                             vdwioffsetptr0+vdwjidx0H,
 555                                             &c6_00,&c12_00);
 556
 557             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 558                                                                   vdwgridioffsetptr0+vdwjidx0B,
 559                                                                   vdwgridioffsetptr0+vdwjidx0C,
 560                                                                   vdwgridioffsetptr0+vdwjidx0D,
 561                                                                   vdwgridioffsetptr0+vdwjidx0E,
 562                                                                   vdwgridioffsetptr0+vdwjidx0F,
 563                                                                   vdwgridioffsetptr0+vdwjidx0G,
 564                                                                   vdwgridioffsetptr0+vdwjidx0H);
 565
 566             /* EWALD ELECTROSTATICS */
 567
 568             /* Analytical PME correction */
 569             zeta2            = _mm256_mul_ps(beta2,rsq00);
 570             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 571             pmecorrF         = avx256_pmecorrF_f(zeta2);
 572             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 573             felec            = _mm256_mul_ps(qq00,felec);
 574             pmecorrV         = avx256_pmecorrV_f(zeta2);
 575             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 576             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 577             velec            = _mm256_mul_ps(qq00,velec);
 578
 579             /* Analytical LJ-PME */
 580             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 581             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 582             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 583             exponent         = avx256_exp_f(ewcljrsq);
 584             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 585             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 586             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 587             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 588             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 589             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 590             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 591             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);
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm256_andnot_ps(dummy_mask,velec);
 595             velecsum         = _mm256_add_ps(velecsum,velec);
 596             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 597             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 598
 599             fscal            = _mm256_add_ps(felec,fvdw);
 600
 601             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 602
 603             /* Calculate temporary vectorial force */
 604             tx               = _mm256_mul_ps(fscal,dx00);
 605             ty               = _mm256_mul_ps(fscal,dy00);
 606             tz               = _mm256_mul_ps(fscal,dz00);
 607
 608             /* Update vectorial force */
 609             fix0             = _mm256_add_ps(fix0,tx);
 610             fiy0             = _mm256_add_ps(fiy0,ty);
 611             fiz0             = _mm256_add_ps(fiz0,tz);
 612
 613             fjx0             = _mm256_add_ps(fjx0,tx);
 614             fjy0             = _mm256_add_ps(fjy0,ty);
 615             fjz0             = _mm256_add_ps(fjz0,tz);
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             r10              = _mm256_mul_ps(rsq10,rinv10);
 622             r10              = _mm256_andnot_ps(dummy_mask,r10);
 623
 624             /* Compute parameters for interactions between i and j atoms */
 625             qq10             = _mm256_mul_ps(iq1,jq0);
 626
 627             /* EWALD ELECTROSTATICS */
 628
 629             /* Analytical PME correction */
 630             zeta2            = _mm256_mul_ps(beta2,rsq10);
 631             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 632             pmecorrF         = avx256_pmecorrF_f(zeta2);
 633             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 634             felec            = _mm256_mul_ps(qq10,felec);
 635             pmecorrV         = avx256_pmecorrV_f(zeta2);
 636             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 637             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 638             velec            = _mm256_mul_ps(qq10,velec);
 639
 640             /* Update potential sum for this i atom from the interaction with this j atom. */
 641             velec            = _mm256_andnot_ps(dummy_mask,velec);
 642             velecsum         = _mm256_add_ps(velecsum,velec);
 643
 644             fscal            = felec;
 645
 646             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 647
 648             /* Calculate temporary vectorial force */
 649             tx               = _mm256_mul_ps(fscal,dx10);
 650             ty               = _mm256_mul_ps(fscal,dy10);
 651             tz               = _mm256_mul_ps(fscal,dz10);
 652
 653             /* Update vectorial force */
 654             fix1             = _mm256_add_ps(fix1,tx);
 655             fiy1             = _mm256_add_ps(fiy1,ty);
 656             fiz1             = _mm256_add_ps(fiz1,tz);
 657
 658             fjx0             = _mm256_add_ps(fjx0,tx);
 659             fjy0             = _mm256_add_ps(fjy0,ty);
 660             fjz0             = _mm256_add_ps(fjz0,tz);
 661
 662             /**************************
 663              * CALCULATE INTERACTIONS *
 664              **************************/
 665
 666             r20              = _mm256_mul_ps(rsq20,rinv20);
 667             r20              = _mm256_andnot_ps(dummy_mask,r20);
 668
 669             /* Compute parameters for interactions between i and j atoms */
 670             qq20             = _mm256_mul_ps(iq2,jq0);
 671
 672             /* EWALD ELECTROSTATICS */
 673
 674             /* Analytical PME correction */
 675             zeta2            = _mm256_mul_ps(beta2,rsq20);
 676             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 677             pmecorrF         = avx256_pmecorrF_f(zeta2);
 678             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 679             felec            = _mm256_mul_ps(qq20,felec);
 680             pmecorrV         = avx256_pmecorrV_f(zeta2);
 681             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 682             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 683             velec            = _mm256_mul_ps(qq20,velec);
 684
 685             /* Update potential sum for this i atom from the interaction with this j atom. */
 686             velec            = _mm256_andnot_ps(dummy_mask,velec);
 687             velecsum         = _mm256_add_ps(velecsum,velec);
 688
 689             fscal            = felec;
 690
 691             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm256_mul_ps(fscal,dx20);
 695             ty               = _mm256_mul_ps(fscal,dy20);
 696             tz               = _mm256_mul_ps(fscal,dz20);
 697
 698             /* Update vectorial force */
 699             fix2             = _mm256_add_ps(fix2,tx);
 700             fiy2             = _mm256_add_ps(fiy2,ty);
 701             fiz2             = _mm256_add_ps(fiz2,tz);
 702
 703             fjx0             = _mm256_add_ps(fjx0,tx);
 704             fjy0             = _mm256_add_ps(fjy0,ty);
 705             fjz0             = _mm256_add_ps(fjz0,tz);
 706
 707             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 708             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 709             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 710             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 711             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 712             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 713             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 714             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 715
 716             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 717
 718             /* Inner loop uses 286 flops */
 719         }
 720
 721         /* End of innermost loop */
 722
 723         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 724                                                  f+i_coord_offset,fshift+i_shift_offset);
 725
 726         ggid                        = gid[iidx];
 727         /* Update potential energies */
 728         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 729         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 730
 731         /* Increment number of inner iterations */
 732         inneriter                  += j_index_end - j_index_start;
 733
 734         /* Outer loop uses 20 flops */
 735     }
 736
 737     /* Increment number of outer iterations */
 738     outeriter        += nri;
 739
 740     /* Update outer/inner flops */
 741
 742     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*286);
 743 }
 744 /*
 745  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_256_single
 746  * Electrostatics interaction: Ewald
 747  * VdW interaction:            LJEwald
 748  * Geometry:                   Water3-Particle
 749  * Calculate force/pot:        Force
 750  */
 751 void
 752 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_256_single
 753                     (t_nblist                    * gmx_restrict       nlist,
 754                      rvec                        * gmx_restrict          xx,
 755                      rvec                        * gmx_restrict          ff,
 756                      struct t_forcerec           * gmx_restrict          fr,
 757                      t_mdatoms                   * gmx_restrict     mdatoms,
 758                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 759                      t_nrnb                      * gmx_restrict        nrnb)
 760 {
 761     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 762      * just 0 for non-waters.
 763      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 764      * jnr indices corresponding to data put in the four positions in the SIMD register.
 765      */
 766     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 767     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 768     int              jnrA,jnrB,jnrC,jnrD;
 769     int              jnrE,jnrF,jnrG,jnrH;
 770     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 771     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 772     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 773     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 774     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 775     real             rcutoff_scalar;
 776     real             *shiftvec,*fshift,*x,*f;
 777     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 778     real             scratch[4*DIM];
 779     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 780     real *           vdwioffsetptr0;
 781     real *           vdwgridioffsetptr0;
 782     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 783     real *           vdwioffsetptr1;
 784     real *           vdwgridioffsetptr1;
 785     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 786     real *           vdwioffsetptr2;
 787     real *           vdwgridioffsetptr2;
 788     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 789     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 790     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 791     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 792     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 793     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 794     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 795     real             *charge;
 796     int              nvdwtype;
 797     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 798     int              *vdwtype;
 799     real             *vdwparam;
 800     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 801     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 802     __m256           c6grid_00;
 803     __m256           c6grid_10;
 804     __m256           c6grid_20;
 805     real             *vdwgridparam;
 806     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 807     __m256           one_half  = _mm256_set1_ps(0.5);
 808     __m256           minus_one = _mm256_set1_ps(-1.0);
 809     __m256i          ewitab;
 810     __m128i          ewitab_lo,ewitab_hi;
 811     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 812     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 813     real             *ewtab;
 814     __m256           dummy_mask,cutoff_mask;
 815     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 816     __m256           one     = _mm256_set1_ps(1.0);
 817     __m256           two     = _mm256_set1_ps(2.0);
 818     x                = xx[0];
 819     f                = ff[0];
 820
 821     nri              = nlist->nri;
 822     iinr             = nlist->iinr;
 823     jindex           = nlist->jindex;
 824     jjnr             = nlist->jjnr;
 825     shiftidx         = nlist->shift;
 826     gid              = nlist->gid;
 827     shiftvec         = fr->shift_vec[0];
 828     fshift           = fr->fshift[0];
 829     facel            = _mm256_set1_ps(fr->ic->epsfac);
 830     charge           = mdatoms->chargeA;
 831     nvdwtype         = fr->ntype;
 832     vdwparam         = fr->nbfp;
 833     vdwtype          = mdatoms->typeA;
 834     vdwgridparam     = fr->ljpme_c6grid;
 835     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 836     ewclj            = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
 837     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 838
 839     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 840     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 841     beta2            = _mm256_mul_ps(beta,beta);
 842     beta3            = _mm256_mul_ps(beta,beta2);
 843
 844     ewtab            = fr->ic->tabq_coul_F;
 845     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 846     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 847
 848     /* Setup water-specific parameters */
 849     inr              = nlist->iinr[0];
 850     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 851     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 852     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 853     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 854     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 855
 856     /* Avoid stupid compiler warnings */
 857     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 858     j_coord_offsetA = 0;
 859     j_coord_offsetB = 0;
 860     j_coord_offsetC = 0;
 861     j_coord_offsetD = 0;
 862     j_coord_offsetE = 0;
 863     j_coord_offsetF = 0;
 864     j_coord_offsetG = 0;
 865     j_coord_offsetH = 0;
 866
 867     outeriter        = 0;
 868     inneriter        = 0;
 869
 870     for(iidx=0;iidx<4*DIM;iidx++)
 871     {
 872         scratch[iidx] = 0.0;
 873     }
 874
 875     /* Start outer loop over neighborlists */
 876     for(iidx=0; iidx<nri; iidx++)
 877     {
 878         /* Load shift vector for this list */
 879         i_shift_offset   = DIM*shiftidx[iidx];
 880
 881         /* Load limits for loop over neighbors */
 882         j_index_start    = jindex[iidx];
 883         j_index_end      = jindex[iidx+1];
 884
 885         /* Get outer coordinate index */
 886         inr              = iinr[iidx];
 887         i_coord_offset   = DIM*inr;
 888
 889         /* Load i particle coords and add shift vector */
 890         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 891                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 892
 893         fix0             = _mm256_setzero_ps();
 894         fiy0             = _mm256_setzero_ps();
 895         fiz0             = _mm256_setzero_ps();
 896         fix1             = _mm256_setzero_ps();
 897         fiy1             = _mm256_setzero_ps();
 898         fiz1             = _mm256_setzero_ps();
 899         fix2             = _mm256_setzero_ps();
 900         fiy2             = _mm256_setzero_ps();
 901         fiz2             = _mm256_setzero_ps();
 902
 903         /* Start inner kernel loop */
 904         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 905         {
 906
 907             /* Get j neighbor index, and coordinate index */
 908             jnrA             = jjnr[jidx];
 909             jnrB             = jjnr[jidx+1];
 910             jnrC             = jjnr[jidx+2];
 911             jnrD             = jjnr[jidx+3];
 912             jnrE             = jjnr[jidx+4];
 913             jnrF             = jjnr[jidx+5];
 914             jnrG             = jjnr[jidx+6];
 915             jnrH             = jjnr[jidx+7];
 916             j_coord_offsetA  = DIM*jnrA;
 917             j_coord_offsetB  = DIM*jnrB;
 918             j_coord_offsetC  = DIM*jnrC;
 919             j_coord_offsetD  = DIM*jnrD;
 920             j_coord_offsetE  = DIM*jnrE;
 921             j_coord_offsetF  = DIM*jnrF;
 922             j_coord_offsetG  = DIM*jnrG;
 923             j_coord_offsetH  = DIM*jnrH;
 924
 925             /* load j atom coordinates */
 926             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 927                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 928                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 929                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 930                                                  &jx0,&jy0,&jz0);
 931
 932             /* Calculate displacement vector */
 933             dx00             = _mm256_sub_ps(ix0,jx0);
 934             dy00             = _mm256_sub_ps(iy0,jy0);
 935             dz00             = _mm256_sub_ps(iz0,jz0);
 936             dx10             = _mm256_sub_ps(ix1,jx0);
 937             dy10             = _mm256_sub_ps(iy1,jy0);
 938             dz10             = _mm256_sub_ps(iz1,jz0);
 939             dx20             = _mm256_sub_ps(ix2,jx0);
 940             dy20             = _mm256_sub_ps(iy2,jy0);
 941             dz20             = _mm256_sub_ps(iz2,jz0);
 942
 943             /* Calculate squared distance and things based on it */
 944             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 945             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 946             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 947
 948             rinv00           = avx256_invsqrt_f(rsq00);
 949             rinv10           = avx256_invsqrt_f(rsq10);
 950             rinv20           = avx256_invsqrt_f(rsq20);
 951
 952             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 953             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 954             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 955
 956             /* Load parameters for j particles */
 957             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 958                                                                  charge+jnrC+0,charge+jnrD+0,
 959                                                                  charge+jnrE+0,charge+jnrF+0,
 960                                                                  charge+jnrG+0,charge+jnrH+0);
 961             vdwjidx0A        = 2*vdwtype[jnrA+0];
 962             vdwjidx0B        = 2*vdwtype[jnrB+0];
 963             vdwjidx0C        = 2*vdwtype[jnrC+0];
 964             vdwjidx0D        = 2*vdwtype[jnrD+0];
 965             vdwjidx0E        = 2*vdwtype[jnrE+0];
 966             vdwjidx0F        = 2*vdwtype[jnrF+0];
 967             vdwjidx0G        = 2*vdwtype[jnrG+0];
 968             vdwjidx0H        = 2*vdwtype[jnrH+0];
 969
 970             fjx0             = _mm256_setzero_ps();
 971             fjy0             = _mm256_setzero_ps();
 972             fjz0             = _mm256_setzero_ps();
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             r00              = _mm256_mul_ps(rsq00,rinv00);
 979
 980             /* Compute parameters for interactions between i and j atoms */
 981             qq00             = _mm256_mul_ps(iq0,jq0);
 982             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 983                                             vdwioffsetptr0+vdwjidx0B,
 984                                             vdwioffsetptr0+vdwjidx0C,
 985                                             vdwioffsetptr0+vdwjidx0D,
 986                                             vdwioffsetptr0+vdwjidx0E,
 987                                             vdwioffsetptr0+vdwjidx0F,
 988                                             vdwioffsetptr0+vdwjidx0G,
 989                                             vdwioffsetptr0+vdwjidx0H,
 990                                             &c6_00,&c12_00);
 991
 992             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 993                                                                   vdwgridioffsetptr0+vdwjidx0B,
 994                                                                   vdwgridioffsetptr0+vdwjidx0C,
 995                                                                   vdwgridioffsetptr0+vdwjidx0D,
 996                                                                   vdwgridioffsetptr0+vdwjidx0E,
 997                                                                   vdwgridioffsetptr0+vdwjidx0F,
 998                                                                   vdwgridioffsetptr0+vdwjidx0G,
 999                                                                   vdwgridioffsetptr0+vdwjidx0H);
1000
1001             /* EWALD ELECTROSTATICS */
1002
1003             /* Analytical PME correction */
1004             zeta2            = _mm256_mul_ps(beta2,rsq00);
1005             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1006             pmecorrF         = avx256_pmecorrF_f(zeta2);
1007             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1008             felec            = _mm256_mul_ps(qq00,felec);
1009
1010             /* Analytical LJ-PME */
1011             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1012             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1013             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1014             exponent         = avx256_exp_f(ewcljrsq);
1015             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1016             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1017             /* f6A = 6 * C6grid * (1 - poly) */
1018             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1019             /* f6B = C6grid * exponent * beta^6 */
1020             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1021             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1022             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);
1023
1024             fscal            = _mm256_add_ps(felec,fvdw);
1025
1026             /* Calculate temporary vectorial force */
1027             tx               = _mm256_mul_ps(fscal,dx00);
1028             ty               = _mm256_mul_ps(fscal,dy00);
1029             tz               = _mm256_mul_ps(fscal,dz00);
1030
1031             /* Update vectorial force */
1032             fix0             = _mm256_add_ps(fix0,tx);
1033             fiy0             = _mm256_add_ps(fiy0,ty);
1034             fiz0             = _mm256_add_ps(fiz0,tz);
1035
1036             fjx0             = _mm256_add_ps(fjx0,tx);
1037             fjy0             = _mm256_add_ps(fjy0,ty);
1038             fjz0             = _mm256_add_ps(fjz0,tz);
1039
1040             /**************************
1041              * CALCULATE INTERACTIONS *
1042              **************************/
1043
1044             r10              = _mm256_mul_ps(rsq10,rinv10);
1045
1046             /* Compute parameters for interactions between i and j atoms */
1047             qq10             = _mm256_mul_ps(iq1,jq0);
1048
1049             /* EWALD ELECTROSTATICS */
1050
1051             /* Analytical PME correction */
1052             zeta2            = _mm256_mul_ps(beta2,rsq10);
1053             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1054             pmecorrF         = avx256_pmecorrF_f(zeta2);
1055             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1056             felec            = _mm256_mul_ps(qq10,felec);
1057
1058             fscal            = felec;
1059
1060             /* Calculate temporary vectorial force */
1061             tx               = _mm256_mul_ps(fscal,dx10);
1062             ty               = _mm256_mul_ps(fscal,dy10);
1063             tz               = _mm256_mul_ps(fscal,dz10);
1064
1065             /* Update vectorial force */
1066             fix1             = _mm256_add_ps(fix1,tx);
1067             fiy1             = _mm256_add_ps(fiy1,ty);
1068             fiz1             = _mm256_add_ps(fiz1,tz);
1069
1070             fjx0             = _mm256_add_ps(fjx0,tx);
1071             fjy0             = _mm256_add_ps(fjy0,ty);
1072             fjz0             = _mm256_add_ps(fjz0,tz);
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             r20              = _mm256_mul_ps(rsq20,rinv20);
1079
1080             /* Compute parameters for interactions between i and j atoms */
1081             qq20             = _mm256_mul_ps(iq2,jq0);
1082
1083             /* EWALD ELECTROSTATICS */
1084
1085             /* Analytical PME correction */
1086             zeta2            = _mm256_mul_ps(beta2,rsq20);
1087             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1088             pmecorrF         = avx256_pmecorrF_f(zeta2);
1089             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1090             felec            = _mm256_mul_ps(qq20,felec);
1091
1092             fscal            = felec;
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm256_mul_ps(fscal,dx20);
1096             ty               = _mm256_mul_ps(fscal,dy20);
1097             tz               = _mm256_mul_ps(fscal,dz20);
1098
1099             /* Update vectorial force */
1100             fix2             = _mm256_add_ps(fix2,tx);
1101             fiy2             = _mm256_add_ps(fiy2,ty);
1102             fiz2             = _mm256_add_ps(fiz2,tz);
1103
1104             fjx0             = _mm256_add_ps(fjx0,tx);
1105             fjy0             = _mm256_add_ps(fjy0,ty);
1106             fjz0             = _mm256_add_ps(fjz0,tz);
1107
1108             fjptrA             = f+j_coord_offsetA;
1109             fjptrB             = f+j_coord_offsetB;
1110             fjptrC             = f+j_coord_offsetC;
1111             fjptrD             = f+j_coord_offsetD;
1112             fjptrE             = f+j_coord_offsetE;
1113             fjptrF             = f+j_coord_offsetF;
1114             fjptrG             = f+j_coord_offsetG;
1115             fjptrH             = f+j_coord_offsetH;
1116
1117             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1118
1119             /* Inner loop uses 194 flops */
1120         }
1121
1122         if(jidx<j_index_end)
1123         {
1124
1125             /* Get j neighbor index, and coordinate index */
1126             jnrlistA         = jjnr[jidx];
1127             jnrlistB         = jjnr[jidx+1];
1128             jnrlistC         = jjnr[jidx+2];
1129             jnrlistD         = jjnr[jidx+3];
1130             jnrlistE         = jjnr[jidx+4];
1131             jnrlistF         = jjnr[jidx+5];
1132             jnrlistG         = jjnr[jidx+6];
1133             jnrlistH         = jjnr[jidx+7];
1134             /* Sign of each element will be negative for non-real atoms.
1135              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1136              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1137              */
1138             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1139                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1140
1141             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1142             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1143             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1144             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1145             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1146             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1147             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1148             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1149             j_coord_offsetA  = DIM*jnrA;
1150             j_coord_offsetB  = DIM*jnrB;
1151             j_coord_offsetC  = DIM*jnrC;
1152             j_coord_offsetD  = DIM*jnrD;
1153             j_coord_offsetE  = DIM*jnrE;
1154             j_coord_offsetF  = DIM*jnrF;
1155             j_coord_offsetG  = DIM*jnrG;
1156             j_coord_offsetH  = DIM*jnrH;
1157
1158             /* load j atom coordinates */
1159             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1160                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1161                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1162                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1163                                                  &jx0,&jy0,&jz0);
1164
1165             /* Calculate displacement vector */
1166             dx00             = _mm256_sub_ps(ix0,jx0);
1167             dy00             = _mm256_sub_ps(iy0,jy0);
1168             dz00             = _mm256_sub_ps(iz0,jz0);
1169             dx10             = _mm256_sub_ps(ix1,jx0);
1170             dy10             = _mm256_sub_ps(iy1,jy0);
1171             dz10             = _mm256_sub_ps(iz1,jz0);
1172             dx20             = _mm256_sub_ps(ix2,jx0);
1173             dy20             = _mm256_sub_ps(iy2,jy0);
1174             dz20             = _mm256_sub_ps(iz2,jz0);
1175
1176             /* Calculate squared distance and things based on it */
1177             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1178             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1179             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1180
1181             rinv00           = avx256_invsqrt_f(rsq00);
1182             rinv10           = avx256_invsqrt_f(rsq10);
1183             rinv20           = avx256_invsqrt_f(rsq20);
1184
1185             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1186             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1187             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1188
1189             /* Load parameters for j particles */
1190             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1191                                                                  charge+jnrC+0,charge+jnrD+0,
1192                                                                  charge+jnrE+0,charge+jnrF+0,
1193                                                                  charge+jnrG+0,charge+jnrH+0);
1194             vdwjidx0A        = 2*vdwtype[jnrA+0];
1195             vdwjidx0B        = 2*vdwtype[jnrB+0];
1196             vdwjidx0C        = 2*vdwtype[jnrC+0];
1197             vdwjidx0D        = 2*vdwtype[jnrD+0];
1198             vdwjidx0E        = 2*vdwtype[jnrE+0];
1199             vdwjidx0F        = 2*vdwtype[jnrF+0];
1200             vdwjidx0G        = 2*vdwtype[jnrG+0];
1201             vdwjidx0H        = 2*vdwtype[jnrH+0];
1202
1203             fjx0             = _mm256_setzero_ps();
1204             fjy0             = _mm256_setzero_ps();
1205             fjz0             = _mm256_setzero_ps();
1206
1207             /**************************
1208              * CALCULATE INTERACTIONS *
1209              **************************/
1210
1211             r00              = _mm256_mul_ps(rsq00,rinv00);
1212             r00              = _mm256_andnot_ps(dummy_mask,r00);
1213
1214             /* Compute parameters for interactions between i and j atoms */
1215             qq00             = _mm256_mul_ps(iq0,jq0);
1216             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1217                                             vdwioffsetptr0+vdwjidx0B,
1218                                             vdwioffsetptr0+vdwjidx0C,
1219                                             vdwioffsetptr0+vdwjidx0D,
1220                                             vdwioffsetptr0+vdwjidx0E,
1221                                             vdwioffsetptr0+vdwjidx0F,
1222                                             vdwioffsetptr0+vdwjidx0G,
1223                                             vdwioffsetptr0+vdwjidx0H,
1224                                             &c6_00,&c12_00);
1225
1226             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1227                                                                   vdwgridioffsetptr0+vdwjidx0B,
1228                                                                   vdwgridioffsetptr0+vdwjidx0C,
1229                                                                   vdwgridioffsetptr0+vdwjidx0D,
1230                                                                   vdwgridioffsetptr0+vdwjidx0E,
1231                                                                   vdwgridioffsetptr0+vdwjidx0F,
1232                                                                   vdwgridioffsetptr0+vdwjidx0G,
1233                                                                   vdwgridioffsetptr0+vdwjidx0H);
1234
1235             /* EWALD ELECTROSTATICS */
1236
1237             /* Analytical PME correction */
1238             zeta2            = _mm256_mul_ps(beta2,rsq00);
1239             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1240             pmecorrF         = avx256_pmecorrF_f(zeta2);
1241             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1242             felec            = _mm256_mul_ps(qq00,felec);
1243
1244             /* Analytical LJ-PME */
1245             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1246             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1247             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1248             exponent         = avx256_exp_f(ewcljrsq);
1249             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1250             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1251             /* f6A = 6 * C6grid * (1 - poly) */
1252             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1253             /* f6B = C6grid * exponent * beta^6 */
1254             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1255             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1256             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);
1257
1258             fscal            = _mm256_add_ps(felec,fvdw);
1259
1260             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1261
1262             /* Calculate temporary vectorial force */
1263             tx               = _mm256_mul_ps(fscal,dx00);
1264             ty               = _mm256_mul_ps(fscal,dy00);
1265             tz               = _mm256_mul_ps(fscal,dz00);
1266
1267             /* Update vectorial force */
1268             fix0             = _mm256_add_ps(fix0,tx);
1269             fiy0             = _mm256_add_ps(fiy0,ty);
1270             fiz0             = _mm256_add_ps(fiz0,tz);
1271
1272             fjx0             = _mm256_add_ps(fjx0,tx);
1273             fjy0             = _mm256_add_ps(fjy0,ty);
1274             fjz0             = _mm256_add_ps(fjz0,tz);
1275
1276             /**************************
1277              * CALCULATE INTERACTIONS *
1278              **************************/
1279
1280             r10              = _mm256_mul_ps(rsq10,rinv10);
1281             r10              = _mm256_andnot_ps(dummy_mask,r10);
1282
1283             /* Compute parameters for interactions between i and j atoms */
1284             qq10             = _mm256_mul_ps(iq1,jq0);
1285
1286             /* EWALD ELECTROSTATICS */
1287
1288             /* Analytical PME correction */
1289             zeta2            = _mm256_mul_ps(beta2,rsq10);
1290             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1291             pmecorrF         = avx256_pmecorrF_f(zeta2);
1292             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1293             felec            = _mm256_mul_ps(qq10,felec);
1294
1295             fscal            = felec;
1296
1297             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1298
1299             /* Calculate temporary vectorial force */
1300             tx               = _mm256_mul_ps(fscal,dx10);
1301             ty               = _mm256_mul_ps(fscal,dy10);
1302             tz               = _mm256_mul_ps(fscal,dz10);
1303
1304             /* Update vectorial force */
1305             fix1             = _mm256_add_ps(fix1,tx);
1306             fiy1             = _mm256_add_ps(fiy1,ty);
1307             fiz1             = _mm256_add_ps(fiz1,tz);
1308
1309             fjx0             = _mm256_add_ps(fjx0,tx);
1310             fjy0             = _mm256_add_ps(fjy0,ty);
1311             fjz0             = _mm256_add_ps(fjz0,tz);
1312
1313             /**************************
1314              * CALCULATE INTERACTIONS *
1315              **************************/
1316
1317             r20              = _mm256_mul_ps(rsq20,rinv20);
1318             r20              = _mm256_andnot_ps(dummy_mask,r20);
1319
1320             /* Compute parameters for interactions between i and j atoms */
1321             qq20             = _mm256_mul_ps(iq2,jq0);
1322
1323             /* EWALD ELECTROSTATICS */
1324
1325             /* Analytical PME correction */
1326             zeta2            = _mm256_mul_ps(beta2,rsq20);
1327             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1328             pmecorrF         = avx256_pmecorrF_f(zeta2);
1329             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1330             felec            = _mm256_mul_ps(qq20,felec);
1331
1332             fscal            = felec;
1333
1334             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1335
1336             /* Calculate temporary vectorial force */
1337             tx               = _mm256_mul_ps(fscal,dx20);
1338             ty               = _mm256_mul_ps(fscal,dy20);
1339             tz               = _mm256_mul_ps(fscal,dz20);
1340
1341             /* Update vectorial force */
1342             fix2             = _mm256_add_ps(fix2,tx);
1343             fiy2             = _mm256_add_ps(fiy2,ty);
1344             fiz2             = _mm256_add_ps(fiz2,tz);
1345
1346             fjx0             = _mm256_add_ps(fjx0,tx);
1347             fjy0             = _mm256_add_ps(fjy0,ty);
1348             fjz0             = _mm256_add_ps(fjz0,tz);
1349
1350             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1351             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1352             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1353             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1354             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1355             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1356             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1357             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1358
1359             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1360
1361             /* Inner loop uses 197 flops */
1362         }
1363
1364         /* End of innermost loop */
1365
1366         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1367                                                  f+i_coord_offset,fshift+i_shift_offset);
1368
1369         /* Increment number of inner iterations */
1370         inneriter                  += j_index_end - j_index_start;
1371
1372         /* Outer loop uses 18 flops */
1373     }
1374
1375     /* Increment number of outer iterations */
1376     outeriter        += nri;
1377
1378     /* Update outer/inner flops */
1379
1380     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*197);
1381 }