src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     real *           vdwgridioffsetptr0;
  88     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  89     real *           vdwioffsetptr1;
  90     real *           vdwgridioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     real *           vdwgridioffsetptr2;
  94     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  95     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  96     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  97     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  98     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  99     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 100     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 107     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 108     __m256           c6grid_00;
 109     __m256           c6grid_10;
 110     __m256           c6grid_20;
 111     real             *vdwgridparam;
 112     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 113     __m256           one_half  = _mm256_set1_ps(0.5);
 114     __m256           minus_one = _mm256_set1_ps(-1.0);
 115     __m256i          ewitab;
 116     __m128i          ewitab_lo,ewitab_hi;
 117     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 118     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 119     real             *ewtab;
 120     __m256           dummy_mask,cutoff_mask;
 121     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 122     __m256           one     = _mm256_set1_ps(1.0);
 123     __m256           two     = _mm256_set1_ps(2.0);
 124     x                = xx[0];
 125     f                = ff[0];
 126
 127     nri              = nlist->nri;
 128     iinr             = nlist->iinr;
 129     jindex           = nlist->jindex;
 130     jjnr             = nlist->jjnr;
 131     shiftidx         = nlist->shift;
 132     gid              = nlist->gid;
 133     shiftvec         = fr->shift_vec[0];
 134     fshift           = fr->fshift[0];
 135     facel            = _mm256_set1_ps(fr->epsfac);
 136     charge           = mdatoms->chargeA;
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140     vdwgridparam     = fr->ljpme_c6grid;
 141     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 142     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 143     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 144
 145     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 146     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 147     beta2            = _mm256_mul_ps(beta,beta);
 148     beta3            = _mm256_mul_ps(beta,beta2);
 149
 150     ewtab            = fr->ic->tabq_coul_FDV0;
 151     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 152     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 153
 154     /* Setup water-specific parameters */
 155     inr              = nlist->iinr[0];
 156     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 157     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 158     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 159     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 160     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 161
 162     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 163     rcutoff_scalar   = fr->rcoulomb;
 164     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 165     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 166
 167     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 168     rvdw             = _mm256_set1_ps(fr->rvdw);
 169
 170     /* Avoid stupid compiler warnings */
 171     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 172     j_coord_offsetA = 0;
 173     j_coord_offsetB = 0;
 174     j_coord_offsetC = 0;
 175     j_coord_offsetD = 0;
 176     j_coord_offsetE = 0;
 177     j_coord_offsetF = 0;
 178     j_coord_offsetG = 0;
 179     j_coord_offsetH = 0;
 180
 181     outeriter        = 0;
 182     inneriter        = 0;
 183
 184     for(iidx=0;iidx<4*DIM;iidx++)
 185     {
 186         scratch[iidx] = 0.0;
 187     }
 188
 189     /* Start outer loop over neighborlists */
 190     for(iidx=0; iidx<nri; iidx++)
 191     {
 192         /* Load shift vector for this list */
 193         i_shift_offset   = DIM*shiftidx[iidx];
 194
 195         /* Load limits for loop over neighbors */
 196         j_index_start    = jindex[iidx];
 197         j_index_end      = jindex[iidx+1];
 198
 199         /* Get outer coordinate index */
 200         inr              = iinr[iidx];
 201         i_coord_offset   = DIM*inr;
 202
 203         /* Load i particle coords and add shift vector */
 204         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 205                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 206
 207         fix0             = _mm256_setzero_ps();
 208         fiy0             = _mm256_setzero_ps();
 209         fiz0             = _mm256_setzero_ps();
 210         fix1             = _mm256_setzero_ps();
 211         fiy1             = _mm256_setzero_ps();
 212         fiz1             = _mm256_setzero_ps();
 213         fix2             = _mm256_setzero_ps();
 214         fiy2             = _mm256_setzero_ps();
 215         fiz2             = _mm256_setzero_ps();
 216
 217         /* Reset potential sums */
 218         velecsum         = _mm256_setzero_ps();
 219         vvdwsum          = _mm256_setzero_ps();
 220
 221         /* Start inner kernel loop */
 222         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 223         {
 224
 225             /* Get j neighbor index, and coordinate index */
 226             jnrA             = jjnr[jidx];
 227             jnrB             = jjnr[jidx+1];
 228             jnrC             = jjnr[jidx+2];
 229             jnrD             = jjnr[jidx+3];
 230             jnrE             = jjnr[jidx+4];
 231             jnrF             = jjnr[jidx+5];
 232             jnrG             = jjnr[jidx+6];
 233             jnrH             = jjnr[jidx+7];
 234             j_coord_offsetA  = DIM*jnrA;
 235             j_coord_offsetB  = DIM*jnrB;
 236             j_coord_offsetC  = DIM*jnrC;
 237             j_coord_offsetD  = DIM*jnrD;
 238             j_coord_offsetE  = DIM*jnrE;
 239             j_coord_offsetF  = DIM*jnrF;
 240             j_coord_offsetG  = DIM*jnrG;
 241             j_coord_offsetH  = DIM*jnrH;
 242
 243             /* load j atom coordinates */
 244             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 245                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 246                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 247                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 248                                                  &jx0,&jy0,&jz0);
 249
 250             /* Calculate displacement vector */
 251             dx00             = _mm256_sub_ps(ix0,jx0);
 252             dy00             = _mm256_sub_ps(iy0,jy0);
 253             dz00             = _mm256_sub_ps(iz0,jz0);
 254             dx10             = _mm256_sub_ps(ix1,jx0);
 255             dy10             = _mm256_sub_ps(iy1,jy0);
 256             dz10             = _mm256_sub_ps(iz1,jz0);
 257             dx20             = _mm256_sub_ps(ix2,jx0);
 258             dy20             = _mm256_sub_ps(iy2,jy0);
 259             dz20             = _mm256_sub_ps(iz2,jz0);
 260
 261             /* Calculate squared distance and things based on it */
 262             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 263             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 264             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 265
 266             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 267             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 268             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 269
 270             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 271             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 272             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 273
 274             /* Load parameters for j particles */
 275             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 276                                                                  charge+jnrC+0,charge+jnrD+0,
 277                                                                  charge+jnrE+0,charge+jnrF+0,
 278                                                                  charge+jnrG+0,charge+jnrH+0);
 279             vdwjidx0A        = 2*vdwtype[jnrA+0];
 280             vdwjidx0B        = 2*vdwtype[jnrB+0];
 281             vdwjidx0C        = 2*vdwtype[jnrC+0];
 282             vdwjidx0D        = 2*vdwtype[jnrD+0];
 283             vdwjidx0E        = 2*vdwtype[jnrE+0];
 284             vdwjidx0F        = 2*vdwtype[jnrF+0];
 285             vdwjidx0G        = 2*vdwtype[jnrG+0];
 286             vdwjidx0H        = 2*vdwtype[jnrH+0];
 287
 288             fjx0             = _mm256_setzero_ps();
 289             fjy0             = _mm256_setzero_ps();
 290             fjz0             = _mm256_setzero_ps();
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 297             {
 298
 299             r00              = _mm256_mul_ps(rsq00,rinv00);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq00             = _mm256_mul_ps(iq0,jq0);
 303             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 304                                             vdwioffsetptr0+vdwjidx0B,
 305                                             vdwioffsetptr0+vdwjidx0C,
 306                                             vdwioffsetptr0+vdwjidx0D,
 307                                             vdwioffsetptr0+vdwjidx0E,
 308                                             vdwioffsetptr0+vdwjidx0F,
 309                                             vdwioffsetptr0+vdwjidx0G,
 310                                             vdwioffsetptr0+vdwjidx0H,
 311                                             &c6_00,&c12_00);
 312
 313             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 314                                                                   vdwgridioffsetptr0+vdwjidx0B,
 315                                                                   vdwgridioffsetptr0+vdwjidx0C,
 316                                                                   vdwgridioffsetptr0+vdwjidx0D,
 317                                                                   vdwgridioffsetptr0+vdwjidx0E,
 318                                                                   vdwgridioffsetptr0+vdwjidx0F,
 319                                                                   vdwgridioffsetptr0+vdwjidx0G,
 320                                                                   vdwgridioffsetptr0+vdwjidx0H);
 321
 322             /* EWALD ELECTROSTATICS */
 323
 324             /* Analytical PME correction */
 325             zeta2            = _mm256_mul_ps(beta2,rsq00);
 326             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 327             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 328             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 329             felec            = _mm256_mul_ps(qq00,felec);
 330             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 331             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 332             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 333             velec            = _mm256_mul_ps(qq00,velec);
 334
 335             /* Analytical LJ-PME */
 336             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 337             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 338             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 339             exponent         = gmx_simd_exp_r(ewcljrsq);
 340             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 341             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 342             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 343             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 344             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 345             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) ,
 346                                           _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));
 347             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 348             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);
 349
 350             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velec            = _mm256_and_ps(velec,cutoff_mask);
 354             velecsum         = _mm256_add_ps(velecsum,velec);
 355             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 356             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 357
 358             fscal            = _mm256_add_ps(felec,fvdw);
 359
 360             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm256_mul_ps(fscal,dx00);
 364             ty               = _mm256_mul_ps(fscal,dy00);
 365             tz               = _mm256_mul_ps(fscal,dz00);
 366
 367             /* Update vectorial force */
 368             fix0             = _mm256_add_ps(fix0,tx);
 369             fiy0             = _mm256_add_ps(fiy0,ty);
 370             fiz0             = _mm256_add_ps(fiz0,tz);
 371
 372             fjx0             = _mm256_add_ps(fjx0,tx);
 373             fjy0             = _mm256_add_ps(fjy0,ty);
 374             fjz0             = _mm256_add_ps(fjz0,tz);
 375
 376             }
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 383             {
 384
 385             r10              = _mm256_mul_ps(rsq10,rinv10);
 386
 387             /* Compute parameters for interactions between i and j atoms */
 388             qq10             = _mm256_mul_ps(iq1,jq0);
 389
 390             /* EWALD ELECTROSTATICS */
 391
 392             /* Analytical PME correction */
 393             zeta2            = _mm256_mul_ps(beta2,rsq10);
 394             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 395             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 396             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 397             felec            = _mm256_mul_ps(qq10,felec);
 398             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 399             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 400             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 401             velec            = _mm256_mul_ps(qq10,velec);
 402
 403             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velec            = _mm256_and_ps(velec,cutoff_mask);
 407             velecsum         = _mm256_add_ps(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 412
 413             /* Calculate temporary vectorial force */
 414             tx               = _mm256_mul_ps(fscal,dx10);
 415             ty               = _mm256_mul_ps(fscal,dy10);
 416             tz               = _mm256_mul_ps(fscal,dz10);
 417
 418             /* Update vectorial force */
 419             fix1             = _mm256_add_ps(fix1,tx);
 420             fiy1             = _mm256_add_ps(fiy1,ty);
 421             fiz1             = _mm256_add_ps(fiz1,tz);
 422
 423             fjx0             = _mm256_add_ps(fjx0,tx);
 424             fjy0             = _mm256_add_ps(fjy0,ty);
 425             fjz0             = _mm256_add_ps(fjz0,tz);
 426
 427             }
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 434             {
 435
 436             r20              = _mm256_mul_ps(rsq20,rinv20);
 437
 438             /* Compute parameters for interactions between i and j atoms */
 439             qq20             = _mm256_mul_ps(iq2,jq0);
 440
 441             /* EWALD ELECTROSTATICS */
 442
 443             /* Analytical PME correction */
 444             zeta2            = _mm256_mul_ps(beta2,rsq20);
 445             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 446             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 447             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 448             felec            = _mm256_mul_ps(qq20,felec);
 449             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 450             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 451             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 452             velec            = _mm256_mul_ps(qq20,velec);
 453
 454             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm256_and_ps(velec,cutoff_mask);
 458             velecsum         = _mm256_add_ps(velecsum,velec);
 459
 460             fscal            = felec;
 461
 462             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm256_mul_ps(fscal,dx20);
 466             ty               = _mm256_mul_ps(fscal,dy20);
 467             tz               = _mm256_mul_ps(fscal,dz20);
 468
 469             /* Update vectorial force */
 470             fix2             = _mm256_add_ps(fix2,tx);
 471             fiy2             = _mm256_add_ps(fiy2,ty);
 472             fiz2             = _mm256_add_ps(fiz2,tz);
 473
 474             fjx0             = _mm256_add_ps(fjx0,tx);
 475             fjy0             = _mm256_add_ps(fjy0,ty);
 476             fjz0             = _mm256_add_ps(fjz0,tz);
 477
 478             }
 479
 480             fjptrA             = f+j_coord_offsetA;
 481             fjptrB             = f+j_coord_offsetB;
 482             fjptrC             = f+j_coord_offsetC;
 483             fjptrD             = f+j_coord_offsetD;
 484             fjptrE             = f+j_coord_offsetE;
 485             fjptrF             = f+j_coord_offsetF;
 486             fjptrG             = f+j_coord_offsetG;
 487             fjptrH             = f+j_coord_offsetH;
 488
 489             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 490
 491             /* Inner loop uses 366 flops */
 492         }
 493
 494         if(jidx<j_index_end)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrlistA         = jjnr[jidx];
 499             jnrlistB         = jjnr[jidx+1];
 500             jnrlistC         = jjnr[jidx+2];
 501             jnrlistD         = jjnr[jidx+3];
 502             jnrlistE         = jjnr[jidx+4];
 503             jnrlistF         = jjnr[jidx+5];
 504             jnrlistG         = jjnr[jidx+6];
 505             jnrlistH         = jjnr[jidx+7];
 506             /* Sign of each element will be negative for non-real atoms.
 507              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 508              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 509              */
 510             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 511                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 512
 513             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 514             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 515             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 516             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 517             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 518             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 519             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 520             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 521             j_coord_offsetA  = DIM*jnrA;
 522             j_coord_offsetB  = DIM*jnrB;
 523             j_coord_offsetC  = DIM*jnrC;
 524             j_coord_offsetD  = DIM*jnrD;
 525             j_coord_offsetE  = DIM*jnrE;
 526             j_coord_offsetF  = DIM*jnrF;
 527             j_coord_offsetG  = DIM*jnrG;
 528             j_coord_offsetH  = DIM*jnrH;
 529
 530             /* load j atom coordinates */
 531             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 532                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 533                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 534                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 535                                                  &jx0,&jy0,&jz0);
 536
 537             /* Calculate displacement vector */
 538             dx00             = _mm256_sub_ps(ix0,jx0);
 539             dy00             = _mm256_sub_ps(iy0,jy0);
 540             dz00             = _mm256_sub_ps(iz0,jz0);
 541             dx10             = _mm256_sub_ps(ix1,jx0);
 542             dy10             = _mm256_sub_ps(iy1,jy0);
 543             dz10             = _mm256_sub_ps(iz1,jz0);
 544             dx20             = _mm256_sub_ps(ix2,jx0);
 545             dy20             = _mm256_sub_ps(iy2,jy0);
 546             dz20             = _mm256_sub_ps(iz2,jz0);
 547
 548             /* Calculate squared distance and things based on it */
 549             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 550             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 551             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 552
 553             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 554             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 555             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 556
 557             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 558             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 559             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 560
 561             /* Load parameters for j particles */
 562             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 563                                                                  charge+jnrC+0,charge+jnrD+0,
 564                                                                  charge+jnrE+0,charge+jnrF+0,
 565                                                                  charge+jnrG+0,charge+jnrH+0);
 566             vdwjidx0A        = 2*vdwtype[jnrA+0];
 567             vdwjidx0B        = 2*vdwtype[jnrB+0];
 568             vdwjidx0C        = 2*vdwtype[jnrC+0];
 569             vdwjidx0D        = 2*vdwtype[jnrD+0];
 570             vdwjidx0E        = 2*vdwtype[jnrE+0];
 571             vdwjidx0F        = 2*vdwtype[jnrF+0];
 572             vdwjidx0G        = 2*vdwtype[jnrG+0];
 573             vdwjidx0H        = 2*vdwtype[jnrH+0];
 574
 575             fjx0             = _mm256_setzero_ps();
 576             fjy0             = _mm256_setzero_ps();
 577             fjz0             = _mm256_setzero_ps();
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 584             {
 585
 586             r00              = _mm256_mul_ps(rsq00,rinv00);
 587             r00              = _mm256_andnot_ps(dummy_mask,r00);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq00             = _mm256_mul_ps(iq0,jq0);
 591             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 592                                             vdwioffsetptr0+vdwjidx0B,
 593                                             vdwioffsetptr0+vdwjidx0C,
 594                                             vdwioffsetptr0+vdwjidx0D,
 595                                             vdwioffsetptr0+vdwjidx0E,
 596                                             vdwioffsetptr0+vdwjidx0F,
 597                                             vdwioffsetptr0+vdwjidx0G,
 598                                             vdwioffsetptr0+vdwjidx0H,
 599                                             &c6_00,&c12_00);
 600
 601             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 602                                                                   vdwgridioffsetptr0+vdwjidx0B,
 603                                                                   vdwgridioffsetptr0+vdwjidx0C,
 604                                                                   vdwgridioffsetptr0+vdwjidx0D,
 605                                                                   vdwgridioffsetptr0+vdwjidx0E,
 606                                                                   vdwgridioffsetptr0+vdwjidx0F,
 607                                                                   vdwgridioffsetptr0+vdwjidx0G,
 608                                                                   vdwgridioffsetptr0+vdwjidx0H);
 609
 610             /* EWALD ELECTROSTATICS */
 611
 612             /* Analytical PME correction */
 613             zeta2            = _mm256_mul_ps(beta2,rsq00);
 614             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 615             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 616             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 617             felec            = _mm256_mul_ps(qq00,felec);
 618             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 619             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 620             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 621             velec            = _mm256_mul_ps(qq00,velec);
 622
 623             /* Analytical LJ-PME */
 624             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 625             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 626             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 627             exponent         = gmx_simd_exp_r(ewcljrsq);
 628             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 629             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 630             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 631             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 632             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 633             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) ,
 634                                           _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));
 635             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 636             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);
 637
 638             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 639
 640             /* Update potential sum for this i atom from the interaction with this j atom. */
 641             velec            = _mm256_and_ps(velec,cutoff_mask);
 642             velec            = _mm256_andnot_ps(dummy_mask,velec);
 643             velecsum         = _mm256_add_ps(velecsum,velec);
 644             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 645             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 646             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 647
 648             fscal            = _mm256_add_ps(felec,fvdw);
 649
 650             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 651
 652             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 653
 654             /* Calculate temporary vectorial force */
 655             tx               = _mm256_mul_ps(fscal,dx00);
 656             ty               = _mm256_mul_ps(fscal,dy00);
 657             tz               = _mm256_mul_ps(fscal,dz00);
 658
 659             /* Update vectorial force */
 660             fix0             = _mm256_add_ps(fix0,tx);
 661             fiy0             = _mm256_add_ps(fiy0,ty);
 662             fiz0             = _mm256_add_ps(fiz0,tz);
 663
 664             fjx0             = _mm256_add_ps(fjx0,tx);
 665             fjy0             = _mm256_add_ps(fjy0,ty);
 666             fjz0             = _mm256_add_ps(fjz0,tz);
 667
 668             }
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 675             {
 676
 677             r10              = _mm256_mul_ps(rsq10,rinv10);
 678             r10              = _mm256_andnot_ps(dummy_mask,r10);
 679
 680             /* Compute parameters for interactions between i and j atoms */
 681             qq10             = _mm256_mul_ps(iq1,jq0);
 682
 683             /* EWALD ELECTROSTATICS */
 684
 685             /* Analytical PME correction */
 686             zeta2            = _mm256_mul_ps(beta2,rsq10);
 687             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 688             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 689             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 690             felec            = _mm256_mul_ps(qq10,felec);
 691             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 692             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 693             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 694             velec            = _mm256_mul_ps(qq10,velec);
 695
 696             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 697
 698             /* Update potential sum for this i atom from the interaction with this j atom. */
 699             velec            = _mm256_and_ps(velec,cutoff_mask);
 700             velec            = _mm256_andnot_ps(dummy_mask,velec);
 701             velecsum         = _mm256_add_ps(velecsum,velec);
 702
 703             fscal            = felec;
 704
 705             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 706
 707             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 708
 709             /* Calculate temporary vectorial force */
 710             tx               = _mm256_mul_ps(fscal,dx10);
 711             ty               = _mm256_mul_ps(fscal,dy10);
 712             tz               = _mm256_mul_ps(fscal,dz10);
 713
 714             /* Update vectorial force */
 715             fix1             = _mm256_add_ps(fix1,tx);
 716             fiy1             = _mm256_add_ps(fiy1,ty);
 717             fiz1             = _mm256_add_ps(fiz1,tz);
 718
 719             fjx0             = _mm256_add_ps(fjx0,tx);
 720             fjy0             = _mm256_add_ps(fjy0,ty);
 721             fjz0             = _mm256_add_ps(fjz0,tz);
 722
 723             }
 724
 725             /**************************
 726              * CALCULATE INTERACTIONS *
 727              **************************/
 728
 729             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 730             {
 731
 732             r20              = _mm256_mul_ps(rsq20,rinv20);
 733             r20              = _mm256_andnot_ps(dummy_mask,r20);
 734
 735             /* Compute parameters for interactions between i and j atoms */
 736             qq20             = _mm256_mul_ps(iq2,jq0);
 737
 738             /* EWALD ELECTROSTATICS */
 739
 740             /* Analytical PME correction */
 741             zeta2            = _mm256_mul_ps(beta2,rsq20);
 742             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 743             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 744             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 745             felec            = _mm256_mul_ps(qq20,felec);
 746             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 747             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 748             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 749             velec            = _mm256_mul_ps(qq20,velec);
 750
 751             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 752
 753             /* Update potential sum for this i atom from the interaction with this j atom. */
 754             velec            = _mm256_and_ps(velec,cutoff_mask);
 755             velec            = _mm256_andnot_ps(dummy_mask,velec);
 756             velecsum         = _mm256_add_ps(velecsum,velec);
 757
 758             fscal            = felec;
 759
 760             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 761
 762             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 763
 764             /* Calculate temporary vectorial force */
 765             tx               = _mm256_mul_ps(fscal,dx20);
 766             ty               = _mm256_mul_ps(fscal,dy20);
 767             tz               = _mm256_mul_ps(fscal,dz20);
 768
 769             /* Update vectorial force */
 770             fix2             = _mm256_add_ps(fix2,tx);
 771             fiy2             = _mm256_add_ps(fiy2,ty);
 772             fiz2             = _mm256_add_ps(fiz2,tz);
 773
 774             fjx0             = _mm256_add_ps(fjx0,tx);
 775             fjy0             = _mm256_add_ps(fjy0,ty);
 776             fjz0             = _mm256_add_ps(fjz0,tz);
 777
 778             }
 779
 780             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 781             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 782             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 783             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 784             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 785             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 786             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 787             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 788
 789             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 790
 791             /* Inner loop uses 369 flops */
 792         }
 793
 794         /* End of innermost loop */
 795
 796         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 797                                                  f+i_coord_offset,fshift+i_shift_offset);
 798
 799         ggid                        = gid[iidx];
 800         /* Update potential energies */
 801         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 802         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 803
 804         /* Increment number of inner iterations */
 805         inneriter                  += j_index_end - j_index_start;
 806
 807         /* Outer loop uses 20 flops */
 808     }
 809
 810     /* Increment number of outer iterations */
 811     outeriter        += nri;
 812
 813     /* Update outer/inner flops */
 814
 815     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*369);
 816 }
 817 /*
 818  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_single
 819  * Electrostatics interaction: Ewald
 820  * VdW interaction:            LJEwald
 821  * Geometry:                   Water3-Particle
 822  * Calculate force/pot:        Force
 823  */
 824 void
 825 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_single
 826                     (t_nblist                    * gmx_restrict       nlist,
 827                      rvec                        * gmx_restrict          xx,
 828                      rvec                        * gmx_restrict          ff,
 829                      t_forcerec                  * gmx_restrict          fr,
 830                      t_mdatoms                   * gmx_restrict     mdatoms,
 831                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 832                      t_nrnb                      * gmx_restrict        nrnb)
 833 {
 834     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 835      * just 0 for non-waters.
 836      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 837      * jnr indices corresponding to data put in the four positions in the SIMD register.
 838      */
 839     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 840     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 841     int              jnrA,jnrB,jnrC,jnrD;
 842     int              jnrE,jnrF,jnrG,jnrH;
 843     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 844     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 845     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 846     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 847     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 848     real             rcutoff_scalar;
 849     real             *shiftvec,*fshift,*x,*f;
 850     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 851     real             scratch[4*DIM];
 852     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 853     real *           vdwioffsetptr0;
 854     real *           vdwgridioffsetptr0;
 855     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 856     real *           vdwioffsetptr1;
 857     real *           vdwgridioffsetptr1;
 858     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 859     real *           vdwioffsetptr2;
 860     real *           vdwgridioffsetptr2;
 861     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 862     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 863     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 864     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 865     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 866     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 867     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 868     real             *charge;
 869     int              nvdwtype;
 870     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 871     int              *vdwtype;
 872     real             *vdwparam;
 873     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 874     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 875     __m256           c6grid_00;
 876     __m256           c6grid_10;
 877     __m256           c6grid_20;
 878     real             *vdwgridparam;
 879     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 880     __m256           one_half  = _mm256_set1_ps(0.5);
 881     __m256           minus_one = _mm256_set1_ps(-1.0);
 882     __m256i          ewitab;
 883     __m128i          ewitab_lo,ewitab_hi;
 884     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 885     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 886     real             *ewtab;
 887     __m256           dummy_mask,cutoff_mask;
 888     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 889     __m256           one     = _mm256_set1_ps(1.0);
 890     __m256           two     = _mm256_set1_ps(2.0);
 891     x                = xx[0];
 892     f                = ff[0];
 893
 894     nri              = nlist->nri;
 895     iinr             = nlist->iinr;
 896     jindex           = nlist->jindex;
 897     jjnr             = nlist->jjnr;
 898     shiftidx         = nlist->shift;
 899     gid              = nlist->gid;
 900     shiftvec         = fr->shift_vec[0];
 901     fshift           = fr->fshift[0];
 902     facel            = _mm256_set1_ps(fr->epsfac);
 903     charge           = mdatoms->chargeA;
 904     nvdwtype         = fr->ntype;
 905     vdwparam         = fr->nbfp;
 906     vdwtype          = mdatoms->typeA;
 907     vdwgridparam     = fr->ljpme_c6grid;
 908     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 909     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 910     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 911
 912     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 913     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 914     beta2            = _mm256_mul_ps(beta,beta);
 915     beta3            = _mm256_mul_ps(beta,beta2);
 916
 917     ewtab            = fr->ic->tabq_coul_F;
 918     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 919     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 920
 921     /* Setup water-specific parameters */
 922     inr              = nlist->iinr[0];
 923     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 924     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 925     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 926     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 927     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 928
 929     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 930     rcutoff_scalar   = fr->rcoulomb;
 931     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 932     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 933
 934     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 935     rvdw             = _mm256_set1_ps(fr->rvdw);
 936
 937     /* Avoid stupid compiler warnings */
 938     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 939     j_coord_offsetA = 0;
 940     j_coord_offsetB = 0;
 941     j_coord_offsetC = 0;
 942     j_coord_offsetD = 0;
 943     j_coord_offsetE = 0;
 944     j_coord_offsetF = 0;
 945     j_coord_offsetG = 0;
 946     j_coord_offsetH = 0;
 947
 948     outeriter        = 0;
 949     inneriter        = 0;
 950
 951     for(iidx=0;iidx<4*DIM;iidx++)
 952     {
 953         scratch[iidx] = 0.0;
 954     }
 955
 956     /* Start outer loop over neighborlists */
 957     for(iidx=0; iidx<nri; iidx++)
 958     {
 959         /* Load shift vector for this list */
 960         i_shift_offset   = DIM*shiftidx[iidx];
 961
 962         /* Load limits for loop over neighbors */
 963         j_index_start    = jindex[iidx];
 964         j_index_end      = jindex[iidx+1];
 965
 966         /* Get outer coordinate index */
 967         inr              = iinr[iidx];
 968         i_coord_offset   = DIM*inr;
 969
 970         /* Load i particle coords and add shift vector */
 971         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 972                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 973
 974         fix0             = _mm256_setzero_ps();
 975         fiy0             = _mm256_setzero_ps();
 976         fiz0             = _mm256_setzero_ps();
 977         fix1             = _mm256_setzero_ps();
 978         fiy1             = _mm256_setzero_ps();
 979         fiz1             = _mm256_setzero_ps();
 980         fix2             = _mm256_setzero_ps();
 981         fiy2             = _mm256_setzero_ps();
 982         fiz2             = _mm256_setzero_ps();
 983
 984         /* Start inner kernel loop */
 985         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 986         {
 987
 988             /* Get j neighbor index, and coordinate index */
 989             jnrA             = jjnr[jidx];
 990             jnrB             = jjnr[jidx+1];
 991             jnrC             = jjnr[jidx+2];
 992             jnrD             = jjnr[jidx+3];
 993             jnrE             = jjnr[jidx+4];
 994             jnrF             = jjnr[jidx+5];
 995             jnrG             = jjnr[jidx+6];
 996             jnrH             = jjnr[jidx+7];
 997             j_coord_offsetA  = DIM*jnrA;
 998             j_coord_offsetB  = DIM*jnrB;
 999             j_coord_offsetC  = DIM*jnrC;
1000             j_coord_offsetD  = DIM*jnrD;
1001             j_coord_offsetE  = DIM*jnrE;
1002             j_coord_offsetF  = DIM*jnrF;
1003             j_coord_offsetG  = DIM*jnrG;
1004             j_coord_offsetH  = DIM*jnrH;
1005
1006             /* load j atom coordinates */
1007             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1008                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1009                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1010                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1011                                                  &jx0,&jy0,&jz0);
1012
1013             /* Calculate displacement vector */
1014             dx00             = _mm256_sub_ps(ix0,jx0);
1015             dy00             = _mm256_sub_ps(iy0,jy0);
1016             dz00             = _mm256_sub_ps(iz0,jz0);
1017             dx10             = _mm256_sub_ps(ix1,jx0);
1018             dy10             = _mm256_sub_ps(iy1,jy0);
1019             dz10             = _mm256_sub_ps(iz1,jz0);
1020             dx20             = _mm256_sub_ps(ix2,jx0);
1021             dy20             = _mm256_sub_ps(iy2,jy0);
1022             dz20             = _mm256_sub_ps(iz2,jz0);
1023
1024             /* Calculate squared distance and things based on it */
1025             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1026             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1027             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1028
1029             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1030             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1031             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1032
1033             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1034             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1035             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1036
1037             /* Load parameters for j particles */
1038             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1039                                                                  charge+jnrC+0,charge+jnrD+0,
1040                                                                  charge+jnrE+0,charge+jnrF+0,
1041                                                                  charge+jnrG+0,charge+jnrH+0);
1042             vdwjidx0A        = 2*vdwtype[jnrA+0];
1043             vdwjidx0B        = 2*vdwtype[jnrB+0];
1044             vdwjidx0C        = 2*vdwtype[jnrC+0];
1045             vdwjidx0D        = 2*vdwtype[jnrD+0];
1046             vdwjidx0E        = 2*vdwtype[jnrE+0];
1047             vdwjidx0F        = 2*vdwtype[jnrF+0];
1048             vdwjidx0G        = 2*vdwtype[jnrG+0];
1049             vdwjidx0H        = 2*vdwtype[jnrH+0];
1050
1051             fjx0             = _mm256_setzero_ps();
1052             fjy0             = _mm256_setzero_ps();
1053             fjz0             = _mm256_setzero_ps();
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1060             {
1061
1062             r00              = _mm256_mul_ps(rsq00,rinv00);
1063
1064             /* Compute parameters for interactions between i and j atoms */
1065             qq00             = _mm256_mul_ps(iq0,jq0);
1066             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1067                                             vdwioffsetptr0+vdwjidx0B,
1068                                             vdwioffsetptr0+vdwjidx0C,
1069                                             vdwioffsetptr0+vdwjidx0D,
1070                                             vdwioffsetptr0+vdwjidx0E,
1071                                             vdwioffsetptr0+vdwjidx0F,
1072                                             vdwioffsetptr0+vdwjidx0G,
1073                                             vdwioffsetptr0+vdwjidx0H,
1074                                             &c6_00,&c12_00);
1075
1076             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1077                                                                   vdwgridioffsetptr0+vdwjidx0B,
1078                                                                   vdwgridioffsetptr0+vdwjidx0C,
1079                                                                   vdwgridioffsetptr0+vdwjidx0D,
1080                                                                   vdwgridioffsetptr0+vdwjidx0E,
1081                                                                   vdwgridioffsetptr0+vdwjidx0F,
1082                                                                   vdwgridioffsetptr0+vdwjidx0G,
1083                                                                   vdwgridioffsetptr0+vdwjidx0H);
1084
1085             /* EWALD ELECTROSTATICS */
1086
1087             /* Analytical PME correction */
1088             zeta2            = _mm256_mul_ps(beta2,rsq00);
1089             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1090             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1091             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1092             felec            = _mm256_mul_ps(qq00,felec);
1093
1094             /* Analytical LJ-PME */
1095             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1096             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1097             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1098             exponent         = gmx_simd_exp_r(ewcljrsq);
1099             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1100             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1101             /* f6A = 6 * C6grid * (1 - poly) */
1102             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1103             /* f6B = C6grid * exponent * beta^6 */
1104             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1105             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1106             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);
1107
1108             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1109
1110             fscal            = _mm256_add_ps(felec,fvdw);
1111
1112             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1113
1114             /* Calculate temporary vectorial force */
1115             tx               = _mm256_mul_ps(fscal,dx00);
1116             ty               = _mm256_mul_ps(fscal,dy00);
1117             tz               = _mm256_mul_ps(fscal,dz00);
1118
1119             /* Update vectorial force */
1120             fix0             = _mm256_add_ps(fix0,tx);
1121             fiy0             = _mm256_add_ps(fiy0,ty);
1122             fiz0             = _mm256_add_ps(fiz0,tz);
1123
1124             fjx0             = _mm256_add_ps(fjx0,tx);
1125             fjy0             = _mm256_add_ps(fjy0,ty);
1126             fjz0             = _mm256_add_ps(fjz0,tz);
1127
1128             }
1129
1130             /**************************
1131              * CALCULATE INTERACTIONS *
1132              **************************/
1133
1134             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1135             {
1136
1137             r10              = _mm256_mul_ps(rsq10,rinv10);
1138
1139             /* Compute parameters for interactions between i and j atoms */
1140             qq10             = _mm256_mul_ps(iq1,jq0);
1141
1142             /* EWALD ELECTROSTATICS */
1143
1144             /* Analytical PME correction */
1145             zeta2            = _mm256_mul_ps(beta2,rsq10);
1146             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1147             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1148             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1149             felec            = _mm256_mul_ps(qq10,felec);
1150
1151             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1152
1153             fscal            = felec;
1154
1155             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1156
1157             /* Calculate temporary vectorial force */
1158             tx               = _mm256_mul_ps(fscal,dx10);
1159             ty               = _mm256_mul_ps(fscal,dy10);
1160             tz               = _mm256_mul_ps(fscal,dz10);
1161
1162             /* Update vectorial force */
1163             fix1             = _mm256_add_ps(fix1,tx);
1164             fiy1             = _mm256_add_ps(fiy1,ty);
1165             fiz1             = _mm256_add_ps(fiz1,tz);
1166
1167             fjx0             = _mm256_add_ps(fjx0,tx);
1168             fjy0             = _mm256_add_ps(fjy0,ty);
1169             fjz0             = _mm256_add_ps(fjz0,tz);
1170
1171             }
1172
1173             /**************************
1174              * CALCULATE INTERACTIONS *
1175              **************************/
1176
1177             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1178             {
1179
1180             r20              = _mm256_mul_ps(rsq20,rinv20);
1181
1182             /* Compute parameters for interactions between i and j atoms */
1183             qq20             = _mm256_mul_ps(iq2,jq0);
1184
1185             /* EWALD ELECTROSTATICS */
1186
1187             /* Analytical PME correction */
1188             zeta2            = _mm256_mul_ps(beta2,rsq20);
1189             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1190             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1191             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1192             felec            = _mm256_mul_ps(qq20,felec);
1193
1194             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1195
1196             fscal            = felec;
1197
1198             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1199
1200             /* Calculate temporary vectorial force */
1201             tx               = _mm256_mul_ps(fscal,dx20);
1202             ty               = _mm256_mul_ps(fscal,dy20);
1203             tz               = _mm256_mul_ps(fscal,dz20);
1204
1205             /* Update vectorial force */
1206             fix2             = _mm256_add_ps(fix2,tx);
1207             fiy2             = _mm256_add_ps(fiy2,ty);
1208             fiz2             = _mm256_add_ps(fiz2,tz);
1209
1210             fjx0             = _mm256_add_ps(fjx0,tx);
1211             fjy0             = _mm256_add_ps(fjy0,ty);
1212             fjz0             = _mm256_add_ps(fjz0,tz);
1213
1214             }
1215
1216             fjptrA             = f+j_coord_offsetA;
1217             fjptrB             = f+j_coord_offsetB;
1218             fjptrC             = f+j_coord_offsetC;
1219             fjptrD             = f+j_coord_offsetD;
1220             fjptrE             = f+j_coord_offsetE;
1221             fjptrF             = f+j_coord_offsetF;
1222             fjptrG             = f+j_coord_offsetG;
1223             fjptrH             = f+j_coord_offsetH;
1224
1225             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1226
1227             /* Inner loop uses 203 flops */
1228         }
1229
1230         if(jidx<j_index_end)
1231         {
1232
1233             /* Get j neighbor index, and coordinate index */
1234             jnrlistA         = jjnr[jidx];
1235             jnrlistB         = jjnr[jidx+1];
1236             jnrlistC         = jjnr[jidx+2];
1237             jnrlistD         = jjnr[jidx+3];
1238             jnrlistE         = jjnr[jidx+4];
1239             jnrlistF         = jjnr[jidx+5];
1240             jnrlistG         = jjnr[jidx+6];
1241             jnrlistH         = jjnr[jidx+7];
1242             /* Sign of each element will be negative for non-real atoms.
1243              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1244              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1245              */
1246             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1247                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1248
1249             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1250             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1251             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1252             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1253             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1254             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1255             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1256             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1257             j_coord_offsetA  = DIM*jnrA;
1258             j_coord_offsetB  = DIM*jnrB;
1259             j_coord_offsetC  = DIM*jnrC;
1260             j_coord_offsetD  = DIM*jnrD;
1261             j_coord_offsetE  = DIM*jnrE;
1262             j_coord_offsetF  = DIM*jnrF;
1263             j_coord_offsetG  = DIM*jnrG;
1264             j_coord_offsetH  = DIM*jnrH;
1265
1266             /* load j atom coordinates */
1267             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1268                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1269                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1270                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1271                                                  &jx0,&jy0,&jz0);
1272
1273             /* Calculate displacement vector */
1274             dx00             = _mm256_sub_ps(ix0,jx0);
1275             dy00             = _mm256_sub_ps(iy0,jy0);
1276             dz00             = _mm256_sub_ps(iz0,jz0);
1277             dx10             = _mm256_sub_ps(ix1,jx0);
1278             dy10             = _mm256_sub_ps(iy1,jy0);
1279             dz10             = _mm256_sub_ps(iz1,jz0);
1280             dx20             = _mm256_sub_ps(ix2,jx0);
1281             dy20             = _mm256_sub_ps(iy2,jy0);
1282             dz20             = _mm256_sub_ps(iz2,jz0);
1283
1284             /* Calculate squared distance and things based on it */
1285             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1286             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1287             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1288
1289             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1290             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1291             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1292
1293             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1294             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1295             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1296
1297             /* Load parameters for j particles */
1298             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1299                                                                  charge+jnrC+0,charge+jnrD+0,
1300                                                                  charge+jnrE+0,charge+jnrF+0,
1301                                                                  charge+jnrG+0,charge+jnrH+0);
1302             vdwjidx0A        = 2*vdwtype[jnrA+0];
1303             vdwjidx0B        = 2*vdwtype[jnrB+0];
1304             vdwjidx0C        = 2*vdwtype[jnrC+0];
1305             vdwjidx0D        = 2*vdwtype[jnrD+0];
1306             vdwjidx0E        = 2*vdwtype[jnrE+0];
1307             vdwjidx0F        = 2*vdwtype[jnrF+0];
1308             vdwjidx0G        = 2*vdwtype[jnrG+0];
1309             vdwjidx0H        = 2*vdwtype[jnrH+0];
1310
1311             fjx0             = _mm256_setzero_ps();
1312             fjy0             = _mm256_setzero_ps();
1313             fjz0             = _mm256_setzero_ps();
1314
1315             /**************************
1316              * CALCULATE INTERACTIONS *
1317              **************************/
1318
1319             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1320             {
1321
1322             r00              = _mm256_mul_ps(rsq00,rinv00);
1323             r00              = _mm256_andnot_ps(dummy_mask,r00);
1324
1325             /* Compute parameters for interactions between i and j atoms */
1326             qq00             = _mm256_mul_ps(iq0,jq0);
1327             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1328                                             vdwioffsetptr0+vdwjidx0B,
1329                                             vdwioffsetptr0+vdwjidx0C,
1330                                             vdwioffsetptr0+vdwjidx0D,
1331                                             vdwioffsetptr0+vdwjidx0E,
1332                                             vdwioffsetptr0+vdwjidx0F,
1333                                             vdwioffsetptr0+vdwjidx0G,
1334                                             vdwioffsetptr0+vdwjidx0H,
1335                                             &c6_00,&c12_00);
1336
1337             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1338                                                                   vdwgridioffsetptr0+vdwjidx0B,
1339                                                                   vdwgridioffsetptr0+vdwjidx0C,
1340                                                                   vdwgridioffsetptr0+vdwjidx0D,
1341                                                                   vdwgridioffsetptr0+vdwjidx0E,
1342                                                                   vdwgridioffsetptr0+vdwjidx0F,
1343                                                                   vdwgridioffsetptr0+vdwjidx0G,
1344                                                                   vdwgridioffsetptr0+vdwjidx0H);
1345
1346             /* EWALD ELECTROSTATICS */
1347
1348             /* Analytical PME correction */
1349             zeta2            = _mm256_mul_ps(beta2,rsq00);
1350             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1351             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1352             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1353             felec            = _mm256_mul_ps(qq00,felec);
1354
1355             /* Analytical LJ-PME */
1356             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1357             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1358             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1359             exponent         = gmx_simd_exp_r(ewcljrsq);
1360             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1361             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1362             /* f6A = 6 * C6grid * (1 - poly) */
1363             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1364             /* f6B = C6grid * exponent * beta^6 */
1365             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1366             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1367             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);
1368
1369             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1370
1371             fscal            = _mm256_add_ps(felec,fvdw);
1372
1373             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1374
1375             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1376
1377             /* Calculate temporary vectorial force */
1378             tx               = _mm256_mul_ps(fscal,dx00);
1379             ty               = _mm256_mul_ps(fscal,dy00);
1380             tz               = _mm256_mul_ps(fscal,dz00);
1381
1382             /* Update vectorial force */
1383             fix0             = _mm256_add_ps(fix0,tx);
1384             fiy0             = _mm256_add_ps(fiy0,ty);
1385             fiz0             = _mm256_add_ps(fiz0,tz);
1386
1387             fjx0             = _mm256_add_ps(fjx0,tx);
1388             fjy0             = _mm256_add_ps(fjy0,ty);
1389             fjz0             = _mm256_add_ps(fjz0,tz);
1390
1391             }
1392
1393             /**************************
1394              * CALCULATE INTERACTIONS *
1395              **************************/
1396
1397             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1398             {
1399
1400             r10              = _mm256_mul_ps(rsq10,rinv10);
1401             r10              = _mm256_andnot_ps(dummy_mask,r10);
1402
1403             /* Compute parameters for interactions between i and j atoms */
1404             qq10             = _mm256_mul_ps(iq1,jq0);
1405
1406             /* EWALD ELECTROSTATICS */
1407
1408             /* Analytical PME correction */
1409             zeta2            = _mm256_mul_ps(beta2,rsq10);
1410             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1411             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1412             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1413             felec            = _mm256_mul_ps(qq10,felec);
1414
1415             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1416
1417             fscal            = felec;
1418
1419             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1420
1421             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1422
1423             /* Calculate temporary vectorial force */
1424             tx               = _mm256_mul_ps(fscal,dx10);
1425             ty               = _mm256_mul_ps(fscal,dy10);
1426             tz               = _mm256_mul_ps(fscal,dz10);
1427
1428             /* Update vectorial force */
1429             fix1             = _mm256_add_ps(fix1,tx);
1430             fiy1             = _mm256_add_ps(fiy1,ty);
1431             fiz1             = _mm256_add_ps(fiz1,tz);
1432
1433             fjx0             = _mm256_add_ps(fjx0,tx);
1434             fjy0             = _mm256_add_ps(fjy0,ty);
1435             fjz0             = _mm256_add_ps(fjz0,tz);
1436
1437             }
1438
1439             /**************************
1440              * CALCULATE INTERACTIONS *
1441              **************************/
1442
1443             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1444             {
1445
1446             r20              = _mm256_mul_ps(rsq20,rinv20);
1447             r20              = _mm256_andnot_ps(dummy_mask,r20);
1448
1449             /* Compute parameters for interactions between i and j atoms */
1450             qq20             = _mm256_mul_ps(iq2,jq0);
1451
1452             /* EWALD ELECTROSTATICS */
1453
1454             /* Analytical PME correction */
1455             zeta2            = _mm256_mul_ps(beta2,rsq20);
1456             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1457             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1458             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1459             felec            = _mm256_mul_ps(qq20,felec);
1460
1461             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1462
1463             fscal            = felec;
1464
1465             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1466
1467             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1468
1469             /* Calculate temporary vectorial force */
1470             tx               = _mm256_mul_ps(fscal,dx20);
1471             ty               = _mm256_mul_ps(fscal,dy20);
1472             tz               = _mm256_mul_ps(fscal,dz20);
1473
1474             /* Update vectorial force */
1475             fix2             = _mm256_add_ps(fix2,tx);
1476             fiy2             = _mm256_add_ps(fiy2,ty);
1477             fiz2             = _mm256_add_ps(fiz2,tz);
1478
1479             fjx0             = _mm256_add_ps(fjx0,tx);
1480             fjy0             = _mm256_add_ps(fjy0,ty);
1481             fjz0             = _mm256_add_ps(fjz0,tz);
1482
1483             }
1484
1485             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1486             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1487             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1488             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1489             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1490             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1491             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1492             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1493
1494             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1495
1496             /* Inner loop uses 206 flops */
1497         }
1498
1499         /* End of innermost loop */
1500
1501         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1502                                                  f+i_coord_offset,fshift+i_shift_offset);
1503
1504         /* Increment number of inner iterations */
1505         inneriter                  += j_index_end - j_index_start;
1506
1507         /* Outer loop uses 18 flops */
1508     }
1509
1510     /* Increment number of outer iterations */
1511     outeriter        += nri;
1512
1513     /* Update outer/inner flops */
1514
1515     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*206);
1516 }