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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     real *           vdwgridioffsetptr0;
  90     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  91     real *           vdwioffsetptr1;
  92     real *           vdwgridioffsetptr1;
  93     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  94     real *           vdwioffsetptr2;
  95     real *           vdwgridioffsetptr2;
  96     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  97     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  98     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  99     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 100     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 101     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 102     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 103     real             *charge;
 104     int              nvdwtype;
 105     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 106     int              *vdwtype;
 107     real             *vdwparam;
 108     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 109     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 110     __m256           c6grid_00;
 111     __m256           c6grid_10;
 112     __m256           c6grid_20;
 113     real             *vdwgridparam;
 114     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 115     __m256           one_half  = _mm256_set1_ps(0.5);
 116     __m256           minus_one = _mm256_set1_ps(-1.0);
 117     __m256i          ewitab;
 118     __m128i          ewitab_lo,ewitab_hi;
 119     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 120     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 121     real             *ewtab;
 122     __m256           dummy_mask,cutoff_mask;
 123     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 124     __m256           one     = _mm256_set1_ps(1.0);
 125     __m256           two     = _mm256_set1_ps(2.0);
 126     x                = xx[0];
 127     f                = ff[0];
 128
 129     nri              = nlist->nri;
 130     iinr             = nlist->iinr;
 131     jindex           = nlist->jindex;
 132     jjnr             = nlist->jjnr;
 133     shiftidx         = nlist->shift;
 134     gid              = nlist->gid;
 135     shiftvec         = fr->shift_vec[0];
 136     fshift           = fr->fshift[0];
 137     facel            = _mm256_set1_ps(fr->epsfac);
 138     charge           = mdatoms->chargeA;
 139     nvdwtype         = fr->ntype;
 140     vdwparam         = fr->nbfp;
 141     vdwtype          = mdatoms->typeA;
 142     vdwgridparam     = fr->ljpme_c6grid;
 143     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 144     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 145     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 146
 147     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 148     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 149     beta2            = _mm256_mul_ps(beta,beta);
 150     beta3            = _mm256_mul_ps(beta,beta2);
 151
 152     ewtab            = fr->ic->tabq_coul_FDV0;
 153     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 154     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 155
 156     /* Setup water-specific parameters */
 157     inr              = nlist->iinr[0];
 158     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 159     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 160     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 161     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 162     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 163
 164     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 165     rcutoff_scalar   = fr->rcoulomb;
 166     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 167     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 168
 169     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 170     rvdw             = _mm256_set1_ps(fr->rvdw);
 171
 172     /* Avoid stupid compiler warnings */
 173     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 174     j_coord_offsetA = 0;
 175     j_coord_offsetB = 0;
 176     j_coord_offsetC = 0;
 177     j_coord_offsetD = 0;
 178     j_coord_offsetE = 0;
 179     j_coord_offsetF = 0;
 180     j_coord_offsetG = 0;
 181     j_coord_offsetH = 0;
 182
 183     outeriter        = 0;
 184     inneriter        = 0;
 185
 186     for(iidx=0;iidx<4*DIM;iidx++)
 187     {
 188         scratch[iidx] = 0.0;
 189     }
 190
 191     /* Start outer loop over neighborlists */
 192     for(iidx=0; iidx<nri; iidx++)
 193     {
 194         /* Load shift vector for this list */
 195         i_shift_offset   = DIM*shiftidx[iidx];
 196
 197         /* Load limits for loop over neighbors */
 198         j_index_start    = jindex[iidx];
 199         j_index_end      = jindex[iidx+1];
 200
 201         /* Get outer coordinate index */
 202         inr              = iinr[iidx];
 203         i_coord_offset   = DIM*inr;
 204
 205         /* Load i particle coords and add shift vector */
 206         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 207                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 208
 209         fix0             = _mm256_setzero_ps();
 210         fiy0             = _mm256_setzero_ps();
 211         fiz0             = _mm256_setzero_ps();
 212         fix1             = _mm256_setzero_ps();
 213         fiy1             = _mm256_setzero_ps();
 214         fiz1             = _mm256_setzero_ps();
 215         fix2             = _mm256_setzero_ps();
 216         fiy2             = _mm256_setzero_ps();
 217         fiz2             = _mm256_setzero_ps();
 218
 219         /* Reset potential sums */
 220         velecsum         = _mm256_setzero_ps();
 221         vvdwsum          = _mm256_setzero_ps();
 222
 223         /* Start inner kernel loop */
 224         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 225         {
 226
 227             /* Get j neighbor index, and coordinate index */
 228             jnrA             = jjnr[jidx];
 229             jnrB             = jjnr[jidx+1];
 230             jnrC             = jjnr[jidx+2];
 231             jnrD             = jjnr[jidx+3];
 232             jnrE             = jjnr[jidx+4];
 233             jnrF             = jjnr[jidx+5];
 234             jnrG             = jjnr[jidx+6];
 235             jnrH             = jjnr[jidx+7];
 236             j_coord_offsetA  = DIM*jnrA;
 237             j_coord_offsetB  = DIM*jnrB;
 238             j_coord_offsetC  = DIM*jnrC;
 239             j_coord_offsetD  = DIM*jnrD;
 240             j_coord_offsetE  = DIM*jnrE;
 241             j_coord_offsetF  = DIM*jnrF;
 242             j_coord_offsetG  = DIM*jnrG;
 243             j_coord_offsetH  = DIM*jnrH;
 244
 245             /* load j atom coordinates */
 246             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 247                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 248                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 249                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 250                                                  &jx0,&jy0,&jz0);
 251
 252             /* Calculate displacement vector */
 253             dx00             = _mm256_sub_ps(ix0,jx0);
 254             dy00             = _mm256_sub_ps(iy0,jy0);
 255             dz00             = _mm256_sub_ps(iz0,jz0);
 256             dx10             = _mm256_sub_ps(ix1,jx0);
 257             dy10             = _mm256_sub_ps(iy1,jy0);
 258             dz10             = _mm256_sub_ps(iz1,jz0);
 259             dx20             = _mm256_sub_ps(ix2,jx0);
 260             dy20             = _mm256_sub_ps(iy2,jy0);
 261             dz20             = _mm256_sub_ps(iz2,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 265             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 266             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 267
 268             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 269             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 270             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 271
 272             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 273             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 274             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 275
 276             /* Load parameters for j particles */
 277             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 278                                                                  charge+jnrC+0,charge+jnrD+0,
 279                                                                  charge+jnrE+0,charge+jnrF+0,
 280                                                                  charge+jnrG+0,charge+jnrH+0);
 281             vdwjidx0A        = 2*vdwtype[jnrA+0];
 282             vdwjidx0B        = 2*vdwtype[jnrB+0];
 283             vdwjidx0C        = 2*vdwtype[jnrC+0];
 284             vdwjidx0D        = 2*vdwtype[jnrD+0];
 285             vdwjidx0E        = 2*vdwtype[jnrE+0];
 286             vdwjidx0F        = 2*vdwtype[jnrF+0];
 287             vdwjidx0G        = 2*vdwtype[jnrG+0];
 288             vdwjidx0H        = 2*vdwtype[jnrH+0];
 289
 290             fjx0             = _mm256_setzero_ps();
 291             fjy0             = _mm256_setzero_ps();
 292             fjz0             = _mm256_setzero_ps();
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 299             {
 300
 301             r00              = _mm256_mul_ps(rsq00,rinv00);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq00             = _mm256_mul_ps(iq0,jq0);
 305             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 306                                             vdwioffsetptr0+vdwjidx0B,
 307                                             vdwioffsetptr0+vdwjidx0C,
 308                                             vdwioffsetptr0+vdwjidx0D,
 309                                             vdwioffsetptr0+vdwjidx0E,
 310                                             vdwioffsetptr0+vdwjidx0F,
 311                                             vdwioffsetptr0+vdwjidx0G,
 312                                             vdwioffsetptr0+vdwjidx0H,
 313                                             &c6_00,&c12_00);
 314
 315             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 316                                                                   vdwgridioffsetptr0+vdwjidx0B,
 317                                                                   vdwgridioffsetptr0+vdwjidx0C,
 318                                                                   vdwgridioffsetptr0+vdwjidx0D,
 319                                                                   vdwgridioffsetptr0+vdwjidx0E,
 320                                                                   vdwgridioffsetptr0+vdwjidx0F,
 321                                                                   vdwgridioffsetptr0+vdwjidx0G,
 322                                                                   vdwgridioffsetptr0+vdwjidx0H);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Analytical PME correction */
 327             zeta2            = _mm256_mul_ps(beta2,rsq00);
 328             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 329             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 330             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 331             felec            = _mm256_mul_ps(qq00,felec);
 332             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 333             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 334             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 335             velec            = _mm256_mul_ps(qq00,velec);
 336
 337             /* Analytical LJ-PME */
 338             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 339             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 340             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 341             exponent         = gmx_simd_exp_r(ewcljrsq);
 342             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 343             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 344             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 345             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 346             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 347             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) ,
 348                                           _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));
 349             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 350             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);
 351
 352             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velec            = _mm256_and_ps(velec,cutoff_mask);
 356             velecsum         = _mm256_add_ps(velecsum,velec);
 357             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 358             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 359
 360             fscal            = _mm256_add_ps(felec,fvdw);
 361
 362             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm256_mul_ps(fscal,dx00);
 366             ty               = _mm256_mul_ps(fscal,dy00);
 367             tz               = _mm256_mul_ps(fscal,dz00);
 368
 369             /* Update vectorial force */
 370             fix0             = _mm256_add_ps(fix0,tx);
 371             fiy0             = _mm256_add_ps(fiy0,ty);
 372             fiz0             = _mm256_add_ps(fiz0,tz);
 373
 374             fjx0             = _mm256_add_ps(fjx0,tx);
 375             fjy0             = _mm256_add_ps(fjy0,ty);
 376             fjz0             = _mm256_add_ps(fjz0,tz);
 377
 378             }
 379
 380             /**************************
 381              * CALCULATE INTERACTIONS *
 382              **************************/
 383
 384             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 385             {
 386
 387             r10              = _mm256_mul_ps(rsq10,rinv10);
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq10             = _mm256_mul_ps(iq1,jq0);
 391
 392             /* EWALD ELECTROSTATICS */
 393
 394             /* Analytical PME correction */
 395             zeta2            = _mm256_mul_ps(beta2,rsq10);
 396             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 397             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 398             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 399             felec            = _mm256_mul_ps(qq10,felec);
 400             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 401             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 402             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 403             velec            = _mm256_mul_ps(qq10,velec);
 404
 405             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velec            = _mm256_and_ps(velec,cutoff_mask);
 409             velecsum         = _mm256_add_ps(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 414
 415             /* Calculate temporary vectorial force */
 416             tx               = _mm256_mul_ps(fscal,dx10);
 417             ty               = _mm256_mul_ps(fscal,dy10);
 418             tz               = _mm256_mul_ps(fscal,dz10);
 419
 420             /* Update vectorial force */
 421             fix1             = _mm256_add_ps(fix1,tx);
 422             fiy1             = _mm256_add_ps(fiy1,ty);
 423             fiz1             = _mm256_add_ps(fiz1,tz);
 424
 425             fjx0             = _mm256_add_ps(fjx0,tx);
 426             fjy0             = _mm256_add_ps(fjy0,ty);
 427             fjz0             = _mm256_add_ps(fjz0,tz);
 428
 429             }
 430
 431             /**************************
 432              * CALCULATE INTERACTIONS *
 433              **************************/
 434
 435             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 436             {
 437
 438             r20              = _mm256_mul_ps(rsq20,rinv20);
 439
 440             /* Compute parameters for interactions between i and j atoms */
 441             qq20             = _mm256_mul_ps(iq2,jq0);
 442
 443             /* EWALD ELECTROSTATICS */
 444
 445             /* Analytical PME correction */
 446             zeta2            = _mm256_mul_ps(beta2,rsq20);
 447             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 448             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 449             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 450             felec            = _mm256_mul_ps(qq20,felec);
 451             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 452             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 453             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 454             velec            = _mm256_mul_ps(qq20,velec);
 455
 456             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velec            = _mm256_and_ps(velec,cutoff_mask);
 460             velecsum         = _mm256_add_ps(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm256_mul_ps(fscal,dx20);
 468             ty               = _mm256_mul_ps(fscal,dy20);
 469             tz               = _mm256_mul_ps(fscal,dz20);
 470
 471             /* Update vectorial force */
 472             fix2             = _mm256_add_ps(fix2,tx);
 473             fiy2             = _mm256_add_ps(fiy2,ty);
 474             fiz2             = _mm256_add_ps(fiz2,tz);
 475
 476             fjx0             = _mm256_add_ps(fjx0,tx);
 477             fjy0             = _mm256_add_ps(fjy0,ty);
 478             fjz0             = _mm256_add_ps(fjz0,tz);
 479
 480             }
 481
 482             fjptrA             = f+j_coord_offsetA;
 483             fjptrB             = f+j_coord_offsetB;
 484             fjptrC             = f+j_coord_offsetC;
 485             fjptrD             = f+j_coord_offsetD;
 486             fjptrE             = f+j_coord_offsetE;
 487             fjptrF             = f+j_coord_offsetF;
 488             fjptrG             = f+j_coord_offsetG;
 489             fjptrH             = f+j_coord_offsetH;
 490
 491             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 492
 493             /* Inner loop uses 366 flops */
 494         }
 495
 496         if(jidx<j_index_end)
 497         {
 498
 499             /* Get j neighbor index, and coordinate index */
 500             jnrlistA         = jjnr[jidx];
 501             jnrlistB         = jjnr[jidx+1];
 502             jnrlistC         = jjnr[jidx+2];
 503             jnrlistD         = jjnr[jidx+3];
 504             jnrlistE         = jjnr[jidx+4];
 505             jnrlistF         = jjnr[jidx+5];
 506             jnrlistG         = jjnr[jidx+6];
 507             jnrlistH         = jjnr[jidx+7];
 508             /* Sign of each element will be negative for non-real atoms.
 509              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 510              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 511              */
 512             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 513                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 514
 515             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 516             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 517             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 518             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 519             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 520             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 521             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 522             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 523             j_coord_offsetA  = DIM*jnrA;
 524             j_coord_offsetB  = DIM*jnrB;
 525             j_coord_offsetC  = DIM*jnrC;
 526             j_coord_offsetD  = DIM*jnrD;
 527             j_coord_offsetE  = DIM*jnrE;
 528             j_coord_offsetF  = DIM*jnrF;
 529             j_coord_offsetG  = DIM*jnrG;
 530             j_coord_offsetH  = DIM*jnrH;
 531
 532             /* load j atom coordinates */
 533             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 534                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 535                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 536                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 537                                                  &jx0,&jy0,&jz0);
 538
 539             /* Calculate displacement vector */
 540             dx00             = _mm256_sub_ps(ix0,jx0);
 541             dy00             = _mm256_sub_ps(iy0,jy0);
 542             dz00             = _mm256_sub_ps(iz0,jz0);
 543             dx10             = _mm256_sub_ps(ix1,jx0);
 544             dy10             = _mm256_sub_ps(iy1,jy0);
 545             dz10             = _mm256_sub_ps(iz1,jz0);
 546             dx20             = _mm256_sub_ps(ix2,jx0);
 547             dy20             = _mm256_sub_ps(iy2,jy0);
 548             dz20             = _mm256_sub_ps(iz2,jz0);
 549
 550             /* Calculate squared distance and things based on it */
 551             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 552             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 553             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 554
 555             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 556             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 557             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 558
 559             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 560             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 561             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 562
 563             /* Load parameters for j particles */
 564             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 565                                                                  charge+jnrC+0,charge+jnrD+0,
 566                                                                  charge+jnrE+0,charge+jnrF+0,
 567                                                                  charge+jnrG+0,charge+jnrH+0);
 568             vdwjidx0A        = 2*vdwtype[jnrA+0];
 569             vdwjidx0B        = 2*vdwtype[jnrB+0];
 570             vdwjidx0C        = 2*vdwtype[jnrC+0];
 571             vdwjidx0D        = 2*vdwtype[jnrD+0];
 572             vdwjidx0E        = 2*vdwtype[jnrE+0];
 573             vdwjidx0F        = 2*vdwtype[jnrF+0];
 574             vdwjidx0G        = 2*vdwtype[jnrG+0];
 575             vdwjidx0H        = 2*vdwtype[jnrH+0];
 576
 577             fjx0             = _mm256_setzero_ps();
 578             fjy0             = _mm256_setzero_ps();
 579             fjz0             = _mm256_setzero_ps();
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 586             {
 587
 588             r00              = _mm256_mul_ps(rsq00,rinv00);
 589             r00              = _mm256_andnot_ps(dummy_mask,r00);
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq00             = _mm256_mul_ps(iq0,jq0);
 593             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 594                                             vdwioffsetptr0+vdwjidx0B,
 595                                             vdwioffsetptr0+vdwjidx0C,
 596                                             vdwioffsetptr0+vdwjidx0D,
 597                                             vdwioffsetptr0+vdwjidx0E,
 598                                             vdwioffsetptr0+vdwjidx0F,
 599                                             vdwioffsetptr0+vdwjidx0G,
 600                                             vdwioffsetptr0+vdwjidx0H,
 601                                             &c6_00,&c12_00);
 602
 603             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 604                                                                   vdwgridioffsetptr0+vdwjidx0B,
 605                                                                   vdwgridioffsetptr0+vdwjidx0C,
 606                                                                   vdwgridioffsetptr0+vdwjidx0D,
 607                                                                   vdwgridioffsetptr0+vdwjidx0E,
 608                                                                   vdwgridioffsetptr0+vdwjidx0F,
 609                                                                   vdwgridioffsetptr0+vdwjidx0G,
 610                                                                   vdwgridioffsetptr0+vdwjidx0H);
 611
 612             /* EWALD ELECTROSTATICS */
 613
 614             /* Analytical PME correction */
 615             zeta2            = _mm256_mul_ps(beta2,rsq00);
 616             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 617             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 618             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 619             felec            = _mm256_mul_ps(qq00,felec);
 620             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 621             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 622             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 623             velec            = _mm256_mul_ps(qq00,velec);
 624
 625             /* Analytical LJ-PME */
 626             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 627             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 628             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 629             exponent         = gmx_simd_exp_r(ewcljrsq);
 630             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 631             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 632             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 633             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 634             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 635             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) ,
 636                                           _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));
 637             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 638             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);
 639
 640             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 641
 642             /* Update potential sum for this i atom from the interaction with this j atom. */
 643             velec            = _mm256_and_ps(velec,cutoff_mask);
 644             velec            = _mm256_andnot_ps(dummy_mask,velec);
 645             velecsum         = _mm256_add_ps(velecsum,velec);
 646             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 647             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 648             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 649
 650             fscal            = _mm256_add_ps(felec,fvdw);
 651
 652             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 653
 654             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 655
 656             /* Calculate temporary vectorial force */
 657             tx               = _mm256_mul_ps(fscal,dx00);
 658             ty               = _mm256_mul_ps(fscal,dy00);
 659             tz               = _mm256_mul_ps(fscal,dz00);
 660
 661             /* Update vectorial force */
 662             fix0             = _mm256_add_ps(fix0,tx);
 663             fiy0             = _mm256_add_ps(fiy0,ty);
 664             fiz0             = _mm256_add_ps(fiz0,tz);
 665
 666             fjx0             = _mm256_add_ps(fjx0,tx);
 667             fjy0             = _mm256_add_ps(fjy0,ty);
 668             fjz0             = _mm256_add_ps(fjz0,tz);
 669
 670             }
 671
 672             /**************************
 673              * CALCULATE INTERACTIONS *
 674              **************************/
 675
 676             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 677             {
 678
 679             r10              = _mm256_mul_ps(rsq10,rinv10);
 680             r10              = _mm256_andnot_ps(dummy_mask,r10);
 681
 682             /* Compute parameters for interactions between i and j atoms */
 683             qq10             = _mm256_mul_ps(iq1,jq0);
 684
 685             /* EWALD ELECTROSTATICS */
 686
 687             /* Analytical PME correction */
 688             zeta2            = _mm256_mul_ps(beta2,rsq10);
 689             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 690             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 691             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 692             felec            = _mm256_mul_ps(qq10,felec);
 693             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 694             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 695             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 696             velec            = _mm256_mul_ps(qq10,velec);
 697
 698             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 699
 700             /* Update potential sum for this i atom from the interaction with this j atom. */
 701             velec            = _mm256_and_ps(velec,cutoff_mask);
 702             velec            = _mm256_andnot_ps(dummy_mask,velec);
 703             velecsum         = _mm256_add_ps(velecsum,velec);
 704
 705             fscal            = felec;
 706
 707             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 708
 709             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 710
 711             /* Calculate temporary vectorial force */
 712             tx               = _mm256_mul_ps(fscal,dx10);
 713             ty               = _mm256_mul_ps(fscal,dy10);
 714             tz               = _mm256_mul_ps(fscal,dz10);
 715
 716             /* Update vectorial force */
 717             fix1             = _mm256_add_ps(fix1,tx);
 718             fiy1             = _mm256_add_ps(fiy1,ty);
 719             fiz1             = _mm256_add_ps(fiz1,tz);
 720
 721             fjx0             = _mm256_add_ps(fjx0,tx);
 722             fjy0             = _mm256_add_ps(fjy0,ty);
 723             fjz0             = _mm256_add_ps(fjz0,tz);
 724
 725             }
 726
 727             /**************************
 728              * CALCULATE INTERACTIONS *
 729              **************************/
 730
 731             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 732             {
 733
 734             r20              = _mm256_mul_ps(rsq20,rinv20);
 735             r20              = _mm256_andnot_ps(dummy_mask,r20);
 736
 737             /* Compute parameters for interactions between i and j atoms */
 738             qq20             = _mm256_mul_ps(iq2,jq0);
 739
 740             /* EWALD ELECTROSTATICS */
 741
 742             /* Analytical PME correction */
 743             zeta2            = _mm256_mul_ps(beta2,rsq20);
 744             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 745             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 746             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 747             felec            = _mm256_mul_ps(qq20,felec);
 748             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 749             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 750             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 751             velec            = _mm256_mul_ps(qq20,velec);
 752
 753             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 754
 755             /* Update potential sum for this i atom from the interaction with this j atom. */
 756             velec            = _mm256_and_ps(velec,cutoff_mask);
 757             velec            = _mm256_andnot_ps(dummy_mask,velec);
 758             velecsum         = _mm256_add_ps(velecsum,velec);
 759
 760             fscal            = felec;
 761
 762             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 763
 764             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 765
 766             /* Calculate temporary vectorial force */
 767             tx               = _mm256_mul_ps(fscal,dx20);
 768             ty               = _mm256_mul_ps(fscal,dy20);
 769             tz               = _mm256_mul_ps(fscal,dz20);
 770
 771             /* Update vectorial force */
 772             fix2             = _mm256_add_ps(fix2,tx);
 773             fiy2             = _mm256_add_ps(fiy2,ty);
 774             fiz2             = _mm256_add_ps(fiz2,tz);
 775
 776             fjx0             = _mm256_add_ps(fjx0,tx);
 777             fjy0             = _mm256_add_ps(fjy0,ty);
 778             fjz0             = _mm256_add_ps(fjz0,tz);
 779
 780             }
 781
 782             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 783             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 784             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 785             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 786             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 787             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 788             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 789             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 790
 791             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 792
 793             /* Inner loop uses 369 flops */
 794         }
 795
 796         /* End of innermost loop */
 797
 798         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 799                                                  f+i_coord_offset,fshift+i_shift_offset);
 800
 801         ggid                        = gid[iidx];
 802         /* Update potential energies */
 803         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 804         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 805
 806         /* Increment number of inner iterations */
 807         inneriter                  += j_index_end - j_index_start;
 808
 809         /* Outer loop uses 20 flops */
 810     }
 811
 812     /* Increment number of outer iterations */
 813     outeriter        += nri;
 814
 815     /* Update outer/inner flops */
 816
 817     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*369);
 818 }
 819 /*
 820  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_single
 821  * Electrostatics interaction: Ewald
 822  * VdW interaction:            LJEwald
 823  * Geometry:                   Water3-Particle
 824  * Calculate force/pot:        Force
 825  */
 826 void
 827 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_single
 828                     (t_nblist                    * gmx_restrict       nlist,
 829                      rvec                        * gmx_restrict          xx,
 830                      rvec                        * gmx_restrict          ff,
 831                      t_forcerec                  * gmx_restrict          fr,
 832                      t_mdatoms                   * gmx_restrict     mdatoms,
 833                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 834                      t_nrnb                      * gmx_restrict        nrnb)
 835 {
 836     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 837      * just 0 for non-waters.
 838      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 839      * jnr indices corresponding to data put in the four positions in the SIMD register.
 840      */
 841     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 842     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 843     int              jnrA,jnrB,jnrC,jnrD;
 844     int              jnrE,jnrF,jnrG,jnrH;
 845     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 846     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 847     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 848     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 849     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 850     real             rcutoff_scalar;
 851     real             *shiftvec,*fshift,*x,*f;
 852     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 853     real             scratch[4*DIM];
 854     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 855     real *           vdwioffsetptr0;
 856     real *           vdwgridioffsetptr0;
 857     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 858     real *           vdwioffsetptr1;
 859     real *           vdwgridioffsetptr1;
 860     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 861     real *           vdwioffsetptr2;
 862     real *           vdwgridioffsetptr2;
 863     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 864     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 865     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 866     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 867     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 868     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 869     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 870     real             *charge;
 871     int              nvdwtype;
 872     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 873     int              *vdwtype;
 874     real             *vdwparam;
 875     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 876     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 877     __m256           c6grid_00;
 878     __m256           c6grid_10;
 879     __m256           c6grid_20;
 880     real             *vdwgridparam;
 881     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 882     __m256           one_half  = _mm256_set1_ps(0.5);
 883     __m256           minus_one = _mm256_set1_ps(-1.0);
 884     __m256i          ewitab;
 885     __m128i          ewitab_lo,ewitab_hi;
 886     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 887     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 888     real             *ewtab;
 889     __m256           dummy_mask,cutoff_mask;
 890     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 891     __m256           one     = _mm256_set1_ps(1.0);
 892     __m256           two     = _mm256_set1_ps(2.0);
 893     x                = xx[0];
 894     f                = ff[0];
 895
 896     nri              = nlist->nri;
 897     iinr             = nlist->iinr;
 898     jindex           = nlist->jindex;
 899     jjnr             = nlist->jjnr;
 900     shiftidx         = nlist->shift;
 901     gid              = nlist->gid;
 902     shiftvec         = fr->shift_vec[0];
 903     fshift           = fr->fshift[0];
 904     facel            = _mm256_set1_ps(fr->epsfac);
 905     charge           = mdatoms->chargeA;
 906     nvdwtype         = fr->ntype;
 907     vdwparam         = fr->nbfp;
 908     vdwtype          = mdatoms->typeA;
 909     vdwgridparam     = fr->ljpme_c6grid;
 910     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 911     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 912     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 913
 914     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 915     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 916     beta2            = _mm256_mul_ps(beta,beta);
 917     beta3            = _mm256_mul_ps(beta,beta2);
 918
 919     ewtab            = fr->ic->tabq_coul_F;
 920     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 921     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 922
 923     /* Setup water-specific parameters */
 924     inr              = nlist->iinr[0];
 925     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 926     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 927     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 928     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 929     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 930
 931     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 932     rcutoff_scalar   = fr->rcoulomb;
 933     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 934     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 935
 936     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 937     rvdw             = _mm256_set1_ps(fr->rvdw);
 938
 939     /* Avoid stupid compiler warnings */
 940     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 941     j_coord_offsetA = 0;
 942     j_coord_offsetB = 0;
 943     j_coord_offsetC = 0;
 944     j_coord_offsetD = 0;
 945     j_coord_offsetE = 0;
 946     j_coord_offsetF = 0;
 947     j_coord_offsetG = 0;
 948     j_coord_offsetH = 0;
 949
 950     outeriter        = 0;
 951     inneriter        = 0;
 952
 953     for(iidx=0;iidx<4*DIM;iidx++)
 954     {
 955         scratch[iidx] = 0.0;
 956     }
 957
 958     /* Start outer loop over neighborlists */
 959     for(iidx=0; iidx<nri; iidx++)
 960     {
 961         /* Load shift vector for this list */
 962         i_shift_offset   = DIM*shiftidx[iidx];
 963
 964         /* Load limits for loop over neighbors */
 965         j_index_start    = jindex[iidx];
 966         j_index_end      = jindex[iidx+1];
 967
 968         /* Get outer coordinate index */
 969         inr              = iinr[iidx];
 970         i_coord_offset   = DIM*inr;
 971
 972         /* Load i particle coords and add shift vector */
 973         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 974                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 975
 976         fix0             = _mm256_setzero_ps();
 977         fiy0             = _mm256_setzero_ps();
 978         fiz0             = _mm256_setzero_ps();
 979         fix1             = _mm256_setzero_ps();
 980         fiy1             = _mm256_setzero_ps();
 981         fiz1             = _mm256_setzero_ps();
 982         fix2             = _mm256_setzero_ps();
 983         fiy2             = _mm256_setzero_ps();
 984         fiz2             = _mm256_setzero_ps();
 985
 986         /* Start inner kernel loop */
 987         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 988         {
 989
 990             /* Get j neighbor index, and coordinate index */
 991             jnrA             = jjnr[jidx];
 992             jnrB             = jjnr[jidx+1];
 993             jnrC             = jjnr[jidx+2];
 994             jnrD             = jjnr[jidx+3];
 995             jnrE             = jjnr[jidx+4];
 996             jnrF             = jjnr[jidx+5];
 997             jnrG             = jjnr[jidx+6];
 998             jnrH             = jjnr[jidx+7];
 999             j_coord_offsetA  = DIM*jnrA;
1000             j_coord_offsetB  = DIM*jnrB;
1001             j_coord_offsetC  = DIM*jnrC;
1002             j_coord_offsetD  = DIM*jnrD;
1003             j_coord_offsetE  = DIM*jnrE;
1004             j_coord_offsetF  = DIM*jnrF;
1005             j_coord_offsetG  = DIM*jnrG;
1006             j_coord_offsetH  = DIM*jnrH;
1007
1008             /* load j atom coordinates */
1009             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1010                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1011                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1012                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1013                                                  &jx0,&jy0,&jz0);
1014
1015             /* Calculate displacement vector */
1016             dx00             = _mm256_sub_ps(ix0,jx0);
1017             dy00             = _mm256_sub_ps(iy0,jy0);
1018             dz00             = _mm256_sub_ps(iz0,jz0);
1019             dx10             = _mm256_sub_ps(ix1,jx0);
1020             dy10             = _mm256_sub_ps(iy1,jy0);
1021             dz10             = _mm256_sub_ps(iz1,jz0);
1022             dx20             = _mm256_sub_ps(ix2,jx0);
1023             dy20             = _mm256_sub_ps(iy2,jy0);
1024             dz20             = _mm256_sub_ps(iz2,jz0);
1025
1026             /* Calculate squared distance and things based on it */
1027             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1028             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1029             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1030
1031             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1032             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1033             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1034
1035             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1036             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1037             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1038
1039             /* Load parameters for j particles */
1040             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1041                                                                  charge+jnrC+0,charge+jnrD+0,
1042                                                                  charge+jnrE+0,charge+jnrF+0,
1043                                                                  charge+jnrG+0,charge+jnrH+0);
1044             vdwjidx0A        = 2*vdwtype[jnrA+0];
1045             vdwjidx0B        = 2*vdwtype[jnrB+0];
1046             vdwjidx0C        = 2*vdwtype[jnrC+0];
1047             vdwjidx0D        = 2*vdwtype[jnrD+0];
1048             vdwjidx0E        = 2*vdwtype[jnrE+0];
1049             vdwjidx0F        = 2*vdwtype[jnrF+0];
1050             vdwjidx0G        = 2*vdwtype[jnrG+0];
1051             vdwjidx0H        = 2*vdwtype[jnrH+0];
1052
1053             fjx0             = _mm256_setzero_ps();
1054             fjy0             = _mm256_setzero_ps();
1055             fjz0             = _mm256_setzero_ps();
1056
1057             /**************************
1058              * CALCULATE INTERACTIONS *
1059              **************************/
1060
1061             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1062             {
1063
1064             r00              = _mm256_mul_ps(rsq00,rinv00);
1065
1066             /* Compute parameters for interactions between i and j atoms */
1067             qq00             = _mm256_mul_ps(iq0,jq0);
1068             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1069                                             vdwioffsetptr0+vdwjidx0B,
1070                                             vdwioffsetptr0+vdwjidx0C,
1071                                             vdwioffsetptr0+vdwjidx0D,
1072                                             vdwioffsetptr0+vdwjidx0E,
1073                                             vdwioffsetptr0+vdwjidx0F,
1074                                             vdwioffsetptr0+vdwjidx0G,
1075                                             vdwioffsetptr0+vdwjidx0H,
1076                                             &c6_00,&c12_00);
1077
1078             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1079                                                                   vdwgridioffsetptr0+vdwjidx0B,
1080                                                                   vdwgridioffsetptr0+vdwjidx0C,
1081                                                                   vdwgridioffsetptr0+vdwjidx0D,
1082                                                                   vdwgridioffsetptr0+vdwjidx0E,
1083                                                                   vdwgridioffsetptr0+vdwjidx0F,
1084                                                                   vdwgridioffsetptr0+vdwjidx0G,
1085                                                                   vdwgridioffsetptr0+vdwjidx0H);
1086
1087             /* EWALD ELECTROSTATICS */
1088
1089             /* Analytical PME correction */
1090             zeta2            = _mm256_mul_ps(beta2,rsq00);
1091             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1092             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1093             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1094             felec            = _mm256_mul_ps(qq00,felec);
1095
1096             /* Analytical LJ-PME */
1097             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1098             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1099             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1100             exponent         = gmx_simd_exp_r(ewcljrsq);
1101             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1102             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1103             /* f6A = 6 * C6grid * (1 - poly) */
1104             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1105             /* f6B = C6grid * exponent * beta^6 */
1106             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1107             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1108             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);
1109
1110             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1111
1112             fscal            = _mm256_add_ps(felec,fvdw);
1113
1114             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1115
1116             /* Calculate temporary vectorial force */
1117             tx               = _mm256_mul_ps(fscal,dx00);
1118             ty               = _mm256_mul_ps(fscal,dy00);
1119             tz               = _mm256_mul_ps(fscal,dz00);
1120
1121             /* Update vectorial force */
1122             fix0             = _mm256_add_ps(fix0,tx);
1123             fiy0             = _mm256_add_ps(fiy0,ty);
1124             fiz0             = _mm256_add_ps(fiz0,tz);
1125
1126             fjx0             = _mm256_add_ps(fjx0,tx);
1127             fjy0             = _mm256_add_ps(fjy0,ty);
1128             fjz0             = _mm256_add_ps(fjz0,tz);
1129
1130             }
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1137             {
1138
1139             r10              = _mm256_mul_ps(rsq10,rinv10);
1140
1141             /* Compute parameters for interactions between i and j atoms */
1142             qq10             = _mm256_mul_ps(iq1,jq0);
1143
1144             /* EWALD ELECTROSTATICS */
1145
1146             /* Analytical PME correction */
1147             zeta2            = _mm256_mul_ps(beta2,rsq10);
1148             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1149             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1150             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1151             felec            = _mm256_mul_ps(qq10,felec);
1152
1153             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1154
1155             fscal            = felec;
1156
1157             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1158
1159             /* Calculate temporary vectorial force */
1160             tx               = _mm256_mul_ps(fscal,dx10);
1161             ty               = _mm256_mul_ps(fscal,dy10);
1162             tz               = _mm256_mul_ps(fscal,dz10);
1163
1164             /* Update vectorial force */
1165             fix1             = _mm256_add_ps(fix1,tx);
1166             fiy1             = _mm256_add_ps(fiy1,ty);
1167             fiz1             = _mm256_add_ps(fiz1,tz);
1168
1169             fjx0             = _mm256_add_ps(fjx0,tx);
1170             fjy0             = _mm256_add_ps(fjy0,ty);
1171             fjz0             = _mm256_add_ps(fjz0,tz);
1172
1173             }
1174
1175             /**************************
1176              * CALCULATE INTERACTIONS *
1177              **************************/
1178
1179             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1180             {
1181
1182             r20              = _mm256_mul_ps(rsq20,rinv20);
1183
1184             /* Compute parameters for interactions between i and j atoms */
1185             qq20             = _mm256_mul_ps(iq2,jq0);
1186
1187             /* EWALD ELECTROSTATICS */
1188
1189             /* Analytical PME correction */
1190             zeta2            = _mm256_mul_ps(beta2,rsq20);
1191             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1192             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1193             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1194             felec            = _mm256_mul_ps(qq20,felec);
1195
1196             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1201
1202             /* Calculate temporary vectorial force */
1203             tx               = _mm256_mul_ps(fscal,dx20);
1204             ty               = _mm256_mul_ps(fscal,dy20);
1205             tz               = _mm256_mul_ps(fscal,dz20);
1206
1207             /* Update vectorial force */
1208             fix2             = _mm256_add_ps(fix2,tx);
1209             fiy2             = _mm256_add_ps(fiy2,ty);
1210             fiz2             = _mm256_add_ps(fiz2,tz);
1211
1212             fjx0             = _mm256_add_ps(fjx0,tx);
1213             fjy0             = _mm256_add_ps(fjy0,ty);
1214             fjz0             = _mm256_add_ps(fjz0,tz);
1215
1216             }
1217
1218             fjptrA             = f+j_coord_offsetA;
1219             fjptrB             = f+j_coord_offsetB;
1220             fjptrC             = f+j_coord_offsetC;
1221             fjptrD             = f+j_coord_offsetD;
1222             fjptrE             = f+j_coord_offsetE;
1223             fjptrF             = f+j_coord_offsetF;
1224             fjptrG             = f+j_coord_offsetG;
1225             fjptrH             = f+j_coord_offsetH;
1226
1227             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1228
1229             /* Inner loop uses 203 flops */
1230         }
1231
1232         if(jidx<j_index_end)
1233         {
1234
1235             /* Get j neighbor index, and coordinate index */
1236             jnrlistA         = jjnr[jidx];
1237             jnrlistB         = jjnr[jidx+1];
1238             jnrlistC         = jjnr[jidx+2];
1239             jnrlistD         = jjnr[jidx+3];
1240             jnrlistE         = jjnr[jidx+4];
1241             jnrlistF         = jjnr[jidx+5];
1242             jnrlistG         = jjnr[jidx+6];
1243             jnrlistH         = jjnr[jidx+7];
1244             /* Sign of each element will be negative for non-real atoms.
1245              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1246              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1247              */
1248             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1249                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1250
1251             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1252             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1253             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1254             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1255             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1256             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1257             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1258             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1259             j_coord_offsetA  = DIM*jnrA;
1260             j_coord_offsetB  = DIM*jnrB;
1261             j_coord_offsetC  = DIM*jnrC;
1262             j_coord_offsetD  = DIM*jnrD;
1263             j_coord_offsetE  = DIM*jnrE;
1264             j_coord_offsetF  = DIM*jnrF;
1265             j_coord_offsetG  = DIM*jnrG;
1266             j_coord_offsetH  = DIM*jnrH;
1267
1268             /* load j atom coordinates */
1269             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1270                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1271                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1272                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1273                                                  &jx0,&jy0,&jz0);
1274
1275             /* Calculate displacement vector */
1276             dx00             = _mm256_sub_ps(ix0,jx0);
1277             dy00             = _mm256_sub_ps(iy0,jy0);
1278             dz00             = _mm256_sub_ps(iz0,jz0);
1279             dx10             = _mm256_sub_ps(ix1,jx0);
1280             dy10             = _mm256_sub_ps(iy1,jy0);
1281             dz10             = _mm256_sub_ps(iz1,jz0);
1282             dx20             = _mm256_sub_ps(ix2,jx0);
1283             dy20             = _mm256_sub_ps(iy2,jy0);
1284             dz20             = _mm256_sub_ps(iz2,jz0);
1285
1286             /* Calculate squared distance and things based on it */
1287             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1288             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1289             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1290
1291             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1292             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1293             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1294
1295             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1296             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1297             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1298
1299             /* Load parameters for j particles */
1300             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1301                                                                  charge+jnrC+0,charge+jnrD+0,
1302                                                                  charge+jnrE+0,charge+jnrF+0,
1303                                                                  charge+jnrG+0,charge+jnrH+0);
1304             vdwjidx0A        = 2*vdwtype[jnrA+0];
1305             vdwjidx0B        = 2*vdwtype[jnrB+0];
1306             vdwjidx0C        = 2*vdwtype[jnrC+0];
1307             vdwjidx0D        = 2*vdwtype[jnrD+0];
1308             vdwjidx0E        = 2*vdwtype[jnrE+0];
1309             vdwjidx0F        = 2*vdwtype[jnrF+0];
1310             vdwjidx0G        = 2*vdwtype[jnrG+0];
1311             vdwjidx0H        = 2*vdwtype[jnrH+0];
1312
1313             fjx0             = _mm256_setzero_ps();
1314             fjy0             = _mm256_setzero_ps();
1315             fjz0             = _mm256_setzero_ps();
1316
1317             /**************************
1318              * CALCULATE INTERACTIONS *
1319              **************************/
1320
1321             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1322             {
1323
1324             r00              = _mm256_mul_ps(rsq00,rinv00);
1325             r00              = _mm256_andnot_ps(dummy_mask,r00);
1326
1327             /* Compute parameters for interactions between i and j atoms */
1328             qq00             = _mm256_mul_ps(iq0,jq0);
1329             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1330                                             vdwioffsetptr0+vdwjidx0B,
1331                                             vdwioffsetptr0+vdwjidx0C,
1332                                             vdwioffsetptr0+vdwjidx0D,
1333                                             vdwioffsetptr0+vdwjidx0E,
1334                                             vdwioffsetptr0+vdwjidx0F,
1335                                             vdwioffsetptr0+vdwjidx0G,
1336                                             vdwioffsetptr0+vdwjidx0H,
1337                                             &c6_00,&c12_00);
1338
1339             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1340                                                                   vdwgridioffsetptr0+vdwjidx0B,
1341                                                                   vdwgridioffsetptr0+vdwjidx0C,
1342                                                                   vdwgridioffsetptr0+vdwjidx0D,
1343                                                                   vdwgridioffsetptr0+vdwjidx0E,
1344                                                                   vdwgridioffsetptr0+vdwjidx0F,
1345                                                                   vdwgridioffsetptr0+vdwjidx0G,
1346                                                                   vdwgridioffsetptr0+vdwjidx0H);
1347
1348             /* EWALD ELECTROSTATICS */
1349
1350             /* Analytical PME correction */
1351             zeta2            = _mm256_mul_ps(beta2,rsq00);
1352             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1353             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1354             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1355             felec            = _mm256_mul_ps(qq00,felec);
1356
1357             /* Analytical LJ-PME */
1358             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1359             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1360             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1361             exponent         = gmx_simd_exp_r(ewcljrsq);
1362             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1363             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1364             /* f6A = 6 * C6grid * (1 - poly) */
1365             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1366             /* f6B = C6grid * exponent * beta^6 */
1367             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1368             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1369             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);
1370
1371             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1372
1373             fscal            = _mm256_add_ps(felec,fvdw);
1374
1375             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1376
1377             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1378
1379             /* Calculate temporary vectorial force */
1380             tx               = _mm256_mul_ps(fscal,dx00);
1381             ty               = _mm256_mul_ps(fscal,dy00);
1382             tz               = _mm256_mul_ps(fscal,dz00);
1383
1384             /* Update vectorial force */
1385             fix0             = _mm256_add_ps(fix0,tx);
1386             fiy0             = _mm256_add_ps(fiy0,ty);
1387             fiz0             = _mm256_add_ps(fiz0,tz);
1388
1389             fjx0             = _mm256_add_ps(fjx0,tx);
1390             fjy0             = _mm256_add_ps(fjy0,ty);
1391             fjz0             = _mm256_add_ps(fjz0,tz);
1392
1393             }
1394
1395             /**************************
1396              * CALCULATE INTERACTIONS *
1397              **************************/
1398
1399             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1400             {
1401
1402             r10              = _mm256_mul_ps(rsq10,rinv10);
1403             r10              = _mm256_andnot_ps(dummy_mask,r10);
1404
1405             /* Compute parameters for interactions between i and j atoms */
1406             qq10             = _mm256_mul_ps(iq1,jq0);
1407
1408             /* EWALD ELECTROSTATICS */
1409
1410             /* Analytical PME correction */
1411             zeta2            = _mm256_mul_ps(beta2,rsq10);
1412             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1413             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1414             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1415             felec            = _mm256_mul_ps(qq10,felec);
1416
1417             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1418
1419             fscal            = felec;
1420
1421             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1422
1423             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1424
1425             /* Calculate temporary vectorial force */
1426             tx               = _mm256_mul_ps(fscal,dx10);
1427             ty               = _mm256_mul_ps(fscal,dy10);
1428             tz               = _mm256_mul_ps(fscal,dz10);
1429
1430             /* Update vectorial force */
1431             fix1             = _mm256_add_ps(fix1,tx);
1432             fiy1             = _mm256_add_ps(fiy1,ty);
1433             fiz1             = _mm256_add_ps(fiz1,tz);
1434
1435             fjx0             = _mm256_add_ps(fjx0,tx);
1436             fjy0             = _mm256_add_ps(fjy0,ty);
1437             fjz0             = _mm256_add_ps(fjz0,tz);
1438
1439             }
1440
1441             /**************************
1442              * CALCULATE INTERACTIONS *
1443              **************************/
1444
1445             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1446             {
1447
1448             r20              = _mm256_mul_ps(rsq20,rinv20);
1449             r20              = _mm256_andnot_ps(dummy_mask,r20);
1450
1451             /* Compute parameters for interactions between i and j atoms */
1452             qq20             = _mm256_mul_ps(iq2,jq0);
1453
1454             /* EWALD ELECTROSTATICS */
1455
1456             /* Analytical PME correction */
1457             zeta2            = _mm256_mul_ps(beta2,rsq20);
1458             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1459             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1460             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1461             felec            = _mm256_mul_ps(qq20,felec);
1462
1463             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1464
1465             fscal            = felec;
1466
1467             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1468
1469             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1470
1471             /* Calculate temporary vectorial force */
1472             tx               = _mm256_mul_ps(fscal,dx20);
1473             ty               = _mm256_mul_ps(fscal,dy20);
1474             tz               = _mm256_mul_ps(fscal,dz20);
1475
1476             /* Update vectorial force */
1477             fix2             = _mm256_add_ps(fix2,tx);
1478             fiy2             = _mm256_add_ps(fiy2,ty);
1479             fiz2             = _mm256_add_ps(fiz2,tz);
1480
1481             fjx0             = _mm256_add_ps(fjx0,tx);
1482             fjy0             = _mm256_add_ps(fjy0,ty);
1483             fjz0             = _mm256_add_ps(fjz0,tz);
1484
1485             }
1486
1487             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1488             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1489             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1490             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1491             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1492             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1493             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1494             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1495
1496             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1497
1498             /* Inner loop uses 206 flops */
1499         }
1500
1501         /* End of innermost loop */
1502
1503         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1504                                                  f+i_coord_offset,fshift+i_shift_offset);
1505
1506         /* Increment number of inner iterations */
1507         inneriter                  += j_index_end - j_index_start;
1508
1509         /* Outer loop uses 18 flops */
1510     }
1511
1512     /* Increment number of outer iterations */
1513     outeriter        += nri;
1514
1515     /* Update outer/inner flops */
1516
1517     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*206);
1518 }