src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEw_VdwLJEw_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_ElecEw_VdwLJEw_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_ElecEw_VdwLJEw_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     /* Avoid stupid compiler warnings */
 163     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 164     j_coord_offsetA = 0;
 165     j_coord_offsetB = 0;
 166     j_coord_offsetC = 0;
 167     j_coord_offsetD = 0;
 168     j_coord_offsetE = 0;
 169     j_coord_offsetF = 0;
 170     j_coord_offsetG = 0;
 171     j_coord_offsetH = 0;
 172
 173     outeriter        = 0;
 174     inneriter        = 0;
 175
 176     for(iidx=0;iidx<4*DIM;iidx++)
 177     {
 178         scratch[iidx] = 0.0;
 179     }
 180
 181     /* Start outer loop over neighborlists */
 182     for(iidx=0; iidx<nri; iidx++)
 183     {
 184         /* Load shift vector for this list */
 185         i_shift_offset   = DIM*shiftidx[iidx];
 186
 187         /* Load limits for loop over neighbors */
 188         j_index_start    = jindex[iidx];
 189         j_index_end      = jindex[iidx+1];
 190
 191         /* Get outer coordinate index */
 192         inr              = iinr[iidx];
 193         i_coord_offset   = DIM*inr;
 194
 195         /* Load i particle coords and add shift vector */
 196         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 197                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 198
 199         fix0             = _mm256_setzero_ps();
 200         fiy0             = _mm256_setzero_ps();
 201         fiz0             = _mm256_setzero_ps();
 202         fix1             = _mm256_setzero_ps();
 203         fiy1             = _mm256_setzero_ps();
 204         fiz1             = _mm256_setzero_ps();
 205         fix2             = _mm256_setzero_ps();
 206         fiy2             = _mm256_setzero_ps();
 207         fiz2             = _mm256_setzero_ps();
 208
 209         /* Reset potential sums */
 210         velecsum         = _mm256_setzero_ps();
 211         vvdwsum          = _mm256_setzero_ps();
 212
 213         /* Start inner kernel loop */
 214         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 215         {
 216
 217             /* Get j neighbor index, and coordinate index */
 218             jnrA             = jjnr[jidx];
 219             jnrB             = jjnr[jidx+1];
 220             jnrC             = jjnr[jidx+2];
 221             jnrD             = jjnr[jidx+3];
 222             jnrE             = jjnr[jidx+4];
 223             jnrF             = jjnr[jidx+5];
 224             jnrG             = jjnr[jidx+6];
 225             jnrH             = jjnr[jidx+7];
 226             j_coord_offsetA  = DIM*jnrA;
 227             j_coord_offsetB  = DIM*jnrB;
 228             j_coord_offsetC  = DIM*jnrC;
 229             j_coord_offsetD  = DIM*jnrD;
 230             j_coord_offsetE  = DIM*jnrE;
 231             j_coord_offsetF  = DIM*jnrF;
 232             j_coord_offsetG  = DIM*jnrG;
 233             j_coord_offsetH  = DIM*jnrH;
 234
 235             /* load j atom coordinates */
 236             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 237                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 238                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 239                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 240                                                  &jx0,&jy0,&jz0);
 241
 242             /* Calculate displacement vector */
 243             dx00             = _mm256_sub_ps(ix0,jx0);
 244             dy00             = _mm256_sub_ps(iy0,jy0);
 245             dz00             = _mm256_sub_ps(iz0,jz0);
 246             dx10             = _mm256_sub_ps(ix1,jx0);
 247             dy10             = _mm256_sub_ps(iy1,jy0);
 248             dz10             = _mm256_sub_ps(iz1,jz0);
 249             dx20             = _mm256_sub_ps(ix2,jx0);
 250             dy20             = _mm256_sub_ps(iy2,jy0);
 251             dz20             = _mm256_sub_ps(iz2,jz0);
 252
 253             /* Calculate squared distance and things based on it */
 254             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 255             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 256             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 257
 258             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 259             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 260             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 261
 262             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 263             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 264             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 265
 266             /* Load parameters for j particles */
 267             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 268                                                                  charge+jnrC+0,charge+jnrD+0,
 269                                                                  charge+jnrE+0,charge+jnrF+0,
 270                                                                  charge+jnrG+0,charge+jnrH+0);
 271             vdwjidx0A        = 2*vdwtype[jnrA+0];
 272             vdwjidx0B        = 2*vdwtype[jnrB+0];
 273             vdwjidx0C        = 2*vdwtype[jnrC+0];
 274             vdwjidx0D        = 2*vdwtype[jnrD+0];
 275             vdwjidx0E        = 2*vdwtype[jnrE+0];
 276             vdwjidx0F        = 2*vdwtype[jnrF+0];
 277             vdwjidx0G        = 2*vdwtype[jnrG+0];
 278             vdwjidx0H        = 2*vdwtype[jnrH+0];
 279
 280             fjx0             = _mm256_setzero_ps();
 281             fjy0             = _mm256_setzero_ps();
 282             fjz0             = _mm256_setzero_ps();
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             r00              = _mm256_mul_ps(rsq00,rinv00);
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq00             = _mm256_mul_ps(iq0,jq0);
 292             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 293                                             vdwioffsetptr0+vdwjidx0B,
 294                                             vdwioffsetptr0+vdwjidx0C,
 295                                             vdwioffsetptr0+vdwjidx0D,
 296                                             vdwioffsetptr0+vdwjidx0E,
 297                                             vdwioffsetptr0+vdwjidx0F,
 298                                             vdwioffsetptr0+vdwjidx0G,
 299                                             vdwioffsetptr0+vdwjidx0H,
 300                                             &c6_00,&c12_00);
 301
 302             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 303                                                                   vdwgridioffsetptr0+vdwjidx0B,
 304                                                                   vdwgridioffsetptr0+vdwjidx0C,
 305                                                                   vdwgridioffsetptr0+vdwjidx0D,
 306                                                                   vdwgridioffsetptr0+vdwjidx0E,
 307                                                                   vdwgridioffsetptr0+vdwjidx0F,
 308                                                                   vdwgridioffsetptr0+vdwjidx0G,
 309                                                                   vdwgridioffsetptr0+vdwjidx0H);
 310
 311             /* EWALD ELECTROSTATICS */
 312
 313             /* Analytical PME correction */
 314             zeta2            = _mm256_mul_ps(beta2,rsq00);
 315             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 316             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 317             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 318             felec            = _mm256_mul_ps(qq00,felec);
 319             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 320             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 321             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 322             velec            = _mm256_mul_ps(qq00,velec);
 323
 324             /* Analytical LJ-PME */
 325             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 326             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 327             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 328             exponent         = gmx_simd_exp_r(ewcljrsq);
 329             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 330             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 331             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 332             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 333             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 334             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 335             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 336             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);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velecsum         = _mm256_add_ps(velecsum,velec);
 340             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 341
 342             fscal            = _mm256_add_ps(felec,fvdw);
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm256_mul_ps(fscal,dx00);
 346             ty               = _mm256_mul_ps(fscal,dy00);
 347             tz               = _mm256_mul_ps(fscal,dz00);
 348
 349             /* Update vectorial force */
 350             fix0             = _mm256_add_ps(fix0,tx);
 351             fiy0             = _mm256_add_ps(fiy0,ty);
 352             fiz0             = _mm256_add_ps(fiz0,tz);
 353
 354             fjx0             = _mm256_add_ps(fjx0,tx);
 355             fjy0             = _mm256_add_ps(fjy0,ty);
 356             fjz0             = _mm256_add_ps(fjz0,tz);
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             r10              = _mm256_mul_ps(rsq10,rinv10);
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq10             = _mm256_mul_ps(iq1,jq0);
 366
 367             /* EWALD ELECTROSTATICS */
 368
 369             /* Analytical PME correction */
 370             zeta2            = _mm256_mul_ps(beta2,rsq10);
 371             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 372             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 373             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 374             felec            = _mm256_mul_ps(qq10,felec);
 375             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 376             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 377             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 378             velec            = _mm256_mul_ps(qq10,velec);
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velecsum         = _mm256_add_ps(velecsum,velec);
 382
 383             fscal            = felec;
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm256_mul_ps(fscal,dx10);
 387             ty               = _mm256_mul_ps(fscal,dy10);
 388             tz               = _mm256_mul_ps(fscal,dz10);
 389
 390             /* Update vectorial force */
 391             fix1             = _mm256_add_ps(fix1,tx);
 392             fiy1             = _mm256_add_ps(fiy1,ty);
 393             fiz1             = _mm256_add_ps(fiz1,tz);
 394
 395             fjx0             = _mm256_add_ps(fjx0,tx);
 396             fjy0             = _mm256_add_ps(fjy0,ty);
 397             fjz0             = _mm256_add_ps(fjz0,tz);
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             r20              = _mm256_mul_ps(rsq20,rinv20);
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq20             = _mm256_mul_ps(iq2,jq0);
 407
 408             /* EWALD ELECTROSTATICS */
 409
 410             /* Analytical PME correction */
 411             zeta2            = _mm256_mul_ps(beta2,rsq20);
 412             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 413             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 414             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 415             felec            = _mm256_mul_ps(qq20,felec);
 416             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 417             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 418             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 419             velec            = _mm256_mul_ps(qq20,velec);
 420
 421             /* Update potential sum for this i atom from the interaction with this j atom. */
 422             velecsum         = _mm256_add_ps(velecsum,velec);
 423
 424             fscal            = felec;
 425
 426             /* Calculate temporary vectorial force */
 427             tx               = _mm256_mul_ps(fscal,dx20);
 428             ty               = _mm256_mul_ps(fscal,dy20);
 429             tz               = _mm256_mul_ps(fscal,dz20);
 430
 431             /* Update vectorial force */
 432             fix2             = _mm256_add_ps(fix2,tx);
 433             fiy2             = _mm256_add_ps(fiy2,ty);
 434             fiz2             = _mm256_add_ps(fiz2,tz);
 435
 436             fjx0             = _mm256_add_ps(fjx0,tx);
 437             fjy0             = _mm256_add_ps(fjy0,ty);
 438             fjz0             = _mm256_add_ps(fjz0,tz);
 439
 440             fjptrA             = f+j_coord_offsetA;
 441             fjptrB             = f+j_coord_offsetB;
 442             fjptrC             = f+j_coord_offsetC;
 443             fjptrD             = f+j_coord_offsetD;
 444             fjptrE             = f+j_coord_offsetE;
 445             fjptrF             = f+j_coord_offsetF;
 446             fjptrG             = f+j_coord_offsetG;
 447             fjptrH             = f+j_coord_offsetH;
 448
 449             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 450
 451             /* Inner loop uses 283 flops */
 452         }
 453
 454         if(jidx<j_index_end)
 455         {
 456
 457             /* Get j neighbor index, and coordinate index */
 458             jnrlistA         = jjnr[jidx];
 459             jnrlistB         = jjnr[jidx+1];
 460             jnrlistC         = jjnr[jidx+2];
 461             jnrlistD         = jjnr[jidx+3];
 462             jnrlistE         = jjnr[jidx+4];
 463             jnrlistF         = jjnr[jidx+5];
 464             jnrlistG         = jjnr[jidx+6];
 465             jnrlistH         = jjnr[jidx+7];
 466             /* Sign of each element will be negative for non-real atoms.
 467              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 468              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 469              */
 470             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 471                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 472
 473             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 474             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 475             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 476             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 477             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 478             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 479             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 480             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 481             j_coord_offsetA  = DIM*jnrA;
 482             j_coord_offsetB  = DIM*jnrB;
 483             j_coord_offsetC  = DIM*jnrC;
 484             j_coord_offsetD  = DIM*jnrD;
 485             j_coord_offsetE  = DIM*jnrE;
 486             j_coord_offsetF  = DIM*jnrF;
 487             j_coord_offsetG  = DIM*jnrG;
 488             j_coord_offsetH  = DIM*jnrH;
 489
 490             /* load j atom coordinates */
 491             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 492                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 493                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 494                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 495                                                  &jx0,&jy0,&jz0);
 496
 497             /* Calculate displacement vector */
 498             dx00             = _mm256_sub_ps(ix0,jx0);
 499             dy00             = _mm256_sub_ps(iy0,jy0);
 500             dz00             = _mm256_sub_ps(iz0,jz0);
 501             dx10             = _mm256_sub_ps(ix1,jx0);
 502             dy10             = _mm256_sub_ps(iy1,jy0);
 503             dz10             = _mm256_sub_ps(iz1,jz0);
 504             dx20             = _mm256_sub_ps(ix2,jx0);
 505             dy20             = _mm256_sub_ps(iy2,jy0);
 506             dz20             = _mm256_sub_ps(iz2,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 510             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 511             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 512
 513             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 514             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 515             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 516
 517             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 518             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 519             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 520
 521             /* Load parameters for j particles */
 522             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 523                                                                  charge+jnrC+0,charge+jnrD+0,
 524                                                                  charge+jnrE+0,charge+jnrF+0,
 525                                                                  charge+jnrG+0,charge+jnrH+0);
 526             vdwjidx0A        = 2*vdwtype[jnrA+0];
 527             vdwjidx0B        = 2*vdwtype[jnrB+0];
 528             vdwjidx0C        = 2*vdwtype[jnrC+0];
 529             vdwjidx0D        = 2*vdwtype[jnrD+0];
 530             vdwjidx0E        = 2*vdwtype[jnrE+0];
 531             vdwjidx0F        = 2*vdwtype[jnrF+0];
 532             vdwjidx0G        = 2*vdwtype[jnrG+0];
 533             vdwjidx0H        = 2*vdwtype[jnrH+0];
 534
 535             fjx0             = _mm256_setzero_ps();
 536             fjy0             = _mm256_setzero_ps();
 537             fjz0             = _mm256_setzero_ps();
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             r00              = _mm256_mul_ps(rsq00,rinv00);
 544             r00              = _mm256_andnot_ps(dummy_mask,r00);
 545
 546             /* Compute parameters for interactions between i and j atoms */
 547             qq00             = _mm256_mul_ps(iq0,jq0);
 548             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 549                                             vdwioffsetptr0+vdwjidx0B,
 550                                             vdwioffsetptr0+vdwjidx0C,
 551                                             vdwioffsetptr0+vdwjidx0D,
 552                                             vdwioffsetptr0+vdwjidx0E,
 553                                             vdwioffsetptr0+vdwjidx0F,
 554                                             vdwioffsetptr0+vdwjidx0G,
 555                                             vdwioffsetptr0+vdwjidx0H,
 556                                             &c6_00,&c12_00);
 557
 558             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 559                                                                   vdwgridioffsetptr0+vdwjidx0B,
 560                                                                   vdwgridioffsetptr0+vdwjidx0C,
 561                                                                   vdwgridioffsetptr0+vdwjidx0D,
 562                                                                   vdwgridioffsetptr0+vdwjidx0E,
 563                                                                   vdwgridioffsetptr0+vdwjidx0F,
 564                                                                   vdwgridioffsetptr0+vdwjidx0G,
 565                                                                   vdwgridioffsetptr0+vdwjidx0H);
 566
 567             /* EWALD ELECTROSTATICS */
 568
 569             /* Analytical PME correction */
 570             zeta2            = _mm256_mul_ps(beta2,rsq00);
 571             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 572             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 573             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 574             felec            = _mm256_mul_ps(qq00,felec);
 575             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 576             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 577             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 578             velec            = _mm256_mul_ps(qq00,velec);
 579
 580             /* Analytical LJ-PME */
 581             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 582             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
 583             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 584             exponent         = gmx_simd_exp_r(ewcljrsq);
 585             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 586             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 587             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 588             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_00,_mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly))),rinvsix);
 589             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 590             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 591             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 592             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);
 593
 594             /* Update potential sum for this i atom from the interaction with this j atom. */
 595             velec            = _mm256_andnot_ps(dummy_mask,velec);
 596             velecsum         = _mm256_add_ps(velecsum,velec);
 597             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 598             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 599
 600             fscal            = _mm256_add_ps(felec,fvdw);
 601
 602             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 603
 604             /* Calculate temporary vectorial force */
 605             tx               = _mm256_mul_ps(fscal,dx00);
 606             ty               = _mm256_mul_ps(fscal,dy00);
 607             tz               = _mm256_mul_ps(fscal,dz00);
 608
 609             /* Update vectorial force */
 610             fix0             = _mm256_add_ps(fix0,tx);
 611             fiy0             = _mm256_add_ps(fiy0,ty);
 612             fiz0             = _mm256_add_ps(fiz0,tz);
 613
 614             fjx0             = _mm256_add_ps(fjx0,tx);
 615             fjy0             = _mm256_add_ps(fjy0,ty);
 616             fjz0             = _mm256_add_ps(fjz0,tz);
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             r10              = _mm256_mul_ps(rsq10,rinv10);
 623             r10              = _mm256_andnot_ps(dummy_mask,r10);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq10             = _mm256_mul_ps(iq1,jq0);
 627
 628             /* EWALD ELECTROSTATICS */
 629
 630             /* Analytical PME correction */
 631             zeta2            = _mm256_mul_ps(beta2,rsq10);
 632             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 633             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 634             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 635             felec            = _mm256_mul_ps(qq10,felec);
 636             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 637             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 638             velec            = _mm256_sub_ps(rinv10,pmecorrV);
 639             velec            = _mm256_mul_ps(qq10,velec);
 640
 641             /* Update potential sum for this i atom from the interaction with this j atom. */
 642             velec            = _mm256_andnot_ps(dummy_mask,velec);
 643             velecsum         = _mm256_add_ps(velecsum,velec);
 644
 645             fscal            = felec;
 646
 647             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 648
 649             /* Calculate temporary vectorial force */
 650             tx               = _mm256_mul_ps(fscal,dx10);
 651             ty               = _mm256_mul_ps(fscal,dy10);
 652             tz               = _mm256_mul_ps(fscal,dz10);
 653
 654             /* Update vectorial force */
 655             fix1             = _mm256_add_ps(fix1,tx);
 656             fiy1             = _mm256_add_ps(fiy1,ty);
 657             fiz1             = _mm256_add_ps(fiz1,tz);
 658
 659             fjx0             = _mm256_add_ps(fjx0,tx);
 660             fjy0             = _mm256_add_ps(fjy0,ty);
 661             fjz0             = _mm256_add_ps(fjz0,tz);
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             r20              = _mm256_mul_ps(rsq20,rinv20);
 668             r20              = _mm256_andnot_ps(dummy_mask,r20);
 669
 670             /* Compute parameters for interactions between i and j atoms */
 671             qq20             = _mm256_mul_ps(iq2,jq0);
 672
 673             /* EWALD ELECTROSTATICS */
 674
 675             /* Analytical PME correction */
 676             zeta2            = _mm256_mul_ps(beta2,rsq20);
 677             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 678             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 679             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 680             felec            = _mm256_mul_ps(qq20,felec);
 681             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 682             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 683             velec            = _mm256_sub_ps(rinv20,pmecorrV);
 684             velec            = _mm256_mul_ps(qq20,velec);
 685
 686             /* Update potential sum for this i atom from the interaction with this j atom. */
 687             velec            = _mm256_andnot_ps(dummy_mask,velec);
 688             velecsum         = _mm256_add_ps(velecsum,velec);
 689
 690             fscal            = felec;
 691
 692             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 693
 694             /* Calculate temporary vectorial force */
 695             tx               = _mm256_mul_ps(fscal,dx20);
 696             ty               = _mm256_mul_ps(fscal,dy20);
 697             tz               = _mm256_mul_ps(fscal,dz20);
 698
 699             /* Update vectorial force */
 700             fix2             = _mm256_add_ps(fix2,tx);
 701             fiy2             = _mm256_add_ps(fiy2,ty);
 702             fiz2             = _mm256_add_ps(fiz2,tz);
 703
 704             fjx0             = _mm256_add_ps(fjx0,tx);
 705             fjy0             = _mm256_add_ps(fjy0,ty);
 706             fjz0             = _mm256_add_ps(fjz0,tz);
 707
 708             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 709             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 710             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 711             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 712             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 713             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 714             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 715             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 716
 717             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 718
 719             /* Inner loop uses 286 flops */
 720         }
 721
 722         /* End of innermost loop */
 723
 724         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 725                                                  f+i_coord_offset,fshift+i_shift_offset);
 726
 727         ggid                        = gid[iidx];
 728         /* Update potential energies */
 729         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 730         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 731
 732         /* Increment number of inner iterations */
 733         inneriter                  += j_index_end - j_index_start;
 734
 735         /* Outer loop uses 20 flops */
 736     }
 737
 738     /* Increment number of outer iterations */
 739     outeriter        += nri;
 740
 741     /* Update outer/inner flops */
 742
 743     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*286);
 744 }
 745 /*
 746  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_256_single
 747  * Electrostatics interaction: Ewald
 748  * VdW interaction:            LJEwald
 749  * Geometry:                   Water3-Particle
 750  * Calculate force/pot:        Force
 751  */
 752 void
 753 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_256_single
 754                     (t_nblist                    * gmx_restrict       nlist,
 755                      rvec                        * gmx_restrict          xx,
 756                      rvec                        * gmx_restrict          ff,
 757                      t_forcerec                  * gmx_restrict          fr,
 758                      t_mdatoms                   * gmx_restrict     mdatoms,
 759                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 760                      t_nrnb                      * gmx_restrict        nrnb)
 761 {
 762     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 763      * just 0 for non-waters.
 764      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 765      * jnr indices corresponding to data put in the four positions in the SIMD register.
 766      */
 767     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 768     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 769     int              jnrA,jnrB,jnrC,jnrD;
 770     int              jnrE,jnrF,jnrG,jnrH;
 771     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 772     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 773     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 774     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 775     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 776     real             rcutoff_scalar;
 777     real             *shiftvec,*fshift,*x,*f;
 778     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 779     real             scratch[4*DIM];
 780     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 781     real *           vdwioffsetptr0;
 782     real *           vdwgridioffsetptr0;
 783     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 784     real *           vdwioffsetptr1;
 785     real *           vdwgridioffsetptr1;
 786     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 787     real *           vdwioffsetptr2;
 788     real *           vdwgridioffsetptr2;
 789     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 790     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 791     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 792     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 793     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 794     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 795     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 796     real             *charge;
 797     int              nvdwtype;
 798     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 799     int              *vdwtype;
 800     real             *vdwparam;
 801     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 802     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 803     __m256           c6grid_00;
 804     __m256           c6grid_10;
 805     __m256           c6grid_20;
 806     real             *vdwgridparam;
 807     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 808     __m256           one_half  = _mm256_set1_ps(0.5);
 809     __m256           minus_one = _mm256_set1_ps(-1.0);
 810     __m256i          ewitab;
 811     __m128i          ewitab_lo,ewitab_hi;
 812     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 813     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 814     real             *ewtab;
 815     __m256           dummy_mask,cutoff_mask;
 816     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 817     __m256           one     = _mm256_set1_ps(1.0);
 818     __m256           two     = _mm256_set1_ps(2.0);
 819     x                = xx[0];
 820     f                = ff[0];
 821
 822     nri              = nlist->nri;
 823     iinr             = nlist->iinr;
 824     jindex           = nlist->jindex;
 825     jjnr             = nlist->jjnr;
 826     shiftidx         = nlist->shift;
 827     gid              = nlist->gid;
 828     shiftvec         = fr->shift_vec[0];
 829     fshift           = fr->fshift[0];
 830     facel            = _mm256_set1_ps(fr->epsfac);
 831     charge           = mdatoms->chargeA;
 832     nvdwtype         = fr->ntype;
 833     vdwparam         = fr->nbfp;
 834     vdwtype          = mdatoms->typeA;
 835     vdwgridparam     = fr->ljpme_c6grid;
 836     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 837     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 838     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 839
 840     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 841     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 842     beta2            = _mm256_mul_ps(beta,beta);
 843     beta3            = _mm256_mul_ps(beta,beta2);
 844
 845     ewtab            = fr->ic->tabq_coul_F;
 846     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 847     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 848
 849     /* Setup water-specific parameters */
 850     inr              = nlist->iinr[0];
 851     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 852     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 853     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 854     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 855     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 856
 857     /* Avoid stupid compiler warnings */
 858     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 859     j_coord_offsetA = 0;
 860     j_coord_offsetB = 0;
 861     j_coord_offsetC = 0;
 862     j_coord_offsetD = 0;
 863     j_coord_offsetE = 0;
 864     j_coord_offsetF = 0;
 865     j_coord_offsetG = 0;
 866     j_coord_offsetH = 0;
 867
 868     outeriter        = 0;
 869     inneriter        = 0;
 870
 871     for(iidx=0;iidx<4*DIM;iidx++)
 872     {
 873         scratch[iidx] = 0.0;
 874     }
 875
 876     /* Start outer loop over neighborlists */
 877     for(iidx=0; iidx<nri; iidx++)
 878     {
 879         /* Load shift vector for this list */
 880         i_shift_offset   = DIM*shiftidx[iidx];
 881
 882         /* Load limits for loop over neighbors */
 883         j_index_start    = jindex[iidx];
 884         j_index_end      = jindex[iidx+1];
 885
 886         /* Get outer coordinate index */
 887         inr              = iinr[iidx];
 888         i_coord_offset   = DIM*inr;
 889
 890         /* Load i particle coords and add shift vector */
 891         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 892                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 893
 894         fix0             = _mm256_setzero_ps();
 895         fiy0             = _mm256_setzero_ps();
 896         fiz0             = _mm256_setzero_ps();
 897         fix1             = _mm256_setzero_ps();
 898         fiy1             = _mm256_setzero_ps();
 899         fiz1             = _mm256_setzero_ps();
 900         fix2             = _mm256_setzero_ps();
 901         fiy2             = _mm256_setzero_ps();
 902         fiz2             = _mm256_setzero_ps();
 903
 904         /* Start inner kernel loop */
 905         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 906         {
 907
 908             /* Get j neighbor index, and coordinate index */
 909             jnrA             = jjnr[jidx];
 910             jnrB             = jjnr[jidx+1];
 911             jnrC             = jjnr[jidx+2];
 912             jnrD             = jjnr[jidx+3];
 913             jnrE             = jjnr[jidx+4];
 914             jnrF             = jjnr[jidx+5];
 915             jnrG             = jjnr[jidx+6];
 916             jnrH             = jjnr[jidx+7];
 917             j_coord_offsetA  = DIM*jnrA;
 918             j_coord_offsetB  = DIM*jnrB;
 919             j_coord_offsetC  = DIM*jnrC;
 920             j_coord_offsetD  = DIM*jnrD;
 921             j_coord_offsetE  = DIM*jnrE;
 922             j_coord_offsetF  = DIM*jnrF;
 923             j_coord_offsetG  = DIM*jnrG;
 924             j_coord_offsetH  = DIM*jnrH;
 925
 926             /* load j atom coordinates */
 927             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 928                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 929                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 930                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 931                                                  &jx0,&jy0,&jz0);
 932
 933             /* Calculate displacement vector */
 934             dx00             = _mm256_sub_ps(ix0,jx0);
 935             dy00             = _mm256_sub_ps(iy0,jy0);
 936             dz00             = _mm256_sub_ps(iz0,jz0);
 937             dx10             = _mm256_sub_ps(ix1,jx0);
 938             dy10             = _mm256_sub_ps(iy1,jy0);
 939             dz10             = _mm256_sub_ps(iz1,jz0);
 940             dx20             = _mm256_sub_ps(ix2,jx0);
 941             dy20             = _mm256_sub_ps(iy2,jy0);
 942             dz20             = _mm256_sub_ps(iz2,jz0);
 943
 944             /* Calculate squared distance and things based on it */
 945             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 946             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 947             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 948
 949             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 950             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 951             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 952
 953             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 954             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 955             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 956
 957             /* Load parameters for j particles */
 958             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 959                                                                  charge+jnrC+0,charge+jnrD+0,
 960                                                                  charge+jnrE+0,charge+jnrF+0,
 961                                                                  charge+jnrG+0,charge+jnrH+0);
 962             vdwjidx0A        = 2*vdwtype[jnrA+0];
 963             vdwjidx0B        = 2*vdwtype[jnrB+0];
 964             vdwjidx0C        = 2*vdwtype[jnrC+0];
 965             vdwjidx0D        = 2*vdwtype[jnrD+0];
 966             vdwjidx0E        = 2*vdwtype[jnrE+0];
 967             vdwjidx0F        = 2*vdwtype[jnrF+0];
 968             vdwjidx0G        = 2*vdwtype[jnrG+0];
 969             vdwjidx0H        = 2*vdwtype[jnrH+0];
 970
 971             fjx0             = _mm256_setzero_ps();
 972             fjy0             = _mm256_setzero_ps();
 973             fjz0             = _mm256_setzero_ps();
 974
 975             /**************************
 976              * CALCULATE INTERACTIONS *
 977              **************************/
 978
 979             r00              = _mm256_mul_ps(rsq00,rinv00);
 980
 981             /* Compute parameters for interactions between i and j atoms */
 982             qq00             = _mm256_mul_ps(iq0,jq0);
 983             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 984                                             vdwioffsetptr0+vdwjidx0B,
 985                                             vdwioffsetptr0+vdwjidx0C,
 986                                             vdwioffsetptr0+vdwjidx0D,
 987                                             vdwioffsetptr0+vdwjidx0E,
 988                                             vdwioffsetptr0+vdwjidx0F,
 989                                             vdwioffsetptr0+vdwjidx0G,
 990                                             vdwioffsetptr0+vdwjidx0H,
 991                                             &c6_00,&c12_00);
 992
 993             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
 994                                                                   vdwgridioffsetptr0+vdwjidx0B,
 995                                                                   vdwgridioffsetptr0+vdwjidx0C,
 996                                                                   vdwgridioffsetptr0+vdwjidx0D,
 997                                                                   vdwgridioffsetptr0+vdwjidx0E,
 998                                                                   vdwgridioffsetptr0+vdwjidx0F,
 999                                                                   vdwgridioffsetptr0+vdwjidx0G,
1000                                                                   vdwgridioffsetptr0+vdwjidx0H);
1001
1002             /* EWALD ELECTROSTATICS */
1003
1004             /* Analytical PME correction */
1005             zeta2            = _mm256_mul_ps(beta2,rsq00);
1006             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1007             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1008             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1009             felec            = _mm256_mul_ps(qq00,felec);
1010
1011             /* Analytical LJ-PME */
1012             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1013             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1014             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1015             exponent         = gmx_simd_exp_r(ewcljrsq);
1016             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1017             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1018             /* f6A = 6 * C6grid * (1 - poly) */
1019             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1020             /* f6B = C6grid * exponent * beta^6 */
1021             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1022             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1023             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);
1024
1025             fscal            = _mm256_add_ps(felec,fvdw);
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm256_mul_ps(fscal,dx00);
1029             ty               = _mm256_mul_ps(fscal,dy00);
1030             tz               = _mm256_mul_ps(fscal,dz00);
1031
1032             /* Update vectorial force */
1033             fix0             = _mm256_add_ps(fix0,tx);
1034             fiy0             = _mm256_add_ps(fiy0,ty);
1035             fiz0             = _mm256_add_ps(fiz0,tz);
1036
1037             fjx0             = _mm256_add_ps(fjx0,tx);
1038             fjy0             = _mm256_add_ps(fjy0,ty);
1039             fjz0             = _mm256_add_ps(fjz0,tz);
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             r10              = _mm256_mul_ps(rsq10,rinv10);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq10             = _mm256_mul_ps(iq1,jq0);
1049
1050             /* EWALD ELECTROSTATICS */
1051
1052             /* Analytical PME correction */
1053             zeta2            = _mm256_mul_ps(beta2,rsq10);
1054             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1055             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1056             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1057             felec            = _mm256_mul_ps(qq10,felec);
1058
1059             fscal            = felec;
1060
1061             /* Calculate temporary vectorial force */
1062             tx               = _mm256_mul_ps(fscal,dx10);
1063             ty               = _mm256_mul_ps(fscal,dy10);
1064             tz               = _mm256_mul_ps(fscal,dz10);
1065
1066             /* Update vectorial force */
1067             fix1             = _mm256_add_ps(fix1,tx);
1068             fiy1             = _mm256_add_ps(fiy1,ty);
1069             fiz1             = _mm256_add_ps(fiz1,tz);
1070
1071             fjx0             = _mm256_add_ps(fjx0,tx);
1072             fjy0             = _mm256_add_ps(fjy0,ty);
1073             fjz0             = _mm256_add_ps(fjz0,tz);
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             r20              = _mm256_mul_ps(rsq20,rinv20);
1080
1081             /* Compute parameters for interactions between i and j atoms */
1082             qq20             = _mm256_mul_ps(iq2,jq0);
1083
1084             /* EWALD ELECTROSTATICS */
1085
1086             /* Analytical PME correction */
1087             zeta2            = _mm256_mul_ps(beta2,rsq20);
1088             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1089             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1090             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1091             felec            = _mm256_mul_ps(qq20,felec);
1092
1093             fscal            = felec;
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm256_mul_ps(fscal,dx20);
1097             ty               = _mm256_mul_ps(fscal,dy20);
1098             tz               = _mm256_mul_ps(fscal,dz20);
1099
1100             /* Update vectorial force */
1101             fix2             = _mm256_add_ps(fix2,tx);
1102             fiy2             = _mm256_add_ps(fiy2,ty);
1103             fiz2             = _mm256_add_ps(fiz2,tz);
1104
1105             fjx0             = _mm256_add_ps(fjx0,tx);
1106             fjy0             = _mm256_add_ps(fjy0,ty);
1107             fjz0             = _mm256_add_ps(fjz0,tz);
1108
1109             fjptrA             = f+j_coord_offsetA;
1110             fjptrB             = f+j_coord_offsetB;
1111             fjptrC             = f+j_coord_offsetC;
1112             fjptrD             = f+j_coord_offsetD;
1113             fjptrE             = f+j_coord_offsetE;
1114             fjptrF             = f+j_coord_offsetF;
1115             fjptrG             = f+j_coord_offsetG;
1116             fjptrH             = f+j_coord_offsetH;
1117
1118             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1119
1120             /* Inner loop uses 194 flops */
1121         }
1122
1123         if(jidx<j_index_end)
1124         {
1125
1126             /* Get j neighbor index, and coordinate index */
1127             jnrlistA         = jjnr[jidx];
1128             jnrlistB         = jjnr[jidx+1];
1129             jnrlistC         = jjnr[jidx+2];
1130             jnrlistD         = jjnr[jidx+3];
1131             jnrlistE         = jjnr[jidx+4];
1132             jnrlistF         = jjnr[jidx+5];
1133             jnrlistG         = jjnr[jidx+6];
1134             jnrlistH         = jjnr[jidx+7];
1135             /* Sign of each element will be negative for non-real atoms.
1136              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1137              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1138              */
1139             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1140                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1141
1142             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1143             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1144             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1145             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1146             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1147             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1148             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1149             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1150             j_coord_offsetA  = DIM*jnrA;
1151             j_coord_offsetB  = DIM*jnrB;
1152             j_coord_offsetC  = DIM*jnrC;
1153             j_coord_offsetD  = DIM*jnrD;
1154             j_coord_offsetE  = DIM*jnrE;
1155             j_coord_offsetF  = DIM*jnrF;
1156             j_coord_offsetG  = DIM*jnrG;
1157             j_coord_offsetH  = DIM*jnrH;
1158
1159             /* load j atom coordinates */
1160             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1161                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1162                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1163                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1164                                                  &jx0,&jy0,&jz0);
1165
1166             /* Calculate displacement vector */
1167             dx00             = _mm256_sub_ps(ix0,jx0);
1168             dy00             = _mm256_sub_ps(iy0,jy0);
1169             dz00             = _mm256_sub_ps(iz0,jz0);
1170             dx10             = _mm256_sub_ps(ix1,jx0);
1171             dy10             = _mm256_sub_ps(iy1,jy0);
1172             dz10             = _mm256_sub_ps(iz1,jz0);
1173             dx20             = _mm256_sub_ps(ix2,jx0);
1174             dy20             = _mm256_sub_ps(iy2,jy0);
1175             dz20             = _mm256_sub_ps(iz2,jz0);
1176
1177             /* Calculate squared distance and things based on it */
1178             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1179             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1180             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1181
1182             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1183             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1184             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1185
1186             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1187             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1188             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1189
1190             /* Load parameters for j particles */
1191             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1192                                                                  charge+jnrC+0,charge+jnrD+0,
1193                                                                  charge+jnrE+0,charge+jnrF+0,
1194                                                                  charge+jnrG+0,charge+jnrH+0);
1195             vdwjidx0A        = 2*vdwtype[jnrA+0];
1196             vdwjidx0B        = 2*vdwtype[jnrB+0];
1197             vdwjidx0C        = 2*vdwtype[jnrC+0];
1198             vdwjidx0D        = 2*vdwtype[jnrD+0];
1199             vdwjidx0E        = 2*vdwtype[jnrE+0];
1200             vdwjidx0F        = 2*vdwtype[jnrF+0];
1201             vdwjidx0G        = 2*vdwtype[jnrG+0];
1202             vdwjidx0H        = 2*vdwtype[jnrH+0];
1203
1204             fjx0             = _mm256_setzero_ps();
1205             fjy0             = _mm256_setzero_ps();
1206             fjz0             = _mm256_setzero_ps();
1207
1208             /**************************
1209              * CALCULATE INTERACTIONS *
1210              **************************/
1211
1212             r00              = _mm256_mul_ps(rsq00,rinv00);
1213             r00              = _mm256_andnot_ps(dummy_mask,r00);
1214
1215             /* Compute parameters for interactions between i and j atoms */
1216             qq00             = _mm256_mul_ps(iq0,jq0);
1217             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1218                                             vdwioffsetptr0+vdwjidx0B,
1219                                             vdwioffsetptr0+vdwjidx0C,
1220                                             vdwioffsetptr0+vdwjidx0D,
1221                                             vdwioffsetptr0+vdwjidx0E,
1222                                             vdwioffsetptr0+vdwjidx0F,
1223                                             vdwioffsetptr0+vdwjidx0G,
1224                                             vdwioffsetptr0+vdwjidx0H,
1225                                             &c6_00,&c12_00);
1226
1227             c6grid_00       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr0+vdwjidx0A,
1228                                                                   vdwgridioffsetptr0+vdwjidx0B,
1229                                                                   vdwgridioffsetptr0+vdwjidx0C,
1230                                                                   vdwgridioffsetptr0+vdwjidx0D,
1231                                                                   vdwgridioffsetptr0+vdwjidx0E,
1232                                                                   vdwgridioffsetptr0+vdwjidx0F,
1233                                                                   vdwgridioffsetptr0+vdwjidx0G,
1234                                                                   vdwgridioffsetptr0+vdwjidx0H);
1235
1236             /* EWALD ELECTROSTATICS */
1237
1238             /* Analytical PME correction */
1239             zeta2            = _mm256_mul_ps(beta2,rsq00);
1240             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1241             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1242             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1243             felec            = _mm256_mul_ps(qq00,felec);
1244
1245             /* Analytical LJ-PME */
1246             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1247             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq00);
1248             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
1249             exponent         = gmx_simd_exp_r(ewcljrsq);
1250             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1251             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1252             /* f6A = 6 * C6grid * (1 - poly) */
1253             f6A              = _mm256_mul_ps(c6grid_00,_mm256_sub_ps(one,poly));
1254             /* f6B = C6grid * exponent * beta^6 */
1255             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_00,one_sixth),_mm256_mul_ps(exponent,ewclj6));
1256             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1257             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);
1258
1259             fscal            = _mm256_add_ps(felec,fvdw);
1260
1261             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1262
1263             /* Calculate temporary vectorial force */
1264             tx               = _mm256_mul_ps(fscal,dx00);
1265             ty               = _mm256_mul_ps(fscal,dy00);
1266             tz               = _mm256_mul_ps(fscal,dz00);
1267
1268             /* Update vectorial force */
1269             fix0             = _mm256_add_ps(fix0,tx);
1270             fiy0             = _mm256_add_ps(fiy0,ty);
1271             fiz0             = _mm256_add_ps(fiz0,tz);
1272
1273             fjx0             = _mm256_add_ps(fjx0,tx);
1274             fjy0             = _mm256_add_ps(fjy0,ty);
1275             fjz0             = _mm256_add_ps(fjz0,tz);
1276
1277             /**************************
1278              * CALCULATE INTERACTIONS *
1279              **************************/
1280
1281             r10              = _mm256_mul_ps(rsq10,rinv10);
1282             r10              = _mm256_andnot_ps(dummy_mask,r10);
1283
1284             /* Compute parameters for interactions between i and j atoms */
1285             qq10             = _mm256_mul_ps(iq1,jq0);
1286
1287             /* EWALD ELECTROSTATICS */
1288
1289             /* Analytical PME correction */
1290             zeta2            = _mm256_mul_ps(beta2,rsq10);
1291             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1292             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1293             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1294             felec            = _mm256_mul_ps(qq10,felec);
1295
1296             fscal            = felec;
1297
1298             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1299
1300             /* Calculate temporary vectorial force */
1301             tx               = _mm256_mul_ps(fscal,dx10);
1302             ty               = _mm256_mul_ps(fscal,dy10);
1303             tz               = _mm256_mul_ps(fscal,dz10);
1304
1305             /* Update vectorial force */
1306             fix1             = _mm256_add_ps(fix1,tx);
1307             fiy1             = _mm256_add_ps(fiy1,ty);
1308             fiz1             = _mm256_add_ps(fiz1,tz);
1309
1310             fjx0             = _mm256_add_ps(fjx0,tx);
1311             fjy0             = _mm256_add_ps(fjy0,ty);
1312             fjz0             = _mm256_add_ps(fjz0,tz);
1313
1314             /**************************
1315              * CALCULATE INTERACTIONS *
1316              **************************/
1317
1318             r20              = _mm256_mul_ps(rsq20,rinv20);
1319             r20              = _mm256_andnot_ps(dummy_mask,r20);
1320
1321             /* Compute parameters for interactions between i and j atoms */
1322             qq20             = _mm256_mul_ps(iq2,jq0);
1323
1324             /* EWALD ELECTROSTATICS */
1325
1326             /* Analytical PME correction */
1327             zeta2            = _mm256_mul_ps(beta2,rsq20);
1328             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1329             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1330             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1331             felec            = _mm256_mul_ps(qq20,felec);
1332
1333             fscal            = felec;
1334
1335             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1336
1337             /* Calculate temporary vectorial force */
1338             tx               = _mm256_mul_ps(fscal,dx20);
1339             ty               = _mm256_mul_ps(fscal,dy20);
1340             tz               = _mm256_mul_ps(fscal,dz20);
1341
1342             /* Update vectorial force */
1343             fix2             = _mm256_add_ps(fix2,tx);
1344             fiy2             = _mm256_add_ps(fiy2,ty);
1345             fiz2             = _mm256_add_ps(fiz2,tz);
1346
1347             fjx0             = _mm256_add_ps(fjx0,tx);
1348             fjy0             = _mm256_add_ps(fjy0,ty);
1349             fjz0             = _mm256_add_ps(fjz0,tz);
1350
1351             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1352             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1353             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1354             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1355             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1356             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1357             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1358             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1359
1360             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1361
1362             /* Inner loop uses 197 flops */
1363         }
1364
1365         /* End of innermost loop */
1366
1367         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1368                                                  f+i_coord_offset,fshift+i_shift_offset);
1369
1370         /* Increment number of inner iterations */
1371         inneriter                  += j_index_end - j_index_start;
1372
1373         /* Outer loop uses 18 flops */
1374     }
1375
1376     /* Increment number of outer iterations */
1377     outeriter        += nri;
1378
1379     /* Update outer/inner flops */
1380
1381     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*197);
1382 }