src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwLJ_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     real *           vdwioffsetptr1;
  86     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 101     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 102     __m256d          dummy_mask,cutoff_mask;
 103     __m128           tmpmask0,tmpmask1;
 104     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 105     __m256d          one     = _mm256_set1_pd(1.0);
 106     __m256d          two     = _mm256_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm256_set1_pd(fr->ic->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 127     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 128     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 129     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 163
 164         fix0             = _mm256_setzero_pd();
 165         fiy0             = _mm256_setzero_pd();
 166         fiz0             = _mm256_setzero_pd();
 167         fix1             = _mm256_setzero_pd();
 168         fiy1             = _mm256_setzero_pd();
 169         fiz1             = _mm256_setzero_pd();
 170         fix2             = _mm256_setzero_pd();
 171         fiy2             = _mm256_setzero_pd();
 172         fiz2             = _mm256_setzero_pd();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm256_setzero_pd();
 176         vvdwsum          = _mm256_setzero_pd();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 195                                                  &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm256_sub_pd(ix0,jx0);
 199             dy00             = _mm256_sub_pd(iy0,jy0);
 200             dz00             = _mm256_sub_pd(iz0,jz0);
 201             dx10             = _mm256_sub_pd(ix1,jx0);
 202             dy10             = _mm256_sub_pd(iy1,jy0);
 203             dz10             = _mm256_sub_pd(iz1,jz0);
 204             dx20             = _mm256_sub_pd(ix2,jx0);
 205             dy20             = _mm256_sub_pd(iy2,jy0);
 206             dz20             = _mm256_sub_pd(iz2,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 210             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 211             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 212
 213             rinv00           = avx256_invsqrt_d(rsq00);
 214             rinv10           = avx256_invsqrt_d(rsq10);
 215             rinv20           = avx256_invsqrt_d(rsq20);
 216
 217             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 218             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 219             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 223                                                                  charge+jnrC+0,charge+jnrD+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228
 229             fjx0             = _mm256_setzero_pd();
 230             fjy0             = _mm256_setzero_pd();
 231             fjz0             = _mm256_setzero_pd();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq00             = _mm256_mul_pd(iq0,jq0);
 239             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 240                                             vdwioffsetptr0+vdwjidx0B,
 241                                             vdwioffsetptr0+vdwjidx0C,
 242                                             vdwioffsetptr0+vdwjidx0D,
 243                                             &c6_00,&c12_00);
 244
 245             /* COULOMB ELECTROSTATICS */
 246             velec            = _mm256_mul_pd(qq00,rinv00);
 247             felec            = _mm256_mul_pd(velec,rinvsq00);
 248
 249             /* LENNARD-JONES DISPERSION/REPULSION */
 250
 251             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 252             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 253             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 254             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 255             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 256
 257             /* Update potential sum for this i atom from the interaction with this j atom. */
 258             velecsum         = _mm256_add_pd(velecsum,velec);
 259             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 260
 261             fscal            = _mm256_add_pd(felec,fvdw);
 262
 263             /* Calculate temporary vectorial force */
 264             tx               = _mm256_mul_pd(fscal,dx00);
 265             ty               = _mm256_mul_pd(fscal,dy00);
 266             tz               = _mm256_mul_pd(fscal,dz00);
 267
 268             /* Update vectorial force */
 269             fix0             = _mm256_add_pd(fix0,tx);
 270             fiy0             = _mm256_add_pd(fiy0,ty);
 271             fiz0             = _mm256_add_pd(fiz0,tz);
 272
 273             fjx0             = _mm256_add_pd(fjx0,tx);
 274             fjy0             = _mm256_add_pd(fjy0,ty);
 275             fjz0             = _mm256_add_pd(fjz0,tz);
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             /* Compute parameters for interactions between i and j atoms */
 282             qq10             = _mm256_mul_pd(iq1,jq0);
 283
 284             /* COULOMB ELECTROSTATICS */
 285             velec            = _mm256_mul_pd(qq10,rinv10);
 286             felec            = _mm256_mul_pd(velec,rinvsq10);
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             velecsum         = _mm256_add_pd(velecsum,velec);
 290
 291             fscal            = felec;
 292
 293             /* Calculate temporary vectorial force */
 294             tx               = _mm256_mul_pd(fscal,dx10);
 295             ty               = _mm256_mul_pd(fscal,dy10);
 296             tz               = _mm256_mul_pd(fscal,dz10);
 297
 298             /* Update vectorial force */
 299             fix1             = _mm256_add_pd(fix1,tx);
 300             fiy1             = _mm256_add_pd(fiy1,ty);
 301             fiz1             = _mm256_add_pd(fiz1,tz);
 302
 303             fjx0             = _mm256_add_pd(fjx0,tx);
 304             fjy0             = _mm256_add_pd(fjy0,ty);
 305             fjz0             = _mm256_add_pd(fjz0,tz);
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq20             = _mm256_mul_pd(iq2,jq0);
 313
 314             /* COULOMB ELECTROSTATICS */
 315             velec            = _mm256_mul_pd(qq20,rinv20);
 316             felec            = _mm256_mul_pd(velec,rinvsq20);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velecsum         = _mm256_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm256_mul_pd(fscal,dx20);
 325             ty               = _mm256_mul_pd(fscal,dy20);
 326             tz               = _mm256_mul_pd(fscal,dz20);
 327
 328             /* Update vectorial force */
 329             fix2             = _mm256_add_pd(fix2,tx);
 330             fiy2             = _mm256_add_pd(fiy2,ty);
 331             fiz2             = _mm256_add_pd(fiz2,tz);
 332
 333             fjx0             = _mm256_add_pd(fjx0,tx);
 334             fjy0             = _mm256_add_pd(fjy0,ty);
 335             fjz0             = _mm256_add_pd(fjz0,tz);
 336
 337             fjptrA             = f+j_coord_offsetA;
 338             fjptrB             = f+j_coord_offsetB;
 339             fjptrC             = f+j_coord_offsetC;
 340             fjptrD             = f+j_coord_offsetD;
 341
 342             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 343
 344             /* Inner loop uses 96 flops */
 345         }
 346
 347         if(jidx<j_index_end)
 348         {
 349
 350             /* Get j neighbor index, and coordinate index */
 351             jnrlistA         = jjnr[jidx];
 352             jnrlistB         = jjnr[jidx+1];
 353             jnrlistC         = jjnr[jidx+2];
 354             jnrlistD         = jjnr[jidx+3];
 355             /* Sign of each element will be negative for non-real atoms.
 356              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 357              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 358              */
 359             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 360
 361             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 362             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 363             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 364
 365             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 366             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 367             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 368             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 369             j_coord_offsetA  = DIM*jnrA;
 370             j_coord_offsetB  = DIM*jnrB;
 371             j_coord_offsetC  = DIM*jnrC;
 372             j_coord_offsetD  = DIM*jnrD;
 373
 374             /* load j atom coordinates */
 375             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 376                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 377                                                  &jx0,&jy0,&jz0);
 378
 379             /* Calculate displacement vector */
 380             dx00             = _mm256_sub_pd(ix0,jx0);
 381             dy00             = _mm256_sub_pd(iy0,jy0);
 382             dz00             = _mm256_sub_pd(iz0,jz0);
 383             dx10             = _mm256_sub_pd(ix1,jx0);
 384             dy10             = _mm256_sub_pd(iy1,jy0);
 385             dz10             = _mm256_sub_pd(iz1,jz0);
 386             dx20             = _mm256_sub_pd(ix2,jx0);
 387             dy20             = _mm256_sub_pd(iy2,jy0);
 388             dz20             = _mm256_sub_pd(iz2,jz0);
 389
 390             /* Calculate squared distance and things based on it */
 391             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 392             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 393             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 394
 395             rinv00           = avx256_invsqrt_d(rsq00);
 396             rinv10           = avx256_invsqrt_d(rsq10);
 397             rinv20           = avx256_invsqrt_d(rsq20);
 398
 399             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 400             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 401             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 402
 403             /* Load parameters for j particles */
 404             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 405                                                                  charge+jnrC+0,charge+jnrD+0);
 406             vdwjidx0A        = 2*vdwtype[jnrA+0];
 407             vdwjidx0B        = 2*vdwtype[jnrB+0];
 408             vdwjidx0C        = 2*vdwtype[jnrC+0];
 409             vdwjidx0D        = 2*vdwtype[jnrD+0];
 410
 411             fjx0             = _mm256_setzero_pd();
 412             fjy0             = _mm256_setzero_pd();
 413             fjz0             = _mm256_setzero_pd();
 414
 415             /**************************
 416              * CALCULATE INTERACTIONS *
 417              **************************/
 418
 419             /* Compute parameters for interactions between i and j atoms */
 420             qq00             = _mm256_mul_pd(iq0,jq0);
 421             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 422                                             vdwioffsetptr0+vdwjidx0B,
 423                                             vdwioffsetptr0+vdwjidx0C,
 424                                             vdwioffsetptr0+vdwjidx0D,
 425                                             &c6_00,&c12_00);
 426
 427             /* COULOMB ELECTROSTATICS */
 428             velec            = _mm256_mul_pd(qq00,rinv00);
 429             felec            = _mm256_mul_pd(velec,rinvsq00);
 430
 431             /* LENNARD-JONES DISPERSION/REPULSION */
 432
 433             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 434             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 435             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 436             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 437             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 438
 439             /* Update potential sum for this i atom from the interaction with this j atom. */
 440             velec            = _mm256_andnot_pd(dummy_mask,velec);
 441             velecsum         = _mm256_add_pd(velecsum,velec);
 442             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 443             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 444
 445             fscal            = _mm256_add_pd(felec,fvdw);
 446
 447             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 448
 449             /* Calculate temporary vectorial force */
 450             tx               = _mm256_mul_pd(fscal,dx00);
 451             ty               = _mm256_mul_pd(fscal,dy00);
 452             tz               = _mm256_mul_pd(fscal,dz00);
 453
 454             /* Update vectorial force */
 455             fix0             = _mm256_add_pd(fix0,tx);
 456             fiy0             = _mm256_add_pd(fiy0,ty);
 457             fiz0             = _mm256_add_pd(fiz0,tz);
 458
 459             fjx0             = _mm256_add_pd(fjx0,tx);
 460             fjy0             = _mm256_add_pd(fjy0,ty);
 461             fjz0             = _mm256_add_pd(fjz0,tz);
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             qq10             = _mm256_mul_pd(iq1,jq0);
 469
 470             /* COULOMB ELECTROSTATICS */
 471             velec            = _mm256_mul_pd(qq10,rinv10);
 472             felec            = _mm256_mul_pd(velec,rinvsq10);
 473
 474             /* Update potential sum for this i atom from the interaction with this j atom. */
 475             velec            = _mm256_andnot_pd(dummy_mask,velec);
 476             velecsum         = _mm256_add_pd(velecsum,velec);
 477
 478             fscal            = felec;
 479
 480             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 481
 482             /* Calculate temporary vectorial force */
 483             tx               = _mm256_mul_pd(fscal,dx10);
 484             ty               = _mm256_mul_pd(fscal,dy10);
 485             tz               = _mm256_mul_pd(fscal,dz10);
 486
 487             /* Update vectorial force */
 488             fix1             = _mm256_add_pd(fix1,tx);
 489             fiy1             = _mm256_add_pd(fiy1,ty);
 490             fiz1             = _mm256_add_pd(fiz1,tz);
 491
 492             fjx0             = _mm256_add_pd(fjx0,tx);
 493             fjy0             = _mm256_add_pd(fjy0,ty);
 494             fjz0             = _mm256_add_pd(fjz0,tz);
 495
 496             /**************************
 497              * CALCULATE INTERACTIONS *
 498              **************************/
 499
 500             /* Compute parameters for interactions between i and j atoms */
 501             qq20             = _mm256_mul_pd(iq2,jq0);
 502
 503             /* COULOMB ELECTROSTATICS */
 504             velec            = _mm256_mul_pd(qq20,rinv20);
 505             felec            = _mm256_mul_pd(velec,rinvsq20);
 506
 507             /* Update potential sum for this i atom from the interaction with this j atom. */
 508             velec            = _mm256_andnot_pd(dummy_mask,velec);
 509             velecsum         = _mm256_add_pd(velecsum,velec);
 510
 511             fscal            = felec;
 512
 513             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 514
 515             /* Calculate temporary vectorial force */
 516             tx               = _mm256_mul_pd(fscal,dx20);
 517             ty               = _mm256_mul_pd(fscal,dy20);
 518             tz               = _mm256_mul_pd(fscal,dz20);
 519
 520             /* Update vectorial force */
 521             fix2             = _mm256_add_pd(fix2,tx);
 522             fiy2             = _mm256_add_pd(fiy2,ty);
 523             fiz2             = _mm256_add_pd(fiz2,tz);
 524
 525             fjx0             = _mm256_add_pd(fjx0,tx);
 526             fjy0             = _mm256_add_pd(fjy0,ty);
 527             fjz0             = _mm256_add_pd(fjz0,tz);
 528
 529             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 530             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 531             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 532             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 533
 534             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 535
 536             /* Inner loop uses 96 flops */
 537         }
 538
 539         /* End of innermost loop */
 540
 541         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 542                                                  f+i_coord_offset,fshift+i_shift_offset);
 543
 544         ggid                        = gid[iidx];
 545         /* Update potential energies */
 546         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 547         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 548
 549         /* Increment number of inner iterations */
 550         inneriter                  += j_index_end - j_index_start;
 551
 552         /* Outer loop uses 20 flops */
 553     }
 554
 555     /* Increment number of outer iterations */
 556     outeriter        += nri;
 557
 558     /* Update outer/inner flops */
 559
 560     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*96);
 561 }
 562 /*
 563  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_double
 564  * Electrostatics interaction: Coulomb
 565  * VdW interaction:            LennardJones
 566  * Geometry:                   Water3-Particle
 567  * Calculate force/pot:        Force
 568  */
 569 void
 570 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_256_double
 571                     (t_nblist                    * gmx_restrict       nlist,
 572                      rvec                        * gmx_restrict          xx,
 573                      rvec                        * gmx_restrict          ff,
 574                      struct t_forcerec           * gmx_restrict          fr,
 575                      t_mdatoms                   * gmx_restrict     mdatoms,
 576                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 577                      t_nrnb                      * gmx_restrict        nrnb)
 578 {
 579     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 580      * just 0 for non-waters.
 581      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 582      * jnr indices corresponding to data put in the four positions in the SIMD register.
 583      */
 584     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 585     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 586     int              jnrA,jnrB,jnrC,jnrD;
 587     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 588     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 589     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 590     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 591     real             rcutoff_scalar;
 592     real             *shiftvec,*fshift,*x,*f;
 593     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 594     real             scratch[4*DIM];
 595     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 596     real *           vdwioffsetptr0;
 597     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 598     real *           vdwioffsetptr1;
 599     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 600     real *           vdwioffsetptr2;
 601     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 602     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 603     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 604     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 605     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 606     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 607     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 608     real             *charge;
 609     int              nvdwtype;
 610     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 611     int              *vdwtype;
 612     real             *vdwparam;
 613     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 614     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 615     __m256d          dummy_mask,cutoff_mask;
 616     __m128           tmpmask0,tmpmask1;
 617     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 618     __m256d          one     = _mm256_set1_pd(1.0);
 619     __m256d          two     = _mm256_set1_pd(2.0);
 620     x                = xx[0];
 621     f                = ff[0];
 622
 623     nri              = nlist->nri;
 624     iinr             = nlist->iinr;
 625     jindex           = nlist->jindex;
 626     jjnr             = nlist->jjnr;
 627     shiftidx         = nlist->shift;
 628     gid              = nlist->gid;
 629     shiftvec         = fr->shift_vec[0];
 630     fshift           = fr->fshift[0];
 631     facel            = _mm256_set1_pd(fr->ic->epsfac);
 632     charge           = mdatoms->chargeA;
 633     nvdwtype         = fr->ntype;
 634     vdwparam         = fr->nbfp;
 635     vdwtype          = mdatoms->typeA;
 636
 637     /* Setup water-specific parameters */
 638     inr              = nlist->iinr[0];
 639     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 640     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 641     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 642     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 643
 644     /* Avoid stupid compiler warnings */
 645     jnrA = jnrB = jnrC = jnrD = 0;
 646     j_coord_offsetA = 0;
 647     j_coord_offsetB = 0;
 648     j_coord_offsetC = 0;
 649     j_coord_offsetD = 0;
 650
 651     outeriter        = 0;
 652     inneriter        = 0;
 653
 654     for(iidx=0;iidx<4*DIM;iidx++)
 655     {
 656         scratch[iidx] = 0.0;
 657     }
 658
 659     /* Start outer loop over neighborlists */
 660     for(iidx=0; iidx<nri; iidx++)
 661     {
 662         /* Load shift vector for this list */
 663         i_shift_offset   = DIM*shiftidx[iidx];
 664
 665         /* Load limits for loop over neighbors */
 666         j_index_start    = jindex[iidx];
 667         j_index_end      = jindex[iidx+1];
 668
 669         /* Get outer coordinate index */
 670         inr              = iinr[iidx];
 671         i_coord_offset   = DIM*inr;
 672
 673         /* Load i particle coords and add shift vector */
 674         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 675                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 676
 677         fix0             = _mm256_setzero_pd();
 678         fiy0             = _mm256_setzero_pd();
 679         fiz0             = _mm256_setzero_pd();
 680         fix1             = _mm256_setzero_pd();
 681         fiy1             = _mm256_setzero_pd();
 682         fiz1             = _mm256_setzero_pd();
 683         fix2             = _mm256_setzero_pd();
 684         fiy2             = _mm256_setzero_pd();
 685         fiz2             = _mm256_setzero_pd();
 686
 687         /* Start inner kernel loop */
 688         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 689         {
 690
 691             /* Get j neighbor index, and coordinate index */
 692             jnrA             = jjnr[jidx];
 693             jnrB             = jjnr[jidx+1];
 694             jnrC             = jjnr[jidx+2];
 695             jnrD             = jjnr[jidx+3];
 696             j_coord_offsetA  = DIM*jnrA;
 697             j_coord_offsetB  = DIM*jnrB;
 698             j_coord_offsetC  = DIM*jnrC;
 699             j_coord_offsetD  = DIM*jnrD;
 700
 701             /* load j atom coordinates */
 702             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 703                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 704                                                  &jx0,&jy0,&jz0);
 705
 706             /* Calculate displacement vector */
 707             dx00             = _mm256_sub_pd(ix0,jx0);
 708             dy00             = _mm256_sub_pd(iy0,jy0);
 709             dz00             = _mm256_sub_pd(iz0,jz0);
 710             dx10             = _mm256_sub_pd(ix1,jx0);
 711             dy10             = _mm256_sub_pd(iy1,jy0);
 712             dz10             = _mm256_sub_pd(iz1,jz0);
 713             dx20             = _mm256_sub_pd(ix2,jx0);
 714             dy20             = _mm256_sub_pd(iy2,jy0);
 715             dz20             = _mm256_sub_pd(iz2,jz0);
 716
 717             /* Calculate squared distance and things based on it */
 718             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 719             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 720             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 721
 722             rinv00           = avx256_invsqrt_d(rsq00);
 723             rinv10           = avx256_invsqrt_d(rsq10);
 724             rinv20           = avx256_invsqrt_d(rsq20);
 725
 726             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 727             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 728             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 729
 730             /* Load parameters for j particles */
 731             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 732                                                                  charge+jnrC+0,charge+jnrD+0);
 733             vdwjidx0A        = 2*vdwtype[jnrA+0];
 734             vdwjidx0B        = 2*vdwtype[jnrB+0];
 735             vdwjidx0C        = 2*vdwtype[jnrC+0];
 736             vdwjidx0D        = 2*vdwtype[jnrD+0];
 737
 738             fjx0             = _mm256_setzero_pd();
 739             fjy0             = _mm256_setzero_pd();
 740             fjz0             = _mm256_setzero_pd();
 741
 742             /**************************
 743              * CALCULATE INTERACTIONS *
 744              **************************/
 745
 746             /* Compute parameters for interactions between i and j atoms */
 747             qq00             = _mm256_mul_pd(iq0,jq0);
 748             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 749                                             vdwioffsetptr0+vdwjidx0B,
 750                                             vdwioffsetptr0+vdwjidx0C,
 751                                             vdwioffsetptr0+vdwjidx0D,
 752                                             &c6_00,&c12_00);
 753
 754             /* COULOMB ELECTROSTATICS */
 755             velec            = _mm256_mul_pd(qq00,rinv00);
 756             felec            = _mm256_mul_pd(velec,rinvsq00);
 757
 758             /* LENNARD-JONES DISPERSION/REPULSION */
 759
 760             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 761             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 762
 763             fscal            = _mm256_add_pd(felec,fvdw);
 764
 765             /* Calculate temporary vectorial force */
 766             tx               = _mm256_mul_pd(fscal,dx00);
 767             ty               = _mm256_mul_pd(fscal,dy00);
 768             tz               = _mm256_mul_pd(fscal,dz00);
 769
 770             /* Update vectorial force */
 771             fix0             = _mm256_add_pd(fix0,tx);
 772             fiy0             = _mm256_add_pd(fiy0,ty);
 773             fiz0             = _mm256_add_pd(fiz0,tz);
 774
 775             fjx0             = _mm256_add_pd(fjx0,tx);
 776             fjy0             = _mm256_add_pd(fjy0,ty);
 777             fjz0             = _mm256_add_pd(fjz0,tz);
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             /* Compute parameters for interactions between i and j atoms */
 784             qq10             = _mm256_mul_pd(iq1,jq0);
 785
 786             /* COULOMB ELECTROSTATICS */
 787             velec            = _mm256_mul_pd(qq10,rinv10);
 788             felec            = _mm256_mul_pd(velec,rinvsq10);
 789
 790             fscal            = felec;
 791
 792             /* Calculate temporary vectorial force */
 793             tx               = _mm256_mul_pd(fscal,dx10);
 794             ty               = _mm256_mul_pd(fscal,dy10);
 795             tz               = _mm256_mul_pd(fscal,dz10);
 796
 797             /* Update vectorial force */
 798             fix1             = _mm256_add_pd(fix1,tx);
 799             fiy1             = _mm256_add_pd(fiy1,ty);
 800             fiz1             = _mm256_add_pd(fiz1,tz);
 801
 802             fjx0             = _mm256_add_pd(fjx0,tx);
 803             fjy0             = _mm256_add_pd(fjy0,ty);
 804             fjz0             = _mm256_add_pd(fjz0,tz);
 805
 806             /**************************
 807              * CALCULATE INTERACTIONS *
 808              **************************/
 809
 810             /* Compute parameters for interactions between i and j atoms */
 811             qq20             = _mm256_mul_pd(iq2,jq0);
 812
 813             /* COULOMB ELECTROSTATICS */
 814             velec            = _mm256_mul_pd(qq20,rinv20);
 815             felec            = _mm256_mul_pd(velec,rinvsq20);
 816
 817             fscal            = felec;
 818
 819             /* Calculate temporary vectorial force */
 820             tx               = _mm256_mul_pd(fscal,dx20);
 821             ty               = _mm256_mul_pd(fscal,dy20);
 822             tz               = _mm256_mul_pd(fscal,dz20);
 823
 824             /* Update vectorial force */
 825             fix2             = _mm256_add_pd(fix2,tx);
 826             fiy2             = _mm256_add_pd(fiy2,ty);
 827             fiz2             = _mm256_add_pd(fiz2,tz);
 828
 829             fjx0             = _mm256_add_pd(fjx0,tx);
 830             fjy0             = _mm256_add_pd(fjy0,ty);
 831             fjz0             = _mm256_add_pd(fjz0,tz);
 832
 833             fjptrA             = f+j_coord_offsetA;
 834             fjptrB             = f+j_coord_offsetB;
 835             fjptrC             = f+j_coord_offsetC;
 836             fjptrD             = f+j_coord_offsetD;
 837
 838             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 839
 840             /* Inner loop uses 88 flops */
 841         }
 842
 843         if(jidx<j_index_end)
 844         {
 845
 846             /* Get j neighbor index, and coordinate index */
 847             jnrlistA         = jjnr[jidx];
 848             jnrlistB         = jjnr[jidx+1];
 849             jnrlistC         = jjnr[jidx+2];
 850             jnrlistD         = jjnr[jidx+3];
 851             /* Sign of each element will be negative for non-real atoms.
 852              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 853              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 854              */
 855             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 856
 857             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 858             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 859             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 860
 861             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 862             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 863             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 864             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 865             j_coord_offsetA  = DIM*jnrA;
 866             j_coord_offsetB  = DIM*jnrB;
 867             j_coord_offsetC  = DIM*jnrC;
 868             j_coord_offsetD  = DIM*jnrD;
 869
 870             /* load j atom coordinates */
 871             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 872                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 873                                                  &jx0,&jy0,&jz0);
 874
 875             /* Calculate displacement vector */
 876             dx00             = _mm256_sub_pd(ix0,jx0);
 877             dy00             = _mm256_sub_pd(iy0,jy0);
 878             dz00             = _mm256_sub_pd(iz0,jz0);
 879             dx10             = _mm256_sub_pd(ix1,jx0);
 880             dy10             = _mm256_sub_pd(iy1,jy0);
 881             dz10             = _mm256_sub_pd(iz1,jz0);
 882             dx20             = _mm256_sub_pd(ix2,jx0);
 883             dy20             = _mm256_sub_pd(iy2,jy0);
 884             dz20             = _mm256_sub_pd(iz2,jz0);
 885
 886             /* Calculate squared distance and things based on it */
 887             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 888             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 889             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 890
 891             rinv00           = avx256_invsqrt_d(rsq00);
 892             rinv10           = avx256_invsqrt_d(rsq10);
 893             rinv20           = avx256_invsqrt_d(rsq20);
 894
 895             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 896             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 897             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 898
 899             /* Load parameters for j particles */
 900             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 901                                                                  charge+jnrC+0,charge+jnrD+0);
 902             vdwjidx0A        = 2*vdwtype[jnrA+0];
 903             vdwjidx0B        = 2*vdwtype[jnrB+0];
 904             vdwjidx0C        = 2*vdwtype[jnrC+0];
 905             vdwjidx0D        = 2*vdwtype[jnrD+0];
 906
 907             fjx0             = _mm256_setzero_pd();
 908             fjy0             = _mm256_setzero_pd();
 909             fjz0             = _mm256_setzero_pd();
 910
 911             /**************************
 912              * CALCULATE INTERACTIONS *
 913              **************************/
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq00             = _mm256_mul_pd(iq0,jq0);
 917             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 918                                             vdwioffsetptr0+vdwjidx0B,
 919                                             vdwioffsetptr0+vdwjidx0C,
 920                                             vdwioffsetptr0+vdwjidx0D,
 921                                             &c6_00,&c12_00);
 922
 923             /* COULOMB ELECTROSTATICS */
 924             velec            = _mm256_mul_pd(qq00,rinv00);
 925             felec            = _mm256_mul_pd(velec,rinvsq00);
 926
 927             /* LENNARD-JONES DISPERSION/REPULSION */
 928
 929             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 930             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 931
 932             fscal            = _mm256_add_pd(felec,fvdw);
 933
 934             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 935
 936             /* Calculate temporary vectorial force */
 937             tx               = _mm256_mul_pd(fscal,dx00);
 938             ty               = _mm256_mul_pd(fscal,dy00);
 939             tz               = _mm256_mul_pd(fscal,dz00);
 940
 941             /* Update vectorial force */
 942             fix0             = _mm256_add_pd(fix0,tx);
 943             fiy0             = _mm256_add_pd(fiy0,ty);
 944             fiz0             = _mm256_add_pd(fiz0,tz);
 945
 946             fjx0             = _mm256_add_pd(fjx0,tx);
 947             fjy0             = _mm256_add_pd(fjy0,ty);
 948             fjz0             = _mm256_add_pd(fjz0,tz);
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq10             = _mm256_mul_pd(iq1,jq0);
 956
 957             /* COULOMB ELECTROSTATICS */
 958             velec            = _mm256_mul_pd(qq10,rinv10);
 959             felec            = _mm256_mul_pd(velec,rinvsq10);
 960
 961             fscal            = felec;
 962
 963             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 964
 965             /* Calculate temporary vectorial force */
 966             tx               = _mm256_mul_pd(fscal,dx10);
 967             ty               = _mm256_mul_pd(fscal,dy10);
 968             tz               = _mm256_mul_pd(fscal,dz10);
 969
 970             /* Update vectorial force */
 971             fix1             = _mm256_add_pd(fix1,tx);
 972             fiy1             = _mm256_add_pd(fiy1,ty);
 973             fiz1             = _mm256_add_pd(fiz1,tz);
 974
 975             fjx0             = _mm256_add_pd(fjx0,tx);
 976             fjy0             = _mm256_add_pd(fjy0,ty);
 977             fjz0             = _mm256_add_pd(fjz0,tz);
 978
 979             /**************************
 980              * CALCULATE INTERACTIONS *
 981              **************************/
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             qq20             = _mm256_mul_pd(iq2,jq0);
 985
 986             /* COULOMB ELECTROSTATICS */
 987             velec            = _mm256_mul_pd(qq20,rinv20);
 988             felec            = _mm256_mul_pd(velec,rinvsq20);
 989
 990             fscal            = felec;
 991
 992             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 993
 994             /* Calculate temporary vectorial force */
 995             tx               = _mm256_mul_pd(fscal,dx20);
 996             ty               = _mm256_mul_pd(fscal,dy20);
 997             tz               = _mm256_mul_pd(fscal,dz20);
 998
 999             /* Update vectorial force */
1000             fix2             = _mm256_add_pd(fix2,tx);
1001             fiy2             = _mm256_add_pd(fiy2,ty);
1002             fiz2             = _mm256_add_pd(fiz2,tz);
1003
1004             fjx0             = _mm256_add_pd(fjx0,tx);
1005             fjy0             = _mm256_add_pd(fjy0,ty);
1006             fjz0             = _mm256_add_pd(fjz0,tz);
1007
1008             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1009             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1010             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1011             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1012
1013             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1014
1015             /* Inner loop uses 88 flops */
1016         }
1017
1018         /* End of innermost loop */
1019
1020         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1021                                                  f+i_coord_offset,fshift+i_shift_offset);
1022
1023         /* Increment number of inner iterations */
1024         inneriter                  += j_index_end - j_index_start;
1025
1026         /* Outer loop uses 18 flops */
1027     }
1028
1029     /* Increment number of outer iterations */
1030     outeriter        += nri;
1031
1032     /* Update outer/inner flops */
1033
1034     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*88);
1035 }