src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_avx_256_double
  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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  96     real             *charge;
  97     int              nvdwtype;
  98     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  99     int              *vdwtype;
 100     real             *vdwparam;
 101     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 102     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 103     __m256d          dummy_mask,cutoff_mask;
 104     __m128           tmpmask0,tmpmask1;
 105     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 106     __m256d          one     = _mm256_set1_pd(1.0);
 107     __m256d          two     = _mm256_set1_pd(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm256_set1_pd(fr->epsfac);
 120     charge           = mdatoms->chargeA;
 121     krf              = _mm256_set1_pd(fr->ic->k_rf);
 122     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 123     crf              = _mm256_set1_pd(fr->ic->c_rf);
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 131     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 132     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 133     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 167
 168         fix0             = _mm256_setzero_pd();
 169         fiy0             = _mm256_setzero_pd();
 170         fiz0             = _mm256_setzero_pd();
 171         fix1             = _mm256_setzero_pd();
 172         fiy1             = _mm256_setzero_pd();
 173         fiz1             = _mm256_setzero_pd();
 174         fix2             = _mm256_setzero_pd();
 175         fiy2             = _mm256_setzero_pd();
 176         fiz2             = _mm256_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm256_setzero_pd();
 180         vvdwsum          = _mm256_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             jnrC             = jjnr[jidx+2];
 190             jnrD             = jjnr[jidx+3];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195
 196             /* load j atom coordinates */
 197             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 199                                                  &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm256_sub_pd(ix0,jx0);
 203             dy00             = _mm256_sub_pd(iy0,jy0);
 204             dz00             = _mm256_sub_pd(iz0,jz0);
 205             dx10             = _mm256_sub_pd(ix1,jx0);
 206             dy10             = _mm256_sub_pd(iy1,jy0);
 207             dz10             = _mm256_sub_pd(iz1,jz0);
 208             dx20             = _mm256_sub_pd(ix2,jx0);
 209             dy20             = _mm256_sub_pd(iy2,jy0);
 210             dz20             = _mm256_sub_pd(iz2,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 214             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 215             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 216
 217             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 218             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 219             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 220
 221             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 222             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 223             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 227                                                                  charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             fjx0             = _mm256_setzero_pd();
 234             fjy0             = _mm256_setzero_pd();
 235             fjz0             = _mm256_setzero_pd();
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _mm256_mul_pd(iq0,jq0);
 243             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 244                                             vdwioffsetptr0+vdwjidx0B,
 245                                             vdwioffsetptr0+vdwjidx0C,
 246                                             vdwioffsetptr0+vdwjidx0D,
 247                                             &c6_00,&c12_00);
 248
 249             /* REACTION-FIELD ELECTROSTATICS */
 250             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 251             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 252
 253             /* LENNARD-JONES DISPERSION/REPULSION */
 254
 255             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 256             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 257             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 258             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 259             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velecsum         = _mm256_add_pd(velecsum,velec);
 263             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 264
 265             fscal            = _mm256_add_pd(felec,fvdw);
 266
 267             /* Calculate temporary vectorial force */
 268             tx               = _mm256_mul_pd(fscal,dx00);
 269             ty               = _mm256_mul_pd(fscal,dy00);
 270             tz               = _mm256_mul_pd(fscal,dz00);
 271
 272             /* Update vectorial force */
 273             fix0             = _mm256_add_pd(fix0,tx);
 274             fiy0             = _mm256_add_pd(fiy0,ty);
 275             fiz0             = _mm256_add_pd(fiz0,tz);
 276
 277             fjx0             = _mm256_add_pd(fjx0,tx);
 278             fjy0             = _mm256_add_pd(fjy0,ty);
 279             fjz0             = _mm256_add_pd(fjz0,tz);
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq10             = _mm256_mul_pd(iq1,jq0);
 287
 288             /* REACTION-FIELD ELECTROSTATICS */
 289             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 290             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velecsum         = _mm256_add_pd(velecsum,velec);
 294
 295             fscal            = felec;
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm256_mul_pd(fscal,dx10);
 299             ty               = _mm256_mul_pd(fscal,dy10);
 300             tz               = _mm256_mul_pd(fscal,dz10);
 301
 302             /* Update vectorial force */
 303             fix1             = _mm256_add_pd(fix1,tx);
 304             fiy1             = _mm256_add_pd(fiy1,ty);
 305             fiz1             = _mm256_add_pd(fiz1,tz);
 306
 307             fjx0             = _mm256_add_pd(fjx0,tx);
 308             fjy0             = _mm256_add_pd(fjy0,ty);
 309             fjz0             = _mm256_add_pd(fjz0,tz);
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             qq20             = _mm256_mul_pd(iq2,jq0);
 317
 318             /* REACTION-FIELD ELECTROSTATICS */
 319             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 320             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velecsum         = _mm256_add_pd(velecsum,velec);
 324
 325             fscal            = felec;
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm256_mul_pd(fscal,dx20);
 329             ty               = _mm256_mul_pd(fscal,dy20);
 330             tz               = _mm256_mul_pd(fscal,dz20);
 331
 332             /* Update vectorial force */
 333             fix2             = _mm256_add_pd(fix2,tx);
 334             fiy2             = _mm256_add_pd(fiy2,ty);
 335             fiz2             = _mm256_add_pd(fiz2,tz);
 336
 337             fjx0             = _mm256_add_pd(fjx0,tx);
 338             fjy0             = _mm256_add_pd(fjy0,ty);
 339             fjz0             = _mm256_add_pd(fjz0,tz);
 340
 341             fjptrA             = f+j_coord_offsetA;
 342             fjptrB             = f+j_coord_offsetB;
 343             fjptrC             = f+j_coord_offsetC;
 344             fjptrD             = f+j_coord_offsetD;
 345
 346             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 347
 348             /* Inner loop uses 111 flops */
 349         }
 350
 351         if(jidx<j_index_end)
 352         {
 353
 354             /* Get j neighbor index, and coordinate index */
 355             jnrlistA         = jjnr[jidx];
 356             jnrlistB         = jjnr[jidx+1];
 357             jnrlistC         = jjnr[jidx+2];
 358             jnrlistD         = jjnr[jidx+3];
 359             /* Sign of each element will be negative for non-real atoms.
 360              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 361              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 362              */
 363             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 364
 365             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 366             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 367             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 368
 369             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 370             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 371             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 372             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 373             j_coord_offsetA  = DIM*jnrA;
 374             j_coord_offsetB  = DIM*jnrB;
 375             j_coord_offsetC  = DIM*jnrC;
 376             j_coord_offsetD  = DIM*jnrD;
 377
 378             /* load j atom coordinates */
 379             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 380                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 381                                                  &jx0,&jy0,&jz0);
 382
 383             /* Calculate displacement vector */
 384             dx00             = _mm256_sub_pd(ix0,jx0);
 385             dy00             = _mm256_sub_pd(iy0,jy0);
 386             dz00             = _mm256_sub_pd(iz0,jz0);
 387             dx10             = _mm256_sub_pd(ix1,jx0);
 388             dy10             = _mm256_sub_pd(iy1,jy0);
 389             dz10             = _mm256_sub_pd(iz1,jz0);
 390             dx20             = _mm256_sub_pd(ix2,jx0);
 391             dy20             = _mm256_sub_pd(iy2,jy0);
 392             dz20             = _mm256_sub_pd(iz2,jz0);
 393
 394             /* Calculate squared distance and things based on it */
 395             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 396             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 397             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 398
 399             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 400             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 401             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 402
 403             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 404             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 405             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 406
 407             /* Load parameters for j particles */
 408             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 409                                                                  charge+jnrC+0,charge+jnrD+0);
 410             vdwjidx0A        = 2*vdwtype[jnrA+0];
 411             vdwjidx0B        = 2*vdwtype[jnrB+0];
 412             vdwjidx0C        = 2*vdwtype[jnrC+0];
 413             vdwjidx0D        = 2*vdwtype[jnrD+0];
 414
 415             fjx0             = _mm256_setzero_pd();
 416             fjy0             = _mm256_setzero_pd();
 417             fjz0             = _mm256_setzero_pd();
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             /* Compute parameters for interactions between i and j atoms */
 424             qq00             = _mm256_mul_pd(iq0,jq0);
 425             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 426                                             vdwioffsetptr0+vdwjidx0B,
 427                                             vdwioffsetptr0+vdwjidx0C,
 428                                             vdwioffsetptr0+vdwjidx0D,
 429                                             &c6_00,&c12_00);
 430
 431             /* REACTION-FIELD ELECTROSTATICS */
 432             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 433             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 434
 435             /* LENNARD-JONES DISPERSION/REPULSION */
 436
 437             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 438             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 439             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 440             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 441             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 442
 443             /* Update potential sum for this i atom from the interaction with this j atom. */
 444             velec            = _mm256_andnot_pd(dummy_mask,velec);
 445             velecsum         = _mm256_add_pd(velecsum,velec);
 446             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 447             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 448
 449             fscal            = _mm256_add_pd(felec,fvdw);
 450
 451             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 452
 453             /* Calculate temporary vectorial force */
 454             tx               = _mm256_mul_pd(fscal,dx00);
 455             ty               = _mm256_mul_pd(fscal,dy00);
 456             tz               = _mm256_mul_pd(fscal,dz00);
 457
 458             /* Update vectorial force */
 459             fix0             = _mm256_add_pd(fix0,tx);
 460             fiy0             = _mm256_add_pd(fiy0,ty);
 461             fiz0             = _mm256_add_pd(fiz0,tz);
 462
 463             fjx0             = _mm256_add_pd(fjx0,tx);
 464             fjy0             = _mm256_add_pd(fjy0,ty);
 465             fjz0             = _mm256_add_pd(fjz0,tz);
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq10             = _mm256_mul_pd(iq1,jq0);
 473
 474             /* REACTION-FIELD ELECTROSTATICS */
 475             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 476             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 477
 478             /* Update potential sum for this i atom from the interaction with this j atom. */
 479             velec            = _mm256_andnot_pd(dummy_mask,velec);
 480             velecsum         = _mm256_add_pd(velecsum,velec);
 481
 482             fscal            = felec;
 483
 484             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 485
 486             /* Calculate temporary vectorial force */
 487             tx               = _mm256_mul_pd(fscal,dx10);
 488             ty               = _mm256_mul_pd(fscal,dy10);
 489             tz               = _mm256_mul_pd(fscal,dz10);
 490
 491             /* Update vectorial force */
 492             fix1             = _mm256_add_pd(fix1,tx);
 493             fiy1             = _mm256_add_pd(fiy1,ty);
 494             fiz1             = _mm256_add_pd(fiz1,tz);
 495
 496             fjx0             = _mm256_add_pd(fjx0,tx);
 497             fjy0             = _mm256_add_pd(fjy0,ty);
 498             fjz0             = _mm256_add_pd(fjz0,tz);
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             qq20             = _mm256_mul_pd(iq2,jq0);
 506
 507             /* REACTION-FIELD ELECTROSTATICS */
 508             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 509             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 510
 511             /* Update potential sum for this i atom from the interaction with this j atom. */
 512             velec            = _mm256_andnot_pd(dummy_mask,velec);
 513             velecsum         = _mm256_add_pd(velecsum,velec);
 514
 515             fscal            = felec;
 516
 517             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 518
 519             /* Calculate temporary vectorial force */
 520             tx               = _mm256_mul_pd(fscal,dx20);
 521             ty               = _mm256_mul_pd(fscal,dy20);
 522             tz               = _mm256_mul_pd(fscal,dz20);
 523
 524             /* Update vectorial force */
 525             fix2             = _mm256_add_pd(fix2,tx);
 526             fiy2             = _mm256_add_pd(fiy2,ty);
 527             fiz2             = _mm256_add_pd(fiz2,tz);
 528
 529             fjx0             = _mm256_add_pd(fjx0,tx);
 530             fjy0             = _mm256_add_pd(fjy0,ty);
 531             fjz0             = _mm256_add_pd(fjz0,tz);
 532
 533             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 534             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 535             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 536             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 537
 538             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 539
 540             /* Inner loop uses 111 flops */
 541         }
 542
 543         /* End of innermost loop */
 544
 545         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 546                                                  f+i_coord_offset,fshift+i_shift_offset);
 547
 548         ggid                        = gid[iidx];
 549         /* Update potential energies */
 550         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 551         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 552
 553         /* Increment number of inner iterations */
 554         inneriter                  += j_index_end - j_index_start;
 555
 556         /* Outer loop uses 20 flops */
 557     }
 558
 559     /* Increment number of outer iterations */
 560     outeriter        += nri;
 561
 562     /* Update outer/inner flops */
 563
 564     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*111);
 565 }
 566 /*
 567  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_256_double
 568  * Electrostatics interaction: ReactionField
 569  * VdW interaction:            LennardJones
 570  * Geometry:                   Water3-Particle
 571  * Calculate force/pot:        Force
 572  */
 573 void
 574 nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_256_double
 575                     (t_nblist                    * gmx_restrict       nlist,
 576                      rvec                        * gmx_restrict          xx,
 577                      rvec                        * gmx_restrict          ff,
 578                      t_forcerec                  * gmx_restrict          fr,
 579                      t_mdatoms                   * gmx_restrict     mdatoms,
 580                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 581                      t_nrnb                      * gmx_restrict        nrnb)
 582 {
 583     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 584      * just 0 for non-waters.
 585      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 586      * jnr indices corresponding to data put in the four positions in the SIMD register.
 587      */
 588     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 589     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 590     int              jnrA,jnrB,jnrC,jnrD;
 591     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 592     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 593     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 594     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 595     real             rcutoff_scalar;
 596     real             *shiftvec,*fshift,*x,*f;
 597     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 598     real             scratch[4*DIM];
 599     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 600     real *           vdwioffsetptr0;
 601     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 602     real *           vdwioffsetptr1;
 603     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 604     real *           vdwioffsetptr2;
 605     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 606     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 607     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 608     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 609     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 610     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 611     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 612     real             *charge;
 613     int              nvdwtype;
 614     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 615     int              *vdwtype;
 616     real             *vdwparam;
 617     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 618     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 619     __m256d          dummy_mask,cutoff_mask;
 620     __m128           tmpmask0,tmpmask1;
 621     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 622     __m256d          one     = _mm256_set1_pd(1.0);
 623     __m256d          two     = _mm256_set1_pd(2.0);
 624     x                = xx[0];
 625     f                = ff[0];
 626
 627     nri              = nlist->nri;
 628     iinr             = nlist->iinr;
 629     jindex           = nlist->jindex;
 630     jjnr             = nlist->jjnr;
 631     shiftidx         = nlist->shift;
 632     gid              = nlist->gid;
 633     shiftvec         = fr->shift_vec[0];
 634     fshift           = fr->fshift[0];
 635     facel            = _mm256_set1_pd(fr->epsfac);
 636     charge           = mdatoms->chargeA;
 637     krf              = _mm256_set1_pd(fr->ic->k_rf);
 638     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 639     crf              = _mm256_set1_pd(fr->ic->c_rf);
 640     nvdwtype         = fr->ntype;
 641     vdwparam         = fr->nbfp;
 642     vdwtype          = mdatoms->typeA;
 643
 644     /* Setup water-specific parameters */
 645     inr              = nlist->iinr[0];
 646     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 647     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 648     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 649     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 650
 651     /* Avoid stupid compiler warnings */
 652     jnrA = jnrB = jnrC = jnrD = 0;
 653     j_coord_offsetA = 0;
 654     j_coord_offsetB = 0;
 655     j_coord_offsetC = 0;
 656     j_coord_offsetD = 0;
 657
 658     outeriter        = 0;
 659     inneriter        = 0;
 660
 661     for(iidx=0;iidx<4*DIM;iidx++)
 662     {
 663         scratch[iidx] = 0.0;
 664     }
 665
 666     /* Start outer loop over neighborlists */
 667     for(iidx=0; iidx<nri; iidx++)
 668     {
 669         /* Load shift vector for this list */
 670         i_shift_offset   = DIM*shiftidx[iidx];
 671
 672         /* Load limits for loop over neighbors */
 673         j_index_start    = jindex[iidx];
 674         j_index_end      = jindex[iidx+1];
 675
 676         /* Get outer coordinate index */
 677         inr              = iinr[iidx];
 678         i_coord_offset   = DIM*inr;
 679
 680         /* Load i particle coords and add shift vector */
 681         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 682                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 683
 684         fix0             = _mm256_setzero_pd();
 685         fiy0             = _mm256_setzero_pd();
 686         fiz0             = _mm256_setzero_pd();
 687         fix1             = _mm256_setzero_pd();
 688         fiy1             = _mm256_setzero_pd();
 689         fiz1             = _mm256_setzero_pd();
 690         fix2             = _mm256_setzero_pd();
 691         fiy2             = _mm256_setzero_pd();
 692         fiz2             = _mm256_setzero_pd();
 693
 694         /* Start inner kernel loop */
 695         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 696         {
 697
 698             /* Get j neighbor index, and coordinate index */
 699             jnrA             = jjnr[jidx];
 700             jnrB             = jjnr[jidx+1];
 701             jnrC             = jjnr[jidx+2];
 702             jnrD             = jjnr[jidx+3];
 703             j_coord_offsetA  = DIM*jnrA;
 704             j_coord_offsetB  = DIM*jnrB;
 705             j_coord_offsetC  = DIM*jnrC;
 706             j_coord_offsetD  = DIM*jnrD;
 707
 708             /* load j atom coordinates */
 709             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 710                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 711                                                  &jx0,&jy0,&jz0);
 712
 713             /* Calculate displacement vector */
 714             dx00             = _mm256_sub_pd(ix0,jx0);
 715             dy00             = _mm256_sub_pd(iy0,jy0);
 716             dz00             = _mm256_sub_pd(iz0,jz0);
 717             dx10             = _mm256_sub_pd(ix1,jx0);
 718             dy10             = _mm256_sub_pd(iy1,jy0);
 719             dz10             = _mm256_sub_pd(iz1,jz0);
 720             dx20             = _mm256_sub_pd(ix2,jx0);
 721             dy20             = _mm256_sub_pd(iy2,jy0);
 722             dz20             = _mm256_sub_pd(iz2,jz0);
 723
 724             /* Calculate squared distance and things based on it */
 725             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 726             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 727             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 728
 729             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 730             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 731             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 732
 733             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 734             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 735             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 736
 737             /* Load parameters for j particles */
 738             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 739                                                                  charge+jnrC+0,charge+jnrD+0);
 740             vdwjidx0A        = 2*vdwtype[jnrA+0];
 741             vdwjidx0B        = 2*vdwtype[jnrB+0];
 742             vdwjidx0C        = 2*vdwtype[jnrC+0];
 743             vdwjidx0D        = 2*vdwtype[jnrD+0];
 744
 745             fjx0             = _mm256_setzero_pd();
 746             fjy0             = _mm256_setzero_pd();
 747             fjz0             = _mm256_setzero_pd();
 748
 749             /**************************
 750              * CALCULATE INTERACTIONS *
 751              **************************/
 752
 753             /* Compute parameters for interactions between i and j atoms */
 754             qq00             = _mm256_mul_pd(iq0,jq0);
 755             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 756                                             vdwioffsetptr0+vdwjidx0B,
 757                                             vdwioffsetptr0+vdwjidx0C,
 758                                             vdwioffsetptr0+vdwjidx0D,
 759                                             &c6_00,&c12_00);
 760
 761             /* REACTION-FIELD ELECTROSTATICS */
 762             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 763
 764             /* LENNARD-JONES DISPERSION/REPULSION */
 765
 766             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 767             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 768
 769             fscal            = _mm256_add_pd(felec,fvdw);
 770
 771             /* Calculate temporary vectorial force */
 772             tx               = _mm256_mul_pd(fscal,dx00);
 773             ty               = _mm256_mul_pd(fscal,dy00);
 774             tz               = _mm256_mul_pd(fscal,dz00);
 775
 776             /* Update vectorial force */
 777             fix0             = _mm256_add_pd(fix0,tx);
 778             fiy0             = _mm256_add_pd(fiy0,ty);
 779             fiz0             = _mm256_add_pd(fiz0,tz);
 780
 781             fjx0             = _mm256_add_pd(fjx0,tx);
 782             fjy0             = _mm256_add_pd(fjy0,ty);
 783             fjz0             = _mm256_add_pd(fjz0,tz);
 784
 785             /**************************
 786              * CALCULATE INTERACTIONS *
 787              **************************/
 788
 789             /* Compute parameters for interactions between i and j atoms */
 790             qq10             = _mm256_mul_pd(iq1,jq0);
 791
 792             /* REACTION-FIELD ELECTROSTATICS */
 793             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 794
 795             fscal            = felec;
 796
 797             /* Calculate temporary vectorial force */
 798             tx               = _mm256_mul_pd(fscal,dx10);
 799             ty               = _mm256_mul_pd(fscal,dy10);
 800             tz               = _mm256_mul_pd(fscal,dz10);
 801
 802             /* Update vectorial force */
 803             fix1             = _mm256_add_pd(fix1,tx);
 804             fiy1             = _mm256_add_pd(fiy1,ty);
 805             fiz1             = _mm256_add_pd(fiz1,tz);
 806
 807             fjx0             = _mm256_add_pd(fjx0,tx);
 808             fjy0             = _mm256_add_pd(fjy0,ty);
 809             fjz0             = _mm256_add_pd(fjz0,tz);
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             /* Compute parameters for interactions between i and j atoms */
 816             qq20             = _mm256_mul_pd(iq2,jq0);
 817
 818             /* REACTION-FIELD ELECTROSTATICS */
 819             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 820
 821             fscal            = felec;
 822
 823             /* Calculate temporary vectorial force */
 824             tx               = _mm256_mul_pd(fscal,dx20);
 825             ty               = _mm256_mul_pd(fscal,dy20);
 826             tz               = _mm256_mul_pd(fscal,dz20);
 827
 828             /* Update vectorial force */
 829             fix2             = _mm256_add_pd(fix2,tx);
 830             fiy2             = _mm256_add_pd(fiy2,ty);
 831             fiz2             = _mm256_add_pd(fiz2,tz);
 832
 833             fjx0             = _mm256_add_pd(fjx0,tx);
 834             fjy0             = _mm256_add_pd(fjy0,ty);
 835             fjz0             = _mm256_add_pd(fjz0,tz);
 836
 837             fjptrA             = f+j_coord_offsetA;
 838             fjptrB             = f+j_coord_offsetB;
 839             fjptrC             = f+j_coord_offsetC;
 840             fjptrD             = f+j_coord_offsetD;
 841
 842             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 843
 844             /* Inner loop uses 91 flops */
 845         }
 846
 847         if(jidx<j_index_end)
 848         {
 849
 850             /* Get j neighbor index, and coordinate index */
 851             jnrlistA         = jjnr[jidx];
 852             jnrlistB         = jjnr[jidx+1];
 853             jnrlistC         = jjnr[jidx+2];
 854             jnrlistD         = jjnr[jidx+3];
 855             /* Sign of each element will be negative for non-real atoms.
 856              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 857              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 858              */
 859             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 860
 861             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 862             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 863             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 864
 865             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 866             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 867             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 868             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 869             j_coord_offsetA  = DIM*jnrA;
 870             j_coord_offsetB  = DIM*jnrB;
 871             j_coord_offsetC  = DIM*jnrC;
 872             j_coord_offsetD  = DIM*jnrD;
 873
 874             /* load j atom coordinates */
 875             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 876                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 877                                                  &jx0,&jy0,&jz0);
 878
 879             /* Calculate displacement vector */
 880             dx00             = _mm256_sub_pd(ix0,jx0);
 881             dy00             = _mm256_sub_pd(iy0,jy0);
 882             dz00             = _mm256_sub_pd(iz0,jz0);
 883             dx10             = _mm256_sub_pd(ix1,jx0);
 884             dy10             = _mm256_sub_pd(iy1,jy0);
 885             dz10             = _mm256_sub_pd(iz1,jz0);
 886             dx20             = _mm256_sub_pd(ix2,jx0);
 887             dy20             = _mm256_sub_pd(iy2,jy0);
 888             dz20             = _mm256_sub_pd(iz2,jz0);
 889
 890             /* Calculate squared distance and things based on it */
 891             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 892             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 893             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 894
 895             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 896             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 897             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 898
 899             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 900             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 901             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 902
 903             /* Load parameters for j particles */
 904             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 905                                                                  charge+jnrC+0,charge+jnrD+0);
 906             vdwjidx0A        = 2*vdwtype[jnrA+0];
 907             vdwjidx0B        = 2*vdwtype[jnrB+0];
 908             vdwjidx0C        = 2*vdwtype[jnrC+0];
 909             vdwjidx0D        = 2*vdwtype[jnrD+0];
 910
 911             fjx0             = _mm256_setzero_pd();
 912             fjy0             = _mm256_setzero_pd();
 913             fjz0             = _mm256_setzero_pd();
 914
 915             /**************************
 916              * CALCULATE INTERACTIONS *
 917              **************************/
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq00             = _mm256_mul_pd(iq0,jq0);
 921             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 922                                             vdwioffsetptr0+vdwjidx0B,
 923                                             vdwioffsetptr0+vdwjidx0C,
 924                                             vdwioffsetptr0+vdwjidx0D,
 925                                             &c6_00,&c12_00);
 926
 927             /* REACTION-FIELD ELECTROSTATICS */
 928             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 929
 930             /* LENNARD-JONES DISPERSION/REPULSION */
 931
 932             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 933             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 934
 935             fscal            = _mm256_add_pd(felec,fvdw);
 936
 937             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 938
 939             /* Calculate temporary vectorial force */
 940             tx               = _mm256_mul_pd(fscal,dx00);
 941             ty               = _mm256_mul_pd(fscal,dy00);
 942             tz               = _mm256_mul_pd(fscal,dz00);
 943
 944             /* Update vectorial force */
 945             fix0             = _mm256_add_pd(fix0,tx);
 946             fiy0             = _mm256_add_pd(fiy0,ty);
 947             fiz0             = _mm256_add_pd(fiz0,tz);
 948
 949             fjx0             = _mm256_add_pd(fjx0,tx);
 950             fjy0             = _mm256_add_pd(fjy0,ty);
 951             fjz0             = _mm256_add_pd(fjz0,tz);
 952
 953             /**************************
 954              * CALCULATE INTERACTIONS *
 955              **************************/
 956
 957             /* Compute parameters for interactions between i and j atoms */
 958             qq10             = _mm256_mul_pd(iq1,jq0);
 959
 960             /* REACTION-FIELD ELECTROSTATICS */
 961             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 962
 963             fscal            = felec;
 964
 965             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 966
 967             /* Calculate temporary vectorial force */
 968             tx               = _mm256_mul_pd(fscal,dx10);
 969             ty               = _mm256_mul_pd(fscal,dy10);
 970             tz               = _mm256_mul_pd(fscal,dz10);
 971
 972             /* Update vectorial force */
 973             fix1             = _mm256_add_pd(fix1,tx);
 974             fiy1             = _mm256_add_pd(fiy1,ty);
 975             fiz1             = _mm256_add_pd(fiz1,tz);
 976
 977             fjx0             = _mm256_add_pd(fjx0,tx);
 978             fjy0             = _mm256_add_pd(fjy0,ty);
 979             fjz0             = _mm256_add_pd(fjz0,tz);
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             /* Compute parameters for interactions between i and j atoms */
 986             qq20             = _mm256_mul_pd(iq2,jq0);
 987
 988             /* REACTION-FIELD ELECTROSTATICS */
 989             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 990
 991             fscal            = felec;
 992
 993             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 994
 995             /* Calculate temporary vectorial force */
 996             tx               = _mm256_mul_pd(fscal,dx20);
 997             ty               = _mm256_mul_pd(fscal,dy20);
 998             tz               = _mm256_mul_pd(fscal,dz20);
 999
1000             /* Update vectorial force */
1001             fix2             = _mm256_add_pd(fix2,tx);
1002             fiy2             = _mm256_add_pd(fiy2,ty);
1003             fiz2             = _mm256_add_pd(fiz2,tz);
1004
1005             fjx0             = _mm256_add_pd(fjx0,tx);
1006             fjy0             = _mm256_add_pd(fjy0,ty);
1007             fjz0             = _mm256_add_pd(fjz0,tz);
1008
1009             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1010             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1011             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1012             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1013
1014             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1015
1016             /* Inner loop uses 91 flops */
1017         }
1018
1019         /* End of innermost loop */
1020
1021         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1022                                                  f+i_coord_offset,fshift+i_shift_offset);
1023
1024         /* Increment number of inner iterations */
1025         inneriter                  += j_index_end - j_index_start;
1026
1027         /* Outer loop uses 18 flops */
1028     }
1029
1030     /* Increment number of outer iterations */
1031     outeriter        += nri;
1032
1033     /* Update outer/inner flops */
1034
1035     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*91);
1036 }