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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m256d          dummy_mask,cutoff_mask;
 106     __m128           tmpmask0,tmpmask1;
 107     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 108     __m256d          one     = _mm256_set1_pd(1.0);
 109     __m256d          two     = _mm256_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_pd(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm256_set1_pd(fr->ic->k_rf);
 124     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 125     crf              = _mm256_set1_pd(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 133     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 134     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 135     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 136
 137     /* Avoid stupid compiler warnings */
 138     jnrA = jnrB = jnrC = jnrD = 0;
 139     j_coord_offsetA = 0;
 140     j_coord_offsetB = 0;
 141     j_coord_offsetC = 0;
 142     j_coord_offsetD = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     for(iidx=0;iidx<4*DIM;iidx++)
 148     {
 149         scratch[iidx] = 0.0;
 150     }
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm256_setzero_pd();
 171         fiy0             = _mm256_setzero_pd();
 172         fiz0             = _mm256_setzero_pd();
 173         fix1             = _mm256_setzero_pd();
 174         fiy1             = _mm256_setzero_pd();
 175         fiz1             = _mm256_setzero_pd();
 176         fix2             = _mm256_setzero_pd();
 177         fiy2             = _mm256_setzero_pd();
 178         fiz2             = _mm256_setzero_pd();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm256_setzero_pd();
 182         vvdwsum          = _mm256_setzero_pd();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 201                                                  &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm256_sub_pd(ix0,jx0);
 205             dy00             = _mm256_sub_pd(iy0,jy0);
 206             dz00             = _mm256_sub_pd(iz0,jz0);
 207             dx10             = _mm256_sub_pd(ix1,jx0);
 208             dy10             = _mm256_sub_pd(iy1,jy0);
 209             dz10             = _mm256_sub_pd(iz1,jz0);
 210             dx20             = _mm256_sub_pd(ix2,jx0);
 211             dy20             = _mm256_sub_pd(iy2,jy0);
 212             dz20             = _mm256_sub_pd(iz2,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 216             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 217             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 218
 219             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 220             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 221             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 222
 223             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 224             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 225             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 229                                                                  charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             fjx0             = _mm256_setzero_pd();
 236             fjy0             = _mm256_setzero_pd();
 237             fjz0             = _mm256_setzero_pd();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm256_mul_pd(iq0,jq0);
 245             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 246                                             vdwioffsetptr0+vdwjidx0B,
 247                                             vdwioffsetptr0+vdwjidx0C,
 248                                             vdwioffsetptr0+vdwjidx0D,
 249                                             &c6_00,&c12_00);
 250
 251             /* REACTION-FIELD ELECTROSTATICS */
 252             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 253             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 254
 255             /* LENNARD-JONES DISPERSION/REPULSION */
 256
 257             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 258             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 259             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 260             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 261             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velecsum         = _mm256_add_pd(velecsum,velec);
 265             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 266
 267             fscal            = _mm256_add_pd(felec,fvdw);
 268
 269             /* Calculate temporary vectorial force */
 270             tx               = _mm256_mul_pd(fscal,dx00);
 271             ty               = _mm256_mul_pd(fscal,dy00);
 272             tz               = _mm256_mul_pd(fscal,dz00);
 273
 274             /* Update vectorial force */
 275             fix0             = _mm256_add_pd(fix0,tx);
 276             fiy0             = _mm256_add_pd(fiy0,ty);
 277             fiz0             = _mm256_add_pd(fiz0,tz);
 278
 279             fjx0             = _mm256_add_pd(fjx0,tx);
 280             fjy0             = _mm256_add_pd(fjy0,ty);
 281             fjz0             = _mm256_add_pd(fjz0,tz);
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq10             = _mm256_mul_pd(iq1,jq0);
 289
 290             /* REACTION-FIELD ELECTROSTATICS */
 291             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 292             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             velecsum         = _mm256_add_pd(velecsum,velec);
 296
 297             fscal            = felec;
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm256_mul_pd(fscal,dx10);
 301             ty               = _mm256_mul_pd(fscal,dy10);
 302             tz               = _mm256_mul_pd(fscal,dz10);
 303
 304             /* Update vectorial force */
 305             fix1             = _mm256_add_pd(fix1,tx);
 306             fiy1             = _mm256_add_pd(fiy1,ty);
 307             fiz1             = _mm256_add_pd(fiz1,tz);
 308
 309             fjx0             = _mm256_add_pd(fjx0,tx);
 310             fjy0             = _mm256_add_pd(fjy0,ty);
 311             fjz0             = _mm256_add_pd(fjz0,tz);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq20             = _mm256_mul_pd(iq2,jq0);
 319
 320             /* REACTION-FIELD ELECTROSTATICS */
 321             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 322             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velecsum         = _mm256_add_pd(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm256_mul_pd(fscal,dx20);
 331             ty               = _mm256_mul_pd(fscal,dy20);
 332             tz               = _mm256_mul_pd(fscal,dz20);
 333
 334             /* Update vectorial force */
 335             fix2             = _mm256_add_pd(fix2,tx);
 336             fiy2             = _mm256_add_pd(fiy2,ty);
 337             fiz2             = _mm256_add_pd(fiz2,tz);
 338
 339             fjx0             = _mm256_add_pd(fjx0,tx);
 340             fjy0             = _mm256_add_pd(fjy0,ty);
 341             fjz0             = _mm256_add_pd(fjz0,tz);
 342
 343             fjptrA             = f+j_coord_offsetA;
 344             fjptrB             = f+j_coord_offsetB;
 345             fjptrC             = f+j_coord_offsetC;
 346             fjptrD             = f+j_coord_offsetD;
 347
 348             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 349
 350             /* Inner loop uses 111 flops */
 351         }
 352
 353         if(jidx<j_index_end)
 354         {
 355
 356             /* Get j neighbor index, and coordinate index */
 357             jnrlistA         = jjnr[jidx];
 358             jnrlistB         = jjnr[jidx+1];
 359             jnrlistC         = jjnr[jidx+2];
 360             jnrlistD         = jjnr[jidx+3];
 361             /* Sign of each element will be negative for non-real atoms.
 362              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 363              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 364              */
 365             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 366
 367             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 368             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 369             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 370
 371             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 372             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 373             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 374             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 375             j_coord_offsetA  = DIM*jnrA;
 376             j_coord_offsetB  = DIM*jnrB;
 377             j_coord_offsetC  = DIM*jnrC;
 378             j_coord_offsetD  = DIM*jnrD;
 379
 380             /* load j atom coordinates */
 381             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 382                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 383                                                  &jx0,&jy0,&jz0);
 384
 385             /* Calculate displacement vector */
 386             dx00             = _mm256_sub_pd(ix0,jx0);
 387             dy00             = _mm256_sub_pd(iy0,jy0);
 388             dz00             = _mm256_sub_pd(iz0,jz0);
 389             dx10             = _mm256_sub_pd(ix1,jx0);
 390             dy10             = _mm256_sub_pd(iy1,jy0);
 391             dz10             = _mm256_sub_pd(iz1,jz0);
 392             dx20             = _mm256_sub_pd(ix2,jx0);
 393             dy20             = _mm256_sub_pd(iy2,jy0);
 394             dz20             = _mm256_sub_pd(iz2,jz0);
 395
 396             /* Calculate squared distance and things based on it */
 397             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 398             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 399             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 400
 401             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 402             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 403             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 404
 405             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 406             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 407             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 408
 409             /* Load parameters for j particles */
 410             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 411                                                                  charge+jnrC+0,charge+jnrD+0);
 412             vdwjidx0A        = 2*vdwtype[jnrA+0];
 413             vdwjidx0B        = 2*vdwtype[jnrB+0];
 414             vdwjidx0C        = 2*vdwtype[jnrC+0];
 415             vdwjidx0D        = 2*vdwtype[jnrD+0];
 416
 417             fjx0             = _mm256_setzero_pd();
 418             fjy0             = _mm256_setzero_pd();
 419             fjz0             = _mm256_setzero_pd();
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             /* Compute parameters for interactions between i and j atoms */
 426             qq00             = _mm256_mul_pd(iq0,jq0);
 427             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 428                                             vdwioffsetptr0+vdwjidx0B,
 429                                             vdwioffsetptr0+vdwjidx0C,
 430                                             vdwioffsetptr0+vdwjidx0D,
 431                                             &c6_00,&c12_00);
 432
 433             /* REACTION-FIELD ELECTROSTATICS */
 434             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_add_pd(rinv00,_mm256_mul_pd(krf,rsq00)),crf));
 435             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 436
 437             /* LENNARD-JONES DISPERSION/REPULSION */
 438
 439             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 440             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 441             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 442             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 443             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 444
 445             /* Update potential sum for this i atom from the interaction with this j atom. */
 446             velec            = _mm256_andnot_pd(dummy_mask,velec);
 447             velecsum         = _mm256_add_pd(velecsum,velec);
 448             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 449             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 450
 451             fscal            = _mm256_add_pd(felec,fvdw);
 452
 453             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 454
 455             /* Calculate temporary vectorial force */
 456             tx               = _mm256_mul_pd(fscal,dx00);
 457             ty               = _mm256_mul_pd(fscal,dy00);
 458             tz               = _mm256_mul_pd(fscal,dz00);
 459
 460             /* Update vectorial force */
 461             fix0             = _mm256_add_pd(fix0,tx);
 462             fiy0             = _mm256_add_pd(fiy0,ty);
 463             fiz0             = _mm256_add_pd(fiz0,tz);
 464
 465             fjx0             = _mm256_add_pd(fjx0,tx);
 466             fjy0             = _mm256_add_pd(fjy0,ty);
 467             fjz0             = _mm256_add_pd(fjz0,tz);
 468
 469             /**************************
 470              * CALCULATE INTERACTIONS *
 471              **************************/
 472
 473             /* Compute parameters for interactions between i and j atoms */
 474             qq10             = _mm256_mul_pd(iq1,jq0);
 475
 476             /* REACTION-FIELD ELECTROSTATICS */
 477             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 478             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             velec            = _mm256_andnot_pd(dummy_mask,velec);
 482             velecsum         = _mm256_add_pd(velecsum,velec);
 483
 484             fscal            = felec;
 485
 486             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 487
 488             /* Calculate temporary vectorial force */
 489             tx               = _mm256_mul_pd(fscal,dx10);
 490             ty               = _mm256_mul_pd(fscal,dy10);
 491             tz               = _mm256_mul_pd(fscal,dz10);
 492
 493             /* Update vectorial force */
 494             fix1             = _mm256_add_pd(fix1,tx);
 495             fiy1             = _mm256_add_pd(fiy1,ty);
 496             fiz1             = _mm256_add_pd(fiz1,tz);
 497
 498             fjx0             = _mm256_add_pd(fjx0,tx);
 499             fjy0             = _mm256_add_pd(fjy0,ty);
 500             fjz0             = _mm256_add_pd(fjz0,tz);
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             /* Compute parameters for interactions between i and j atoms */
 507             qq20             = _mm256_mul_pd(iq2,jq0);
 508
 509             /* REACTION-FIELD ELECTROSTATICS */
 510             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 511             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             velec            = _mm256_andnot_pd(dummy_mask,velec);
 515             velecsum         = _mm256_add_pd(velecsum,velec);
 516
 517             fscal            = felec;
 518
 519             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm256_mul_pd(fscal,dx20);
 523             ty               = _mm256_mul_pd(fscal,dy20);
 524             tz               = _mm256_mul_pd(fscal,dz20);
 525
 526             /* Update vectorial force */
 527             fix2             = _mm256_add_pd(fix2,tx);
 528             fiy2             = _mm256_add_pd(fiy2,ty);
 529             fiz2             = _mm256_add_pd(fiz2,tz);
 530
 531             fjx0             = _mm256_add_pd(fjx0,tx);
 532             fjy0             = _mm256_add_pd(fjy0,ty);
 533             fjz0             = _mm256_add_pd(fjz0,tz);
 534
 535             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 536             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 537             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 538             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 539
 540             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 541
 542             /* Inner loop uses 111 flops */
 543         }
 544
 545         /* End of innermost loop */
 546
 547         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 548                                                  f+i_coord_offset,fshift+i_shift_offset);
 549
 550         ggid                        = gid[iidx];
 551         /* Update potential energies */
 552         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 553         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 554
 555         /* Increment number of inner iterations */
 556         inneriter                  += j_index_end - j_index_start;
 557
 558         /* Outer loop uses 20 flops */
 559     }
 560
 561     /* Increment number of outer iterations */
 562     outeriter        += nri;
 563
 564     /* Update outer/inner flops */
 565
 566     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*111);
 567 }
 568 /*
 569  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_256_double
 570  * Electrostatics interaction: ReactionField
 571  * VdW interaction:            LennardJones
 572  * Geometry:                   Water3-Particle
 573  * Calculate force/pot:        Force
 574  */
 575 void
 576 nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_256_double
 577                     (t_nblist                    * gmx_restrict       nlist,
 578                      rvec                        * gmx_restrict          xx,
 579                      rvec                        * gmx_restrict          ff,
 580                      t_forcerec                  * gmx_restrict          fr,
 581                      t_mdatoms                   * gmx_restrict     mdatoms,
 582                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 583                      t_nrnb                      * gmx_restrict        nrnb)
 584 {
 585     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 586      * just 0 for non-waters.
 587      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 588      * jnr indices corresponding to data put in the four positions in the SIMD register.
 589      */
 590     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 591     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 592     int              jnrA,jnrB,jnrC,jnrD;
 593     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 594     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 595     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 596     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 597     real             rcutoff_scalar;
 598     real             *shiftvec,*fshift,*x,*f;
 599     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 600     real             scratch[4*DIM];
 601     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 602     real *           vdwioffsetptr0;
 603     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 604     real *           vdwioffsetptr1;
 605     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 606     real *           vdwioffsetptr2;
 607     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 608     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 609     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 610     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 611     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 612     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 613     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 614     real             *charge;
 615     int              nvdwtype;
 616     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 617     int              *vdwtype;
 618     real             *vdwparam;
 619     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 620     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 621     __m256d          dummy_mask,cutoff_mask;
 622     __m128           tmpmask0,tmpmask1;
 623     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 624     __m256d          one     = _mm256_set1_pd(1.0);
 625     __m256d          two     = _mm256_set1_pd(2.0);
 626     x                = xx[0];
 627     f                = ff[0];
 628
 629     nri              = nlist->nri;
 630     iinr             = nlist->iinr;
 631     jindex           = nlist->jindex;
 632     jjnr             = nlist->jjnr;
 633     shiftidx         = nlist->shift;
 634     gid              = nlist->gid;
 635     shiftvec         = fr->shift_vec[0];
 636     fshift           = fr->fshift[0];
 637     facel            = _mm256_set1_pd(fr->epsfac);
 638     charge           = mdatoms->chargeA;
 639     krf              = _mm256_set1_pd(fr->ic->k_rf);
 640     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 641     crf              = _mm256_set1_pd(fr->ic->c_rf);
 642     nvdwtype         = fr->ntype;
 643     vdwparam         = fr->nbfp;
 644     vdwtype          = mdatoms->typeA;
 645
 646     /* Setup water-specific parameters */
 647     inr              = nlist->iinr[0];
 648     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 649     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 650     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 651     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 652
 653     /* Avoid stupid compiler warnings */
 654     jnrA = jnrB = jnrC = jnrD = 0;
 655     j_coord_offsetA = 0;
 656     j_coord_offsetB = 0;
 657     j_coord_offsetC = 0;
 658     j_coord_offsetD = 0;
 659
 660     outeriter        = 0;
 661     inneriter        = 0;
 662
 663     for(iidx=0;iidx<4*DIM;iidx++)
 664     {
 665         scratch[iidx] = 0.0;
 666     }
 667
 668     /* Start outer loop over neighborlists */
 669     for(iidx=0; iidx<nri; iidx++)
 670     {
 671         /* Load shift vector for this list */
 672         i_shift_offset   = DIM*shiftidx[iidx];
 673
 674         /* Load limits for loop over neighbors */
 675         j_index_start    = jindex[iidx];
 676         j_index_end      = jindex[iidx+1];
 677
 678         /* Get outer coordinate index */
 679         inr              = iinr[iidx];
 680         i_coord_offset   = DIM*inr;
 681
 682         /* Load i particle coords and add shift vector */
 683         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 684                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 685
 686         fix0             = _mm256_setzero_pd();
 687         fiy0             = _mm256_setzero_pd();
 688         fiz0             = _mm256_setzero_pd();
 689         fix1             = _mm256_setzero_pd();
 690         fiy1             = _mm256_setzero_pd();
 691         fiz1             = _mm256_setzero_pd();
 692         fix2             = _mm256_setzero_pd();
 693         fiy2             = _mm256_setzero_pd();
 694         fiz2             = _mm256_setzero_pd();
 695
 696         /* Start inner kernel loop */
 697         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 698         {
 699
 700             /* Get j neighbor index, and coordinate index */
 701             jnrA             = jjnr[jidx];
 702             jnrB             = jjnr[jidx+1];
 703             jnrC             = jjnr[jidx+2];
 704             jnrD             = jjnr[jidx+3];
 705             j_coord_offsetA  = DIM*jnrA;
 706             j_coord_offsetB  = DIM*jnrB;
 707             j_coord_offsetC  = DIM*jnrC;
 708             j_coord_offsetD  = DIM*jnrD;
 709
 710             /* load j atom coordinates */
 711             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 712                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 713                                                  &jx0,&jy0,&jz0);
 714
 715             /* Calculate displacement vector */
 716             dx00             = _mm256_sub_pd(ix0,jx0);
 717             dy00             = _mm256_sub_pd(iy0,jy0);
 718             dz00             = _mm256_sub_pd(iz0,jz0);
 719             dx10             = _mm256_sub_pd(ix1,jx0);
 720             dy10             = _mm256_sub_pd(iy1,jy0);
 721             dz10             = _mm256_sub_pd(iz1,jz0);
 722             dx20             = _mm256_sub_pd(ix2,jx0);
 723             dy20             = _mm256_sub_pd(iy2,jy0);
 724             dz20             = _mm256_sub_pd(iz2,jz0);
 725
 726             /* Calculate squared distance and things based on it */
 727             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 728             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 729             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 730
 731             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 732             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 733             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 734
 735             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 736             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 737             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 738
 739             /* Load parameters for j particles */
 740             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 741                                                                  charge+jnrC+0,charge+jnrD+0);
 742             vdwjidx0A        = 2*vdwtype[jnrA+0];
 743             vdwjidx0B        = 2*vdwtype[jnrB+0];
 744             vdwjidx0C        = 2*vdwtype[jnrC+0];
 745             vdwjidx0D        = 2*vdwtype[jnrD+0];
 746
 747             fjx0             = _mm256_setzero_pd();
 748             fjy0             = _mm256_setzero_pd();
 749             fjz0             = _mm256_setzero_pd();
 750
 751             /**************************
 752              * CALCULATE INTERACTIONS *
 753              **************************/
 754
 755             /* Compute parameters for interactions between i and j atoms */
 756             qq00             = _mm256_mul_pd(iq0,jq0);
 757             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 758                                             vdwioffsetptr0+vdwjidx0B,
 759                                             vdwioffsetptr0+vdwjidx0C,
 760                                             vdwioffsetptr0+vdwjidx0D,
 761                                             &c6_00,&c12_00);
 762
 763             /* REACTION-FIELD ELECTROSTATICS */
 764             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 765
 766             /* LENNARD-JONES DISPERSION/REPULSION */
 767
 768             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 769             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 770
 771             fscal            = _mm256_add_pd(felec,fvdw);
 772
 773             /* Calculate temporary vectorial force */
 774             tx               = _mm256_mul_pd(fscal,dx00);
 775             ty               = _mm256_mul_pd(fscal,dy00);
 776             tz               = _mm256_mul_pd(fscal,dz00);
 777
 778             /* Update vectorial force */
 779             fix0             = _mm256_add_pd(fix0,tx);
 780             fiy0             = _mm256_add_pd(fiy0,ty);
 781             fiz0             = _mm256_add_pd(fiz0,tz);
 782
 783             fjx0             = _mm256_add_pd(fjx0,tx);
 784             fjy0             = _mm256_add_pd(fjy0,ty);
 785             fjz0             = _mm256_add_pd(fjz0,tz);
 786
 787             /**************************
 788              * CALCULATE INTERACTIONS *
 789              **************************/
 790
 791             /* Compute parameters for interactions between i and j atoms */
 792             qq10             = _mm256_mul_pd(iq1,jq0);
 793
 794             /* REACTION-FIELD ELECTROSTATICS */
 795             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 796
 797             fscal            = felec;
 798
 799             /* Calculate temporary vectorial force */
 800             tx               = _mm256_mul_pd(fscal,dx10);
 801             ty               = _mm256_mul_pd(fscal,dy10);
 802             tz               = _mm256_mul_pd(fscal,dz10);
 803
 804             /* Update vectorial force */
 805             fix1             = _mm256_add_pd(fix1,tx);
 806             fiy1             = _mm256_add_pd(fiy1,ty);
 807             fiz1             = _mm256_add_pd(fiz1,tz);
 808
 809             fjx0             = _mm256_add_pd(fjx0,tx);
 810             fjy0             = _mm256_add_pd(fjy0,ty);
 811             fjz0             = _mm256_add_pd(fjz0,tz);
 812
 813             /**************************
 814              * CALCULATE INTERACTIONS *
 815              **************************/
 816
 817             /* Compute parameters for interactions between i and j atoms */
 818             qq20             = _mm256_mul_pd(iq2,jq0);
 819
 820             /* REACTION-FIELD ELECTROSTATICS */
 821             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 822
 823             fscal            = felec;
 824
 825             /* Calculate temporary vectorial force */
 826             tx               = _mm256_mul_pd(fscal,dx20);
 827             ty               = _mm256_mul_pd(fscal,dy20);
 828             tz               = _mm256_mul_pd(fscal,dz20);
 829
 830             /* Update vectorial force */
 831             fix2             = _mm256_add_pd(fix2,tx);
 832             fiy2             = _mm256_add_pd(fiy2,ty);
 833             fiz2             = _mm256_add_pd(fiz2,tz);
 834
 835             fjx0             = _mm256_add_pd(fjx0,tx);
 836             fjy0             = _mm256_add_pd(fjy0,ty);
 837             fjz0             = _mm256_add_pd(fjz0,tz);
 838
 839             fjptrA             = f+j_coord_offsetA;
 840             fjptrB             = f+j_coord_offsetB;
 841             fjptrC             = f+j_coord_offsetC;
 842             fjptrD             = f+j_coord_offsetD;
 843
 844             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 845
 846             /* Inner loop uses 91 flops */
 847         }
 848
 849         if(jidx<j_index_end)
 850         {
 851
 852             /* Get j neighbor index, and coordinate index */
 853             jnrlistA         = jjnr[jidx];
 854             jnrlistB         = jjnr[jidx+1];
 855             jnrlistC         = jjnr[jidx+2];
 856             jnrlistD         = jjnr[jidx+3];
 857             /* Sign of each element will be negative for non-real atoms.
 858              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 859              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 860              */
 861             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 862
 863             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 864             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 865             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 866
 867             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 868             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 869             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 870             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 871             j_coord_offsetA  = DIM*jnrA;
 872             j_coord_offsetB  = DIM*jnrB;
 873             j_coord_offsetC  = DIM*jnrC;
 874             j_coord_offsetD  = DIM*jnrD;
 875
 876             /* load j atom coordinates */
 877             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 878                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 879                                                  &jx0,&jy0,&jz0);
 880
 881             /* Calculate displacement vector */
 882             dx00             = _mm256_sub_pd(ix0,jx0);
 883             dy00             = _mm256_sub_pd(iy0,jy0);
 884             dz00             = _mm256_sub_pd(iz0,jz0);
 885             dx10             = _mm256_sub_pd(ix1,jx0);
 886             dy10             = _mm256_sub_pd(iy1,jy0);
 887             dz10             = _mm256_sub_pd(iz1,jz0);
 888             dx20             = _mm256_sub_pd(ix2,jx0);
 889             dy20             = _mm256_sub_pd(iy2,jy0);
 890             dz20             = _mm256_sub_pd(iz2,jz0);
 891
 892             /* Calculate squared distance and things based on it */
 893             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 894             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 895             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 896
 897             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 898             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 899             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 900
 901             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 902             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 903             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 904
 905             /* Load parameters for j particles */
 906             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 907                                                                  charge+jnrC+0,charge+jnrD+0);
 908             vdwjidx0A        = 2*vdwtype[jnrA+0];
 909             vdwjidx0B        = 2*vdwtype[jnrB+0];
 910             vdwjidx0C        = 2*vdwtype[jnrC+0];
 911             vdwjidx0D        = 2*vdwtype[jnrD+0];
 912
 913             fjx0             = _mm256_setzero_pd();
 914             fjy0             = _mm256_setzero_pd();
 915             fjz0             = _mm256_setzero_pd();
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             /* Compute parameters for interactions between i and j atoms */
 922             qq00             = _mm256_mul_pd(iq0,jq0);
 923             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 924                                             vdwioffsetptr0+vdwjidx0B,
 925                                             vdwioffsetptr0+vdwjidx0C,
 926                                             vdwioffsetptr0+vdwjidx0D,
 927                                             &c6_00,&c12_00);
 928
 929             /* REACTION-FIELD ELECTROSTATICS */
 930             felec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_mul_pd(rinv00,rinvsq00),krf2));
 931
 932             /* LENNARD-JONES DISPERSION/REPULSION */
 933
 934             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 935             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 936
 937             fscal            = _mm256_add_pd(felec,fvdw);
 938
 939             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 940
 941             /* Calculate temporary vectorial force */
 942             tx               = _mm256_mul_pd(fscal,dx00);
 943             ty               = _mm256_mul_pd(fscal,dy00);
 944             tz               = _mm256_mul_pd(fscal,dz00);
 945
 946             /* Update vectorial force */
 947             fix0             = _mm256_add_pd(fix0,tx);
 948             fiy0             = _mm256_add_pd(fiy0,ty);
 949             fiz0             = _mm256_add_pd(fiz0,tz);
 950
 951             fjx0             = _mm256_add_pd(fjx0,tx);
 952             fjy0             = _mm256_add_pd(fjy0,ty);
 953             fjz0             = _mm256_add_pd(fjz0,tz);
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             qq10             = _mm256_mul_pd(iq1,jq0);
 961
 962             /* REACTION-FIELD ELECTROSTATICS */
 963             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 964
 965             fscal            = felec;
 966
 967             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 968
 969             /* Calculate temporary vectorial force */
 970             tx               = _mm256_mul_pd(fscal,dx10);
 971             ty               = _mm256_mul_pd(fscal,dy10);
 972             tz               = _mm256_mul_pd(fscal,dz10);
 973
 974             /* Update vectorial force */
 975             fix1             = _mm256_add_pd(fix1,tx);
 976             fiy1             = _mm256_add_pd(fiy1,ty);
 977             fiz1             = _mm256_add_pd(fiz1,tz);
 978
 979             fjx0             = _mm256_add_pd(fjx0,tx);
 980             fjy0             = _mm256_add_pd(fjy0,ty);
 981             fjz0             = _mm256_add_pd(fjz0,tz);
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             qq20             = _mm256_mul_pd(iq2,jq0);
 989
 990             /* REACTION-FIELD ELECTROSTATICS */
 991             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 992
 993             fscal            = felec;
 994
 995             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 996
 997             /* Calculate temporary vectorial force */
 998             tx               = _mm256_mul_pd(fscal,dx20);
 999             ty               = _mm256_mul_pd(fscal,dy20);
1000             tz               = _mm256_mul_pd(fscal,dz20);
1001
1002             /* Update vectorial force */
1003             fix2             = _mm256_add_pd(fix2,tx);
1004             fiy2             = _mm256_add_pd(fiy2,ty);
1005             fiz2             = _mm256_add_pd(fiz2,tz);
1006
1007             fjx0             = _mm256_add_pd(fjx0,tx);
1008             fjy0             = _mm256_add_pd(fjy0,ty);
1009             fjz0             = _mm256_add_pd(fjz0,tz);
1010
1011             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1012             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1013             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1014             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1015
1016             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1017
1018             /* Inner loop uses 91 flops */
1019         }
1020
1021         /* End of innermost loop */
1022
1023         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1024                                                  f+i_coord_offset,fshift+i_shift_offset);
1025
1026         /* Increment number of inner iterations */
1027         inneriter                  += j_index_end - j_index_start;
1028
1029         /* Outer loop uses 18 flops */
1030     }
1031
1032     /* Increment number of outer iterations */
1033     outeriter        += nri;
1034
1035     /* Update outer/inner flops */
1036
1037     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*91);
1038 }