src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRF_VdwLJ_GeomW4P1_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRF_VdwLJ_GeomW4P1_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     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 107     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 108     __m256d          dummy_mask,cutoff_mask;
 109     __m128           tmpmask0,tmpmask1;
 110     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 111     __m256d          one     = _mm256_set1_pd(1.0);
 112     __m256d          two     = _mm256_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     krf              = _mm256_set1_pd(fr->ic->k_rf);
 127     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 128     crf              = _mm256_set1_pd(fr->ic->c_rf);
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 136     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 137     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 138     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = jnrC = jnrD = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144     j_coord_offsetC = 0;
 145     j_coord_offsetD = 0;
 146
 147     outeriter        = 0;
 148     inneriter        = 0;
 149
 150     for(iidx=0;iidx<4*DIM;iidx++)
 151     {
 152         scratch[iidx] = 0.0;
 153     }
 154
 155     /* Start outer loop over neighborlists */
 156     for(iidx=0; iidx<nri; iidx++)
 157     {
 158         /* Load shift vector for this list */
 159         i_shift_offset   = DIM*shiftidx[iidx];
 160
 161         /* Load limits for loop over neighbors */
 162         j_index_start    = jindex[iidx];
 163         j_index_end      = jindex[iidx+1];
 164
 165         /* Get outer coordinate index */
 166         inr              = iinr[iidx];
 167         i_coord_offset   = DIM*inr;
 168
 169         /* Load i particle coords and add shift vector */
 170         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 171                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 172
 173         fix0             = _mm256_setzero_pd();
 174         fiy0             = _mm256_setzero_pd();
 175         fiz0             = _mm256_setzero_pd();
 176         fix1             = _mm256_setzero_pd();
 177         fiy1             = _mm256_setzero_pd();
 178         fiz1             = _mm256_setzero_pd();
 179         fix2             = _mm256_setzero_pd();
 180         fiy2             = _mm256_setzero_pd();
 181         fiz2             = _mm256_setzero_pd();
 182         fix3             = _mm256_setzero_pd();
 183         fiy3             = _mm256_setzero_pd();
 184         fiz3             = _mm256_setzero_pd();
 185
 186         /* Reset potential sums */
 187         velecsum         = _mm256_setzero_pd();
 188         vvdwsum          = _mm256_setzero_pd();
 189
 190         /* Start inner kernel loop */
 191         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 192         {
 193
 194             /* Get j neighbor index, and coordinate index */
 195             jnrA             = jjnr[jidx];
 196             jnrB             = jjnr[jidx+1];
 197             jnrC             = jjnr[jidx+2];
 198             jnrD             = jjnr[jidx+3];
 199             j_coord_offsetA  = DIM*jnrA;
 200             j_coord_offsetB  = DIM*jnrB;
 201             j_coord_offsetC  = DIM*jnrC;
 202             j_coord_offsetD  = DIM*jnrD;
 203
 204             /* load j atom coordinates */
 205             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 206                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 207                                                  &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm256_sub_pd(ix0,jx0);
 211             dy00             = _mm256_sub_pd(iy0,jy0);
 212             dz00             = _mm256_sub_pd(iz0,jz0);
 213             dx10             = _mm256_sub_pd(ix1,jx0);
 214             dy10             = _mm256_sub_pd(iy1,jy0);
 215             dz10             = _mm256_sub_pd(iz1,jz0);
 216             dx20             = _mm256_sub_pd(ix2,jx0);
 217             dy20             = _mm256_sub_pd(iy2,jy0);
 218             dz20             = _mm256_sub_pd(iz2,jz0);
 219             dx30             = _mm256_sub_pd(ix3,jx0);
 220             dy30             = _mm256_sub_pd(iy3,jy0);
 221             dz30             = _mm256_sub_pd(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 225             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 226             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 227             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 228
 229             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 230             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 231             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 232
 233             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 234             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 235             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 236             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 237
 238             /* Load parameters for j particles */
 239             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 240                                                                  charge+jnrC+0,charge+jnrD+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243             vdwjidx0C        = 2*vdwtype[jnrC+0];
 244             vdwjidx0D        = 2*vdwtype[jnrD+0];
 245
 246             fjx0             = _mm256_setzero_pd();
 247             fjy0             = _mm256_setzero_pd();
 248             fjz0             = _mm256_setzero_pd();
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 256                                             vdwioffsetptr0+vdwjidx0B,
 257                                             vdwioffsetptr0+vdwjidx0C,
 258                                             vdwioffsetptr0+vdwjidx0D,
 259                                             &c6_00,&c12_00);
 260
 261             /* LENNARD-JONES DISPERSION/REPULSION */
 262
 263             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 264             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 265             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 266             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 267             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 268
 269             /* Update potential sum for this i atom from the interaction with this j atom. */
 270             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 271
 272             fscal            = fvdw;
 273
 274             /* Calculate temporary vectorial force */
 275             tx               = _mm256_mul_pd(fscal,dx00);
 276             ty               = _mm256_mul_pd(fscal,dy00);
 277             tz               = _mm256_mul_pd(fscal,dz00);
 278
 279             /* Update vectorial force */
 280             fix0             = _mm256_add_pd(fix0,tx);
 281             fiy0             = _mm256_add_pd(fiy0,ty);
 282             fiz0             = _mm256_add_pd(fiz0,tz);
 283
 284             fjx0             = _mm256_add_pd(fjx0,tx);
 285             fjy0             = _mm256_add_pd(fjy0,ty);
 286             fjz0             = _mm256_add_pd(fjz0,tz);
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             /* Compute parameters for interactions between i and j atoms */
 293             qq10             = _mm256_mul_pd(iq1,jq0);
 294
 295             /* REACTION-FIELD ELECTROSTATICS */
 296             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 297             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             velecsum         = _mm256_add_pd(velecsum,velec);
 301
 302             fscal            = felec;
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm256_mul_pd(fscal,dx10);
 306             ty               = _mm256_mul_pd(fscal,dy10);
 307             tz               = _mm256_mul_pd(fscal,dz10);
 308
 309             /* Update vectorial force */
 310             fix1             = _mm256_add_pd(fix1,tx);
 311             fiy1             = _mm256_add_pd(fiy1,ty);
 312             fiz1             = _mm256_add_pd(fiz1,tz);
 313
 314             fjx0             = _mm256_add_pd(fjx0,tx);
 315             fjy0             = _mm256_add_pd(fjy0,ty);
 316             fjz0             = _mm256_add_pd(fjz0,tz);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq20             = _mm256_mul_pd(iq2,jq0);
 324
 325             /* REACTION-FIELD ELECTROSTATICS */
 326             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 327             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velecsum         = _mm256_add_pd(velecsum,velec);
 331
 332             fscal            = felec;
 333
 334             /* Calculate temporary vectorial force */
 335             tx               = _mm256_mul_pd(fscal,dx20);
 336             ty               = _mm256_mul_pd(fscal,dy20);
 337             tz               = _mm256_mul_pd(fscal,dz20);
 338
 339             /* Update vectorial force */
 340             fix2             = _mm256_add_pd(fix2,tx);
 341             fiy2             = _mm256_add_pd(fiy2,ty);
 342             fiz2             = _mm256_add_pd(fiz2,tz);
 343
 344             fjx0             = _mm256_add_pd(fjx0,tx);
 345             fjy0             = _mm256_add_pd(fjy0,ty);
 346             fjz0             = _mm256_add_pd(fjz0,tz);
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             /* Compute parameters for interactions between i and j atoms */
 353             qq30             = _mm256_mul_pd(iq3,jq0);
 354
 355             /* REACTION-FIELD ELECTROSTATICS */
 356             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 357             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm256_add_pd(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm256_mul_pd(fscal,dx30);
 366             ty               = _mm256_mul_pd(fscal,dy30);
 367             tz               = _mm256_mul_pd(fscal,dz30);
 368
 369             /* Update vectorial force */
 370             fix3             = _mm256_add_pd(fix3,tx);
 371             fiy3             = _mm256_add_pd(fiy3,ty);
 372             fiz3             = _mm256_add_pd(fiz3,tz);
 373
 374             fjx0             = _mm256_add_pd(fjx0,tx);
 375             fjy0             = _mm256_add_pd(fjy0,ty);
 376             fjz0             = _mm256_add_pd(fjz0,tz);
 377
 378             fjptrA             = f+j_coord_offsetA;
 379             fjptrB             = f+j_coord_offsetB;
 380             fjptrC             = f+j_coord_offsetC;
 381             fjptrD             = f+j_coord_offsetD;
 382
 383             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 384
 385             /* Inner loop uses 131 flops */
 386         }
 387
 388         if(jidx<j_index_end)
 389         {
 390
 391             /* Get j neighbor index, and coordinate index */
 392             jnrlistA         = jjnr[jidx];
 393             jnrlistB         = jjnr[jidx+1];
 394             jnrlistC         = jjnr[jidx+2];
 395             jnrlistD         = jjnr[jidx+3];
 396             /* Sign of each element will be negative for non-real atoms.
 397              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 398              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 399              */
 400             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 401
 402             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 403             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 404             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 405
 406             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 407             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 408             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 409             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 410             j_coord_offsetA  = DIM*jnrA;
 411             j_coord_offsetB  = DIM*jnrB;
 412             j_coord_offsetC  = DIM*jnrC;
 413             j_coord_offsetD  = DIM*jnrD;
 414
 415             /* load j atom coordinates */
 416             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 417                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 418                                                  &jx0,&jy0,&jz0);
 419
 420             /* Calculate displacement vector */
 421             dx00             = _mm256_sub_pd(ix0,jx0);
 422             dy00             = _mm256_sub_pd(iy0,jy0);
 423             dz00             = _mm256_sub_pd(iz0,jz0);
 424             dx10             = _mm256_sub_pd(ix1,jx0);
 425             dy10             = _mm256_sub_pd(iy1,jy0);
 426             dz10             = _mm256_sub_pd(iz1,jz0);
 427             dx20             = _mm256_sub_pd(ix2,jx0);
 428             dy20             = _mm256_sub_pd(iy2,jy0);
 429             dz20             = _mm256_sub_pd(iz2,jz0);
 430             dx30             = _mm256_sub_pd(ix3,jx0);
 431             dy30             = _mm256_sub_pd(iy3,jy0);
 432             dz30             = _mm256_sub_pd(iz3,jz0);
 433
 434             /* Calculate squared distance and things based on it */
 435             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 436             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 437             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 438             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 439
 440             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 441             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 442             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 443
 444             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 445             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 446             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 447             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 448
 449             /* Load parameters for j particles */
 450             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 451                                                                  charge+jnrC+0,charge+jnrD+0);
 452             vdwjidx0A        = 2*vdwtype[jnrA+0];
 453             vdwjidx0B        = 2*vdwtype[jnrB+0];
 454             vdwjidx0C        = 2*vdwtype[jnrC+0];
 455             vdwjidx0D        = 2*vdwtype[jnrD+0];
 456
 457             fjx0             = _mm256_setzero_pd();
 458             fjy0             = _mm256_setzero_pd();
 459             fjz0             = _mm256_setzero_pd();
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             /* Compute parameters for interactions between i and j atoms */
 466             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 467                                             vdwioffsetptr0+vdwjidx0B,
 468                                             vdwioffsetptr0+vdwjidx0C,
 469                                             vdwioffsetptr0+vdwjidx0D,
 470                                             &c6_00,&c12_00);
 471
 472             /* LENNARD-JONES DISPERSION/REPULSION */
 473
 474             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 475             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 476             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 477             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 478             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 479
 480             /* Update potential sum for this i atom from the interaction with this j atom. */
 481             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 482             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 483
 484             fscal            = fvdw;
 485
 486             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 487
 488             /* Calculate temporary vectorial force */
 489             tx               = _mm256_mul_pd(fscal,dx00);
 490             ty               = _mm256_mul_pd(fscal,dy00);
 491             tz               = _mm256_mul_pd(fscal,dz00);
 492
 493             /* Update vectorial force */
 494             fix0             = _mm256_add_pd(fix0,tx);
 495             fiy0             = _mm256_add_pd(fiy0,ty);
 496             fiz0             = _mm256_add_pd(fiz0,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             qq10             = _mm256_mul_pd(iq1,jq0);
 508
 509             /* REACTION-FIELD ELECTROSTATICS */
 510             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 511             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),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,dx10);
 523             ty               = _mm256_mul_pd(fscal,dy10);
 524             tz               = _mm256_mul_pd(fscal,dz10);
 525
 526             /* Update vectorial force */
 527             fix1             = _mm256_add_pd(fix1,tx);
 528             fiy1             = _mm256_add_pd(fiy1,ty);
 529             fiz1             = _mm256_add_pd(fiz1,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             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq20             = _mm256_mul_pd(iq2,jq0);
 541
 542             /* REACTION-FIELD ELECTROSTATICS */
 543             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 544             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 545
 546             /* Update potential sum for this i atom from the interaction with this j atom. */
 547             velec            = _mm256_andnot_pd(dummy_mask,velec);
 548             velecsum         = _mm256_add_pd(velecsum,velec);
 549
 550             fscal            = felec;
 551
 552             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm256_mul_pd(fscal,dx20);
 556             ty               = _mm256_mul_pd(fscal,dy20);
 557             tz               = _mm256_mul_pd(fscal,dz20);
 558
 559             /* Update vectorial force */
 560             fix2             = _mm256_add_pd(fix2,tx);
 561             fiy2             = _mm256_add_pd(fiy2,ty);
 562             fiz2             = _mm256_add_pd(fiz2,tz);
 563
 564             fjx0             = _mm256_add_pd(fjx0,tx);
 565             fjy0             = _mm256_add_pd(fjy0,ty);
 566             fjz0             = _mm256_add_pd(fjz0,tz);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq30             = _mm256_mul_pd(iq3,jq0);
 574
 575             /* REACTION-FIELD ELECTROSTATICS */
 576             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 577             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velec            = _mm256_andnot_pd(dummy_mask,velec);
 581             velecsum         = _mm256_add_pd(velecsum,velec);
 582
 583             fscal            = felec;
 584
 585             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 586
 587             /* Calculate temporary vectorial force */
 588             tx               = _mm256_mul_pd(fscal,dx30);
 589             ty               = _mm256_mul_pd(fscal,dy30);
 590             tz               = _mm256_mul_pd(fscal,dz30);
 591
 592             /* Update vectorial force */
 593             fix3             = _mm256_add_pd(fix3,tx);
 594             fiy3             = _mm256_add_pd(fiy3,ty);
 595             fiz3             = _mm256_add_pd(fiz3,tz);
 596
 597             fjx0             = _mm256_add_pd(fjx0,tx);
 598             fjy0             = _mm256_add_pd(fjy0,ty);
 599             fjz0             = _mm256_add_pd(fjz0,tz);
 600
 601             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 602             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 603             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 604             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 605
 606             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 607
 608             /* Inner loop uses 131 flops */
 609         }
 610
 611         /* End of innermost loop */
 612
 613         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 614                                                  f+i_coord_offset,fshift+i_shift_offset);
 615
 616         ggid                        = gid[iidx];
 617         /* Update potential energies */
 618         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 619         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 26 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*131);
 633 }
 634 /*
 635  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_256_double
 636  * Electrostatics interaction: ReactionField
 637  * VdW interaction:            LennardJones
 638  * Geometry:                   Water4-Particle
 639  * Calculate force/pot:        Force
 640  */
 641 void
 642 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_256_double
 643                     (t_nblist                    * gmx_restrict       nlist,
 644                      rvec                        * gmx_restrict          xx,
 645                      rvec                        * gmx_restrict          ff,
 646                      t_forcerec                  * gmx_restrict          fr,
 647                      t_mdatoms                   * gmx_restrict     mdatoms,
 648                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 649                      t_nrnb                      * gmx_restrict        nrnb)
 650 {
 651     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 652      * just 0 for non-waters.
 653      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 654      * jnr indices corresponding to data put in the four positions in the SIMD register.
 655      */
 656     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 657     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 658     int              jnrA,jnrB,jnrC,jnrD;
 659     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 660     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 661     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 662     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 663     real             rcutoff_scalar;
 664     real             *shiftvec,*fshift,*x,*f;
 665     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 666     real             scratch[4*DIM];
 667     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 668     real *           vdwioffsetptr0;
 669     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 670     real *           vdwioffsetptr1;
 671     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 672     real *           vdwioffsetptr2;
 673     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 674     real *           vdwioffsetptr3;
 675     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 676     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 677     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 678     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 679     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 680     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 681     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 682     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 683     real             *charge;
 684     int              nvdwtype;
 685     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 686     int              *vdwtype;
 687     real             *vdwparam;
 688     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 689     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 690     __m256d          dummy_mask,cutoff_mask;
 691     __m128           tmpmask0,tmpmask1;
 692     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 693     __m256d          one     = _mm256_set1_pd(1.0);
 694     __m256d          two     = _mm256_set1_pd(2.0);
 695     x                = xx[0];
 696     f                = ff[0];
 697
 698     nri              = nlist->nri;
 699     iinr             = nlist->iinr;
 700     jindex           = nlist->jindex;
 701     jjnr             = nlist->jjnr;
 702     shiftidx         = nlist->shift;
 703     gid              = nlist->gid;
 704     shiftvec         = fr->shift_vec[0];
 705     fshift           = fr->fshift[0];
 706     facel            = _mm256_set1_pd(fr->epsfac);
 707     charge           = mdatoms->chargeA;
 708     krf              = _mm256_set1_pd(fr->ic->k_rf);
 709     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 710     crf              = _mm256_set1_pd(fr->ic->c_rf);
 711     nvdwtype         = fr->ntype;
 712     vdwparam         = fr->nbfp;
 713     vdwtype          = mdatoms->typeA;
 714
 715     /* Setup water-specific parameters */
 716     inr              = nlist->iinr[0];
 717     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 718     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 719     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 720     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 721
 722     /* Avoid stupid compiler warnings */
 723     jnrA = jnrB = jnrC = jnrD = 0;
 724     j_coord_offsetA = 0;
 725     j_coord_offsetB = 0;
 726     j_coord_offsetC = 0;
 727     j_coord_offsetD = 0;
 728
 729     outeriter        = 0;
 730     inneriter        = 0;
 731
 732     for(iidx=0;iidx<4*DIM;iidx++)
 733     {
 734         scratch[iidx] = 0.0;
 735     }
 736
 737     /* Start outer loop over neighborlists */
 738     for(iidx=0; iidx<nri; iidx++)
 739     {
 740         /* Load shift vector for this list */
 741         i_shift_offset   = DIM*shiftidx[iidx];
 742
 743         /* Load limits for loop over neighbors */
 744         j_index_start    = jindex[iidx];
 745         j_index_end      = jindex[iidx+1];
 746
 747         /* Get outer coordinate index */
 748         inr              = iinr[iidx];
 749         i_coord_offset   = DIM*inr;
 750
 751         /* Load i particle coords and add shift vector */
 752         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 753                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 754
 755         fix0             = _mm256_setzero_pd();
 756         fiy0             = _mm256_setzero_pd();
 757         fiz0             = _mm256_setzero_pd();
 758         fix1             = _mm256_setzero_pd();
 759         fiy1             = _mm256_setzero_pd();
 760         fiz1             = _mm256_setzero_pd();
 761         fix2             = _mm256_setzero_pd();
 762         fiy2             = _mm256_setzero_pd();
 763         fiz2             = _mm256_setzero_pd();
 764         fix3             = _mm256_setzero_pd();
 765         fiy3             = _mm256_setzero_pd();
 766         fiz3             = _mm256_setzero_pd();
 767
 768         /* Start inner kernel loop */
 769         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 770         {
 771
 772             /* Get j neighbor index, and coordinate index */
 773             jnrA             = jjnr[jidx];
 774             jnrB             = jjnr[jidx+1];
 775             jnrC             = jjnr[jidx+2];
 776             jnrD             = jjnr[jidx+3];
 777             j_coord_offsetA  = DIM*jnrA;
 778             j_coord_offsetB  = DIM*jnrB;
 779             j_coord_offsetC  = DIM*jnrC;
 780             j_coord_offsetD  = DIM*jnrD;
 781
 782             /* load j atom coordinates */
 783             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 784                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 785                                                  &jx0,&jy0,&jz0);
 786
 787             /* Calculate displacement vector */
 788             dx00             = _mm256_sub_pd(ix0,jx0);
 789             dy00             = _mm256_sub_pd(iy0,jy0);
 790             dz00             = _mm256_sub_pd(iz0,jz0);
 791             dx10             = _mm256_sub_pd(ix1,jx0);
 792             dy10             = _mm256_sub_pd(iy1,jy0);
 793             dz10             = _mm256_sub_pd(iz1,jz0);
 794             dx20             = _mm256_sub_pd(ix2,jx0);
 795             dy20             = _mm256_sub_pd(iy2,jy0);
 796             dz20             = _mm256_sub_pd(iz2,jz0);
 797             dx30             = _mm256_sub_pd(ix3,jx0);
 798             dy30             = _mm256_sub_pd(iy3,jy0);
 799             dz30             = _mm256_sub_pd(iz3,jz0);
 800
 801             /* Calculate squared distance and things based on it */
 802             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 803             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 804             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 805             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 806
 807             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 808             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 809             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 810
 811             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 812             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 813             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 814             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 815
 816             /* Load parameters for j particles */
 817             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 818                                                                  charge+jnrC+0,charge+jnrD+0);
 819             vdwjidx0A        = 2*vdwtype[jnrA+0];
 820             vdwjidx0B        = 2*vdwtype[jnrB+0];
 821             vdwjidx0C        = 2*vdwtype[jnrC+0];
 822             vdwjidx0D        = 2*vdwtype[jnrD+0];
 823
 824             fjx0             = _mm256_setzero_pd();
 825             fjy0             = _mm256_setzero_pd();
 826             fjz0             = _mm256_setzero_pd();
 827
 828             /**************************
 829              * CALCULATE INTERACTIONS *
 830              **************************/
 831
 832             /* Compute parameters for interactions between i and j atoms */
 833             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 834                                             vdwioffsetptr0+vdwjidx0B,
 835                                             vdwioffsetptr0+vdwjidx0C,
 836                                             vdwioffsetptr0+vdwjidx0D,
 837                                             &c6_00,&c12_00);
 838
 839             /* LENNARD-JONES DISPERSION/REPULSION */
 840
 841             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 842             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 843
 844             fscal            = fvdw;
 845
 846             /* Calculate temporary vectorial force */
 847             tx               = _mm256_mul_pd(fscal,dx00);
 848             ty               = _mm256_mul_pd(fscal,dy00);
 849             tz               = _mm256_mul_pd(fscal,dz00);
 850
 851             /* Update vectorial force */
 852             fix0             = _mm256_add_pd(fix0,tx);
 853             fiy0             = _mm256_add_pd(fiy0,ty);
 854             fiz0             = _mm256_add_pd(fiz0,tz);
 855
 856             fjx0             = _mm256_add_pd(fjx0,tx);
 857             fjy0             = _mm256_add_pd(fjy0,ty);
 858             fjz0             = _mm256_add_pd(fjz0,tz);
 859
 860             /**************************
 861              * CALCULATE INTERACTIONS *
 862              **************************/
 863
 864             /* Compute parameters for interactions between i and j atoms */
 865             qq10             = _mm256_mul_pd(iq1,jq0);
 866
 867             /* REACTION-FIELD ELECTROSTATICS */
 868             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 869
 870             fscal            = felec;
 871
 872             /* Calculate temporary vectorial force */
 873             tx               = _mm256_mul_pd(fscal,dx10);
 874             ty               = _mm256_mul_pd(fscal,dy10);
 875             tz               = _mm256_mul_pd(fscal,dz10);
 876
 877             /* Update vectorial force */
 878             fix1             = _mm256_add_pd(fix1,tx);
 879             fiy1             = _mm256_add_pd(fiy1,ty);
 880             fiz1             = _mm256_add_pd(fiz1,tz);
 881
 882             fjx0             = _mm256_add_pd(fjx0,tx);
 883             fjy0             = _mm256_add_pd(fjy0,ty);
 884             fjz0             = _mm256_add_pd(fjz0,tz);
 885
 886             /**************************
 887              * CALCULATE INTERACTIONS *
 888              **************************/
 889
 890             /* Compute parameters for interactions between i and j atoms */
 891             qq20             = _mm256_mul_pd(iq2,jq0);
 892
 893             /* REACTION-FIELD ELECTROSTATICS */
 894             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 895
 896             fscal            = felec;
 897
 898             /* Calculate temporary vectorial force */
 899             tx               = _mm256_mul_pd(fscal,dx20);
 900             ty               = _mm256_mul_pd(fscal,dy20);
 901             tz               = _mm256_mul_pd(fscal,dz20);
 902
 903             /* Update vectorial force */
 904             fix2             = _mm256_add_pd(fix2,tx);
 905             fiy2             = _mm256_add_pd(fiy2,ty);
 906             fiz2             = _mm256_add_pd(fiz2,tz);
 907
 908             fjx0             = _mm256_add_pd(fjx0,tx);
 909             fjy0             = _mm256_add_pd(fjy0,ty);
 910             fjz0             = _mm256_add_pd(fjz0,tz);
 911
 912             /**************************
 913              * CALCULATE INTERACTIONS *
 914              **************************/
 915
 916             /* Compute parameters for interactions between i and j atoms */
 917             qq30             = _mm256_mul_pd(iq3,jq0);
 918
 919             /* REACTION-FIELD ELECTROSTATICS */
 920             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 921
 922             fscal            = felec;
 923
 924             /* Calculate temporary vectorial force */
 925             tx               = _mm256_mul_pd(fscal,dx30);
 926             ty               = _mm256_mul_pd(fscal,dy30);
 927             tz               = _mm256_mul_pd(fscal,dz30);
 928
 929             /* Update vectorial force */
 930             fix3             = _mm256_add_pd(fix3,tx);
 931             fiy3             = _mm256_add_pd(fiy3,ty);
 932             fiz3             = _mm256_add_pd(fiz3,tz);
 933
 934             fjx0             = _mm256_add_pd(fjx0,tx);
 935             fjy0             = _mm256_add_pd(fjy0,ty);
 936             fjz0             = _mm256_add_pd(fjz0,tz);
 937
 938             fjptrA             = f+j_coord_offsetA;
 939             fjptrB             = f+j_coord_offsetB;
 940             fjptrC             = f+j_coord_offsetC;
 941             fjptrD             = f+j_coord_offsetD;
 942
 943             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 944
 945             /* Inner loop uses 111 flops */
 946         }
 947
 948         if(jidx<j_index_end)
 949         {
 950
 951             /* Get j neighbor index, and coordinate index */
 952             jnrlistA         = jjnr[jidx];
 953             jnrlistB         = jjnr[jidx+1];
 954             jnrlistC         = jjnr[jidx+2];
 955             jnrlistD         = jjnr[jidx+3];
 956             /* Sign of each element will be negative for non-real atoms.
 957              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 958              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 959              */
 960             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 961
 962             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 963             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 964             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 965
 966             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 967             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 968             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 969             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 970             j_coord_offsetA  = DIM*jnrA;
 971             j_coord_offsetB  = DIM*jnrB;
 972             j_coord_offsetC  = DIM*jnrC;
 973             j_coord_offsetD  = DIM*jnrD;
 974
 975             /* load j atom coordinates */
 976             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 977                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 978                                                  &jx0,&jy0,&jz0);
 979
 980             /* Calculate displacement vector */
 981             dx00             = _mm256_sub_pd(ix0,jx0);
 982             dy00             = _mm256_sub_pd(iy0,jy0);
 983             dz00             = _mm256_sub_pd(iz0,jz0);
 984             dx10             = _mm256_sub_pd(ix1,jx0);
 985             dy10             = _mm256_sub_pd(iy1,jy0);
 986             dz10             = _mm256_sub_pd(iz1,jz0);
 987             dx20             = _mm256_sub_pd(ix2,jx0);
 988             dy20             = _mm256_sub_pd(iy2,jy0);
 989             dz20             = _mm256_sub_pd(iz2,jz0);
 990             dx30             = _mm256_sub_pd(ix3,jx0);
 991             dy30             = _mm256_sub_pd(iy3,jy0);
 992             dz30             = _mm256_sub_pd(iz3,jz0);
 993
 994             /* Calculate squared distance and things based on it */
 995             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 996             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 997             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 998             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 999
1000             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1001             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1002             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1003
1004             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1005             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1006             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1007             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1008
1009             /* Load parameters for j particles */
1010             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1011                                                                  charge+jnrC+0,charge+jnrD+0);
1012             vdwjidx0A        = 2*vdwtype[jnrA+0];
1013             vdwjidx0B        = 2*vdwtype[jnrB+0];
1014             vdwjidx0C        = 2*vdwtype[jnrC+0];
1015             vdwjidx0D        = 2*vdwtype[jnrD+0];
1016
1017             fjx0             = _mm256_setzero_pd();
1018             fjy0             = _mm256_setzero_pd();
1019             fjz0             = _mm256_setzero_pd();
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             /* Compute parameters for interactions between i and j atoms */
1026             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1027                                             vdwioffsetptr0+vdwjidx0B,
1028                                             vdwioffsetptr0+vdwjidx0C,
1029                                             vdwioffsetptr0+vdwjidx0D,
1030                                             &c6_00,&c12_00);
1031
1032             /* LENNARD-JONES DISPERSION/REPULSION */
1033
1034             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1035             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1036
1037             fscal            = fvdw;
1038
1039             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1040
1041             /* Calculate temporary vectorial force */
1042             tx               = _mm256_mul_pd(fscal,dx00);
1043             ty               = _mm256_mul_pd(fscal,dy00);
1044             tz               = _mm256_mul_pd(fscal,dz00);
1045
1046             /* Update vectorial force */
1047             fix0             = _mm256_add_pd(fix0,tx);
1048             fiy0             = _mm256_add_pd(fiy0,ty);
1049             fiz0             = _mm256_add_pd(fiz0,tz);
1050
1051             fjx0             = _mm256_add_pd(fjx0,tx);
1052             fjy0             = _mm256_add_pd(fjy0,ty);
1053             fjz0             = _mm256_add_pd(fjz0,tz);
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             /* Compute parameters for interactions between i and j atoms */
1060             qq10             = _mm256_mul_pd(iq1,jq0);
1061
1062             /* REACTION-FIELD ELECTROSTATICS */
1063             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1064
1065             fscal            = felec;
1066
1067             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1068
1069             /* Calculate temporary vectorial force */
1070             tx               = _mm256_mul_pd(fscal,dx10);
1071             ty               = _mm256_mul_pd(fscal,dy10);
1072             tz               = _mm256_mul_pd(fscal,dz10);
1073
1074             /* Update vectorial force */
1075             fix1             = _mm256_add_pd(fix1,tx);
1076             fiy1             = _mm256_add_pd(fiy1,ty);
1077             fiz1             = _mm256_add_pd(fiz1,tz);
1078
1079             fjx0             = _mm256_add_pd(fjx0,tx);
1080             fjy0             = _mm256_add_pd(fjy0,ty);
1081             fjz0             = _mm256_add_pd(fjz0,tz);
1082
1083             /**************************
1084              * CALCULATE INTERACTIONS *
1085              **************************/
1086
1087             /* Compute parameters for interactions between i and j atoms */
1088             qq20             = _mm256_mul_pd(iq2,jq0);
1089
1090             /* REACTION-FIELD ELECTROSTATICS */
1091             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1092
1093             fscal            = felec;
1094
1095             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1096
1097             /* Calculate temporary vectorial force */
1098             tx               = _mm256_mul_pd(fscal,dx20);
1099             ty               = _mm256_mul_pd(fscal,dy20);
1100             tz               = _mm256_mul_pd(fscal,dz20);
1101
1102             /* Update vectorial force */
1103             fix2             = _mm256_add_pd(fix2,tx);
1104             fiy2             = _mm256_add_pd(fiy2,ty);
1105             fiz2             = _mm256_add_pd(fiz2,tz);
1106
1107             fjx0             = _mm256_add_pd(fjx0,tx);
1108             fjy0             = _mm256_add_pd(fjy0,ty);
1109             fjz0             = _mm256_add_pd(fjz0,tz);
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             /* Compute parameters for interactions between i and j atoms */
1116             qq30             = _mm256_mul_pd(iq3,jq0);
1117
1118             /* REACTION-FIELD ELECTROSTATICS */
1119             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1120
1121             fscal            = felec;
1122
1123             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1124
1125             /* Calculate temporary vectorial force */
1126             tx               = _mm256_mul_pd(fscal,dx30);
1127             ty               = _mm256_mul_pd(fscal,dy30);
1128             tz               = _mm256_mul_pd(fscal,dz30);
1129
1130             /* Update vectorial force */
1131             fix3             = _mm256_add_pd(fix3,tx);
1132             fiy3             = _mm256_add_pd(fiy3,ty);
1133             fiz3             = _mm256_add_pd(fiz3,tz);
1134
1135             fjx0             = _mm256_add_pd(fjx0,tx);
1136             fjy0             = _mm256_add_pd(fjy0,ty);
1137             fjz0             = _mm256_add_pd(fjz0,tz);
1138
1139             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1140             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1141             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1142             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1143
1144             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1145
1146             /* Inner loop uses 111 flops */
1147         }
1148
1149         /* End of innermost loop */
1150
1151         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1152                                                  f+i_coord_offset,fshift+i_shift_offset);
1153
1154         /* Increment number of inner iterations */
1155         inneriter                  += j_index_end - j_index_start;
1156
1157         /* Outer loop uses 24 flops */
1158     }
1159
1160     /* Increment number of outer iterations */
1161     outeriter        += nri;
1162
1163     /* Update outer/inner flops */
1164
1165     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*111);
1166 }