src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_avx_256_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     real *           vdwioffsetptr1;
  86     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     real *           vdwioffsetptr3;
  90     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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     __m128i          vfitab;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m256d          dummy_mask,cutoff_mask;
 110     __m128           tmpmask0,tmpmask1;
 111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 112     __m256d          one     = _mm256_set1_pd(1.0);
 113     __m256d          two     = _mm256_set1_pd(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_pd(fr->ic->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     vftab            = kernel_data->table_vdw->data;
 132     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 137     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 138     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 139     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 173
 174         fix0             = _mm256_setzero_pd();
 175         fiy0             = _mm256_setzero_pd();
 176         fiz0             = _mm256_setzero_pd();
 177         fix1             = _mm256_setzero_pd();
 178         fiy1             = _mm256_setzero_pd();
 179         fiz1             = _mm256_setzero_pd();
 180         fix2             = _mm256_setzero_pd();
 181         fiy2             = _mm256_setzero_pd();
 182         fiz2             = _mm256_setzero_pd();
 183         fix3             = _mm256_setzero_pd();
 184         fiy3             = _mm256_setzero_pd();
 185         fiz3             = _mm256_setzero_pd();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm256_setzero_pd();
 189         vvdwsum          = _mm256_setzero_pd();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             jnrC             = jjnr[jidx+2];
 199             jnrD             = jjnr[jidx+3];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204
 205             /* load j atom coordinates */
 206             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 208                                                  &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm256_sub_pd(ix0,jx0);
 212             dy00             = _mm256_sub_pd(iy0,jy0);
 213             dz00             = _mm256_sub_pd(iz0,jz0);
 214             dx10             = _mm256_sub_pd(ix1,jx0);
 215             dy10             = _mm256_sub_pd(iy1,jy0);
 216             dz10             = _mm256_sub_pd(iz1,jz0);
 217             dx20             = _mm256_sub_pd(ix2,jx0);
 218             dy20             = _mm256_sub_pd(iy2,jy0);
 219             dz20             = _mm256_sub_pd(iz2,jz0);
 220             dx30             = _mm256_sub_pd(ix3,jx0);
 221             dy30             = _mm256_sub_pd(iy3,jy0);
 222             dz30             = _mm256_sub_pd(iz3,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 226             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 227             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 228             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 229
 230             rinv00           = avx256_invsqrt_d(rsq00);
 231             rinv10           = avx256_invsqrt_d(rsq10);
 232             rinv20           = avx256_invsqrt_d(rsq20);
 233             rinv30           = avx256_invsqrt_d(rsq30);
 234
 235             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 236             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 237             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 241                                                                  charge+jnrC+0,charge+jnrD+0);
 242             vdwjidx0A        = 2*vdwtype[jnrA+0];
 243             vdwjidx0B        = 2*vdwtype[jnrB+0];
 244             vdwjidx0C        = 2*vdwtype[jnrC+0];
 245             vdwjidx0D        = 2*vdwtype[jnrD+0];
 246
 247             fjx0             = _mm256_setzero_pd();
 248             fjy0             = _mm256_setzero_pd();
 249             fjz0             = _mm256_setzero_pd();
 250
 251             /**************************
 252              * CALCULATE INTERACTIONS *
 253              **************************/
 254
 255             r00              = _mm256_mul_pd(rsq00,rinv00);
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 259                                             vdwioffsetptr0+vdwjidx0B,
 260                                             vdwioffsetptr0+vdwjidx0C,
 261                                             vdwioffsetptr0+vdwjidx0D,
 262                                             &c6_00,&c12_00);
 263
 264             /* Calculate table index by multiplying r with table scale and truncate to integer */
 265             rt               = _mm256_mul_pd(r00,vftabscale);
 266             vfitab           = _mm256_cvttpd_epi32(rt);
 267             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 268             vfitab           = _mm_slli_epi32(vfitab,3);
 269
 270             /* CUBIC SPLINE TABLE DISPERSION */
 271             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 272             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 273             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 274             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 275             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 276             Heps             = _mm256_mul_pd(vfeps,H);
 277             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 278             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 279             vvdw6            = _mm256_mul_pd(c6_00,VV);
 280             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 281             fvdw6            = _mm256_mul_pd(c6_00,FF);
 282
 283             /* CUBIC SPLINE TABLE REPULSION */
 284             vfitab           = _mm_add_epi32(vfitab,ifour);
 285             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 286             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 287             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 288             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 289             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 290             Heps             = _mm256_mul_pd(vfeps,H);
 291             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 292             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 293             vvdw12           = _mm256_mul_pd(c12_00,VV);
 294             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 295             fvdw12           = _mm256_mul_pd(c12_00,FF);
 296             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 297             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 298
 299             /* Update potential sum for this i atom from the interaction with this j atom. */
 300             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 301
 302             fscal            = fvdw;
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm256_mul_pd(fscal,dx00);
 306             ty               = _mm256_mul_pd(fscal,dy00);
 307             tz               = _mm256_mul_pd(fscal,dz00);
 308
 309             /* Update vectorial force */
 310             fix0             = _mm256_add_pd(fix0,tx);
 311             fiy0             = _mm256_add_pd(fiy0,ty);
 312             fiz0             = _mm256_add_pd(fiz0,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             qq10             = _mm256_mul_pd(iq1,jq0);
 324
 325             /* COULOMB ELECTROSTATICS */
 326             velec            = _mm256_mul_pd(qq10,rinv10);
 327             felec            = _mm256_mul_pd(velec,rinvsq10);
 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,dx10);
 336             ty               = _mm256_mul_pd(fscal,dy10);
 337             tz               = _mm256_mul_pd(fscal,dz10);
 338
 339             /* Update vectorial force */
 340             fix1             = _mm256_add_pd(fix1,tx);
 341             fiy1             = _mm256_add_pd(fiy1,ty);
 342             fiz1             = _mm256_add_pd(fiz1,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             qq20             = _mm256_mul_pd(iq2,jq0);
 354
 355             /* COULOMB ELECTROSTATICS */
 356             velec            = _mm256_mul_pd(qq20,rinv20);
 357             felec            = _mm256_mul_pd(velec,rinvsq20);
 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,dx20);
 366             ty               = _mm256_mul_pd(fscal,dy20);
 367             tz               = _mm256_mul_pd(fscal,dz20);
 368
 369             /* Update vectorial force */
 370             fix2             = _mm256_add_pd(fix2,tx);
 371             fiy2             = _mm256_add_pd(fiy2,ty);
 372             fiz2             = _mm256_add_pd(fiz2,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             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq30             = _mm256_mul_pd(iq3,jq0);
 384
 385             /* COULOMB ELECTROSTATICS */
 386             velec            = _mm256_mul_pd(qq30,rinv30);
 387             felec            = _mm256_mul_pd(velec,rinvsq30);
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velecsum         = _mm256_add_pd(velecsum,velec);
 391
 392             fscal            = felec;
 393
 394             /* Calculate temporary vectorial force */
 395             tx               = _mm256_mul_pd(fscal,dx30);
 396             ty               = _mm256_mul_pd(fscal,dy30);
 397             tz               = _mm256_mul_pd(fscal,dz30);
 398
 399             /* Update vectorial force */
 400             fix3             = _mm256_add_pd(fix3,tx);
 401             fiy3             = _mm256_add_pd(fiy3,ty);
 402             fiz3             = _mm256_add_pd(fiz3,tz);
 403
 404             fjx0             = _mm256_add_pd(fjx0,tx);
 405             fjy0             = _mm256_add_pd(fjy0,ty);
 406             fjz0             = _mm256_add_pd(fjz0,tz);
 407
 408             fjptrA             = f+j_coord_offsetA;
 409             fjptrB             = f+j_coord_offsetB;
 410             fjptrC             = f+j_coord_offsetC;
 411             fjptrD             = f+j_coord_offsetD;
 412
 413             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 414
 415             /* Inner loop uses 140 flops */
 416         }
 417
 418         if(jidx<j_index_end)
 419         {
 420
 421             /* Get j neighbor index, and coordinate index */
 422             jnrlistA         = jjnr[jidx];
 423             jnrlistB         = jjnr[jidx+1];
 424             jnrlistC         = jjnr[jidx+2];
 425             jnrlistD         = jjnr[jidx+3];
 426             /* Sign of each element will be negative for non-real atoms.
 427              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 428              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 429              */
 430             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 431
 432             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 433             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 434             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 435
 436             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 437             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 438             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 439             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 440             j_coord_offsetA  = DIM*jnrA;
 441             j_coord_offsetB  = DIM*jnrB;
 442             j_coord_offsetC  = DIM*jnrC;
 443             j_coord_offsetD  = DIM*jnrD;
 444
 445             /* load j atom coordinates */
 446             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 447                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 448                                                  &jx0,&jy0,&jz0);
 449
 450             /* Calculate displacement vector */
 451             dx00             = _mm256_sub_pd(ix0,jx0);
 452             dy00             = _mm256_sub_pd(iy0,jy0);
 453             dz00             = _mm256_sub_pd(iz0,jz0);
 454             dx10             = _mm256_sub_pd(ix1,jx0);
 455             dy10             = _mm256_sub_pd(iy1,jy0);
 456             dz10             = _mm256_sub_pd(iz1,jz0);
 457             dx20             = _mm256_sub_pd(ix2,jx0);
 458             dy20             = _mm256_sub_pd(iy2,jy0);
 459             dz20             = _mm256_sub_pd(iz2,jz0);
 460             dx30             = _mm256_sub_pd(ix3,jx0);
 461             dy30             = _mm256_sub_pd(iy3,jy0);
 462             dz30             = _mm256_sub_pd(iz3,jz0);
 463
 464             /* Calculate squared distance and things based on it */
 465             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 466             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 467             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 468             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 469
 470             rinv00           = avx256_invsqrt_d(rsq00);
 471             rinv10           = avx256_invsqrt_d(rsq10);
 472             rinv20           = avx256_invsqrt_d(rsq20);
 473             rinv30           = avx256_invsqrt_d(rsq30);
 474
 475             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 476             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 477             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 478
 479             /* Load parameters for j particles */
 480             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 481                                                                  charge+jnrC+0,charge+jnrD+0);
 482             vdwjidx0A        = 2*vdwtype[jnrA+0];
 483             vdwjidx0B        = 2*vdwtype[jnrB+0];
 484             vdwjidx0C        = 2*vdwtype[jnrC+0];
 485             vdwjidx0D        = 2*vdwtype[jnrD+0];
 486
 487             fjx0             = _mm256_setzero_pd();
 488             fjy0             = _mm256_setzero_pd();
 489             fjz0             = _mm256_setzero_pd();
 490
 491             /**************************
 492              * CALCULATE INTERACTIONS *
 493              **************************/
 494
 495             r00              = _mm256_mul_pd(rsq00,rinv00);
 496             r00              = _mm256_andnot_pd(dummy_mask,r00);
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 500                                             vdwioffsetptr0+vdwjidx0B,
 501                                             vdwioffsetptr0+vdwjidx0C,
 502                                             vdwioffsetptr0+vdwjidx0D,
 503                                             &c6_00,&c12_00);
 504
 505             /* Calculate table index by multiplying r with table scale and truncate to integer */
 506             rt               = _mm256_mul_pd(r00,vftabscale);
 507             vfitab           = _mm256_cvttpd_epi32(rt);
 508             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 509             vfitab           = _mm_slli_epi32(vfitab,3);
 510
 511             /* CUBIC SPLINE TABLE DISPERSION */
 512             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 513             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 514             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 515             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 516             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 517             Heps             = _mm256_mul_pd(vfeps,H);
 518             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 519             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 520             vvdw6            = _mm256_mul_pd(c6_00,VV);
 521             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 522             fvdw6            = _mm256_mul_pd(c6_00,FF);
 523
 524             /* CUBIC SPLINE TABLE REPULSION */
 525             vfitab           = _mm_add_epi32(vfitab,ifour);
 526             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 527             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 528             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 529             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 530             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 531             Heps             = _mm256_mul_pd(vfeps,H);
 532             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 533             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 534             vvdw12           = _mm256_mul_pd(c12_00,VV);
 535             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 536             fvdw12           = _mm256_mul_pd(c12_00,FF);
 537             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 538             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 542             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 543
 544             fscal            = fvdw;
 545
 546             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 547
 548             /* Calculate temporary vectorial force */
 549             tx               = _mm256_mul_pd(fscal,dx00);
 550             ty               = _mm256_mul_pd(fscal,dy00);
 551             tz               = _mm256_mul_pd(fscal,dz00);
 552
 553             /* Update vectorial force */
 554             fix0             = _mm256_add_pd(fix0,tx);
 555             fiy0             = _mm256_add_pd(fiy0,ty);
 556             fiz0             = _mm256_add_pd(fiz0,tz);
 557
 558             fjx0             = _mm256_add_pd(fjx0,tx);
 559             fjy0             = _mm256_add_pd(fjy0,ty);
 560             fjz0             = _mm256_add_pd(fjz0,tz);
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq10             = _mm256_mul_pd(iq1,jq0);
 568
 569             /* COULOMB ELECTROSTATICS */
 570             velec            = _mm256_mul_pd(qq10,rinv10);
 571             felec            = _mm256_mul_pd(velec,rinvsq10);
 572
 573             /* Update potential sum for this i atom from the interaction with this j atom. */
 574             velec            = _mm256_andnot_pd(dummy_mask,velec);
 575             velecsum         = _mm256_add_pd(velecsum,velec);
 576
 577             fscal            = felec;
 578
 579             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm256_mul_pd(fscal,dx10);
 583             ty               = _mm256_mul_pd(fscal,dy10);
 584             tz               = _mm256_mul_pd(fscal,dz10);
 585
 586             /* Update vectorial force */
 587             fix1             = _mm256_add_pd(fix1,tx);
 588             fiy1             = _mm256_add_pd(fiy1,ty);
 589             fiz1             = _mm256_add_pd(fiz1,tz);
 590
 591             fjx0             = _mm256_add_pd(fjx0,tx);
 592             fjy0             = _mm256_add_pd(fjy0,ty);
 593             fjz0             = _mm256_add_pd(fjz0,tz);
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             /* Compute parameters for interactions between i and j atoms */
 600             qq20             = _mm256_mul_pd(iq2,jq0);
 601
 602             /* COULOMB ELECTROSTATICS */
 603             velec            = _mm256_mul_pd(qq20,rinv20);
 604             felec            = _mm256_mul_pd(velec,rinvsq20);
 605
 606             /* Update potential sum for this i atom from the interaction with this j atom. */
 607             velec            = _mm256_andnot_pd(dummy_mask,velec);
 608             velecsum         = _mm256_add_pd(velecsum,velec);
 609
 610             fscal            = felec;
 611
 612             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 613
 614             /* Calculate temporary vectorial force */
 615             tx               = _mm256_mul_pd(fscal,dx20);
 616             ty               = _mm256_mul_pd(fscal,dy20);
 617             tz               = _mm256_mul_pd(fscal,dz20);
 618
 619             /* Update vectorial force */
 620             fix2             = _mm256_add_pd(fix2,tx);
 621             fiy2             = _mm256_add_pd(fiy2,ty);
 622             fiz2             = _mm256_add_pd(fiz2,tz);
 623
 624             fjx0             = _mm256_add_pd(fjx0,tx);
 625             fjy0             = _mm256_add_pd(fjy0,ty);
 626             fjz0             = _mm256_add_pd(fjz0,tz);
 627
 628             /**************************
 629              * CALCULATE INTERACTIONS *
 630              **************************/
 631
 632             /* Compute parameters for interactions between i and j atoms */
 633             qq30             = _mm256_mul_pd(iq3,jq0);
 634
 635             /* COULOMB ELECTROSTATICS */
 636             velec            = _mm256_mul_pd(qq30,rinv30);
 637             felec            = _mm256_mul_pd(velec,rinvsq30);
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm256_andnot_pd(dummy_mask,velec);
 641             velecsum         = _mm256_add_pd(velecsum,velec);
 642
 643             fscal            = felec;
 644
 645             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm256_mul_pd(fscal,dx30);
 649             ty               = _mm256_mul_pd(fscal,dy30);
 650             tz               = _mm256_mul_pd(fscal,dz30);
 651
 652             /* Update vectorial force */
 653             fix3             = _mm256_add_pd(fix3,tx);
 654             fiy3             = _mm256_add_pd(fiy3,ty);
 655             fiz3             = _mm256_add_pd(fiz3,tz);
 656
 657             fjx0             = _mm256_add_pd(fjx0,tx);
 658             fjy0             = _mm256_add_pd(fjy0,ty);
 659             fjz0             = _mm256_add_pd(fjz0,tz);
 660
 661             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 662             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 663             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 664             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 665
 666             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 667
 668             /* Inner loop uses 141 flops */
 669         }
 670
 671         /* End of innermost loop */
 672
 673         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 674                                                  f+i_coord_offset,fshift+i_shift_offset);
 675
 676         ggid                        = gid[iidx];
 677         /* Update potential energies */
 678         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 679         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 680
 681         /* Increment number of inner iterations */
 682         inneriter                  += j_index_end - j_index_start;
 683
 684         /* Outer loop uses 26 flops */
 685     }
 686
 687     /* Increment number of outer iterations */
 688     outeriter        += nri;
 689
 690     /* Update outer/inner flops */
 691
 692     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*141);
 693 }
 694 /*
 695  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_double
 696  * Electrostatics interaction: Coulomb
 697  * VdW interaction:            CubicSplineTable
 698  * Geometry:                   Water4-Particle
 699  * Calculate force/pot:        Force
 700  */
 701 void
 702 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_256_double
 703                     (t_nblist                    * gmx_restrict       nlist,
 704                      rvec                        * gmx_restrict          xx,
 705                      rvec                        * gmx_restrict          ff,
 706                      struct t_forcerec           * gmx_restrict          fr,
 707                      t_mdatoms                   * gmx_restrict     mdatoms,
 708                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 709                      t_nrnb                      * gmx_restrict        nrnb)
 710 {
 711     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 712      * just 0 for non-waters.
 713      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 714      * jnr indices corresponding to data put in the four positions in the SIMD register.
 715      */
 716     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 717     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 718     int              jnrA,jnrB,jnrC,jnrD;
 719     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 720     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 721     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 722     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 723     real             rcutoff_scalar;
 724     real             *shiftvec,*fshift,*x,*f;
 725     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 726     real             scratch[4*DIM];
 727     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 728     real *           vdwioffsetptr0;
 729     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 730     real *           vdwioffsetptr1;
 731     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 732     real *           vdwioffsetptr2;
 733     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 734     real *           vdwioffsetptr3;
 735     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 736     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 737     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 738     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 739     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 740     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 741     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 742     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 743     real             *charge;
 744     int              nvdwtype;
 745     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 746     int              *vdwtype;
 747     real             *vdwparam;
 748     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 749     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 750     __m128i          vfitab;
 751     __m128i          ifour       = _mm_set1_epi32(4);
 752     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 753     real             *vftab;
 754     __m256d          dummy_mask,cutoff_mask;
 755     __m128           tmpmask0,tmpmask1;
 756     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 757     __m256d          one     = _mm256_set1_pd(1.0);
 758     __m256d          two     = _mm256_set1_pd(2.0);
 759     x                = xx[0];
 760     f                = ff[0];
 761
 762     nri              = nlist->nri;
 763     iinr             = nlist->iinr;
 764     jindex           = nlist->jindex;
 765     jjnr             = nlist->jjnr;
 766     shiftidx         = nlist->shift;
 767     gid              = nlist->gid;
 768     shiftvec         = fr->shift_vec[0];
 769     fshift           = fr->fshift[0];
 770     facel            = _mm256_set1_pd(fr->ic->epsfac);
 771     charge           = mdatoms->chargeA;
 772     nvdwtype         = fr->ntype;
 773     vdwparam         = fr->nbfp;
 774     vdwtype          = mdatoms->typeA;
 775
 776     vftab            = kernel_data->table_vdw->data;
 777     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 778
 779     /* Setup water-specific parameters */
 780     inr              = nlist->iinr[0];
 781     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 782     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 783     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 784     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 785
 786     /* Avoid stupid compiler warnings */
 787     jnrA = jnrB = jnrC = jnrD = 0;
 788     j_coord_offsetA = 0;
 789     j_coord_offsetB = 0;
 790     j_coord_offsetC = 0;
 791     j_coord_offsetD = 0;
 792
 793     outeriter        = 0;
 794     inneriter        = 0;
 795
 796     for(iidx=0;iidx<4*DIM;iidx++)
 797     {
 798         scratch[iidx] = 0.0;
 799     }
 800
 801     /* Start outer loop over neighborlists */
 802     for(iidx=0; iidx<nri; iidx++)
 803     {
 804         /* Load shift vector for this list */
 805         i_shift_offset   = DIM*shiftidx[iidx];
 806
 807         /* Load limits for loop over neighbors */
 808         j_index_start    = jindex[iidx];
 809         j_index_end      = jindex[iidx+1];
 810
 811         /* Get outer coordinate index */
 812         inr              = iinr[iidx];
 813         i_coord_offset   = DIM*inr;
 814
 815         /* Load i particle coords and add shift vector */
 816         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 817                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 818
 819         fix0             = _mm256_setzero_pd();
 820         fiy0             = _mm256_setzero_pd();
 821         fiz0             = _mm256_setzero_pd();
 822         fix1             = _mm256_setzero_pd();
 823         fiy1             = _mm256_setzero_pd();
 824         fiz1             = _mm256_setzero_pd();
 825         fix2             = _mm256_setzero_pd();
 826         fiy2             = _mm256_setzero_pd();
 827         fiz2             = _mm256_setzero_pd();
 828         fix3             = _mm256_setzero_pd();
 829         fiy3             = _mm256_setzero_pd();
 830         fiz3             = _mm256_setzero_pd();
 831
 832         /* Start inner kernel loop */
 833         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 834         {
 835
 836             /* Get j neighbor index, and coordinate index */
 837             jnrA             = jjnr[jidx];
 838             jnrB             = jjnr[jidx+1];
 839             jnrC             = jjnr[jidx+2];
 840             jnrD             = jjnr[jidx+3];
 841             j_coord_offsetA  = DIM*jnrA;
 842             j_coord_offsetB  = DIM*jnrB;
 843             j_coord_offsetC  = DIM*jnrC;
 844             j_coord_offsetD  = DIM*jnrD;
 845
 846             /* load j atom coordinates */
 847             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 848                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 849                                                  &jx0,&jy0,&jz0);
 850
 851             /* Calculate displacement vector */
 852             dx00             = _mm256_sub_pd(ix0,jx0);
 853             dy00             = _mm256_sub_pd(iy0,jy0);
 854             dz00             = _mm256_sub_pd(iz0,jz0);
 855             dx10             = _mm256_sub_pd(ix1,jx0);
 856             dy10             = _mm256_sub_pd(iy1,jy0);
 857             dz10             = _mm256_sub_pd(iz1,jz0);
 858             dx20             = _mm256_sub_pd(ix2,jx0);
 859             dy20             = _mm256_sub_pd(iy2,jy0);
 860             dz20             = _mm256_sub_pd(iz2,jz0);
 861             dx30             = _mm256_sub_pd(ix3,jx0);
 862             dy30             = _mm256_sub_pd(iy3,jy0);
 863             dz30             = _mm256_sub_pd(iz3,jz0);
 864
 865             /* Calculate squared distance and things based on it */
 866             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 867             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 868             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 869             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 870
 871             rinv00           = avx256_invsqrt_d(rsq00);
 872             rinv10           = avx256_invsqrt_d(rsq10);
 873             rinv20           = avx256_invsqrt_d(rsq20);
 874             rinv30           = avx256_invsqrt_d(rsq30);
 875
 876             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 877             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 878             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 879
 880             /* Load parameters for j particles */
 881             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 882                                                                  charge+jnrC+0,charge+jnrD+0);
 883             vdwjidx0A        = 2*vdwtype[jnrA+0];
 884             vdwjidx0B        = 2*vdwtype[jnrB+0];
 885             vdwjidx0C        = 2*vdwtype[jnrC+0];
 886             vdwjidx0D        = 2*vdwtype[jnrD+0];
 887
 888             fjx0             = _mm256_setzero_pd();
 889             fjy0             = _mm256_setzero_pd();
 890             fjz0             = _mm256_setzero_pd();
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             r00              = _mm256_mul_pd(rsq00,rinv00);
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 900                                             vdwioffsetptr0+vdwjidx0B,
 901                                             vdwioffsetptr0+vdwjidx0C,
 902                                             vdwioffsetptr0+vdwjidx0D,
 903                                             &c6_00,&c12_00);
 904
 905             /* Calculate table index by multiplying r with table scale and truncate to integer */
 906             rt               = _mm256_mul_pd(r00,vftabscale);
 907             vfitab           = _mm256_cvttpd_epi32(rt);
 908             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 909             vfitab           = _mm_slli_epi32(vfitab,3);
 910
 911             /* CUBIC SPLINE TABLE DISPERSION */
 912             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 913             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 914             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 915             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 916             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 917             Heps             = _mm256_mul_pd(vfeps,H);
 918             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 919             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 920             fvdw6            = _mm256_mul_pd(c6_00,FF);
 921
 922             /* CUBIC SPLINE TABLE REPULSION */
 923             vfitab           = _mm_add_epi32(vfitab,ifour);
 924             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 925             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 926             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 927             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 928             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 929             Heps             = _mm256_mul_pd(vfeps,H);
 930             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 931             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 932             fvdw12           = _mm256_mul_pd(c12_00,FF);
 933             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 934
 935             fscal            = fvdw;
 936
 937             /* Calculate temporary vectorial force */
 938             tx               = _mm256_mul_pd(fscal,dx00);
 939             ty               = _mm256_mul_pd(fscal,dy00);
 940             tz               = _mm256_mul_pd(fscal,dz00);
 941
 942             /* Update vectorial force */
 943             fix0             = _mm256_add_pd(fix0,tx);
 944             fiy0             = _mm256_add_pd(fiy0,ty);
 945             fiz0             = _mm256_add_pd(fiz0,tz);
 946
 947             fjx0             = _mm256_add_pd(fjx0,tx);
 948             fjy0             = _mm256_add_pd(fjy0,ty);
 949             fjz0             = _mm256_add_pd(fjz0,tz);
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             /* Compute parameters for interactions between i and j atoms */
 956             qq10             = _mm256_mul_pd(iq1,jq0);
 957
 958             /* COULOMB ELECTROSTATICS */
 959             velec            = _mm256_mul_pd(qq10,rinv10);
 960             felec            = _mm256_mul_pd(velec,rinvsq10);
 961
 962             fscal            = felec;
 963
 964             /* Calculate temporary vectorial force */
 965             tx               = _mm256_mul_pd(fscal,dx10);
 966             ty               = _mm256_mul_pd(fscal,dy10);
 967             tz               = _mm256_mul_pd(fscal,dz10);
 968
 969             /* Update vectorial force */
 970             fix1             = _mm256_add_pd(fix1,tx);
 971             fiy1             = _mm256_add_pd(fiy1,ty);
 972             fiz1             = _mm256_add_pd(fiz1,tz);
 973
 974             fjx0             = _mm256_add_pd(fjx0,tx);
 975             fjy0             = _mm256_add_pd(fjy0,ty);
 976             fjz0             = _mm256_add_pd(fjz0,tz);
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             qq20             = _mm256_mul_pd(iq2,jq0);
 984
 985             /* COULOMB ELECTROSTATICS */
 986             velec            = _mm256_mul_pd(qq20,rinv20);
 987             felec            = _mm256_mul_pd(velec,rinvsq20);
 988
 989             fscal            = felec;
 990
 991             /* Calculate temporary vectorial force */
 992             tx               = _mm256_mul_pd(fscal,dx20);
 993             ty               = _mm256_mul_pd(fscal,dy20);
 994             tz               = _mm256_mul_pd(fscal,dz20);
 995
 996             /* Update vectorial force */
 997             fix2             = _mm256_add_pd(fix2,tx);
 998             fiy2             = _mm256_add_pd(fiy2,ty);
 999             fiz2             = _mm256_add_pd(fiz2,tz);
1000
1001             fjx0             = _mm256_add_pd(fjx0,tx);
1002             fjy0             = _mm256_add_pd(fjy0,ty);
1003             fjz0             = _mm256_add_pd(fjz0,tz);
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             qq30             = _mm256_mul_pd(iq3,jq0);
1011
1012             /* COULOMB ELECTROSTATICS */
1013             velec            = _mm256_mul_pd(qq30,rinv30);
1014             felec            = _mm256_mul_pd(velec,rinvsq30);
1015
1016             fscal            = felec;
1017
1018             /* Calculate temporary vectorial force */
1019             tx               = _mm256_mul_pd(fscal,dx30);
1020             ty               = _mm256_mul_pd(fscal,dy30);
1021             tz               = _mm256_mul_pd(fscal,dz30);
1022
1023             /* Update vectorial force */
1024             fix3             = _mm256_add_pd(fix3,tx);
1025             fiy3             = _mm256_add_pd(fiy3,ty);
1026             fiz3             = _mm256_add_pd(fiz3,tz);
1027
1028             fjx0             = _mm256_add_pd(fjx0,tx);
1029             fjy0             = _mm256_add_pd(fjy0,ty);
1030             fjz0             = _mm256_add_pd(fjz0,tz);
1031
1032             fjptrA             = f+j_coord_offsetA;
1033             fjptrB             = f+j_coord_offsetB;
1034             fjptrC             = f+j_coord_offsetC;
1035             fjptrD             = f+j_coord_offsetD;
1036
1037             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1038
1039             /* Inner loop uses 129 flops */
1040         }
1041
1042         if(jidx<j_index_end)
1043         {
1044
1045             /* Get j neighbor index, and coordinate index */
1046             jnrlistA         = jjnr[jidx];
1047             jnrlistB         = jjnr[jidx+1];
1048             jnrlistC         = jjnr[jidx+2];
1049             jnrlistD         = jjnr[jidx+3];
1050             /* Sign of each element will be negative for non-real atoms.
1051              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1052              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1053              */
1054             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1055
1056             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1057             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1058             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1059
1060             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1061             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1062             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1063             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1064             j_coord_offsetA  = DIM*jnrA;
1065             j_coord_offsetB  = DIM*jnrB;
1066             j_coord_offsetC  = DIM*jnrC;
1067             j_coord_offsetD  = DIM*jnrD;
1068
1069             /* load j atom coordinates */
1070             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1071                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1072                                                  &jx0,&jy0,&jz0);
1073
1074             /* Calculate displacement vector */
1075             dx00             = _mm256_sub_pd(ix0,jx0);
1076             dy00             = _mm256_sub_pd(iy0,jy0);
1077             dz00             = _mm256_sub_pd(iz0,jz0);
1078             dx10             = _mm256_sub_pd(ix1,jx0);
1079             dy10             = _mm256_sub_pd(iy1,jy0);
1080             dz10             = _mm256_sub_pd(iz1,jz0);
1081             dx20             = _mm256_sub_pd(ix2,jx0);
1082             dy20             = _mm256_sub_pd(iy2,jy0);
1083             dz20             = _mm256_sub_pd(iz2,jz0);
1084             dx30             = _mm256_sub_pd(ix3,jx0);
1085             dy30             = _mm256_sub_pd(iy3,jy0);
1086             dz30             = _mm256_sub_pd(iz3,jz0);
1087
1088             /* Calculate squared distance and things based on it */
1089             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1090             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1091             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1092             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1093
1094             rinv00           = avx256_invsqrt_d(rsq00);
1095             rinv10           = avx256_invsqrt_d(rsq10);
1096             rinv20           = avx256_invsqrt_d(rsq20);
1097             rinv30           = avx256_invsqrt_d(rsq30);
1098
1099             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1100             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1101             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1102
1103             /* Load parameters for j particles */
1104             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1105                                                                  charge+jnrC+0,charge+jnrD+0);
1106             vdwjidx0A        = 2*vdwtype[jnrA+0];
1107             vdwjidx0B        = 2*vdwtype[jnrB+0];
1108             vdwjidx0C        = 2*vdwtype[jnrC+0];
1109             vdwjidx0D        = 2*vdwtype[jnrD+0];
1110
1111             fjx0             = _mm256_setzero_pd();
1112             fjy0             = _mm256_setzero_pd();
1113             fjz0             = _mm256_setzero_pd();
1114
1115             /**************************
1116              * CALCULATE INTERACTIONS *
1117              **************************/
1118
1119             r00              = _mm256_mul_pd(rsq00,rinv00);
1120             r00              = _mm256_andnot_pd(dummy_mask,r00);
1121
1122             /* Compute parameters for interactions between i and j atoms */
1123             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1124                                             vdwioffsetptr0+vdwjidx0B,
1125                                             vdwioffsetptr0+vdwjidx0C,
1126                                             vdwioffsetptr0+vdwjidx0D,
1127                                             &c6_00,&c12_00);
1128
1129             /* Calculate table index by multiplying r with table scale and truncate to integer */
1130             rt               = _mm256_mul_pd(r00,vftabscale);
1131             vfitab           = _mm256_cvttpd_epi32(rt);
1132             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1133             vfitab           = _mm_slli_epi32(vfitab,3);
1134
1135             /* CUBIC SPLINE TABLE DISPERSION */
1136             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1137             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1138             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1139             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1140             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1141             Heps             = _mm256_mul_pd(vfeps,H);
1142             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1143             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1144             fvdw6            = _mm256_mul_pd(c6_00,FF);
1145
1146             /* CUBIC SPLINE TABLE REPULSION */
1147             vfitab           = _mm_add_epi32(vfitab,ifour);
1148             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1149             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1150             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1151             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1152             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1153             Heps             = _mm256_mul_pd(vfeps,H);
1154             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1155             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1156             fvdw12           = _mm256_mul_pd(c12_00,FF);
1157             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1158
1159             fscal            = fvdw;
1160
1161             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1162
1163             /* Calculate temporary vectorial force */
1164             tx               = _mm256_mul_pd(fscal,dx00);
1165             ty               = _mm256_mul_pd(fscal,dy00);
1166             tz               = _mm256_mul_pd(fscal,dz00);
1167
1168             /* Update vectorial force */
1169             fix0             = _mm256_add_pd(fix0,tx);
1170             fiy0             = _mm256_add_pd(fiy0,ty);
1171             fiz0             = _mm256_add_pd(fiz0,tz);
1172
1173             fjx0             = _mm256_add_pd(fjx0,tx);
1174             fjy0             = _mm256_add_pd(fjy0,ty);
1175             fjz0             = _mm256_add_pd(fjz0,tz);
1176
1177             /**************************
1178              * CALCULATE INTERACTIONS *
1179              **************************/
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             qq10             = _mm256_mul_pd(iq1,jq0);
1183
1184             /* COULOMB ELECTROSTATICS */
1185             velec            = _mm256_mul_pd(qq10,rinv10);
1186             felec            = _mm256_mul_pd(velec,rinvsq10);
1187
1188             fscal            = felec;
1189
1190             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1191
1192             /* Calculate temporary vectorial force */
1193             tx               = _mm256_mul_pd(fscal,dx10);
1194             ty               = _mm256_mul_pd(fscal,dy10);
1195             tz               = _mm256_mul_pd(fscal,dz10);
1196
1197             /* Update vectorial force */
1198             fix1             = _mm256_add_pd(fix1,tx);
1199             fiy1             = _mm256_add_pd(fiy1,ty);
1200             fiz1             = _mm256_add_pd(fiz1,tz);
1201
1202             fjx0             = _mm256_add_pd(fjx0,tx);
1203             fjy0             = _mm256_add_pd(fjy0,ty);
1204             fjz0             = _mm256_add_pd(fjz0,tz);
1205
1206             /**************************
1207              * CALCULATE INTERACTIONS *
1208              **************************/
1209
1210             /* Compute parameters for interactions between i and j atoms */
1211             qq20             = _mm256_mul_pd(iq2,jq0);
1212
1213             /* COULOMB ELECTROSTATICS */
1214             velec            = _mm256_mul_pd(qq20,rinv20);
1215             felec            = _mm256_mul_pd(velec,rinvsq20);
1216
1217             fscal            = felec;
1218
1219             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1220
1221             /* Calculate temporary vectorial force */
1222             tx               = _mm256_mul_pd(fscal,dx20);
1223             ty               = _mm256_mul_pd(fscal,dy20);
1224             tz               = _mm256_mul_pd(fscal,dz20);
1225
1226             /* Update vectorial force */
1227             fix2             = _mm256_add_pd(fix2,tx);
1228             fiy2             = _mm256_add_pd(fiy2,ty);
1229             fiz2             = _mm256_add_pd(fiz2,tz);
1230
1231             fjx0             = _mm256_add_pd(fjx0,tx);
1232             fjy0             = _mm256_add_pd(fjy0,ty);
1233             fjz0             = _mm256_add_pd(fjz0,tz);
1234
1235             /**************************
1236              * CALCULATE INTERACTIONS *
1237              **************************/
1238
1239             /* Compute parameters for interactions between i and j atoms */
1240             qq30             = _mm256_mul_pd(iq3,jq0);
1241
1242             /* COULOMB ELECTROSTATICS */
1243             velec            = _mm256_mul_pd(qq30,rinv30);
1244             felec            = _mm256_mul_pd(velec,rinvsq30);
1245
1246             fscal            = felec;
1247
1248             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1249
1250             /* Calculate temporary vectorial force */
1251             tx               = _mm256_mul_pd(fscal,dx30);
1252             ty               = _mm256_mul_pd(fscal,dy30);
1253             tz               = _mm256_mul_pd(fscal,dz30);
1254
1255             /* Update vectorial force */
1256             fix3             = _mm256_add_pd(fix3,tx);
1257             fiy3             = _mm256_add_pd(fiy3,ty);
1258             fiz3             = _mm256_add_pd(fiz3,tz);
1259
1260             fjx0             = _mm256_add_pd(fjx0,tx);
1261             fjy0             = _mm256_add_pd(fjy0,ty);
1262             fjz0             = _mm256_add_pd(fjz0,tz);
1263
1264             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1265             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1266             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1267             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1268
1269             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1270
1271             /* Inner loop uses 130 flops */
1272         }
1273
1274         /* End of innermost loop */
1275
1276         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1277                                                  f+i_coord_offset,fshift+i_shift_offset);
1278
1279         /* Increment number of inner iterations */
1280         inneriter                  += j_index_end - j_index_start;
1281
1282         /* Outer loop uses 24 flops */
1283     }
1284
1285     /* Increment number of outer iterations */
1286     outeriter        += nri;
1287
1288     /* Update outer/inner flops */
1289
1290     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1291 }