src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __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->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     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 137     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 138     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 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_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 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
 184         /* Reset potential sums */
 185         velecsum         = _mm256_setzero_pd();
 186         vvdwsum          = _mm256_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_pd(ix0,jx0);
 209             dy00             = _mm256_sub_pd(iy0,jy0);
 210             dz00             = _mm256_sub_pd(iz0,jz0);
 211             dx10             = _mm256_sub_pd(ix1,jx0);
 212             dy10             = _mm256_sub_pd(iy1,jy0);
 213             dz10             = _mm256_sub_pd(iz1,jz0);
 214             dx20             = _mm256_sub_pd(ix2,jx0);
 215             dy20             = _mm256_sub_pd(iy2,jy0);
 216             dz20             = _mm256_sub_pd(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 220             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 221             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 222
 223             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 224             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 225             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 226
 227             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 228             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 229             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 233                                                                  charge+jnrC+0,charge+jnrD+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236             vdwjidx0C        = 2*vdwtype[jnrC+0];
 237             vdwjidx0D        = 2*vdwtype[jnrD+0];
 238
 239             fjx0             = _mm256_setzero_pd();
 240             fjy0             = _mm256_setzero_pd();
 241             fjz0             = _mm256_setzero_pd();
 242
 243             /**************************
 244              * CALCULATE INTERACTIONS *
 245              **************************/
 246
 247             r00              = _mm256_mul_pd(rsq00,rinv00);
 248
 249             /* Compute parameters for interactions between i and j atoms */
 250             qq00             = _mm256_mul_pd(iq0,jq0);
 251             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 252                                             vdwioffsetptr0+vdwjidx0B,
 253                                             vdwioffsetptr0+vdwjidx0C,
 254                                             vdwioffsetptr0+vdwjidx0D,
 255                                             &c6_00,&c12_00);
 256
 257             /* Calculate table index by multiplying r with table scale and truncate to integer */
 258             rt               = _mm256_mul_pd(r00,vftabscale);
 259             vfitab           = _mm256_cvttpd_epi32(rt);
 260             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 261             vfitab           = _mm_slli_epi32(vfitab,3);
 262
 263             /* COULOMB ELECTROSTATICS */
 264             velec            = _mm256_mul_pd(qq00,rinv00);
 265             felec            = _mm256_mul_pd(velec,rinvsq00);
 266
 267             /* CUBIC SPLINE TABLE DISPERSION */
 268             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 269             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 270             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 271             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 272             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 273             Heps             = _mm256_mul_pd(vfeps,H);
 274             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 275             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 276             vvdw6            = _mm256_mul_pd(c6_00,VV);
 277             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 278             fvdw6            = _mm256_mul_pd(c6_00,FF);
 279
 280             /* CUBIC SPLINE TABLE REPULSION */
 281             vfitab           = _mm_add_epi32(vfitab,ifour);
 282             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 283             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 284             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 285             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 286             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 287             Heps             = _mm256_mul_pd(vfeps,H);
 288             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 289             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 290             vvdw12           = _mm256_mul_pd(c12_00,VV);
 291             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 292             fvdw12           = _mm256_mul_pd(c12_00,FF);
 293             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 294             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velecsum         = _mm256_add_pd(velecsum,velec);
 298             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 299
 300             fscal            = _mm256_add_pd(felec,fvdw);
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm256_mul_pd(fscal,dx00);
 304             ty               = _mm256_mul_pd(fscal,dy00);
 305             tz               = _mm256_mul_pd(fscal,dz00);
 306
 307             /* Update vectorial force */
 308             fix0             = _mm256_add_pd(fix0,tx);
 309             fiy0             = _mm256_add_pd(fiy0,ty);
 310             fiz0             = _mm256_add_pd(fiz0,tz);
 311
 312             fjx0             = _mm256_add_pd(fjx0,tx);
 313             fjy0             = _mm256_add_pd(fjy0,ty);
 314             fjz0             = _mm256_add_pd(fjz0,tz);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm256_mul_pd(iq1,jq0);
 322
 323             /* COULOMB ELECTROSTATICS */
 324             velec            = _mm256_mul_pd(qq10,rinv10);
 325             felec            = _mm256_mul_pd(velec,rinvsq10);
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm256_add_pd(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm256_mul_pd(fscal,dx10);
 334             ty               = _mm256_mul_pd(fscal,dy10);
 335             tz               = _mm256_mul_pd(fscal,dz10);
 336
 337             /* Update vectorial force */
 338             fix1             = _mm256_add_pd(fix1,tx);
 339             fiy1             = _mm256_add_pd(fiy1,ty);
 340             fiz1             = _mm256_add_pd(fiz1,tz);
 341
 342             fjx0             = _mm256_add_pd(fjx0,tx);
 343             fjy0             = _mm256_add_pd(fjy0,ty);
 344             fjz0             = _mm256_add_pd(fjz0,tz);
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm256_mul_pd(iq2,jq0);
 352
 353             /* COULOMB ELECTROSTATICS */
 354             velec            = _mm256_mul_pd(qq20,rinv20);
 355             felec            = _mm256_mul_pd(velec,rinvsq20);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velecsum         = _mm256_add_pd(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             /* Calculate temporary vectorial force */
 363             tx               = _mm256_mul_pd(fscal,dx20);
 364             ty               = _mm256_mul_pd(fscal,dy20);
 365             tz               = _mm256_mul_pd(fscal,dz20);
 366
 367             /* Update vectorial force */
 368             fix2             = _mm256_add_pd(fix2,tx);
 369             fiy2             = _mm256_add_pd(fiy2,ty);
 370             fiz2             = _mm256_add_pd(fiz2,tz);
 371
 372             fjx0             = _mm256_add_pd(fjx0,tx);
 373             fjy0             = _mm256_add_pd(fjy0,ty);
 374             fjz0             = _mm256_add_pd(fjz0,tz);
 375
 376             fjptrA             = f+j_coord_offsetA;
 377             fjptrB             = f+j_coord_offsetB;
 378             fjptrC             = f+j_coord_offsetC;
 379             fjptrD             = f+j_coord_offsetD;
 380
 381             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 382
 383             /* Inner loop uses 119 flops */
 384         }
 385
 386         if(jidx<j_index_end)
 387         {
 388
 389             /* Get j neighbor index, and coordinate index */
 390             jnrlistA         = jjnr[jidx];
 391             jnrlistB         = jjnr[jidx+1];
 392             jnrlistC         = jjnr[jidx+2];
 393             jnrlistD         = jjnr[jidx+3];
 394             /* Sign of each element will be negative for non-real atoms.
 395              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 396              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 397              */
 398             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 399
 400             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 401             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 402             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 403
 404             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 405             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 406             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 407             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 408             j_coord_offsetA  = DIM*jnrA;
 409             j_coord_offsetB  = DIM*jnrB;
 410             j_coord_offsetC  = DIM*jnrC;
 411             j_coord_offsetD  = DIM*jnrD;
 412
 413             /* load j atom coordinates */
 414             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 415                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 416                                                  &jx0,&jy0,&jz0);
 417
 418             /* Calculate displacement vector */
 419             dx00             = _mm256_sub_pd(ix0,jx0);
 420             dy00             = _mm256_sub_pd(iy0,jy0);
 421             dz00             = _mm256_sub_pd(iz0,jz0);
 422             dx10             = _mm256_sub_pd(ix1,jx0);
 423             dy10             = _mm256_sub_pd(iy1,jy0);
 424             dz10             = _mm256_sub_pd(iz1,jz0);
 425             dx20             = _mm256_sub_pd(ix2,jx0);
 426             dy20             = _mm256_sub_pd(iy2,jy0);
 427             dz20             = _mm256_sub_pd(iz2,jz0);
 428
 429             /* Calculate squared distance and things based on it */
 430             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 431             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 432             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 433
 434             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 435             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 436             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 437
 438             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 439             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 440             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 441
 442             /* Load parameters for j particles */
 443             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 444                                                                  charge+jnrC+0,charge+jnrD+0);
 445             vdwjidx0A        = 2*vdwtype[jnrA+0];
 446             vdwjidx0B        = 2*vdwtype[jnrB+0];
 447             vdwjidx0C        = 2*vdwtype[jnrC+0];
 448             vdwjidx0D        = 2*vdwtype[jnrD+0];
 449
 450             fjx0             = _mm256_setzero_pd();
 451             fjy0             = _mm256_setzero_pd();
 452             fjz0             = _mm256_setzero_pd();
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             r00              = _mm256_mul_pd(rsq00,rinv00);
 459             r00              = _mm256_andnot_pd(dummy_mask,r00);
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             qq00             = _mm256_mul_pd(iq0,jq0);
 463             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 464                                             vdwioffsetptr0+vdwjidx0B,
 465                                             vdwioffsetptr0+vdwjidx0C,
 466                                             vdwioffsetptr0+vdwjidx0D,
 467                                             &c6_00,&c12_00);
 468
 469             /* Calculate table index by multiplying r with table scale and truncate to integer */
 470             rt               = _mm256_mul_pd(r00,vftabscale);
 471             vfitab           = _mm256_cvttpd_epi32(rt);
 472             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 473             vfitab           = _mm_slli_epi32(vfitab,3);
 474
 475             /* COULOMB ELECTROSTATICS */
 476             velec            = _mm256_mul_pd(qq00,rinv00);
 477             felec            = _mm256_mul_pd(velec,rinvsq00);
 478
 479             /* CUBIC SPLINE TABLE DISPERSION */
 480             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 481             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 482             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 483             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 484             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 485             Heps             = _mm256_mul_pd(vfeps,H);
 486             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 487             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 488             vvdw6            = _mm256_mul_pd(c6_00,VV);
 489             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 490             fvdw6            = _mm256_mul_pd(c6_00,FF);
 491
 492             /* CUBIC SPLINE TABLE REPULSION */
 493             vfitab           = _mm_add_epi32(vfitab,ifour);
 494             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 495             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 496             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 497             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 498             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 499             Heps             = _mm256_mul_pd(vfeps,H);
 500             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 501             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 502             vvdw12           = _mm256_mul_pd(c12_00,VV);
 503             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 504             fvdw12           = _mm256_mul_pd(c12_00,FF);
 505             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 506             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 507
 508             /* Update potential sum for this i atom from the interaction with this j atom. */
 509             velec            = _mm256_andnot_pd(dummy_mask,velec);
 510             velecsum         = _mm256_add_pd(velecsum,velec);
 511             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 512             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 513
 514             fscal            = _mm256_add_pd(felec,fvdw);
 515
 516             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm256_mul_pd(fscal,dx00);
 520             ty               = _mm256_mul_pd(fscal,dy00);
 521             tz               = _mm256_mul_pd(fscal,dz00);
 522
 523             /* Update vectorial force */
 524             fix0             = _mm256_add_pd(fix0,tx);
 525             fiy0             = _mm256_add_pd(fiy0,ty);
 526             fiz0             = _mm256_add_pd(fiz0,tz);
 527
 528             fjx0             = _mm256_add_pd(fjx0,tx);
 529             fjy0             = _mm256_add_pd(fjy0,ty);
 530             fjz0             = _mm256_add_pd(fjz0,tz);
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq10             = _mm256_mul_pd(iq1,jq0);
 538
 539             /* COULOMB ELECTROSTATICS */
 540             velec            = _mm256_mul_pd(qq10,rinv10);
 541             felec            = _mm256_mul_pd(velec,rinvsq10);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm256_andnot_pd(dummy_mask,velec);
 545             velecsum         = _mm256_add_pd(velecsum,velec);
 546
 547             fscal            = felec;
 548
 549             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 550
 551             /* Calculate temporary vectorial force */
 552             tx               = _mm256_mul_pd(fscal,dx10);
 553             ty               = _mm256_mul_pd(fscal,dy10);
 554             tz               = _mm256_mul_pd(fscal,dz10);
 555
 556             /* Update vectorial force */
 557             fix1             = _mm256_add_pd(fix1,tx);
 558             fiy1             = _mm256_add_pd(fiy1,ty);
 559             fiz1             = _mm256_add_pd(fiz1,tz);
 560
 561             fjx0             = _mm256_add_pd(fjx0,tx);
 562             fjy0             = _mm256_add_pd(fjy0,ty);
 563             fjz0             = _mm256_add_pd(fjz0,tz);
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             /* Compute parameters for interactions between i and j atoms */
 570             qq20             = _mm256_mul_pd(iq2,jq0);
 571
 572             /* COULOMB ELECTROSTATICS */
 573             velec            = _mm256_mul_pd(qq20,rinv20);
 574             felec            = _mm256_mul_pd(velec,rinvsq20);
 575
 576             /* Update potential sum for this i atom from the interaction with this j atom. */
 577             velec            = _mm256_andnot_pd(dummy_mask,velec);
 578             velecsum         = _mm256_add_pd(velecsum,velec);
 579
 580             fscal            = felec;
 581
 582             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 583
 584             /* Calculate temporary vectorial force */
 585             tx               = _mm256_mul_pd(fscal,dx20);
 586             ty               = _mm256_mul_pd(fscal,dy20);
 587             tz               = _mm256_mul_pd(fscal,dz20);
 588
 589             /* Update vectorial force */
 590             fix2             = _mm256_add_pd(fix2,tx);
 591             fiy2             = _mm256_add_pd(fiy2,ty);
 592             fiz2             = _mm256_add_pd(fiz2,tz);
 593
 594             fjx0             = _mm256_add_pd(fjx0,tx);
 595             fjy0             = _mm256_add_pd(fjy0,ty);
 596             fjz0             = _mm256_add_pd(fjz0,tz);
 597
 598             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 599             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 600             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 601             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 602
 603             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 604
 605             /* Inner loop uses 120 flops */
 606         }
 607
 608         /* End of innermost loop */
 609
 610         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 611                                                  f+i_coord_offset,fshift+i_shift_offset);
 612
 613         ggid                        = gid[iidx];
 614         /* Update potential energies */
 615         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 616         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 617
 618         /* Increment number of inner iterations */
 619         inneriter                  += j_index_end - j_index_start;
 620
 621         /* Outer loop uses 20 flops */
 622     }
 623
 624     /* Increment number of outer iterations */
 625     outeriter        += nri;
 626
 627     /* Update outer/inner flops */
 628
 629     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 630 }
 631 /*
 632  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_double
 633  * Electrostatics interaction: Coulomb
 634  * VdW interaction:            CubicSplineTable
 635  * Geometry:                   Water3-Particle
 636  * Calculate force/pot:        Force
 637  */
 638 void
 639 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_avx_256_double
 640                     (t_nblist                    * gmx_restrict       nlist,
 641                      rvec                        * gmx_restrict          xx,
 642                      rvec                        * gmx_restrict          ff,
 643                      t_forcerec                  * gmx_restrict          fr,
 644                      t_mdatoms                   * gmx_restrict     mdatoms,
 645                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 646                      t_nrnb                      * gmx_restrict        nrnb)
 647 {
 648     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 649      * just 0 for non-waters.
 650      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 651      * jnr indices corresponding to data put in the four positions in the SIMD register.
 652      */
 653     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 654     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 655     int              jnrA,jnrB,jnrC,jnrD;
 656     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 657     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 658     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 659     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 660     real             rcutoff_scalar;
 661     real             *shiftvec,*fshift,*x,*f;
 662     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 663     real             scratch[4*DIM];
 664     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 665     real *           vdwioffsetptr0;
 666     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 667     real *           vdwioffsetptr1;
 668     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 669     real *           vdwioffsetptr2;
 670     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 671     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 672     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 673     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 674     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 675     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 676     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 677     real             *charge;
 678     int              nvdwtype;
 679     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 680     int              *vdwtype;
 681     real             *vdwparam;
 682     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 683     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 684     __m128i          vfitab;
 685     __m128i          ifour       = _mm_set1_epi32(4);
 686     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 687     real             *vftab;
 688     __m256d          dummy_mask,cutoff_mask;
 689     __m128           tmpmask0,tmpmask1;
 690     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 691     __m256d          one     = _mm256_set1_pd(1.0);
 692     __m256d          two     = _mm256_set1_pd(2.0);
 693     x                = xx[0];
 694     f                = ff[0];
 695
 696     nri              = nlist->nri;
 697     iinr             = nlist->iinr;
 698     jindex           = nlist->jindex;
 699     jjnr             = nlist->jjnr;
 700     shiftidx         = nlist->shift;
 701     gid              = nlist->gid;
 702     shiftvec         = fr->shift_vec[0];
 703     fshift           = fr->fshift[0];
 704     facel            = _mm256_set1_pd(fr->epsfac);
 705     charge           = mdatoms->chargeA;
 706     nvdwtype         = fr->ntype;
 707     vdwparam         = fr->nbfp;
 708     vdwtype          = mdatoms->typeA;
 709
 710     vftab            = kernel_data->table_vdw->data;
 711     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 712
 713     /* Setup water-specific parameters */
 714     inr              = nlist->iinr[0];
 715     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 716     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 717     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 718     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 719
 720     /* Avoid stupid compiler warnings */
 721     jnrA = jnrB = jnrC = jnrD = 0;
 722     j_coord_offsetA = 0;
 723     j_coord_offsetB = 0;
 724     j_coord_offsetC = 0;
 725     j_coord_offsetD = 0;
 726
 727     outeriter        = 0;
 728     inneriter        = 0;
 729
 730     for(iidx=0;iidx<4*DIM;iidx++)
 731     {
 732         scratch[iidx] = 0.0;
 733     }
 734
 735     /* Start outer loop over neighborlists */
 736     for(iidx=0; iidx<nri; iidx++)
 737     {
 738         /* Load shift vector for this list */
 739         i_shift_offset   = DIM*shiftidx[iidx];
 740
 741         /* Load limits for loop over neighbors */
 742         j_index_start    = jindex[iidx];
 743         j_index_end      = jindex[iidx+1];
 744
 745         /* Get outer coordinate index */
 746         inr              = iinr[iidx];
 747         i_coord_offset   = DIM*inr;
 748
 749         /* Load i particle coords and add shift vector */
 750         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 751                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 752
 753         fix0             = _mm256_setzero_pd();
 754         fiy0             = _mm256_setzero_pd();
 755         fiz0             = _mm256_setzero_pd();
 756         fix1             = _mm256_setzero_pd();
 757         fiy1             = _mm256_setzero_pd();
 758         fiz1             = _mm256_setzero_pd();
 759         fix2             = _mm256_setzero_pd();
 760         fiy2             = _mm256_setzero_pd();
 761         fiz2             = _mm256_setzero_pd();
 762
 763         /* Start inner kernel loop */
 764         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 765         {
 766
 767             /* Get j neighbor index, and coordinate index */
 768             jnrA             = jjnr[jidx];
 769             jnrB             = jjnr[jidx+1];
 770             jnrC             = jjnr[jidx+2];
 771             jnrD             = jjnr[jidx+3];
 772             j_coord_offsetA  = DIM*jnrA;
 773             j_coord_offsetB  = DIM*jnrB;
 774             j_coord_offsetC  = DIM*jnrC;
 775             j_coord_offsetD  = DIM*jnrD;
 776
 777             /* load j atom coordinates */
 778             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 779                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 780                                                  &jx0,&jy0,&jz0);
 781
 782             /* Calculate displacement vector */
 783             dx00             = _mm256_sub_pd(ix0,jx0);
 784             dy00             = _mm256_sub_pd(iy0,jy0);
 785             dz00             = _mm256_sub_pd(iz0,jz0);
 786             dx10             = _mm256_sub_pd(ix1,jx0);
 787             dy10             = _mm256_sub_pd(iy1,jy0);
 788             dz10             = _mm256_sub_pd(iz1,jz0);
 789             dx20             = _mm256_sub_pd(ix2,jx0);
 790             dy20             = _mm256_sub_pd(iy2,jy0);
 791             dz20             = _mm256_sub_pd(iz2,jz0);
 792
 793             /* Calculate squared distance and things based on it */
 794             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 795             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 796             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 797
 798             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 799             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 800             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 801
 802             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 803             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 804             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 805
 806             /* Load parameters for j particles */
 807             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 808                                                                  charge+jnrC+0,charge+jnrD+0);
 809             vdwjidx0A        = 2*vdwtype[jnrA+0];
 810             vdwjidx0B        = 2*vdwtype[jnrB+0];
 811             vdwjidx0C        = 2*vdwtype[jnrC+0];
 812             vdwjidx0D        = 2*vdwtype[jnrD+0];
 813
 814             fjx0             = _mm256_setzero_pd();
 815             fjy0             = _mm256_setzero_pd();
 816             fjz0             = _mm256_setzero_pd();
 817
 818             /**************************
 819              * CALCULATE INTERACTIONS *
 820              **************************/
 821
 822             r00              = _mm256_mul_pd(rsq00,rinv00);
 823
 824             /* Compute parameters for interactions between i and j atoms */
 825             qq00             = _mm256_mul_pd(iq0,jq0);
 826             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 827                                             vdwioffsetptr0+vdwjidx0B,
 828                                             vdwioffsetptr0+vdwjidx0C,
 829                                             vdwioffsetptr0+vdwjidx0D,
 830                                             &c6_00,&c12_00);
 831
 832             /* Calculate table index by multiplying r with table scale and truncate to integer */
 833             rt               = _mm256_mul_pd(r00,vftabscale);
 834             vfitab           = _mm256_cvttpd_epi32(rt);
 835             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 836             vfitab           = _mm_slli_epi32(vfitab,3);
 837
 838             /* COULOMB ELECTROSTATICS */
 839             velec            = _mm256_mul_pd(qq00,rinv00);
 840             felec            = _mm256_mul_pd(velec,rinvsq00);
 841
 842             /* CUBIC SPLINE TABLE DISPERSION */
 843             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 844             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 845             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 846             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 847             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 848             Heps             = _mm256_mul_pd(vfeps,H);
 849             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 850             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 851             fvdw6            = _mm256_mul_pd(c6_00,FF);
 852
 853             /* CUBIC SPLINE TABLE REPULSION */
 854             vfitab           = _mm_add_epi32(vfitab,ifour);
 855             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 856             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 857             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 858             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 859             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 860             Heps             = _mm256_mul_pd(vfeps,H);
 861             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 862             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 863             fvdw12           = _mm256_mul_pd(c12_00,FF);
 864             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 865
 866             fscal            = _mm256_add_pd(felec,fvdw);
 867
 868             /* Calculate temporary vectorial force */
 869             tx               = _mm256_mul_pd(fscal,dx00);
 870             ty               = _mm256_mul_pd(fscal,dy00);
 871             tz               = _mm256_mul_pd(fscal,dz00);
 872
 873             /* Update vectorial force */
 874             fix0             = _mm256_add_pd(fix0,tx);
 875             fiy0             = _mm256_add_pd(fiy0,ty);
 876             fiz0             = _mm256_add_pd(fiz0,tz);
 877
 878             fjx0             = _mm256_add_pd(fjx0,tx);
 879             fjy0             = _mm256_add_pd(fjy0,ty);
 880             fjz0             = _mm256_add_pd(fjz0,tz);
 881
 882             /**************************
 883              * CALCULATE INTERACTIONS *
 884              **************************/
 885
 886             /* Compute parameters for interactions between i and j atoms */
 887             qq10             = _mm256_mul_pd(iq1,jq0);
 888
 889             /* COULOMB ELECTROSTATICS */
 890             velec            = _mm256_mul_pd(qq10,rinv10);
 891             felec            = _mm256_mul_pd(velec,rinvsq10);
 892
 893             fscal            = felec;
 894
 895             /* Calculate temporary vectorial force */
 896             tx               = _mm256_mul_pd(fscal,dx10);
 897             ty               = _mm256_mul_pd(fscal,dy10);
 898             tz               = _mm256_mul_pd(fscal,dz10);
 899
 900             /* Update vectorial force */
 901             fix1             = _mm256_add_pd(fix1,tx);
 902             fiy1             = _mm256_add_pd(fiy1,ty);
 903             fiz1             = _mm256_add_pd(fiz1,tz);
 904
 905             fjx0             = _mm256_add_pd(fjx0,tx);
 906             fjy0             = _mm256_add_pd(fjy0,ty);
 907             fjz0             = _mm256_add_pd(fjz0,tz);
 908
 909             /**************************
 910              * CALCULATE INTERACTIONS *
 911              **************************/
 912
 913             /* Compute parameters for interactions between i and j atoms */
 914             qq20             = _mm256_mul_pd(iq2,jq0);
 915
 916             /* COULOMB ELECTROSTATICS */
 917             velec            = _mm256_mul_pd(qq20,rinv20);
 918             felec            = _mm256_mul_pd(velec,rinvsq20);
 919
 920             fscal            = felec;
 921
 922             /* Calculate temporary vectorial force */
 923             tx               = _mm256_mul_pd(fscal,dx20);
 924             ty               = _mm256_mul_pd(fscal,dy20);
 925             tz               = _mm256_mul_pd(fscal,dz20);
 926
 927             /* Update vectorial force */
 928             fix2             = _mm256_add_pd(fix2,tx);
 929             fiy2             = _mm256_add_pd(fiy2,ty);
 930             fiz2             = _mm256_add_pd(fiz2,tz);
 931
 932             fjx0             = _mm256_add_pd(fjx0,tx);
 933             fjy0             = _mm256_add_pd(fjy0,ty);
 934             fjz0             = _mm256_add_pd(fjz0,tz);
 935
 936             fjptrA             = f+j_coord_offsetA;
 937             fjptrB             = f+j_coord_offsetB;
 938             fjptrC             = f+j_coord_offsetC;
 939             fjptrD             = f+j_coord_offsetD;
 940
 941             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 942
 943             /* Inner loop uses 108 flops */
 944         }
 945
 946         if(jidx<j_index_end)
 947         {
 948
 949             /* Get j neighbor index, and coordinate index */
 950             jnrlistA         = jjnr[jidx];
 951             jnrlistB         = jjnr[jidx+1];
 952             jnrlistC         = jjnr[jidx+2];
 953             jnrlistD         = jjnr[jidx+3];
 954             /* Sign of each element will be negative for non-real atoms.
 955              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 956              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 957              */
 958             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 959
 960             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 961             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 962             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 963
 964             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 965             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 966             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 967             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 968             j_coord_offsetA  = DIM*jnrA;
 969             j_coord_offsetB  = DIM*jnrB;
 970             j_coord_offsetC  = DIM*jnrC;
 971             j_coord_offsetD  = DIM*jnrD;
 972
 973             /* load j atom coordinates */
 974             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 975                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 976                                                  &jx0,&jy0,&jz0);
 977
 978             /* Calculate displacement vector */
 979             dx00             = _mm256_sub_pd(ix0,jx0);
 980             dy00             = _mm256_sub_pd(iy0,jy0);
 981             dz00             = _mm256_sub_pd(iz0,jz0);
 982             dx10             = _mm256_sub_pd(ix1,jx0);
 983             dy10             = _mm256_sub_pd(iy1,jy0);
 984             dz10             = _mm256_sub_pd(iz1,jz0);
 985             dx20             = _mm256_sub_pd(ix2,jx0);
 986             dy20             = _mm256_sub_pd(iy2,jy0);
 987             dz20             = _mm256_sub_pd(iz2,jz0);
 988
 989             /* Calculate squared distance and things based on it */
 990             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 991             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 992             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 993
 994             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 995             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 996             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 997
 998             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 999             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1000             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1001
1002             /* Load parameters for j particles */
1003             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1004                                                                  charge+jnrC+0,charge+jnrD+0);
1005             vdwjidx0A        = 2*vdwtype[jnrA+0];
1006             vdwjidx0B        = 2*vdwtype[jnrB+0];
1007             vdwjidx0C        = 2*vdwtype[jnrC+0];
1008             vdwjidx0D        = 2*vdwtype[jnrD+0];
1009
1010             fjx0             = _mm256_setzero_pd();
1011             fjy0             = _mm256_setzero_pd();
1012             fjz0             = _mm256_setzero_pd();
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             r00              = _mm256_mul_pd(rsq00,rinv00);
1019             r00              = _mm256_andnot_pd(dummy_mask,r00);
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq00             = _mm256_mul_pd(iq0,jq0);
1023             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1024                                             vdwioffsetptr0+vdwjidx0B,
1025                                             vdwioffsetptr0+vdwjidx0C,
1026                                             vdwioffsetptr0+vdwjidx0D,
1027                                             &c6_00,&c12_00);
1028
1029             /* Calculate table index by multiplying r with table scale and truncate to integer */
1030             rt               = _mm256_mul_pd(r00,vftabscale);
1031             vfitab           = _mm256_cvttpd_epi32(rt);
1032             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1033             vfitab           = _mm_slli_epi32(vfitab,3);
1034
1035             /* COULOMB ELECTROSTATICS */
1036             velec            = _mm256_mul_pd(qq00,rinv00);
1037             felec            = _mm256_mul_pd(velec,rinvsq00);
1038
1039             /* CUBIC SPLINE TABLE DISPERSION */
1040             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1041             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1042             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1043             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1044             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1045             Heps             = _mm256_mul_pd(vfeps,H);
1046             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1047             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1048             fvdw6            = _mm256_mul_pd(c6_00,FF);
1049
1050             /* CUBIC SPLINE TABLE REPULSION */
1051             vfitab           = _mm_add_epi32(vfitab,ifour);
1052             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1053             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1054             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1055             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1056             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1057             Heps             = _mm256_mul_pd(vfeps,H);
1058             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1059             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1060             fvdw12           = _mm256_mul_pd(c12_00,FF);
1061             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1062
1063             fscal            = _mm256_add_pd(felec,fvdw);
1064
1065             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm256_mul_pd(fscal,dx00);
1069             ty               = _mm256_mul_pd(fscal,dy00);
1070             tz               = _mm256_mul_pd(fscal,dz00);
1071
1072             /* Update vectorial force */
1073             fix0             = _mm256_add_pd(fix0,tx);
1074             fiy0             = _mm256_add_pd(fiy0,ty);
1075             fiz0             = _mm256_add_pd(fiz0,tz);
1076
1077             fjx0             = _mm256_add_pd(fjx0,tx);
1078             fjy0             = _mm256_add_pd(fjy0,ty);
1079             fjz0             = _mm256_add_pd(fjz0,tz);
1080
1081             /**************************
1082              * CALCULATE INTERACTIONS *
1083              **************************/
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             qq10             = _mm256_mul_pd(iq1,jq0);
1087
1088             /* COULOMB ELECTROSTATICS */
1089             velec            = _mm256_mul_pd(qq10,rinv10);
1090             felec            = _mm256_mul_pd(velec,rinvsq10);
1091
1092             fscal            = felec;
1093
1094             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1095
1096             /* Calculate temporary vectorial force */
1097             tx               = _mm256_mul_pd(fscal,dx10);
1098             ty               = _mm256_mul_pd(fscal,dy10);
1099             tz               = _mm256_mul_pd(fscal,dz10);
1100
1101             /* Update vectorial force */
1102             fix1             = _mm256_add_pd(fix1,tx);
1103             fiy1             = _mm256_add_pd(fiy1,ty);
1104             fiz1             = _mm256_add_pd(fiz1,tz);
1105
1106             fjx0             = _mm256_add_pd(fjx0,tx);
1107             fjy0             = _mm256_add_pd(fjy0,ty);
1108             fjz0             = _mm256_add_pd(fjz0,tz);
1109
1110             /**************************
1111              * CALCULATE INTERACTIONS *
1112              **************************/
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             qq20             = _mm256_mul_pd(iq2,jq0);
1116
1117             /* COULOMB ELECTROSTATICS */
1118             velec            = _mm256_mul_pd(qq20,rinv20);
1119             felec            = _mm256_mul_pd(velec,rinvsq20);
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,dx20);
1127             ty               = _mm256_mul_pd(fscal,dy20);
1128             tz               = _mm256_mul_pd(fscal,dz20);
1129
1130             /* Update vectorial force */
1131             fix2             = _mm256_add_pd(fix2,tx);
1132             fiy2             = _mm256_add_pd(fiy2,ty);
1133             fiz2             = _mm256_add_pd(fiz2,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 109 flops */
1147         }
1148
1149         /* End of innermost loop */
1150
1151         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
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 18 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_W3_F,outeriter*18 + inneriter*109);
1166 }