src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_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     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 101     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 102     __m128i          vfitab;
 103     __m128i          ifour       = _mm_set1_epi32(4);
 104     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 105     real             *vftab;
 106     __m256d          dummy_mask,cutoff_mask;
 107     __m128           tmpmask0,tmpmask1;
 108     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 109     __m256d          one     = _mm256_set1_pd(1.0);
 110     __m256d          two     = _mm256_set1_pd(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm256_set1_pd(fr->ic->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     vftab            = kernel_data->table_elec_vdw->data;
 129     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 134     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 135     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 136     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = jnrC = jnrD = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142     j_coord_offsetC = 0;
 143     j_coord_offsetD = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     for(iidx=0;iidx<4*DIM;iidx++)
 149     {
 150         scratch[iidx] = 0.0;
 151     }
 152
 153     /* Start outer loop over neighborlists */
 154     for(iidx=0; iidx<nri; iidx++)
 155     {
 156         /* Load shift vector for this list */
 157         i_shift_offset   = DIM*shiftidx[iidx];
 158
 159         /* Load limits for loop over neighbors */
 160         j_index_start    = jindex[iidx];
 161         j_index_end      = jindex[iidx+1];
 162
 163         /* Get outer coordinate index */
 164         inr              = iinr[iidx];
 165         i_coord_offset   = DIM*inr;
 166
 167         /* Load i particle coords and add shift vector */
 168         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 169                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 170
 171         fix0             = _mm256_setzero_pd();
 172         fiy0             = _mm256_setzero_pd();
 173         fiz0             = _mm256_setzero_pd();
 174         fix1             = _mm256_setzero_pd();
 175         fiy1             = _mm256_setzero_pd();
 176         fiz1             = _mm256_setzero_pd();
 177         fix2             = _mm256_setzero_pd();
 178         fiy2             = _mm256_setzero_pd();
 179         fiz2             = _mm256_setzero_pd();
 180
 181         /* Reset potential sums */
 182         velecsum         = _mm256_setzero_pd();
 183         vvdwsum          = _mm256_setzero_pd();
 184
 185         /* Start inner kernel loop */
 186         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 187         {
 188
 189             /* Get j neighbor index, and coordinate index */
 190             jnrA             = jjnr[jidx];
 191             jnrB             = jjnr[jidx+1];
 192             jnrC             = jjnr[jidx+2];
 193             jnrD             = jjnr[jidx+3];
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 202                                                  &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm256_sub_pd(ix0,jx0);
 206             dy00             = _mm256_sub_pd(iy0,jy0);
 207             dz00             = _mm256_sub_pd(iz0,jz0);
 208             dx10             = _mm256_sub_pd(ix1,jx0);
 209             dy10             = _mm256_sub_pd(iy1,jy0);
 210             dz10             = _mm256_sub_pd(iz1,jz0);
 211             dx20             = _mm256_sub_pd(ix2,jx0);
 212             dy20             = _mm256_sub_pd(iy2,jy0);
 213             dz20             = _mm256_sub_pd(iz2,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 217             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 218             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 219
 220             rinv00           = avx256_invsqrt_d(rsq00);
 221             rinv10           = avx256_invsqrt_d(rsq10);
 222             rinv20           = avx256_invsqrt_d(rsq20);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 226                                                                  charge+jnrC+0,charge+jnrD+0);
 227             vdwjidx0A        = 2*vdwtype[jnrA+0];
 228             vdwjidx0B        = 2*vdwtype[jnrB+0];
 229             vdwjidx0C        = 2*vdwtype[jnrC+0];
 230             vdwjidx0D        = 2*vdwtype[jnrD+0];
 231
 232             fjx0             = _mm256_setzero_pd();
 233             fjy0             = _mm256_setzero_pd();
 234             fjz0             = _mm256_setzero_pd();
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             r00              = _mm256_mul_pd(rsq00,rinv00);
 241
 242             /* Compute parameters for interactions between i and j atoms */
 243             qq00             = _mm256_mul_pd(iq0,jq0);
 244             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 245                                             vdwioffsetptr0+vdwjidx0B,
 246                                             vdwioffsetptr0+vdwjidx0C,
 247                                             vdwioffsetptr0+vdwjidx0D,
 248                                             &c6_00,&c12_00);
 249
 250             /* Calculate table index by multiplying r with table scale and truncate to integer */
 251             rt               = _mm256_mul_pd(r00,vftabscale);
 252             vfitab           = _mm256_cvttpd_epi32(rt);
 253             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 254             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 255
 256             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 257             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 258             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 259             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 260             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 261             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 262             Heps             = _mm256_mul_pd(vfeps,H);
 263             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 264             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 265             velec            = _mm256_mul_pd(qq00,VV);
 266             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 267             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 268
 269             /* CUBIC SPLINE TABLE DISPERSION */
 270             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             velecsum         = _mm256_add_pd(velecsum,velec);
 301             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 302
 303             fscal            = _mm256_add_pd(felec,fvdw);
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm256_mul_pd(fscal,dx00);
 307             ty               = _mm256_mul_pd(fscal,dy00);
 308             tz               = _mm256_mul_pd(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm256_add_pd(fix0,tx);
 312             fiy0             = _mm256_add_pd(fiy0,ty);
 313             fiz0             = _mm256_add_pd(fiz0,tz);
 314
 315             fjx0             = _mm256_add_pd(fjx0,tx);
 316             fjy0             = _mm256_add_pd(fjy0,ty);
 317             fjz0             = _mm256_add_pd(fjz0,tz);
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             r10              = _mm256_mul_pd(rsq10,rinv10);
 324
 325             /* Compute parameters for interactions between i and j atoms */
 326             qq10             = _mm256_mul_pd(iq1,jq0);
 327
 328             /* Calculate table index by multiplying r with table scale and truncate to integer */
 329             rt               = _mm256_mul_pd(r10,vftabscale);
 330             vfitab           = _mm256_cvttpd_epi32(rt);
 331             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 332             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 333
 334             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 335             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 336             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 337             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 338             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 339             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 340             Heps             = _mm256_mul_pd(vfeps,H);
 341             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 342             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 343             velec            = _mm256_mul_pd(qq10,VV);
 344             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 345             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velecsum         = _mm256_add_pd(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm256_mul_pd(fscal,dx10);
 354             ty               = _mm256_mul_pd(fscal,dy10);
 355             tz               = _mm256_mul_pd(fscal,dz10);
 356
 357             /* Update vectorial force */
 358             fix1             = _mm256_add_pd(fix1,tx);
 359             fiy1             = _mm256_add_pd(fiy1,ty);
 360             fiz1             = _mm256_add_pd(fiz1,tz);
 361
 362             fjx0             = _mm256_add_pd(fjx0,tx);
 363             fjy0             = _mm256_add_pd(fjy0,ty);
 364             fjz0             = _mm256_add_pd(fjz0,tz);
 365
 366             /**************************
 367              * CALCULATE INTERACTIONS *
 368              **************************/
 369
 370             r20              = _mm256_mul_pd(rsq20,rinv20);
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq20             = _mm256_mul_pd(iq2,jq0);
 374
 375             /* Calculate table index by multiplying r with table scale and truncate to integer */
 376             rt               = _mm256_mul_pd(r20,vftabscale);
 377             vfitab           = _mm256_cvttpd_epi32(rt);
 378             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 379             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 380
 381             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 382             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 383             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 384             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 385             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 386             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 387             Heps             = _mm256_mul_pd(vfeps,H);
 388             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 389             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 390             velec            = _mm256_mul_pd(qq20,VV);
 391             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 392             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 393
 394             /* Update potential sum for this i atom from the interaction with this j atom. */
 395             velecsum         = _mm256_add_pd(velecsum,velec);
 396
 397             fscal            = felec;
 398
 399             /* Calculate temporary vectorial force */
 400             tx               = _mm256_mul_pd(fscal,dx20);
 401             ty               = _mm256_mul_pd(fscal,dy20);
 402             tz               = _mm256_mul_pd(fscal,dz20);
 403
 404             /* Update vectorial force */
 405             fix2             = _mm256_add_pd(fix2,tx);
 406             fiy2             = _mm256_add_pd(fiy2,ty);
 407             fiz2             = _mm256_add_pd(fiz2,tz);
 408
 409             fjx0             = _mm256_add_pd(fjx0,tx);
 410             fjy0             = _mm256_add_pd(fjy0,ty);
 411             fjz0             = _mm256_add_pd(fjz0,tz);
 412
 413             fjptrA             = f+j_coord_offsetA;
 414             fjptrB             = f+j_coord_offsetB;
 415             fjptrC             = f+j_coord_offsetC;
 416             fjptrD             = f+j_coord_offsetD;
 417
 418             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 419
 420             /* Inner loop uses 162 flops */
 421         }
 422
 423         if(jidx<j_index_end)
 424         {
 425
 426             /* Get j neighbor index, and coordinate index */
 427             jnrlistA         = jjnr[jidx];
 428             jnrlistB         = jjnr[jidx+1];
 429             jnrlistC         = jjnr[jidx+2];
 430             jnrlistD         = jjnr[jidx+3];
 431             /* Sign of each element will be negative for non-real atoms.
 432              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 433              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 434              */
 435             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 436
 437             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 438             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 439             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 440
 441             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 442             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 443             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 444             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 445             j_coord_offsetA  = DIM*jnrA;
 446             j_coord_offsetB  = DIM*jnrB;
 447             j_coord_offsetC  = DIM*jnrC;
 448             j_coord_offsetD  = DIM*jnrD;
 449
 450             /* load j atom coordinates */
 451             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 452                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 453                                                  &jx0,&jy0,&jz0);
 454
 455             /* Calculate displacement vector */
 456             dx00             = _mm256_sub_pd(ix0,jx0);
 457             dy00             = _mm256_sub_pd(iy0,jy0);
 458             dz00             = _mm256_sub_pd(iz0,jz0);
 459             dx10             = _mm256_sub_pd(ix1,jx0);
 460             dy10             = _mm256_sub_pd(iy1,jy0);
 461             dz10             = _mm256_sub_pd(iz1,jz0);
 462             dx20             = _mm256_sub_pd(ix2,jx0);
 463             dy20             = _mm256_sub_pd(iy2,jy0);
 464             dz20             = _mm256_sub_pd(iz2,jz0);
 465
 466             /* Calculate squared distance and things based on it */
 467             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 468             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 469             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 470
 471             rinv00           = avx256_invsqrt_d(rsq00);
 472             rinv10           = avx256_invsqrt_d(rsq10);
 473             rinv20           = avx256_invsqrt_d(rsq20);
 474
 475             /* Load parameters for j particles */
 476             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 477                                                                  charge+jnrC+0,charge+jnrD+0);
 478             vdwjidx0A        = 2*vdwtype[jnrA+0];
 479             vdwjidx0B        = 2*vdwtype[jnrB+0];
 480             vdwjidx0C        = 2*vdwtype[jnrC+0];
 481             vdwjidx0D        = 2*vdwtype[jnrD+0];
 482
 483             fjx0             = _mm256_setzero_pd();
 484             fjy0             = _mm256_setzero_pd();
 485             fjz0             = _mm256_setzero_pd();
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             r00              = _mm256_mul_pd(rsq00,rinv00);
 492             r00              = _mm256_andnot_pd(dummy_mask,r00);
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             qq00             = _mm256_mul_pd(iq0,jq0);
 496             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 497                                             vdwioffsetptr0+vdwjidx0B,
 498                                             vdwioffsetptr0+vdwjidx0C,
 499                                             vdwioffsetptr0+vdwjidx0D,
 500                                             &c6_00,&c12_00);
 501
 502             /* Calculate table index by multiplying r with table scale and truncate to integer */
 503             rt               = _mm256_mul_pd(r00,vftabscale);
 504             vfitab           = _mm256_cvttpd_epi32(rt);
 505             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 506             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 507
 508             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 509             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 510             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 511             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 512             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 513             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 514             Heps             = _mm256_mul_pd(vfeps,H);
 515             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 516             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 517             velec            = _mm256_mul_pd(qq00,VV);
 518             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 519             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 520
 521             /* CUBIC SPLINE TABLE DISPERSION */
 522             vfitab           = _mm_add_epi32(vfitab,ifour);
 523             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 524             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 525             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 526             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 527             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 528             Heps             = _mm256_mul_pd(vfeps,H);
 529             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 530             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 531             vvdw6            = _mm256_mul_pd(c6_00,VV);
 532             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 533             fvdw6            = _mm256_mul_pd(c6_00,FF);
 534
 535             /* CUBIC SPLINE TABLE REPULSION */
 536             vfitab           = _mm_add_epi32(vfitab,ifour);
 537             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 538             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 539             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 540             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 541             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 542             Heps             = _mm256_mul_pd(vfeps,H);
 543             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 544             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 545             vvdw12           = _mm256_mul_pd(c12_00,VV);
 546             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 547             fvdw12           = _mm256_mul_pd(c12_00,FF);
 548             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 549             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm256_andnot_pd(dummy_mask,velec);
 553             velecsum         = _mm256_add_pd(velecsum,velec);
 554             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 555             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 556
 557             fscal            = _mm256_add_pd(felec,fvdw);
 558
 559             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm256_mul_pd(fscal,dx00);
 563             ty               = _mm256_mul_pd(fscal,dy00);
 564             tz               = _mm256_mul_pd(fscal,dz00);
 565
 566             /* Update vectorial force */
 567             fix0             = _mm256_add_pd(fix0,tx);
 568             fiy0             = _mm256_add_pd(fiy0,ty);
 569             fiz0             = _mm256_add_pd(fiz0,tz);
 570
 571             fjx0             = _mm256_add_pd(fjx0,tx);
 572             fjy0             = _mm256_add_pd(fjy0,ty);
 573             fjz0             = _mm256_add_pd(fjz0,tz);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r10              = _mm256_mul_pd(rsq10,rinv10);
 580             r10              = _mm256_andnot_pd(dummy_mask,r10);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq10             = _mm256_mul_pd(iq1,jq0);
 584
 585             /* Calculate table index by multiplying r with table scale and truncate to integer */
 586             rt               = _mm256_mul_pd(r10,vftabscale);
 587             vfitab           = _mm256_cvttpd_epi32(rt);
 588             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 589             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 590
 591             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 592             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 593             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 594             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 595             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 596             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 597             Heps             = _mm256_mul_pd(vfeps,H);
 598             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 599             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 600             velec            = _mm256_mul_pd(qq10,VV);
 601             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 602             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 603
 604             /* Update potential sum for this i atom from the interaction with this j atom. */
 605             velec            = _mm256_andnot_pd(dummy_mask,velec);
 606             velecsum         = _mm256_add_pd(velecsum,velec);
 607
 608             fscal            = felec;
 609
 610             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 611
 612             /* Calculate temporary vectorial force */
 613             tx               = _mm256_mul_pd(fscal,dx10);
 614             ty               = _mm256_mul_pd(fscal,dy10);
 615             tz               = _mm256_mul_pd(fscal,dz10);
 616
 617             /* Update vectorial force */
 618             fix1             = _mm256_add_pd(fix1,tx);
 619             fiy1             = _mm256_add_pd(fiy1,ty);
 620             fiz1             = _mm256_add_pd(fiz1,tz);
 621
 622             fjx0             = _mm256_add_pd(fjx0,tx);
 623             fjy0             = _mm256_add_pd(fjy0,ty);
 624             fjz0             = _mm256_add_pd(fjz0,tz);
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             r20              = _mm256_mul_pd(rsq20,rinv20);
 631             r20              = _mm256_andnot_pd(dummy_mask,r20);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq20             = _mm256_mul_pd(iq2,jq0);
 635
 636             /* Calculate table index by multiplying r with table scale and truncate to integer */
 637             rt               = _mm256_mul_pd(r20,vftabscale);
 638             vfitab           = _mm256_cvttpd_epi32(rt);
 639             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 640             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 641
 642             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 643             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 644             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 645             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 646             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 647             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 648             Heps             = _mm256_mul_pd(vfeps,H);
 649             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 650             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 651             velec            = _mm256_mul_pd(qq20,VV);
 652             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 653             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velec            = _mm256_andnot_pd(dummy_mask,velec);
 657             velecsum         = _mm256_add_pd(velecsum,velec);
 658
 659             fscal            = felec;
 660
 661             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 662
 663             /* Calculate temporary vectorial force */
 664             tx               = _mm256_mul_pd(fscal,dx20);
 665             ty               = _mm256_mul_pd(fscal,dy20);
 666             tz               = _mm256_mul_pd(fscal,dz20);
 667
 668             /* Update vectorial force */
 669             fix2             = _mm256_add_pd(fix2,tx);
 670             fiy2             = _mm256_add_pd(fiy2,ty);
 671             fiz2             = _mm256_add_pd(fiz2,tz);
 672
 673             fjx0             = _mm256_add_pd(fjx0,tx);
 674             fjy0             = _mm256_add_pd(fjy0,ty);
 675             fjz0             = _mm256_add_pd(fjz0,tz);
 676
 677             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 678             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 679             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 680             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 681
 682             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 683
 684             /* Inner loop uses 165 flops */
 685         }
 686
 687         /* End of innermost loop */
 688
 689         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 690                                                  f+i_coord_offset,fshift+i_shift_offset);
 691
 692         ggid                        = gid[iidx];
 693         /* Update potential energies */
 694         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 695         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 696
 697         /* Increment number of inner iterations */
 698         inneriter                  += j_index_end - j_index_start;
 699
 700         /* Outer loop uses 20 flops */
 701     }
 702
 703     /* Increment number of outer iterations */
 704     outeriter        += nri;
 705
 706     /* Update outer/inner flops */
 707
 708     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*165);
 709 }
 710 /*
 711  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_double
 712  * Electrostatics interaction: CubicSplineTable
 713  * VdW interaction:            CubicSplineTable
 714  * Geometry:                   Water3-Particle
 715  * Calculate force/pot:        Force
 716  */
 717 void
 718 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_double
 719                     (t_nblist                    * gmx_restrict       nlist,
 720                      rvec                        * gmx_restrict          xx,
 721                      rvec                        * gmx_restrict          ff,
 722                      struct t_forcerec           * gmx_restrict          fr,
 723                      t_mdatoms                   * gmx_restrict     mdatoms,
 724                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 725                      t_nrnb                      * gmx_restrict        nrnb)
 726 {
 727     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 728      * just 0 for non-waters.
 729      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 730      * jnr indices corresponding to data put in the four positions in the SIMD register.
 731      */
 732     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 733     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 734     int              jnrA,jnrB,jnrC,jnrD;
 735     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 736     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 737     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 738     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 739     real             rcutoff_scalar;
 740     real             *shiftvec,*fshift,*x,*f;
 741     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 742     real             scratch[4*DIM];
 743     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 744     real *           vdwioffsetptr0;
 745     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 746     real *           vdwioffsetptr1;
 747     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 748     real *           vdwioffsetptr2;
 749     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 750     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 751     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 752     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 753     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 754     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 755     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 756     real             *charge;
 757     int              nvdwtype;
 758     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 759     int              *vdwtype;
 760     real             *vdwparam;
 761     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 762     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 763     __m128i          vfitab;
 764     __m128i          ifour       = _mm_set1_epi32(4);
 765     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 766     real             *vftab;
 767     __m256d          dummy_mask,cutoff_mask;
 768     __m128           tmpmask0,tmpmask1;
 769     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 770     __m256d          one     = _mm256_set1_pd(1.0);
 771     __m256d          two     = _mm256_set1_pd(2.0);
 772     x                = xx[0];
 773     f                = ff[0];
 774
 775     nri              = nlist->nri;
 776     iinr             = nlist->iinr;
 777     jindex           = nlist->jindex;
 778     jjnr             = nlist->jjnr;
 779     shiftidx         = nlist->shift;
 780     gid              = nlist->gid;
 781     shiftvec         = fr->shift_vec[0];
 782     fshift           = fr->fshift[0];
 783     facel            = _mm256_set1_pd(fr->ic->epsfac);
 784     charge           = mdatoms->chargeA;
 785     nvdwtype         = fr->ntype;
 786     vdwparam         = fr->nbfp;
 787     vdwtype          = mdatoms->typeA;
 788
 789     vftab            = kernel_data->table_elec_vdw->data;
 790     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 791
 792     /* Setup water-specific parameters */
 793     inr              = nlist->iinr[0];
 794     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 795     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 796     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 797     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 798
 799     /* Avoid stupid compiler warnings */
 800     jnrA = jnrB = jnrC = jnrD = 0;
 801     j_coord_offsetA = 0;
 802     j_coord_offsetB = 0;
 803     j_coord_offsetC = 0;
 804     j_coord_offsetD = 0;
 805
 806     outeriter        = 0;
 807     inneriter        = 0;
 808
 809     for(iidx=0;iidx<4*DIM;iidx++)
 810     {
 811         scratch[iidx] = 0.0;
 812     }
 813
 814     /* Start outer loop over neighborlists */
 815     for(iidx=0; iidx<nri; iidx++)
 816     {
 817         /* Load shift vector for this list */
 818         i_shift_offset   = DIM*shiftidx[iidx];
 819
 820         /* Load limits for loop over neighbors */
 821         j_index_start    = jindex[iidx];
 822         j_index_end      = jindex[iidx+1];
 823
 824         /* Get outer coordinate index */
 825         inr              = iinr[iidx];
 826         i_coord_offset   = DIM*inr;
 827
 828         /* Load i particle coords and add shift vector */
 829         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 830                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 831
 832         fix0             = _mm256_setzero_pd();
 833         fiy0             = _mm256_setzero_pd();
 834         fiz0             = _mm256_setzero_pd();
 835         fix1             = _mm256_setzero_pd();
 836         fiy1             = _mm256_setzero_pd();
 837         fiz1             = _mm256_setzero_pd();
 838         fix2             = _mm256_setzero_pd();
 839         fiy2             = _mm256_setzero_pd();
 840         fiz2             = _mm256_setzero_pd();
 841
 842         /* Start inner kernel loop */
 843         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 844         {
 845
 846             /* Get j neighbor index, and coordinate index */
 847             jnrA             = jjnr[jidx];
 848             jnrB             = jjnr[jidx+1];
 849             jnrC             = jjnr[jidx+2];
 850             jnrD             = jjnr[jidx+3];
 851             j_coord_offsetA  = DIM*jnrA;
 852             j_coord_offsetB  = DIM*jnrB;
 853             j_coord_offsetC  = DIM*jnrC;
 854             j_coord_offsetD  = DIM*jnrD;
 855
 856             /* load j atom coordinates */
 857             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 858                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 859                                                  &jx0,&jy0,&jz0);
 860
 861             /* Calculate displacement vector */
 862             dx00             = _mm256_sub_pd(ix0,jx0);
 863             dy00             = _mm256_sub_pd(iy0,jy0);
 864             dz00             = _mm256_sub_pd(iz0,jz0);
 865             dx10             = _mm256_sub_pd(ix1,jx0);
 866             dy10             = _mm256_sub_pd(iy1,jy0);
 867             dz10             = _mm256_sub_pd(iz1,jz0);
 868             dx20             = _mm256_sub_pd(ix2,jx0);
 869             dy20             = _mm256_sub_pd(iy2,jy0);
 870             dz20             = _mm256_sub_pd(iz2,jz0);
 871
 872             /* Calculate squared distance and things based on it */
 873             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 874             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 875             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 876
 877             rinv00           = avx256_invsqrt_d(rsq00);
 878             rinv10           = avx256_invsqrt_d(rsq10);
 879             rinv20           = avx256_invsqrt_d(rsq20);
 880
 881             /* Load parameters for j particles */
 882             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 883                                                                  charge+jnrC+0,charge+jnrD+0);
 884             vdwjidx0A        = 2*vdwtype[jnrA+0];
 885             vdwjidx0B        = 2*vdwtype[jnrB+0];
 886             vdwjidx0C        = 2*vdwtype[jnrC+0];
 887             vdwjidx0D        = 2*vdwtype[jnrD+0];
 888
 889             fjx0             = _mm256_setzero_pd();
 890             fjy0             = _mm256_setzero_pd();
 891             fjz0             = _mm256_setzero_pd();
 892
 893             /**************************
 894              * CALCULATE INTERACTIONS *
 895              **************************/
 896
 897             r00              = _mm256_mul_pd(rsq00,rinv00);
 898
 899             /* Compute parameters for interactions between i and j atoms */
 900             qq00             = _mm256_mul_pd(iq0,jq0);
 901             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 902                                             vdwioffsetptr0+vdwjidx0B,
 903                                             vdwioffsetptr0+vdwjidx0C,
 904                                             vdwioffsetptr0+vdwjidx0D,
 905                                             &c6_00,&c12_00);
 906
 907             /* Calculate table index by multiplying r with table scale and truncate to integer */
 908             rt               = _mm256_mul_pd(r00,vftabscale);
 909             vfitab           = _mm256_cvttpd_epi32(rt);
 910             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 911             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 912
 913             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 914             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 915             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 916             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 917             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 918             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 919             Heps             = _mm256_mul_pd(vfeps,H);
 920             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 921             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 922             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 923
 924             /* CUBIC SPLINE TABLE DISPERSION */
 925             vfitab           = _mm_add_epi32(vfitab,ifour);
 926             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 927             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 928             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 929             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 930             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 931             Heps             = _mm256_mul_pd(vfeps,H);
 932             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 933             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 934             fvdw6            = _mm256_mul_pd(c6_00,FF);
 935
 936             /* CUBIC SPLINE TABLE REPULSION */
 937             vfitab           = _mm_add_epi32(vfitab,ifour);
 938             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 939             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 940             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 941             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 942             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 943             Heps             = _mm256_mul_pd(vfeps,H);
 944             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 945             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 946             fvdw12           = _mm256_mul_pd(c12_00,FF);
 947             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 948
 949             fscal            = _mm256_add_pd(felec,fvdw);
 950
 951             /* Calculate temporary vectorial force */
 952             tx               = _mm256_mul_pd(fscal,dx00);
 953             ty               = _mm256_mul_pd(fscal,dy00);
 954             tz               = _mm256_mul_pd(fscal,dz00);
 955
 956             /* Update vectorial force */
 957             fix0             = _mm256_add_pd(fix0,tx);
 958             fiy0             = _mm256_add_pd(fiy0,ty);
 959             fiz0             = _mm256_add_pd(fiz0,tz);
 960
 961             fjx0             = _mm256_add_pd(fjx0,tx);
 962             fjy0             = _mm256_add_pd(fjy0,ty);
 963             fjz0             = _mm256_add_pd(fjz0,tz);
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             r10              = _mm256_mul_pd(rsq10,rinv10);
 970
 971             /* Compute parameters for interactions between i and j atoms */
 972             qq10             = _mm256_mul_pd(iq1,jq0);
 973
 974             /* Calculate table index by multiplying r with table scale and truncate to integer */
 975             rt               = _mm256_mul_pd(r10,vftabscale);
 976             vfitab           = _mm256_cvttpd_epi32(rt);
 977             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 978             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 979
 980             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 981             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 982             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 983             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 984             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 985             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 986             Heps             = _mm256_mul_pd(vfeps,H);
 987             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 988             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 989             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 990
 991             fscal            = felec;
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm256_mul_pd(fscal,dx10);
 995             ty               = _mm256_mul_pd(fscal,dy10);
 996             tz               = _mm256_mul_pd(fscal,dz10);
 997
 998             /* Update vectorial force */
 999             fix1             = _mm256_add_pd(fix1,tx);
1000             fiy1             = _mm256_add_pd(fiy1,ty);
1001             fiz1             = _mm256_add_pd(fiz1,tz);
1002
1003             fjx0             = _mm256_add_pd(fjx0,tx);
1004             fjy0             = _mm256_add_pd(fjy0,ty);
1005             fjz0             = _mm256_add_pd(fjz0,tz);
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             r20              = _mm256_mul_pd(rsq20,rinv20);
1012
1013             /* Compute parameters for interactions between i and j atoms */
1014             qq20             = _mm256_mul_pd(iq2,jq0);
1015
1016             /* Calculate table index by multiplying r with table scale and truncate to integer */
1017             rt               = _mm256_mul_pd(r20,vftabscale);
1018             vfitab           = _mm256_cvttpd_epi32(rt);
1019             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1020             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1021
1022             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1023             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1024             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1025             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1026             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1027             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1028             Heps             = _mm256_mul_pd(vfeps,H);
1029             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1030             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1031             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1032
1033             fscal            = felec;
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm256_mul_pd(fscal,dx20);
1037             ty               = _mm256_mul_pd(fscal,dy20);
1038             tz               = _mm256_mul_pd(fscal,dz20);
1039
1040             /* Update vectorial force */
1041             fix2             = _mm256_add_pd(fix2,tx);
1042             fiy2             = _mm256_add_pd(fiy2,ty);
1043             fiz2             = _mm256_add_pd(fiz2,tz);
1044
1045             fjx0             = _mm256_add_pd(fjx0,tx);
1046             fjy0             = _mm256_add_pd(fjy0,ty);
1047             fjz0             = _mm256_add_pd(fjz0,tz);
1048
1049             fjptrA             = f+j_coord_offsetA;
1050             fjptrB             = f+j_coord_offsetB;
1051             fjptrC             = f+j_coord_offsetC;
1052             fjptrD             = f+j_coord_offsetD;
1053
1054             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1055
1056             /* Inner loop uses 142 flops */
1057         }
1058
1059         if(jidx<j_index_end)
1060         {
1061
1062             /* Get j neighbor index, and coordinate index */
1063             jnrlistA         = jjnr[jidx];
1064             jnrlistB         = jjnr[jidx+1];
1065             jnrlistC         = jjnr[jidx+2];
1066             jnrlistD         = jjnr[jidx+3];
1067             /* Sign of each element will be negative for non-real atoms.
1068              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1069              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1070              */
1071             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1072
1073             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1074             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1075             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1076
1077             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1078             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1079             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1080             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1081             j_coord_offsetA  = DIM*jnrA;
1082             j_coord_offsetB  = DIM*jnrB;
1083             j_coord_offsetC  = DIM*jnrC;
1084             j_coord_offsetD  = DIM*jnrD;
1085
1086             /* load j atom coordinates */
1087             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1088                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1089                                                  &jx0,&jy0,&jz0);
1090
1091             /* Calculate displacement vector */
1092             dx00             = _mm256_sub_pd(ix0,jx0);
1093             dy00             = _mm256_sub_pd(iy0,jy0);
1094             dz00             = _mm256_sub_pd(iz0,jz0);
1095             dx10             = _mm256_sub_pd(ix1,jx0);
1096             dy10             = _mm256_sub_pd(iy1,jy0);
1097             dz10             = _mm256_sub_pd(iz1,jz0);
1098             dx20             = _mm256_sub_pd(ix2,jx0);
1099             dy20             = _mm256_sub_pd(iy2,jy0);
1100             dz20             = _mm256_sub_pd(iz2,jz0);
1101
1102             /* Calculate squared distance and things based on it */
1103             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1104             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1105             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1106
1107             rinv00           = avx256_invsqrt_d(rsq00);
1108             rinv10           = avx256_invsqrt_d(rsq10);
1109             rinv20           = avx256_invsqrt_d(rsq20);
1110
1111             /* Load parameters for j particles */
1112             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1113                                                                  charge+jnrC+0,charge+jnrD+0);
1114             vdwjidx0A        = 2*vdwtype[jnrA+0];
1115             vdwjidx0B        = 2*vdwtype[jnrB+0];
1116             vdwjidx0C        = 2*vdwtype[jnrC+0];
1117             vdwjidx0D        = 2*vdwtype[jnrD+0];
1118
1119             fjx0             = _mm256_setzero_pd();
1120             fjy0             = _mm256_setzero_pd();
1121             fjz0             = _mm256_setzero_pd();
1122
1123             /**************************
1124              * CALCULATE INTERACTIONS *
1125              **************************/
1126
1127             r00              = _mm256_mul_pd(rsq00,rinv00);
1128             r00              = _mm256_andnot_pd(dummy_mask,r00);
1129
1130             /* Compute parameters for interactions between i and j atoms */
1131             qq00             = _mm256_mul_pd(iq0,jq0);
1132             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1133                                             vdwioffsetptr0+vdwjidx0B,
1134                                             vdwioffsetptr0+vdwjidx0C,
1135                                             vdwioffsetptr0+vdwjidx0D,
1136                                             &c6_00,&c12_00);
1137
1138             /* Calculate table index by multiplying r with table scale and truncate to integer */
1139             rt               = _mm256_mul_pd(r00,vftabscale);
1140             vfitab           = _mm256_cvttpd_epi32(rt);
1141             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1142             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1143
1144             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1145             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1146             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1147             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1148             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1149             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1150             Heps             = _mm256_mul_pd(vfeps,H);
1151             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1152             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1153             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1154
1155             /* CUBIC SPLINE TABLE DISPERSION */
1156             vfitab           = _mm_add_epi32(vfitab,ifour);
1157             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1158             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1159             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1160             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1161             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1162             Heps             = _mm256_mul_pd(vfeps,H);
1163             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1164             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1165             fvdw6            = _mm256_mul_pd(c6_00,FF);
1166
1167             /* CUBIC SPLINE TABLE REPULSION */
1168             vfitab           = _mm_add_epi32(vfitab,ifour);
1169             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1170             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1171             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1172             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1173             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1174             Heps             = _mm256_mul_pd(vfeps,H);
1175             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1176             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1177             fvdw12           = _mm256_mul_pd(c12_00,FF);
1178             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1179
1180             fscal            = _mm256_add_pd(felec,fvdw);
1181
1182             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1183
1184             /* Calculate temporary vectorial force */
1185             tx               = _mm256_mul_pd(fscal,dx00);
1186             ty               = _mm256_mul_pd(fscal,dy00);
1187             tz               = _mm256_mul_pd(fscal,dz00);
1188
1189             /* Update vectorial force */
1190             fix0             = _mm256_add_pd(fix0,tx);
1191             fiy0             = _mm256_add_pd(fiy0,ty);
1192             fiz0             = _mm256_add_pd(fiz0,tz);
1193
1194             fjx0             = _mm256_add_pd(fjx0,tx);
1195             fjy0             = _mm256_add_pd(fjy0,ty);
1196             fjz0             = _mm256_add_pd(fjz0,tz);
1197
1198             /**************************
1199              * CALCULATE INTERACTIONS *
1200              **************************/
1201
1202             r10              = _mm256_mul_pd(rsq10,rinv10);
1203             r10              = _mm256_andnot_pd(dummy_mask,r10);
1204
1205             /* Compute parameters for interactions between i and j atoms */
1206             qq10             = _mm256_mul_pd(iq1,jq0);
1207
1208             /* Calculate table index by multiplying r with table scale and truncate to integer */
1209             rt               = _mm256_mul_pd(r10,vftabscale);
1210             vfitab           = _mm256_cvttpd_epi32(rt);
1211             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1212             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1213
1214             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1215             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1216             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1217             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1218             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1219             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1220             Heps             = _mm256_mul_pd(vfeps,H);
1221             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1222             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1223             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1224
1225             fscal            = felec;
1226
1227             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1228
1229             /* Calculate temporary vectorial force */
1230             tx               = _mm256_mul_pd(fscal,dx10);
1231             ty               = _mm256_mul_pd(fscal,dy10);
1232             tz               = _mm256_mul_pd(fscal,dz10);
1233
1234             /* Update vectorial force */
1235             fix1             = _mm256_add_pd(fix1,tx);
1236             fiy1             = _mm256_add_pd(fiy1,ty);
1237             fiz1             = _mm256_add_pd(fiz1,tz);
1238
1239             fjx0             = _mm256_add_pd(fjx0,tx);
1240             fjy0             = _mm256_add_pd(fjy0,ty);
1241             fjz0             = _mm256_add_pd(fjz0,tz);
1242
1243             /**************************
1244              * CALCULATE INTERACTIONS *
1245              **************************/
1246
1247             r20              = _mm256_mul_pd(rsq20,rinv20);
1248             r20              = _mm256_andnot_pd(dummy_mask,r20);
1249
1250             /* Compute parameters for interactions between i and j atoms */
1251             qq20             = _mm256_mul_pd(iq2,jq0);
1252
1253             /* Calculate table index by multiplying r with table scale and truncate to integer */
1254             rt               = _mm256_mul_pd(r20,vftabscale);
1255             vfitab           = _mm256_cvttpd_epi32(rt);
1256             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1257             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1258
1259             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1260             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1261             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1262             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1263             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1264             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1265             Heps             = _mm256_mul_pd(vfeps,H);
1266             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1267             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1268             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1269
1270             fscal            = felec;
1271
1272             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1273
1274             /* Calculate temporary vectorial force */
1275             tx               = _mm256_mul_pd(fscal,dx20);
1276             ty               = _mm256_mul_pd(fscal,dy20);
1277             tz               = _mm256_mul_pd(fscal,dz20);
1278
1279             /* Update vectorial force */
1280             fix2             = _mm256_add_pd(fix2,tx);
1281             fiy2             = _mm256_add_pd(fiy2,ty);
1282             fiz2             = _mm256_add_pd(fiz2,tz);
1283
1284             fjx0             = _mm256_add_pd(fjx0,tx);
1285             fjy0             = _mm256_add_pd(fjy0,ty);
1286             fjz0             = _mm256_add_pd(fjz0,tz);
1287
1288             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1289             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1290             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1291             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1292
1293             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1294
1295             /* Inner loop uses 145 flops */
1296         }
1297
1298         /* End of innermost loop */
1299
1300         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1301                                                  f+i_coord_offset,fshift+i_shift_offset);
1302
1303         /* Increment number of inner iterations */
1304         inneriter                  += j_index_end - j_index_start;
1305
1306         /* Outer loop uses 18 flops */
1307     }
1308
1309     /* Increment number of outer iterations */
1310     outeriter        += nri;
1311
1312     /* Update outer/inner flops */
1313
1314     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*145);
1315 }