src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_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_ElecEw_VdwCSTab_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_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     __m128i          ewitab;
 110     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 111     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 112     real             *ewtab;
 113     __m256d          dummy_mask,cutoff_mask;
 114     __m128           tmpmask0,tmpmask1;
 115     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 116     __m256d          one     = _mm256_set1_pd(1.0);
 117     __m256d          two     = _mm256_set1_pd(2.0);
 118     x                = xx[0];
 119     f                = ff[0];
 120
 121     nri              = nlist->nri;
 122     iinr             = nlist->iinr;
 123     jindex           = nlist->jindex;
 124     jjnr             = nlist->jjnr;
 125     shiftidx         = nlist->shift;
 126     gid              = nlist->gid;
 127     shiftvec         = fr->shift_vec[0];
 128     fshift           = fr->fshift[0];
 129     facel            = _mm256_set1_pd(fr->epsfac);
 130     charge           = mdatoms->chargeA;
 131     nvdwtype         = fr->ntype;
 132     vdwparam         = fr->nbfp;
 133     vdwtype          = mdatoms->typeA;
 134
 135     vftab            = kernel_data->table_vdw->data;
 136     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 137
 138     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 139     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 140     beta2            = _mm256_mul_pd(beta,beta);
 141     beta3            = _mm256_mul_pd(beta,beta2);
 142
 143     ewtab            = fr->ic->tabq_coul_FDV0;
 144     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 145     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 146
 147     /* Setup water-specific parameters */
 148     inr              = nlist->iinr[0];
 149     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 150     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 151     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 152     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = jnrC = jnrD = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158     j_coord_offsetC = 0;
 159     j_coord_offsetD = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     for(iidx=0;iidx<4*DIM;iidx++)
 165     {
 166         scratch[iidx] = 0.0;
 167     }
 168
 169     /* Start outer loop over neighborlists */
 170     for(iidx=0; iidx<nri; iidx++)
 171     {
 172         /* Load shift vector for this list */
 173         i_shift_offset   = DIM*shiftidx[iidx];
 174
 175         /* Load limits for loop over neighbors */
 176         j_index_start    = jindex[iidx];
 177         j_index_end      = jindex[iidx+1];
 178
 179         /* Get outer coordinate index */
 180         inr              = iinr[iidx];
 181         i_coord_offset   = DIM*inr;
 182
 183         /* Load i particle coords and add shift vector */
 184         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 185                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 186
 187         fix0             = _mm256_setzero_pd();
 188         fiy0             = _mm256_setzero_pd();
 189         fiz0             = _mm256_setzero_pd();
 190         fix1             = _mm256_setzero_pd();
 191         fiy1             = _mm256_setzero_pd();
 192         fiz1             = _mm256_setzero_pd();
 193         fix2             = _mm256_setzero_pd();
 194         fiy2             = _mm256_setzero_pd();
 195         fiz2             = _mm256_setzero_pd();
 196
 197         /* Reset potential sums */
 198         velecsum         = _mm256_setzero_pd();
 199         vvdwsum          = _mm256_setzero_pd();
 200
 201         /* Start inner kernel loop */
 202         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 203         {
 204
 205             /* Get j neighbor index, and coordinate index */
 206             jnrA             = jjnr[jidx];
 207             jnrB             = jjnr[jidx+1];
 208             jnrC             = jjnr[jidx+2];
 209             jnrD             = jjnr[jidx+3];
 210             j_coord_offsetA  = DIM*jnrA;
 211             j_coord_offsetB  = DIM*jnrB;
 212             j_coord_offsetC  = DIM*jnrC;
 213             j_coord_offsetD  = DIM*jnrD;
 214
 215             /* load j atom coordinates */
 216             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 217                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 218                                                  &jx0,&jy0,&jz0);
 219
 220             /* Calculate displacement vector */
 221             dx00             = _mm256_sub_pd(ix0,jx0);
 222             dy00             = _mm256_sub_pd(iy0,jy0);
 223             dz00             = _mm256_sub_pd(iz0,jz0);
 224             dx10             = _mm256_sub_pd(ix1,jx0);
 225             dy10             = _mm256_sub_pd(iy1,jy0);
 226             dz10             = _mm256_sub_pd(iz1,jz0);
 227             dx20             = _mm256_sub_pd(ix2,jx0);
 228             dy20             = _mm256_sub_pd(iy2,jy0);
 229             dz20             = _mm256_sub_pd(iz2,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 233             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 234             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 235
 236             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 237             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 238             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 239
 240             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 241             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 242             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 243
 244             /* Load parameters for j particles */
 245             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 246                                                                  charge+jnrC+0,charge+jnrD+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249             vdwjidx0C        = 2*vdwtype[jnrC+0];
 250             vdwjidx0D        = 2*vdwtype[jnrD+0];
 251
 252             fjx0             = _mm256_setzero_pd();
 253             fjy0             = _mm256_setzero_pd();
 254             fjz0             = _mm256_setzero_pd();
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             r00              = _mm256_mul_pd(rsq00,rinv00);
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             qq00             = _mm256_mul_pd(iq0,jq0);
 264             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 265                                             vdwioffsetptr0+vdwjidx0B,
 266                                             vdwioffsetptr0+vdwjidx0C,
 267                                             vdwioffsetptr0+vdwjidx0D,
 268                                             &c6_00,&c12_00);
 269
 270             /* Calculate table index by multiplying r with table scale and truncate to integer */
 271             rt               = _mm256_mul_pd(r00,vftabscale);
 272             vfitab           = _mm256_cvttpd_epi32(rt);
 273             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 274             vfitab           = _mm_slli_epi32(vfitab,3);
 275
 276             /* EWALD ELECTROSTATICS */
 277
 278             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 279             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 280             ewitab           = _mm256_cvttpd_epi32(ewrt);
 281             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 282             ewitab           = _mm_slli_epi32(ewitab,2);
 283             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 284             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 285             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 286             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 287             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 288             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 289             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 290             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 291             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 292
 293             /* CUBIC SPLINE TABLE DISPERSION */
 294             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 295             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 296             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 297             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 298             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 299             Heps             = _mm256_mul_pd(vfeps,H);
 300             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 301             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 302             vvdw6            = _mm256_mul_pd(c6_00,VV);
 303             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 304             fvdw6            = _mm256_mul_pd(c6_00,FF);
 305
 306             /* CUBIC SPLINE TABLE REPULSION */
 307             vfitab           = _mm_add_epi32(vfitab,ifour);
 308             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 309             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 310             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 311             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 312             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 313             Heps             = _mm256_mul_pd(vfeps,H);
 314             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 315             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 316             vvdw12           = _mm256_mul_pd(c12_00,VV);
 317             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 318             fvdw12           = _mm256_mul_pd(c12_00,FF);
 319             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 320             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velecsum         = _mm256_add_pd(velecsum,velec);
 324             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 325
 326             fscal            = _mm256_add_pd(felec,fvdw);
 327
 328             /* Calculate temporary vectorial force */
 329             tx               = _mm256_mul_pd(fscal,dx00);
 330             ty               = _mm256_mul_pd(fscal,dy00);
 331             tz               = _mm256_mul_pd(fscal,dz00);
 332
 333             /* Update vectorial force */
 334             fix0             = _mm256_add_pd(fix0,tx);
 335             fiy0             = _mm256_add_pd(fiy0,ty);
 336             fiz0             = _mm256_add_pd(fiz0,tz);
 337
 338             fjx0             = _mm256_add_pd(fjx0,tx);
 339             fjy0             = _mm256_add_pd(fjy0,ty);
 340             fjz0             = _mm256_add_pd(fjz0,tz);
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             r10              = _mm256_mul_pd(rsq10,rinv10);
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq10             = _mm256_mul_pd(iq1,jq0);
 350
 351             /* EWALD ELECTROSTATICS */
 352
 353             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 354             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 355             ewitab           = _mm256_cvttpd_epi32(ewrt);
 356             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 357             ewitab           = _mm_slli_epi32(ewitab,2);
 358             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 359             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 360             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 361             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 362             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 363             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 364             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 365             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 366             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velecsum         = _mm256_add_pd(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             /* Calculate temporary vectorial force */
 374             tx               = _mm256_mul_pd(fscal,dx10);
 375             ty               = _mm256_mul_pd(fscal,dy10);
 376             tz               = _mm256_mul_pd(fscal,dz10);
 377
 378             /* Update vectorial force */
 379             fix1             = _mm256_add_pd(fix1,tx);
 380             fiy1             = _mm256_add_pd(fiy1,ty);
 381             fiz1             = _mm256_add_pd(fiz1,tz);
 382
 383             fjx0             = _mm256_add_pd(fjx0,tx);
 384             fjy0             = _mm256_add_pd(fjy0,ty);
 385             fjz0             = _mm256_add_pd(fjz0,tz);
 386
 387             /**************************
 388              * CALCULATE INTERACTIONS *
 389              **************************/
 390
 391             r20              = _mm256_mul_pd(rsq20,rinv20);
 392
 393             /* Compute parameters for interactions between i and j atoms */
 394             qq20             = _mm256_mul_pd(iq2,jq0);
 395
 396             /* EWALD ELECTROSTATICS */
 397
 398             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 399             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 400             ewitab           = _mm256_cvttpd_epi32(ewrt);
 401             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 402             ewitab           = _mm_slli_epi32(ewitab,2);
 403             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 404             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 405             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 406             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 407             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 408             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 409             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 410             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 411             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 412
 413             /* Update potential sum for this i atom from the interaction with this j atom. */
 414             velecsum         = _mm256_add_pd(velecsum,velec);
 415
 416             fscal            = felec;
 417
 418             /* Calculate temporary vectorial force */
 419             tx               = _mm256_mul_pd(fscal,dx20);
 420             ty               = _mm256_mul_pd(fscal,dy20);
 421             tz               = _mm256_mul_pd(fscal,dz20);
 422
 423             /* Update vectorial force */
 424             fix2             = _mm256_add_pd(fix2,tx);
 425             fiy2             = _mm256_add_pd(fiy2,ty);
 426             fiz2             = _mm256_add_pd(fiz2,tz);
 427
 428             fjx0             = _mm256_add_pd(fjx0,tx);
 429             fjy0             = _mm256_add_pd(fjy0,ty);
 430             fjz0             = _mm256_add_pd(fjz0,tz);
 431
 432             fjptrA             = f+j_coord_offsetA;
 433             fjptrB             = f+j_coord_offsetB;
 434             fjptrC             = f+j_coord_offsetC;
 435             fjptrD             = f+j_coord_offsetD;
 436
 437             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 438
 439             /* Inner loop uses 160 flops */
 440         }
 441
 442         if(jidx<j_index_end)
 443         {
 444
 445             /* Get j neighbor index, and coordinate index */
 446             jnrlistA         = jjnr[jidx];
 447             jnrlistB         = jjnr[jidx+1];
 448             jnrlistC         = jjnr[jidx+2];
 449             jnrlistD         = jjnr[jidx+3];
 450             /* Sign of each element will be negative for non-real atoms.
 451              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 452              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 453              */
 454             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 455
 456             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 457             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 458             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 459
 460             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 461             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 462             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 463             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 464             j_coord_offsetA  = DIM*jnrA;
 465             j_coord_offsetB  = DIM*jnrB;
 466             j_coord_offsetC  = DIM*jnrC;
 467             j_coord_offsetD  = DIM*jnrD;
 468
 469             /* load j atom coordinates */
 470             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 471                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 472                                                  &jx0,&jy0,&jz0);
 473
 474             /* Calculate displacement vector */
 475             dx00             = _mm256_sub_pd(ix0,jx0);
 476             dy00             = _mm256_sub_pd(iy0,jy0);
 477             dz00             = _mm256_sub_pd(iz0,jz0);
 478             dx10             = _mm256_sub_pd(ix1,jx0);
 479             dy10             = _mm256_sub_pd(iy1,jy0);
 480             dz10             = _mm256_sub_pd(iz1,jz0);
 481             dx20             = _mm256_sub_pd(ix2,jx0);
 482             dy20             = _mm256_sub_pd(iy2,jy0);
 483             dz20             = _mm256_sub_pd(iz2,jz0);
 484
 485             /* Calculate squared distance and things based on it */
 486             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 487             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 488             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 489
 490             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 491             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 492             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 493
 494             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 495             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 496             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 497
 498             /* Load parameters for j particles */
 499             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 500                                                                  charge+jnrC+0,charge+jnrD+0);
 501             vdwjidx0A        = 2*vdwtype[jnrA+0];
 502             vdwjidx0B        = 2*vdwtype[jnrB+0];
 503             vdwjidx0C        = 2*vdwtype[jnrC+0];
 504             vdwjidx0D        = 2*vdwtype[jnrD+0];
 505
 506             fjx0             = _mm256_setzero_pd();
 507             fjy0             = _mm256_setzero_pd();
 508             fjz0             = _mm256_setzero_pd();
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             r00              = _mm256_mul_pd(rsq00,rinv00);
 515             r00              = _mm256_andnot_pd(dummy_mask,r00);
 516
 517             /* Compute parameters for interactions between i and j atoms */
 518             qq00             = _mm256_mul_pd(iq0,jq0);
 519             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 520                                             vdwioffsetptr0+vdwjidx0B,
 521                                             vdwioffsetptr0+vdwjidx0C,
 522                                             vdwioffsetptr0+vdwjidx0D,
 523                                             &c6_00,&c12_00);
 524
 525             /* Calculate table index by multiplying r with table scale and truncate to integer */
 526             rt               = _mm256_mul_pd(r00,vftabscale);
 527             vfitab           = _mm256_cvttpd_epi32(rt);
 528             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 529             vfitab           = _mm_slli_epi32(vfitab,3);
 530
 531             /* EWALD ELECTROSTATICS */
 532
 533             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 534             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 535             ewitab           = _mm256_cvttpd_epi32(ewrt);
 536             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 537             ewitab           = _mm_slli_epi32(ewitab,2);
 538             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 539             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 540             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 541             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 542             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 543             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 544             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 545             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 546             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 547
 548             /* CUBIC SPLINE TABLE DISPERSION */
 549             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 550             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 551             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 552             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 553             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 554             Heps             = _mm256_mul_pd(vfeps,H);
 555             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 556             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 557             vvdw6            = _mm256_mul_pd(c6_00,VV);
 558             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 559             fvdw6            = _mm256_mul_pd(c6_00,FF);
 560
 561             /* CUBIC SPLINE TABLE REPULSION */
 562             vfitab           = _mm_add_epi32(vfitab,ifour);
 563             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 564             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 565             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 566             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 567             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 568             Heps             = _mm256_mul_pd(vfeps,H);
 569             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 570             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 571             vvdw12           = _mm256_mul_pd(c12_00,VV);
 572             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 573             fvdw12           = _mm256_mul_pd(c12_00,FF);
 574             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 575             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_andnot_pd(dummy_mask,velec);
 579             velecsum         = _mm256_add_pd(velecsum,velec);
 580             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 581             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 582
 583             fscal            = _mm256_add_pd(felec,fvdw);
 584
 585             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 586
 587             /* Calculate temporary vectorial force */
 588             tx               = _mm256_mul_pd(fscal,dx00);
 589             ty               = _mm256_mul_pd(fscal,dy00);
 590             tz               = _mm256_mul_pd(fscal,dz00);
 591
 592             /* Update vectorial force */
 593             fix0             = _mm256_add_pd(fix0,tx);
 594             fiy0             = _mm256_add_pd(fiy0,ty);
 595             fiz0             = _mm256_add_pd(fiz0,tz);
 596
 597             fjx0             = _mm256_add_pd(fjx0,tx);
 598             fjy0             = _mm256_add_pd(fjy0,ty);
 599             fjz0             = _mm256_add_pd(fjz0,tz);
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             r10              = _mm256_mul_pd(rsq10,rinv10);
 606             r10              = _mm256_andnot_pd(dummy_mask,r10);
 607
 608             /* Compute parameters for interactions between i and j atoms */
 609             qq10             = _mm256_mul_pd(iq1,jq0);
 610
 611             /* EWALD ELECTROSTATICS */
 612
 613             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 614             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 615             ewitab           = _mm256_cvttpd_epi32(ewrt);
 616             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 617             ewitab           = _mm_slli_epi32(ewitab,2);
 618             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 619             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 620             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 621             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 622             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 623             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 624             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 625             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 626             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 627
 628             /* Update potential sum for this i atom from the interaction with this j atom. */
 629             velec            = _mm256_andnot_pd(dummy_mask,velec);
 630             velecsum         = _mm256_add_pd(velecsum,velec);
 631
 632             fscal            = felec;
 633
 634             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 635
 636             /* Calculate temporary vectorial force */
 637             tx               = _mm256_mul_pd(fscal,dx10);
 638             ty               = _mm256_mul_pd(fscal,dy10);
 639             tz               = _mm256_mul_pd(fscal,dz10);
 640
 641             /* Update vectorial force */
 642             fix1             = _mm256_add_pd(fix1,tx);
 643             fiy1             = _mm256_add_pd(fiy1,ty);
 644             fiz1             = _mm256_add_pd(fiz1,tz);
 645
 646             fjx0             = _mm256_add_pd(fjx0,tx);
 647             fjy0             = _mm256_add_pd(fjy0,ty);
 648             fjz0             = _mm256_add_pd(fjz0,tz);
 649
 650             /**************************
 651              * CALCULATE INTERACTIONS *
 652              **************************/
 653
 654             r20              = _mm256_mul_pd(rsq20,rinv20);
 655             r20              = _mm256_andnot_pd(dummy_mask,r20);
 656
 657             /* Compute parameters for interactions between i and j atoms */
 658             qq20             = _mm256_mul_pd(iq2,jq0);
 659
 660             /* EWALD ELECTROSTATICS */
 661
 662             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 663             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 664             ewitab           = _mm256_cvttpd_epi32(ewrt);
 665             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 666             ewitab           = _mm_slli_epi32(ewitab,2);
 667             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 668             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 669             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 670             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 671             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 672             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 673             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 674             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 675             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 676
 677             /* Update potential sum for this i atom from the interaction with this j atom. */
 678             velec            = _mm256_andnot_pd(dummy_mask,velec);
 679             velecsum         = _mm256_add_pd(velecsum,velec);
 680
 681             fscal            = felec;
 682
 683             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 684
 685             /* Calculate temporary vectorial force */
 686             tx               = _mm256_mul_pd(fscal,dx20);
 687             ty               = _mm256_mul_pd(fscal,dy20);
 688             tz               = _mm256_mul_pd(fscal,dz20);
 689
 690             /* Update vectorial force */
 691             fix2             = _mm256_add_pd(fix2,tx);
 692             fiy2             = _mm256_add_pd(fiy2,ty);
 693             fiz2             = _mm256_add_pd(fiz2,tz);
 694
 695             fjx0             = _mm256_add_pd(fjx0,tx);
 696             fjy0             = _mm256_add_pd(fjy0,ty);
 697             fjz0             = _mm256_add_pd(fjz0,tz);
 698
 699             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 700             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 701             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 702             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 703
 704             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 705
 706             /* Inner loop uses 163 flops */
 707         }
 708
 709         /* End of innermost loop */
 710
 711         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 712                                                  f+i_coord_offset,fshift+i_shift_offset);
 713
 714         ggid                        = gid[iidx];
 715         /* Update potential energies */
 716         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 717         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 718
 719         /* Increment number of inner iterations */
 720         inneriter                  += j_index_end - j_index_start;
 721
 722         /* Outer loop uses 20 flops */
 723     }
 724
 725     /* Increment number of outer iterations */
 726     outeriter        += nri;
 727
 728     /* Update outer/inner flops */
 729
 730     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*163);
 731 }
 732 /*
 733  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_double
 734  * Electrostatics interaction: Ewald
 735  * VdW interaction:            CubicSplineTable
 736  * Geometry:                   Water3-Particle
 737  * Calculate force/pot:        Force
 738  */
 739 void
 740 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_256_double
 741                     (t_nblist                    * gmx_restrict       nlist,
 742                      rvec                        * gmx_restrict          xx,
 743                      rvec                        * gmx_restrict          ff,
 744                      t_forcerec                  * gmx_restrict          fr,
 745                      t_mdatoms                   * gmx_restrict     mdatoms,
 746                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 747                      t_nrnb                      * gmx_restrict        nrnb)
 748 {
 749     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 750      * just 0 for non-waters.
 751      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 752      * jnr indices corresponding to data put in the four positions in the SIMD register.
 753      */
 754     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 755     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 756     int              jnrA,jnrB,jnrC,jnrD;
 757     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 758     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 759     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 760     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 761     real             rcutoff_scalar;
 762     real             *shiftvec,*fshift,*x,*f;
 763     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 764     real             scratch[4*DIM];
 765     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 766     real *           vdwioffsetptr0;
 767     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 768     real *           vdwioffsetptr1;
 769     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 770     real *           vdwioffsetptr2;
 771     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 772     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 773     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 774     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 775     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 776     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 777     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 778     real             *charge;
 779     int              nvdwtype;
 780     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 781     int              *vdwtype;
 782     real             *vdwparam;
 783     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 784     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 785     __m128i          vfitab;
 786     __m128i          ifour       = _mm_set1_epi32(4);
 787     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 788     real             *vftab;
 789     __m128i          ewitab;
 790     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 791     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 792     real             *ewtab;
 793     __m256d          dummy_mask,cutoff_mask;
 794     __m128           tmpmask0,tmpmask1;
 795     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 796     __m256d          one     = _mm256_set1_pd(1.0);
 797     __m256d          two     = _mm256_set1_pd(2.0);
 798     x                = xx[0];
 799     f                = ff[0];
 800
 801     nri              = nlist->nri;
 802     iinr             = nlist->iinr;
 803     jindex           = nlist->jindex;
 804     jjnr             = nlist->jjnr;
 805     shiftidx         = nlist->shift;
 806     gid              = nlist->gid;
 807     shiftvec         = fr->shift_vec[0];
 808     fshift           = fr->fshift[0];
 809     facel            = _mm256_set1_pd(fr->epsfac);
 810     charge           = mdatoms->chargeA;
 811     nvdwtype         = fr->ntype;
 812     vdwparam         = fr->nbfp;
 813     vdwtype          = mdatoms->typeA;
 814
 815     vftab            = kernel_data->table_vdw->data;
 816     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 817
 818     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 819     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 820     beta2            = _mm256_mul_pd(beta,beta);
 821     beta3            = _mm256_mul_pd(beta,beta2);
 822
 823     ewtab            = fr->ic->tabq_coul_F;
 824     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 825     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 826
 827     /* Setup water-specific parameters */
 828     inr              = nlist->iinr[0];
 829     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 830     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 831     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 832     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 833
 834     /* Avoid stupid compiler warnings */
 835     jnrA = jnrB = jnrC = jnrD = 0;
 836     j_coord_offsetA = 0;
 837     j_coord_offsetB = 0;
 838     j_coord_offsetC = 0;
 839     j_coord_offsetD = 0;
 840
 841     outeriter        = 0;
 842     inneriter        = 0;
 843
 844     for(iidx=0;iidx<4*DIM;iidx++)
 845     {
 846         scratch[iidx] = 0.0;
 847     }
 848
 849     /* Start outer loop over neighborlists */
 850     for(iidx=0; iidx<nri; iidx++)
 851     {
 852         /* Load shift vector for this list */
 853         i_shift_offset   = DIM*shiftidx[iidx];
 854
 855         /* Load limits for loop over neighbors */
 856         j_index_start    = jindex[iidx];
 857         j_index_end      = jindex[iidx+1];
 858
 859         /* Get outer coordinate index */
 860         inr              = iinr[iidx];
 861         i_coord_offset   = DIM*inr;
 862
 863         /* Load i particle coords and add shift vector */
 864         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 865                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 866
 867         fix0             = _mm256_setzero_pd();
 868         fiy0             = _mm256_setzero_pd();
 869         fiz0             = _mm256_setzero_pd();
 870         fix1             = _mm256_setzero_pd();
 871         fiy1             = _mm256_setzero_pd();
 872         fiz1             = _mm256_setzero_pd();
 873         fix2             = _mm256_setzero_pd();
 874         fiy2             = _mm256_setzero_pd();
 875         fiz2             = _mm256_setzero_pd();
 876
 877         /* Start inner kernel loop */
 878         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 879         {
 880
 881             /* Get j neighbor index, and coordinate index */
 882             jnrA             = jjnr[jidx];
 883             jnrB             = jjnr[jidx+1];
 884             jnrC             = jjnr[jidx+2];
 885             jnrD             = jjnr[jidx+3];
 886             j_coord_offsetA  = DIM*jnrA;
 887             j_coord_offsetB  = DIM*jnrB;
 888             j_coord_offsetC  = DIM*jnrC;
 889             j_coord_offsetD  = DIM*jnrD;
 890
 891             /* load j atom coordinates */
 892             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 893                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 894                                                  &jx0,&jy0,&jz0);
 895
 896             /* Calculate displacement vector */
 897             dx00             = _mm256_sub_pd(ix0,jx0);
 898             dy00             = _mm256_sub_pd(iy0,jy0);
 899             dz00             = _mm256_sub_pd(iz0,jz0);
 900             dx10             = _mm256_sub_pd(ix1,jx0);
 901             dy10             = _mm256_sub_pd(iy1,jy0);
 902             dz10             = _mm256_sub_pd(iz1,jz0);
 903             dx20             = _mm256_sub_pd(ix2,jx0);
 904             dy20             = _mm256_sub_pd(iy2,jy0);
 905             dz20             = _mm256_sub_pd(iz2,jz0);
 906
 907             /* Calculate squared distance and things based on it */
 908             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 909             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 910             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 911
 912             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 913             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 914             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 915
 916             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 917             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 918             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 919
 920             /* Load parameters for j particles */
 921             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 922                                                                  charge+jnrC+0,charge+jnrD+0);
 923             vdwjidx0A        = 2*vdwtype[jnrA+0];
 924             vdwjidx0B        = 2*vdwtype[jnrB+0];
 925             vdwjidx0C        = 2*vdwtype[jnrC+0];
 926             vdwjidx0D        = 2*vdwtype[jnrD+0];
 927
 928             fjx0             = _mm256_setzero_pd();
 929             fjy0             = _mm256_setzero_pd();
 930             fjz0             = _mm256_setzero_pd();
 931
 932             /**************************
 933              * CALCULATE INTERACTIONS *
 934              **************************/
 935
 936             r00              = _mm256_mul_pd(rsq00,rinv00);
 937
 938             /* Compute parameters for interactions between i and j atoms */
 939             qq00             = _mm256_mul_pd(iq0,jq0);
 940             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 941                                             vdwioffsetptr0+vdwjidx0B,
 942                                             vdwioffsetptr0+vdwjidx0C,
 943                                             vdwioffsetptr0+vdwjidx0D,
 944                                             &c6_00,&c12_00);
 945
 946             /* Calculate table index by multiplying r with table scale and truncate to integer */
 947             rt               = _mm256_mul_pd(r00,vftabscale);
 948             vfitab           = _mm256_cvttpd_epi32(rt);
 949             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 950             vfitab           = _mm_slli_epi32(vfitab,3);
 951
 952             /* EWALD ELECTROSTATICS */
 953
 954             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 955             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 956             ewitab           = _mm256_cvttpd_epi32(ewrt);
 957             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 958             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 959                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 960                                             &ewtabF,&ewtabFn);
 961             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 962             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 963
 964             /* CUBIC SPLINE TABLE DISPERSION */
 965             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 966             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 967             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 968             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 969             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 970             Heps             = _mm256_mul_pd(vfeps,H);
 971             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 972             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 973             fvdw6            = _mm256_mul_pd(c6_00,FF);
 974
 975             /* CUBIC SPLINE TABLE REPULSION */
 976             vfitab           = _mm_add_epi32(vfitab,ifour);
 977             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 978             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 979             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 980             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 981             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 982             Heps             = _mm256_mul_pd(vfeps,H);
 983             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 984             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 985             fvdw12           = _mm256_mul_pd(c12_00,FF);
 986             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 987
 988             fscal            = _mm256_add_pd(felec,fvdw);
 989
 990             /* Calculate temporary vectorial force */
 991             tx               = _mm256_mul_pd(fscal,dx00);
 992             ty               = _mm256_mul_pd(fscal,dy00);
 993             tz               = _mm256_mul_pd(fscal,dz00);
 994
 995             /* Update vectorial force */
 996             fix0             = _mm256_add_pd(fix0,tx);
 997             fiy0             = _mm256_add_pd(fiy0,ty);
 998             fiz0             = _mm256_add_pd(fiz0,tz);
 999
1000             fjx0             = _mm256_add_pd(fjx0,tx);
1001             fjy0             = _mm256_add_pd(fjy0,ty);
1002             fjz0             = _mm256_add_pd(fjz0,tz);
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             r10              = _mm256_mul_pd(rsq10,rinv10);
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq10             = _mm256_mul_pd(iq1,jq0);
1012
1013             /* EWALD ELECTROSTATICS */
1014
1015             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1016             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1017             ewitab           = _mm256_cvttpd_epi32(ewrt);
1018             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1019             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1020                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1021                                             &ewtabF,&ewtabFn);
1022             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1023             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1024
1025             fscal            = felec;
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm256_mul_pd(fscal,dx10);
1029             ty               = _mm256_mul_pd(fscal,dy10);
1030             tz               = _mm256_mul_pd(fscal,dz10);
1031
1032             /* Update vectorial force */
1033             fix1             = _mm256_add_pd(fix1,tx);
1034             fiy1             = _mm256_add_pd(fiy1,ty);
1035             fiz1             = _mm256_add_pd(fiz1,tz);
1036
1037             fjx0             = _mm256_add_pd(fjx0,tx);
1038             fjy0             = _mm256_add_pd(fjy0,ty);
1039             fjz0             = _mm256_add_pd(fjz0,tz);
1040
1041             /**************************
1042              * CALCULATE INTERACTIONS *
1043              **************************/
1044
1045             r20              = _mm256_mul_pd(rsq20,rinv20);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq20             = _mm256_mul_pd(iq2,jq0);
1049
1050             /* EWALD ELECTROSTATICS */
1051
1052             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1053             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1054             ewitab           = _mm256_cvttpd_epi32(ewrt);
1055             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1056             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1057                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1058                                             &ewtabF,&ewtabFn);
1059             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1060             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1061
1062             fscal            = felec;
1063
1064             /* Calculate temporary vectorial force */
1065             tx               = _mm256_mul_pd(fscal,dx20);
1066             ty               = _mm256_mul_pd(fscal,dy20);
1067             tz               = _mm256_mul_pd(fscal,dz20);
1068
1069             /* Update vectorial force */
1070             fix2             = _mm256_add_pd(fix2,tx);
1071             fiy2             = _mm256_add_pd(fiy2,ty);
1072             fiz2             = _mm256_add_pd(fiz2,tz);
1073
1074             fjx0             = _mm256_add_pd(fjx0,tx);
1075             fjy0             = _mm256_add_pd(fjy0,ty);
1076             fjz0             = _mm256_add_pd(fjz0,tz);
1077
1078             fjptrA             = f+j_coord_offsetA;
1079             fjptrB             = f+j_coord_offsetB;
1080             fjptrC             = f+j_coord_offsetC;
1081             fjptrD             = f+j_coord_offsetD;
1082
1083             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1084
1085             /* Inner loop uses 137 flops */
1086         }
1087
1088         if(jidx<j_index_end)
1089         {
1090
1091             /* Get j neighbor index, and coordinate index */
1092             jnrlistA         = jjnr[jidx];
1093             jnrlistB         = jjnr[jidx+1];
1094             jnrlistC         = jjnr[jidx+2];
1095             jnrlistD         = jjnr[jidx+3];
1096             /* Sign of each element will be negative for non-real atoms.
1097              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1098              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1099              */
1100             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1101
1102             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1103             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1104             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1105
1106             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1107             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1108             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1109             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1110             j_coord_offsetA  = DIM*jnrA;
1111             j_coord_offsetB  = DIM*jnrB;
1112             j_coord_offsetC  = DIM*jnrC;
1113             j_coord_offsetD  = DIM*jnrD;
1114
1115             /* load j atom coordinates */
1116             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1117                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1118                                                  &jx0,&jy0,&jz0);
1119
1120             /* Calculate displacement vector */
1121             dx00             = _mm256_sub_pd(ix0,jx0);
1122             dy00             = _mm256_sub_pd(iy0,jy0);
1123             dz00             = _mm256_sub_pd(iz0,jz0);
1124             dx10             = _mm256_sub_pd(ix1,jx0);
1125             dy10             = _mm256_sub_pd(iy1,jy0);
1126             dz10             = _mm256_sub_pd(iz1,jz0);
1127             dx20             = _mm256_sub_pd(ix2,jx0);
1128             dy20             = _mm256_sub_pd(iy2,jy0);
1129             dz20             = _mm256_sub_pd(iz2,jz0);
1130
1131             /* Calculate squared distance and things based on it */
1132             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1133             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1134             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1135
1136             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1137             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1138             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1139
1140             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1141             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1142             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1143
1144             /* Load parameters for j particles */
1145             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1146                                                                  charge+jnrC+0,charge+jnrD+0);
1147             vdwjidx0A        = 2*vdwtype[jnrA+0];
1148             vdwjidx0B        = 2*vdwtype[jnrB+0];
1149             vdwjidx0C        = 2*vdwtype[jnrC+0];
1150             vdwjidx0D        = 2*vdwtype[jnrD+0];
1151
1152             fjx0             = _mm256_setzero_pd();
1153             fjy0             = _mm256_setzero_pd();
1154             fjz0             = _mm256_setzero_pd();
1155
1156             /**************************
1157              * CALCULATE INTERACTIONS *
1158              **************************/
1159
1160             r00              = _mm256_mul_pd(rsq00,rinv00);
1161             r00              = _mm256_andnot_pd(dummy_mask,r00);
1162
1163             /* Compute parameters for interactions between i and j atoms */
1164             qq00             = _mm256_mul_pd(iq0,jq0);
1165             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1166                                             vdwioffsetptr0+vdwjidx0B,
1167                                             vdwioffsetptr0+vdwjidx0C,
1168                                             vdwioffsetptr0+vdwjidx0D,
1169                                             &c6_00,&c12_00);
1170
1171             /* Calculate table index by multiplying r with table scale and truncate to integer */
1172             rt               = _mm256_mul_pd(r00,vftabscale);
1173             vfitab           = _mm256_cvttpd_epi32(rt);
1174             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1175             vfitab           = _mm_slli_epi32(vfitab,3);
1176
1177             /* EWALD ELECTROSTATICS */
1178
1179             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1180             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1181             ewitab           = _mm256_cvttpd_epi32(ewrt);
1182             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1183             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1184                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1185                                             &ewtabF,&ewtabFn);
1186             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1187             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1188
1189             /* CUBIC SPLINE TABLE DISPERSION */
1190             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1191             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1192             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1193             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1194             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1195             Heps             = _mm256_mul_pd(vfeps,H);
1196             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1197             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1198             fvdw6            = _mm256_mul_pd(c6_00,FF);
1199
1200             /* CUBIC SPLINE TABLE REPULSION */
1201             vfitab           = _mm_add_epi32(vfitab,ifour);
1202             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1203             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1204             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1205             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1206             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1207             Heps             = _mm256_mul_pd(vfeps,H);
1208             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1209             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1210             fvdw12           = _mm256_mul_pd(c12_00,FF);
1211             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1212
1213             fscal            = _mm256_add_pd(felec,fvdw);
1214
1215             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1216
1217             /* Calculate temporary vectorial force */
1218             tx               = _mm256_mul_pd(fscal,dx00);
1219             ty               = _mm256_mul_pd(fscal,dy00);
1220             tz               = _mm256_mul_pd(fscal,dz00);
1221
1222             /* Update vectorial force */
1223             fix0             = _mm256_add_pd(fix0,tx);
1224             fiy0             = _mm256_add_pd(fiy0,ty);
1225             fiz0             = _mm256_add_pd(fiz0,tz);
1226
1227             fjx0             = _mm256_add_pd(fjx0,tx);
1228             fjy0             = _mm256_add_pd(fjy0,ty);
1229             fjz0             = _mm256_add_pd(fjz0,tz);
1230
1231             /**************************
1232              * CALCULATE INTERACTIONS *
1233              **************************/
1234
1235             r10              = _mm256_mul_pd(rsq10,rinv10);
1236             r10              = _mm256_andnot_pd(dummy_mask,r10);
1237
1238             /* Compute parameters for interactions between i and j atoms */
1239             qq10             = _mm256_mul_pd(iq1,jq0);
1240
1241             /* EWALD ELECTROSTATICS */
1242
1243             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1244             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1245             ewitab           = _mm256_cvttpd_epi32(ewrt);
1246             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1247             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1248                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1249                                             &ewtabF,&ewtabFn);
1250             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1251             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1252
1253             fscal            = felec;
1254
1255             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1256
1257             /* Calculate temporary vectorial force */
1258             tx               = _mm256_mul_pd(fscal,dx10);
1259             ty               = _mm256_mul_pd(fscal,dy10);
1260             tz               = _mm256_mul_pd(fscal,dz10);
1261
1262             /* Update vectorial force */
1263             fix1             = _mm256_add_pd(fix1,tx);
1264             fiy1             = _mm256_add_pd(fiy1,ty);
1265             fiz1             = _mm256_add_pd(fiz1,tz);
1266
1267             fjx0             = _mm256_add_pd(fjx0,tx);
1268             fjy0             = _mm256_add_pd(fjy0,ty);
1269             fjz0             = _mm256_add_pd(fjz0,tz);
1270
1271             /**************************
1272              * CALCULATE INTERACTIONS *
1273              **************************/
1274
1275             r20              = _mm256_mul_pd(rsq20,rinv20);
1276             r20              = _mm256_andnot_pd(dummy_mask,r20);
1277
1278             /* Compute parameters for interactions between i and j atoms */
1279             qq20             = _mm256_mul_pd(iq2,jq0);
1280
1281             /* EWALD ELECTROSTATICS */
1282
1283             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1284             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1285             ewitab           = _mm256_cvttpd_epi32(ewrt);
1286             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1287             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1288                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1289                                             &ewtabF,&ewtabFn);
1290             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1291             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1292
1293             fscal            = felec;
1294
1295             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1296
1297             /* Calculate temporary vectorial force */
1298             tx               = _mm256_mul_pd(fscal,dx20);
1299             ty               = _mm256_mul_pd(fscal,dy20);
1300             tz               = _mm256_mul_pd(fscal,dz20);
1301
1302             /* Update vectorial force */
1303             fix2             = _mm256_add_pd(fix2,tx);
1304             fiy2             = _mm256_add_pd(fiy2,ty);
1305             fiz2             = _mm256_add_pd(fiz2,tz);
1306
1307             fjx0             = _mm256_add_pd(fjx0,tx);
1308             fjy0             = _mm256_add_pd(fjy0,ty);
1309             fjz0             = _mm256_add_pd(fjz0,tz);
1310
1311             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1312             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1313             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1314             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1315
1316             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1317
1318             /* Inner loop uses 140 flops */
1319         }
1320
1321         /* End of innermost loop */
1322
1323         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1324                                                  f+i_coord_offset,fshift+i_shift_offset);
1325
1326         /* Increment number of inner iterations */
1327         inneriter                  += j_index_end - j_index_start;
1328
1329         /* Outer loop uses 18 flops */
1330     }
1331
1332     /* Increment number of outer iterations */
1333     outeriter        += nri;
1334
1335     /* Update outer/inner flops */
1336
1337     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*140);
1338 }