src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_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_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 107     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 108     __m128i          vfitab;
 109     __m128i          ifour       = _mm_set1_epi32(4);
 110     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 111     real             *vftab;
 112     __m256d          dummy_mask,cutoff_mask;
 113     __m128           tmpmask0,tmpmask1;
 114     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 115     __m256d          one     = _mm256_set1_pd(1.0);
 116     __m256d          two     = _mm256_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     vftab            = kernel_data->table_elec_vdw->data;
 135     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 136
 137     /* Setup water-specific parameters */
 138     inr              = nlist->iinr[0];
 139     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 140     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 141     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 142     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = jnrC = jnrD = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148     j_coord_offsetC = 0;
 149     j_coord_offsetD = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     for(iidx=0;iidx<4*DIM;iidx++)
 155     {
 156         scratch[iidx] = 0.0;
 157     }
 158
 159     /* Start outer loop over neighborlists */
 160     for(iidx=0; iidx<nri; iidx++)
 161     {
 162         /* Load shift vector for this list */
 163         i_shift_offset   = DIM*shiftidx[iidx];
 164
 165         /* Load limits for loop over neighbors */
 166         j_index_start    = jindex[iidx];
 167         j_index_end      = jindex[iidx+1];
 168
 169         /* Get outer coordinate index */
 170         inr              = iinr[iidx];
 171         i_coord_offset   = DIM*inr;
 172
 173         /* Load i particle coords and add shift vector */
 174         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 175                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 176
 177         fix0             = _mm256_setzero_pd();
 178         fiy0             = _mm256_setzero_pd();
 179         fiz0             = _mm256_setzero_pd();
 180         fix1             = _mm256_setzero_pd();
 181         fiy1             = _mm256_setzero_pd();
 182         fiz1             = _mm256_setzero_pd();
 183         fix2             = _mm256_setzero_pd();
 184         fiy2             = _mm256_setzero_pd();
 185         fiz2             = _mm256_setzero_pd();
 186         fix3             = _mm256_setzero_pd();
 187         fiy3             = _mm256_setzero_pd();
 188         fiz3             = _mm256_setzero_pd();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm256_setzero_pd();
 192         vvdwsum          = _mm256_setzero_pd();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 211                                                  &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm256_sub_pd(ix0,jx0);
 215             dy00             = _mm256_sub_pd(iy0,jy0);
 216             dz00             = _mm256_sub_pd(iz0,jz0);
 217             dx10             = _mm256_sub_pd(ix1,jx0);
 218             dy10             = _mm256_sub_pd(iy1,jy0);
 219             dz10             = _mm256_sub_pd(iz1,jz0);
 220             dx20             = _mm256_sub_pd(ix2,jx0);
 221             dy20             = _mm256_sub_pd(iy2,jy0);
 222             dz20             = _mm256_sub_pd(iz2,jz0);
 223             dx30             = _mm256_sub_pd(ix3,jx0);
 224             dy30             = _mm256_sub_pd(iy3,jy0);
 225             dz30             = _mm256_sub_pd(iz3,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 229             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 230             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 231             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 232
 233             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 234             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 235             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 236             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 237
 238             /* Load parameters for j particles */
 239             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 240                                                                  charge+jnrC+0,charge+jnrD+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243             vdwjidx0C        = 2*vdwtype[jnrC+0];
 244             vdwjidx0D        = 2*vdwtype[jnrD+0];
 245
 246             fjx0             = _mm256_setzero_pd();
 247             fjy0             = _mm256_setzero_pd();
 248             fjz0             = _mm256_setzero_pd();
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             r00              = _mm256_mul_pd(rsq00,rinv00);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 258                                             vdwioffsetptr0+vdwjidx0B,
 259                                             vdwioffsetptr0+vdwjidx0C,
 260                                             vdwioffsetptr0+vdwjidx0D,
 261                                             &c6_00,&c12_00);
 262
 263             /* Calculate table index by multiplying r with table scale and truncate to integer */
 264             rt               = _mm256_mul_pd(r00,vftabscale);
 265             vfitab           = _mm256_cvttpd_epi32(rt);
 266             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 267             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 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             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 301
 302             fscal            = fvdw;
 303
 304             /* Calculate temporary vectorial force */
 305             tx               = _mm256_mul_pd(fscal,dx00);
 306             ty               = _mm256_mul_pd(fscal,dy00);
 307             tz               = _mm256_mul_pd(fscal,dz00);
 308
 309             /* Update vectorial force */
 310             fix0             = _mm256_add_pd(fix0,tx);
 311             fiy0             = _mm256_add_pd(fiy0,ty);
 312             fiz0             = _mm256_add_pd(fiz0,tz);
 313
 314             fjx0             = _mm256_add_pd(fjx0,tx);
 315             fjy0             = _mm256_add_pd(fjy0,ty);
 316             fjz0             = _mm256_add_pd(fjz0,tz);
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             r10              = _mm256_mul_pd(rsq10,rinv10);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq10             = _mm256_mul_pd(iq1,jq0);
 326
 327             /* Calculate table index by multiplying r with table scale and truncate to integer */
 328             rt               = _mm256_mul_pd(r10,vftabscale);
 329             vfitab           = _mm256_cvttpd_epi32(rt);
 330             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 331             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 332
 333             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 334             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 335             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 336             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 337             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 338             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 339             Heps             = _mm256_mul_pd(vfeps,H);
 340             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 341             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 342             velec            = _mm256_mul_pd(qq10,VV);
 343             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 344             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm256_add_pd(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm256_mul_pd(fscal,dx10);
 353             ty               = _mm256_mul_pd(fscal,dy10);
 354             tz               = _mm256_mul_pd(fscal,dz10);
 355
 356             /* Update vectorial force */
 357             fix1             = _mm256_add_pd(fix1,tx);
 358             fiy1             = _mm256_add_pd(fiy1,ty);
 359             fiz1             = _mm256_add_pd(fiz1,tz);
 360
 361             fjx0             = _mm256_add_pd(fjx0,tx);
 362             fjy0             = _mm256_add_pd(fjy0,ty);
 363             fjz0             = _mm256_add_pd(fjz0,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             r20              = _mm256_mul_pd(rsq20,rinv20);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm256_mul_pd(iq2,jq0);
 373
 374             /* Calculate table index by multiplying r with table scale and truncate to integer */
 375             rt               = _mm256_mul_pd(r20,vftabscale);
 376             vfitab           = _mm256_cvttpd_epi32(rt);
 377             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 378             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 379
 380             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 381             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 382             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 383             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 384             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 385             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 386             Heps             = _mm256_mul_pd(vfeps,H);
 387             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 388             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 389             velec            = _mm256_mul_pd(qq20,VV);
 390             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 391             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm256_add_pd(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm256_mul_pd(fscal,dx20);
 400             ty               = _mm256_mul_pd(fscal,dy20);
 401             tz               = _mm256_mul_pd(fscal,dz20);
 402
 403             /* Update vectorial force */
 404             fix2             = _mm256_add_pd(fix2,tx);
 405             fiy2             = _mm256_add_pd(fiy2,ty);
 406             fiz2             = _mm256_add_pd(fiz2,tz);
 407
 408             fjx0             = _mm256_add_pd(fjx0,tx);
 409             fjy0             = _mm256_add_pd(fjy0,ty);
 410             fjz0             = _mm256_add_pd(fjz0,tz);
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             r30              = _mm256_mul_pd(rsq30,rinv30);
 417
 418             /* Compute parameters for interactions between i and j atoms */
 419             qq30             = _mm256_mul_pd(iq3,jq0);
 420
 421             /* Calculate table index by multiplying r with table scale and truncate to integer */
 422             rt               = _mm256_mul_pd(r30,vftabscale);
 423             vfitab           = _mm256_cvttpd_epi32(rt);
 424             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 425             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 426
 427             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 428             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 429             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 430             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 431             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 432             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 433             Heps             = _mm256_mul_pd(vfeps,H);
 434             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 435             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 436             velec            = _mm256_mul_pd(qq30,VV);
 437             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 438             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 439
 440             /* Update potential sum for this i atom from the interaction with this j atom. */
 441             velecsum         = _mm256_add_pd(velecsum,velec);
 442
 443             fscal            = felec;
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm256_mul_pd(fscal,dx30);
 447             ty               = _mm256_mul_pd(fscal,dy30);
 448             tz               = _mm256_mul_pd(fscal,dz30);
 449
 450             /* Update vectorial force */
 451             fix3             = _mm256_add_pd(fix3,tx);
 452             fiy3             = _mm256_add_pd(fiy3,ty);
 453             fiz3             = _mm256_add_pd(fiz3,tz);
 454
 455             fjx0             = _mm256_add_pd(fjx0,tx);
 456             fjy0             = _mm256_add_pd(fjy0,ty);
 457             fjz0             = _mm256_add_pd(fjz0,tz);
 458
 459             fjptrA             = f+j_coord_offsetA;
 460             fjptrB             = f+j_coord_offsetB;
 461             fjptrC             = f+j_coord_offsetC;
 462             fjptrD             = f+j_coord_offsetD;
 463
 464             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 465
 466             /* Inner loop uses 188 flops */
 467         }
 468
 469         if(jidx<j_index_end)
 470         {
 471
 472             /* Get j neighbor index, and coordinate index */
 473             jnrlistA         = jjnr[jidx];
 474             jnrlistB         = jjnr[jidx+1];
 475             jnrlistC         = jjnr[jidx+2];
 476             jnrlistD         = jjnr[jidx+3];
 477             /* Sign of each element will be negative for non-real atoms.
 478              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 479              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 480              */
 481             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 482
 483             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 484             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 485             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 486
 487             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 488             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 489             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 490             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 491             j_coord_offsetA  = DIM*jnrA;
 492             j_coord_offsetB  = DIM*jnrB;
 493             j_coord_offsetC  = DIM*jnrC;
 494             j_coord_offsetD  = DIM*jnrD;
 495
 496             /* load j atom coordinates */
 497             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 498                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 499                                                  &jx0,&jy0,&jz0);
 500
 501             /* Calculate displacement vector */
 502             dx00             = _mm256_sub_pd(ix0,jx0);
 503             dy00             = _mm256_sub_pd(iy0,jy0);
 504             dz00             = _mm256_sub_pd(iz0,jz0);
 505             dx10             = _mm256_sub_pd(ix1,jx0);
 506             dy10             = _mm256_sub_pd(iy1,jy0);
 507             dz10             = _mm256_sub_pd(iz1,jz0);
 508             dx20             = _mm256_sub_pd(ix2,jx0);
 509             dy20             = _mm256_sub_pd(iy2,jy0);
 510             dz20             = _mm256_sub_pd(iz2,jz0);
 511             dx30             = _mm256_sub_pd(ix3,jx0);
 512             dy30             = _mm256_sub_pd(iy3,jy0);
 513             dz30             = _mm256_sub_pd(iz3,jz0);
 514
 515             /* Calculate squared distance and things based on it */
 516             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 517             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 518             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 519             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 520
 521             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 522             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 523             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 524             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 525
 526             /* Load parameters for j particles */
 527             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 528                                                                  charge+jnrC+0,charge+jnrD+0);
 529             vdwjidx0A        = 2*vdwtype[jnrA+0];
 530             vdwjidx0B        = 2*vdwtype[jnrB+0];
 531             vdwjidx0C        = 2*vdwtype[jnrC+0];
 532             vdwjidx0D        = 2*vdwtype[jnrD+0];
 533
 534             fjx0             = _mm256_setzero_pd();
 535             fjy0             = _mm256_setzero_pd();
 536             fjz0             = _mm256_setzero_pd();
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             r00              = _mm256_mul_pd(rsq00,rinv00);
 543             r00              = _mm256_andnot_pd(dummy_mask,r00);
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 547                                             vdwioffsetptr0+vdwjidx0B,
 548                                             vdwioffsetptr0+vdwjidx0C,
 549                                             vdwioffsetptr0+vdwjidx0D,
 550                                             &c6_00,&c12_00);
 551
 552             /* Calculate table index by multiplying r with table scale and truncate to integer */
 553             rt               = _mm256_mul_pd(r00,vftabscale);
 554             vfitab           = _mm256_cvttpd_epi32(rt);
 555             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 556             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 557
 558             /* CUBIC SPLINE TABLE DISPERSION */
 559             vfitab           = _mm_add_epi32(vfitab,ifour);
 560             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 561             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 562             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 563             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 564             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 565             Heps             = _mm256_mul_pd(vfeps,H);
 566             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 567             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 568             vvdw6            = _mm256_mul_pd(c6_00,VV);
 569             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 570             fvdw6            = _mm256_mul_pd(c6_00,FF);
 571
 572             /* CUBIC SPLINE TABLE REPULSION */
 573             vfitab           = _mm_add_epi32(vfitab,ifour);
 574             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 575             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 576             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 577             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 578             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 579             Heps             = _mm256_mul_pd(vfeps,H);
 580             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 581             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 582             vvdw12           = _mm256_mul_pd(c12_00,VV);
 583             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 584             fvdw12           = _mm256_mul_pd(c12_00,FF);
 585             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 586             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 590             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 591
 592             fscal            = fvdw;
 593
 594             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm256_mul_pd(fscal,dx00);
 598             ty               = _mm256_mul_pd(fscal,dy00);
 599             tz               = _mm256_mul_pd(fscal,dz00);
 600
 601             /* Update vectorial force */
 602             fix0             = _mm256_add_pd(fix0,tx);
 603             fiy0             = _mm256_add_pd(fiy0,ty);
 604             fiz0             = _mm256_add_pd(fiz0,tz);
 605
 606             fjx0             = _mm256_add_pd(fjx0,tx);
 607             fjy0             = _mm256_add_pd(fjy0,ty);
 608             fjz0             = _mm256_add_pd(fjz0,tz);
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             r10              = _mm256_mul_pd(rsq10,rinv10);
 615             r10              = _mm256_andnot_pd(dummy_mask,r10);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq10             = _mm256_mul_pd(iq1,jq0);
 619
 620             /* Calculate table index by multiplying r with table scale and truncate to integer */
 621             rt               = _mm256_mul_pd(r10,vftabscale);
 622             vfitab           = _mm256_cvttpd_epi32(rt);
 623             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 624             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 625
 626             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 627             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 628             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 629             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 630             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 631             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 632             Heps             = _mm256_mul_pd(vfeps,H);
 633             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 634             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 635             velec            = _mm256_mul_pd(qq10,VV);
 636             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 637             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm256_andnot_pd(dummy_mask,velec);
 641             velecsum         = _mm256_add_pd(velecsum,velec);
 642
 643             fscal            = felec;
 644
 645             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm256_mul_pd(fscal,dx10);
 649             ty               = _mm256_mul_pd(fscal,dy10);
 650             tz               = _mm256_mul_pd(fscal,dz10);
 651
 652             /* Update vectorial force */
 653             fix1             = _mm256_add_pd(fix1,tx);
 654             fiy1             = _mm256_add_pd(fiy1,ty);
 655             fiz1             = _mm256_add_pd(fiz1,tz);
 656
 657             fjx0             = _mm256_add_pd(fjx0,tx);
 658             fjy0             = _mm256_add_pd(fjy0,ty);
 659             fjz0             = _mm256_add_pd(fjz0,tz);
 660
 661             /**************************
 662              * CALCULATE INTERACTIONS *
 663              **************************/
 664
 665             r20              = _mm256_mul_pd(rsq20,rinv20);
 666             r20              = _mm256_andnot_pd(dummy_mask,r20);
 667
 668             /* Compute parameters for interactions between i and j atoms */
 669             qq20             = _mm256_mul_pd(iq2,jq0);
 670
 671             /* Calculate table index by multiplying r with table scale and truncate to integer */
 672             rt               = _mm256_mul_pd(r20,vftabscale);
 673             vfitab           = _mm256_cvttpd_epi32(rt);
 674             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 675             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 676
 677             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 678             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 679             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 680             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 681             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 682             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 683             Heps             = _mm256_mul_pd(vfeps,H);
 684             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 685             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 686             velec            = _mm256_mul_pd(qq20,VV);
 687             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 688             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 689
 690             /* Update potential sum for this i atom from the interaction with this j atom. */
 691             velec            = _mm256_andnot_pd(dummy_mask,velec);
 692             velecsum         = _mm256_add_pd(velecsum,velec);
 693
 694             fscal            = felec;
 695
 696             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 697
 698             /* Calculate temporary vectorial force */
 699             tx               = _mm256_mul_pd(fscal,dx20);
 700             ty               = _mm256_mul_pd(fscal,dy20);
 701             tz               = _mm256_mul_pd(fscal,dz20);
 702
 703             /* Update vectorial force */
 704             fix2             = _mm256_add_pd(fix2,tx);
 705             fiy2             = _mm256_add_pd(fiy2,ty);
 706             fiz2             = _mm256_add_pd(fiz2,tz);
 707
 708             fjx0             = _mm256_add_pd(fjx0,tx);
 709             fjy0             = _mm256_add_pd(fjy0,ty);
 710             fjz0             = _mm256_add_pd(fjz0,tz);
 711
 712             /**************************
 713              * CALCULATE INTERACTIONS *
 714              **************************/
 715
 716             r30              = _mm256_mul_pd(rsq30,rinv30);
 717             r30              = _mm256_andnot_pd(dummy_mask,r30);
 718
 719             /* Compute parameters for interactions between i and j atoms */
 720             qq30             = _mm256_mul_pd(iq3,jq0);
 721
 722             /* Calculate table index by multiplying r with table scale and truncate to integer */
 723             rt               = _mm256_mul_pd(r30,vftabscale);
 724             vfitab           = _mm256_cvttpd_epi32(rt);
 725             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 726             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 727
 728             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 729             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 730             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 731             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 732             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 733             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 734             Heps             = _mm256_mul_pd(vfeps,H);
 735             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 736             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 737             velec            = _mm256_mul_pd(qq30,VV);
 738             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 739             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
 740
 741             /* Update potential sum for this i atom from the interaction with this j atom. */
 742             velec            = _mm256_andnot_pd(dummy_mask,velec);
 743             velecsum         = _mm256_add_pd(velecsum,velec);
 744
 745             fscal            = felec;
 746
 747             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 748
 749             /* Calculate temporary vectorial force */
 750             tx               = _mm256_mul_pd(fscal,dx30);
 751             ty               = _mm256_mul_pd(fscal,dy30);
 752             tz               = _mm256_mul_pd(fscal,dz30);
 753
 754             /* Update vectorial force */
 755             fix3             = _mm256_add_pd(fix3,tx);
 756             fiy3             = _mm256_add_pd(fiy3,ty);
 757             fiz3             = _mm256_add_pd(fiz3,tz);
 758
 759             fjx0             = _mm256_add_pd(fjx0,tx);
 760             fjy0             = _mm256_add_pd(fjy0,ty);
 761             fjz0             = _mm256_add_pd(fjz0,tz);
 762
 763             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 764             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 765             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 766             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 767
 768             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 769
 770             /* Inner loop uses 192 flops */
 771         }
 772
 773         /* End of innermost loop */
 774
 775         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 776                                                  f+i_coord_offset,fshift+i_shift_offset);
 777
 778         ggid                        = gid[iidx];
 779         /* Update potential energies */
 780         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 781         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 782
 783         /* Increment number of inner iterations */
 784         inneriter                  += j_index_end - j_index_start;
 785
 786         /* Outer loop uses 26 flops */
 787     }
 788
 789     /* Increment number of outer iterations */
 790     outeriter        += nri;
 791
 792     /* Update outer/inner flops */
 793
 794     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*192);
 795 }
 796 /*
 797  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_256_double
 798  * Electrostatics interaction: CubicSplineTable
 799  * VdW interaction:            CubicSplineTable
 800  * Geometry:                   Water4-Particle
 801  * Calculate force/pot:        Force
 802  */
 803 void
 804 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_256_double
 805                     (t_nblist                    * gmx_restrict       nlist,
 806                      rvec                        * gmx_restrict          xx,
 807                      rvec                        * gmx_restrict          ff,
 808                      t_forcerec                  * gmx_restrict          fr,
 809                      t_mdatoms                   * gmx_restrict     mdatoms,
 810                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 811                      t_nrnb                      * gmx_restrict        nrnb)
 812 {
 813     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 814      * just 0 for non-waters.
 815      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 816      * jnr indices corresponding to data put in the four positions in the SIMD register.
 817      */
 818     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 819     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 820     int              jnrA,jnrB,jnrC,jnrD;
 821     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 822     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 823     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 824     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 825     real             rcutoff_scalar;
 826     real             *shiftvec,*fshift,*x,*f;
 827     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 828     real             scratch[4*DIM];
 829     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 830     real *           vdwioffsetptr0;
 831     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 832     real *           vdwioffsetptr1;
 833     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 834     real *           vdwioffsetptr2;
 835     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 836     real *           vdwioffsetptr3;
 837     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 838     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 839     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 840     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 841     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 842     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 843     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 844     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 845     real             *charge;
 846     int              nvdwtype;
 847     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 848     int              *vdwtype;
 849     real             *vdwparam;
 850     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 851     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 852     __m128i          vfitab;
 853     __m128i          ifour       = _mm_set1_epi32(4);
 854     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 855     real             *vftab;
 856     __m256d          dummy_mask,cutoff_mask;
 857     __m128           tmpmask0,tmpmask1;
 858     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 859     __m256d          one     = _mm256_set1_pd(1.0);
 860     __m256d          two     = _mm256_set1_pd(2.0);
 861     x                = xx[0];
 862     f                = ff[0];
 863
 864     nri              = nlist->nri;
 865     iinr             = nlist->iinr;
 866     jindex           = nlist->jindex;
 867     jjnr             = nlist->jjnr;
 868     shiftidx         = nlist->shift;
 869     gid              = nlist->gid;
 870     shiftvec         = fr->shift_vec[0];
 871     fshift           = fr->fshift[0];
 872     facel            = _mm256_set1_pd(fr->epsfac);
 873     charge           = mdatoms->chargeA;
 874     nvdwtype         = fr->ntype;
 875     vdwparam         = fr->nbfp;
 876     vdwtype          = mdatoms->typeA;
 877
 878     vftab            = kernel_data->table_elec_vdw->data;
 879     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 880
 881     /* Setup water-specific parameters */
 882     inr              = nlist->iinr[0];
 883     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 884     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 885     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 886     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 887
 888     /* Avoid stupid compiler warnings */
 889     jnrA = jnrB = jnrC = jnrD = 0;
 890     j_coord_offsetA = 0;
 891     j_coord_offsetB = 0;
 892     j_coord_offsetC = 0;
 893     j_coord_offsetD = 0;
 894
 895     outeriter        = 0;
 896     inneriter        = 0;
 897
 898     for(iidx=0;iidx<4*DIM;iidx++)
 899     {
 900         scratch[iidx] = 0.0;
 901     }
 902
 903     /* Start outer loop over neighborlists */
 904     for(iidx=0; iidx<nri; iidx++)
 905     {
 906         /* Load shift vector for this list */
 907         i_shift_offset   = DIM*shiftidx[iidx];
 908
 909         /* Load limits for loop over neighbors */
 910         j_index_start    = jindex[iidx];
 911         j_index_end      = jindex[iidx+1];
 912
 913         /* Get outer coordinate index */
 914         inr              = iinr[iidx];
 915         i_coord_offset   = DIM*inr;
 916
 917         /* Load i particle coords and add shift vector */
 918         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 919                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 920
 921         fix0             = _mm256_setzero_pd();
 922         fiy0             = _mm256_setzero_pd();
 923         fiz0             = _mm256_setzero_pd();
 924         fix1             = _mm256_setzero_pd();
 925         fiy1             = _mm256_setzero_pd();
 926         fiz1             = _mm256_setzero_pd();
 927         fix2             = _mm256_setzero_pd();
 928         fiy2             = _mm256_setzero_pd();
 929         fiz2             = _mm256_setzero_pd();
 930         fix3             = _mm256_setzero_pd();
 931         fiy3             = _mm256_setzero_pd();
 932         fiz3             = _mm256_setzero_pd();
 933
 934         /* Start inner kernel loop */
 935         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 936         {
 937
 938             /* Get j neighbor index, and coordinate index */
 939             jnrA             = jjnr[jidx];
 940             jnrB             = jjnr[jidx+1];
 941             jnrC             = jjnr[jidx+2];
 942             jnrD             = jjnr[jidx+3];
 943             j_coord_offsetA  = DIM*jnrA;
 944             j_coord_offsetB  = DIM*jnrB;
 945             j_coord_offsetC  = DIM*jnrC;
 946             j_coord_offsetD  = DIM*jnrD;
 947
 948             /* load j atom coordinates */
 949             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 950                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 951                                                  &jx0,&jy0,&jz0);
 952
 953             /* Calculate displacement vector */
 954             dx00             = _mm256_sub_pd(ix0,jx0);
 955             dy00             = _mm256_sub_pd(iy0,jy0);
 956             dz00             = _mm256_sub_pd(iz0,jz0);
 957             dx10             = _mm256_sub_pd(ix1,jx0);
 958             dy10             = _mm256_sub_pd(iy1,jy0);
 959             dz10             = _mm256_sub_pd(iz1,jz0);
 960             dx20             = _mm256_sub_pd(ix2,jx0);
 961             dy20             = _mm256_sub_pd(iy2,jy0);
 962             dz20             = _mm256_sub_pd(iz2,jz0);
 963             dx30             = _mm256_sub_pd(ix3,jx0);
 964             dy30             = _mm256_sub_pd(iy3,jy0);
 965             dz30             = _mm256_sub_pd(iz3,jz0);
 966
 967             /* Calculate squared distance and things based on it */
 968             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 969             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 970             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 971             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 972
 973             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 974             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 975             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 976             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 977
 978             /* Load parameters for j particles */
 979             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 980                                                                  charge+jnrC+0,charge+jnrD+0);
 981             vdwjidx0A        = 2*vdwtype[jnrA+0];
 982             vdwjidx0B        = 2*vdwtype[jnrB+0];
 983             vdwjidx0C        = 2*vdwtype[jnrC+0];
 984             vdwjidx0D        = 2*vdwtype[jnrD+0];
 985
 986             fjx0             = _mm256_setzero_pd();
 987             fjy0             = _mm256_setzero_pd();
 988             fjz0             = _mm256_setzero_pd();
 989
 990             /**************************
 991              * CALCULATE INTERACTIONS *
 992              **************************/
 993
 994             r00              = _mm256_mul_pd(rsq00,rinv00);
 995
 996             /* Compute parameters for interactions between i and j atoms */
 997             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 998                                             vdwioffsetptr0+vdwjidx0B,
 999                                             vdwioffsetptr0+vdwjidx0C,
1000                                             vdwioffsetptr0+vdwjidx0D,
1001                                             &c6_00,&c12_00);
1002
1003             /* Calculate table index by multiplying r with table scale and truncate to integer */
1004             rt               = _mm256_mul_pd(r00,vftabscale);
1005             vfitab           = _mm256_cvttpd_epi32(rt);
1006             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1007             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1008
1009             /* CUBIC SPLINE TABLE DISPERSION */
1010             vfitab           = _mm_add_epi32(vfitab,ifour);
1011             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1012             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1013             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1014             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1015             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1016             Heps             = _mm256_mul_pd(vfeps,H);
1017             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1018             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1019             fvdw6            = _mm256_mul_pd(c6_00,FF);
1020
1021             /* CUBIC SPLINE TABLE REPULSION */
1022             vfitab           = _mm_add_epi32(vfitab,ifour);
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             fvdw12           = _mm256_mul_pd(c12_00,FF);
1032             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1033
1034             fscal            = fvdw;
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm256_mul_pd(fscal,dx00);
1038             ty               = _mm256_mul_pd(fscal,dy00);
1039             tz               = _mm256_mul_pd(fscal,dz00);
1040
1041             /* Update vectorial force */
1042             fix0             = _mm256_add_pd(fix0,tx);
1043             fiy0             = _mm256_add_pd(fiy0,ty);
1044             fiz0             = _mm256_add_pd(fiz0,tz);
1045
1046             fjx0             = _mm256_add_pd(fjx0,tx);
1047             fjy0             = _mm256_add_pd(fjy0,ty);
1048             fjz0             = _mm256_add_pd(fjz0,tz);
1049
1050             /**************************
1051              * CALCULATE INTERACTIONS *
1052              **************************/
1053
1054             r10              = _mm256_mul_pd(rsq10,rinv10);
1055
1056             /* Compute parameters for interactions between i and j atoms */
1057             qq10             = _mm256_mul_pd(iq1,jq0);
1058
1059             /* Calculate table index by multiplying r with table scale and truncate to integer */
1060             rt               = _mm256_mul_pd(r10,vftabscale);
1061             vfitab           = _mm256_cvttpd_epi32(rt);
1062             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1063             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1064
1065             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1066             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1067             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1068             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1069             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1070             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1071             Heps             = _mm256_mul_pd(vfeps,H);
1072             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1073             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1074             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1075
1076             fscal            = felec;
1077
1078             /* Calculate temporary vectorial force */
1079             tx               = _mm256_mul_pd(fscal,dx10);
1080             ty               = _mm256_mul_pd(fscal,dy10);
1081             tz               = _mm256_mul_pd(fscal,dz10);
1082
1083             /* Update vectorial force */
1084             fix1             = _mm256_add_pd(fix1,tx);
1085             fiy1             = _mm256_add_pd(fiy1,ty);
1086             fiz1             = _mm256_add_pd(fiz1,tz);
1087
1088             fjx0             = _mm256_add_pd(fjx0,tx);
1089             fjy0             = _mm256_add_pd(fjy0,ty);
1090             fjz0             = _mm256_add_pd(fjz0,tz);
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             r20              = _mm256_mul_pd(rsq20,rinv20);
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq20             = _mm256_mul_pd(iq2,jq0);
1100
1101             /* Calculate table index by multiplying r with table scale and truncate to integer */
1102             rt               = _mm256_mul_pd(r20,vftabscale);
1103             vfitab           = _mm256_cvttpd_epi32(rt);
1104             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1105             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1106
1107             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1108             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1109             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1110             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1111             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1112             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1113             Heps             = _mm256_mul_pd(vfeps,H);
1114             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1115             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1116             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1117
1118             fscal            = felec;
1119
1120             /* Calculate temporary vectorial force */
1121             tx               = _mm256_mul_pd(fscal,dx20);
1122             ty               = _mm256_mul_pd(fscal,dy20);
1123             tz               = _mm256_mul_pd(fscal,dz20);
1124
1125             /* Update vectorial force */
1126             fix2             = _mm256_add_pd(fix2,tx);
1127             fiy2             = _mm256_add_pd(fiy2,ty);
1128             fiz2             = _mm256_add_pd(fiz2,tz);
1129
1130             fjx0             = _mm256_add_pd(fjx0,tx);
1131             fjy0             = _mm256_add_pd(fjy0,ty);
1132             fjz0             = _mm256_add_pd(fjz0,tz);
1133
1134             /**************************
1135              * CALCULATE INTERACTIONS *
1136              **************************/
1137
1138             r30              = _mm256_mul_pd(rsq30,rinv30);
1139
1140             /* Compute parameters for interactions between i and j atoms */
1141             qq30             = _mm256_mul_pd(iq3,jq0);
1142
1143             /* Calculate table index by multiplying r with table scale and truncate to integer */
1144             rt               = _mm256_mul_pd(r30,vftabscale);
1145             vfitab           = _mm256_cvttpd_epi32(rt);
1146             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1147             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1148
1149             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1150             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1151             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1152             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1153             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1154             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1155             Heps             = _mm256_mul_pd(vfeps,H);
1156             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1157             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1158             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
1159
1160             fscal            = felec;
1161
1162             /* Calculate temporary vectorial force */
1163             tx               = _mm256_mul_pd(fscal,dx30);
1164             ty               = _mm256_mul_pd(fscal,dy30);
1165             tz               = _mm256_mul_pd(fscal,dz30);
1166
1167             /* Update vectorial force */
1168             fix3             = _mm256_add_pd(fix3,tx);
1169             fiy3             = _mm256_add_pd(fiy3,ty);
1170             fiz3             = _mm256_add_pd(fiz3,tz);
1171
1172             fjx0             = _mm256_add_pd(fjx0,tx);
1173             fjy0             = _mm256_add_pd(fjy0,ty);
1174             fjz0             = _mm256_add_pd(fjz0,tz);
1175
1176             fjptrA             = f+j_coord_offsetA;
1177             fjptrB             = f+j_coord_offsetB;
1178             fjptrC             = f+j_coord_offsetC;
1179             fjptrD             = f+j_coord_offsetD;
1180
1181             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1182
1183             /* Inner loop uses 168 flops */
1184         }
1185
1186         if(jidx<j_index_end)
1187         {
1188
1189             /* Get j neighbor index, and coordinate index */
1190             jnrlistA         = jjnr[jidx];
1191             jnrlistB         = jjnr[jidx+1];
1192             jnrlistC         = jjnr[jidx+2];
1193             jnrlistD         = jjnr[jidx+3];
1194             /* Sign of each element will be negative for non-real atoms.
1195              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1196              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1197              */
1198             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1199
1200             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1201             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1202             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1203
1204             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1205             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1206             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1207             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1208             j_coord_offsetA  = DIM*jnrA;
1209             j_coord_offsetB  = DIM*jnrB;
1210             j_coord_offsetC  = DIM*jnrC;
1211             j_coord_offsetD  = DIM*jnrD;
1212
1213             /* load j atom coordinates */
1214             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1215                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1216                                                  &jx0,&jy0,&jz0);
1217
1218             /* Calculate displacement vector */
1219             dx00             = _mm256_sub_pd(ix0,jx0);
1220             dy00             = _mm256_sub_pd(iy0,jy0);
1221             dz00             = _mm256_sub_pd(iz0,jz0);
1222             dx10             = _mm256_sub_pd(ix1,jx0);
1223             dy10             = _mm256_sub_pd(iy1,jy0);
1224             dz10             = _mm256_sub_pd(iz1,jz0);
1225             dx20             = _mm256_sub_pd(ix2,jx0);
1226             dy20             = _mm256_sub_pd(iy2,jy0);
1227             dz20             = _mm256_sub_pd(iz2,jz0);
1228             dx30             = _mm256_sub_pd(ix3,jx0);
1229             dy30             = _mm256_sub_pd(iy3,jy0);
1230             dz30             = _mm256_sub_pd(iz3,jz0);
1231
1232             /* Calculate squared distance and things based on it */
1233             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1234             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1235             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1236             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1237
1238             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1239             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1240             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1241             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1242
1243             /* Load parameters for j particles */
1244             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1245                                                                  charge+jnrC+0,charge+jnrD+0);
1246             vdwjidx0A        = 2*vdwtype[jnrA+0];
1247             vdwjidx0B        = 2*vdwtype[jnrB+0];
1248             vdwjidx0C        = 2*vdwtype[jnrC+0];
1249             vdwjidx0D        = 2*vdwtype[jnrD+0];
1250
1251             fjx0             = _mm256_setzero_pd();
1252             fjy0             = _mm256_setzero_pd();
1253             fjz0             = _mm256_setzero_pd();
1254
1255             /**************************
1256              * CALCULATE INTERACTIONS *
1257              **************************/
1258
1259             r00              = _mm256_mul_pd(rsq00,rinv00);
1260             r00              = _mm256_andnot_pd(dummy_mask,r00);
1261
1262             /* Compute parameters for interactions between i and j atoms */
1263             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1264                                             vdwioffsetptr0+vdwjidx0B,
1265                                             vdwioffsetptr0+vdwjidx0C,
1266                                             vdwioffsetptr0+vdwjidx0D,
1267                                             &c6_00,&c12_00);
1268
1269             /* Calculate table index by multiplying r with table scale and truncate to integer */
1270             rt               = _mm256_mul_pd(r00,vftabscale);
1271             vfitab           = _mm256_cvttpd_epi32(rt);
1272             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1273             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1274
1275             /* CUBIC SPLINE TABLE DISPERSION */
1276             vfitab           = _mm_add_epi32(vfitab,ifour);
1277             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1278             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1279             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1280             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1281             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1282             Heps             = _mm256_mul_pd(vfeps,H);
1283             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1284             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1285             fvdw6            = _mm256_mul_pd(c6_00,FF);
1286
1287             /* CUBIC SPLINE TABLE REPULSION */
1288             vfitab           = _mm_add_epi32(vfitab,ifour);
1289             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1290             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1291             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1292             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1293             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1294             Heps             = _mm256_mul_pd(vfeps,H);
1295             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1296             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1297             fvdw12           = _mm256_mul_pd(c12_00,FF);
1298             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1299
1300             fscal            = fvdw;
1301
1302             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1303
1304             /* Calculate temporary vectorial force */
1305             tx               = _mm256_mul_pd(fscal,dx00);
1306             ty               = _mm256_mul_pd(fscal,dy00);
1307             tz               = _mm256_mul_pd(fscal,dz00);
1308
1309             /* Update vectorial force */
1310             fix0             = _mm256_add_pd(fix0,tx);
1311             fiy0             = _mm256_add_pd(fiy0,ty);
1312             fiz0             = _mm256_add_pd(fiz0,tz);
1313
1314             fjx0             = _mm256_add_pd(fjx0,tx);
1315             fjy0             = _mm256_add_pd(fjy0,ty);
1316             fjz0             = _mm256_add_pd(fjz0,tz);
1317
1318             /**************************
1319              * CALCULATE INTERACTIONS *
1320              **************************/
1321
1322             r10              = _mm256_mul_pd(rsq10,rinv10);
1323             r10              = _mm256_andnot_pd(dummy_mask,r10);
1324
1325             /* Compute parameters for interactions between i and j atoms */
1326             qq10             = _mm256_mul_pd(iq1,jq0);
1327
1328             /* Calculate table index by multiplying r with table scale and truncate to integer */
1329             rt               = _mm256_mul_pd(r10,vftabscale);
1330             vfitab           = _mm256_cvttpd_epi32(rt);
1331             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1332             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1333
1334             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1335             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1336             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1337             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1338             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1339             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1340             Heps             = _mm256_mul_pd(vfeps,H);
1341             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1342             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1343             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1344
1345             fscal            = felec;
1346
1347             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1348
1349             /* Calculate temporary vectorial force */
1350             tx               = _mm256_mul_pd(fscal,dx10);
1351             ty               = _mm256_mul_pd(fscal,dy10);
1352             tz               = _mm256_mul_pd(fscal,dz10);
1353
1354             /* Update vectorial force */
1355             fix1             = _mm256_add_pd(fix1,tx);
1356             fiy1             = _mm256_add_pd(fiy1,ty);
1357             fiz1             = _mm256_add_pd(fiz1,tz);
1358
1359             fjx0             = _mm256_add_pd(fjx0,tx);
1360             fjy0             = _mm256_add_pd(fjy0,ty);
1361             fjz0             = _mm256_add_pd(fjz0,tz);
1362
1363             /**************************
1364              * CALCULATE INTERACTIONS *
1365              **************************/
1366
1367             r20              = _mm256_mul_pd(rsq20,rinv20);
1368             r20              = _mm256_andnot_pd(dummy_mask,r20);
1369
1370             /* Compute parameters for interactions between i and j atoms */
1371             qq20             = _mm256_mul_pd(iq2,jq0);
1372
1373             /* Calculate table index by multiplying r with table scale and truncate to integer */
1374             rt               = _mm256_mul_pd(r20,vftabscale);
1375             vfitab           = _mm256_cvttpd_epi32(rt);
1376             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1377             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1378
1379             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1380             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1381             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1382             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1383             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1384             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1385             Heps             = _mm256_mul_pd(vfeps,H);
1386             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1387             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1388             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1389
1390             fscal            = felec;
1391
1392             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1393
1394             /* Calculate temporary vectorial force */
1395             tx               = _mm256_mul_pd(fscal,dx20);
1396             ty               = _mm256_mul_pd(fscal,dy20);
1397             tz               = _mm256_mul_pd(fscal,dz20);
1398
1399             /* Update vectorial force */
1400             fix2             = _mm256_add_pd(fix2,tx);
1401             fiy2             = _mm256_add_pd(fiy2,ty);
1402             fiz2             = _mm256_add_pd(fiz2,tz);
1403
1404             fjx0             = _mm256_add_pd(fjx0,tx);
1405             fjy0             = _mm256_add_pd(fjy0,ty);
1406             fjz0             = _mm256_add_pd(fjz0,tz);
1407
1408             /**************************
1409              * CALCULATE INTERACTIONS *
1410              **************************/
1411
1412             r30              = _mm256_mul_pd(rsq30,rinv30);
1413             r30              = _mm256_andnot_pd(dummy_mask,r30);
1414
1415             /* Compute parameters for interactions between i and j atoms */
1416             qq30             = _mm256_mul_pd(iq3,jq0);
1417
1418             /* Calculate table index by multiplying r with table scale and truncate to integer */
1419             rt               = _mm256_mul_pd(r30,vftabscale);
1420             vfitab           = _mm256_cvttpd_epi32(rt);
1421             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1422             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1423
1424             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1425             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1426             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1427             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1428             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1429             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1430             Heps             = _mm256_mul_pd(vfeps,H);
1431             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1432             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1433             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq30,FF),_mm256_mul_pd(vftabscale,rinv30)));
1434
1435             fscal            = felec;
1436
1437             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1438
1439             /* Calculate temporary vectorial force */
1440             tx               = _mm256_mul_pd(fscal,dx30);
1441             ty               = _mm256_mul_pd(fscal,dy30);
1442             tz               = _mm256_mul_pd(fscal,dz30);
1443
1444             /* Update vectorial force */
1445             fix3             = _mm256_add_pd(fix3,tx);
1446             fiy3             = _mm256_add_pd(fiy3,ty);
1447             fiz3             = _mm256_add_pd(fiz3,tz);
1448
1449             fjx0             = _mm256_add_pd(fjx0,tx);
1450             fjy0             = _mm256_add_pd(fjy0,ty);
1451             fjz0             = _mm256_add_pd(fjz0,tz);
1452
1453             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1454             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1455             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1456             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1457
1458             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1459
1460             /* Inner loop uses 172 flops */
1461         }
1462
1463         /* End of innermost loop */
1464
1465         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1466                                                  f+i_coord_offset,fshift+i_shift_offset);
1467
1468         /* Increment number of inner iterations */
1469         inneriter                  += j_index_end - j_index_start;
1470
1471         /* Outer loop uses 24 flops */
1472     }
1473
1474     /* Increment number of outer iterations */
1475     outeriter        += nri;
1476
1477     /* Update outer/inner flops */
1478
1479     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*172);
1480 }