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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m128i          vfitab;
 106     __m128i          ifour       = _mm_set1_epi32(4);
 107     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 108     real             *vftab;
 109     __m256d          dummy_mask,cutoff_mask;
 110     __m128           tmpmask0,tmpmask1;
 111     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 112     __m256d          one     = _mm256_set1_pd(1.0);
 113     __m256d          two     = _mm256_set1_pd(2.0);
 114     x                = xx[0];
 115     f                = ff[0];
 116
 117     nri              = nlist->nri;
 118     iinr             = nlist->iinr;
 119     jindex           = nlist->jindex;
 120     jjnr             = nlist->jjnr;
 121     shiftidx         = nlist->shift;
 122     gid              = nlist->gid;
 123     shiftvec         = fr->shift_vec[0];
 124     fshift           = fr->fshift[0];
 125     facel            = _mm256_set1_pd(fr->epsfac);
 126     charge           = mdatoms->chargeA;
 127     nvdwtype         = fr->ntype;
 128     vdwparam         = fr->nbfp;
 129     vdwtype          = mdatoms->typeA;
 130
 131     vftab            = kernel_data->table_elec_vdw->data;
 132     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 133
 134     /* Setup water-specific parameters */
 135     inr              = nlist->iinr[0];
 136     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 137     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 138     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 139     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 173
 174         fix0             = _mm256_setzero_pd();
 175         fiy0             = _mm256_setzero_pd();
 176         fiz0             = _mm256_setzero_pd();
 177         fix1             = _mm256_setzero_pd();
 178         fiy1             = _mm256_setzero_pd();
 179         fiz1             = _mm256_setzero_pd();
 180         fix2             = _mm256_setzero_pd();
 181         fiy2             = _mm256_setzero_pd();
 182         fiz2             = _mm256_setzero_pd();
 183
 184         /* Reset potential sums */
 185         velecsum         = _mm256_setzero_pd();
 186         vvdwsum          = _mm256_setzero_pd();
 187
 188         /* Start inner kernel loop */
 189         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 190         {
 191
 192             /* Get j neighbor index, and coordinate index */
 193             jnrA             = jjnr[jidx];
 194             jnrB             = jjnr[jidx+1];
 195             jnrC             = jjnr[jidx+2];
 196             jnrD             = jjnr[jidx+3];
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 205                                                  &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm256_sub_pd(ix0,jx0);
 209             dy00             = _mm256_sub_pd(iy0,jy0);
 210             dz00             = _mm256_sub_pd(iz0,jz0);
 211             dx10             = _mm256_sub_pd(ix1,jx0);
 212             dy10             = _mm256_sub_pd(iy1,jy0);
 213             dz10             = _mm256_sub_pd(iz1,jz0);
 214             dx20             = _mm256_sub_pd(ix2,jx0);
 215             dy20             = _mm256_sub_pd(iy2,jy0);
 216             dz20             = _mm256_sub_pd(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 220             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 221             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 222
 223             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 224             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 225             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 229                                                                  charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             fjx0             = _mm256_setzero_pd();
 236             fjy0             = _mm256_setzero_pd();
 237             fjz0             = _mm256_setzero_pd();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             r00              = _mm256_mul_pd(rsq00,rinv00);
 244
 245             /* Compute parameters for interactions between i and j atoms */
 246             qq00             = _mm256_mul_pd(iq0,jq0);
 247             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 248                                             vdwioffsetptr0+vdwjidx0B,
 249                                             vdwioffsetptr0+vdwjidx0C,
 250                                             vdwioffsetptr0+vdwjidx0D,
 251                                             &c6_00,&c12_00);
 252
 253             /* Calculate table index by multiplying r with table scale and truncate to integer */
 254             rt               = _mm256_mul_pd(r00,vftabscale);
 255             vfitab           = _mm256_cvttpd_epi32(rt);
 256             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 257             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 258
 259             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 260             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 261             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 262             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 263             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 264             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 265             Heps             = _mm256_mul_pd(vfeps,H);
 266             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 267             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 268             velec            = _mm256_mul_pd(qq00,VV);
 269             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 270             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 271
 272             /* CUBIC SPLINE TABLE DISPERSION */
 273             vfitab           = _mm_add_epi32(vfitab,ifour);
 274             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 275             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 276             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 277             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 278             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 279             Heps             = _mm256_mul_pd(vfeps,H);
 280             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 281             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 282             vvdw6            = _mm256_mul_pd(c6_00,VV);
 283             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 284             fvdw6            = _mm256_mul_pd(c6_00,FF);
 285
 286             /* CUBIC SPLINE TABLE REPULSION */
 287             vfitab           = _mm_add_epi32(vfitab,ifour);
 288             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 289             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 290             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 291             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 292             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 293             Heps             = _mm256_mul_pd(vfeps,H);
 294             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 295             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 296             vvdw12           = _mm256_mul_pd(c12_00,VV);
 297             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 298             fvdw12           = _mm256_mul_pd(c12_00,FF);
 299             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 300             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velecsum         = _mm256_add_pd(velecsum,velec);
 304             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 305
 306             fscal            = _mm256_add_pd(felec,fvdw);
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm256_mul_pd(fscal,dx00);
 310             ty               = _mm256_mul_pd(fscal,dy00);
 311             tz               = _mm256_mul_pd(fscal,dz00);
 312
 313             /* Update vectorial force */
 314             fix0             = _mm256_add_pd(fix0,tx);
 315             fiy0             = _mm256_add_pd(fiy0,ty);
 316             fiz0             = _mm256_add_pd(fiz0,tz);
 317
 318             fjx0             = _mm256_add_pd(fjx0,tx);
 319             fjy0             = _mm256_add_pd(fjy0,ty);
 320             fjz0             = _mm256_add_pd(fjz0,tz);
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             r10              = _mm256_mul_pd(rsq10,rinv10);
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq10             = _mm256_mul_pd(iq1,jq0);
 330
 331             /* Calculate table index by multiplying r with table scale and truncate to integer */
 332             rt               = _mm256_mul_pd(r10,vftabscale);
 333             vfitab           = _mm256_cvttpd_epi32(rt);
 334             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 335             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 336
 337             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 338             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 339             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 340             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 341             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 342             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 343             Heps             = _mm256_mul_pd(vfeps,H);
 344             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 345             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 346             velec            = _mm256_mul_pd(qq10,VV);
 347             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 348             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 349
 350             /* Update potential sum for this i atom from the interaction with this j atom. */
 351             velecsum         = _mm256_add_pd(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             /* Calculate temporary vectorial force */
 356             tx               = _mm256_mul_pd(fscal,dx10);
 357             ty               = _mm256_mul_pd(fscal,dy10);
 358             tz               = _mm256_mul_pd(fscal,dz10);
 359
 360             /* Update vectorial force */
 361             fix1             = _mm256_add_pd(fix1,tx);
 362             fiy1             = _mm256_add_pd(fiy1,ty);
 363             fiz1             = _mm256_add_pd(fiz1,tz);
 364
 365             fjx0             = _mm256_add_pd(fjx0,tx);
 366             fjy0             = _mm256_add_pd(fjy0,ty);
 367             fjz0             = _mm256_add_pd(fjz0,tz);
 368
 369             /**************************
 370              * CALCULATE INTERACTIONS *
 371              **************************/
 372
 373             r20              = _mm256_mul_pd(rsq20,rinv20);
 374
 375             /* Compute parameters for interactions between i and j atoms */
 376             qq20             = _mm256_mul_pd(iq2,jq0);
 377
 378             /* Calculate table index by multiplying r with table scale and truncate to integer */
 379             rt               = _mm256_mul_pd(r20,vftabscale);
 380             vfitab           = _mm256_cvttpd_epi32(rt);
 381             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 382             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 383
 384             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 385             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 386             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 387             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 388             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 389             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 390             Heps             = _mm256_mul_pd(vfeps,H);
 391             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 392             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 393             velec            = _mm256_mul_pd(qq20,VV);
 394             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 395             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 396
 397             /* Update potential sum for this i atom from the interaction with this j atom. */
 398             velecsum         = _mm256_add_pd(velecsum,velec);
 399
 400             fscal            = felec;
 401
 402             /* Calculate temporary vectorial force */
 403             tx               = _mm256_mul_pd(fscal,dx20);
 404             ty               = _mm256_mul_pd(fscal,dy20);
 405             tz               = _mm256_mul_pd(fscal,dz20);
 406
 407             /* Update vectorial force */
 408             fix2             = _mm256_add_pd(fix2,tx);
 409             fiy2             = _mm256_add_pd(fiy2,ty);
 410             fiz2             = _mm256_add_pd(fiz2,tz);
 411
 412             fjx0             = _mm256_add_pd(fjx0,tx);
 413             fjy0             = _mm256_add_pd(fjy0,ty);
 414             fjz0             = _mm256_add_pd(fjz0,tz);
 415
 416             fjptrA             = f+j_coord_offsetA;
 417             fjptrB             = f+j_coord_offsetB;
 418             fjptrC             = f+j_coord_offsetC;
 419             fjptrD             = f+j_coord_offsetD;
 420
 421             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 422
 423             /* Inner loop uses 162 flops */
 424         }
 425
 426         if(jidx<j_index_end)
 427         {
 428
 429             /* Get j neighbor index, and coordinate index */
 430             jnrlistA         = jjnr[jidx];
 431             jnrlistB         = jjnr[jidx+1];
 432             jnrlistC         = jjnr[jidx+2];
 433             jnrlistD         = jjnr[jidx+3];
 434             /* Sign of each element will be negative for non-real atoms.
 435              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 436              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 437              */
 438             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 439
 440             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 441             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 442             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 443
 444             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 445             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 446             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 447             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 448             j_coord_offsetA  = DIM*jnrA;
 449             j_coord_offsetB  = DIM*jnrB;
 450             j_coord_offsetC  = DIM*jnrC;
 451             j_coord_offsetD  = DIM*jnrD;
 452
 453             /* load j atom coordinates */
 454             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 455                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 456                                                  &jx0,&jy0,&jz0);
 457
 458             /* Calculate displacement vector */
 459             dx00             = _mm256_sub_pd(ix0,jx0);
 460             dy00             = _mm256_sub_pd(iy0,jy0);
 461             dz00             = _mm256_sub_pd(iz0,jz0);
 462             dx10             = _mm256_sub_pd(ix1,jx0);
 463             dy10             = _mm256_sub_pd(iy1,jy0);
 464             dz10             = _mm256_sub_pd(iz1,jz0);
 465             dx20             = _mm256_sub_pd(ix2,jx0);
 466             dy20             = _mm256_sub_pd(iy2,jy0);
 467             dz20             = _mm256_sub_pd(iz2,jz0);
 468
 469             /* Calculate squared distance and things based on it */
 470             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 471             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 472             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 473
 474             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 475             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 476             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 477
 478             /* Load parameters for j particles */
 479             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 480                                                                  charge+jnrC+0,charge+jnrD+0);
 481             vdwjidx0A        = 2*vdwtype[jnrA+0];
 482             vdwjidx0B        = 2*vdwtype[jnrB+0];
 483             vdwjidx0C        = 2*vdwtype[jnrC+0];
 484             vdwjidx0D        = 2*vdwtype[jnrD+0];
 485
 486             fjx0             = _mm256_setzero_pd();
 487             fjy0             = _mm256_setzero_pd();
 488             fjz0             = _mm256_setzero_pd();
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             r00              = _mm256_mul_pd(rsq00,rinv00);
 495             r00              = _mm256_andnot_pd(dummy_mask,r00);
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq00             = _mm256_mul_pd(iq0,jq0);
 499             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 500                                             vdwioffsetptr0+vdwjidx0B,
 501                                             vdwioffsetptr0+vdwjidx0C,
 502                                             vdwioffsetptr0+vdwjidx0D,
 503                                             &c6_00,&c12_00);
 504
 505             /* Calculate table index by multiplying r with table scale and truncate to integer */
 506             rt               = _mm256_mul_pd(r00,vftabscale);
 507             vfitab           = _mm256_cvttpd_epi32(rt);
 508             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 509             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 510
 511             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 512             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 513             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 514             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 515             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 516             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 517             Heps             = _mm256_mul_pd(vfeps,H);
 518             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 519             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 520             velec            = _mm256_mul_pd(qq00,VV);
 521             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 522             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 523
 524             /* CUBIC SPLINE TABLE DISPERSION */
 525             vfitab           = _mm_add_epi32(vfitab,ifour);
 526             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 527             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 528             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 529             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 530             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 531             Heps             = _mm256_mul_pd(vfeps,H);
 532             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 533             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 534             vvdw6            = _mm256_mul_pd(c6_00,VV);
 535             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 536             fvdw6            = _mm256_mul_pd(c6_00,FF);
 537
 538             /* CUBIC SPLINE TABLE REPULSION */
 539             vfitab           = _mm_add_epi32(vfitab,ifour);
 540             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 541             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 542             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 543             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 544             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 545             Heps             = _mm256_mul_pd(vfeps,H);
 546             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 547             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 548             vvdw12           = _mm256_mul_pd(c12_00,VV);
 549             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 550             fvdw12           = _mm256_mul_pd(c12_00,FF);
 551             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 552             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velec            = _mm256_andnot_pd(dummy_mask,velec);
 556             velecsum         = _mm256_add_pd(velecsum,velec);
 557             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 558             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 559
 560             fscal            = _mm256_add_pd(felec,fvdw);
 561
 562             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm256_mul_pd(fscal,dx00);
 566             ty               = _mm256_mul_pd(fscal,dy00);
 567             tz               = _mm256_mul_pd(fscal,dz00);
 568
 569             /* Update vectorial force */
 570             fix0             = _mm256_add_pd(fix0,tx);
 571             fiy0             = _mm256_add_pd(fiy0,ty);
 572             fiz0             = _mm256_add_pd(fiz0,tz);
 573
 574             fjx0             = _mm256_add_pd(fjx0,tx);
 575             fjy0             = _mm256_add_pd(fjy0,ty);
 576             fjz0             = _mm256_add_pd(fjz0,tz);
 577
 578             /**************************
 579              * CALCULATE INTERACTIONS *
 580              **************************/
 581
 582             r10              = _mm256_mul_pd(rsq10,rinv10);
 583             r10              = _mm256_andnot_pd(dummy_mask,r10);
 584
 585             /* Compute parameters for interactions between i and j atoms */
 586             qq10             = _mm256_mul_pd(iq1,jq0);
 587
 588             /* Calculate table index by multiplying r with table scale and truncate to integer */
 589             rt               = _mm256_mul_pd(r10,vftabscale);
 590             vfitab           = _mm256_cvttpd_epi32(rt);
 591             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 592             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 593
 594             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 595             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 596             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 597             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 598             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 599             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 600             Heps             = _mm256_mul_pd(vfeps,H);
 601             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 602             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 603             velec            = _mm256_mul_pd(qq10,VV);
 604             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 605             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 606
 607             /* Update potential sum for this i atom from the interaction with this j atom. */
 608             velec            = _mm256_andnot_pd(dummy_mask,velec);
 609             velecsum         = _mm256_add_pd(velecsum,velec);
 610
 611             fscal            = felec;
 612
 613             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 614
 615             /* Calculate temporary vectorial force */
 616             tx               = _mm256_mul_pd(fscal,dx10);
 617             ty               = _mm256_mul_pd(fscal,dy10);
 618             tz               = _mm256_mul_pd(fscal,dz10);
 619
 620             /* Update vectorial force */
 621             fix1             = _mm256_add_pd(fix1,tx);
 622             fiy1             = _mm256_add_pd(fiy1,ty);
 623             fiz1             = _mm256_add_pd(fiz1,tz);
 624
 625             fjx0             = _mm256_add_pd(fjx0,tx);
 626             fjy0             = _mm256_add_pd(fjy0,ty);
 627             fjz0             = _mm256_add_pd(fjz0,tz);
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             r20              = _mm256_mul_pd(rsq20,rinv20);
 634             r20              = _mm256_andnot_pd(dummy_mask,r20);
 635
 636             /* Compute parameters for interactions between i and j atoms */
 637             qq20             = _mm256_mul_pd(iq2,jq0);
 638
 639             /* Calculate table index by multiplying r with table scale and truncate to integer */
 640             rt               = _mm256_mul_pd(r20,vftabscale);
 641             vfitab           = _mm256_cvttpd_epi32(rt);
 642             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 643             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 644
 645             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 646             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 647             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 648             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 649             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 650             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 651             Heps             = _mm256_mul_pd(vfeps,H);
 652             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 653             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 654             velec            = _mm256_mul_pd(qq20,VV);
 655             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 656             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
 657
 658             /* Update potential sum for this i atom from the interaction with this j atom. */
 659             velec            = _mm256_andnot_pd(dummy_mask,velec);
 660             velecsum         = _mm256_add_pd(velecsum,velec);
 661
 662             fscal            = felec;
 663
 664             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 665
 666             /* Calculate temporary vectorial force */
 667             tx               = _mm256_mul_pd(fscal,dx20);
 668             ty               = _mm256_mul_pd(fscal,dy20);
 669             tz               = _mm256_mul_pd(fscal,dz20);
 670
 671             /* Update vectorial force */
 672             fix2             = _mm256_add_pd(fix2,tx);
 673             fiy2             = _mm256_add_pd(fiy2,ty);
 674             fiz2             = _mm256_add_pd(fiz2,tz);
 675
 676             fjx0             = _mm256_add_pd(fjx0,tx);
 677             fjy0             = _mm256_add_pd(fjy0,ty);
 678             fjz0             = _mm256_add_pd(fjz0,tz);
 679
 680             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 681             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 682             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 683             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 684
 685             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 686
 687             /* Inner loop uses 165 flops */
 688         }
 689
 690         /* End of innermost loop */
 691
 692         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 693                                                  f+i_coord_offset,fshift+i_shift_offset);
 694
 695         ggid                        = gid[iidx];
 696         /* Update potential energies */
 697         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 698         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 699
 700         /* Increment number of inner iterations */
 701         inneriter                  += j_index_end - j_index_start;
 702
 703         /* Outer loop uses 20 flops */
 704     }
 705
 706     /* Increment number of outer iterations */
 707     outeriter        += nri;
 708
 709     /* Update outer/inner flops */
 710
 711     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*165);
 712 }
 713 /*
 714  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_double
 715  * Electrostatics interaction: CubicSplineTable
 716  * VdW interaction:            CubicSplineTable
 717  * Geometry:                   Water3-Particle
 718  * Calculate force/pot:        Force
 719  */
 720 void
 721 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_avx_256_double
 722                     (t_nblist                    * gmx_restrict       nlist,
 723                      rvec                        * gmx_restrict          xx,
 724                      rvec                        * gmx_restrict          ff,
 725                      t_forcerec                  * gmx_restrict          fr,
 726                      t_mdatoms                   * gmx_restrict     mdatoms,
 727                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 728                      t_nrnb                      * gmx_restrict        nrnb)
 729 {
 730     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 731      * just 0 for non-waters.
 732      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 733      * jnr indices corresponding to data put in the four positions in the SIMD register.
 734      */
 735     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 736     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 737     int              jnrA,jnrB,jnrC,jnrD;
 738     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 739     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 740     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 741     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 742     real             rcutoff_scalar;
 743     real             *shiftvec,*fshift,*x,*f;
 744     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 745     real             scratch[4*DIM];
 746     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 747     real *           vdwioffsetptr0;
 748     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 749     real *           vdwioffsetptr1;
 750     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 751     real *           vdwioffsetptr2;
 752     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 753     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 754     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 755     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 756     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 757     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 758     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 759     real             *charge;
 760     int              nvdwtype;
 761     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 762     int              *vdwtype;
 763     real             *vdwparam;
 764     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 765     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 766     __m128i          vfitab;
 767     __m128i          ifour       = _mm_set1_epi32(4);
 768     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 769     real             *vftab;
 770     __m256d          dummy_mask,cutoff_mask;
 771     __m128           tmpmask0,tmpmask1;
 772     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 773     __m256d          one     = _mm256_set1_pd(1.0);
 774     __m256d          two     = _mm256_set1_pd(2.0);
 775     x                = xx[0];
 776     f                = ff[0];
 777
 778     nri              = nlist->nri;
 779     iinr             = nlist->iinr;
 780     jindex           = nlist->jindex;
 781     jjnr             = nlist->jjnr;
 782     shiftidx         = nlist->shift;
 783     gid              = nlist->gid;
 784     shiftvec         = fr->shift_vec[0];
 785     fshift           = fr->fshift[0];
 786     facel            = _mm256_set1_pd(fr->epsfac);
 787     charge           = mdatoms->chargeA;
 788     nvdwtype         = fr->ntype;
 789     vdwparam         = fr->nbfp;
 790     vdwtype          = mdatoms->typeA;
 791
 792     vftab            = kernel_data->table_elec_vdw->data;
 793     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 794
 795     /* Setup water-specific parameters */
 796     inr              = nlist->iinr[0];
 797     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 798     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 799     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 800     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 801
 802     /* Avoid stupid compiler warnings */
 803     jnrA = jnrB = jnrC = jnrD = 0;
 804     j_coord_offsetA = 0;
 805     j_coord_offsetB = 0;
 806     j_coord_offsetC = 0;
 807     j_coord_offsetD = 0;
 808
 809     outeriter        = 0;
 810     inneriter        = 0;
 811
 812     for(iidx=0;iidx<4*DIM;iidx++)
 813     {
 814         scratch[iidx] = 0.0;
 815     }
 816
 817     /* Start outer loop over neighborlists */
 818     for(iidx=0; iidx<nri; iidx++)
 819     {
 820         /* Load shift vector for this list */
 821         i_shift_offset   = DIM*shiftidx[iidx];
 822
 823         /* Load limits for loop over neighbors */
 824         j_index_start    = jindex[iidx];
 825         j_index_end      = jindex[iidx+1];
 826
 827         /* Get outer coordinate index */
 828         inr              = iinr[iidx];
 829         i_coord_offset   = DIM*inr;
 830
 831         /* Load i particle coords and add shift vector */
 832         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 833                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 834
 835         fix0             = _mm256_setzero_pd();
 836         fiy0             = _mm256_setzero_pd();
 837         fiz0             = _mm256_setzero_pd();
 838         fix1             = _mm256_setzero_pd();
 839         fiy1             = _mm256_setzero_pd();
 840         fiz1             = _mm256_setzero_pd();
 841         fix2             = _mm256_setzero_pd();
 842         fiy2             = _mm256_setzero_pd();
 843         fiz2             = _mm256_setzero_pd();
 844
 845         /* Start inner kernel loop */
 846         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 847         {
 848
 849             /* Get j neighbor index, and coordinate index */
 850             jnrA             = jjnr[jidx];
 851             jnrB             = jjnr[jidx+1];
 852             jnrC             = jjnr[jidx+2];
 853             jnrD             = jjnr[jidx+3];
 854             j_coord_offsetA  = DIM*jnrA;
 855             j_coord_offsetB  = DIM*jnrB;
 856             j_coord_offsetC  = DIM*jnrC;
 857             j_coord_offsetD  = DIM*jnrD;
 858
 859             /* load j atom coordinates */
 860             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 861                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 862                                                  &jx0,&jy0,&jz0);
 863
 864             /* Calculate displacement vector */
 865             dx00             = _mm256_sub_pd(ix0,jx0);
 866             dy00             = _mm256_sub_pd(iy0,jy0);
 867             dz00             = _mm256_sub_pd(iz0,jz0);
 868             dx10             = _mm256_sub_pd(ix1,jx0);
 869             dy10             = _mm256_sub_pd(iy1,jy0);
 870             dz10             = _mm256_sub_pd(iz1,jz0);
 871             dx20             = _mm256_sub_pd(ix2,jx0);
 872             dy20             = _mm256_sub_pd(iy2,jy0);
 873             dz20             = _mm256_sub_pd(iz2,jz0);
 874
 875             /* Calculate squared distance and things based on it */
 876             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 877             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 878             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 879
 880             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 881             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 882             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 883
 884             /* Load parameters for j particles */
 885             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 886                                                                  charge+jnrC+0,charge+jnrD+0);
 887             vdwjidx0A        = 2*vdwtype[jnrA+0];
 888             vdwjidx0B        = 2*vdwtype[jnrB+0];
 889             vdwjidx0C        = 2*vdwtype[jnrC+0];
 890             vdwjidx0D        = 2*vdwtype[jnrD+0];
 891
 892             fjx0             = _mm256_setzero_pd();
 893             fjy0             = _mm256_setzero_pd();
 894             fjz0             = _mm256_setzero_pd();
 895
 896             /**************************
 897              * CALCULATE INTERACTIONS *
 898              **************************/
 899
 900             r00              = _mm256_mul_pd(rsq00,rinv00);
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             qq00             = _mm256_mul_pd(iq0,jq0);
 904             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 905                                             vdwioffsetptr0+vdwjidx0B,
 906                                             vdwioffsetptr0+vdwjidx0C,
 907                                             vdwioffsetptr0+vdwjidx0D,
 908                                             &c6_00,&c12_00);
 909
 910             /* Calculate table index by multiplying r with table scale and truncate to integer */
 911             rt               = _mm256_mul_pd(r00,vftabscale);
 912             vfitab           = _mm256_cvttpd_epi32(rt);
 913             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 914             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 915
 916             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 917             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 918             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 919             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 920             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 921             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 922             Heps             = _mm256_mul_pd(vfeps,H);
 923             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 924             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 925             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
 926
 927             /* CUBIC SPLINE TABLE DISPERSION */
 928             vfitab           = _mm_add_epi32(vfitab,ifour);
 929             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 930             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 931             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 932             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 933             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 934             Heps             = _mm256_mul_pd(vfeps,H);
 935             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 936             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 937             fvdw6            = _mm256_mul_pd(c6_00,FF);
 938
 939             /* CUBIC SPLINE TABLE REPULSION */
 940             vfitab           = _mm_add_epi32(vfitab,ifour);
 941             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 942             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 943             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 944             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 945             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 946             Heps             = _mm256_mul_pd(vfeps,H);
 947             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 948             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 949             fvdw12           = _mm256_mul_pd(c12_00,FF);
 950             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 951
 952             fscal            = _mm256_add_pd(felec,fvdw);
 953
 954             /* Calculate temporary vectorial force */
 955             tx               = _mm256_mul_pd(fscal,dx00);
 956             ty               = _mm256_mul_pd(fscal,dy00);
 957             tz               = _mm256_mul_pd(fscal,dz00);
 958
 959             /* Update vectorial force */
 960             fix0             = _mm256_add_pd(fix0,tx);
 961             fiy0             = _mm256_add_pd(fiy0,ty);
 962             fiz0             = _mm256_add_pd(fiz0,tz);
 963
 964             fjx0             = _mm256_add_pd(fjx0,tx);
 965             fjy0             = _mm256_add_pd(fjy0,ty);
 966             fjz0             = _mm256_add_pd(fjz0,tz);
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             r10              = _mm256_mul_pd(rsq10,rinv10);
 973
 974             /* Compute parameters for interactions between i and j atoms */
 975             qq10             = _mm256_mul_pd(iq1,jq0);
 976
 977             /* Calculate table index by multiplying r with table scale and truncate to integer */
 978             rt               = _mm256_mul_pd(r10,vftabscale);
 979             vfitab           = _mm256_cvttpd_epi32(rt);
 980             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 981             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 982
 983             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 984             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 985             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 986             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 987             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 988             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 989             Heps             = _mm256_mul_pd(vfeps,H);
 990             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 991             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 992             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
 993
 994             fscal            = felec;
 995
 996             /* Calculate temporary vectorial force */
 997             tx               = _mm256_mul_pd(fscal,dx10);
 998             ty               = _mm256_mul_pd(fscal,dy10);
 999             tz               = _mm256_mul_pd(fscal,dz10);
1000
1001             /* Update vectorial force */
1002             fix1             = _mm256_add_pd(fix1,tx);
1003             fiy1             = _mm256_add_pd(fiy1,ty);
1004             fiz1             = _mm256_add_pd(fiz1,tz);
1005
1006             fjx0             = _mm256_add_pd(fjx0,tx);
1007             fjy0             = _mm256_add_pd(fjy0,ty);
1008             fjz0             = _mm256_add_pd(fjz0,tz);
1009
1010             /**************************
1011              * CALCULATE INTERACTIONS *
1012              **************************/
1013
1014             r20              = _mm256_mul_pd(rsq20,rinv20);
1015
1016             /* Compute parameters for interactions between i and j atoms */
1017             qq20             = _mm256_mul_pd(iq2,jq0);
1018
1019             /* Calculate table index by multiplying r with table scale and truncate to integer */
1020             rt               = _mm256_mul_pd(r20,vftabscale);
1021             vfitab           = _mm256_cvttpd_epi32(rt);
1022             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1023             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1024
1025             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1026             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1027             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1028             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1029             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1030             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1031             Heps             = _mm256_mul_pd(vfeps,H);
1032             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1033             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1034             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1035
1036             fscal            = felec;
1037
1038             /* Calculate temporary vectorial force */
1039             tx               = _mm256_mul_pd(fscal,dx20);
1040             ty               = _mm256_mul_pd(fscal,dy20);
1041             tz               = _mm256_mul_pd(fscal,dz20);
1042
1043             /* Update vectorial force */
1044             fix2             = _mm256_add_pd(fix2,tx);
1045             fiy2             = _mm256_add_pd(fiy2,ty);
1046             fiz2             = _mm256_add_pd(fiz2,tz);
1047
1048             fjx0             = _mm256_add_pd(fjx0,tx);
1049             fjy0             = _mm256_add_pd(fjy0,ty);
1050             fjz0             = _mm256_add_pd(fjz0,tz);
1051
1052             fjptrA             = f+j_coord_offsetA;
1053             fjptrB             = f+j_coord_offsetB;
1054             fjptrC             = f+j_coord_offsetC;
1055             fjptrD             = f+j_coord_offsetD;
1056
1057             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1058
1059             /* Inner loop uses 142 flops */
1060         }
1061
1062         if(jidx<j_index_end)
1063         {
1064
1065             /* Get j neighbor index, and coordinate index */
1066             jnrlistA         = jjnr[jidx];
1067             jnrlistB         = jjnr[jidx+1];
1068             jnrlistC         = jjnr[jidx+2];
1069             jnrlistD         = jjnr[jidx+3];
1070             /* Sign of each element will be negative for non-real atoms.
1071              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1072              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1073              */
1074             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1075
1076             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1077             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1078             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1079
1080             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1081             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1082             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1083             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1084             j_coord_offsetA  = DIM*jnrA;
1085             j_coord_offsetB  = DIM*jnrB;
1086             j_coord_offsetC  = DIM*jnrC;
1087             j_coord_offsetD  = DIM*jnrD;
1088
1089             /* load j atom coordinates */
1090             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1091                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1092                                                  &jx0,&jy0,&jz0);
1093
1094             /* Calculate displacement vector */
1095             dx00             = _mm256_sub_pd(ix0,jx0);
1096             dy00             = _mm256_sub_pd(iy0,jy0);
1097             dz00             = _mm256_sub_pd(iz0,jz0);
1098             dx10             = _mm256_sub_pd(ix1,jx0);
1099             dy10             = _mm256_sub_pd(iy1,jy0);
1100             dz10             = _mm256_sub_pd(iz1,jz0);
1101             dx20             = _mm256_sub_pd(ix2,jx0);
1102             dy20             = _mm256_sub_pd(iy2,jy0);
1103             dz20             = _mm256_sub_pd(iz2,jz0);
1104
1105             /* Calculate squared distance and things based on it */
1106             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1107             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1108             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1109
1110             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1111             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1112             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1113
1114             /* Load parameters for j particles */
1115             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1116                                                                  charge+jnrC+0,charge+jnrD+0);
1117             vdwjidx0A        = 2*vdwtype[jnrA+0];
1118             vdwjidx0B        = 2*vdwtype[jnrB+0];
1119             vdwjidx0C        = 2*vdwtype[jnrC+0];
1120             vdwjidx0D        = 2*vdwtype[jnrD+0];
1121
1122             fjx0             = _mm256_setzero_pd();
1123             fjy0             = _mm256_setzero_pd();
1124             fjz0             = _mm256_setzero_pd();
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             r00              = _mm256_mul_pd(rsq00,rinv00);
1131             r00              = _mm256_andnot_pd(dummy_mask,r00);
1132
1133             /* Compute parameters for interactions between i and j atoms */
1134             qq00             = _mm256_mul_pd(iq0,jq0);
1135             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1136                                             vdwioffsetptr0+vdwjidx0B,
1137                                             vdwioffsetptr0+vdwjidx0C,
1138                                             vdwioffsetptr0+vdwjidx0D,
1139                                             &c6_00,&c12_00);
1140
1141             /* Calculate table index by multiplying r with table scale and truncate to integer */
1142             rt               = _mm256_mul_pd(r00,vftabscale);
1143             vfitab           = _mm256_cvttpd_epi32(rt);
1144             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1145             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1146
1147             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1148             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1149             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1150             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1151             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1152             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1153             Heps             = _mm256_mul_pd(vfeps,H);
1154             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1155             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1156             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq00,FF),_mm256_mul_pd(vftabscale,rinv00)));
1157
1158             /* CUBIC SPLINE TABLE DISPERSION */
1159             vfitab           = _mm_add_epi32(vfitab,ifour);
1160             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1161             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1162             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1163             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1164             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1165             Heps             = _mm256_mul_pd(vfeps,H);
1166             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1167             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1168             fvdw6            = _mm256_mul_pd(c6_00,FF);
1169
1170             /* CUBIC SPLINE TABLE REPULSION */
1171             vfitab           = _mm_add_epi32(vfitab,ifour);
1172             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1173             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1174             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1175             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1176             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1177             Heps             = _mm256_mul_pd(vfeps,H);
1178             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1179             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1180             fvdw12           = _mm256_mul_pd(c12_00,FF);
1181             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1182
1183             fscal            = _mm256_add_pd(felec,fvdw);
1184
1185             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1186
1187             /* Calculate temporary vectorial force */
1188             tx               = _mm256_mul_pd(fscal,dx00);
1189             ty               = _mm256_mul_pd(fscal,dy00);
1190             tz               = _mm256_mul_pd(fscal,dz00);
1191
1192             /* Update vectorial force */
1193             fix0             = _mm256_add_pd(fix0,tx);
1194             fiy0             = _mm256_add_pd(fiy0,ty);
1195             fiz0             = _mm256_add_pd(fiz0,tz);
1196
1197             fjx0             = _mm256_add_pd(fjx0,tx);
1198             fjy0             = _mm256_add_pd(fjy0,ty);
1199             fjz0             = _mm256_add_pd(fjz0,tz);
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             r10              = _mm256_mul_pd(rsq10,rinv10);
1206             r10              = _mm256_andnot_pd(dummy_mask,r10);
1207
1208             /* Compute parameters for interactions between i and j atoms */
1209             qq10             = _mm256_mul_pd(iq1,jq0);
1210
1211             /* Calculate table index by multiplying r with table scale and truncate to integer */
1212             rt               = _mm256_mul_pd(r10,vftabscale);
1213             vfitab           = _mm256_cvttpd_epi32(rt);
1214             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1215             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1216
1217             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1218             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1219             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1220             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1221             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1222             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1223             Heps             = _mm256_mul_pd(vfeps,H);
1224             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1225             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1226             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq10,FF),_mm256_mul_pd(vftabscale,rinv10)));
1227
1228             fscal            = felec;
1229
1230             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1231
1232             /* Calculate temporary vectorial force */
1233             tx               = _mm256_mul_pd(fscal,dx10);
1234             ty               = _mm256_mul_pd(fscal,dy10);
1235             tz               = _mm256_mul_pd(fscal,dz10);
1236
1237             /* Update vectorial force */
1238             fix1             = _mm256_add_pd(fix1,tx);
1239             fiy1             = _mm256_add_pd(fiy1,ty);
1240             fiz1             = _mm256_add_pd(fiz1,tz);
1241
1242             fjx0             = _mm256_add_pd(fjx0,tx);
1243             fjy0             = _mm256_add_pd(fjy0,ty);
1244             fjz0             = _mm256_add_pd(fjz0,tz);
1245
1246             /**************************
1247              * CALCULATE INTERACTIONS *
1248              **************************/
1249
1250             r20              = _mm256_mul_pd(rsq20,rinv20);
1251             r20              = _mm256_andnot_pd(dummy_mask,r20);
1252
1253             /* Compute parameters for interactions between i and j atoms */
1254             qq20             = _mm256_mul_pd(iq2,jq0);
1255
1256             /* Calculate table index by multiplying r with table scale and truncate to integer */
1257             rt               = _mm256_mul_pd(r20,vftabscale);
1258             vfitab           = _mm256_cvttpd_epi32(rt);
1259             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1260             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1261
1262             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1263             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1264             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1265             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1266             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1267             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1268             Heps             = _mm256_mul_pd(vfeps,H);
1269             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1270             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1271             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq20,FF),_mm256_mul_pd(vftabscale,rinv20)));
1272
1273             fscal            = felec;
1274
1275             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1276
1277             /* Calculate temporary vectorial force */
1278             tx               = _mm256_mul_pd(fscal,dx20);
1279             ty               = _mm256_mul_pd(fscal,dy20);
1280             tz               = _mm256_mul_pd(fscal,dz20);
1281
1282             /* Update vectorial force */
1283             fix2             = _mm256_add_pd(fix2,tx);
1284             fiy2             = _mm256_add_pd(fiy2,ty);
1285             fiz2             = _mm256_add_pd(fiz2,tz);
1286
1287             fjx0             = _mm256_add_pd(fjx0,tx);
1288             fjy0             = _mm256_add_pd(fjy0,ty);
1289             fjz0             = _mm256_add_pd(fjz0,tz);
1290
1291             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1292             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1293             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1294             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1295
1296             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1297
1298             /* Inner loop uses 145 flops */
1299         }
1300
1301         /* End of innermost loop */
1302
1303         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1304                                                  f+i_coord_offset,fshift+i_shift_offset);
1305
1306         /* Increment number of inner iterations */
1307         inneriter                  += j_index_end - j_index_start;
1308
1309         /* Outer loop uses 18 flops */
1310     }
1311
1312     /* Increment number of outer iterations */
1313     outeriter        += nri;
1314
1315     /* Update outer/inner flops */
1316
1317     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*145);
1318 }