src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m128i          vfitab;
 107     __m128i          ifour       = _mm_set1_epi32(4);
 108     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 109     real             *vftab;
 110     __m128i          ewitab;
 111     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 112     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 113     real             *ewtab;
 114     __m256d          dummy_mask,cutoff_mask;
 115     __m128           tmpmask0,tmpmask1;
 116     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 117     __m256d          one     = _mm256_set1_pd(1.0);
 118     __m256d          two     = _mm256_set1_pd(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm256_set1_pd(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135
 136     vftab            = kernel_data->table_vdw->data;
 137     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 138
 139     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 140     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 141     beta2            = _mm256_mul_pd(beta,beta);
 142     beta3            = _mm256_mul_pd(beta,beta2);
 143
 144     ewtab            = fr->ic->tabq_coul_FDV0;
 145     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 146     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 147
 148     /* Setup water-specific parameters */
 149     inr              = nlist->iinr[0];
 150     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 151     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 152     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 153     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 154
 155     /* Avoid stupid compiler warnings */
 156     jnrA = jnrB = jnrC = jnrD = 0;
 157     j_coord_offsetA = 0;
 158     j_coord_offsetB = 0;
 159     j_coord_offsetC = 0;
 160     j_coord_offsetD = 0;
 161
 162     outeriter        = 0;
 163     inneriter        = 0;
 164
 165     for(iidx=0;iidx<4*DIM;iidx++)
 166     {
 167         scratch[iidx] = 0.0;
 168     }
 169
 170     /* Start outer loop over neighborlists */
 171     for(iidx=0; iidx<nri; iidx++)
 172     {
 173         /* Load shift vector for this list */
 174         i_shift_offset   = DIM*shiftidx[iidx];
 175
 176         /* Load limits for loop over neighbors */
 177         j_index_start    = jindex[iidx];
 178         j_index_end      = jindex[iidx+1];
 179
 180         /* Get outer coordinate index */
 181         inr              = iinr[iidx];
 182         i_coord_offset   = DIM*inr;
 183
 184         /* Load i particle coords and add shift vector */
 185         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 186                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 187
 188         fix0             = _mm256_setzero_pd();
 189         fiy0             = _mm256_setzero_pd();
 190         fiz0             = _mm256_setzero_pd();
 191         fix1             = _mm256_setzero_pd();
 192         fiy1             = _mm256_setzero_pd();
 193         fiz1             = _mm256_setzero_pd();
 194         fix2             = _mm256_setzero_pd();
 195         fiy2             = _mm256_setzero_pd();
 196         fiz2             = _mm256_setzero_pd();
 197         fix3             = _mm256_setzero_pd();
 198         fiy3             = _mm256_setzero_pd();
 199         fiz3             = _mm256_setzero_pd();
 200
 201         /* Reset potential sums */
 202         velecsum         = _mm256_setzero_pd();
 203         vvdwsum          = _mm256_setzero_pd();
 204
 205         /* Start inner kernel loop */
 206         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 207         {
 208
 209             /* Get j neighbor index, and coordinate index */
 210             jnrA             = jjnr[jidx];
 211             jnrB             = jjnr[jidx+1];
 212             jnrC             = jjnr[jidx+2];
 213             jnrD             = jjnr[jidx+3];
 214             j_coord_offsetA  = DIM*jnrA;
 215             j_coord_offsetB  = DIM*jnrB;
 216             j_coord_offsetC  = DIM*jnrC;
 217             j_coord_offsetD  = DIM*jnrD;
 218
 219             /* load j atom coordinates */
 220             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 221                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 222                                                  &jx0,&jy0,&jz0);
 223
 224             /* Calculate displacement vector */
 225             dx00             = _mm256_sub_pd(ix0,jx0);
 226             dy00             = _mm256_sub_pd(iy0,jy0);
 227             dz00             = _mm256_sub_pd(iz0,jz0);
 228             dx10             = _mm256_sub_pd(ix1,jx0);
 229             dy10             = _mm256_sub_pd(iy1,jy0);
 230             dz10             = _mm256_sub_pd(iz1,jz0);
 231             dx20             = _mm256_sub_pd(ix2,jx0);
 232             dy20             = _mm256_sub_pd(iy2,jy0);
 233             dz20             = _mm256_sub_pd(iz2,jz0);
 234             dx30             = _mm256_sub_pd(ix3,jx0);
 235             dy30             = _mm256_sub_pd(iy3,jy0);
 236             dz30             = _mm256_sub_pd(iz3,jz0);
 237
 238             /* Calculate squared distance and things based on it */
 239             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 240             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 241             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 242             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 243
 244             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 245             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 246             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 247             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 248
 249             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 250             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 251             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 252
 253             /* Load parameters for j particles */
 254             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 255                                                                  charge+jnrC+0,charge+jnrD+0);
 256             vdwjidx0A        = 2*vdwtype[jnrA+0];
 257             vdwjidx0B        = 2*vdwtype[jnrB+0];
 258             vdwjidx0C        = 2*vdwtype[jnrC+0];
 259             vdwjidx0D        = 2*vdwtype[jnrD+0];
 260
 261             fjx0             = _mm256_setzero_pd();
 262             fjy0             = _mm256_setzero_pd();
 263             fjz0             = _mm256_setzero_pd();
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             r00              = _mm256_mul_pd(rsq00,rinv00);
 270
 271             /* Compute parameters for interactions between i and j atoms */
 272             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 273                                             vdwioffsetptr0+vdwjidx0B,
 274                                             vdwioffsetptr0+vdwjidx0C,
 275                                             vdwioffsetptr0+vdwjidx0D,
 276                                             &c6_00,&c12_00);
 277
 278             /* Calculate table index by multiplying r with table scale and truncate to integer */
 279             rt               = _mm256_mul_pd(r00,vftabscale);
 280             vfitab           = _mm256_cvttpd_epi32(rt);
 281             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 282             vfitab           = _mm_slli_epi32(vfitab,3);
 283
 284             /* CUBIC SPLINE TABLE DISPERSION */
 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             vvdw6            = _mm256_mul_pd(c6_00,VV);
 294             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 295             fvdw6            = _mm256_mul_pd(c6_00,FF);
 296
 297             /* CUBIC SPLINE TABLE REPULSION */
 298             vfitab           = _mm_add_epi32(vfitab,ifour);
 299             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 300             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 301             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 302             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 303             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 304             Heps             = _mm256_mul_pd(vfeps,H);
 305             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 306             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 307             vvdw12           = _mm256_mul_pd(c12_00,VV);
 308             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 309             fvdw12           = _mm256_mul_pd(c12_00,FF);
 310             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 311             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 315
 316             fscal            = fvdw;
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm256_mul_pd(fscal,dx00);
 320             ty               = _mm256_mul_pd(fscal,dy00);
 321             tz               = _mm256_mul_pd(fscal,dz00);
 322
 323             /* Update vectorial force */
 324             fix0             = _mm256_add_pd(fix0,tx);
 325             fiy0             = _mm256_add_pd(fiy0,ty);
 326             fiz0             = _mm256_add_pd(fiz0,tz);
 327
 328             fjx0             = _mm256_add_pd(fjx0,tx);
 329             fjy0             = _mm256_add_pd(fjy0,ty);
 330             fjz0             = _mm256_add_pd(fjz0,tz);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             r10              = _mm256_mul_pd(rsq10,rinv10);
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq10             = _mm256_mul_pd(iq1,jq0);
 340
 341             /* EWALD ELECTROSTATICS */
 342
 343             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 344             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 345             ewitab           = _mm256_cvttpd_epi32(ewrt);
 346             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 347             ewitab           = _mm_slli_epi32(ewitab,2);
 348             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 349             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 350             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 351             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 352             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 353             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 354             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 355             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 356             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm256_add_pd(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm256_mul_pd(fscal,dx10);
 365             ty               = _mm256_mul_pd(fscal,dy10);
 366             tz               = _mm256_mul_pd(fscal,dz10);
 367
 368             /* Update vectorial force */
 369             fix1             = _mm256_add_pd(fix1,tx);
 370             fiy1             = _mm256_add_pd(fiy1,ty);
 371             fiz1             = _mm256_add_pd(fiz1,tz);
 372
 373             fjx0             = _mm256_add_pd(fjx0,tx);
 374             fjy0             = _mm256_add_pd(fjy0,ty);
 375             fjz0             = _mm256_add_pd(fjz0,tz);
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             r20              = _mm256_mul_pd(rsq20,rinv20);
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq20             = _mm256_mul_pd(iq2,jq0);
 385
 386             /* EWALD ELECTROSTATICS */
 387
 388             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 389             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 390             ewitab           = _mm256_cvttpd_epi32(ewrt);
 391             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 392             ewitab           = _mm_slli_epi32(ewitab,2);
 393             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 394             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 395             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 396             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 397             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 398             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 399             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 400             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 401             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velecsum         = _mm256_add_pd(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm256_mul_pd(fscal,dx20);
 410             ty               = _mm256_mul_pd(fscal,dy20);
 411             tz               = _mm256_mul_pd(fscal,dz20);
 412
 413             /* Update vectorial force */
 414             fix2             = _mm256_add_pd(fix2,tx);
 415             fiy2             = _mm256_add_pd(fiy2,ty);
 416             fiz2             = _mm256_add_pd(fiz2,tz);
 417
 418             fjx0             = _mm256_add_pd(fjx0,tx);
 419             fjy0             = _mm256_add_pd(fjy0,ty);
 420             fjz0             = _mm256_add_pd(fjz0,tz);
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             r30              = _mm256_mul_pd(rsq30,rinv30);
 427
 428             /* Compute parameters for interactions between i and j atoms */
 429             qq30             = _mm256_mul_pd(iq3,jq0);
 430
 431             /* EWALD ELECTROSTATICS */
 432
 433             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 434             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 435             ewitab           = _mm256_cvttpd_epi32(ewrt);
 436             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 437             ewitab           = _mm_slli_epi32(ewitab,2);
 438             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 439             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 440             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 441             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 442             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 443             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 444             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 445             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 446             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 447
 448             /* Update potential sum for this i atom from the interaction with this j atom. */
 449             velecsum         = _mm256_add_pd(velecsum,velec);
 450
 451             fscal            = felec;
 452
 453             /* Calculate temporary vectorial force */
 454             tx               = _mm256_mul_pd(fscal,dx30);
 455             ty               = _mm256_mul_pd(fscal,dy30);
 456             tz               = _mm256_mul_pd(fscal,dz30);
 457
 458             /* Update vectorial force */
 459             fix3             = _mm256_add_pd(fix3,tx);
 460             fiy3             = _mm256_add_pd(fiy3,ty);
 461             fiz3             = _mm256_add_pd(fiz3,tz);
 462
 463             fjx0             = _mm256_add_pd(fjx0,tx);
 464             fjy0             = _mm256_add_pd(fjy0,ty);
 465             fjz0             = _mm256_add_pd(fjz0,tz);
 466
 467             fjptrA             = f+j_coord_offsetA;
 468             fjptrB             = f+j_coord_offsetB;
 469             fjptrC             = f+j_coord_offsetC;
 470             fjptrD             = f+j_coord_offsetD;
 471
 472             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 473
 474             /* Inner loop uses 182 flops */
 475         }
 476
 477         if(jidx<j_index_end)
 478         {
 479
 480             /* Get j neighbor index, and coordinate index */
 481             jnrlistA         = jjnr[jidx];
 482             jnrlistB         = jjnr[jidx+1];
 483             jnrlistC         = jjnr[jidx+2];
 484             jnrlistD         = jjnr[jidx+3];
 485             /* Sign of each element will be negative for non-real atoms.
 486              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 487              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 488              */
 489             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 490
 491             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 492             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 493             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 494
 495             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 496             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 497             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 498             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 499             j_coord_offsetA  = DIM*jnrA;
 500             j_coord_offsetB  = DIM*jnrB;
 501             j_coord_offsetC  = DIM*jnrC;
 502             j_coord_offsetD  = DIM*jnrD;
 503
 504             /* load j atom coordinates */
 505             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 506                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 507                                                  &jx0,&jy0,&jz0);
 508
 509             /* Calculate displacement vector */
 510             dx00             = _mm256_sub_pd(ix0,jx0);
 511             dy00             = _mm256_sub_pd(iy0,jy0);
 512             dz00             = _mm256_sub_pd(iz0,jz0);
 513             dx10             = _mm256_sub_pd(ix1,jx0);
 514             dy10             = _mm256_sub_pd(iy1,jy0);
 515             dz10             = _mm256_sub_pd(iz1,jz0);
 516             dx20             = _mm256_sub_pd(ix2,jx0);
 517             dy20             = _mm256_sub_pd(iy2,jy0);
 518             dz20             = _mm256_sub_pd(iz2,jz0);
 519             dx30             = _mm256_sub_pd(ix3,jx0);
 520             dy30             = _mm256_sub_pd(iy3,jy0);
 521             dz30             = _mm256_sub_pd(iz3,jz0);
 522
 523             /* Calculate squared distance and things based on it */
 524             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 525             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 526             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 527             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 528
 529             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 530             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 531             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 532             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 533
 534             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 535             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 536             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 537
 538             /* Load parameters for j particles */
 539             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 540                                                                  charge+jnrC+0,charge+jnrD+0);
 541             vdwjidx0A        = 2*vdwtype[jnrA+0];
 542             vdwjidx0B        = 2*vdwtype[jnrB+0];
 543             vdwjidx0C        = 2*vdwtype[jnrC+0];
 544             vdwjidx0D        = 2*vdwtype[jnrD+0];
 545
 546             fjx0             = _mm256_setzero_pd();
 547             fjy0             = _mm256_setzero_pd();
 548             fjz0             = _mm256_setzero_pd();
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             r00              = _mm256_mul_pd(rsq00,rinv00);
 555             r00              = _mm256_andnot_pd(dummy_mask,r00);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 559                                             vdwioffsetptr0+vdwjidx0B,
 560                                             vdwioffsetptr0+vdwjidx0C,
 561                                             vdwioffsetptr0+vdwjidx0D,
 562                                             &c6_00,&c12_00);
 563
 564             /* Calculate table index by multiplying r with table scale and truncate to integer */
 565             rt               = _mm256_mul_pd(r00,vftabscale);
 566             vfitab           = _mm256_cvttpd_epi32(rt);
 567             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 568             vfitab           = _mm_slli_epi32(vfitab,3);
 569
 570             /* CUBIC SPLINE TABLE DISPERSION */
 571             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 572             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 573             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 574             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 575             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 576             Heps             = _mm256_mul_pd(vfeps,H);
 577             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 578             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 579             vvdw6            = _mm256_mul_pd(c6_00,VV);
 580             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 581             fvdw6            = _mm256_mul_pd(c6_00,FF);
 582
 583             /* CUBIC SPLINE TABLE REPULSION */
 584             vfitab           = _mm_add_epi32(vfitab,ifour);
 585             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 586             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 587             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 588             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 589             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 590             Heps             = _mm256_mul_pd(vfeps,H);
 591             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 592             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 593             vvdw12           = _mm256_mul_pd(c12_00,VV);
 594             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 595             fvdw12           = _mm256_mul_pd(c12_00,FF);
 596             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 597             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
 598
 599             /* Update potential sum for this i atom from the interaction with this j atom. */
 600             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 601             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 602
 603             fscal            = fvdw;
 604
 605             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 606
 607             /* Calculate temporary vectorial force */
 608             tx               = _mm256_mul_pd(fscal,dx00);
 609             ty               = _mm256_mul_pd(fscal,dy00);
 610             tz               = _mm256_mul_pd(fscal,dz00);
 611
 612             /* Update vectorial force */
 613             fix0             = _mm256_add_pd(fix0,tx);
 614             fiy0             = _mm256_add_pd(fiy0,ty);
 615             fiz0             = _mm256_add_pd(fiz0,tz);
 616
 617             fjx0             = _mm256_add_pd(fjx0,tx);
 618             fjy0             = _mm256_add_pd(fjy0,ty);
 619             fjz0             = _mm256_add_pd(fjz0,tz);
 620
 621             /**************************
 622              * CALCULATE INTERACTIONS *
 623              **************************/
 624
 625             r10              = _mm256_mul_pd(rsq10,rinv10);
 626             r10              = _mm256_andnot_pd(dummy_mask,r10);
 627
 628             /* Compute parameters for interactions between i and j atoms */
 629             qq10             = _mm256_mul_pd(iq1,jq0);
 630
 631             /* EWALD ELECTROSTATICS */
 632
 633             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 634             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 635             ewitab           = _mm256_cvttpd_epi32(ewrt);
 636             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 637             ewitab           = _mm_slli_epi32(ewitab,2);
 638             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 639             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 640             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 641             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 642             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 643             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 644             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 645             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 646             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velec            = _mm256_andnot_pd(dummy_mask,velec);
 650             velecsum         = _mm256_add_pd(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 655
 656             /* Calculate temporary vectorial force */
 657             tx               = _mm256_mul_pd(fscal,dx10);
 658             ty               = _mm256_mul_pd(fscal,dy10);
 659             tz               = _mm256_mul_pd(fscal,dz10);
 660
 661             /* Update vectorial force */
 662             fix1             = _mm256_add_pd(fix1,tx);
 663             fiy1             = _mm256_add_pd(fiy1,ty);
 664             fiz1             = _mm256_add_pd(fiz1,tz);
 665
 666             fjx0             = _mm256_add_pd(fjx0,tx);
 667             fjy0             = _mm256_add_pd(fjy0,ty);
 668             fjz0             = _mm256_add_pd(fjz0,tz);
 669
 670             /**************************
 671              * CALCULATE INTERACTIONS *
 672              **************************/
 673
 674             r20              = _mm256_mul_pd(rsq20,rinv20);
 675             r20              = _mm256_andnot_pd(dummy_mask,r20);
 676
 677             /* Compute parameters for interactions between i and j atoms */
 678             qq20             = _mm256_mul_pd(iq2,jq0);
 679
 680             /* EWALD ELECTROSTATICS */
 681
 682             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 683             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 684             ewitab           = _mm256_cvttpd_epi32(ewrt);
 685             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 686             ewitab           = _mm_slli_epi32(ewitab,2);
 687             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 688             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 689             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 690             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 691             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 692             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 693             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 694             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 695             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 696
 697             /* Update potential sum for this i atom from the interaction with this j atom. */
 698             velec            = _mm256_andnot_pd(dummy_mask,velec);
 699             velecsum         = _mm256_add_pd(velecsum,velec);
 700
 701             fscal            = felec;
 702
 703             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm256_mul_pd(fscal,dx20);
 707             ty               = _mm256_mul_pd(fscal,dy20);
 708             tz               = _mm256_mul_pd(fscal,dz20);
 709
 710             /* Update vectorial force */
 711             fix2             = _mm256_add_pd(fix2,tx);
 712             fiy2             = _mm256_add_pd(fiy2,ty);
 713             fiz2             = _mm256_add_pd(fiz2,tz);
 714
 715             fjx0             = _mm256_add_pd(fjx0,tx);
 716             fjy0             = _mm256_add_pd(fjy0,ty);
 717             fjz0             = _mm256_add_pd(fjz0,tz);
 718
 719             /**************************
 720              * CALCULATE INTERACTIONS *
 721              **************************/
 722
 723             r30              = _mm256_mul_pd(rsq30,rinv30);
 724             r30              = _mm256_andnot_pd(dummy_mask,r30);
 725
 726             /* Compute parameters for interactions between i and j atoms */
 727             qq30             = _mm256_mul_pd(iq3,jq0);
 728
 729             /* EWALD ELECTROSTATICS */
 730
 731             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 732             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 733             ewitab           = _mm256_cvttpd_epi32(ewrt);
 734             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 735             ewitab           = _mm_slli_epi32(ewitab,2);
 736             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 737             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 738             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 739             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 740             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 741             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 742             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 743             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 744             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 745
 746             /* Update potential sum for this i atom from the interaction with this j atom. */
 747             velec            = _mm256_andnot_pd(dummy_mask,velec);
 748             velecsum         = _mm256_add_pd(velecsum,velec);
 749
 750             fscal            = felec;
 751
 752             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 753
 754             /* Calculate temporary vectorial force */
 755             tx               = _mm256_mul_pd(fscal,dx30);
 756             ty               = _mm256_mul_pd(fscal,dy30);
 757             tz               = _mm256_mul_pd(fscal,dz30);
 758
 759             /* Update vectorial force */
 760             fix3             = _mm256_add_pd(fix3,tx);
 761             fiy3             = _mm256_add_pd(fiy3,ty);
 762             fiz3             = _mm256_add_pd(fiz3,tz);
 763
 764             fjx0             = _mm256_add_pd(fjx0,tx);
 765             fjy0             = _mm256_add_pd(fjy0,ty);
 766             fjz0             = _mm256_add_pd(fjz0,tz);
 767
 768             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 769             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 770             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 771             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 772
 773             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 774
 775             /* Inner loop uses 186 flops */
 776         }
 777
 778         /* End of innermost loop */
 779
 780         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 781                                                  f+i_coord_offset,fshift+i_shift_offset);
 782
 783         ggid                        = gid[iidx];
 784         /* Update potential energies */
 785         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 786         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 787
 788         /* Increment number of inner iterations */
 789         inneriter                  += j_index_end - j_index_start;
 790
 791         /* Outer loop uses 26 flops */
 792     }
 793
 794     /* Increment number of outer iterations */
 795     outeriter        += nri;
 796
 797     /* Update outer/inner flops */
 798
 799     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*186);
 800 }
 801 /*
 802  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_double
 803  * Electrostatics interaction: Ewald
 804  * VdW interaction:            CubicSplineTable
 805  * Geometry:                   Water4-Particle
 806  * Calculate force/pot:        Force
 807  */
 808 void
 809 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_avx_256_double
 810                     (t_nblist                    * gmx_restrict       nlist,
 811                      rvec                        * gmx_restrict          xx,
 812                      rvec                        * gmx_restrict          ff,
 813                      t_forcerec                  * gmx_restrict          fr,
 814                      t_mdatoms                   * gmx_restrict     mdatoms,
 815                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 816                      t_nrnb                      * gmx_restrict        nrnb)
 817 {
 818     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 819      * just 0 for non-waters.
 820      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 821      * jnr indices corresponding to data put in the four positions in the SIMD register.
 822      */
 823     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 824     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 825     int              jnrA,jnrB,jnrC,jnrD;
 826     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 827     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 828     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 829     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 830     real             rcutoff_scalar;
 831     real             *shiftvec,*fshift,*x,*f;
 832     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 833     real             scratch[4*DIM];
 834     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 835     real *           vdwioffsetptr0;
 836     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 837     real *           vdwioffsetptr1;
 838     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 839     real *           vdwioffsetptr2;
 840     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 841     real *           vdwioffsetptr3;
 842     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 843     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 844     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 845     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 846     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 847     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 848     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 849     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 850     real             *charge;
 851     int              nvdwtype;
 852     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 853     int              *vdwtype;
 854     real             *vdwparam;
 855     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 856     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 857     __m128i          vfitab;
 858     __m128i          ifour       = _mm_set1_epi32(4);
 859     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 860     real             *vftab;
 861     __m128i          ewitab;
 862     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 863     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 864     real             *ewtab;
 865     __m256d          dummy_mask,cutoff_mask;
 866     __m128           tmpmask0,tmpmask1;
 867     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 868     __m256d          one     = _mm256_set1_pd(1.0);
 869     __m256d          two     = _mm256_set1_pd(2.0);
 870     x                = xx[0];
 871     f                = ff[0];
 872
 873     nri              = nlist->nri;
 874     iinr             = nlist->iinr;
 875     jindex           = nlist->jindex;
 876     jjnr             = nlist->jjnr;
 877     shiftidx         = nlist->shift;
 878     gid              = nlist->gid;
 879     shiftvec         = fr->shift_vec[0];
 880     fshift           = fr->fshift[0];
 881     facel            = _mm256_set1_pd(fr->epsfac);
 882     charge           = mdatoms->chargeA;
 883     nvdwtype         = fr->ntype;
 884     vdwparam         = fr->nbfp;
 885     vdwtype          = mdatoms->typeA;
 886
 887     vftab            = kernel_data->table_vdw->data;
 888     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 889
 890     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 891     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 892     beta2            = _mm256_mul_pd(beta,beta);
 893     beta3            = _mm256_mul_pd(beta,beta2);
 894
 895     ewtab            = fr->ic->tabq_coul_F;
 896     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 897     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 898
 899     /* Setup water-specific parameters */
 900     inr              = nlist->iinr[0];
 901     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 902     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 903     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 904     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 905
 906     /* Avoid stupid compiler warnings */
 907     jnrA = jnrB = jnrC = jnrD = 0;
 908     j_coord_offsetA = 0;
 909     j_coord_offsetB = 0;
 910     j_coord_offsetC = 0;
 911     j_coord_offsetD = 0;
 912
 913     outeriter        = 0;
 914     inneriter        = 0;
 915
 916     for(iidx=0;iidx<4*DIM;iidx++)
 917     {
 918         scratch[iidx] = 0.0;
 919     }
 920
 921     /* Start outer loop over neighborlists */
 922     for(iidx=0; iidx<nri; iidx++)
 923     {
 924         /* Load shift vector for this list */
 925         i_shift_offset   = DIM*shiftidx[iidx];
 926
 927         /* Load limits for loop over neighbors */
 928         j_index_start    = jindex[iidx];
 929         j_index_end      = jindex[iidx+1];
 930
 931         /* Get outer coordinate index */
 932         inr              = iinr[iidx];
 933         i_coord_offset   = DIM*inr;
 934
 935         /* Load i particle coords and add shift vector */
 936         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 937                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 938
 939         fix0             = _mm256_setzero_pd();
 940         fiy0             = _mm256_setzero_pd();
 941         fiz0             = _mm256_setzero_pd();
 942         fix1             = _mm256_setzero_pd();
 943         fiy1             = _mm256_setzero_pd();
 944         fiz1             = _mm256_setzero_pd();
 945         fix2             = _mm256_setzero_pd();
 946         fiy2             = _mm256_setzero_pd();
 947         fiz2             = _mm256_setzero_pd();
 948         fix3             = _mm256_setzero_pd();
 949         fiy3             = _mm256_setzero_pd();
 950         fiz3             = _mm256_setzero_pd();
 951
 952         /* Start inner kernel loop */
 953         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 954         {
 955
 956             /* Get j neighbor index, and coordinate index */
 957             jnrA             = jjnr[jidx];
 958             jnrB             = jjnr[jidx+1];
 959             jnrC             = jjnr[jidx+2];
 960             jnrD             = jjnr[jidx+3];
 961             j_coord_offsetA  = DIM*jnrA;
 962             j_coord_offsetB  = DIM*jnrB;
 963             j_coord_offsetC  = DIM*jnrC;
 964             j_coord_offsetD  = DIM*jnrD;
 965
 966             /* load j atom coordinates */
 967             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 968                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 969                                                  &jx0,&jy0,&jz0);
 970
 971             /* Calculate displacement vector */
 972             dx00             = _mm256_sub_pd(ix0,jx0);
 973             dy00             = _mm256_sub_pd(iy0,jy0);
 974             dz00             = _mm256_sub_pd(iz0,jz0);
 975             dx10             = _mm256_sub_pd(ix1,jx0);
 976             dy10             = _mm256_sub_pd(iy1,jy0);
 977             dz10             = _mm256_sub_pd(iz1,jz0);
 978             dx20             = _mm256_sub_pd(ix2,jx0);
 979             dy20             = _mm256_sub_pd(iy2,jy0);
 980             dz20             = _mm256_sub_pd(iz2,jz0);
 981             dx30             = _mm256_sub_pd(ix3,jx0);
 982             dy30             = _mm256_sub_pd(iy3,jy0);
 983             dz30             = _mm256_sub_pd(iz3,jz0);
 984
 985             /* Calculate squared distance and things based on it */
 986             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 987             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 988             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 989             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 990
 991             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 992             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 993             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 994             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 995
 996             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 997             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 998             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 999
1000             /* Load parameters for j particles */
1001             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1002                                                                  charge+jnrC+0,charge+jnrD+0);
1003             vdwjidx0A        = 2*vdwtype[jnrA+0];
1004             vdwjidx0B        = 2*vdwtype[jnrB+0];
1005             vdwjidx0C        = 2*vdwtype[jnrC+0];
1006             vdwjidx0D        = 2*vdwtype[jnrD+0];
1007
1008             fjx0             = _mm256_setzero_pd();
1009             fjy0             = _mm256_setzero_pd();
1010             fjz0             = _mm256_setzero_pd();
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             r00              = _mm256_mul_pd(rsq00,rinv00);
1017
1018             /* Compute parameters for interactions between i and j atoms */
1019             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1020                                             vdwioffsetptr0+vdwjidx0B,
1021                                             vdwioffsetptr0+vdwjidx0C,
1022                                             vdwioffsetptr0+vdwjidx0D,
1023                                             &c6_00,&c12_00);
1024
1025             /* Calculate table index by multiplying r with table scale and truncate to integer */
1026             rt               = _mm256_mul_pd(r00,vftabscale);
1027             vfitab           = _mm256_cvttpd_epi32(rt);
1028             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1029             vfitab           = _mm_slli_epi32(vfitab,3);
1030
1031             /* CUBIC SPLINE TABLE DISPERSION */
1032             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1033             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1034             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1035             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1036             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1037             Heps             = _mm256_mul_pd(vfeps,H);
1038             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1039             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1040             fvdw6            = _mm256_mul_pd(c6_00,FF);
1041
1042             /* CUBIC SPLINE TABLE REPULSION */
1043             vfitab           = _mm_add_epi32(vfitab,ifour);
1044             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1045             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1046             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1047             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1048             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1049             Heps             = _mm256_mul_pd(vfeps,H);
1050             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1051             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1052             fvdw12           = _mm256_mul_pd(c12_00,FF);
1053             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1054
1055             fscal            = fvdw;
1056
1057             /* Calculate temporary vectorial force */
1058             tx               = _mm256_mul_pd(fscal,dx00);
1059             ty               = _mm256_mul_pd(fscal,dy00);
1060             tz               = _mm256_mul_pd(fscal,dz00);
1061
1062             /* Update vectorial force */
1063             fix0             = _mm256_add_pd(fix0,tx);
1064             fiy0             = _mm256_add_pd(fiy0,ty);
1065             fiz0             = _mm256_add_pd(fiz0,tz);
1066
1067             fjx0             = _mm256_add_pd(fjx0,tx);
1068             fjy0             = _mm256_add_pd(fjy0,ty);
1069             fjz0             = _mm256_add_pd(fjz0,tz);
1070
1071             /**************************
1072              * CALCULATE INTERACTIONS *
1073              **************************/
1074
1075             r10              = _mm256_mul_pd(rsq10,rinv10);
1076
1077             /* Compute parameters for interactions between i and j atoms */
1078             qq10             = _mm256_mul_pd(iq1,jq0);
1079
1080             /* EWALD ELECTROSTATICS */
1081
1082             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1083             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1084             ewitab           = _mm256_cvttpd_epi32(ewrt);
1085             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1086             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1087                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1088                                             &ewtabF,&ewtabFn);
1089             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1090             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1091
1092             fscal            = felec;
1093
1094             /* Calculate temporary vectorial force */
1095             tx               = _mm256_mul_pd(fscal,dx10);
1096             ty               = _mm256_mul_pd(fscal,dy10);
1097             tz               = _mm256_mul_pd(fscal,dz10);
1098
1099             /* Update vectorial force */
1100             fix1             = _mm256_add_pd(fix1,tx);
1101             fiy1             = _mm256_add_pd(fiy1,ty);
1102             fiz1             = _mm256_add_pd(fiz1,tz);
1103
1104             fjx0             = _mm256_add_pd(fjx0,tx);
1105             fjy0             = _mm256_add_pd(fjy0,ty);
1106             fjz0             = _mm256_add_pd(fjz0,tz);
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             r20              = _mm256_mul_pd(rsq20,rinv20);
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             qq20             = _mm256_mul_pd(iq2,jq0);
1116
1117             /* EWALD ELECTROSTATICS */
1118
1119             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1120             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1121             ewitab           = _mm256_cvttpd_epi32(ewrt);
1122             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1123             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1124                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1125                                             &ewtabF,&ewtabFn);
1126             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1127             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1128
1129             fscal            = felec;
1130
1131             /* Calculate temporary vectorial force */
1132             tx               = _mm256_mul_pd(fscal,dx20);
1133             ty               = _mm256_mul_pd(fscal,dy20);
1134             tz               = _mm256_mul_pd(fscal,dz20);
1135
1136             /* Update vectorial force */
1137             fix2             = _mm256_add_pd(fix2,tx);
1138             fiy2             = _mm256_add_pd(fiy2,ty);
1139             fiz2             = _mm256_add_pd(fiz2,tz);
1140
1141             fjx0             = _mm256_add_pd(fjx0,tx);
1142             fjy0             = _mm256_add_pd(fjy0,ty);
1143             fjz0             = _mm256_add_pd(fjz0,tz);
1144
1145             /**************************
1146              * CALCULATE INTERACTIONS *
1147              **************************/
1148
1149             r30              = _mm256_mul_pd(rsq30,rinv30);
1150
1151             /* Compute parameters for interactions between i and j atoms */
1152             qq30             = _mm256_mul_pd(iq3,jq0);
1153
1154             /* EWALD ELECTROSTATICS */
1155
1156             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1157             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1158             ewitab           = _mm256_cvttpd_epi32(ewrt);
1159             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1160             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1161                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1162                                             &ewtabF,&ewtabFn);
1163             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1164             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1165
1166             fscal            = felec;
1167
1168             /* Calculate temporary vectorial force */
1169             tx               = _mm256_mul_pd(fscal,dx30);
1170             ty               = _mm256_mul_pd(fscal,dy30);
1171             tz               = _mm256_mul_pd(fscal,dz30);
1172
1173             /* Update vectorial force */
1174             fix3             = _mm256_add_pd(fix3,tx);
1175             fiy3             = _mm256_add_pd(fiy3,ty);
1176             fiz3             = _mm256_add_pd(fiz3,tz);
1177
1178             fjx0             = _mm256_add_pd(fjx0,tx);
1179             fjy0             = _mm256_add_pd(fjy0,ty);
1180             fjz0             = _mm256_add_pd(fjz0,tz);
1181
1182             fjptrA             = f+j_coord_offsetA;
1183             fjptrB             = f+j_coord_offsetB;
1184             fjptrC             = f+j_coord_offsetC;
1185             fjptrD             = f+j_coord_offsetD;
1186
1187             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1188
1189             /* Inner loop uses 159 flops */
1190         }
1191
1192         if(jidx<j_index_end)
1193         {
1194
1195             /* Get j neighbor index, and coordinate index */
1196             jnrlistA         = jjnr[jidx];
1197             jnrlistB         = jjnr[jidx+1];
1198             jnrlistC         = jjnr[jidx+2];
1199             jnrlistD         = jjnr[jidx+3];
1200             /* Sign of each element will be negative for non-real atoms.
1201              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1202              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1203              */
1204             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1205
1206             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1207             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1208             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1209
1210             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1211             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1212             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1213             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1214             j_coord_offsetA  = DIM*jnrA;
1215             j_coord_offsetB  = DIM*jnrB;
1216             j_coord_offsetC  = DIM*jnrC;
1217             j_coord_offsetD  = DIM*jnrD;
1218
1219             /* load j atom coordinates */
1220             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1221                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1222                                                  &jx0,&jy0,&jz0);
1223
1224             /* Calculate displacement vector */
1225             dx00             = _mm256_sub_pd(ix0,jx0);
1226             dy00             = _mm256_sub_pd(iy0,jy0);
1227             dz00             = _mm256_sub_pd(iz0,jz0);
1228             dx10             = _mm256_sub_pd(ix1,jx0);
1229             dy10             = _mm256_sub_pd(iy1,jy0);
1230             dz10             = _mm256_sub_pd(iz1,jz0);
1231             dx20             = _mm256_sub_pd(ix2,jx0);
1232             dy20             = _mm256_sub_pd(iy2,jy0);
1233             dz20             = _mm256_sub_pd(iz2,jz0);
1234             dx30             = _mm256_sub_pd(ix3,jx0);
1235             dy30             = _mm256_sub_pd(iy3,jy0);
1236             dz30             = _mm256_sub_pd(iz3,jz0);
1237
1238             /* Calculate squared distance and things based on it */
1239             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1240             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1241             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1242             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1243
1244             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1245             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1246             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1247             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1248
1249             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1250             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1251             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1252
1253             /* Load parameters for j particles */
1254             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1255                                                                  charge+jnrC+0,charge+jnrD+0);
1256             vdwjidx0A        = 2*vdwtype[jnrA+0];
1257             vdwjidx0B        = 2*vdwtype[jnrB+0];
1258             vdwjidx0C        = 2*vdwtype[jnrC+0];
1259             vdwjidx0D        = 2*vdwtype[jnrD+0];
1260
1261             fjx0             = _mm256_setzero_pd();
1262             fjy0             = _mm256_setzero_pd();
1263             fjz0             = _mm256_setzero_pd();
1264
1265             /**************************
1266              * CALCULATE INTERACTIONS *
1267              **************************/
1268
1269             r00              = _mm256_mul_pd(rsq00,rinv00);
1270             r00              = _mm256_andnot_pd(dummy_mask,r00);
1271
1272             /* Compute parameters for interactions between i and j atoms */
1273             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1274                                             vdwioffsetptr0+vdwjidx0B,
1275                                             vdwioffsetptr0+vdwjidx0C,
1276                                             vdwioffsetptr0+vdwjidx0D,
1277                                             &c6_00,&c12_00);
1278
1279             /* Calculate table index by multiplying r with table scale and truncate to integer */
1280             rt               = _mm256_mul_pd(r00,vftabscale);
1281             vfitab           = _mm256_cvttpd_epi32(rt);
1282             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
1283             vfitab           = _mm_slli_epi32(vfitab,3);
1284
1285             /* CUBIC SPLINE TABLE DISPERSION */
1286             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1287             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1288             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1289             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1290             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1291             Heps             = _mm256_mul_pd(vfeps,H);
1292             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1293             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1294             fvdw6            = _mm256_mul_pd(c6_00,FF);
1295
1296             /* CUBIC SPLINE TABLE REPULSION */
1297             vfitab           = _mm_add_epi32(vfitab,ifour);
1298             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1299             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
1300             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
1301             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
1302             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
1303             Heps             = _mm256_mul_pd(vfeps,H);
1304             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
1305             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
1306             fvdw12           = _mm256_mul_pd(c12_00,FF);
1307             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv00)));
1308
1309             fscal            = fvdw;
1310
1311             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1312
1313             /* Calculate temporary vectorial force */
1314             tx               = _mm256_mul_pd(fscal,dx00);
1315             ty               = _mm256_mul_pd(fscal,dy00);
1316             tz               = _mm256_mul_pd(fscal,dz00);
1317
1318             /* Update vectorial force */
1319             fix0             = _mm256_add_pd(fix0,tx);
1320             fiy0             = _mm256_add_pd(fiy0,ty);
1321             fiz0             = _mm256_add_pd(fiz0,tz);
1322
1323             fjx0             = _mm256_add_pd(fjx0,tx);
1324             fjy0             = _mm256_add_pd(fjy0,ty);
1325             fjz0             = _mm256_add_pd(fjz0,tz);
1326
1327             /**************************
1328              * CALCULATE INTERACTIONS *
1329              **************************/
1330
1331             r10              = _mm256_mul_pd(rsq10,rinv10);
1332             r10              = _mm256_andnot_pd(dummy_mask,r10);
1333
1334             /* Compute parameters for interactions between i and j atoms */
1335             qq10             = _mm256_mul_pd(iq1,jq0);
1336
1337             /* EWALD ELECTROSTATICS */
1338
1339             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1340             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1341             ewitab           = _mm256_cvttpd_epi32(ewrt);
1342             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1343             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1344                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1345                                             &ewtabF,&ewtabFn);
1346             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1347             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1348
1349             fscal            = felec;
1350
1351             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1352
1353             /* Calculate temporary vectorial force */
1354             tx               = _mm256_mul_pd(fscal,dx10);
1355             ty               = _mm256_mul_pd(fscal,dy10);
1356             tz               = _mm256_mul_pd(fscal,dz10);
1357
1358             /* Update vectorial force */
1359             fix1             = _mm256_add_pd(fix1,tx);
1360             fiy1             = _mm256_add_pd(fiy1,ty);
1361             fiz1             = _mm256_add_pd(fiz1,tz);
1362
1363             fjx0             = _mm256_add_pd(fjx0,tx);
1364             fjy0             = _mm256_add_pd(fjy0,ty);
1365             fjz0             = _mm256_add_pd(fjz0,tz);
1366
1367             /**************************
1368              * CALCULATE INTERACTIONS *
1369              **************************/
1370
1371             r20              = _mm256_mul_pd(rsq20,rinv20);
1372             r20              = _mm256_andnot_pd(dummy_mask,r20);
1373
1374             /* Compute parameters for interactions between i and j atoms */
1375             qq20             = _mm256_mul_pd(iq2,jq0);
1376
1377             /* EWALD ELECTROSTATICS */
1378
1379             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1380             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1381             ewitab           = _mm256_cvttpd_epi32(ewrt);
1382             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1383             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1384                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1385                                             &ewtabF,&ewtabFn);
1386             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1387             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1388
1389             fscal            = felec;
1390
1391             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1392
1393             /* Calculate temporary vectorial force */
1394             tx               = _mm256_mul_pd(fscal,dx20);
1395             ty               = _mm256_mul_pd(fscal,dy20);
1396             tz               = _mm256_mul_pd(fscal,dz20);
1397
1398             /* Update vectorial force */
1399             fix2             = _mm256_add_pd(fix2,tx);
1400             fiy2             = _mm256_add_pd(fiy2,ty);
1401             fiz2             = _mm256_add_pd(fiz2,tz);
1402
1403             fjx0             = _mm256_add_pd(fjx0,tx);
1404             fjy0             = _mm256_add_pd(fjy0,ty);
1405             fjz0             = _mm256_add_pd(fjz0,tz);
1406
1407             /**************************
1408              * CALCULATE INTERACTIONS *
1409              **************************/
1410
1411             r30              = _mm256_mul_pd(rsq30,rinv30);
1412             r30              = _mm256_andnot_pd(dummy_mask,r30);
1413
1414             /* Compute parameters for interactions between i and j atoms */
1415             qq30             = _mm256_mul_pd(iq3,jq0);
1416
1417             /* EWALD ELECTROSTATICS */
1418
1419             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1420             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1421             ewitab           = _mm256_cvttpd_epi32(ewrt);
1422             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1423             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1424                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1425                                             &ewtabF,&ewtabFn);
1426             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1427             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1428
1429             fscal            = felec;
1430
1431             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1432
1433             /* Calculate temporary vectorial force */
1434             tx               = _mm256_mul_pd(fscal,dx30);
1435             ty               = _mm256_mul_pd(fscal,dy30);
1436             tz               = _mm256_mul_pd(fscal,dz30);
1437
1438             /* Update vectorial force */
1439             fix3             = _mm256_add_pd(fix3,tx);
1440             fiy3             = _mm256_add_pd(fiy3,ty);
1441             fiz3             = _mm256_add_pd(fiz3,tz);
1442
1443             fjx0             = _mm256_add_pd(fjx0,tx);
1444             fjy0             = _mm256_add_pd(fjy0,ty);
1445             fjz0             = _mm256_add_pd(fjz0,tz);
1446
1447             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1448             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1449             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1450             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1451
1452             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1453
1454             /* Inner loop uses 163 flops */
1455         }
1456
1457         /* End of innermost loop */
1458
1459         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1460                                                  f+i_coord_offset,fshift+i_shift_offset);
1461
1462         /* Increment number of inner iterations */
1463         inneriter                  += j_index_end - j_index_start;
1464
1465         /* Outer loop uses 24 flops */
1466     }
1467
1468     /* Increment number of outer iterations */
1469     outeriter        += nri;
1470
1471     /* Update outer/inner flops */
1472
1473     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*163);
1474 }