src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwLJ_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 "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/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_VdwLJ_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_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          ewitab;
 107     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 109     real             *ewtab;
 110     __m256d          dummy_mask,cutoff_mask;
 111     __m128           tmpmask0,tmpmask1;
 112     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 113     __m256d          one     = _mm256_set1_pd(1.0);
 114     __m256d          two     = _mm256_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 133     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 134     beta2            = _mm256_mul_pd(beta,beta);
 135     beta3            = _mm256_mul_pd(beta,beta2);
 136
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 144     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 145     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 146     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = jnrC = jnrD = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152     j_coord_offsetC = 0;
 153     j_coord_offsetD = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 180
 181         fix0             = _mm256_setzero_pd();
 182         fiy0             = _mm256_setzero_pd();
 183         fiz0             = _mm256_setzero_pd();
 184         fix1             = _mm256_setzero_pd();
 185         fiy1             = _mm256_setzero_pd();
 186         fiz1             = _mm256_setzero_pd();
 187         fix2             = _mm256_setzero_pd();
 188         fiy2             = _mm256_setzero_pd();
 189         fiz2             = _mm256_setzero_pd();
 190         fix3             = _mm256_setzero_pd();
 191         fiy3             = _mm256_setzero_pd();
 192         fiz3             = _mm256_setzero_pd();
 193
 194         /* Reset potential sums */
 195         velecsum         = _mm256_setzero_pd();
 196         vvdwsum          = _mm256_setzero_pd();
 197
 198         /* Start inner kernel loop */
 199         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 200         {
 201
 202             /* Get j neighbor index, and coordinate index */
 203             jnrA             = jjnr[jidx];
 204             jnrB             = jjnr[jidx+1];
 205             jnrC             = jjnr[jidx+2];
 206             jnrD             = jjnr[jidx+3];
 207             j_coord_offsetA  = DIM*jnrA;
 208             j_coord_offsetB  = DIM*jnrB;
 209             j_coord_offsetC  = DIM*jnrC;
 210             j_coord_offsetD  = DIM*jnrD;
 211
 212             /* load j atom coordinates */
 213             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 214                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 215                                                  &jx0,&jy0,&jz0);
 216
 217             /* Calculate displacement vector */
 218             dx00             = _mm256_sub_pd(ix0,jx0);
 219             dy00             = _mm256_sub_pd(iy0,jy0);
 220             dz00             = _mm256_sub_pd(iz0,jz0);
 221             dx10             = _mm256_sub_pd(ix1,jx0);
 222             dy10             = _mm256_sub_pd(iy1,jy0);
 223             dz10             = _mm256_sub_pd(iz1,jz0);
 224             dx20             = _mm256_sub_pd(ix2,jx0);
 225             dy20             = _mm256_sub_pd(iy2,jy0);
 226             dz20             = _mm256_sub_pd(iz2,jz0);
 227             dx30             = _mm256_sub_pd(ix3,jx0);
 228             dy30             = _mm256_sub_pd(iy3,jy0);
 229             dz30             = _mm256_sub_pd(iz3,jz0);
 230
 231             /* Calculate squared distance and things based on it */
 232             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 233             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 234             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 235             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 236
 237             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 238             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 239             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 240
 241             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 242             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 243             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 244             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 245
 246             /* Load parameters for j particles */
 247             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 248                                                                  charge+jnrC+0,charge+jnrD+0);
 249             vdwjidx0A        = 2*vdwtype[jnrA+0];
 250             vdwjidx0B        = 2*vdwtype[jnrB+0];
 251             vdwjidx0C        = 2*vdwtype[jnrC+0];
 252             vdwjidx0D        = 2*vdwtype[jnrD+0];
 253
 254             fjx0             = _mm256_setzero_pd();
 255             fjy0             = _mm256_setzero_pd();
 256             fjz0             = _mm256_setzero_pd();
 257
 258             /**************************
 259              * CALCULATE INTERACTIONS *
 260              **************************/
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 264                                             vdwioffsetptr0+vdwjidx0B,
 265                                             vdwioffsetptr0+vdwjidx0C,
 266                                             vdwioffsetptr0+vdwjidx0D,
 267                                             &c6_00,&c12_00);
 268
 269             /* LENNARD-JONES DISPERSION/REPULSION */
 270
 271             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 272             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 273             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 274             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 275             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 279
 280             fscal            = fvdw;
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm256_mul_pd(fscal,dx00);
 284             ty               = _mm256_mul_pd(fscal,dy00);
 285             tz               = _mm256_mul_pd(fscal,dz00);
 286
 287             /* Update vectorial force */
 288             fix0             = _mm256_add_pd(fix0,tx);
 289             fiy0             = _mm256_add_pd(fiy0,ty);
 290             fiz0             = _mm256_add_pd(fiz0,tz);
 291
 292             fjx0             = _mm256_add_pd(fjx0,tx);
 293             fjy0             = _mm256_add_pd(fjy0,ty);
 294             fjz0             = _mm256_add_pd(fjz0,tz);
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             r10              = _mm256_mul_pd(rsq10,rinv10);
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq10             = _mm256_mul_pd(iq1,jq0);
 304
 305             /* EWALD ELECTROSTATICS */
 306
 307             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 308             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 309             ewitab           = _mm256_cvttpd_epi32(ewrt);
 310             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 311             ewitab           = _mm_slli_epi32(ewitab,2);
 312             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 313             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 314             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 315             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 316             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 317             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 318             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 319             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 320             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             velecsum         = _mm256_add_pd(velecsum,velec);
 324
 325             fscal            = felec;
 326
 327             /* Calculate temporary vectorial force */
 328             tx               = _mm256_mul_pd(fscal,dx10);
 329             ty               = _mm256_mul_pd(fscal,dy10);
 330             tz               = _mm256_mul_pd(fscal,dz10);
 331
 332             /* Update vectorial force */
 333             fix1             = _mm256_add_pd(fix1,tx);
 334             fiy1             = _mm256_add_pd(fiy1,ty);
 335             fiz1             = _mm256_add_pd(fiz1,tz);
 336
 337             fjx0             = _mm256_add_pd(fjx0,tx);
 338             fjy0             = _mm256_add_pd(fjy0,ty);
 339             fjz0             = _mm256_add_pd(fjz0,tz);
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             r20              = _mm256_mul_pd(rsq20,rinv20);
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq20             = _mm256_mul_pd(iq2,jq0);
 349
 350             /* EWALD ELECTROSTATICS */
 351
 352             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 353             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 354             ewitab           = _mm256_cvttpd_epi32(ewrt);
 355             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 356             ewitab           = _mm_slli_epi32(ewitab,2);
 357             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 358             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 359             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 360             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 361             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 362             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 363             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 364             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 365             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velecsum         = _mm256_add_pd(velecsum,velec);
 369
 370             fscal            = felec;
 371
 372             /* Calculate temporary vectorial force */
 373             tx               = _mm256_mul_pd(fscal,dx20);
 374             ty               = _mm256_mul_pd(fscal,dy20);
 375             tz               = _mm256_mul_pd(fscal,dz20);
 376
 377             /* Update vectorial force */
 378             fix2             = _mm256_add_pd(fix2,tx);
 379             fiy2             = _mm256_add_pd(fiy2,ty);
 380             fiz2             = _mm256_add_pd(fiz2,tz);
 381
 382             fjx0             = _mm256_add_pd(fjx0,tx);
 383             fjy0             = _mm256_add_pd(fjy0,ty);
 384             fjz0             = _mm256_add_pd(fjz0,tz);
 385
 386             /**************************
 387              * CALCULATE INTERACTIONS *
 388              **************************/
 389
 390             r30              = _mm256_mul_pd(rsq30,rinv30);
 391
 392             /* Compute parameters for interactions between i and j atoms */
 393             qq30             = _mm256_mul_pd(iq3,jq0);
 394
 395             /* EWALD ELECTROSTATICS */
 396
 397             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 398             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 399             ewitab           = _mm256_cvttpd_epi32(ewrt);
 400             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 401             ewitab           = _mm_slli_epi32(ewitab,2);
 402             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 403             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 404             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 405             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 406             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 407             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 408             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 409             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 410             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 411
 412             /* Update potential sum for this i atom from the interaction with this j atom. */
 413             velecsum         = _mm256_add_pd(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm256_mul_pd(fscal,dx30);
 419             ty               = _mm256_mul_pd(fscal,dy30);
 420             tz               = _mm256_mul_pd(fscal,dz30);
 421
 422             /* Update vectorial force */
 423             fix3             = _mm256_add_pd(fix3,tx);
 424             fiy3             = _mm256_add_pd(fiy3,ty);
 425             fiz3             = _mm256_add_pd(fiz3,tz);
 426
 427             fjx0             = _mm256_add_pd(fjx0,tx);
 428             fjy0             = _mm256_add_pd(fjy0,ty);
 429             fjz0             = _mm256_add_pd(fjz0,tz);
 430
 431             fjptrA             = f+j_coord_offsetA;
 432             fjptrB             = f+j_coord_offsetB;
 433             fjptrC             = f+j_coord_offsetC;
 434             fjptrD             = f+j_coord_offsetD;
 435
 436             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 437
 438             /* Inner loop uses 158 flops */
 439         }
 440
 441         if(jidx<j_index_end)
 442         {
 443
 444             /* Get j neighbor index, and coordinate index */
 445             jnrlistA         = jjnr[jidx];
 446             jnrlistB         = jjnr[jidx+1];
 447             jnrlistC         = jjnr[jidx+2];
 448             jnrlistD         = jjnr[jidx+3];
 449             /* Sign of each element will be negative for non-real atoms.
 450              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 451              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 452              */
 453             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 454
 455             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 456             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 457             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 458
 459             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 460             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 461             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 462             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 463             j_coord_offsetA  = DIM*jnrA;
 464             j_coord_offsetB  = DIM*jnrB;
 465             j_coord_offsetC  = DIM*jnrC;
 466             j_coord_offsetD  = DIM*jnrD;
 467
 468             /* load j atom coordinates */
 469             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 470                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 471                                                  &jx0,&jy0,&jz0);
 472
 473             /* Calculate displacement vector */
 474             dx00             = _mm256_sub_pd(ix0,jx0);
 475             dy00             = _mm256_sub_pd(iy0,jy0);
 476             dz00             = _mm256_sub_pd(iz0,jz0);
 477             dx10             = _mm256_sub_pd(ix1,jx0);
 478             dy10             = _mm256_sub_pd(iy1,jy0);
 479             dz10             = _mm256_sub_pd(iz1,jz0);
 480             dx20             = _mm256_sub_pd(ix2,jx0);
 481             dy20             = _mm256_sub_pd(iy2,jy0);
 482             dz20             = _mm256_sub_pd(iz2,jz0);
 483             dx30             = _mm256_sub_pd(ix3,jx0);
 484             dy30             = _mm256_sub_pd(iy3,jy0);
 485             dz30             = _mm256_sub_pd(iz3,jz0);
 486
 487             /* Calculate squared distance and things based on it */
 488             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 489             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 490             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 491             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 492
 493             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 494             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 495             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 496
 497             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 498             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 499             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 500             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 501
 502             /* Load parameters for j particles */
 503             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 504                                                                  charge+jnrC+0,charge+jnrD+0);
 505             vdwjidx0A        = 2*vdwtype[jnrA+0];
 506             vdwjidx0B        = 2*vdwtype[jnrB+0];
 507             vdwjidx0C        = 2*vdwtype[jnrC+0];
 508             vdwjidx0D        = 2*vdwtype[jnrD+0];
 509
 510             fjx0             = _mm256_setzero_pd();
 511             fjy0             = _mm256_setzero_pd();
 512             fjz0             = _mm256_setzero_pd();
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             /* Compute parameters for interactions between i and j atoms */
 519             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 520                                             vdwioffsetptr0+vdwjidx0B,
 521                                             vdwioffsetptr0+vdwjidx0C,
 522                                             vdwioffsetptr0+vdwjidx0D,
 523                                             &c6_00,&c12_00);
 524
 525             /* LENNARD-JONES DISPERSION/REPULSION */
 526
 527             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 528             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 529             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 530             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 531             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 535             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 536
 537             fscal            = fvdw;
 538
 539             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm256_mul_pd(fscal,dx00);
 543             ty               = _mm256_mul_pd(fscal,dy00);
 544             tz               = _mm256_mul_pd(fscal,dz00);
 545
 546             /* Update vectorial force */
 547             fix0             = _mm256_add_pd(fix0,tx);
 548             fiy0             = _mm256_add_pd(fiy0,ty);
 549             fiz0             = _mm256_add_pd(fiz0,tz);
 550
 551             fjx0             = _mm256_add_pd(fjx0,tx);
 552             fjy0             = _mm256_add_pd(fjy0,ty);
 553             fjz0             = _mm256_add_pd(fjz0,tz);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             r10              = _mm256_mul_pd(rsq10,rinv10);
 560             r10              = _mm256_andnot_pd(dummy_mask,r10);
 561
 562             /* Compute parameters for interactions between i and j atoms */
 563             qq10             = _mm256_mul_pd(iq1,jq0);
 564
 565             /* EWALD ELECTROSTATICS */
 566
 567             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 568             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 569             ewitab           = _mm256_cvttpd_epi32(ewrt);
 570             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 571             ewitab           = _mm_slli_epi32(ewitab,2);
 572             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 573             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 574             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 575             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 576             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 577             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 578             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 579             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 580             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm256_andnot_pd(dummy_mask,velec);
 584             velecsum         = _mm256_add_pd(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 589
 590             /* Calculate temporary vectorial force */
 591             tx               = _mm256_mul_pd(fscal,dx10);
 592             ty               = _mm256_mul_pd(fscal,dy10);
 593             tz               = _mm256_mul_pd(fscal,dz10);
 594
 595             /* Update vectorial force */
 596             fix1             = _mm256_add_pd(fix1,tx);
 597             fiy1             = _mm256_add_pd(fiy1,ty);
 598             fiz1             = _mm256_add_pd(fiz1,tz);
 599
 600             fjx0             = _mm256_add_pd(fjx0,tx);
 601             fjy0             = _mm256_add_pd(fjy0,ty);
 602             fjz0             = _mm256_add_pd(fjz0,tz);
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             r20              = _mm256_mul_pd(rsq20,rinv20);
 609             r20              = _mm256_andnot_pd(dummy_mask,r20);
 610
 611             /* Compute parameters for interactions between i and j atoms */
 612             qq20             = _mm256_mul_pd(iq2,jq0);
 613
 614             /* EWALD ELECTROSTATICS */
 615
 616             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 617             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 618             ewitab           = _mm256_cvttpd_epi32(ewrt);
 619             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 620             ewitab           = _mm_slli_epi32(ewitab,2);
 621             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 622             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 623             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 624             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 625             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 626             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 627             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 628             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 629             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 630
 631             /* Update potential sum for this i atom from the interaction with this j atom. */
 632             velec            = _mm256_andnot_pd(dummy_mask,velec);
 633             velecsum         = _mm256_add_pd(velecsum,velec);
 634
 635             fscal            = felec;
 636
 637             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 638
 639             /* Calculate temporary vectorial force */
 640             tx               = _mm256_mul_pd(fscal,dx20);
 641             ty               = _mm256_mul_pd(fscal,dy20);
 642             tz               = _mm256_mul_pd(fscal,dz20);
 643
 644             /* Update vectorial force */
 645             fix2             = _mm256_add_pd(fix2,tx);
 646             fiy2             = _mm256_add_pd(fiy2,ty);
 647             fiz2             = _mm256_add_pd(fiz2,tz);
 648
 649             fjx0             = _mm256_add_pd(fjx0,tx);
 650             fjy0             = _mm256_add_pd(fjy0,ty);
 651             fjz0             = _mm256_add_pd(fjz0,tz);
 652
 653             /**************************
 654              * CALCULATE INTERACTIONS *
 655              **************************/
 656
 657             r30              = _mm256_mul_pd(rsq30,rinv30);
 658             r30              = _mm256_andnot_pd(dummy_mask,r30);
 659
 660             /* Compute parameters for interactions between i and j atoms */
 661             qq30             = _mm256_mul_pd(iq3,jq0);
 662
 663             /* EWALD ELECTROSTATICS */
 664
 665             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 666             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 667             ewitab           = _mm256_cvttpd_epi32(ewrt);
 668             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 669             ewitab           = _mm_slli_epi32(ewitab,2);
 670             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 671             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 672             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 673             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 674             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 675             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 676             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 677             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 678             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 679
 680             /* Update potential sum for this i atom from the interaction with this j atom. */
 681             velec            = _mm256_andnot_pd(dummy_mask,velec);
 682             velecsum         = _mm256_add_pd(velecsum,velec);
 683
 684             fscal            = felec;
 685
 686             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 687
 688             /* Calculate temporary vectorial force */
 689             tx               = _mm256_mul_pd(fscal,dx30);
 690             ty               = _mm256_mul_pd(fscal,dy30);
 691             tz               = _mm256_mul_pd(fscal,dz30);
 692
 693             /* Update vectorial force */
 694             fix3             = _mm256_add_pd(fix3,tx);
 695             fiy3             = _mm256_add_pd(fiy3,ty);
 696             fiz3             = _mm256_add_pd(fiz3,tz);
 697
 698             fjx0             = _mm256_add_pd(fjx0,tx);
 699             fjy0             = _mm256_add_pd(fjy0,ty);
 700             fjz0             = _mm256_add_pd(fjz0,tz);
 701
 702             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 703             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 704             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 705             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 706
 707             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 708
 709             /* Inner loop uses 161 flops */
 710         }
 711
 712         /* End of innermost loop */
 713
 714         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 715                                                  f+i_coord_offset,fshift+i_shift_offset);
 716
 717         ggid                        = gid[iidx];
 718         /* Update potential energies */
 719         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 720         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 721
 722         /* Increment number of inner iterations */
 723         inneriter                  += j_index_end - j_index_start;
 724
 725         /* Outer loop uses 26 flops */
 726     }
 727
 728     /* Increment number of outer iterations */
 729     outeriter        += nri;
 730
 731     /* Update outer/inner flops */
 732
 733     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*161);
 734 }
 735 /*
 736  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_256_double
 737  * Electrostatics interaction: Ewald
 738  * VdW interaction:            LennardJones
 739  * Geometry:                   Water4-Particle
 740  * Calculate force/pot:        Force
 741  */
 742 void
 743 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_256_double
 744                     (t_nblist                    * gmx_restrict       nlist,
 745                      rvec                        * gmx_restrict          xx,
 746                      rvec                        * gmx_restrict          ff,
 747                      t_forcerec                  * gmx_restrict          fr,
 748                      t_mdatoms                   * gmx_restrict     mdatoms,
 749                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 750                      t_nrnb                      * gmx_restrict        nrnb)
 751 {
 752     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 753      * just 0 for non-waters.
 754      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 755      * jnr indices corresponding to data put in the four positions in the SIMD register.
 756      */
 757     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 758     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 759     int              jnrA,jnrB,jnrC,jnrD;
 760     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 761     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 762     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 763     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 764     real             rcutoff_scalar;
 765     real             *shiftvec,*fshift,*x,*f;
 766     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 767     real             scratch[4*DIM];
 768     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 769     real *           vdwioffsetptr0;
 770     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 771     real *           vdwioffsetptr1;
 772     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 773     real *           vdwioffsetptr2;
 774     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 775     real *           vdwioffsetptr3;
 776     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 777     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 778     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 779     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 780     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 781     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 782     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 783     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 784     real             *charge;
 785     int              nvdwtype;
 786     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 787     int              *vdwtype;
 788     real             *vdwparam;
 789     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 790     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 791     __m128i          ewitab;
 792     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 793     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 794     real             *ewtab;
 795     __m256d          dummy_mask,cutoff_mask;
 796     __m128           tmpmask0,tmpmask1;
 797     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 798     __m256d          one     = _mm256_set1_pd(1.0);
 799     __m256d          two     = _mm256_set1_pd(2.0);
 800     x                = xx[0];
 801     f                = ff[0];
 802
 803     nri              = nlist->nri;
 804     iinr             = nlist->iinr;
 805     jindex           = nlist->jindex;
 806     jjnr             = nlist->jjnr;
 807     shiftidx         = nlist->shift;
 808     gid              = nlist->gid;
 809     shiftvec         = fr->shift_vec[0];
 810     fshift           = fr->fshift[0];
 811     facel            = _mm256_set1_pd(fr->epsfac);
 812     charge           = mdatoms->chargeA;
 813     nvdwtype         = fr->ntype;
 814     vdwparam         = fr->nbfp;
 815     vdwtype          = mdatoms->typeA;
 816
 817     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 818     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 819     beta2            = _mm256_mul_pd(beta,beta);
 820     beta3            = _mm256_mul_pd(beta,beta2);
 821
 822     ewtab            = fr->ic->tabq_coul_F;
 823     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 824     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 825
 826     /* Setup water-specific parameters */
 827     inr              = nlist->iinr[0];
 828     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 829     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 830     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 831     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 832
 833     /* Avoid stupid compiler warnings */
 834     jnrA = jnrB = jnrC = jnrD = 0;
 835     j_coord_offsetA = 0;
 836     j_coord_offsetB = 0;
 837     j_coord_offsetC = 0;
 838     j_coord_offsetD = 0;
 839
 840     outeriter        = 0;
 841     inneriter        = 0;
 842
 843     for(iidx=0;iidx<4*DIM;iidx++)
 844     {
 845         scratch[iidx] = 0.0;
 846     }
 847
 848     /* Start outer loop over neighborlists */
 849     for(iidx=0; iidx<nri; iidx++)
 850     {
 851         /* Load shift vector for this list */
 852         i_shift_offset   = DIM*shiftidx[iidx];
 853
 854         /* Load limits for loop over neighbors */
 855         j_index_start    = jindex[iidx];
 856         j_index_end      = jindex[iidx+1];
 857
 858         /* Get outer coordinate index */
 859         inr              = iinr[iidx];
 860         i_coord_offset   = DIM*inr;
 861
 862         /* Load i particle coords and add shift vector */
 863         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 864                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 865
 866         fix0             = _mm256_setzero_pd();
 867         fiy0             = _mm256_setzero_pd();
 868         fiz0             = _mm256_setzero_pd();
 869         fix1             = _mm256_setzero_pd();
 870         fiy1             = _mm256_setzero_pd();
 871         fiz1             = _mm256_setzero_pd();
 872         fix2             = _mm256_setzero_pd();
 873         fiy2             = _mm256_setzero_pd();
 874         fiz2             = _mm256_setzero_pd();
 875         fix3             = _mm256_setzero_pd();
 876         fiy3             = _mm256_setzero_pd();
 877         fiz3             = _mm256_setzero_pd();
 878
 879         /* Start inner kernel loop */
 880         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 881         {
 882
 883             /* Get j neighbor index, and coordinate index */
 884             jnrA             = jjnr[jidx];
 885             jnrB             = jjnr[jidx+1];
 886             jnrC             = jjnr[jidx+2];
 887             jnrD             = jjnr[jidx+3];
 888             j_coord_offsetA  = DIM*jnrA;
 889             j_coord_offsetB  = DIM*jnrB;
 890             j_coord_offsetC  = DIM*jnrC;
 891             j_coord_offsetD  = DIM*jnrD;
 892
 893             /* load j atom coordinates */
 894             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 895                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 896                                                  &jx0,&jy0,&jz0);
 897
 898             /* Calculate displacement vector */
 899             dx00             = _mm256_sub_pd(ix0,jx0);
 900             dy00             = _mm256_sub_pd(iy0,jy0);
 901             dz00             = _mm256_sub_pd(iz0,jz0);
 902             dx10             = _mm256_sub_pd(ix1,jx0);
 903             dy10             = _mm256_sub_pd(iy1,jy0);
 904             dz10             = _mm256_sub_pd(iz1,jz0);
 905             dx20             = _mm256_sub_pd(ix2,jx0);
 906             dy20             = _mm256_sub_pd(iy2,jy0);
 907             dz20             = _mm256_sub_pd(iz2,jz0);
 908             dx30             = _mm256_sub_pd(ix3,jx0);
 909             dy30             = _mm256_sub_pd(iy3,jy0);
 910             dz30             = _mm256_sub_pd(iz3,jz0);
 911
 912             /* Calculate squared distance and things based on it */
 913             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 914             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 915             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 916             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 917
 918             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 919             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 920             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 921
 922             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 923             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 924             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 925             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 926
 927             /* Load parameters for j particles */
 928             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 929                                                                  charge+jnrC+0,charge+jnrD+0);
 930             vdwjidx0A        = 2*vdwtype[jnrA+0];
 931             vdwjidx0B        = 2*vdwtype[jnrB+0];
 932             vdwjidx0C        = 2*vdwtype[jnrC+0];
 933             vdwjidx0D        = 2*vdwtype[jnrD+0];
 934
 935             fjx0             = _mm256_setzero_pd();
 936             fjy0             = _mm256_setzero_pd();
 937             fjz0             = _mm256_setzero_pd();
 938
 939             /**************************
 940              * CALCULATE INTERACTIONS *
 941              **************************/
 942
 943             /* Compute parameters for interactions between i and j atoms */
 944             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 945                                             vdwioffsetptr0+vdwjidx0B,
 946                                             vdwioffsetptr0+vdwjidx0C,
 947                                             vdwioffsetptr0+vdwjidx0D,
 948                                             &c6_00,&c12_00);
 949
 950             /* LENNARD-JONES DISPERSION/REPULSION */
 951
 952             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 953             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 954
 955             fscal            = fvdw;
 956
 957             /* Calculate temporary vectorial force */
 958             tx               = _mm256_mul_pd(fscal,dx00);
 959             ty               = _mm256_mul_pd(fscal,dy00);
 960             tz               = _mm256_mul_pd(fscal,dz00);
 961
 962             /* Update vectorial force */
 963             fix0             = _mm256_add_pd(fix0,tx);
 964             fiy0             = _mm256_add_pd(fiy0,ty);
 965             fiz0             = _mm256_add_pd(fiz0,tz);
 966
 967             fjx0             = _mm256_add_pd(fjx0,tx);
 968             fjy0             = _mm256_add_pd(fjy0,ty);
 969             fjz0             = _mm256_add_pd(fjz0,tz);
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             r10              = _mm256_mul_pd(rsq10,rinv10);
 976
 977             /* Compute parameters for interactions between i and j atoms */
 978             qq10             = _mm256_mul_pd(iq1,jq0);
 979
 980             /* EWALD ELECTROSTATICS */
 981
 982             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 983             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 984             ewitab           = _mm256_cvttpd_epi32(ewrt);
 985             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 986             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 987                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 988                                             &ewtabF,&ewtabFn);
 989             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 990             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 991
 992             fscal            = felec;
 993
 994             /* Calculate temporary vectorial force */
 995             tx               = _mm256_mul_pd(fscal,dx10);
 996             ty               = _mm256_mul_pd(fscal,dy10);
 997             tz               = _mm256_mul_pd(fscal,dz10);
 998
 999             /* Update vectorial force */
1000             fix1             = _mm256_add_pd(fix1,tx);
1001             fiy1             = _mm256_add_pd(fiy1,ty);
1002             fiz1             = _mm256_add_pd(fiz1,tz);
1003
1004             fjx0             = _mm256_add_pd(fjx0,tx);
1005             fjy0             = _mm256_add_pd(fjy0,ty);
1006             fjz0             = _mm256_add_pd(fjz0,tz);
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             r20              = _mm256_mul_pd(rsq20,rinv20);
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq20             = _mm256_mul_pd(iq2,jq0);
1016
1017             /* EWALD ELECTROSTATICS */
1018
1019             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1020             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1021             ewitab           = _mm256_cvttpd_epi32(ewrt);
1022             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1023             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1024                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1025                                             &ewtabF,&ewtabFn);
1026             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1027             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1028
1029             fscal            = felec;
1030
1031             /* Calculate temporary vectorial force */
1032             tx               = _mm256_mul_pd(fscal,dx20);
1033             ty               = _mm256_mul_pd(fscal,dy20);
1034             tz               = _mm256_mul_pd(fscal,dz20);
1035
1036             /* Update vectorial force */
1037             fix2             = _mm256_add_pd(fix2,tx);
1038             fiy2             = _mm256_add_pd(fiy2,ty);
1039             fiz2             = _mm256_add_pd(fiz2,tz);
1040
1041             fjx0             = _mm256_add_pd(fjx0,tx);
1042             fjy0             = _mm256_add_pd(fjy0,ty);
1043             fjz0             = _mm256_add_pd(fjz0,tz);
1044
1045             /**************************
1046              * CALCULATE INTERACTIONS *
1047              **************************/
1048
1049             r30              = _mm256_mul_pd(rsq30,rinv30);
1050
1051             /* Compute parameters for interactions between i and j atoms */
1052             qq30             = _mm256_mul_pd(iq3,jq0);
1053
1054             /* EWALD ELECTROSTATICS */
1055
1056             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1057             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1058             ewitab           = _mm256_cvttpd_epi32(ewrt);
1059             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1060             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1061                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1062                                             &ewtabF,&ewtabFn);
1063             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1064             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1065
1066             fscal            = felec;
1067
1068             /* Calculate temporary vectorial force */
1069             tx               = _mm256_mul_pd(fscal,dx30);
1070             ty               = _mm256_mul_pd(fscal,dy30);
1071             tz               = _mm256_mul_pd(fscal,dz30);
1072
1073             /* Update vectorial force */
1074             fix3             = _mm256_add_pd(fix3,tx);
1075             fiy3             = _mm256_add_pd(fiy3,ty);
1076             fiz3             = _mm256_add_pd(fiz3,tz);
1077
1078             fjx0             = _mm256_add_pd(fjx0,tx);
1079             fjy0             = _mm256_add_pd(fjy0,ty);
1080             fjz0             = _mm256_add_pd(fjz0,tz);
1081
1082             fjptrA             = f+j_coord_offsetA;
1083             fjptrB             = f+j_coord_offsetB;
1084             fjptrC             = f+j_coord_offsetC;
1085             fjptrD             = f+j_coord_offsetD;
1086
1087             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1088
1089             /* Inner loop uses 138 flops */
1090         }
1091
1092         if(jidx<j_index_end)
1093         {
1094
1095             /* Get j neighbor index, and coordinate index */
1096             jnrlistA         = jjnr[jidx];
1097             jnrlistB         = jjnr[jidx+1];
1098             jnrlistC         = jjnr[jidx+2];
1099             jnrlistD         = jjnr[jidx+3];
1100             /* Sign of each element will be negative for non-real atoms.
1101              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1102              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1103              */
1104             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1105
1106             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1107             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1108             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1109
1110             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1111             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1112             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1113             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1114             j_coord_offsetA  = DIM*jnrA;
1115             j_coord_offsetB  = DIM*jnrB;
1116             j_coord_offsetC  = DIM*jnrC;
1117             j_coord_offsetD  = DIM*jnrD;
1118
1119             /* load j atom coordinates */
1120             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1121                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1122                                                  &jx0,&jy0,&jz0);
1123
1124             /* Calculate displacement vector */
1125             dx00             = _mm256_sub_pd(ix0,jx0);
1126             dy00             = _mm256_sub_pd(iy0,jy0);
1127             dz00             = _mm256_sub_pd(iz0,jz0);
1128             dx10             = _mm256_sub_pd(ix1,jx0);
1129             dy10             = _mm256_sub_pd(iy1,jy0);
1130             dz10             = _mm256_sub_pd(iz1,jz0);
1131             dx20             = _mm256_sub_pd(ix2,jx0);
1132             dy20             = _mm256_sub_pd(iy2,jy0);
1133             dz20             = _mm256_sub_pd(iz2,jz0);
1134             dx30             = _mm256_sub_pd(ix3,jx0);
1135             dy30             = _mm256_sub_pd(iy3,jy0);
1136             dz30             = _mm256_sub_pd(iz3,jz0);
1137
1138             /* Calculate squared distance and things based on it */
1139             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1140             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1141             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1142             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1143
1144             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1145             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1146             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1147
1148             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1149             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1150             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1151             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1152
1153             /* Load parameters for j particles */
1154             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1155                                                                  charge+jnrC+0,charge+jnrD+0);
1156             vdwjidx0A        = 2*vdwtype[jnrA+0];
1157             vdwjidx0B        = 2*vdwtype[jnrB+0];
1158             vdwjidx0C        = 2*vdwtype[jnrC+0];
1159             vdwjidx0D        = 2*vdwtype[jnrD+0];
1160
1161             fjx0             = _mm256_setzero_pd();
1162             fjy0             = _mm256_setzero_pd();
1163             fjz0             = _mm256_setzero_pd();
1164
1165             /**************************
1166              * CALCULATE INTERACTIONS *
1167              **************************/
1168
1169             /* Compute parameters for interactions between i and j atoms */
1170             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1171                                             vdwioffsetptr0+vdwjidx0B,
1172                                             vdwioffsetptr0+vdwjidx0C,
1173                                             vdwioffsetptr0+vdwjidx0D,
1174                                             &c6_00,&c12_00);
1175
1176             /* LENNARD-JONES DISPERSION/REPULSION */
1177
1178             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1179             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1180
1181             fscal            = fvdw;
1182
1183             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1184
1185             /* Calculate temporary vectorial force */
1186             tx               = _mm256_mul_pd(fscal,dx00);
1187             ty               = _mm256_mul_pd(fscal,dy00);
1188             tz               = _mm256_mul_pd(fscal,dz00);
1189
1190             /* Update vectorial force */
1191             fix0             = _mm256_add_pd(fix0,tx);
1192             fiy0             = _mm256_add_pd(fiy0,ty);
1193             fiz0             = _mm256_add_pd(fiz0,tz);
1194
1195             fjx0             = _mm256_add_pd(fjx0,tx);
1196             fjy0             = _mm256_add_pd(fjy0,ty);
1197             fjz0             = _mm256_add_pd(fjz0,tz);
1198
1199             /**************************
1200              * CALCULATE INTERACTIONS *
1201              **************************/
1202
1203             r10              = _mm256_mul_pd(rsq10,rinv10);
1204             r10              = _mm256_andnot_pd(dummy_mask,r10);
1205
1206             /* Compute parameters for interactions between i and j atoms */
1207             qq10             = _mm256_mul_pd(iq1,jq0);
1208
1209             /* EWALD ELECTROSTATICS */
1210
1211             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1212             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1213             ewitab           = _mm256_cvttpd_epi32(ewrt);
1214             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1215             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1216                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1217                                             &ewtabF,&ewtabFn);
1218             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1219             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1220
1221             fscal            = felec;
1222
1223             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1224
1225             /* Calculate temporary vectorial force */
1226             tx               = _mm256_mul_pd(fscal,dx10);
1227             ty               = _mm256_mul_pd(fscal,dy10);
1228             tz               = _mm256_mul_pd(fscal,dz10);
1229
1230             /* Update vectorial force */
1231             fix1             = _mm256_add_pd(fix1,tx);
1232             fiy1             = _mm256_add_pd(fiy1,ty);
1233             fiz1             = _mm256_add_pd(fiz1,tz);
1234
1235             fjx0             = _mm256_add_pd(fjx0,tx);
1236             fjy0             = _mm256_add_pd(fjy0,ty);
1237             fjz0             = _mm256_add_pd(fjz0,tz);
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             r20              = _mm256_mul_pd(rsq20,rinv20);
1244             r20              = _mm256_andnot_pd(dummy_mask,r20);
1245
1246             /* Compute parameters for interactions between i and j atoms */
1247             qq20             = _mm256_mul_pd(iq2,jq0);
1248
1249             /* EWALD ELECTROSTATICS */
1250
1251             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1252             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1253             ewitab           = _mm256_cvttpd_epi32(ewrt);
1254             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1255             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1256                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1257                                             &ewtabF,&ewtabFn);
1258             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1259             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1260
1261             fscal            = felec;
1262
1263             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1264
1265             /* Calculate temporary vectorial force */
1266             tx               = _mm256_mul_pd(fscal,dx20);
1267             ty               = _mm256_mul_pd(fscal,dy20);
1268             tz               = _mm256_mul_pd(fscal,dz20);
1269
1270             /* Update vectorial force */
1271             fix2             = _mm256_add_pd(fix2,tx);
1272             fiy2             = _mm256_add_pd(fiy2,ty);
1273             fiz2             = _mm256_add_pd(fiz2,tz);
1274
1275             fjx0             = _mm256_add_pd(fjx0,tx);
1276             fjy0             = _mm256_add_pd(fjy0,ty);
1277             fjz0             = _mm256_add_pd(fjz0,tz);
1278
1279             /**************************
1280              * CALCULATE INTERACTIONS *
1281              **************************/
1282
1283             r30              = _mm256_mul_pd(rsq30,rinv30);
1284             r30              = _mm256_andnot_pd(dummy_mask,r30);
1285
1286             /* Compute parameters for interactions between i and j atoms */
1287             qq30             = _mm256_mul_pd(iq3,jq0);
1288
1289             /* EWALD ELECTROSTATICS */
1290
1291             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1292             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1293             ewitab           = _mm256_cvttpd_epi32(ewrt);
1294             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1295             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1296                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1297                                             &ewtabF,&ewtabFn);
1298             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1299             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1300
1301             fscal            = felec;
1302
1303             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1304
1305             /* Calculate temporary vectorial force */
1306             tx               = _mm256_mul_pd(fscal,dx30);
1307             ty               = _mm256_mul_pd(fscal,dy30);
1308             tz               = _mm256_mul_pd(fscal,dz30);
1309
1310             /* Update vectorial force */
1311             fix3             = _mm256_add_pd(fix3,tx);
1312             fiy3             = _mm256_add_pd(fiy3,ty);
1313             fiz3             = _mm256_add_pd(fiz3,tz);
1314
1315             fjx0             = _mm256_add_pd(fjx0,tx);
1316             fjy0             = _mm256_add_pd(fjy0,ty);
1317             fjz0             = _mm256_add_pd(fjz0,tz);
1318
1319             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1320             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1321             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1322             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1323
1324             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1325
1326             /* Inner loop uses 141 flops */
1327         }
1328
1329         /* End of innermost loop */
1330
1331         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1332                                                  f+i_coord_offset,fshift+i_shift_offset);
1333
1334         /* Increment number of inner iterations */
1335         inneriter                  += j_index_end - j_index_start;
1336
1337         /* Outer loop uses 24 flops */
1338     }
1339
1340     /* Increment number of outer iterations */
1341     outeriter        += nri;
1342
1343     /* Update outer/inner flops */
1344
1345     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*141);
1346 }