src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJSh_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     real *           vdwioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     real *           vdwioffsetptr3;
  93     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  95     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 100     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 101     real             *charge;
 102     int              nvdwtype;
 103     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 104     int              *vdwtype;
 105     real             *vdwparam;
 106     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 107     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 108     __m128i          ewitab;
 109     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 111     real             *ewtab;
 112     __m256d          dummy_mask,cutoff_mask;
 113     __m128           tmpmask0,tmpmask1;
 114     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 115     __m256d          one     = _mm256_set1_pd(1.0);
 116     __m256d          two     = _mm256_set1_pd(2.0);
 117     x                = xx[0];
 118     f                = ff[0];
 119
 120     nri              = nlist->nri;
 121     iinr             = nlist->iinr;
 122     jindex           = nlist->jindex;
 123     jjnr             = nlist->jjnr;
 124     shiftidx         = nlist->shift;
 125     gid              = nlist->gid;
 126     shiftvec         = fr->shift_vec[0];
 127     fshift           = fr->fshift[0];
 128     facel            = _mm256_set1_pd(fr->epsfac);
 129     charge           = mdatoms->chargeA;
 130     nvdwtype         = fr->ntype;
 131     vdwparam         = fr->nbfp;
 132     vdwtype          = mdatoms->typeA;
 133
 134     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 135     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 136     beta2            = _mm256_mul_pd(beta,beta);
 137     beta3            = _mm256_mul_pd(beta,beta2);
 138
 139     ewtab            = fr->ic->tabq_coul_FDV0;
 140     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 141     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 142
 143     /* Setup water-specific parameters */
 144     inr              = nlist->iinr[0];
 145     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 146     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 147     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 148     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 149
 150     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 151     rcutoff_scalar   = fr->rcoulomb;
 152     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 153     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 154
 155     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 156     rvdw             = _mm256_set1_pd(fr->rvdw);
 157
 158     /* Avoid stupid compiler warnings */
 159     jnrA = jnrB = jnrC = jnrD = 0;
 160     j_coord_offsetA = 0;
 161     j_coord_offsetB = 0;
 162     j_coord_offsetC = 0;
 163     j_coord_offsetD = 0;
 164
 165     outeriter        = 0;
 166     inneriter        = 0;
 167
 168     for(iidx=0;iidx<4*DIM;iidx++)
 169     {
 170         scratch[iidx] = 0.0;
 171     }
 172
 173     /* Start outer loop over neighborlists */
 174     for(iidx=0; iidx<nri; iidx++)
 175     {
 176         /* Load shift vector for this list */
 177         i_shift_offset   = DIM*shiftidx[iidx];
 178
 179         /* Load limits for loop over neighbors */
 180         j_index_start    = jindex[iidx];
 181         j_index_end      = jindex[iidx+1];
 182
 183         /* Get outer coordinate index */
 184         inr              = iinr[iidx];
 185         i_coord_offset   = DIM*inr;
 186
 187         /* Load i particle coords and add shift vector */
 188         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 189                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 190
 191         fix0             = _mm256_setzero_pd();
 192         fiy0             = _mm256_setzero_pd();
 193         fiz0             = _mm256_setzero_pd();
 194         fix1             = _mm256_setzero_pd();
 195         fiy1             = _mm256_setzero_pd();
 196         fiz1             = _mm256_setzero_pd();
 197         fix2             = _mm256_setzero_pd();
 198         fiy2             = _mm256_setzero_pd();
 199         fiz2             = _mm256_setzero_pd();
 200         fix3             = _mm256_setzero_pd();
 201         fiy3             = _mm256_setzero_pd();
 202         fiz3             = _mm256_setzero_pd();
 203
 204         /* Reset potential sums */
 205         velecsum         = _mm256_setzero_pd();
 206         vvdwsum          = _mm256_setzero_pd();
 207
 208         /* Start inner kernel loop */
 209         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 210         {
 211
 212             /* Get j neighbor index, and coordinate index */
 213             jnrA             = jjnr[jidx];
 214             jnrB             = jjnr[jidx+1];
 215             jnrC             = jjnr[jidx+2];
 216             jnrD             = jjnr[jidx+3];
 217             j_coord_offsetA  = DIM*jnrA;
 218             j_coord_offsetB  = DIM*jnrB;
 219             j_coord_offsetC  = DIM*jnrC;
 220             j_coord_offsetD  = DIM*jnrD;
 221
 222             /* load j atom coordinates */
 223             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 224                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 225                                                  &jx0,&jy0,&jz0);
 226
 227             /* Calculate displacement vector */
 228             dx00             = _mm256_sub_pd(ix0,jx0);
 229             dy00             = _mm256_sub_pd(iy0,jy0);
 230             dz00             = _mm256_sub_pd(iz0,jz0);
 231             dx10             = _mm256_sub_pd(ix1,jx0);
 232             dy10             = _mm256_sub_pd(iy1,jy0);
 233             dz10             = _mm256_sub_pd(iz1,jz0);
 234             dx20             = _mm256_sub_pd(ix2,jx0);
 235             dy20             = _mm256_sub_pd(iy2,jy0);
 236             dz20             = _mm256_sub_pd(iz2,jz0);
 237             dx30             = _mm256_sub_pd(ix3,jx0);
 238             dy30             = _mm256_sub_pd(iy3,jy0);
 239             dz30             = _mm256_sub_pd(iz3,jz0);
 240
 241             /* Calculate squared distance and things based on it */
 242             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 243             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 244             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 245             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 246
 247             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 248             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 249             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 250
 251             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 252             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 253             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 254             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 255
 256             /* Load parameters for j particles */
 257             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 258                                                                  charge+jnrC+0,charge+jnrD+0);
 259             vdwjidx0A        = 2*vdwtype[jnrA+0];
 260             vdwjidx0B        = 2*vdwtype[jnrB+0];
 261             vdwjidx0C        = 2*vdwtype[jnrC+0];
 262             vdwjidx0D        = 2*vdwtype[jnrD+0];
 263
 264             fjx0             = _mm256_setzero_pd();
 265             fjy0             = _mm256_setzero_pd();
 266             fjz0             = _mm256_setzero_pd();
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 273             {
 274
 275             /* Compute parameters for interactions between i and j atoms */
 276             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 277                                             vdwioffsetptr0+vdwjidx0B,
 278                                             vdwioffsetptr0+vdwjidx0C,
 279                                             vdwioffsetptr0+vdwjidx0D,
 280                                             &c6_00,&c12_00);
 281
 282             /* LENNARD-JONES DISPERSION/REPULSION */
 283
 284             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 285             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 286             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 287             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 288                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 289             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 290
 291             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 295             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 296
 297             fscal            = fvdw;
 298
 299             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm256_mul_pd(fscal,dx00);
 303             ty               = _mm256_mul_pd(fscal,dy00);
 304             tz               = _mm256_mul_pd(fscal,dz00);
 305
 306             /* Update vectorial force */
 307             fix0             = _mm256_add_pd(fix0,tx);
 308             fiy0             = _mm256_add_pd(fiy0,ty);
 309             fiz0             = _mm256_add_pd(fiz0,tz);
 310
 311             fjx0             = _mm256_add_pd(fjx0,tx);
 312             fjy0             = _mm256_add_pd(fjy0,ty);
 313             fjz0             = _mm256_add_pd(fjz0,tz);
 314
 315             }
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 322             {
 323
 324             r10              = _mm256_mul_pd(rsq10,rinv10);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm256_mul_pd(iq1,jq0);
 328
 329             /* EWALD ELECTROSTATICS */
 330
 331             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 332             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 333             ewitab           = _mm256_cvttpd_epi32(ewrt);
 334             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 335             ewitab           = _mm_slli_epi32(ewitab,2);
 336             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 337             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 338             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 339             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 340             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 341             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 342             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 343             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 344             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 345
 346             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velec            = _mm256_and_pd(velec,cutoff_mask);
 350             velecsum         = _mm256_add_pd(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 355
 356             /* Calculate temporary vectorial force */
 357             tx               = _mm256_mul_pd(fscal,dx10);
 358             ty               = _mm256_mul_pd(fscal,dy10);
 359             tz               = _mm256_mul_pd(fscal,dz10);
 360
 361             /* Update vectorial force */
 362             fix1             = _mm256_add_pd(fix1,tx);
 363             fiy1             = _mm256_add_pd(fiy1,ty);
 364             fiz1             = _mm256_add_pd(fiz1,tz);
 365
 366             fjx0             = _mm256_add_pd(fjx0,tx);
 367             fjy0             = _mm256_add_pd(fjy0,ty);
 368             fjz0             = _mm256_add_pd(fjz0,tz);
 369
 370             }
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 377             {
 378
 379             r20              = _mm256_mul_pd(rsq20,rinv20);
 380
 381             /* Compute parameters for interactions between i and j atoms */
 382             qq20             = _mm256_mul_pd(iq2,jq0);
 383
 384             /* EWALD ELECTROSTATICS */
 385
 386             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 387             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 388             ewitab           = _mm256_cvttpd_epi32(ewrt);
 389             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 390             ewitab           = _mm_slli_epi32(ewitab,2);
 391             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 392             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 393             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 394             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 395             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 396             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 397             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 398             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 399             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 400
 401             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velec            = _mm256_and_pd(velec,cutoff_mask);
 405             velecsum         = _mm256_add_pd(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm256_mul_pd(fscal,dx20);
 413             ty               = _mm256_mul_pd(fscal,dy20);
 414             tz               = _mm256_mul_pd(fscal,dz20);
 415
 416             /* Update vectorial force */
 417             fix2             = _mm256_add_pd(fix2,tx);
 418             fiy2             = _mm256_add_pd(fiy2,ty);
 419             fiz2             = _mm256_add_pd(fiz2,tz);
 420
 421             fjx0             = _mm256_add_pd(fjx0,tx);
 422             fjy0             = _mm256_add_pd(fjy0,ty);
 423             fjz0             = _mm256_add_pd(fjz0,tz);
 424
 425             }
 426
 427             /**************************
 428              * CALCULATE INTERACTIONS *
 429              **************************/
 430
 431             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 432             {
 433
 434             r30              = _mm256_mul_pd(rsq30,rinv30);
 435
 436             /* Compute parameters for interactions between i and j atoms */
 437             qq30             = _mm256_mul_pd(iq3,jq0);
 438
 439             /* EWALD ELECTROSTATICS */
 440
 441             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 442             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 443             ewitab           = _mm256_cvttpd_epi32(ewrt);
 444             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 445             ewitab           = _mm_slli_epi32(ewitab,2);
 446             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 447             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 448             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 449             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 450             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 451             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 452             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 453             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 454             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 455
 456             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velec            = _mm256_and_pd(velec,cutoff_mask);
 460             velecsum         = _mm256_add_pd(velecsum,velec);
 461
 462             fscal            = felec;
 463
 464             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm256_mul_pd(fscal,dx30);
 468             ty               = _mm256_mul_pd(fscal,dy30);
 469             tz               = _mm256_mul_pd(fscal,dz30);
 470
 471             /* Update vectorial force */
 472             fix3             = _mm256_add_pd(fix3,tx);
 473             fiy3             = _mm256_add_pd(fiy3,ty);
 474             fiz3             = _mm256_add_pd(fiz3,tz);
 475
 476             fjx0             = _mm256_add_pd(fjx0,tx);
 477             fjy0             = _mm256_add_pd(fjy0,ty);
 478             fjz0             = _mm256_add_pd(fjz0,tz);
 479
 480             }
 481
 482             fjptrA             = f+j_coord_offsetA;
 483             fjptrB             = f+j_coord_offsetB;
 484             fjptrC             = f+j_coord_offsetC;
 485             fjptrD             = f+j_coord_offsetD;
 486
 487             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 488
 489             /* Inner loop uses 182 flops */
 490         }
 491
 492         if(jidx<j_index_end)
 493         {
 494
 495             /* Get j neighbor index, and coordinate index */
 496             jnrlistA         = jjnr[jidx];
 497             jnrlistB         = jjnr[jidx+1];
 498             jnrlistC         = jjnr[jidx+2];
 499             jnrlistD         = jjnr[jidx+3];
 500             /* Sign of each element will be negative for non-real atoms.
 501              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 502              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 503              */
 504             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 505
 506             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 507             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 508             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 509
 510             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 511             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 512             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 513             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 514             j_coord_offsetA  = DIM*jnrA;
 515             j_coord_offsetB  = DIM*jnrB;
 516             j_coord_offsetC  = DIM*jnrC;
 517             j_coord_offsetD  = DIM*jnrD;
 518
 519             /* load j atom coordinates */
 520             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 521                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 522                                                  &jx0,&jy0,&jz0);
 523
 524             /* Calculate displacement vector */
 525             dx00             = _mm256_sub_pd(ix0,jx0);
 526             dy00             = _mm256_sub_pd(iy0,jy0);
 527             dz00             = _mm256_sub_pd(iz0,jz0);
 528             dx10             = _mm256_sub_pd(ix1,jx0);
 529             dy10             = _mm256_sub_pd(iy1,jy0);
 530             dz10             = _mm256_sub_pd(iz1,jz0);
 531             dx20             = _mm256_sub_pd(ix2,jx0);
 532             dy20             = _mm256_sub_pd(iy2,jy0);
 533             dz20             = _mm256_sub_pd(iz2,jz0);
 534             dx30             = _mm256_sub_pd(ix3,jx0);
 535             dy30             = _mm256_sub_pd(iy3,jy0);
 536             dz30             = _mm256_sub_pd(iz3,jz0);
 537
 538             /* Calculate squared distance and things based on it */
 539             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 540             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 541             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 542             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 543
 544             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 545             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 546             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 547
 548             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 549             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 550             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 551             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 552
 553             /* Load parameters for j particles */
 554             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 555                                                                  charge+jnrC+0,charge+jnrD+0);
 556             vdwjidx0A        = 2*vdwtype[jnrA+0];
 557             vdwjidx0B        = 2*vdwtype[jnrB+0];
 558             vdwjidx0C        = 2*vdwtype[jnrC+0];
 559             vdwjidx0D        = 2*vdwtype[jnrD+0];
 560
 561             fjx0             = _mm256_setzero_pd();
 562             fjy0             = _mm256_setzero_pd();
 563             fjz0             = _mm256_setzero_pd();
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 570             {
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 574                                             vdwioffsetptr0+vdwjidx0B,
 575                                             vdwioffsetptr0+vdwjidx0C,
 576                                             vdwioffsetptr0+vdwjidx0D,
 577                                             &c6_00,&c12_00);
 578
 579             /* LENNARD-JONES DISPERSION/REPULSION */
 580
 581             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 582             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 583             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 584             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 585                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 586             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 587
 588             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 592             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 593             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 594
 595             fscal            = fvdw;
 596
 597             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 598
 599             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 600
 601             /* Calculate temporary vectorial force */
 602             tx               = _mm256_mul_pd(fscal,dx00);
 603             ty               = _mm256_mul_pd(fscal,dy00);
 604             tz               = _mm256_mul_pd(fscal,dz00);
 605
 606             /* Update vectorial force */
 607             fix0             = _mm256_add_pd(fix0,tx);
 608             fiy0             = _mm256_add_pd(fiy0,ty);
 609             fiz0             = _mm256_add_pd(fiz0,tz);
 610
 611             fjx0             = _mm256_add_pd(fjx0,tx);
 612             fjy0             = _mm256_add_pd(fjy0,ty);
 613             fjz0             = _mm256_add_pd(fjz0,tz);
 614
 615             }
 616
 617             /**************************
 618              * CALCULATE INTERACTIONS *
 619              **************************/
 620
 621             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 622             {
 623
 624             r10              = _mm256_mul_pd(rsq10,rinv10);
 625             r10              = _mm256_andnot_pd(dummy_mask,r10);
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq10             = _mm256_mul_pd(iq1,jq0);
 629
 630             /* EWALD ELECTROSTATICS */
 631
 632             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 633             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 634             ewitab           = _mm256_cvttpd_epi32(ewrt);
 635             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 636             ewitab           = _mm_slli_epi32(ewitab,2);
 637             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 638             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 639             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 640             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 641             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 642             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 643             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 644             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 645             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 646
 647             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 648
 649             /* Update potential sum for this i atom from the interaction with this j atom. */
 650             velec            = _mm256_and_pd(velec,cutoff_mask);
 651             velec            = _mm256_andnot_pd(dummy_mask,velec);
 652             velecsum         = _mm256_add_pd(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 657
 658             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 659
 660             /* Calculate temporary vectorial force */
 661             tx               = _mm256_mul_pd(fscal,dx10);
 662             ty               = _mm256_mul_pd(fscal,dy10);
 663             tz               = _mm256_mul_pd(fscal,dz10);
 664
 665             /* Update vectorial force */
 666             fix1             = _mm256_add_pd(fix1,tx);
 667             fiy1             = _mm256_add_pd(fiy1,ty);
 668             fiz1             = _mm256_add_pd(fiz1,tz);
 669
 670             fjx0             = _mm256_add_pd(fjx0,tx);
 671             fjy0             = _mm256_add_pd(fjy0,ty);
 672             fjz0             = _mm256_add_pd(fjz0,tz);
 673
 674             }
 675
 676             /**************************
 677              * CALCULATE INTERACTIONS *
 678              **************************/
 679
 680             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 681             {
 682
 683             r20              = _mm256_mul_pd(rsq20,rinv20);
 684             r20              = _mm256_andnot_pd(dummy_mask,r20);
 685
 686             /* Compute parameters for interactions between i and j atoms */
 687             qq20             = _mm256_mul_pd(iq2,jq0);
 688
 689             /* EWALD ELECTROSTATICS */
 690
 691             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 692             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 693             ewitab           = _mm256_cvttpd_epi32(ewrt);
 694             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 695             ewitab           = _mm_slli_epi32(ewitab,2);
 696             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 697             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 698             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 699             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 700             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 701             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 702             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 703             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 704             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 705
 706             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 707
 708             /* Update potential sum for this i atom from the interaction with this j atom. */
 709             velec            = _mm256_and_pd(velec,cutoff_mask);
 710             velec            = _mm256_andnot_pd(dummy_mask,velec);
 711             velecsum         = _mm256_add_pd(velecsum,velec);
 712
 713             fscal            = felec;
 714
 715             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 716
 717             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 718
 719             /* Calculate temporary vectorial force */
 720             tx               = _mm256_mul_pd(fscal,dx20);
 721             ty               = _mm256_mul_pd(fscal,dy20);
 722             tz               = _mm256_mul_pd(fscal,dz20);
 723
 724             /* Update vectorial force */
 725             fix2             = _mm256_add_pd(fix2,tx);
 726             fiy2             = _mm256_add_pd(fiy2,ty);
 727             fiz2             = _mm256_add_pd(fiz2,tz);
 728
 729             fjx0             = _mm256_add_pd(fjx0,tx);
 730             fjy0             = _mm256_add_pd(fjy0,ty);
 731             fjz0             = _mm256_add_pd(fjz0,tz);
 732
 733             }
 734
 735             /**************************
 736              * CALCULATE INTERACTIONS *
 737              **************************/
 738
 739             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 740             {
 741
 742             r30              = _mm256_mul_pd(rsq30,rinv30);
 743             r30              = _mm256_andnot_pd(dummy_mask,r30);
 744
 745             /* Compute parameters for interactions between i and j atoms */
 746             qq30             = _mm256_mul_pd(iq3,jq0);
 747
 748             /* EWALD ELECTROSTATICS */
 749
 750             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 751             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 752             ewitab           = _mm256_cvttpd_epi32(ewrt);
 753             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 754             ewitab           = _mm_slli_epi32(ewitab,2);
 755             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 756             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 757             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 758             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 759             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 760             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 761             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 762             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 763             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 764
 765             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 766
 767             /* Update potential sum for this i atom from the interaction with this j atom. */
 768             velec            = _mm256_and_pd(velec,cutoff_mask);
 769             velec            = _mm256_andnot_pd(dummy_mask,velec);
 770             velecsum         = _mm256_add_pd(velecsum,velec);
 771
 772             fscal            = felec;
 773
 774             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 775
 776             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 777
 778             /* Calculate temporary vectorial force */
 779             tx               = _mm256_mul_pd(fscal,dx30);
 780             ty               = _mm256_mul_pd(fscal,dy30);
 781             tz               = _mm256_mul_pd(fscal,dz30);
 782
 783             /* Update vectorial force */
 784             fix3             = _mm256_add_pd(fix3,tx);
 785             fiy3             = _mm256_add_pd(fiy3,ty);
 786             fiz3             = _mm256_add_pd(fiz3,tz);
 787
 788             fjx0             = _mm256_add_pd(fjx0,tx);
 789             fjy0             = _mm256_add_pd(fjy0,ty);
 790             fjz0             = _mm256_add_pd(fjz0,tz);
 791
 792             }
 793
 794             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 795             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 796             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 797             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 798
 799             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 800
 801             /* Inner loop uses 185 flops */
 802         }
 803
 804         /* End of innermost loop */
 805
 806         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 807                                                  f+i_coord_offset,fshift+i_shift_offset);
 808
 809         ggid                        = gid[iidx];
 810         /* Update potential energies */
 811         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 812         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 813
 814         /* Increment number of inner iterations */
 815         inneriter                  += j_index_end - j_index_start;
 816
 817         /* Outer loop uses 26 flops */
 818     }
 819
 820     /* Increment number of outer iterations */
 821     outeriter        += nri;
 822
 823     /* Update outer/inner flops */
 824
 825     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*185);
 826 }
 827 /*
 828  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_double
 829  * Electrostatics interaction: Ewald
 830  * VdW interaction:            LennardJones
 831  * Geometry:                   Water4-Particle
 832  * Calculate force/pot:        Force
 833  */
 834 void
 835 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_double
 836                     (t_nblist                    * gmx_restrict       nlist,
 837                      rvec                        * gmx_restrict          xx,
 838                      rvec                        * gmx_restrict          ff,
 839                      t_forcerec                  * gmx_restrict          fr,
 840                      t_mdatoms                   * gmx_restrict     mdatoms,
 841                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 842                      t_nrnb                      * gmx_restrict        nrnb)
 843 {
 844     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 845      * just 0 for non-waters.
 846      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 847      * jnr indices corresponding to data put in the four positions in the SIMD register.
 848      */
 849     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 850     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 851     int              jnrA,jnrB,jnrC,jnrD;
 852     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 853     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 854     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 855     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 856     real             rcutoff_scalar;
 857     real             *shiftvec,*fshift,*x,*f;
 858     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 859     real             scratch[4*DIM];
 860     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 861     real *           vdwioffsetptr0;
 862     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 863     real *           vdwioffsetptr1;
 864     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 865     real *           vdwioffsetptr2;
 866     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 867     real *           vdwioffsetptr3;
 868     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 869     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 870     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 871     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 872     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 873     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 874     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 875     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 876     real             *charge;
 877     int              nvdwtype;
 878     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 879     int              *vdwtype;
 880     real             *vdwparam;
 881     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 882     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 883     __m128i          ewitab;
 884     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 885     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 886     real             *ewtab;
 887     __m256d          dummy_mask,cutoff_mask;
 888     __m128           tmpmask0,tmpmask1;
 889     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 890     __m256d          one     = _mm256_set1_pd(1.0);
 891     __m256d          two     = _mm256_set1_pd(2.0);
 892     x                = xx[0];
 893     f                = ff[0];
 894
 895     nri              = nlist->nri;
 896     iinr             = nlist->iinr;
 897     jindex           = nlist->jindex;
 898     jjnr             = nlist->jjnr;
 899     shiftidx         = nlist->shift;
 900     gid              = nlist->gid;
 901     shiftvec         = fr->shift_vec[0];
 902     fshift           = fr->fshift[0];
 903     facel            = _mm256_set1_pd(fr->epsfac);
 904     charge           = mdatoms->chargeA;
 905     nvdwtype         = fr->ntype;
 906     vdwparam         = fr->nbfp;
 907     vdwtype          = mdatoms->typeA;
 908
 909     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 910     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 911     beta2            = _mm256_mul_pd(beta,beta);
 912     beta3            = _mm256_mul_pd(beta,beta2);
 913
 914     ewtab            = fr->ic->tabq_coul_F;
 915     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 916     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 917
 918     /* Setup water-specific parameters */
 919     inr              = nlist->iinr[0];
 920     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 921     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 922     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 923     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 924
 925     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 926     rcutoff_scalar   = fr->rcoulomb;
 927     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 928     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 929
 930     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 931     rvdw             = _mm256_set1_pd(fr->rvdw);
 932
 933     /* Avoid stupid compiler warnings */
 934     jnrA = jnrB = jnrC = jnrD = 0;
 935     j_coord_offsetA = 0;
 936     j_coord_offsetB = 0;
 937     j_coord_offsetC = 0;
 938     j_coord_offsetD = 0;
 939
 940     outeriter        = 0;
 941     inneriter        = 0;
 942
 943     for(iidx=0;iidx<4*DIM;iidx++)
 944     {
 945         scratch[iidx] = 0.0;
 946     }
 947
 948     /* Start outer loop over neighborlists */
 949     for(iidx=0; iidx<nri; iidx++)
 950     {
 951         /* Load shift vector for this list */
 952         i_shift_offset   = DIM*shiftidx[iidx];
 953
 954         /* Load limits for loop over neighbors */
 955         j_index_start    = jindex[iidx];
 956         j_index_end      = jindex[iidx+1];
 957
 958         /* Get outer coordinate index */
 959         inr              = iinr[iidx];
 960         i_coord_offset   = DIM*inr;
 961
 962         /* Load i particle coords and add shift vector */
 963         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 964                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 965
 966         fix0             = _mm256_setzero_pd();
 967         fiy0             = _mm256_setzero_pd();
 968         fiz0             = _mm256_setzero_pd();
 969         fix1             = _mm256_setzero_pd();
 970         fiy1             = _mm256_setzero_pd();
 971         fiz1             = _mm256_setzero_pd();
 972         fix2             = _mm256_setzero_pd();
 973         fiy2             = _mm256_setzero_pd();
 974         fiz2             = _mm256_setzero_pd();
 975         fix3             = _mm256_setzero_pd();
 976         fiy3             = _mm256_setzero_pd();
 977         fiz3             = _mm256_setzero_pd();
 978
 979         /* Start inner kernel loop */
 980         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 981         {
 982
 983             /* Get j neighbor index, and coordinate index */
 984             jnrA             = jjnr[jidx];
 985             jnrB             = jjnr[jidx+1];
 986             jnrC             = jjnr[jidx+2];
 987             jnrD             = jjnr[jidx+3];
 988             j_coord_offsetA  = DIM*jnrA;
 989             j_coord_offsetB  = DIM*jnrB;
 990             j_coord_offsetC  = DIM*jnrC;
 991             j_coord_offsetD  = DIM*jnrD;
 992
 993             /* load j atom coordinates */
 994             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 995                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 996                                                  &jx0,&jy0,&jz0);
 997
 998             /* Calculate displacement vector */
 999             dx00             = _mm256_sub_pd(ix0,jx0);
1000             dy00             = _mm256_sub_pd(iy0,jy0);
1001             dz00             = _mm256_sub_pd(iz0,jz0);
1002             dx10             = _mm256_sub_pd(ix1,jx0);
1003             dy10             = _mm256_sub_pd(iy1,jy0);
1004             dz10             = _mm256_sub_pd(iz1,jz0);
1005             dx20             = _mm256_sub_pd(ix2,jx0);
1006             dy20             = _mm256_sub_pd(iy2,jy0);
1007             dz20             = _mm256_sub_pd(iz2,jz0);
1008             dx30             = _mm256_sub_pd(ix3,jx0);
1009             dy30             = _mm256_sub_pd(iy3,jy0);
1010             dz30             = _mm256_sub_pd(iz3,jz0);
1011
1012             /* Calculate squared distance and things based on it */
1013             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1014             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1015             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1016             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1017
1018             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1019             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1020             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1021
1022             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1023             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1024             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1025             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1026
1027             /* Load parameters for j particles */
1028             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1029                                                                  charge+jnrC+0,charge+jnrD+0);
1030             vdwjidx0A        = 2*vdwtype[jnrA+0];
1031             vdwjidx0B        = 2*vdwtype[jnrB+0];
1032             vdwjidx0C        = 2*vdwtype[jnrC+0];
1033             vdwjidx0D        = 2*vdwtype[jnrD+0];
1034
1035             fjx0             = _mm256_setzero_pd();
1036             fjy0             = _mm256_setzero_pd();
1037             fjz0             = _mm256_setzero_pd();
1038
1039             /**************************
1040              * CALCULATE INTERACTIONS *
1041              **************************/
1042
1043             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1044             {
1045
1046             /* Compute parameters for interactions between i and j atoms */
1047             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1048                                             vdwioffsetptr0+vdwjidx0B,
1049                                             vdwioffsetptr0+vdwjidx0C,
1050                                             vdwioffsetptr0+vdwjidx0D,
1051                                             &c6_00,&c12_00);
1052
1053             /* LENNARD-JONES DISPERSION/REPULSION */
1054
1055             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1056             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1057
1058             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1059
1060             fscal            = fvdw;
1061
1062             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1063
1064             /* Calculate temporary vectorial force */
1065             tx               = _mm256_mul_pd(fscal,dx00);
1066             ty               = _mm256_mul_pd(fscal,dy00);
1067             tz               = _mm256_mul_pd(fscal,dz00);
1068
1069             /* Update vectorial force */
1070             fix0             = _mm256_add_pd(fix0,tx);
1071             fiy0             = _mm256_add_pd(fiy0,ty);
1072             fiz0             = _mm256_add_pd(fiz0,tz);
1073
1074             fjx0             = _mm256_add_pd(fjx0,tx);
1075             fjy0             = _mm256_add_pd(fjy0,ty);
1076             fjz0             = _mm256_add_pd(fjz0,tz);
1077
1078             }
1079
1080             /**************************
1081              * CALCULATE INTERACTIONS *
1082              **************************/
1083
1084             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1085             {
1086
1087             r10              = _mm256_mul_pd(rsq10,rinv10);
1088
1089             /* Compute parameters for interactions between i and j atoms */
1090             qq10             = _mm256_mul_pd(iq1,jq0);
1091
1092             /* EWALD ELECTROSTATICS */
1093
1094             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1095             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1096             ewitab           = _mm256_cvttpd_epi32(ewrt);
1097             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1098             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1099                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1100                                             &ewtabF,&ewtabFn);
1101             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1102             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1103
1104             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1105
1106             fscal            = felec;
1107
1108             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1109
1110             /* Calculate temporary vectorial force */
1111             tx               = _mm256_mul_pd(fscal,dx10);
1112             ty               = _mm256_mul_pd(fscal,dy10);
1113             tz               = _mm256_mul_pd(fscal,dz10);
1114
1115             /* Update vectorial force */
1116             fix1             = _mm256_add_pd(fix1,tx);
1117             fiy1             = _mm256_add_pd(fiy1,ty);
1118             fiz1             = _mm256_add_pd(fiz1,tz);
1119
1120             fjx0             = _mm256_add_pd(fjx0,tx);
1121             fjy0             = _mm256_add_pd(fjy0,ty);
1122             fjz0             = _mm256_add_pd(fjz0,tz);
1123
1124             }
1125
1126             /**************************
1127              * CALCULATE INTERACTIONS *
1128              **************************/
1129
1130             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1131             {
1132
1133             r20              = _mm256_mul_pd(rsq20,rinv20);
1134
1135             /* Compute parameters for interactions between i and j atoms */
1136             qq20             = _mm256_mul_pd(iq2,jq0);
1137
1138             /* EWALD ELECTROSTATICS */
1139
1140             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1141             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1142             ewitab           = _mm256_cvttpd_epi32(ewrt);
1143             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1144             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1145                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1146                                             &ewtabF,&ewtabFn);
1147             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1148             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1149
1150             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1151
1152             fscal            = felec;
1153
1154             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1155
1156             /* Calculate temporary vectorial force */
1157             tx               = _mm256_mul_pd(fscal,dx20);
1158             ty               = _mm256_mul_pd(fscal,dy20);
1159             tz               = _mm256_mul_pd(fscal,dz20);
1160
1161             /* Update vectorial force */
1162             fix2             = _mm256_add_pd(fix2,tx);
1163             fiy2             = _mm256_add_pd(fiy2,ty);
1164             fiz2             = _mm256_add_pd(fiz2,tz);
1165
1166             fjx0             = _mm256_add_pd(fjx0,tx);
1167             fjy0             = _mm256_add_pd(fjy0,ty);
1168             fjz0             = _mm256_add_pd(fjz0,tz);
1169
1170             }
1171
1172             /**************************
1173              * CALCULATE INTERACTIONS *
1174              **************************/
1175
1176             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1177             {
1178
1179             r30              = _mm256_mul_pd(rsq30,rinv30);
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             qq30             = _mm256_mul_pd(iq3,jq0);
1183
1184             /* EWALD ELECTROSTATICS */
1185
1186             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1187             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1188             ewitab           = _mm256_cvttpd_epi32(ewrt);
1189             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1190             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1191                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1192                                             &ewtabF,&ewtabFn);
1193             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1194             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1195
1196             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1201
1202             /* Calculate temporary vectorial force */
1203             tx               = _mm256_mul_pd(fscal,dx30);
1204             ty               = _mm256_mul_pd(fscal,dy30);
1205             tz               = _mm256_mul_pd(fscal,dz30);
1206
1207             /* Update vectorial force */
1208             fix3             = _mm256_add_pd(fix3,tx);
1209             fiy3             = _mm256_add_pd(fiy3,ty);
1210             fiz3             = _mm256_add_pd(fiz3,tz);
1211
1212             fjx0             = _mm256_add_pd(fjx0,tx);
1213             fjy0             = _mm256_add_pd(fjy0,ty);
1214             fjz0             = _mm256_add_pd(fjz0,tz);
1215
1216             }
1217
1218             fjptrA             = f+j_coord_offsetA;
1219             fjptrB             = f+j_coord_offsetB;
1220             fjptrC             = f+j_coord_offsetC;
1221             fjptrD             = f+j_coord_offsetD;
1222
1223             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1224
1225             /* Inner loop uses 150 flops */
1226         }
1227
1228         if(jidx<j_index_end)
1229         {
1230
1231             /* Get j neighbor index, and coordinate index */
1232             jnrlistA         = jjnr[jidx];
1233             jnrlistB         = jjnr[jidx+1];
1234             jnrlistC         = jjnr[jidx+2];
1235             jnrlistD         = jjnr[jidx+3];
1236             /* Sign of each element will be negative for non-real atoms.
1237              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1238              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1239              */
1240             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1241
1242             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1243             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1244             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1245
1246             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1247             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1248             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1249             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1250             j_coord_offsetA  = DIM*jnrA;
1251             j_coord_offsetB  = DIM*jnrB;
1252             j_coord_offsetC  = DIM*jnrC;
1253             j_coord_offsetD  = DIM*jnrD;
1254
1255             /* load j atom coordinates */
1256             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1257                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1258                                                  &jx0,&jy0,&jz0);
1259
1260             /* Calculate displacement vector */
1261             dx00             = _mm256_sub_pd(ix0,jx0);
1262             dy00             = _mm256_sub_pd(iy0,jy0);
1263             dz00             = _mm256_sub_pd(iz0,jz0);
1264             dx10             = _mm256_sub_pd(ix1,jx0);
1265             dy10             = _mm256_sub_pd(iy1,jy0);
1266             dz10             = _mm256_sub_pd(iz1,jz0);
1267             dx20             = _mm256_sub_pd(ix2,jx0);
1268             dy20             = _mm256_sub_pd(iy2,jy0);
1269             dz20             = _mm256_sub_pd(iz2,jz0);
1270             dx30             = _mm256_sub_pd(ix3,jx0);
1271             dy30             = _mm256_sub_pd(iy3,jy0);
1272             dz30             = _mm256_sub_pd(iz3,jz0);
1273
1274             /* Calculate squared distance and things based on it */
1275             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1276             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1277             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1278             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1279
1280             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1281             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1282             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1283
1284             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1285             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1286             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1287             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1288
1289             /* Load parameters for j particles */
1290             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1291                                                                  charge+jnrC+0,charge+jnrD+0);
1292             vdwjidx0A        = 2*vdwtype[jnrA+0];
1293             vdwjidx0B        = 2*vdwtype[jnrB+0];
1294             vdwjidx0C        = 2*vdwtype[jnrC+0];
1295             vdwjidx0D        = 2*vdwtype[jnrD+0];
1296
1297             fjx0             = _mm256_setzero_pd();
1298             fjy0             = _mm256_setzero_pd();
1299             fjz0             = _mm256_setzero_pd();
1300
1301             /**************************
1302              * CALCULATE INTERACTIONS *
1303              **************************/
1304
1305             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1306             {
1307
1308             /* Compute parameters for interactions between i and j atoms */
1309             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1310                                             vdwioffsetptr0+vdwjidx0B,
1311                                             vdwioffsetptr0+vdwjidx0C,
1312                                             vdwioffsetptr0+vdwjidx0D,
1313                                             &c6_00,&c12_00);
1314
1315             /* LENNARD-JONES DISPERSION/REPULSION */
1316
1317             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1318             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1319
1320             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1321
1322             fscal            = fvdw;
1323
1324             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1325
1326             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1327
1328             /* Calculate temporary vectorial force */
1329             tx               = _mm256_mul_pd(fscal,dx00);
1330             ty               = _mm256_mul_pd(fscal,dy00);
1331             tz               = _mm256_mul_pd(fscal,dz00);
1332
1333             /* Update vectorial force */
1334             fix0             = _mm256_add_pd(fix0,tx);
1335             fiy0             = _mm256_add_pd(fiy0,ty);
1336             fiz0             = _mm256_add_pd(fiz0,tz);
1337
1338             fjx0             = _mm256_add_pd(fjx0,tx);
1339             fjy0             = _mm256_add_pd(fjy0,ty);
1340             fjz0             = _mm256_add_pd(fjz0,tz);
1341
1342             }
1343
1344             /**************************
1345              * CALCULATE INTERACTIONS *
1346              **************************/
1347
1348             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1349             {
1350
1351             r10              = _mm256_mul_pd(rsq10,rinv10);
1352             r10              = _mm256_andnot_pd(dummy_mask,r10);
1353
1354             /* Compute parameters for interactions between i and j atoms */
1355             qq10             = _mm256_mul_pd(iq1,jq0);
1356
1357             /* EWALD ELECTROSTATICS */
1358
1359             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1360             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1361             ewitab           = _mm256_cvttpd_epi32(ewrt);
1362             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1363             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1364                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1365                                             &ewtabF,&ewtabFn);
1366             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1367             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1368
1369             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1370
1371             fscal            = felec;
1372
1373             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1374
1375             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1376
1377             /* Calculate temporary vectorial force */
1378             tx               = _mm256_mul_pd(fscal,dx10);
1379             ty               = _mm256_mul_pd(fscal,dy10);
1380             tz               = _mm256_mul_pd(fscal,dz10);
1381
1382             /* Update vectorial force */
1383             fix1             = _mm256_add_pd(fix1,tx);
1384             fiy1             = _mm256_add_pd(fiy1,ty);
1385             fiz1             = _mm256_add_pd(fiz1,tz);
1386
1387             fjx0             = _mm256_add_pd(fjx0,tx);
1388             fjy0             = _mm256_add_pd(fjy0,ty);
1389             fjz0             = _mm256_add_pd(fjz0,tz);
1390
1391             }
1392
1393             /**************************
1394              * CALCULATE INTERACTIONS *
1395              **************************/
1396
1397             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1398             {
1399
1400             r20              = _mm256_mul_pd(rsq20,rinv20);
1401             r20              = _mm256_andnot_pd(dummy_mask,r20);
1402
1403             /* Compute parameters for interactions between i and j atoms */
1404             qq20             = _mm256_mul_pd(iq2,jq0);
1405
1406             /* EWALD ELECTROSTATICS */
1407
1408             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1409             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1410             ewitab           = _mm256_cvttpd_epi32(ewrt);
1411             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1412             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1413                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1414                                             &ewtabF,&ewtabFn);
1415             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1416             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1417
1418             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1419
1420             fscal            = felec;
1421
1422             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1423
1424             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1425
1426             /* Calculate temporary vectorial force */
1427             tx               = _mm256_mul_pd(fscal,dx20);
1428             ty               = _mm256_mul_pd(fscal,dy20);
1429             tz               = _mm256_mul_pd(fscal,dz20);
1430
1431             /* Update vectorial force */
1432             fix2             = _mm256_add_pd(fix2,tx);
1433             fiy2             = _mm256_add_pd(fiy2,ty);
1434             fiz2             = _mm256_add_pd(fiz2,tz);
1435
1436             fjx0             = _mm256_add_pd(fjx0,tx);
1437             fjy0             = _mm256_add_pd(fjy0,ty);
1438             fjz0             = _mm256_add_pd(fjz0,tz);
1439
1440             }
1441
1442             /**************************
1443              * CALCULATE INTERACTIONS *
1444              **************************/
1445
1446             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1447             {
1448
1449             r30              = _mm256_mul_pd(rsq30,rinv30);
1450             r30              = _mm256_andnot_pd(dummy_mask,r30);
1451
1452             /* Compute parameters for interactions between i and j atoms */
1453             qq30             = _mm256_mul_pd(iq3,jq0);
1454
1455             /* EWALD ELECTROSTATICS */
1456
1457             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1458             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1459             ewitab           = _mm256_cvttpd_epi32(ewrt);
1460             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1461             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1462                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1463                                             &ewtabF,&ewtabFn);
1464             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1465             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1466
1467             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1468
1469             fscal            = felec;
1470
1471             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1472
1473             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1474
1475             /* Calculate temporary vectorial force */
1476             tx               = _mm256_mul_pd(fscal,dx30);
1477             ty               = _mm256_mul_pd(fscal,dy30);
1478             tz               = _mm256_mul_pd(fscal,dz30);
1479
1480             /* Update vectorial force */
1481             fix3             = _mm256_add_pd(fix3,tx);
1482             fiy3             = _mm256_add_pd(fiy3,ty);
1483             fiz3             = _mm256_add_pd(fiz3,tz);
1484
1485             fjx0             = _mm256_add_pd(fjx0,tx);
1486             fjy0             = _mm256_add_pd(fjy0,ty);
1487             fjz0             = _mm256_add_pd(fjz0,tz);
1488
1489             }
1490
1491             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1492             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1493             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1494             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1495
1496             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1497
1498             /* Inner loop uses 153 flops */
1499         }
1500
1501         /* End of innermost loop */
1502
1503         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1504                                                  f+i_coord_offset,fshift+i_shift_offset);
1505
1506         /* Increment number of inner iterations */
1507         inneriter                  += j_index_end - j_index_start;
1508
1509         /* Outer loop uses 24 flops */
1510     }
1511
1512     /* Increment number of outer iterations */
1513     outeriter        += nri;
1514
1515     /* Update outer/inner flops */
1516
1517     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*153);
1518 }