src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     real *           vdwgridioffsetptr0;
  85     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     real *           vdwgridioffsetptr1;
  88     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     real *           vdwgridioffsetptr2;
  91     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m256d           c6grid_00;
 106     __m256d           c6grid_10;
 107     __m256d           c6grid_20;
 108     real             *vdwgridparam;
 109     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 110     __m256d           one_half  = _mm256_set1_pd(0.5);
 111     __m256d           minus_one = _mm256_set1_pd(-1.0);
 112     __m128i          ewitab;
 113     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 115     real             *ewtab;
 116     __m256d          dummy_mask,cutoff_mask;
 117     __m128           tmpmask0,tmpmask1;
 118     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 119     __m256d          one     = _mm256_set1_pd(1.0);
 120     __m256d          two     = _mm256_set1_pd(2.0);
 121     x                = xx[0];
 122     f                = ff[0];
 123
 124     nri              = nlist->nri;
 125     iinr             = nlist->iinr;
 126     jindex           = nlist->jindex;
 127     jjnr             = nlist->jjnr;
 128     shiftidx         = nlist->shift;
 129     gid              = nlist->gid;
 130     shiftvec         = fr->shift_vec[0];
 131     fshift           = fr->fshift[0];
 132     facel            = _mm256_set1_pd(fr->ic->epsfac);
 133     charge           = mdatoms->chargeA;
 134     nvdwtype         = fr->ntype;
 135     vdwparam         = fr->nbfp;
 136     vdwtype          = mdatoms->typeA;
 137     vdwgridparam     = fr->ljpme_c6grid;
 138     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 139     ewclj            = _mm256_set1_pd(fr->ic->ewaldcoeff_lj);
 140     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 141
 142     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 143     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 144     beta2            = _mm256_mul_pd(beta,beta);
 145     beta3            = _mm256_mul_pd(beta,beta2);
 146
 147     ewtab            = fr->ic->tabq_coul_FDV0;
 148     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 149     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 150
 151     /* Setup water-specific parameters */
 152     inr              = nlist->iinr[0];
 153     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 154     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 155     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 156     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 157     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 158
 159     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 160     rcutoff_scalar   = fr->ic->rcoulomb;
 161     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 162     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 163
 164     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 165     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173
 174     outeriter        = 0;
 175     inneriter        = 0;
 176
 177     for(iidx=0;iidx<4*DIM;iidx++)
 178     {
 179         scratch[iidx] = 0.0;
 180     }
 181
 182     /* Start outer loop over neighborlists */
 183     for(iidx=0; iidx<nri; iidx++)
 184     {
 185         /* Load shift vector for this list */
 186         i_shift_offset   = DIM*shiftidx[iidx];
 187
 188         /* Load limits for loop over neighbors */
 189         j_index_start    = jindex[iidx];
 190         j_index_end      = jindex[iidx+1];
 191
 192         /* Get outer coordinate index */
 193         inr              = iinr[iidx];
 194         i_coord_offset   = DIM*inr;
 195
 196         /* Load i particle coords and add shift vector */
 197         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 198                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 199
 200         fix0             = _mm256_setzero_pd();
 201         fiy0             = _mm256_setzero_pd();
 202         fiz0             = _mm256_setzero_pd();
 203         fix1             = _mm256_setzero_pd();
 204         fiy1             = _mm256_setzero_pd();
 205         fiz1             = _mm256_setzero_pd();
 206         fix2             = _mm256_setzero_pd();
 207         fiy2             = _mm256_setzero_pd();
 208         fiz2             = _mm256_setzero_pd();
 209
 210         /* Reset potential sums */
 211         velecsum         = _mm256_setzero_pd();
 212         vvdwsum          = _mm256_setzero_pd();
 213
 214         /* Start inner kernel loop */
 215         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 216         {
 217
 218             /* Get j neighbor index, and coordinate index */
 219             jnrA             = jjnr[jidx];
 220             jnrB             = jjnr[jidx+1];
 221             jnrC             = jjnr[jidx+2];
 222             jnrD             = jjnr[jidx+3];
 223             j_coord_offsetA  = DIM*jnrA;
 224             j_coord_offsetB  = DIM*jnrB;
 225             j_coord_offsetC  = DIM*jnrC;
 226             j_coord_offsetD  = DIM*jnrD;
 227
 228             /* load j atom coordinates */
 229             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 230                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 231                                                  &jx0,&jy0,&jz0);
 232
 233             /* Calculate displacement vector */
 234             dx00             = _mm256_sub_pd(ix0,jx0);
 235             dy00             = _mm256_sub_pd(iy0,jy0);
 236             dz00             = _mm256_sub_pd(iz0,jz0);
 237             dx10             = _mm256_sub_pd(ix1,jx0);
 238             dy10             = _mm256_sub_pd(iy1,jy0);
 239             dz10             = _mm256_sub_pd(iz1,jz0);
 240             dx20             = _mm256_sub_pd(ix2,jx0);
 241             dy20             = _mm256_sub_pd(iy2,jy0);
 242             dz20             = _mm256_sub_pd(iz2,jz0);
 243
 244             /* Calculate squared distance and things based on it */
 245             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 246             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 247             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 248
 249             rinv00           = avx256_invsqrt_d(rsq00);
 250             rinv10           = avx256_invsqrt_d(rsq10);
 251             rinv20           = avx256_invsqrt_d(rsq20);
 252
 253             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 254             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 255             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 256
 257             /* Load parameters for j particles */
 258             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 259                                                                  charge+jnrC+0,charge+jnrD+0);
 260             vdwjidx0A        = 2*vdwtype[jnrA+0];
 261             vdwjidx0B        = 2*vdwtype[jnrB+0];
 262             vdwjidx0C        = 2*vdwtype[jnrC+0];
 263             vdwjidx0D        = 2*vdwtype[jnrD+0];
 264
 265             fjx0             = _mm256_setzero_pd();
 266             fjy0             = _mm256_setzero_pd();
 267             fjz0             = _mm256_setzero_pd();
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 274             {
 275
 276             r00              = _mm256_mul_pd(rsq00,rinv00);
 277
 278             /* Compute parameters for interactions between i and j atoms */
 279             qq00             = _mm256_mul_pd(iq0,jq0);
 280             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 281                                             vdwioffsetptr0+vdwjidx0B,
 282                                             vdwioffsetptr0+vdwjidx0C,
 283                                             vdwioffsetptr0+vdwjidx0D,
 284                                             &c6_00,&c12_00);
 285
 286             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 287                                                                   vdwgridioffsetptr0+vdwjidx0B,
 288                                                                   vdwgridioffsetptr0+vdwjidx0C,
 289                                                                   vdwgridioffsetptr0+vdwjidx0D);
 290
 291             /* EWALD ELECTROSTATICS */
 292
 293             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 294             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 295             ewitab           = _mm256_cvttpd_epi32(ewrt);
 296             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 297             ewitab           = _mm_slli_epi32(ewitab,2);
 298             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 299             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 300             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 301             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 302             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 303             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 304             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 305             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 306             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 307
 308             /* Analytical LJ-PME */
 309             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 310             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 311             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 312             exponent         = avx256_exp_d(ewcljrsq);
 313             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 314             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 315             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 316             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 317             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 318             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) ,
 319                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_add_pd(_mm256_mul_pd(c6_00,sh_vdw_invrcut6),_mm256_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 320             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 321             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 322
 323             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velec            = _mm256_and_pd(velec,cutoff_mask);
 327             velecsum         = _mm256_add_pd(velecsum,velec);
 328             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 329             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 330
 331             fscal            = _mm256_add_pd(felec,fvdw);
 332
 333             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_pd(fscal,dx00);
 337             ty               = _mm256_mul_pd(fscal,dy00);
 338             tz               = _mm256_mul_pd(fscal,dz00);
 339
 340             /* Update vectorial force */
 341             fix0             = _mm256_add_pd(fix0,tx);
 342             fiy0             = _mm256_add_pd(fiy0,ty);
 343             fiz0             = _mm256_add_pd(fiz0,tz);
 344
 345             fjx0             = _mm256_add_pd(fjx0,tx);
 346             fjy0             = _mm256_add_pd(fjy0,ty);
 347             fjz0             = _mm256_add_pd(fjz0,tz);
 348
 349             }
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 356             {
 357
 358             r10              = _mm256_mul_pd(rsq10,rinv10);
 359
 360             /* Compute parameters for interactions between i and j atoms */
 361             qq10             = _mm256_mul_pd(iq1,jq0);
 362
 363             /* EWALD ELECTROSTATICS */
 364
 365             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 366             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 367             ewitab           = _mm256_cvttpd_epi32(ewrt);
 368             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 369             ewitab           = _mm_slli_epi32(ewitab,2);
 370             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 371             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 372             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 373             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 374             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 375             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 376             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 377             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 378             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 379
 380             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velec            = _mm256_and_pd(velec,cutoff_mask);
 384             velecsum         = _mm256_add_pd(velecsum,velec);
 385
 386             fscal            = felec;
 387
 388             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 389
 390             /* Calculate temporary vectorial force */
 391             tx               = _mm256_mul_pd(fscal,dx10);
 392             ty               = _mm256_mul_pd(fscal,dy10);
 393             tz               = _mm256_mul_pd(fscal,dz10);
 394
 395             /* Update vectorial force */
 396             fix1             = _mm256_add_pd(fix1,tx);
 397             fiy1             = _mm256_add_pd(fiy1,ty);
 398             fiz1             = _mm256_add_pd(fiz1,tz);
 399
 400             fjx0             = _mm256_add_pd(fjx0,tx);
 401             fjy0             = _mm256_add_pd(fjy0,ty);
 402             fjz0             = _mm256_add_pd(fjz0,tz);
 403
 404             }
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 411             {
 412
 413             r20              = _mm256_mul_pd(rsq20,rinv20);
 414
 415             /* Compute parameters for interactions between i and j atoms */
 416             qq20             = _mm256_mul_pd(iq2,jq0);
 417
 418             /* EWALD ELECTROSTATICS */
 419
 420             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 421             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 422             ewitab           = _mm256_cvttpd_epi32(ewrt);
 423             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 424             ewitab           = _mm_slli_epi32(ewitab,2);
 425             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 426             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 427             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 428             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 429             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 430             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 431             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 432             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 433             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 434
 435             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             velec            = _mm256_and_pd(velec,cutoff_mask);
 439             velecsum         = _mm256_add_pd(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm256_mul_pd(fscal,dx20);
 447             ty               = _mm256_mul_pd(fscal,dy20);
 448             tz               = _mm256_mul_pd(fscal,dz20);
 449
 450             /* Update vectorial force */
 451             fix2             = _mm256_add_pd(fix2,tx);
 452             fiy2             = _mm256_add_pd(fiy2,ty);
 453             fiz2             = _mm256_add_pd(fiz2,tz);
 454
 455             fjx0             = _mm256_add_pd(fjx0,tx);
 456             fjy0             = _mm256_add_pd(fjy0,ty);
 457             fjz0             = _mm256_add_pd(fjz0,tz);
 458
 459             }
 460
 461             fjptrA             = f+j_coord_offsetA;
 462             fjptrB             = f+j_coord_offsetB;
 463             fjptrC             = f+j_coord_offsetC;
 464             fjptrD             = f+j_coord_offsetD;
 465
 466             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 467
 468             /* Inner loop uses 177 flops */
 469         }
 470
 471         if(jidx<j_index_end)
 472         {
 473
 474             /* Get j neighbor index, and coordinate index */
 475             jnrlistA         = jjnr[jidx];
 476             jnrlistB         = jjnr[jidx+1];
 477             jnrlistC         = jjnr[jidx+2];
 478             jnrlistD         = jjnr[jidx+3];
 479             /* Sign of each element will be negative for non-real atoms.
 480              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 481              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 482              */
 483             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 484
 485             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 486             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 487             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 488
 489             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 490             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 491             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 492             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 493             j_coord_offsetA  = DIM*jnrA;
 494             j_coord_offsetB  = DIM*jnrB;
 495             j_coord_offsetC  = DIM*jnrC;
 496             j_coord_offsetD  = DIM*jnrD;
 497
 498             /* load j atom coordinates */
 499             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 500                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 501                                                  &jx0,&jy0,&jz0);
 502
 503             /* Calculate displacement vector */
 504             dx00             = _mm256_sub_pd(ix0,jx0);
 505             dy00             = _mm256_sub_pd(iy0,jy0);
 506             dz00             = _mm256_sub_pd(iz0,jz0);
 507             dx10             = _mm256_sub_pd(ix1,jx0);
 508             dy10             = _mm256_sub_pd(iy1,jy0);
 509             dz10             = _mm256_sub_pd(iz1,jz0);
 510             dx20             = _mm256_sub_pd(ix2,jx0);
 511             dy20             = _mm256_sub_pd(iy2,jy0);
 512             dz20             = _mm256_sub_pd(iz2,jz0);
 513
 514             /* Calculate squared distance and things based on it */
 515             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 516             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 517             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 518
 519             rinv00           = avx256_invsqrt_d(rsq00);
 520             rinv10           = avx256_invsqrt_d(rsq10);
 521             rinv20           = avx256_invsqrt_d(rsq20);
 522
 523             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 524             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 525             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 526
 527             /* Load parameters for j particles */
 528             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 529                                                                  charge+jnrC+0,charge+jnrD+0);
 530             vdwjidx0A        = 2*vdwtype[jnrA+0];
 531             vdwjidx0B        = 2*vdwtype[jnrB+0];
 532             vdwjidx0C        = 2*vdwtype[jnrC+0];
 533             vdwjidx0D        = 2*vdwtype[jnrD+0];
 534
 535             fjx0             = _mm256_setzero_pd();
 536             fjy0             = _mm256_setzero_pd();
 537             fjz0             = _mm256_setzero_pd();
 538
 539             /**************************
 540              * CALCULATE INTERACTIONS *
 541              **************************/
 542
 543             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 544             {
 545
 546             r00              = _mm256_mul_pd(rsq00,rinv00);
 547             r00              = _mm256_andnot_pd(dummy_mask,r00);
 548
 549             /* Compute parameters for interactions between i and j atoms */
 550             qq00             = _mm256_mul_pd(iq0,jq0);
 551             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 552                                             vdwioffsetptr0+vdwjidx0B,
 553                                             vdwioffsetptr0+vdwjidx0C,
 554                                             vdwioffsetptr0+vdwjidx0D,
 555                                             &c6_00,&c12_00);
 556
 557             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 558                                                                   vdwgridioffsetptr0+vdwjidx0B,
 559                                                                   vdwgridioffsetptr0+vdwjidx0C,
 560                                                                   vdwgridioffsetptr0+vdwjidx0D);
 561
 562             /* EWALD ELECTROSTATICS */
 563
 564             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 565             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 566             ewitab           = _mm256_cvttpd_epi32(ewrt);
 567             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 568             ewitab           = _mm_slli_epi32(ewitab,2);
 569             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 570             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 571             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 572             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 573             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 574             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 575             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 576             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 577             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 578
 579             /* Analytical LJ-PME */
 580             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 581             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 582             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 583             exponent         = avx256_exp_d(ewcljrsq);
 584             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 585             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 586             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 587             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 588             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 589             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) ,
 590                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_add_pd(_mm256_mul_pd(c6_00,sh_vdw_invrcut6),_mm256_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 591             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 592             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,_mm256_sub_pd(vvdw6,_mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6)))),rinvsq00);
 593
 594             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 595
 596             /* Update potential sum for this i atom from the interaction with this j atom. */
 597             velec            = _mm256_and_pd(velec,cutoff_mask);
 598             velec            = _mm256_andnot_pd(dummy_mask,velec);
 599             velecsum         = _mm256_add_pd(velecsum,velec);
 600             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 601             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 602             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 603
 604             fscal            = _mm256_add_pd(felec,fvdw);
 605
 606             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 607
 608             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 609
 610             /* Calculate temporary vectorial force */
 611             tx               = _mm256_mul_pd(fscal,dx00);
 612             ty               = _mm256_mul_pd(fscal,dy00);
 613             tz               = _mm256_mul_pd(fscal,dz00);
 614
 615             /* Update vectorial force */
 616             fix0             = _mm256_add_pd(fix0,tx);
 617             fiy0             = _mm256_add_pd(fiy0,ty);
 618             fiz0             = _mm256_add_pd(fiz0,tz);
 619
 620             fjx0             = _mm256_add_pd(fjx0,tx);
 621             fjy0             = _mm256_add_pd(fjy0,ty);
 622             fjz0             = _mm256_add_pd(fjz0,tz);
 623
 624             }
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 631             {
 632
 633             r10              = _mm256_mul_pd(rsq10,rinv10);
 634             r10              = _mm256_andnot_pd(dummy_mask,r10);
 635
 636             /* Compute parameters for interactions between i and j atoms */
 637             qq10             = _mm256_mul_pd(iq1,jq0);
 638
 639             /* EWALD ELECTROSTATICS */
 640
 641             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 642             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 643             ewitab           = _mm256_cvttpd_epi32(ewrt);
 644             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 645             ewitab           = _mm_slli_epi32(ewitab,2);
 646             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 647             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 648             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 649             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 650             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 651             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 652             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 653             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 654             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 655
 656             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 657
 658             /* Update potential sum for this i atom from the interaction with this j atom. */
 659             velec            = _mm256_and_pd(velec,cutoff_mask);
 660             velec            = _mm256_andnot_pd(dummy_mask,velec);
 661             velecsum         = _mm256_add_pd(velecsum,velec);
 662
 663             fscal            = felec;
 664
 665             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 666
 667             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 668
 669             /* Calculate temporary vectorial force */
 670             tx               = _mm256_mul_pd(fscal,dx10);
 671             ty               = _mm256_mul_pd(fscal,dy10);
 672             tz               = _mm256_mul_pd(fscal,dz10);
 673
 674             /* Update vectorial force */
 675             fix1             = _mm256_add_pd(fix1,tx);
 676             fiy1             = _mm256_add_pd(fiy1,ty);
 677             fiz1             = _mm256_add_pd(fiz1,tz);
 678
 679             fjx0             = _mm256_add_pd(fjx0,tx);
 680             fjy0             = _mm256_add_pd(fjy0,ty);
 681             fjz0             = _mm256_add_pd(fjz0,tz);
 682
 683             }
 684
 685             /**************************
 686              * CALCULATE INTERACTIONS *
 687              **************************/
 688
 689             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 690             {
 691
 692             r20              = _mm256_mul_pd(rsq20,rinv20);
 693             r20              = _mm256_andnot_pd(dummy_mask,r20);
 694
 695             /* Compute parameters for interactions between i and j atoms */
 696             qq20             = _mm256_mul_pd(iq2,jq0);
 697
 698             /* EWALD ELECTROSTATICS */
 699
 700             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 701             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 702             ewitab           = _mm256_cvttpd_epi32(ewrt);
 703             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 704             ewitab           = _mm_slli_epi32(ewitab,2);
 705             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 706             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 707             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 708             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 709             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 710             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 711             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 712             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 713             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 714
 715             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 716
 717             /* Update potential sum for this i atom from the interaction with this j atom. */
 718             velec            = _mm256_and_pd(velec,cutoff_mask);
 719             velec            = _mm256_andnot_pd(dummy_mask,velec);
 720             velecsum         = _mm256_add_pd(velecsum,velec);
 721
 722             fscal            = felec;
 723
 724             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 725
 726             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 727
 728             /* Calculate temporary vectorial force */
 729             tx               = _mm256_mul_pd(fscal,dx20);
 730             ty               = _mm256_mul_pd(fscal,dy20);
 731             tz               = _mm256_mul_pd(fscal,dz20);
 732
 733             /* Update vectorial force */
 734             fix2             = _mm256_add_pd(fix2,tx);
 735             fiy2             = _mm256_add_pd(fiy2,ty);
 736             fiz2             = _mm256_add_pd(fiz2,tz);
 737
 738             fjx0             = _mm256_add_pd(fjx0,tx);
 739             fjy0             = _mm256_add_pd(fjy0,ty);
 740             fjz0             = _mm256_add_pd(fjz0,tz);
 741
 742             }
 743
 744             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 745             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 746             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 747             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 748
 749             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 750
 751             /* Inner loop uses 180 flops */
 752         }
 753
 754         /* End of innermost loop */
 755
 756         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 757                                                  f+i_coord_offset,fshift+i_shift_offset);
 758
 759         ggid                        = gid[iidx];
 760         /* Update potential energies */
 761         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 762         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 763
 764         /* Increment number of inner iterations */
 765         inneriter                  += j_index_end - j_index_start;
 766
 767         /* Outer loop uses 20 flops */
 768     }
 769
 770     /* Increment number of outer iterations */
 771     outeriter        += nri;
 772
 773     /* Update outer/inner flops */
 774
 775     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*180);
 776 }
 777 /*
 778  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_double
 779  * Electrostatics interaction: Ewald
 780  * VdW interaction:            LJEwald
 781  * Geometry:                   Water3-Particle
 782  * Calculate force/pot:        Force
 783  */
 784 void
 785 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_double
 786                     (t_nblist                    * gmx_restrict       nlist,
 787                      rvec                        * gmx_restrict          xx,
 788                      rvec                        * gmx_restrict          ff,
 789                      struct t_forcerec           * gmx_restrict          fr,
 790                      t_mdatoms                   * gmx_restrict     mdatoms,
 791                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 792                      t_nrnb                      * gmx_restrict        nrnb)
 793 {
 794     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 795      * just 0 for non-waters.
 796      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 797      * jnr indices corresponding to data put in the four positions in the SIMD register.
 798      */
 799     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 800     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 801     int              jnrA,jnrB,jnrC,jnrD;
 802     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 803     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 804     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 805     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 806     real             rcutoff_scalar;
 807     real             *shiftvec,*fshift,*x,*f;
 808     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 809     real             scratch[4*DIM];
 810     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 811     real *           vdwioffsetptr0;
 812     real *           vdwgridioffsetptr0;
 813     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 814     real *           vdwioffsetptr1;
 815     real *           vdwgridioffsetptr1;
 816     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 817     real *           vdwioffsetptr2;
 818     real *           vdwgridioffsetptr2;
 819     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 820     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 821     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 822     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 823     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 824     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 825     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 826     real             *charge;
 827     int              nvdwtype;
 828     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 829     int              *vdwtype;
 830     real             *vdwparam;
 831     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 832     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 833     __m256d           c6grid_00;
 834     __m256d           c6grid_10;
 835     __m256d           c6grid_20;
 836     real             *vdwgridparam;
 837     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 838     __m256d           one_half  = _mm256_set1_pd(0.5);
 839     __m256d           minus_one = _mm256_set1_pd(-1.0);
 840     __m128i          ewitab;
 841     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 842     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 843     real             *ewtab;
 844     __m256d          dummy_mask,cutoff_mask;
 845     __m128           tmpmask0,tmpmask1;
 846     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 847     __m256d          one     = _mm256_set1_pd(1.0);
 848     __m256d          two     = _mm256_set1_pd(2.0);
 849     x                = xx[0];
 850     f                = ff[0];
 851
 852     nri              = nlist->nri;
 853     iinr             = nlist->iinr;
 854     jindex           = nlist->jindex;
 855     jjnr             = nlist->jjnr;
 856     shiftidx         = nlist->shift;
 857     gid              = nlist->gid;
 858     shiftvec         = fr->shift_vec[0];
 859     fshift           = fr->fshift[0];
 860     facel            = _mm256_set1_pd(fr->ic->epsfac);
 861     charge           = mdatoms->chargeA;
 862     nvdwtype         = fr->ntype;
 863     vdwparam         = fr->nbfp;
 864     vdwtype          = mdatoms->typeA;
 865     vdwgridparam     = fr->ljpme_c6grid;
 866     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 867     ewclj            = _mm256_set1_pd(fr->ic->ewaldcoeff_lj);
 868     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 869
 870     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 871     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 872     beta2            = _mm256_mul_pd(beta,beta);
 873     beta3            = _mm256_mul_pd(beta,beta2);
 874
 875     ewtab            = fr->ic->tabq_coul_F;
 876     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 877     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 878
 879     /* Setup water-specific parameters */
 880     inr              = nlist->iinr[0];
 881     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 882     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 883     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 884     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 885     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 886
 887     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 888     rcutoff_scalar   = fr->ic->rcoulomb;
 889     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 890     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 891
 892     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 893     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 894
 895     /* Avoid stupid compiler warnings */
 896     jnrA = jnrB = jnrC = jnrD = 0;
 897     j_coord_offsetA = 0;
 898     j_coord_offsetB = 0;
 899     j_coord_offsetC = 0;
 900     j_coord_offsetD = 0;
 901
 902     outeriter        = 0;
 903     inneriter        = 0;
 904
 905     for(iidx=0;iidx<4*DIM;iidx++)
 906     {
 907         scratch[iidx] = 0.0;
 908     }
 909
 910     /* Start outer loop over neighborlists */
 911     for(iidx=0; iidx<nri; iidx++)
 912     {
 913         /* Load shift vector for this list */
 914         i_shift_offset   = DIM*shiftidx[iidx];
 915
 916         /* Load limits for loop over neighbors */
 917         j_index_start    = jindex[iidx];
 918         j_index_end      = jindex[iidx+1];
 919
 920         /* Get outer coordinate index */
 921         inr              = iinr[iidx];
 922         i_coord_offset   = DIM*inr;
 923
 924         /* Load i particle coords and add shift vector */
 925         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 926                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 927
 928         fix0             = _mm256_setzero_pd();
 929         fiy0             = _mm256_setzero_pd();
 930         fiz0             = _mm256_setzero_pd();
 931         fix1             = _mm256_setzero_pd();
 932         fiy1             = _mm256_setzero_pd();
 933         fiz1             = _mm256_setzero_pd();
 934         fix2             = _mm256_setzero_pd();
 935         fiy2             = _mm256_setzero_pd();
 936         fiz2             = _mm256_setzero_pd();
 937
 938         /* Start inner kernel loop */
 939         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 940         {
 941
 942             /* Get j neighbor index, and coordinate index */
 943             jnrA             = jjnr[jidx];
 944             jnrB             = jjnr[jidx+1];
 945             jnrC             = jjnr[jidx+2];
 946             jnrD             = jjnr[jidx+3];
 947             j_coord_offsetA  = DIM*jnrA;
 948             j_coord_offsetB  = DIM*jnrB;
 949             j_coord_offsetC  = DIM*jnrC;
 950             j_coord_offsetD  = DIM*jnrD;
 951
 952             /* load j atom coordinates */
 953             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 954                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 955                                                  &jx0,&jy0,&jz0);
 956
 957             /* Calculate displacement vector */
 958             dx00             = _mm256_sub_pd(ix0,jx0);
 959             dy00             = _mm256_sub_pd(iy0,jy0);
 960             dz00             = _mm256_sub_pd(iz0,jz0);
 961             dx10             = _mm256_sub_pd(ix1,jx0);
 962             dy10             = _mm256_sub_pd(iy1,jy0);
 963             dz10             = _mm256_sub_pd(iz1,jz0);
 964             dx20             = _mm256_sub_pd(ix2,jx0);
 965             dy20             = _mm256_sub_pd(iy2,jy0);
 966             dz20             = _mm256_sub_pd(iz2,jz0);
 967
 968             /* Calculate squared distance and things based on it */
 969             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 970             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 971             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 972
 973             rinv00           = avx256_invsqrt_d(rsq00);
 974             rinv10           = avx256_invsqrt_d(rsq10);
 975             rinv20           = avx256_invsqrt_d(rsq20);
 976
 977             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 978             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 979             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 980
 981             /* Load parameters for j particles */
 982             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 983                                                                  charge+jnrC+0,charge+jnrD+0);
 984             vdwjidx0A        = 2*vdwtype[jnrA+0];
 985             vdwjidx0B        = 2*vdwtype[jnrB+0];
 986             vdwjidx0C        = 2*vdwtype[jnrC+0];
 987             vdwjidx0D        = 2*vdwtype[jnrD+0];
 988
 989             fjx0             = _mm256_setzero_pd();
 990             fjy0             = _mm256_setzero_pd();
 991             fjz0             = _mm256_setzero_pd();
 992
 993             /**************************
 994              * CALCULATE INTERACTIONS *
 995              **************************/
 996
 997             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 998             {
 999
1000             r00              = _mm256_mul_pd(rsq00,rinv00);
1001
1002             /* Compute parameters for interactions between i and j atoms */
1003             qq00             = _mm256_mul_pd(iq0,jq0);
1004             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1005                                             vdwioffsetptr0+vdwjidx0B,
1006                                             vdwioffsetptr0+vdwjidx0C,
1007                                             vdwioffsetptr0+vdwjidx0D,
1008                                             &c6_00,&c12_00);
1009
1010             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1011                                                                   vdwgridioffsetptr0+vdwjidx0B,
1012                                                                   vdwgridioffsetptr0+vdwjidx0C,
1013                                                                   vdwgridioffsetptr0+vdwjidx0D);
1014
1015             /* EWALD ELECTROSTATICS */
1016
1017             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1018             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1019             ewitab           = _mm256_cvttpd_epi32(ewrt);
1020             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1021             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1022                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1023                                             &ewtabF,&ewtabFn);
1024             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1025             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1026
1027             /* Analytical LJ-PME */
1028             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1029             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1030             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1031             exponent         = avx256_exp_d(ewcljrsq);
1032             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1033             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1034             /* f6A = 6 * C6grid * (1 - poly) */
1035             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1036             /* f6B = C6grid * exponent * beta^6 */
1037             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1038             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1039             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1040
1041             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1042
1043             fscal            = _mm256_add_pd(felec,fvdw);
1044
1045             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1046
1047             /* Calculate temporary vectorial force */
1048             tx               = _mm256_mul_pd(fscal,dx00);
1049             ty               = _mm256_mul_pd(fscal,dy00);
1050             tz               = _mm256_mul_pd(fscal,dz00);
1051
1052             /* Update vectorial force */
1053             fix0             = _mm256_add_pd(fix0,tx);
1054             fiy0             = _mm256_add_pd(fiy0,ty);
1055             fiz0             = _mm256_add_pd(fiz0,tz);
1056
1057             fjx0             = _mm256_add_pd(fjx0,tx);
1058             fjy0             = _mm256_add_pd(fjy0,ty);
1059             fjz0             = _mm256_add_pd(fjz0,tz);
1060
1061             }
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1068             {
1069
1070             r10              = _mm256_mul_pd(rsq10,rinv10);
1071
1072             /* Compute parameters for interactions between i and j atoms */
1073             qq10             = _mm256_mul_pd(iq1,jq0);
1074
1075             /* EWALD ELECTROSTATICS */
1076
1077             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1078             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1079             ewitab           = _mm256_cvttpd_epi32(ewrt);
1080             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1081             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1082                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1083                                             &ewtabF,&ewtabFn);
1084             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1085             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1086
1087             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1088
1089             fscal            = felec;
1090
1091             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1092
1093             /* Calculate temporary vectorial force */
1094             tx               = _mm256_mul_pd(fscal,dx10);
1095             ty               = _mm256_mul_pd(fscal,dy10);
1096             tz               = _mm256_mul_pd(fscal,dz10);
1097
1098             /* Update vectorial force */
1099             fix1             = _mm256_add_pd(fix1,tx);
1100             fiy1             = _mm256_add_pd(fiy1,ty);
1101             fiz1             = _mm256_add_pd(fiz1,tz);
1102
1103             fjx0             = _mm256_add_pd(fjx0,tx);
1104             fjy0             = _mm256_add_pd(fjy0,ty);
1105             fjz0             = _mm256_add_pd(fjz0,tz);
1106
1107             }
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1114             {
1115
1116             r20              = _mm256_mul_pd(rsq20,rinv20);
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq20             = _mm256_mul_pd(iq2,jq0);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1124             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1125             ewitab           = _mm256_cvttpd_epi32(ewrt);
1126             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1127             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1128                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1129                                             &ewtabF,&ewtabFn);
1130             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1131             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1132
1133             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1134
1135             fscal            = felec;
1136
1137             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm256_mul_pd(fscal,dx20);
1141             ty               = _mm256_mul_pd(fscal,dy20);
1142             tz               = _mm256_mul_pd(fscal,dz20);
1143
1144             /* Update vectorial force */
1145             fix2             = _mm256_add_pd(fix2,tx);
1146             fiy2             = _mm256_add_pd(fiy2,ty);
1147             fiz2             = _mm256_add_pd(fiz2,tz);
1148
1149             fjx0             = _mm256_add_pd(fjx0,tx);
1150             fjy0             = _mm256_add_pd(fjy0,ty);
1151             fjz0             = _mm256_add_pd(fjz0,tz);
1152
1153             }
1154
1155             fjptrA             = f+j_coord_offsetA;
1156             fjptrB             = f+j_coord_offsetB;
1157             fjptrC             = f+j_coord_offsetC;
1158             fjptrD             = f+j_coord_offsetD;
1159
1160             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1161
1162             /* Inner loop uses 143 flops */
1163         }
1164
1165         if(jidx<j_index_end)
1166         {
1167
1168             /* Get j neighbor index, and coordinate index */
1169             jnrlistA         = jjnr[jidx];
1170             jnrlistB         = jjnr[jidx+1];
1171             jnrlistC         = jjnr[jidx+2];
1172             jnrlistD         = jjnr[jidx+3];
1173             /* Sign of each element will be negative for non-real atoms.
1174              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1175              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1176              */
1177             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1178
1179             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1180             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1181             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1182
1183             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1184             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1185             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1186             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1187             j_coord_offsetA  = DIM*jnrA;
1188             j_coord_offsetB  = DIM*jnrB;
1189             j_coord_offsetC  = DIM*jnrC;
1190             j_coord_offsetD  = DIM*jnrD;
1191
1192             /* load j atom coordinates */
1193             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1194                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1195                                                  &jx0,&jy0,&jz0);
1196
1197             /* Calculate displacement vector */
1198             dx00             = _mm256_sub_pd(ix0,jx0);
1199             dy00             = _mm256_sub_pd(iy0,jy0);
1200             dz00             = _mm256_sub_pd(iz0,jz0);
1201             dx10             = _mm256_sub_pd(ix1,jx0);
1202             dy10             = _mm256_sub_pd(iy1,jy0);
1203             dz10             = _mm256_sub_pd(iz1,jz0);
1204             dx20             = _mm256_sub_pd(ix2,jx0);
1205             dy20             = _mm256_sub_pd(iy2,jy0);
1206             dz20             = _mm256_sub_pd(iz2,jz0);
1207
1208             /* Calculate squared distance and things based on it */
1209             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1210             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1211             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1212
1213             rinv00           = avx256_invsqrt_d(rsq00);
1214             rinv10           = avx256_invsqrt_d(rsq10);
1215             rinv20           = avx256_invsqrt_d(rsq20);
1216
1217             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1218             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1219             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1220
1221             /* Load parameters for j particles */
1222             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1223                                                                  charge+jnrC+0,charge+jnrD+0);
1224             vdwjidx0A        = 2*vdwtype[jnrA+0];
1225             vdwjidx0B        = 2*vdwtype[jnrB+0];
1226             vdwjidx0C        = 2*vdwtype[jnrC+0];
1227             vdwjidx0D        = 2*vdwtype[jnrD+0];
1228
1229             fjx0             = _mm256_setzero_pd();
1230             fjy0             = _mm256_setzero_pd();
1231             fjz0             = _mm256_setzero_pd();
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1238             {
1239
1240             r00              = _mm256_mul_pd(rsq00,rinv00);
1241             r00              = _mm256_andnot_pd(dummy_mask,r00);
1242
1243             /* Compute parameters for interactions between i and j atoms */
1244             qq00             = _mm256_mul_pd(iq0,jq0);
1245             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1246                                             vdwioffsetptr0+vdwjidx0B,
1247                                             vdwioffsetptr0+vdwjidx0C,
1248                                             vdwioffsetptr0+vdwjidx0D,
1249                                             &c6_00,&c12_00);
1250
1251             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1252                                                                   vdwgridioffsetptr0+vdwjidx0B,
1253                                                                   vdwgridioffsetptr0+vdwjidx0C,
1254                                                                   vdwgridioffsetptr0+vdwjidx0D);
1255
1256             /* EWALD ELECTROSTATICS */
1257
1258             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1259             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1260             ewitab           = _mm256_cvttpd_epi32(ewrt);
1261             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1262             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1263                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1264                                             &ewtabF,&ewtabFn);
1265             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1266             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1267
1268             /* Analytical LJ-PME */
1269             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1270             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1271             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1272             exponent         = avx256_exp_d(ewcljrsq);
1273             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1274             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1275             /* f6A = 6 * C6grid * (1 - poly) */
1276             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1277             /* f6B = C6grid * exponent * beta^6 */
1278             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1279             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1280             fvdw              = _mm256_mul_pd(_mm256_add_pd(_mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),_mm256_sub_pd(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1281
1282             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1283
1284             fscal            = _mm256_add_pd(felec,fvdw);
1285
1286             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1287
1288             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1289
1290             /* Calculate temporary vectorial force */
1291             tx               = _mm256_mul_pd(fscal,dx00);
1292             ty               = _mm256_mul_pd(fscal,dy00);
1293             tz               = _mm256_mul_pd(fscal,dz00);
1294
1295             /* Update vectorial force */
1296             fix0             = _mm256_add_pd(fix0,tx);
1297             fiy0             = _mm256_add_pd(fiy0,ty);
1298             fiz0             = _mm256_add_pd(fiz0,tz);
1299
1300             fjx0             = _mm256_add_pd(fjx0,tx);
1301             fjy0             = _mm256_add_pd(fjy0,ty);
1302             fjz0             = _mm256_add_pd(fjz0,tz);
1303
1304             }
1305
1306             /**************************
1307              * CALCULATE INTERACTIONS *
1308              **************************/
1309
1310             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1311             {
1312
1313             r10              = _mm256_mul_pd(rsq10,rinv10);
1314             r10              = _mm256_andnot_pd(dummy_mask,r10);
1315
1316             /* Compute parameters for interactions between i and j atoms */
1317             qq10             = _mm256_mul_pd(iq1,jq0);
1318
1319             /* EWALD ELECTROSTATICS */
1320
1321             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1322             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1323             ewitab           = _mm256_cvttpd_epi32(ewrt);
1324             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1325             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1326                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1327                                             &ewtabF,&ewtabFn);
1328             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1329             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1330
1331             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1332
1333             fscal            = felec;
1334
1335             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1336
1337             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1338
1339             /* Calculate temporary vectorial force */
1340             tx               = _mm256_mul_pd(fscal,dx10);
1341             ty               = _mm256_mul_pd(fscal,dy10);
1342             tz               = _mm256_mul_pd(fscal,dz10);
1343
1344             /* Update vectorial force */
1345             fix1             = _mm256_add_pd(fix1,tx);
1346             fiy1             = _mm256_add_pd(fiy1,ty);
1347             fiz1             = _mm256_add_pd(fiz1,tz);
1348
1349             fjx0             = _mm256_add_pd(fjx0,tx);
1350             fjy0             = _mm256_add_pd(fjy0,ty);
1351             fjz0             = _mm256_add_pd(fjz0,tz);
1352
1353             }
1354
1355             /**************************
1356              * CALCULATE INTERACTIONS *
1357              **************************/
1358
1359             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1360             {
1361
1362             r20              = _mm256_mul_pd(rsq20,rinv20);
1363             r20              = _mm256_andnot_pd(dummy_mask,r20);
1364
1365             /* Compute parameters for interactions between i and j atoms */
1366             qq20             = _mm256_mul_pd(iq2,jq0);
1367
1368             /* EWALD ELECTROSTATICS */
1369
1370             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1371             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1372             ewitab           = _mm256_cvttpd_epi32(ewrt);
1373             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1374             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1375                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1376                                             &ewtabF,&ewtabFn);
1377             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1378             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1379
1380             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1381
1382             fscal            = felec;
1383
1384             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1385
1386             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1387
1388             /* Calculate temporary vectorial force */
1389             tx               = _mm256_mul_pd(fscal,dx20);
1390             ty               = _mm256_mul_pd(fscal,dy20);
1391             tz               = _mm256_mul_pd(fscal,dz20);
1392
1393             /* Update vectorial force */
1394             fix2             = _mm256_add_pd(fix2,tx);
1395             fiy2             = _mm256_add_pd(fiy2,ty);
1396             fiz2             = _mm256_add_pd(fiz2,tz);
1397
1398             fjx0             = _mm256_add_pd(fjx0,tx);
1399             fjy0             = _mm256_add_pd(fjy0,ty);
1400             fjz0             = _mm256_add_pd(fjz0,tz);
1401
1402             }
1403
1404             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1405             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1406             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1407             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1408
1409             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1410
1411             /* Inner loop uses 146 flops */
1412         }
1413
1414         /* End of innermost loop */
1415
1416         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1417                                                  f+i_coord_offset,fshift+i_shift_offset);
1418
1419         /* Increment number of inner iterations */
1420         inneriter                  += j_index_end - j_index_start;
1421
1422         /* Outer loop uses 18 flops */
1423     }
1424
1425     /* Increment number of outer iterations */
1426     outeriter        += nri;
1427
1428     /* Update outer/inner flops */
1429
1430     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*146);
1431 }