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, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "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_VdwLJEwSh_GeomW3P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     real *           vdwgridioffsetptr0;
  88     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  89     real *           vdwioffsetptr1;
  90     real *           vdwgridioffsetptr1;
  91     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     real *           vdwgridioffsetptr2;
  94     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  95     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  96     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  97     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  98     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  99     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 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     __m256d           c6grid_00;
 109     __m256d           c6grid_10;
 110     __m256d           c6grid_20;
 111     real             *vdwgridparam;
 112     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 113     __m256d           one_half  = _mm256_set1_pd(0.5);
 114     __m256d           minus_one = _mm256_set1_pd(-1.0);
 115     __m128i          ewitab;
 116     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 117     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 118     real             *ewtab;
 119     __m256d          dummy_mask,cutoff_mask;
 120     __m128           tmpmask0,tmpmask1;
 121     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 122     __m256d          one     = _mm256_set1_pd(1.0);
 123     __m256d          two     = _mm256_set1_pd(2.0);
 124     x                = xx[0];
 125     f                = ff[0];
 126
 127     nri              = nlist->nri;
 128     iinr             = nlist->iinr;
 129     jindex           = nlist->jindex;
 130     jjnr             = nlist->jjnr;
 131     shiftidx         = nlist->shift;
 132     gid              = nlist->gid;
 133     shiftvec         = fr->shift_vec[0];
 134     fshift           = fr->fshift[0];
 135     facel            = _mm256_set1_pd(fr->epsfac);
 136     charge           = mdatoms->chargeA;
 137     nvdwtype         = fr->ntype;
 138     vdwparam         = fr->nbfp;
 139     vdwtype          = mdatoms->typeA;
 140     vdwgridparam     = fr->ljpme_c6grid;
 141     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 142     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 143     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 144
 145     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 146     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 147     beta2            = _mm256_mul_pd(beta,beta);
 148     beta3            = _mm256_mul_pd(beta,beta2);
 149
 150     ewtab            = fr->ic->tabq_coul_FDV0;
 151     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 152     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 153
 154     /* Setup water-specific parameters */
 155     inr              = nlist->iinr[0];
 156     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 157     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 158     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 159     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 160     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 161
 162     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 163     rcutoff_scalar   = fr->rcoulomb;
 164     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 165     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 166
 167     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 168     rvdw             = _mm256_set1_pd(fr->rvdw);
 169
 170     /* Avoid stupid compiler warnings */
 171     jnrA = jnrB = jnrC = jnrD = 0;
 172     j_coord_offsetA = 0;
 173     j_coord_offsetB = 0;
 174     j_coord_offsetC = 0;
 175     j_coord_offsetD = 0;
 176
 177     outeriter        = 0;
 178     inneriter        = 0;
 179
 180     for(iidx=0;iidx<4*DIM;iidx++)
 181     {
 182         scratch[iidx] = 0.0;
 183     }
 184
 185     /* Start outer loop over neighborlists */
 186     for(iidx=0; iidx<nri; iidx++)
 187     {
 188         /* Load shift vector for this list */
 189         i_shift_offset   = DIM*shiftidx[iidx];
 190
 191         /* Load limits for loop over neighbors */
 192         j_index_start    = jindex[iidx];
 193         j_index_end      = jindex[iidx+1];
 194
 195         /* Get outer coordinate index */
 196         inr              = iinr[iidx];
 197         i_coord_offset   = DIM*inr;
 198
 199         /* Load i particle coords and add shift vector */
 200         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 201                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 202
 203         fix0             = _mm256_setzero_pd();
 204         fiy0             = _mm256_setzero_pd();
 205         fiz0             = _mm256_setzero_pd();
 206         fix1             = _mm256_setzero_pd();
 207         fiy1             = _mm256_setzero_pd();
 208         fiz1             = _mm256_setzero_pd();
 209         fix2             = _mm256_setzero_pd();
 210         fiy2             = _mm256_setzero_pd();
 211         fiz2             = _mm256_setzero_pd();
 212
 213         /* Reset potential sums */
 214         velecsum         = _mm256_setzero_pd();
 215         vvdwsum          = _mm256_setzero_pd();
 216
 217         /* Start inner kernel loop */
 218         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 219         {
 220
 221             /* Get j neighbor index, and coordinate index */
 222             jnrA             = jjnr[jidx];
 223             jnrB             = jjnr[jidx+1];
 224             jnrC             = jjnr[jidx+2];
 225             jnrD             = jjnr[jidx+3];
 226             j_coord_offsetA  = DIM*jnrA;
 227             j_coord_offsetB  = DIM*jnrB;
 228             j_coord_offsetC  = DIM*jnrC;
 229             j_coord_offsetD  = DIM*jnrD;
 230
 231             /* load j atom coordinates */
 232             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 233                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 234                                                  &jx0,&jy0,&jz0);
 235
 236             /* Calculate displacement vector */
 237             dx00             = _mm256_sub_pd(ix0,jx0);
 238             dy00             = _mm256_sub_pd(iy0,jy0);
 239             dz00             = _mm256_sub_pd(iz0,jz0);
 240             dx10             = _mm256_sub_pd(ix1,jx0);
 241             dy10             = _mm256_sub_pd(iy1,jy0);
 242             dz10             = _mm256_sub_pd(iz1,jz0);
 243             dx20             = _mm256_sub_pd(ix2,jx0);
 244             dy20             = _mm256_sub_pd(iy2,jy0);
 245             dz20             = _mm256_sub_pd(iz2,jz0);
 246
 247             /* Calculate squared distance and things based on it */
 248             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 249             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 250             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 251
 252             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 253             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 254             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 255
 256             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 257             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 258             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 259
 260             /* Load parameters for j particles */
 261             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 262                                                                  charge+jnrC+0,charge+jnrD+0);
 263             vdwjidx0A        = 2*vdwtype[jnrA+0];
 264             vdwjidx0B        = 2*vdwtype[jnrB+0];
 265             vdwjidx0C        = 2*vdwtype[jnrC+0];
 266             vdwjidx0D        = 2*vdwtype[jnrD+0];
 267
 268             fjx0             = _mm256_setzero_pd();
 269             fjy0             = _mm256_setzero_pd();
 270             fjz0             = _mm256_setzero_pd();
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 277             {
 278
 279             r00              = _mm256_mul_pd(rsq00,rinv00);
 280
 281             /* Compute parameters for interactions between i and j atoms */
 282             qq00             = _mm256_mul_pd(iq0,jq0);
 283             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 284                                             vdwioffsetptr0+vdwjidx0B,
 285                                             vdwioffsetptr0+vdwjidx0C,
 286                                             vdwioffsetptr0+vdwjidx0D,
 287                                             &c6_00,&c12_00);
 288
 289             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 290                                                                   vdwgridioffsetptr0+vdwjidx0B,
 291                                                                   vdwgridioffsetptr0+vdwjidx0C,
 292                                                                   vdwgridioffsetptr0+vdwjidx0D);
 293
 294             /* EWALD ELECTROSTATICS */
 295
 296             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 297             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 298             ewitab           = _mm256_cvttpd_epi32(ewrt);
 299             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 300             ewitab           = _mm_slli_epi32(ewitab,2);
 301             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 302             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 303             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 304             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 305             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 306             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 307             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 308             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 309             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 310
 311             /* Analytical LJ-PME */
 312             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 313             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 314             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 315             exponent         = gmx_simd_exp_d(ewcljrsq);
 316             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 317             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 318             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 319             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 320             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 321             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) ,
 322                                           _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));
 323             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 324             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);
 325
 326             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velec            = _mm256_and_pd(velec,cutoff_mask);
 330             velecsum         = _mm256_add_pd(velecsum,velec);
 331             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 332             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 333
 334             fscal            = _mm256_add_pd(felec,fvdw);
 335
 336             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm256_mul_pd(fscal,dx00);
 340             ty               = _mm256_mul_pd(fscal,dy00);
 341             tz               = _mm256_mul_pd(fscal,dz00);
 342
 343             /* Update vectorial force */
 344             fix0             = _mm256_add_pd(fix0,tx);
 345             fiy0             = _mm256_add_pd(fiy0,ty);
 346             fiz0             = _mm256_add_pd(fiz0,tz);
 347
 348             fjx0             = _mm256_add_pd(fjx0,tx);
 349             fjy0             = _mm256_add_pd(fjy0,ty);
 350             fjz0             = _mm256_add_pd(fjz0,tz);
 351
 352             }
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 359             {
 360
 361             r10              = _mm256_mul_pd(rsq10,rinv10);
 362
 363             /* Compute parameters for interactions between i and j atoms */
 364             qq10             = _mm256_mul_pd(iq1,jq0);
 365
 366             /* EWALD ELECTROSTATICS */
 367
 368             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 369             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 370             ewitab           = _mm256_cvttpd_epi32(ewrt);
 371             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 372             ewitab           = _mm_slli_epi32(ewitab,2);
 373             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 374             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 375             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 376             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 377             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 378             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 379             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 380             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 381             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 382
 383             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 384
 385             /* Update potential sum for this i atom from the interaction with this j atom. */
 386             velec            = _mm256_and_pd(velec,cutoff_mask);
 387             velecsum         = _mm256_add_pd(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm256_mul_pd(fscal,dx10);
 395             ty               = _mm256_mul_pd(fscal,dy10);
 396             tz               = _mm256_mul_pd(fscal,dz10);
 397
 398             /* Update vectorial force */
 399             fix1             = _mm256_add_pd(fix1,tx);
 400             fiy1             = _mm256_add_pd(fiy1,ty);
 401             fiz1             = _mm256_add_pd(fiz1,tz);
 402
 403             fjx0             = _mm256_add_pd(fjx0,tx);
 404             fjy0             = _mm256_add_pd(fjy0,ty);
 405             fjz0             = _mm256_add_pd(fjz0,tz);
 406
 407             }
 408
 409             /**************************
 410              * CALCULATE INTERACTIONS *
 411              **************************/
 412
 413             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 414             {
 415
 416             r20              = _mm256_mul_pd(rsq20,rinv20);
 417
 418             /* Compute parameters for interactions between i and j atoms */
 419             qq20             = _mm256_mul_pd(iq2,jq0);
 420
 421             /* EWALD ELECTROSTATICS */
 422
 423             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 424             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 425             ewitab           = _mm256_cvttpd_epi32(ewrt);
 426             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 427             ewitab           = _mm_slli_epi32(ewitab,2);
 428             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 429             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 430             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 431             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 432             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 433             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 434             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 435             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 436             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 437
 438             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 439
 440             /* Update potential sum for this i atom from the interaction with this j atom. */
 441             velec            = _mm256_and_pd(velec,cutoff_mask);
 442             velecsum         = _mm256_add_pd(velecsum,velec);
 443
 444             fscal            = felec;
 445
 446             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 447
 448             /* Calculate temporary vectorial force */
 449             tx               = _mm256_mul_pd(fscal,dx20);
 450             ty               = _mm256_mul_pd(fscal,dy20);
 451             tz               = _mm256_mul_pd(fscal,dz20);
 452
 453             /* Update vectorial force */
 454             fix2             = _mm256_add_pd(fix2,tx);
 455             fiy2             = _mm256_add_pd(fiy2,ty);
 456             fiz2             = _mm256_add_pd(fiz2,tz);
 457
 458             fjx0             = _mm256_add_pd(fjx0,tx);
 459             fjy0             = _mm256_add_pd(fjy0,ty);
 460             fjz0             = _mm256_add_pd(fjz0,tz);
 461
 462             }
 463
 464             fjptrA             = f+j_coord_offsetA;
 465             fjptrB             = f+j_coord_offsetB;
 466             fjptrC             = f+j_coord_offsetC;
 467             fjptrD             = f+j_coord_offsetD;
 468
 469             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 470
 471             /* Inner loop uses 177 flops */
 472         }
 473
 474         if(jidx<j_index_end)
 475         {
 476
 477             /* Get j neighbor index, and coordinate index */
 478             jnrlistA         = jjnr[jidx];
 479             jnrlistB         = jjnr[jidx+1];
 480             jnrlistC         = jjnr[jidx+2];
 481             jnrlistD         = jjnr[jidx+3];
 482             /* Sign of each element will be negative for non-real atoms.
 483              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 484              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 485              */
 486             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 487
 488             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 489             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 490             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 491
 492             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 493             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 494             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 495             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 496             j_coord_offsetA  = DIM*jnrA;
 497             j_coord_offsetB  = DIM*jnrB;
 498             j_coord_offsetC  = DIM*jnrC;
 499             j_coord_offsetD  = DIM*jnrD;
 500
 501             /* load j atom coordinates */
 502             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 503                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 504                                                  &jx0,&jy0,&jz0);
 505
 506             /* Calculate displacement vector */
 507             dx00             = _mm256_sub_pd(ix0,jx0);
 508             dy00             = _mm256_sub_pd(iy0,jy0);
 509             dz00             = _mm256_sub_pd(iz0,jz0);
 510             dx10             = _mm256_sub_pd(ix1,jx0);
 511             dy10             = _mm256_sub_pd(iy1,jy0);
 512             dz10             = _mm256_sub_pd(iz1,jz0);
 513             dx20             = _mm256_sub_pd(ix2,jx0);
 514             dy20             = _mm256_sub_pd(iy2,jy0);
 515             dz20             = _mm256_sub_pd(iz2,jz0);
 516
 517             /* Calculate squared distance and things based on it */
 518             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 519             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 520             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 521
 522             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 523             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 524             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 525
 526             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 527             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 528             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 529
 530             /* Load parameters for j particles */
 531             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 532                                                                  charge+jnrC+0,charge+jnrD+0);
 533             vdwjidx0A        = 2*vdwtype[jnrA+0];
 534             vdwjidx0B        = 2*vdwtype[jnrB+0];
 535             vdwjidx0C        = 2*vdwtype[jnrC+0];
 536             vdwjidx0D        = 2*vdwtype[jnrD+0];
 537
 538             fjx0             = _mm256_setzero_pd();
 539             fjy0             = _mm256_setzero_pd();
 540             fjz0             = _mm256_setzero_pd();
 541
 542             /**************************
 543              * CALCULATE INTERACTIONS *
 544              **************************/
 545
 546             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 547             {
 548
 549             r00              = _mm256_mul_pd(rsq00,rinv00);
 550             r00              = _mm256_andnot_pd(dummy_mask,r00);
 551
 552             /* Compute parameters for interactions between i and j atoms */
 553             qq00             = _mm256_mul_pd(iq0,jq0);
 554             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 555                                             vdwioffsetptr0+vdwjidx0B,
 556                                             vdwioffsetptr0+vdwjidx0C,
 557                                             vdwioffsetptr0+vdwjidx0D,
 558                                             &c6_00,&c12_00);
 559
 560             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 561                                                                   vdwgridioffsetptr0+vdwjidx0B,
 562                                                                   vdwgridioffsetptr0+vdwjidx0C,
 563                                                                   vdwgridioffsetptr0+vdwjidx0D);
 564
 565             /* EWALD ELECTROSTATICS */
 566
 567             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 568             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 569             ewitab           = _mm256_cvttpd_epi32(ewrt);
 570             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 571             ewitab           = _mm_slli_epi32(ewitab,2);
 572             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 573             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 574             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 575             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 576             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 577             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 578             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 579             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(_mm256_sub_pd(rinv00,sh_ewald),velec));
 580             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 581
 582             /* Analytical LJ-PME */
 583             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 584             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 585             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 586             exponent         = gmx_simd_exp_d(ewcljrsq);
 587             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 588             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 589             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 590             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 591             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 592             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) ,
 593                                           _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));
 594             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 595             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);
 596
 597             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 598
 599             /* Update potential sum for this i atom from the interaction with this j atom. */
 600             velec            = _mm256_and_pd(velec,cutoff_mask);
 601             velec            = _mm256_andnot_pd(dummy_mask,velec);
 602             velecsum         = _mm256_add_pd(velecsum,velec);
 603             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 604             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 605             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 606
 607             fscal            = _mm256_add_pd(felec,fvdw);
 608
 609             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 610
 611             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm256_mul_pd(fscal,dx00);
 615             ty               = _mm256_mul_pd(fscal,dy00);
 616             tz               = _mm256_mul_pd(fscal,dz00);
 617
 618             /* Update vectorial force */
 619             fix0             = _mm256_add_pd(fix0,tx);
 620             fiy0             = _mm256_add_pd(fiy0,ty);
 621             fiz0             = _mm256_add_pd(fiz0,tz);
 622
 623             fjx0             = _mm256_add_pd(fjx0,tx);
 624             fjy0             = _mm256_add_pd(fjy0,ty);
 625             fjz0             = _mm256_add_pd(fjz0,tz);
 626
 627             }
 628
 629             /**************************
 630              * CALCULATE INTERACTIONS *
 631              **************************/
 632
 633             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 634             {
 635
 636             r10              = _mm256_mul_pd(rsq10,rinv10);
 637             r10              = _mm256_andnot_pd(dummy_mask,r10);
 638
 639             /* Compute parameters for interactions between i and j atoms */
 640             qq10             = _mm256_mul_pd(iq1,jq0);
 641
 642             /* EWALD ELECTROSTATICS */
 643
 644             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 645             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 646             ewitab           = _mm256_cvttpd_epi32(ewrt);
 647             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 648             ewitab           = _mm_slli_epi32(ewitab,2);
 649             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 650             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 651             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 652             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 653             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 654             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 655             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 656             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 657             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 658
 659             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 660
 661             /* Update potential sum for this i atom from the interaction with this j atom. */
 662             velec            = _mm256_and_pd(velec,cutoff_mask);
 663             velec            = _mm256_andnot_pd(dummy_mask,velec);
 664             velecsum         = _mm256_add_pd(velecsum,velec);
 665
 666             fscal            = felec;
 667
 668             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 669
 670             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 671
 672             /* Calculate temporary vectorial force */
 673             tx               = _mm256_mul_pd(fscal,dx10);
 674             ty               = _mm256_mul_pd(fscal,dy10);
 675             tz               = _mm256_mul_pd(fscal,dz10);
 676
 677             /* Update vectorial force */
 678             fix1             = _mm256_add_pd(fix1,tx);
 679             fiy1             = _mm256_add_pd(fiy1,ty);
 680             fiz1             = _mm256_add_pd(fiz1,tz);
 681
 682             fjx0             = _mm256_add_pd(fjx0,tx);
 683             fjy0             = _mm256_add_pd(fjy0,ty);
 684             fjz0             = _mm256_add_pd(fjz0,tz);
 685
 686             }
 687
 688             /**************************
 689              * CALCULATE INTERACTIONS *
 690              **************************/
 691
 692             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 693             {
 694
 695             r20              = _mm256_mul_pd(rsq20,rinv20);
 696             r20              = _mm256_andnot_pd(dummy_mask,r20);
 697
 698             /* Compute parameters for interactions between i and j atoms */
 699             qq20             = _mm256_mul_pd(iq2,jq0);
 700
 701             /* EWALD ELECTROSTATICS */
 702
 703             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 704             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 705             ewitab           = _mm256_cvttpd_epi32(ewrt);
 706             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 707             ewitab           = _mm_slli_epi32(ewitab,2);
 708             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 709             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 710             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 711             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 712             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 713             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 714             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 715             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 716             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 717
 718             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 719
 720             /* Update potential sum for this i atom from the interaction with this j atom. */
 721             velec            = _mm256_and_pd(velec,cutoff_mask);
 722             velec            = _mm256_andnot_pd(dummy_mask,velec);
 723             velecsum         = _mm256_add_pd(velecsum,velec);
 724
 725             fscal            = felec;
 726
 727             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 728
 729             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 730
 731             /* Calculate temporary vectorial force */
 732             tx               = _mm256_mul_pd(fscal,dx20);
 733             ty               = _mm256_mul_pd(fscal,dy20);
 734             tz               = _mm256_mul_pd(fscal,dz20);
 735
 736             /* Update vectorial force */
 737             fix2             = _mm256_add_pd(fix2,tx);
 738             fiy2             = _mm256_add_pd(fiy2,ty);
 739             fiz2             = _mm256_add_pd(fiz2,tz);
 740
 741             fjx0             = _mm256_add_pd(fjx0,tx);
 742             fjy0             = _mm256_add_pd(fjy0,ty);
 743             fjz0             = _mm256_add_pd(fjz0,tz);
 744
 745             }
 746
 747             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 748             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 749             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 750             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 751
 752             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 753
 754             /* Inner loop uses 180 flops */
 755         }
 756
 757         /* End of innermost loop */
 758
 759         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 760                                                  f+i_coord_offset,fshift+i_shift_offset);
 761
 762         ggid                        = gid[iidx];
 763         /* Update potential energies */
 764         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 765         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 766
 767         /* Increment number of inner iterations */
 768         inneriter                  += j_index_end - j_index_start;
 769
 770         /* Outer loop uses 20 flops */
 771     }
 772
 773     /* Increment number of outer iterations */
 774     outeriter        += nri;
 775
 776     /* Update outer/inner flops */
 777
 778     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*180);
 779 }
 780 /*
 781  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_double
 782  * Electrostatics interaction: Ewald
 783  * VdW interaction:            LJEwald
 784  * Geometry:                   Water3-Particle
 785  * Calculate force/pot:        Force
 786  */
 787 void
 788 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW3P1_F_avx_256_double
 789                     (t_nblist                    * gmx_restrict       nlist,
 790                      rvec                        * gmx_restrict          xx,
 791                      rvec                        * gmx_restrict          ff,
 792                      t_forcerec                  * gmx_restrict          fr,
 793                      t_mdatoms                   * gmx_restrict     mdatoms,
 794                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 795                      t_nrnb                      * gmx_restrict        nrnb)
 796 {
 797     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 798      * just 0 for non-waters.
 799      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 800      * jnr indices corresponding to data put in the four positions in the SIMD register.
 801      */
 802     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 803     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 804     int              jnrA,jnrB,jnrC,jnrD;
 805     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 806     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 807     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 808     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 809     real             rcutoff_scalar;
 810     real             *shiftvec,*fshift,*x,*f;
 811     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 812     real             scratch[4*DIM];
 813     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 814     real *           vdwioffsetptr0;
 815     real *           vdwgridioffsetptr0;
 816     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 817     real *           vdwioffsetptr1;
 818     real *           vdwgridioffsetptr1;
 819     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 820     real *           vdwioffsetptr2;
 821     real *           vdwgridioffsetptr2;
 822     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 823     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 824     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 825     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 826     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 827     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 828     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 829     real             *charge;
 830     int              nvdwtype;
 831     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 832     int              *vdwtype;
 833     real             *vdwparam;
 834     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 835     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 836     __m256d           c6grid_00;
 837     __m256d           c6grid_10;
 838     __m256d           c6grid_20;
 839     real             *vdwgridparam;
 840     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 841     __m256d           one_half  = _mm256_set1_pd(0.5);
 842     __m256d           minus_one = _mm256_set1_pd(-1.0);
 843     __m128i          ewitab;
 844     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 845     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 846     real             *ewtab;
 847     __m256d          dummy_mask,cutoff_mask;
 848     __m128           tmpmask0,tmpmask1;
 849     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 850     __m256d          one     = _mm256_set1_pd(1.0);
 851     __m256d          two     = _mm256_set1_pd(2.0);
 852     x                = xx[0];
 853     f                = ff[0];
 854
 855     nri              = nlist->nri;
 856     iinr             = nlist->iinr;
 857     jindex           = nlist->jindex;
 858     jjnr             = nlist->jjnr;
 859     shiftidx         = nlist->shift;
 860     gid              = nlist->gid;
 861     shiftvec         = fr->shift_vec[0];
 862     fshift           = fr->fshift[0];
 863     facel            = _mm256_set1_pd(fr->epsfac);
 864     charge           = mdatoms->chargeA;
 865     nvdwtype         = fr->ntype;
 866     vdwparam         = fr->nbfp;
 867     vdwtype          = mdatoms->typeA;
 868     vdwgridparam     = fr->ljpme_c6grid;
 869     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 870     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 871     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 872
 873     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 874     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 875     beta2            = _mm256_mul_pd(beta,beta);
 876     beta3            = _mm256_mul_pd(beta,beta2);
 877
 878     ewtab            = fr->ic->tabq_coul_F;
 879     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 880     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 881
 882     /* Setup water-specific parameters */
 883     inr              = nlist->iinr[0];
 884     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 885     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 886     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 887     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 888     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 889
 890     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 891     rcutoff_scalar   = fr->rcoulomb;
 892     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 893     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 894
 895     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 896     rvdw             = _mm256_set1_pd(fr->rvdw);
 897
 898     /* Avoid stupid compiler warnings */
 899     jnrA = jnrB = jnrC = jnrD = 0;
 900     j_coord_offsetA = 0;
 901     j_coord_offsetB = 0;
 902     j_coord_offsetC = 0;
 903     j_coord_offsetD = 0;
 904
 905     outeriter        = 0;
 906     inneriter        = 0;
 907
 908     for(iidx=0;iidx<4*DIM;iidx++)
 909     {
 910         scratch[iidx] = 0.0;
 911     }
 912
 913     /* Start outer loop over neighborlists */
 914     for(iidx=0; iidx<nri; iidx++)
 915     {
 916         /* Load shift vector for this list */
 917         i_shift_offset   = DIM*shiftidx[iidx];
 918
 919         /* Load limits for loop over neighbors */
 920         j_index_start    = jindex[iidx];
 921         j_index_end      = jindex[iidx+1];
 922
 923         /* Get outer coordinate index */
 924         inr              = iinr[iidx];
 925         i_coord_offset   = DIM*inr;
 926
 927         /* Load i particle coords and add shift vector */
 928         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 929                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 930
 931         fix0             = _mm256_setzero_pd();
 932         fiy0             = _mm256_setzero_pd();
 933         fiz0             = _mm256_setzero_pd();
 934         fix1             = _mm256_setzero_pd();
 935         fiy1             = _mm256_setzero_pd();
 936         fiz1             = _mm256_setzero_pd();
 937         fix2             = _mm256_setzero_pd();
 938         fiy2             = _mm256_setzero_pd();
 939         fiz2             = _mm256_setzero_pd();
 940
 941         /* Start inner kernel loop */
 942         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 943         {
 944
 945             /* Get j neighbor index, and coordinate index */
 946             jnrA             = jjnr[jidx];
 947             jnrB             = jjnr[jidx+1];
 948             jnrC             = jjnr[jidx+2];
 949             jnrD             = jjnr[jidx+3];
 950             j_coord_offsetA  = DIM*jnrA;
 951             j_coord_offsetB  = DIM*jnrB;
 952             j_coord_offsetC  = DIM*jnrC;
 953             j_coord_offsetD  = DIM*jnrD;
 954
 955             /* load j atom coordinates */
 956             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 957                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 958                                                  &jx0,&jy0,&jz0);
 959
 960             /* Calculate displacement vector */
 961             dx00             = _mm256_sub_pd(ix0,jx0);
 962             dy00             = _mm256_sub_pd(iy0,jy0);
 963             dz00             = _mm256_sub_pd(iz0,jz0);
 964             dx10             = _mm256_sub_pd(ix1,jx0);
 965             dy10             = _mm256_sub_pd(iy1,jy0);
 966             dz10             = _mm256_sub_pd(iz1,jz0);
 967             dx20             = _mm256_sub_pd(ix2,jx0);
 968             dy20             = _mm256_sub_pd(iy2,jy0);
 969             dz20             = _mm256_sub_pd(iz2,jz0);
 970
 971             /* Calculate squared distance and things based on it */
 972             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 973             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 974             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 975
 976             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 977             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 978             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 979
 980             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 981             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 982             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 983
 984             /* Load parameters for j particles */
 985             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 986                                                                  charge+jnrC+0,charge+jnrD+0);
 987             vdwjidx0A        = 2*vdwtype[jnrA+0];
 988             vdwjidx0B        = 2*vdwtype[jnrB+0];
 989             vdwjidx0C        = 2*vdwtype[jnrC+0];
 990             vdwjidx0D        = 2*vdwtype[jnrD+0];
 991
 992             fjx0             = _mm256_setzero_pd();
 993             fjy0             = _mm256_setzero_pd();
 994             fjz0             = _mm256_setzero_pd();
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1001             {
1002
1003             r00              = _mm256_mul_pd(rsq00,rinv00);
1004
1005             /* Compute parameters for interactions between i and j atoms */
1006             qq00             = _mm256_mul_pd(iq0,jq0);
1007             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1008                                             vdwioffsetptr0+vdwjidx0B,
1009                                             vdwioffsetptr0+vdwjidx0C,
1010                                             vdwioffsetptr0+vdwjidx0D,
1011                                             &c6_00,&c12_00);
1012
1013             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1014                                                                   vdwgridioffsetptr0+vdwjidx0B,
1015                                                                   vdwgridioffsetptr0+vdwjidx0C,
1016                                                                   vdwgridioffsetptr0+vdwjidx0D);
1017
1018             /* EWALD ELECTROSTATICS */
1019
1020             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1021             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1022             ewitab           = _mm256_cvttpd_epi32(ewrt);
1023             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1024             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1025                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1026                                             &ewtabF,&ewtabFn);
1027             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1028             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1029
1030             /* Analytical LJ-PME */
1031             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1032             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1033             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1034             exponent         = gmx_simd_exp_d(ewcljrsq);
1035             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1036             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1037             /* f6A = 6 * C6grid * (1 - poly) */
1038             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1039             /* f6B = C6grid * exponent * beta^6 */
1040             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1041             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1042             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);
1043
1044             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1045
1046             fscal            = _mm256_add_pd(felec,fvdw);
1047
1048             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1049
1050             /* Calculate temporary vectorial force */
1051             tx               = _mm256_mul_pd(fscal,dx00);
1052             ty               = _mm256_mul_pd(fscal,dy00);
1053             tz               = _mm256_mul_pd(fscal,dz00);
1054
1055             /* Update vectorial force */
1056             fix0             = _mm256_add_pd(fix0,tx);
1057             fiy0             = _mm256_add_pd(fiy0,ty);
1058             fiz0             = _mm256_add_pd(fiz0,tz);
1059
1060             fjx0             = _mm256_add_pd(fjx0,tx);
1061             fjy0             = _mm256_add_pd(fjy0,ty);
1062             fjz0             = _mm256_add_pd(fjz0,tz);
1063
1064             }
1065
1066             /**************************
1067              * CALCULATE INTERACTIONS *
1068              **************************/
1069
1070             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1071             {
1072
1073             r10              = _mm256_mul_pd(rsq10,rinv10);
1074
1075             /* Compute parameters for interactions between i and j atoms */
1076             qq10             = _mm256_mul_pd(iq1,jq0);
1077
1078             /* EWALD ELECTROSTATICS */
1079
1080             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1081             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1082             ewitab           = _mm256_cvttpd_epi32(ewrt);
1083             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1084             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1085                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1086                                             &ewtabF,&ewtabFn);
1087             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1088             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1089
1090             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1091
1092             fscal            = felec;
1093
1094             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1095
1096             /* Calculate temporary vectorial force */
1097             tx               = _mm256_mul_pd(fscal,dx10);
1098             ty               = _mm256_mul_pd(fscal,dy10);
1099             tz               = _mm256_mul_pd(fscal,dz10);
1100
1101             /* Update vectorial force */
1102             fix1             = _mm256_add_pd(fix1,tx);
1103             fiy1             = _mm256_add_pd(fiy1,ty);
1104             fiz1             = _mm256_add_pd(fiz1,tz);
1105
1106             fjx0             = _mm256_add_pd(fjx0,tx);
1107             fjy0             = _mm256_add_pd(fjy0,ty);
1108             fjz0             = _mm256_add_pd(fjz0,tz);
1109
1110             }
1111
1112             /**************************
1113              * CALCULATE INTERACTIONS *
1114              **************************/
1115
1116             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1117             {
1118
1119             r20              = _mm256_mul_pd(rsq20,rinv20);
1120
1121             /* Compute parameters for interactions between i and j atoms */
1122             qq20             = _mm256_mul_pd(iq2,jq0);
1123
1124             /* EWALD ELECTROSTATICS */
1125
1126             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1127             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1128             ewitab           = _mm256_cvttpd_epi32(ewrt);
1129             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1130             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1131                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1132                                             &ewtabF,&ewtabFn);
1133             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1134             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1135
1136             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1137
1138             fscal            = felec;
1139
1140             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1141
1142             /* Calculate temporary vectorial force */
1143             tx               = _mm256_mul_pd(fscal,dx20);
1144             ty               = _mm256_mul_pd(fscal,dy20);
1145             tz               = _mm256_mul_pd(fscal,dz20);
1146
1147             /* Update vectorial force */
1148             fix2             = _mm256_add_pd(fix2,tx);
1149             fiy2             = _mm256_add_pd(fiy2,ty);
1150             fiz2             = _mm256_add_pd(fiz2,tz);
1151
1152             fjx0             = _mm256_add_pd(fjx0,tx);
1153             fjy0             = _mm256_add_pd(fjy0,ty);
1154             fjz0             = _mm256_add_pd(fjz0,tz);
1155
1156             }
1157
1158             fjptrA             = f+j_coord_offsetA;
1159             fjptrB             = f+j_coord_offsetB;
1160             fjptrC             = f+j_coord_offsetC;
1161             fjptrD             = f+j_coord_offsetD;
1162
1163             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1164
1165             /* Inner loop uses 143 flops */
1166         }
1167
1168         if(jidx<j_index_end)
1169         {
1170
1171             /* Get j neighbor index, and coordinate index */
1172             jnrlistA         = jjnr[jidx];
1173             jnrlistB         = jjnr[jidx+1];
1174             jnrlistC         = jjnr[jidx+2];
1175             jnrlistD         = jjnr[jidx+3];
1176             /* Sign of each element will be negative for non-real atoms.
1177              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1178              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1179              */
1180             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1181
1182             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1183             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1184             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1185
1186             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1187             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1188             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1189             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1190             j_coord_offsetA  = DIM*jnrA;
1191             j_coord_offsetB  = DIM*jnrB;
1192             j_coord_offsetC  = DIM*jnrC;
1193             j_coord_offsetD  = DIM*jnrD;
1194
1195             /* load j atom coordinates */
1196             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1197                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1198                                                  &jx0,&jy0,&jz0);
1199
1200             /* Calculate displacement vector */
1201             dx00             = _mm256_sub_pd(ix0,jx0);
1202             dy00             = _mm256_sub_pd(iy0,jy0);
1203             dz00             = _mm256_sub_pd(iz0,jz0);
1204             dx10             = _mm256_sub_pd(ix1,jx0);
1205             dy10             = _mm256_sub_pd(iy1,jy0);
1206             dz10             = _mm256_sub_pd(iz1,jz0);
1207             dx20             = _mm256_sub_pd(ix2,jx0);
1208             dy20             = _mm256_sub_pd(iy2,jy0);
1209             dz20             = _mm256_sub_pd(iz2,jz0);
1210
1211             /* Calculate squared distance and things based on it */
1212             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1213             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1214             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1215
1216             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1217             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1218             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1219
1220             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1221             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1222             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1223
1224             /* Load parameters for j particles */
1225             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1226                                                                  charge+jnrC+0,charge+jnrD+0);
1227             vdwjidx0A        = 2*vdwtype[jnrA+0];
1228             vdwjidx0B        = 2*vdwtype[jnrB+0];
1229             vdwjidx0C        = 2*vdwtype[jnrC+0];
1230             vdwjidx0D        = 2*vdwtype[jnrD+0];
1231
1232             fjx0             = _mm256_setzero_pd();
1233             fjy0             = _mm256_setzero_pd();
1234             fjz0             = _mm256_setzero_pd();
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1241             {
1242
1243             r00              = _mm256_mul_pd(rsq00,rinv00);
1244             r00              = _mm256_andnot_pd(dummy_mask,r00);
1245
1246             /* Compute parameters for interactions between i and j atoms */
1247             qq00             = _mm256_mul_pd(iq0,jq0);
1248             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1249                                             vdwioffsetptr0+vdwjidx0B,
1250                                             vdwioffsetptr0+vdwjidx0C,
1251                                             vdwioffsetptr0+vdwjidx0D,
1252                                             &c6_00,&c12_00);
1253
1254             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1255                                                                   vdwgridioffsetptr0+vdwjidx0B,
1256                                                                   vdwgridioffsetptr0+vdwjidx0C,
1257                                                                   vdwgridioffsetptr0+vdwjidx0D);
1258
1259             /* EWALD ELECTROSTATICS */
1260
1261             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1262             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1263             ewitab           = _mm256_cvttpd_epi32(ewrt);
1264             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1265             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1266                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1267                                             &ewtabF,&ewtabFn);
1268             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1269             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1270
1271             /* Analytical LJ-PME */
1272             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1273             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1274             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1275             exponent         = gmx_simd_exp_d(ewcljrsq);
1276             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1277             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1278             /* f6A = 6 * C6grid * (1 - poly) */
1279             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1280             /* f6B = C6grid * exponent * beta^6 */
1281             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1282             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1283             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);
1284
1285             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1286
1287             fscal            = _mm256_add_pd(felec,fvdw);
1288
1289             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1290
1291             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1292
1293             /* Calculate temporary vectorial force */
1294             tx               = _mm256_mul_pd(fscal,dx00);
1295             ty               = _mm256_mul_pd(fscal,dy00);
1296             tz               = _mm256_mul_pd(fscal,dz00);
1297
1298             /* Update vectorial force */
1299             fix0             = _mm256_add_pd(fix0,tx);
1300             fiy0             = _mm256_add_pd(fiy0,ty);
1301             fiz0             = _mm256_add_pd(fiz0,tz);
1302
1303             fjx0             = _mm256_add_pd(fjx0,tx);
1304             fjy0             = _mm256_add_pd(fjy0,ty);
1305             fjz0             = _mm256_add_pd(fjz0,tz);
1306
1307             }
1308
1309             /**************************
1310              * CALCULATE INTERACTIONS *
1311              **************************/
1312
1313             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1314             {
1315
1316             r10              = _mm256_mul_pd(rsq10,rinv10);
1317             r10              = _mm256_andnot_pd(dummy_mask,r10);
1318
1319             /* Compute parameters for interactions between i and j atoms */
1320             qq10             = _mm256_mul_pd(iq1,jq0);
1321
1322             /* EWALD ELECTROSTATICS */
1323
1324             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1325             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1326             ewitab           = _mm256_cvttpd_epi32(ewrt);
1327             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1328             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1329                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1330                                             &ewtabF,&ewtabFn);
1331             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1332             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1333
1334             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1335
1336             fscal            = felec;
1337
1338             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1339
1340             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1341
1342             /* Calculate temporary vectorial force */
1343             tx               = _mm256_mul_pd(fscal,dx10);
1344             ty               = _mm256_mul_pd(fscal,dy10);
1345             tz               = _mm256_mul_pd(fscal,dz10);
1346
1347             /* Update vectorial force */
1348             fix1             = _mm256_add_pd(fix1,tx);
1349             fiy1             = _mm256_add_pd(fiy1,ty);
1350             fiz1             = _mm256_add_pd(fiz1,tz);
1351
1352             fjx0             = _mm256_add_pd(fjx0,tx);
1353             fjy0             = _mm256_add_pd(fjy0,ty);
1354             fjz0             = _mm256_add_pd(fjz0,tz);
1355
1356             }
1357
1358             /**************************
1359              * CALCULATE INTERACTIONS *
1360              **************************/
1361
1362             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1363             {
1364
1365             r20              = _mm256_mul_pd(rsq20,rinv20);
1366             r20              = _mm256_andnot_pd(dummy_mask,r20);
1367
1368             /* Compute parameters for interactions between i and j atoms */
1369             qq20             = _mm256_mul_pd(iq2,jq0);
1370
1371             /* EWALD ELECTROSTATICS */
1372
1373             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1374             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1375             ewitab           = _mm256_cvttpd_epi32(ewrt);
1376             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1377             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1378                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1379                                             &ewtabF,&ewtabFn);
1380             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1381             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1382
1383             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1384
1385             fscal            = felec;
1386
1387             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1388
1389             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1390
1391             /* Calculate temporary vectorial force */
1392             tx               = _mm256_mul_pd(fscal,dx20);
1393             ty               = _mm256_mul_pd(fscal,dy20);
1394             tz               = _mm256_mul_pd(fscal,dz20);
1395
1396             /* Update vectorial force */
1397             fix2             = _mm256_add_pd(fix2,tx);
1398             fiy2             = _mm256_add_pd(fiy2,ty);
1399             fiz2             = _mm256_add_pd(fiz2,tz);
1400
1401             fjx0             = _mm256_add_pd(fjx0,tx);
1402             fjy0             = _mm256_add_pd(fjy0,ty);
1403             fjz0             = _mm256_add_pd(fjz0,tz);
1404
1405             }
1406
1407             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1408             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1409             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1410             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1411
1412             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1413
1414             /* Inner loop uses 146 flops */
1415         }
1416
1417         /* End of innermost loop */
1418
1419         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1420                                                  f+i_coord_offset,fshift+i_shift_offset);
1421
1422         /* Increment number of inner iterations */
1423         inneriter                  += j_index_end - j_index_start;
1424
1425         /* Outer loop uses 18 flops */
1426     }
1427
1428     /* Increment number of outer iterations */
1429     outeriter        += nri;
1430
1431     /* Update outer/inner flops */
1432
1433     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*146);
1434 }