src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwLJEw_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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     real *           vdwgridioffsetptr0;
  86     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     real *           vdwgridioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     real *           vdwgridioffsetptr2;
  92     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m256d           c6grid_00;
 107     __m256d           c6grid_10;
 108     __m256d           c6grid_20;
 109     real             *vdwgridparam;
 110     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 111     __m256d           one_half  = _mm256_set1_pd(0.5);
 112     __m256d           minus_one = _mm256_set1_pd(-1.0);
 113     __m128i          ewitab;
 114     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 115     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 116     real             *ewtab;
 117     __m256d          dummy_mask,cutoff_mask;
 118     __m128           tmpmask0,tmpmask1;
 119     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 120     __m256d          one     = _mm256_set1_pd(1.0);
 121     __m256d          two     = _mm256_set1_pd(2.0);
 122     x                = xx[0];
 123     f                = ff[0];
 124
 125     nri              = nlist->nri;
 126     iinr             = nlist->iinr;
 127     jindex           = nlist->jindex;
 128     jjnr             = nlist->jjnr;
 129     shiftidx         = nlist->shift;
 130     gid              = nlist->gid;
 131     shiftvec         = fr->shift_vec[0];
 132     fshift           = fr->fshift[0];
 133     facel            = _mm256_set1_pd(fr->epsfac);
 134     charge           = mdatoms->chargeA;
 135     nvdwtype         = fr->ntype;
 136     vdwparam         = fr->nbfp;
 137     vdwtype          = mdatoms->typeA;
 138     vdwgridparam     = fr->ljpme_c6grid;
 139     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 140     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 141     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 142
 143     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 144     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 145     beta2            = _mm256_mul_pd(beta,beta);
 146     beta3            = _mm256_mul_pd(beta,beta2);
 147
 148     ewtab            = fr->ic->tabq_coul_FDV0;
 149     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 150     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 151
 152     /* Setup water-specific parameters */
 153     inr              = nlist->iinr[0];
 154     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 155     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 156     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 157     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 158     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 159
 160     /* Avoid stupid compiler warnings */
 161     jnrA = jnrB = jnrC = jnrD = 0;
 162     j_coord_offsetA = 0;
 163     j_coord_offsetB = 0;
 164     j_coord_offsetC = 0;
 165     j_coord_offsetD = 0;
 166
 167     outeriter        = 0;
 168     inneriter        = 0;
 169
 170     for(iidx=0;iidx<4*DIM;iidx++)
 171     {
 172         scratch[iidx] = 0.0;
 173     }
 174
 175     /* Start outer loop over neighborlists */
 176     for(iidx=0; iidx<nri; iidx++)
 177     {
 178         /* Load shift vector for this list */
 179         i_shift_offset   = DIM*shiftidx[iidx];
 180
 181         /* Load limits for loop over neighbors */
 182         j_index_start    = jindex[iidx];
 183         j_index_end      = jindex[iidx+1];
 184
 185         /* Get outer coordinate index */
 186         inr              = iinr[iidx];
 187         i_coord_offset   = DIM*inr;
 188
 189         /* Load i particle coords and add shift vector */
 190         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 191                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 192
 193         fix0             = _mm256_setzero_pd();
 194         fiy0             = _mm256_setzero_pd();
 195         fiz0             = _mm256_setzero_pd();
 196         fix1             = _mm256_setzero_pd();
 197         fiy1             = _mm256_setzero_pd();
 198         fiz1             = _mm256_setzero_pd();
 199         fix2             = _mm256_setzero_pd();
 200         fiy2             = _mm256_setzero_pd();
 201         fiz2             = _mm256_setzero_pd();
 202
 203         /* Reset potential sums */
 204         velecsum         = _mm256_setzero_pd();
 205         vvdwsum          = _mm256_setzero_pd();
 206
 207         /* Start inner kernel loop */
 208         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 209         {
 210
 211             /* Get j neighbor index, and coordinate index */
 212             jnrA             = jjnr[jidx];
 213             jnrB             = jjnr[jidx+1];
 214             jnrC             = jjnr[jidx+2];
 215             jnrD             = jjnr[jidx+3];
 216             j_coord_offsetA  = DIM*jnrA;
 217             j_coord_offsetB  = DIM*jnrB;
 218             j_coord_offsetC  = DIM*jnrC;
 219             j_coord_offsetD  = DIM*jnrD;
 220
 221             /* load j atom coordinates */
 222             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 223                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 224                                                  &jx0,&jy0,&jz0);
 225
 226             /* Calculate displacement vector */
 227             dx00             = _mm256_sub_pd(ix0,jx0);
 228             dy00             = _mm256_sub_pd(iy0,jy0);
 229             dz00             = _mm256_sub_pd(iz0,jz0);
 230             dx10             = _mm256_sub_pd(ix1,jx0);
 231             dy10             = _mm256_sub_pd(iy1,jy0);
 232             dz10             = _mm256_sub_pd(iz1,jz0);
 233             dx20             = _mm256_sub_pd(ix2,jx0);
 234             dy20             = _mm256_sub_pd(iy2,jy0);
 235             dz20             = _mm256_sub_pd(iz2,jz0);
 236
 237             /* Calculate squared distance and things based on it */
 238             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 239             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 240             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 241
 242             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 243             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 244             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 245
 246             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 247             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 248             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 249
 250             /* Load parameters for j particles */
 251             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 252                                                                  charge+jnrC+0,charge+jnrD+0);
 253             vdwjidx0A        = 2*vdwtype[jnrA+0];
 254             vdwjidx0B        = 2*vdwtype[jnrB+0];
 255             vdwjidx0C        = 2*vdwtype[jnrC+0];
 256             vdwjidx0D        = 2*vdwtype[jnrD+0];
 257
 258             fjx0             = _mm256_setzero_pd();
 259             fjy0             = _mm256_setzero_pd();
 260             fjz0             = _mm256_setzero_pd();
 261
 262             /**************************
 263              * CALCULATE INTERACTIONS *
 264              **************************/
 265
 266             r00              = _mm256_mul_pd(rsq00,rinv00);
 267
 268             /* Compute parameters for interactions between i and j atoms */
 269             qq00             = _mm256_mul_pd(iq0,jq0);
 270             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 271                                             vdwioffsetptr0+vdwjidx0B,
 272                                             vdwioffsetptr0+vdwjidx0C,
 273                                             vdwioffsetptr0+vdwjidx0D,
 274                                             &c6_00,&c12_00);
 275
 276             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 277                                                                   vdwgridioffsetptr0+vdwjidx0B,
 278                                                                   vdwgridioffsetptr0+vdwjidx0C,
 279                                                                   vdwgridioffsetptr0+vdwjidx0D);
 280
 281             /* EWALD ELECTROSTATICS */
 282
 283             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 284             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 285             ewitab           = _mm256_cvttpd_epi32(ewrt);
 286             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 287             ewitab           = _mm_slli_epi32(ewitab,2);
 288             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 289             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 290             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 291             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 292             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 293             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 294             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 295             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 296             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 297
 298             /* Analytical LJ-PME */
 299             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 300             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 301             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 302             exponent         = gmx_simd_exp_d(ewcljrsq);
 303             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 304             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 305             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 306             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 307             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 308             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 309             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 310             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);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm256_add_pd(velecsum,velec);
 314             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 315
 316             fscal            = _mm256_add_pd(felec,fvdw);
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm256_mul_pd(fscal,dx00);
 320             ty               = _mm256_mul_pd(fscal,dy00);
 321             tz               = _mm256_mul_pd(fscal,dz00);
 322
 323             /* Update vectorial force */
 324             fix0             = _mm256_add_pd(fix0,tx);
 325             fiy0             = _mm256_add_pd(fiy0,ty);
 326             fiz0             = _mm256_add_pd(fiz0,tz);
 327
 328             fjx0             = _mm256_add_pd(fjx0,tx);
 329             fjy0             = _mm256_add_pd(fjy0,ty);
 330             fjz0             = _mm256_add_pd(fjz0,tz);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             r10              = _mm256_mul_pd(rsq10,rinv10);
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq10             = _mm256_mul_pd(iq1,jq0);
 340
 341             /* EWALD ELECTROSTATICS */
 342
 343             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 344             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 345             ewitab           = _mm256_cvttpd_epi32(ewrt);
 346             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 347             ewitab           = _mm_slli_epi32(ewitab,2);
 348             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 349             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 350             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 351             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 352             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 353             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 354             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 355             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 356             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm256_add_pd(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm256_mul_pd(fscal,dx10);
 365             ty               = _mm256_mul_pd(fscal,dy10);
 366             tz               = _mm256_mul_pd(fscal,dz10);
 367
 368             /* Update vectorial force */
 369             fix1             = _mm256_add_pd(fix1,tx);
 370             fiy1             = _mm256_add_pd(fiy1,ty);
 371             fiz1             = _mm256_add_pd(fiz1,tz);
 372
 373             fjx0             = _mm256_add_pd(fjx0,tx);
 374             fjy0             = _mm256_add_pd(fjy0,ty);
 375             fjz0             = _mm256_add_pd(fjz0,tz);
 376
 377             /**************************
 378              * CALCULATE INTERACTIONS *
 379              **************************/
 380
 381             r20              = _mm256_mul_pd(rsq20,rinv20);
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq20             = _mm256_mul_pd(iq2,jq0);
 385
 386             /* EWALD ELECTROSTATICS */
 387
 388             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 389             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 390             ewitab           = _mm256_cvttpd_epi32(ewrt);
 391             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 392             ewitab           = _mm_slli_epi32(ewitab,2);
 393             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 394             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 395             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 396             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 397             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 398             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 399             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 400             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 401             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velecsum         = _mm256_add_pd(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm256_mul_pd(fscal,dx20);
 410             ty               = _mm256_mul_pd(fscal,dy20);
 411             tz               = _mm256_mul_pd(fscal,dz20);
 412
 413             /* Update vectorial force */
 414             fix2             = _mm256_add_pd(fix2,tx);
 415             fiy2             = _mm256_add_pd(fiy2,ty);
 416             fiz2             = _mm256_add_pd(fiz2,tz);
 417
 418             fjx0             = _mm256_add_pd(fjx0,tx);
 419             fjy0             = _mm256_add_pd(fjy0,ty);
 420             fjz0             = _mm256_add_pd(fjz0,tz);
 421
 422             fjptrA             = f+j_coord_offsetA;
 423             fjptrB             = f+j_coord_offsetB;
 424             fjptrC             = f+j_coord_offsetC;
 425             fjptrD             = f+j_coord_offsetD;
 426
 427             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 428
 429             /* Inner loop uses 157 flops */
 430         }
 431
 432         if(jidx<j_index_end)
 433         {
 434
 435             /* Get j neighbor index, and coordinate index */
 436             jnrlistA         = jjnr[jidx];
 437             jnrlistB         = jjnr[jidx+1];
 438             jnrlistC         = jjnr[jidx+2];
 439             jnrlistD         = jjnr[jidx+3];
 440             /* Sign of each element will be negative for non-real atoms.
 441              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 442              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 443              */
 444             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 445
 446             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 447             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 448             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 449
 450             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 451             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 452             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 453             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 454             j_coord_offsetA  = DIM*jnrA;
 455             j_coord_offsetB  = DIM*jnrB;
 456             j_coord_offsetC  = DIM*jnrC;
 457             j_coord_offsetD  = DIM*jnrD;
 458
 459             /* load j atom coordinates */
 460             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 461                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 462                                                  &jx0,&jy0,&jz0);
 463
 464             /* Calculate displacement vector */
 465             dx00             = _mm256_sub_pd(ix0,jx0);
 466             dy00             = _mm256_sub_pd(iy0,jy0);
 467             dz00             = _mm256_sub_pd(iz0,jz0);
 468             dx10             = _mm256_sub_pd(ix1,jx0);
 469             dy10             = _mm256_sub_pd(iy1,jy0);
 470             dz10             = _mm256_sub_pd(iz1,jz0);
 471             dx20             = _mm256_sub_pd(ix2,jx0);
 472             dy20             = _mm256_sub_pd(iy2,jy0);
 473             dz20             = _mm256_sub_pd(iz2,jz0);
 474
 475             /* Calculate squared distance and things based on it */
 476             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 477             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 478             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 479
 480             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 481             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 482             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 483
 484             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 485             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 486             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 487
 488             /* Load parameters for j particles */
 489             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 490                                                                  charge+jnrC+0,charge+jnrD+0);
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493             vdwjidx0C        = 2*vdwtype[jnrC+0];
 494             vdwjidx0D        = 2*vdwtype[jnrD+0];
 495
 496             fjx0             = _mm256_setzero_pd();
 497             fjy0             = _mm256_setzero_pd();
 498             fjz0             = _mm256_setzero_pd();
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             r00              = _mm256_mul_pd(rsq00,rinv00);
 505             r00              = _mm256_andnot_pd(dummy_mask,r00);
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             qq00             = _mm256_mul_pd(iq0,jq0);
 509             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 510                                             vdwioffsetptr0+vdwjidx0B,
 511                                             vdwioffsetptr0+vdwjidx0C,
 512                                             vdwioffsetptr0+vdwjidx0D,
 513                                             &c6_00,&c12_00);
 514
 515             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 516                                                                   vdwgridioffsetptr0+vdwjidx0B,
 517                                                                   vdwgridioffsetptr0+vdwjidx0C,
 518                                                                   vdwgridioffsetptr0+vdwjidx0D);
 519
 520             /* EWALD ELECTROSTATICS */
 521
 522             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 523             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 524             ewitab           = _mm256_cvttpd_epi32(ewrt);
 525             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 526             ewitab           = _mm_slli_epi32(ewitab,2);
 527             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 528             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 529             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 530             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 531             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 532             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 533             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 534             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 535             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 536
 537             /* Analytical LJ-PME */
 538             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 539             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 540             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 541             exponent         = gmx_simd_exp_d(ewcljrsq);
 542             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 543             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 544             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 545             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 546             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 547             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 548             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 549             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);
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm256_andnot_pd(dummy_mask,velec);
 553             velecsum         = _mm256_add_pd(velecsum,velec);
 554             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 555             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 556
 557             fscal            = _mm256_add_pd(felec,fvdw);
 558
 559             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm256_mul_pd(fscal,dx00);
 563             ty               = _mm256_mul_pd(fscal,dy00);
 564             tz               = _mm256_mul_pd(fscal,dz00);
 565
 566             /* Update vectorial force */
 567             fix0             = _mm256_add_pd(fix0,tx);
 568             fiy0             = _mm256_add_pd(fiy0,ty);
 569             fiz0             = _mm256_add_pd(fiz0,tz);
 570
 571             fjx0             = _mm256_add_pd(fjx0,tx);
 572             fjy0             = _mm256_add_pd(fjy0,ty);
 573             fjz0             = _mm256_add_pd(fjz0,tz);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r10              = _mm256_mul_pd(rsq10,rinv10);
 580             r10              = _mm256_andnot_pd(dummy_mask,r10);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq10             = _mm256_mul_pd(iq1,jq0);
 584
 585             /* EWALD ELECTROSTATICS */
 586
 587             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 588             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 589             ewitab           = _mm256_cvttpd_epi32(ewrt);
 590             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 591             ewitab           = _mm_slli_epi32(ewitab,2);
 592             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 593             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 594             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 595             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 596             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 597             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 598             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 599             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 600             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 601
 602             /* Update potential sum for this i atom from the interaction with this j atom. */
 603             velec            = _mm256_andnot_pd(dummy_mask,velec);
 604             velecsum         = _mm256_add_pd(velecsum,velec);
 605
 606             fscal            = felec;
 607
 608             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 609
 610             /* Calculate temporary vectorial force */
 611             tx               = _mm256_mul_pd(fscal,dx10);
 612             ty               = _mm256_mul_pd(fscal,dy10);
 613             tz               = _mm256_mul_pd(fscal,dz10);
 614
 615             /* Update vectorial force */
 616             fix1             = _mm256_add_pd(fix1,tx);
 617             fiy1             = _mm256_add_pd(fiy1,ty);
 618             fiz1             = _mm256_add_pd(fiz1,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              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             r20              = _mm256_mul_pd(rsq20,rinv20);
 629             r20              = _mm256_andnot_pd(dummy_mask,r20);
 630
 631             /* Compute parameters for interactions between i and j atoms */
 632             qq20             = _mm256_mul_pd(iq2,jq0);
 633
 634             /* EWALD ELECTROSTATICS */
 635
 636             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 637             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 638             ewitab           = _mm256_cvttpd_epi32(ewrt);
 639             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 640             ewitab           = _mm_slli_epi32(ewitab,2);
 641             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 642             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 643             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 644             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 645             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 646             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 647             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 648             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 649             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 650
 651             /* Update potential sum for this i atom from the interaction with this j atom. */
 652             velec            = _mm256_andnot_pd(dummy_mask,velec);
 653             velecsum         = _mm256_add_pd(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 658
 659             /* Calculate temporary vectorial force */
 660             tx               = _mm256_mul_pd(fscal,dx20);
 661             ty               = _mm256_mul_pd(fscal,dy20);
 662             tz               = _mm256_mul_pd(fscal,dz20);
 663
 664             /* Update vectorial force */
 665             fix2             = _mm256_add_pd(fix2,tx);
 666             fiy2             = _mm256_add_pd(fiy2,ty);
 667             fiz2             = _mm256_add_pd(fiz2,tz);
 668
 669             fjx0             = _mm256_add_pd(fjx0,tx);
 670             fjy0             = _mm256_add_pd(fjy0,ty);
 671             fjz0             = _mm256_add_pd(fjz0,tz);
 672
 673             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 674             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 675             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 676             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 677
 678             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 679
 680             /* Inner loop uses 160 flops */
 681         }
 682
 683         /* End of innermost loop */
 684
 685         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 686                                                  f+i_coord_offset,fshift+i_shift_offset);
 687
 688         ggid                        = gid[iidx];
 689         /* Update potential energies */
 690         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 691         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 692
 693         /* Increment number of inner iterations */
 694         inneriter                  += j_index_end - j_index_start;
 695
 696         /* Outer loop uses 20 flops */
 697     }
 698
 699     /* Increment number of outer iterations */
 700     outeriter        += nri;
 701
 702     /* Update outer/inner flops */
 703
 704     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 705 }
 706 /*
 707  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_256_double
 708  * Electrostatics interaction: Ewald
 709  * VdW interaction:            LJEwald
 710  * Geometry:                   Water3-Particle
 711  * Calculate force/pot:        Force
 712  */
 713 void
 714 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_256_double
 715                     (t_nblist                    * gmx_restrict       nlist,
 716                      rvec                        * gmx_restrict          xx,
 717                      rvec                        * gmx_restrict          ff,
 718                      t_forcerec                  * gmx_restrict          fr,
 719                      t_mdatoms                   * gmx_restrict     mdatoms,
 720                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 721                      t_nrnb                      * gmx_restrict        nrnb)
 722 {
 723     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 724      * just 0 for non-waters.
 725      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 726      * jnr indices corresponding to data put in the four positions in the SIMD register.
 727      */
 728     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 729     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 730     int              jnrA,jnrB,jnrC,jnrD;
 731     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 732     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 733     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 734     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 735     real             rcutoff_scalar;
 736     real             *shiftvec,*fshift,*x,*f;
 737     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 738     real             scratch[4*DIM];
 739     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 740     real *           vdwioffsetptr0;
 741     real *           vdwgridioffsetptr0;
 742     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 743     real *           vdwioffsetptr1;
 744     real *           vdwgridioffsetptr1;
 745     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 746     real *           vdwioffsetptr2;
 747     real *           vdwgridioffsetptr2;
 748     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 749     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 750     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 751     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 752     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 753     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 754     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 755     real             *charge;
 756     int              nvdwtype;
 757     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 758     int              *vdwtype;
 759     real             *vdwparam;
 760     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 761     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 762     __m256d           c6grid_00;
 763     __m256d           c6grid_10;
 764     __m256d           c6grid_20;
 765     real             *vdwgridparam;
 766     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 767     __m256d           one_half  = _mm256_set1_pd(0.5);
 768     __m256d           minus_one = _mm256_set1_pd(-1.0);
 769     __m128i          ewitab;
 770     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 771     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 772     real             *ewtab;
 773     __m256d          dummy_mask,cutoff_mask;
 774     __m128           tmpmask0,tmpmask1;
 775     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 776     __m256d          one     = _mm256_set1_pd(1.0);
 777     __m256d          two     = _mm256_set1_pd(2.0);
 778     x                = xx[0];
 779     f                = ff[0];
 780
 781     nri              = nlist->nri;
 782     iinr             = nlist->iinr;
 783     jindex           = nlist->jindex;
 784     jjnr             = nlist->jjnr;
 785     shiftidx         = nlist->shift;
 786     gid              = nlist->gid;
 787     shiftvec         = fr->shift_vec[0];
 788     fshift           = fr->fshift[0];
 789     facel            = _mm256_set1_pd(fr->epsfac);
 790     charge           = mdatoms->chargeA;
 791     nvdwtype         = fr->ntype;
 792     vdwparam         = fr->nbfp;
 793     vdwtype          = mdatoms->typeA;
 794     vdwgridparam     = fr->ljpme_c6grid;
 795     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 796     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 797     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 798
 799     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 800     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 801     beta2            = _mm256_mul_pd(beta,beta);
 802     beta3            = _mm256_mul_pd(beta,beta2);
 803
 804     ewtab            = fr->ic->tabq_coul_F;
 805     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 806     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 807
 808     /* Setup water-specific parameters */
 809     inr              = nlist->iinr[0];
 810     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 811     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 812     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 813     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 814     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 815
 816     /* Avoid stupid compiler warnings */
 817     jnrA = jnrB = jnrC = jnrD = 0;
 818     j_coord_offsetA = 0;
 819     j_coord_offsetB = 0;
 820     j_coord_offsetC = 0;
 821     j_coord_offsetD = 0;
 822
 823     outeriter        = 0;
 824     inneriter        = 0;
 825
 826     for(iidx=0;iidx<4*DIM;iidx++)
 827     {
 828         scratch[iidx] = 0.0;
 829     }
 830
 831     /* Start outer loop over neighborlists */
 832     for(iidx=0; iidx<nri; iidx++)
 833     {
 834         /* Load shift vector for this list */
 835         i_shift_offset   = DIM*shiftidx[iidx];
 836
 837         /* Load limits for loop over neighbors */
 838         j_index_start    = jindex[iidx];
 839         j_index_end      = jindex[iidx+1];
 840
 841         /* Get outer coordinate index */
 842         inr              = iinr[iidx];
 843         i_coord_offset   = DIM*inr;
 844
 845         /* Load i particle coords and add shift vector */
 846         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 847                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 848
 849         fix0             = _mm256_setzero_pd();
 850         fiy0             = _mm256_setzero_pd();
 851         fiz0             = _mm256_setzero_pd();
 852         fix1             = _mm256_setzero_pd();
 853         fiy1             = _mm256_setzero_pd();
 854         fiz1             = _mm256_setzero_pd();
 855         fix2             = _mm256_setzero_pd();
 856         fiy2             = _mm256_setzero_pd();
 857         fiz2             = _mm256_setzero_pd();
 858
 859         /* Start inner kernel loop */
 860         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 861         {
 862
 863             /* Get j neighbor index, and coordinate index */
 864             jnrA             = jjnr[jidx];
 865             jnrB             = jjnr[jidx+1];
 866             jnrC             = jjnr[jidx+2];
 867             jnrD             = jjnr[jidx+3];
 868             j_coord_offsetA  = DIM*jnrA;
 869             j_coord_offsetB  = DIM*jnrB;
 870             j_coord_offsetC  = DIM*jnrC;
 871             j_coord_offsetD  = DIM*jnrD;
 872
 873             /* load j atom coordinates */
 874             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 875                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 876                                                  &jx0,&jy0,&jz0);
 877
 878             /* Calculate displacement vector */
 879             dx00             = _mm256_sub_pd(ix0,jx0);
 880             dy00             = _mm256_sub_pd(iy0,jy0);
 881             dz00             = _mm256_sub_pd(iz0,jz0);
 882             dx10             = _mm256_sub_pd(ix1,jx0);
 883             dy10             = _mm256_sub_pd(iy1,jy0);
 884             dz10             = _mm256_sub_pd(iz1,jz0);
 885             dx20             = _mm256_sub_pd(ix2,jx0);
 886             dy20             = _mm256_sub_pd(iy2,jy0);
 887             dz20             = _mm256_sub_pd(iz2,jz0);
 888
 889             /* Calculate squared distance and things based on it */
 890             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 891             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 892             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 893
 894             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 895             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 896             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 897
 898             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 899             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 900             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 901
 902             /* Load parameters for j particles */
 903             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 904                                                                  charge+jnrC+0,charge+jnrD+0);
 905             vdwjidx0A        = 2*vdwtype[jnrA+0];
 906             vdwjidx0B        = 2*vdwtype[jnrB+0];
 907             vdwjidx0C        = 2*vdwtype[jnrC+0];
 908             vdwjidx0D        = 2*vdwtype[jnrD+0];
 909
 910             fjx0             = _mm256_setzero_pd();
 911             fjy0             = _mm256_setzero_pd();
 912             fjz0             = _mm256_setzero_pd();
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             r00              = _mm256_mul_pd(rsq00,rinv00);
 919
 920             /* Compute parameters for interactions between i and j atoms */
 921             qq00             = _mm256_mul_pd(iq0,jq0);
 922             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 923                                             vdwioffsetptr0+vdwjidx0B,
 924                                             vdwioffsetptr0+vdwjidx0C,
 925                                             vdwioffsetptr0+vdwjidx0D,
 926                                             &c6_00,&c12_00);
 927
 928             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 929                                                                   vdwgridioffsetptr0+vdwjidx0B,
 930                                                                   vdwgridioffsetptr0+vdwjidx0C,
 931                                                                   vdwgridioffsetptr0+vdwjidx0D);
 932
 933             /* EWALD ELECTROSTATICS */
 934
 935             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 936             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 937             ewitab           = _mm256_cvttpd_epi32(ewrt);
 938             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 939             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 940                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 941                                             &ewtabF,&ewtabFn);
 942             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 943             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 944
 945             /* Analytical LJ-PME */
 946             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 947             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 948             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 949             exponent         = gmx_simd_exp_d(ewcljrsq);
 950             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 951             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 952             /* f6A = 6 * C6grid * (1 - poly) */
 953             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 954             /* f6B = C6grid * exponent * beta^6 */
 955             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 956             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 957             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);
 958
 959             fscal            = _mm256_add_pd(felec,fvdw);
 960
 961             /* Calculate temporary vectorial force */
 962             tx               = _mm256_mul_pd(fscal,dx00);
 963             ty               = _mm256_mul_pd(fscal,dy00);
 964             tz               = _mm256_mul_pd(fscal,dz00);
 965
 966             /* Update vectorial force */
 967             fix0             = _mm256_add_pd(fix0,tx);
 968             fiy0             = _mm256_add_pd(fiy0,ty);
 969             fiz0             = _mm256_add_pd(fiz0,tz);
 970
 971             fjx0             = _mm256_add_pd(fjx0,tx);
 972             fjy0             = _mm256_add_pd(fjy0,ty);
 973             fjz0             = _mm256_add_pd(fjz0,tz);
 974
 975             /**************************
 976              * CALCULATE INTERACTIONS *
 977              **************************/
 978
 979             r10              = _mm256_mul_pd(rsq10,rinv10);
 980
 981             /* Compute parameters for interactions between i and j atoms */
 982             qq10             = _mm256_mul_pd(iq1,jq0);
 983
 984             /* EWALD ELECTROSTATICS */
 985
 986             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 987             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 988             ewitab           = _mm256_cvttpd_epi32(ewrt);
 989             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 990             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 991                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 992                                             &ewtabF,&ewtabFn);
 993             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 994             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 995
 996             fscal            = felec;
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm256_mul_pd(fscal,dx10);
1000             ty               = _mm256_mul_pd(fscal,dy10);
1001             tz               = _mm256_mul_pd(fscal,dz10);
1002
1003             /* Update vectorial force */
1004             fix1             = _mm256_add_pd(fix1,tx);
1005             fiy1             = _mm256_add_pd(fiy1,ty);
1006             fiz1             = _mm256_add_pd(fiz1,tz);
1007
1008             fjx0             = _mm256_add_pd(fjx0,tx);
1009             fjy0             = _mm256_add_pd(fjy0,ty);
1010             fjz0             = _mm256_add_pd(fjz0,tz);
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             r20              = _mm256_mul_pd(rsq20,rinv20);
1017
1018             /* Compute parameters for interactions between i and j atoms */
1019             qq20             = _mm256_mul_pd(iq2,jq0);
1020
1021             /* EWALD ELECTROSTATICS */
1022
1023             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1024             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1025             ewitab           = _mm256_cvttpd_epi32(ewrt);
1026             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1027             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1028                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1029                                             &ewtabF,&ewtabFn);
1030             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1031             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1032
1033             fscal            = felec;
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm256_mul_pd(fscal,dx20);
1037             ty               = _mm256_mul_pd(fscal,dy20);
1038             tz               = _mm256_mul_pd(fscal,dz20);
1039
1040             /* Update vectorial force */
1041             fix2             = _mm256_add_pd(fix2,tx);
1042             fiy2             = _mm256_add_pd(fiy2,ty);
1043             fiz2             = _mm256_add_pd(fiz2,tz);
1044
1045             fjx0             = _mm256_add_pd(fjx0,tx);
1046             fjy0             = _mm256_add_pd(fjy0,ty);
1047             fjz0             = _mm256_add_pd(fjz0,tz);
1048
1049             fjptrA             = f+j_coord_offsetA;
1050             fjptrB             = f+j_coord_offsetB;
1051             fjptrC             = f+j_coord_offsetC;
1052             fjptrD             = f+j_coord_offsetD;
1053
1054             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1055
1056             /* Inner loop uses 134 flops */
1057         }
1058
1059         if(jidx<j_index_end)
1060         {
1061
1062             /* Get j neighbor index, and coordinate index */
1063             jnrlistA         = jjnr[jidx];
1064             jnrlistB         = jjnr[jidx+1];
1065             jnrlistC         = jjnr[jidx+2];
1066             jnrlistD         = jjnr[jidx+3];
1067             /* Sign of each element will be negative for non-real atoms.
1068              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1069              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1070              */
1071             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1072
1073             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1074             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1075             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1076
1077             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1078             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1079             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1080             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1081             j_coord_offsetA  = DIM*jnrA;
1082             j_coord_offsetB  = DIM*jnrB;
1083             j_coord_offsetC  = DIM*jnrC;
1084             j_coord_offsetD  = DIM*jnrD;
1085
1086             /* load j atom coordinates */
1087             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1088                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1089                                                  &jx0,&jy0,&jz0);
1090
1091             /* Calculate displacement vector */
1092             dx00             = _mm256_sub_pd(ix0,jx0);
1093             dy00             = _mm256_sub_pd(iy0,jy0);
1094             dz00             = _mm256_sub_pd(iz0,jz0);
1095             dx10             = _mm256_sub_pd(ix1,jx0);
1096             dy10             = _mm256_sub_pd(iy1,jy0);
1097             dz10             = _mm256_sub_pd(iz1,jz0);
1098             dx20             = _mm256_sub_pd(ix2,jx0);
1099             dy20             = _mm256_sub_pd(iy2,jy0);
1100             dz20             = _mm256_sub_pd(iz2,jz0);
1101
1102             /* Calculate squared distance and things based on it */
1103             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1104             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1105             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1106
1107             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1108             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1109             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1110
1111             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1112             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1113             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1114
1115             /* Load parameters for j particles */
1116             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1117                                                                  charge+jnrC+0,charge+jnrD+0);
1118             vdwjidx0A        = 2*vdwtype[jnrA+0];
1119             vdwjidx0B        = 2*vdwtype[jnrB+0];
1120             vdwjidx0C        = 2*vdwtype[jnrC+0];
1121             vdwjidx0D        = 2*vdwtype[jnrD+0];
1122
1123             fjx0             = _mm256_setzero_pd();
1124             fjy0             = _mm256_setzero_pd();
1125             fjz0             = _mm256_setzero_pd();
1126
1127             /**************************
1128              * CALCULATE INTERACTIONS *
1129              **************************/
1130
1131             r00              = _mm256_mul_pd(rsq00,rinv00);
1132             r00              = _mm256_andnot_pd(dummy_mask,r00);
1133
1134             /* Compute parameters for interactions between i and j atoms */
1135             qq00             = _mm256_mul_pd(iq0,jq0);
1136             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1137                                             vdwioffsetptr0+vdwjidx0B,
1138                                             vdwioffsetptr0+vdwjidx0C,
1139                                             vdwioffsetptr0+vdwjidx0D,
1140                                             &c6_00,&c12_00);
1141
1142             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1143                                                                   vdwgridioffsetptr0+vdwjidx0B,
1144                                                                   vdwgridioffsetptr0+vdwjidx0C,
1145                                                                   vdwgridioffsetptr0+vdwjidx0D);
1146
1147             /* EWALD ELECTROSTATICS */
1148
1149             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1150             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1151             ewitab           = _mm256_cvttpd_epi32(ewrt);
1152             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1153             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1154                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1155                                             &ewtabF,&ewtabFn);
1156             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1157             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1158
1159             /* Analytical LJ-PME */
1160             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1161             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1162             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1163             exponent         = gmx_simd_exp_d(ewcljrsq);
1164             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1165             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1166             /* f6A = 6 * C6grid * (1 - poly) */
1167             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1168             /* f6B = C6grid * exponent * beta^6 */
1169             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1170             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1171             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);
1172
1173             fscal            = _mm256_add_pd(felec,fvdw);
1174
1175             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1176
1177             /* Calculate temporary vectorial force */
1178             tx               = _mm256_mul_pd(fscal,dx00);
1179             ty               = _mm256_mul_pd(fscal,dy00);
1180             tz               = _mm256_mul_pd(fscal,dz00);
1181
1182             /* Update vectorial force */
1183             fix0             = _mm256_add_pd(fix0,tx);
1184             fiy0             = _mm256_add_pd(fiy0,ty);
1185             fiz0             = _mm256_add_pd(fiz0,tz);
1186
1187             fjx0             = _mm256_add_pd(fjx0,tx);
1188             fjy0             = _mm256_add_pd(fjy0,ty);
1189             fjz0             = _mm256_add_pd(fjz0,tz);
1190
1191             /**************************
1192              * CALCULATE INTERACTIONS *
1193              **************************/
1194
1195             r10              = _mm256_mul_pd(rsq10,rinv10);
1196             r10              = _mm256_andnot_pd(dummy_mask,r10);
1197
1198             /* Compute parameters for interactions between i and j atoms */
1199             qq10             = _mm256_mul_pd(iq1,jq0);
1200
1201             /* EWALD ELECTROSTATICS */
1202
1203             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1204             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1205             ewitab           = _mm256_cvttpd_epi32(ewrt);
1206             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1207             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1208                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1209                                             &ewtabF,&ewtabFn);
1210             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1211             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1212
1213             fscal            = felec;
1214
1215             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1216
1217             /* Calculate temporary vectorial force */
1218             tx               = _mm256_mul_pd(fscal,dx10);
1219             ty               = _mm256_mul_pd(fscal,dy10);
1220             tz               = _mm256_mul_pd(fscal,dz10);
1221
1222             /* Update vectorial force */
1223             fix1             = _mm256_add_pd(fix1,tx);
1224             fiy1             = _mm256_add_pd(fiy1,ty);
1225             fiz1             = _mm256_add_pd(fiz1,tz);
1226
1227             fjx0             = _mm256_add_pd(fjx0,tx);
1228             fjy0             = _mm256_add_pd(fjy0,ty);
1229             fjz0             = _mm256_add_pd(fjz0,tz);
1230
1231             /**************************
1232              * CALCULATE INTERACTIONS *
1233              **************************/
1234
1235             r20              = _mm256_mul_pd(rsq20,rinv20);
1236             r20              = _mm256_andnot_pd(dummy_mask,r20);
1237
1238             /* Compute parameters for interactions between i and j atoms */
1239             qq20             = _mm256_mul_pd(iq2,jq0);
1240
1241             /* EWALD ELECTROSTATICS */
1242
1243             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1244             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1245             ewitab           = _mm256_cvttpd_epi32(ewrt);
1246             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1247             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1248                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1249                                             &ewtabF,&ewtabFn);
1250             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1251             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1252
1253             fscal            = felec;
1254
1255             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1256
1257             /* Calculate temporary vectorial force */
1258             tx               = _mm256_mul_pd(fscal,dx20);
1259             ty               = _mm256_mul_pd(fscal,dy20);
1260             tz               = _mm256_mul_pd(fscal,dz20);
1261
1262             /* Update vectorial force */
1263             fix2             = _mm256_add_pd(fix2,tx);
1264             fiy2             = _mm256_add_pd(fiy2,ty);
1265             fiz2             = _mm256_add_pd(fiz2,tz);
1266
1267             fjx0             = _mm256_add_pd(fjx0,tx);
1268             fjy0             = _mm256_add_pd(fjy0,ty);
1269             fjz0             = _mm256_add_pd(fjz0,tz);
1270
1271             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1272             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1273             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1274             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1275
1276             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1277
1278             /* Inner loop uses 137 flops */
1279         }
1280
1281         /* End of innermost loop */
1282
1283         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1284                                                  f+i_coord_offset,fshift+i_shift_offset);
1285
1286         /* Increment number of inner iterations */
1287         inneriter                  += j_index_end - j_index_start;
1288
1289         /* Outer loop uses 18 flops */
1290     }
1291
1292     /* Increment number of outer iterations */
1293     outeriter        += nri;
1294
1295     /* Update outer/inner flops */
1296
1297     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1298 }