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 #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_ElecEw_VdwLJEw_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_ElecEw_VdwLJEw_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     /* Avoid stupid compiler warnings */
 163     jnrA = jnrB = jnrC = jnrD = 0;
 164     j_coord_offsetA = 0;
 165     j_coord_offsetB = 0;
 166     j_coord_offsetC = 0;
 167     j_coord_offsetD = 0;
 168
 169     outeriter        = 0;
 170     inneriter        = 0;
 171
 172     for(iidx=0;iidx<4*DIM;iidx++)
 173     {
 174         scratch[iidx] = 0.0;
 175     }
 176
 177     /* Start outer loop over neighborlists */
 178     for(iidx=0; iidx<nri; iidx++)
 179     {
 180         /* Load shift vector for this list */
 181         i_shift_offset   = DIM*shiftidx[iidx];
 182
 183         /* Load limits for loop over neighbors */
 184         j_index_start    = jindex[iidx];
 185         j_index_end      = jindex[iidx+1];
 186
 187         /* Get outer coordinate index */
 188         inr              = iinr[iidx];
 189         i_coord_offset   = DIM*inr;
 190
 191         /* Load i particle coords and add shift vector */
 192         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 193                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 194
 195         fix0             = _mm256_setzero_pd();
 196         fiy0             = _mm256_setzero_pd();
 197         fiz0             = _mm256_setzero_pd();
 198         fix1             = _mm256_setzero_pd();
 199         fiy1             = _mm256_setzero_pd();
 200         fiz1             = _mm256_setzero_pd();
 201         fix2             = _mm256_setzero_pd();
 202         fiy2             = _mm256_setzero_pd();
 203         fiz2             = _mm256_setzero_pd();
 204
 205         /* Reset potential sums */
 206         velecsum         = _mm256_setzero_pd();
 207         vvdwsum          = _mm256_setzero_pd();
 208
 209         /* Start inner kernel loop */
 210         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 211         {
 212
 213             /* Get j neighbor index, and coordinate index */
 214             jnrA             = jjnr[jidx];
 215             jnrB             = jjnr[jidx+1];
 216             jnrC             = jjnr[jidx+2];
 217             jnrD             = jjnr[jidx+3];
 218             j_coord_offsetA  = DIM*jnrA;
 219             j_coord_offsetB  = DIM*jnrB;
 220             j_coord_offsetC  = DIM*jnrC;
 221             j_coord_offsetD  = DIM*jnrD;
 222
 223             /* load j atom coordinates */
 224             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 225                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 226                                                  &jx0,&jy0,&jz0);
 227
 228             /* Calculate displacement vector */
 229             dx00             = _mm256_sub_pd(ix0,jx0);
 230             dy00             = _mm256_sub_pd(iy0,jy0);
 231             dz00             = _mm256_sub_pd(iz0,jz0);
 232             dx10             = _mm256_sub_pd(ix1,jx0);
 233             dy10             = _mm256_sub_pd(iy1,jy0);
 234             dz10             = _mm256_sub_pd(iz1,jz0);
 235             dx20             = _mm256_sub_pd(ix2,jx0);
 236             dy20             = _mm256_sub_pd(iy2,jy0);
 237             dz20             = _mm256_sub_pd(iz2,jz0);
 238
 239             /* Calculate squared distance and things based on it */
 240             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 241             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 242             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 243
 244             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 245             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 246             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 247
 248             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 249             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 250             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 251
 252             /* Load parameters for j particles */
 253             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 254                                                                  charge+jnrC+0,charge+jnrD+0);
 255             vdwjidx0A        = 2*vdwtype[jnrA+0];
 256             vdwjidx0B        = 2*vdwtype[jnrB+0];
 257             vdwjidx0C        = 2*vdwtype[jnrC+0];
 258             vdwjidx0D        = 2*vdwtype[jnrD+0];
 259
 260             fjx0             = _mm256_setzero_pd();
 261             fjy0             = _mm256_setzero_pd();
 262             fjz0             = _mm256_setzero_pd();
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             r00              = _mm256_mul_pd(rsq00,rinv00);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq00             = _mm256_mul_pd(iq0,jq0);
 272             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 273                                             vdwioffsetptr0+vdwjidx0B,
 274                                             vdwioffsetptr0+vdwjidx0C,
 275                                             vdwioffsetptr0+vdwjidx0D,
 276                                             &c6_00,&c12_00);
 277
 278             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 279                                                                   vdwgridioffsetptr0+vdwjidx0B,
 280                                                                   vdwgridioffsetptr0+vdwjidx0C,
 281                                                                   vdwgridioffsetptr0+vdwjidx0D);
 282
 283             /* EWALD ELECTROSTATICS */
 284
 285             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 286             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 287             ewitab           = _mm256_cvttpd_epi32(ewrt);
 288             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 289             ewitab           = _mm_slli_epi32(ewitab,2);
 290             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 291             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 292             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 293             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 294             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 295             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 296             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 297             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 298             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 299
 300             /* Analytical LJ-PME */
 301             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 302             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 303             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 304             exponent         = gmx_simd_exp_d(ewcljrsq);
 305             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 306             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 307             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 308             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 309             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 310             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 311             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 312             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);
 313
 314             /* Update potential sum for this i atom from the interaction with this j atom. */
 315             velecsum         = _mm256_add_pd(velecsum,velec);
 316             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 317
 318             fscal            = _mm256_add_pd(felec,fvdw);
 319
 320             /* Calculate temporary vectorial force */
 321             tx               = _mm256_mul_pd(fscal,dx00);
 322             ty               = _mm256_mul_pd(fscal,dy00);
 323             tz               = _mm256_mul_pd(fscal,dz00);
 324
 325             /* Update vectorial force */
 326             fix0             = _mm256_add_pd(fix0,tx);
 327             fiy0             = _mm256_add_pd(fiy0,ty);
 328             fiz0             = _mm256_add_pd(fiz0,tz);
 329
 330             fjx0             = _mm256_add_pd(fjx0,tx);
 331             fjy0             = _mm256_add_pd(fjy0,ty);
 332             fjz0             = _mm256_add_pd(fjz0,tz);
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             r10              = _mm256_mul_pd(rsq10,rinv10);
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq10             = _mm256_mul_pd(iq1,jq0);
 342
 343             /* EWALD ELECTROSTATICS */
 344
 345             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 346             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 347             ewitab           = _mm256_cvttpd_epi32(ewrt);
 348             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 349             ewitab           = _mm_slli_epi32(ewitab,2);
 350             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 351             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 352             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 353             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 354             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 355             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 356             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 357             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 358             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velecsum         = _mm256_add_pd(velecsum,velec);
 362
 363             fscal            = felec;
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm256_mul_pd(fscal,dx10);
 367             ty               = _mm256_mul_pd(fscal,dy10);
 368             tz               = _mm256_mul_pd(fscal,dz10);
 369
 370             /* Update vectorial force */
 371             fix1             = _mm256_add_pd(fix1,tx);
 372             fiy1             = _mm256_add_pd(fiy1,ty);
 373             fiz1             = _mm256_add_pd(fiz1,tz);
 374
 375             fjx0             = _mm256_add_pd(fjx0,tx);
 376             fjy0             = _mm256_add_pd(fjy0,ty);
 377             fjz0             = _mm256_add_pd(fjz0,tz);
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             r20              = _mm256_mul_pd(rsq20,rinv20);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq20             = _mm256_mul_pd(iq2,jq0);
 387
 388             /* EWALD ELECTROSTATICS */
 389
 390             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 391             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 392             ewitab           = _mm256_cvttpd_epi32(ewrt);
 393             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 394             ewitab           = _mm_slli_epi32(ewitab,2);
 395             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 396             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 397             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 398             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 399             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 400             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 401             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 402             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 403             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 404
 405             /* Update potential sum for this i atom from the interaction with this j atom. */
 406             velecsum         = _mm256_add_pd(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             /* Calculate temporary vectorial force */
 411             tx               = _mm256_mul_pd(fscal,dx20);
 412             ty               = _mm256_mul_pd(fscal,dy20);
 413             tz               = _mm256_mul_pd(fscal,dz20);
 414
 415             /* Update vectorial force */
 416             fix2             = _mm256_add_pd(fix2,tx);
 417             fiy2             = _mm256_add_pd(fiy2,ty);
 418             fiz2             = _mm256_add_pd(fiz2,tz);
 419
 420             fjx0             = _mm256_add_pd(fjx0,tx);
 421             fjy0             = _mm256_add_pd(fjy0,ty);
 422             fjz0             = _mm256_add_pd(fjz0,tz);
 423
 424             fjptrA             = f+j_coord_offsetA;
 425             fjptrB             = f+j_coord_offsetB;
 426             fjptrC             = f+j_coord_offsetC;
 427             fjptrD             = f+j_coord_offsetD;
 428
 429             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 430
 431             /* Inner loop uses 157 flops */
 432         }
 433
 434         if(jidx<j_index_end)
 435         {
 436
 437             /* Get j neighbor index, and coordinate index */
 438             jnrlistA         = jjnr[jidx];
 439             jnrlistB         = jjnr[jidx+1];
 440             jnrlistC         = jjnr[jidx+2];
 441             jnrlistD         = jjnr[jidx+3];
 442             /* Sign of each element will be negative for non-real atoms.
 443              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 444              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 445              */
 446             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 447
 448             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 449             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 450             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 451
 452             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 453             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 454             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 455             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 456             j_coord_offsetA  = DIM*jnrA;
 457             j_coord_offsetB  = DIM*jnrB;
 458             j_coord_offsetC  = DIM*jnrC;
 459             j_coord_offsetD  = DIM*jnrD;
 460
 461             /* load j atom coordinates */
 462             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 463                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 464                                                  &jx0,&jy0,&jz0);
 465
 466             /* Calculate displacement vector */
 467             dx00             = _mm256_sub_pd(ix0,jx0);
 468             dy00             = _mm256_sub_pd(iy0,jy0);
 469             dz00             = _mm256_sub_pd(iz0,jz0);
 470             dx10             = _mm256_sub_pd(ix1,jx0);
 471             dy10             = _mm256_sub_pd(iy1,jy0);
 472             dz10             = _mm256_sub_pd(iz1,jz0);
 473             dx20             = _mm256_sub_pd(ix2,jx0);
 474             dy20             = _mm256_sub_pd(iy2,jy0);
 475             dz20             = _mm256_sub_pd(iz2,jz0);
 476
 477             /* Calculate squared distance and things based on it */
 478             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 479             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 480             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 481
 482             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 483             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 484             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 485
 486             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 487             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 488             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 489
 490             /* Load parameters for j particles */
 491             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 492                                                                  charge+jnrC+0,charge+jnrD+0);
 493             vdwjidx0A        = 2*vdwtype[jnrA+0];
 494             vdwjidx0B        = 2*vdwtype[jnrB+0];
 495             vdwjidx0C        = 2*vdwtype[jnrC+0];
 496             vdwjidx0D        = 2*vdwtype[jnrD+0];
 497
 498             fjx0             = _mm256_setzero_pd();
 499             fjy0             = _mm256_setzero_pd();
 500             fjz0             = _mm256_setzero_pd();
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             r00              = _mm256_mul_pd(rsq00,rinv00);
 507             r00              = _mm256_andnot_pd(dummy_mask,r00);
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq00             = _mm256_mul_pd(iq0,jq0);
 511             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 512                                             vdwioffsetptr0+vdwjidx0B,
 513                                             vdwioffsetptr0+vdwjidx0C,
 514                                             vdwioffsetptr0+vdwjidx0D,
 515                                             &c6_00,&c12_00);
 516
 517             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 518                                                                   vdwgridioffsetptr0+vdwjidx0B,
 519                                                                   vdwgridioffsetptr0+vdwjidx0C,
 520                                                                   vdwgridioffsetptr0+vdwjidx0D);
 521
 522             /* EWALD ELECTROSTATICS */
 523
 524             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 525             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 526             ewitab           = _mm256_cvttpd_epi32(ewrt);
 527             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 528             ewitab           = _mm_slli_epi32(ewitab,2);
 529             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 530             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 531             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 532             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 533             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 534             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 535             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 536             velec            = _mm256_mul_pd(qq00,_mm256_sub_pd(rinv00,velec));
 537             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 538
 539             /* Analytical LJ-PME */
 540             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 541             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 542             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 543             exponent         = gmx_simd_exp_d(ewcljrsq);
 544             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 545             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 546             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 547             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 548             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 549             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 550             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 551             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);
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velec            = _mm256_andnot_pd(dummy_mask,velec);
 555             velecsum         = _mm256_add_pd(velecsum,velec);
 556             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 557             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 558
 559             fscal            = _mm256_add_pd(felec,fvdw);
 560
 561             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 562
 563             /* Calculate temporary vectorial force */
 564             tx               = _mm256_mul_pd(fscal,dx00);
 565             ty               = _mm256_mul_pd(fscal,dy00);
 566             tz               = _mm256_mul_pd(fscal,dz00);
 567
 568             /* Update vectorial force */
 569             fix0             = _mm256_add_pd(fix0,tx);
 570             fiy0             = _mm256_add_pd(fiy0,ty);
 571             fiz0             = _mm256_add_pd(fiz0,tz);
 572
 573             fjx0             = _mm256_add_pd(fjx0,tx);
 574             fjy0             = _mm256_add_pd(fjy0,ty);
 575             fjz0             = _mm256_add_pd(fjz0,tz);
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             r10              = _mm256_mul_pd(rsq10,rinv10);
 582             r10              = _mm256_andnot_pd(dummy_mask,r10);
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq10             = _mm256_mul_pd(iq1,jq0);
 586
 587             /* EWALD ELECTROSTATICS */
 588
 589             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 590             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 591             ewitab           = _mm256_cvttpd_epi32(ewrt);
 592             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 593             ewitab           = _mm_slli_epi32(ewitab,2);
 594             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 595             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 596             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 597             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 598             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 599             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 600             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 601             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 602             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 603
 604             /* Update potential sum for this i atom from the interaction with this j atom. */
 605             velec            = _mm256_andnot_pd(dummy_mask,velec);
 606             velecsum         = _mm256_add_pd(velecsum,velec);
 607
 608             fscal            = felec;
 609
 610             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 611
 612             /* Calculate temporary vectorial force */
 613             tx               = _mm256_mul_pd(fscal,dx10);
 614             ty               = _mm256_mul_pd(fscal,dy10);
 615             tz               = _mm256_mul_pd(fscal,dz10);
 616
 617             /* Update vectorial force */
 618             fix1             = _mm256_add_pd(fix1,tx);
 619             fiy1             = _mm256_add_pd(fiy1,ty);
 620             fiz1             = _mm256_add_pd(fiz1,tz);
 621
 622             fjx0             = _mm256_add_pd(fjx0,tx);
 623             fjy0             = _mm256_add_pd(fjy0,ty);
 624             fjz0             = _mm256_add_pd(fjz0,tz);
 625
 626             /**************************
 627              * CALCULATE INTERACTIONS *
 628              **************************/
 629
 630             r20              = _mm256_mul_pd(rsq20,rinv20);
 631             r20              = _mm256_andnot_pd(dummy_mask,r20);
 632
 633             /* Compute parameters for interactions between i and j atoms */
 634             qq20             = _mm256_mul_pd(iq2,jq0);
 635
 636             /* EWALD ELECTROSTATICS */
 637
 638             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 639             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 640             ewitab           = _mm256_cvttpd_epi32(ewrt);
 641             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 642             ewitab           = _mm_slli_epi32(ewitab,2);
 643             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 644             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 645             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 646             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 647             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 648             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 649             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 650             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 651             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 652
 653             /* Update potential sum for this i atom from the interaction with this j atom. */
 654             velec            = _mm256_andnot_pd(dummy_mask,velec);
 655             velecsum         = _mm256_add_pd(velecsum,velec);
 656
 657             fscal            = felec;
 658
 659             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm256_mul_pd(fscal,dx20);
 663             ty               = _mm256_mul_pd(fscal,dy20);
 664             tz               = _mm256_mul_pd(fscal,dz20);
 665
 666             /* Update vectorial force */
 667             fix2             = _mm256_add_pd(fix2,tx);
 668             fiy2             = _mm256_add_pd(fiy2,ty);
 669             fiz2             = _mm256_add_pd(fiz2,tz);
 670
 671             fjx0             = _mm256_add_pd(fjx0,tx);
 672             fjy0             = _mm256_add_pd(fjy0,ty);
 673             fjz0             = _mm256_add_pd(fjz0,tz);
 674
 675             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 676             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 677             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 678             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 679
 680             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 681
 682             /* Inner loop uses 160 flops */
 683         }
 684
 685         /* End of innermost loop */
 686
 687         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 688                                                  f+i_coord_offset,fshift+i_shift_offset);
 689
 690         ggid                        = gid[iidx];
 691         /* Update potential energies */
 692         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 693         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 694
 695         /* Increment number of inner iterations */
 696         inneriter                  += j_index_end - j_index_start;
 697
 698         /* Outer loop uses 20 flops */
 699     }
 700
 701     /* Increment number of outer iterations */
 702     outeriter        += nri;
 703
 704     /* Update outer/inner flops */
 705
 706     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 707 }
 708 /*
 709  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_256_double
 710  * Electrostatics interaction: Ewald
 711  * VdW interaction:            LJEwald
 712  * Geometry:                   Water3-Particle
 713  * Calculate force/pot:        Force
 714  */
 715 void
 716 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_256_double
 717                     (t_nblist                    * gmx_restrict       nlist,
 718                      rvec                        * gmx_restrict          xx,
 719                      rvec                        * gmx_restrict          ff,
 720                      t_forcerec                  * gmx_restrict          fr,
 721                      t_mdatoms                   * gmx_restrict     mdatoms,
 722                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 723                      t_nrnb                      * gmx_restrict        nrnb)
 724 {
 725     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 726      * just 0 for non-waters.
 727      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 728      * jnr indices corresponding to data put in the four positions in the SIMD register.
 729      */
 730     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 731     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 732     int              jnrA,jnrB,jnrC,jnrD;
 733     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 734     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 735     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 736     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 737     real             rcutoff_scalar;
 738     real             *shiftvec,*fshift,*x,*f;
 739     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 740     real             scratch[4*DIM];
 741     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 742     real *           vdwioffsetptr0;
 743     real *           vdwgridioffsetptr0;
 744     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 745     real *           vdwioffsetptr1;
 746     real *           vdwgridioffsetptr1;
 747     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 748     real *           vdwioffsetptr2;
 749     real *           vdwgridioffsetptr2;
 750     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 751     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 752     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 753     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 754     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 755     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 756     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 757     real             *charge;
 758     int              nvdwtype;
 759     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 760     int              *vdwtype;
 761     real             *vdwparam;
 762     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 763     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 764     __m256d           c6grid_00;
 765     __m256d           c6grid_10;
 766     __m256d           c6grid_20;
 767     real             *vdwgridparam;
 768     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 769     __m256d           one_half  = _mm256_set1_pd(0.5);
 770     __m256d           minus_one = _mm256_set1_pd(-1.0);
 771     __m128i          ewitab;
 772     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 773     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 774     real             *ewtab;
 775     __m256d          dummy_mask,cutoff_mask;
 776     __m128           tmpmask0,tmpmask1;
 777     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 778     __m256d          one     = _mm256_set1_pd(1.0);
 779     __m256d          two     = _mm256_set1_pd(2.0);
 780     x                = xx[0];
 781     f                = ff[0];
 782
 783     nri              = nlist->nri;
 784     iinr             = nlist->iinr;
 785     jindex           = nlist->jindex;
 786     jjnr             = nlist->jjnr;
 787     shiftidx         = nlist->shift;
 788     gid              = nlist->gid;
 789     shiftvec         = fr->shift_vec[0];
 790     fshift           = fr->fshift[0];
 791     facel            = _mm256_set1_pd(fr->epsfac);
 792     charge           = mdatoms->chargeA;
 793     nvdwtype         = fr->ntype;
 794     vdwparam         = fr->nbfp;
 795     vdwtype          = mdatoms->typeA;
 796     vdwgridparam     = fr->ljpme_c6grid;
 797     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 798     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 799     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 800
 801     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 802     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 803     beta2            = _mm256_mul_pd(beta,beta);
 804     beta3            = _mm256_mul_pd(beta,beta2);
 805
 806     ewtab            = fr->ic->tabq_coul_F;
 807     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 808     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 809
 810     /* Setup water-specific parameters */
 811     inr              = nlist->iinr[0];
 812     iq0              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+0]));
 813     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 814     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 815     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 816     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 817
 818     /* Avoid stupid compiler warnings */
 819     jnrA = jnrB = jnrC = jnrD = 0;
 820     j_coord_offsetA = 0;
 821     j_coord_offsetB = 0;
 822     j_coord_offsetC = 0;
 823     j_coord_offsetD = 0;
 824
 825     outeriter        = 0;
 826     inneriter        = 0;
 827
 828     for(iidx=0;iidx<4*DIM;iidx++)
 829     {
 830         scratch[iidx] = 0.0;
 831     }
 832
 833     /* Start outer loop over neighborlists */
 834     for(iidx=0; iidx<nri; iidx++)
 835     {
 836         /* Load shift vector for this list */
 837         i_shift_offset   = DIM*shiftidx[iidx];
 838
 839         /* Load limits for loop over neighbors */
 840         j_index_start    = jindex[iidx];
 841         j_index_end      = jindex[iidx+1];
 842
 843         /* Get outer coordinate index */
 844         inr              = iinr[iidx];
 845         i_coord_offset   = DIM*inr;
 846
 847         /* Load i particle coords and add shift vector */
 848         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 849                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 850
 851         fix0             = _mm256_setzero_pd();
 852         fiy0             = _mm256_setzero_pd();
 853         fiz0             = _mm256_setzero_pd();
 854         fix1             = _mm256_setzero_pd();
 855         fiy1             = _mm256_setzero_pd();
 856         fiz1             = _mm256_setzero_pd();
 857         fix2             = _mm256_setzero_pd();
 858         fiy2             = _mm256_setzero_pd();
 859         fiz2             = _mm256_setzero_pd();
 860
 861         /* Start inner kernel loop */
 862         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 863         {
 864
 865             /* Get j neighbor index, and coordinate index */
 866             jnrA             = jjnr[jidx];
 867             jnrB             = jjnr[jidx+1];
 868             jnrC             = jjnr[jidx+2];
 869             jnrD             = jjnr[jidx+3];
 870             j_coord_offsetA  = DIM*jnrA;
 871             j_coord_offsetB  = DIM*jnrB;
 872             j_coord_offsetC  = DIM*jnrC;
 873             j_coord_offsetD  = DIM*jnrD;
 874
 875             /* load j atom coordinates */
 876             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 877                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 878                                                  &jx0,&jy0,&jz0);
 879
 880             /* Calculate displacement vector */
 881             dx00             = _mm256_sub_pd(ix0,jx0);
 882             dy00             = _mm256_sub_pd(iy0,jy0);
 883             dz00             = _mm256_sub_pd(iz0,jz0);
 884             dx10             = _mm256_sub_pd(ix1,jx0);
 885             dy10             = _mm256_sub_pd(iy1,jy0);
 886             dz10             = _mm256_sub_pd(iz1,jz0);
 887             dx20             = _mm256_sub_pd(ix2,jx0);
 888             dy20             = _mm256_sub_pd(iy2,jy0);
 889             dz20             = _mm256_sub_pd(iz2,jz0);
 890
 891             /* Calculate squared distance and things based on it */
 892             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 893             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 894             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 895
 896             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 897             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 898             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 899
 900             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 901             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 902             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 903
 904             /* Load parameters for j particles */
 905             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 906                                                                  charge+jnrC+0,charge+jnrD+0);
 907             vdwjidx0A        = 2*vdwtype[jnrA+0];
 908             vdwjidx0B        = 2*vdwtype[jnrB+0];
 909             vdwjidx0C        = 2*vdwtype[jnrC+0];
 910             vdwjidx0D        = 2*vdwtype[jnrD+0];
 911
 912             fjx0             = _mm256_setzero_pd();
 913             fjy0             = _mm256_setzero_pd();
 914             fjz0             = _mm256_setzero_pd();
 915
 916             /**************************
 917              * CALCULATE INTERACTIONS *
 918              **************************/
 919
 920             r00              = _mm256_mul_pd(rsq00,rinv00);
 921
 922             /* Compute parameters for interactions between i and j atoms */
 923             qq00             = _mm256_mul_pd(iq0,jq0);
 924             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 925                                             vdwioffsetptr0+vdwjidx0B,
 926                                             vdwioffsetptr0+vdwjidx0C,
 927                                             vdwioffsetptr0+vdwjidx0D,
 928                                             &c6_00,&c12_00);
 929
 930             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 931                                                                   vdwgridioffsetptr0+vdwjidx0B,
 932                                                                   vdwgridioffsetptr0+vdwjidx0C,
 933                                                                   vdwgridioffsetptr0+vdwjidx0D);
 934
 935             /* EWALD ELECTROSTATICS */
 936
 937             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 938             ewrt             = _mm256_mul_pd(r00,ewtabscale);
 939             ewitab           = _mm256_cvttpd_epi32(ewrt);
 940             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 941             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 942                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 943                                             &ewtabF,&ewtabFn);
 944             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 945             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
 946
 947             /* Analytical LJ-PME */
 948             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 949             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 950             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 951             exponent         = gmx_simd_exp_d(ewcljrsq);
 952             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 953             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 954             /* f6A = 6 * C6grid * (1 - poly) */
 955             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
 956             /* f6B = C6grid * exponent * beta^6 */
 957             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
 958             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 959             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);
 960
 961             fscal            = _mm256_add_pd(felec,fvdw);
 962
 963             /* Calculate temporary vectorial force */
 964             tx               = _mm256_mul_pd(fscal,dx00);
 965             ty               = _mm256_mul_pd(fscal,dy00);
 966             tz               = _mm256_mul_pd(fscal,dz00);
 967
 968             /* Update vectorial force */
 969             fix0             = _mm256_add_pd(fix0,tx);
 970             fiy0             = _mm256_add_pd(fiy0,ty);
 971             fiz0             = _mm256_add_pd(fiz0,tz);
 972
 973             fjx0             = _mm256_add_pd(fjx0,tx);
 974             fjy0             = _mm256_add_pd(fjy0,ty);
 975             fjz0             = _mm256_add_pd(fjz0,tz);
 976
 977             /**************************
 978              * CALCULATE INTERACTIONS *
 979              **************************/
 980
 981             r10              = _mm256_mul_pd(rsq10,rinv10);
 982
 983             /* Compute parameters for interactions between i and j atoms */
 984             qq10             = _mm256_mul_pd(iq1,jq0);
 985
 986             /* EWALD ELECTROSTATICS */
 987
 988             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 989             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 990             ewitab           = _mm256_cvttpd_epi32(ewrt);
 991             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 992             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 993                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 994                                             &ewtabF,&ewtabFn);
 995             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 996             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 997
 998             fscal            = felec;
 999
1000             /* Calculate temporary vectorial force */
1001             tx               = _mm256_mul_pd(fscal,dx10);
1002             ty               = _mm256_mul_pd(fscal,dy10);
1003             tz               = _mm256_mul_pd(fscal,dz10);
1004
1005             /* Update vectorial force */
1006             fix1             = _mm256_add_pd(fix1,tx);
1007             fiy1             = _mm256_add_pd(fiy1,ty);
1008             fiz1             = _mm256_add_pd(fiz1,tz);
1009
1010             fjx0             = _mm256_add_pd(fjx0,tx);
1011             fjy0             = _mm256_add_pd(fjy0,ty);
1012             fjz0             = _mm256_add_pd(fjz0,tz);
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             r20              = _mm256_mul_pd(rsq20,rinv20);
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq20             = _mm256_mul_pd(iq2,jq0);
1022
1023             /* EWALD ELECTROSTATICS */
1024
1025             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1026             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1027             ewitab           = _mm256_cvttpd_epi32(ewrt);
1028             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1029             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1030                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1031                                             &ewtabF,&ewtabFn);
1032             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1033             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1034
1035             fscal            = felec;
1036
1037             /* Calculate temporary vectorial force */
1038             tx               = _mm256_mul_pd(fscal,dx20);
1039             ty               = _mm256_mul_pd(fscal,dy20);
1040             tz               = _mm256_mul_pd(fscal,dz20);
1041
1042             /* Update vectorial force */
1043             fix2             = _mm256_add_pd(fix2,tx);
1044             fiy2             = _mm256_add_pd(fiy2,ty);
1045             fiz2             = _mm256_add_pd(fiz2,tz);
1046
1047             fjx0             = _mm256_add_pd(fjx0,tx);
1048             fjy0             = _mm256_add_pd(fjy0,ty);
1049             fjz0             = _mm256_add_pd(fjz0,tz);
1050
1051             fjptrA             = f+j_coord_offsetA;
1052             fjptrB             = f+j_coord_offsetB;
1053             fjptrC             = f+j_coord_offsetC;
1054             fjptrD             = f+j_coord_offsetD;
1055
1056             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1057
1058             /* Inner loop uses 134 flops */
1059         }
1060
1061         if(jidx<j_index_end)
1062         {
1063
1064             /* Get j neighbor index, and coordinate index */
1065             jnrlistA         = jjnr[jidx];
1066             jnrlistB         = jjnr[jidx+1];
1067             jnrlistC         = jjnr[jidx+2];
1068             jnrlistD         = jjnr[jidx+3];
1069             /* Sign of each element will be negative for non-real atoms.
1070              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1071              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1072              */
1073             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1074
1075             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1076             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1077             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1078
1079             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1080             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1081             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1082             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1083             j_coord_offsetA  = DIM*jnrA;
1084             j_coord_offsetB  = DIM*jnrB;
1085             j_coord_offsetC  = DIM*jnrC;
1086             j_coord_offsetD  = DIM*jnrD;
1087
1088             /* load j atom coordinates */
1089             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1090                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1091                                                  &jx0,&jy0,&jz0);
1092
1093             /* Calculate displacement vector */
1094             dx00             = _mm256_sub_pd(ix0,jx0);
1095             dy00             = _mm256_sub_pd(iy0,jy0);
1096             dz00             = _mm256_sub_pd(iz0,jz0);
1097             dx10             = _mm256_sub_pd(ix1,jx0);
1098             dy10             = _mm256_sub_pd(iy1,jy0);
1099             dz10             = _mm256_sub_pd(iz1,jz0);
1100             dx20             = _mm256_sub_pd(ix2,jx0);
1101             dy20             = _mm256_sub_pd(iy2,jy0);
1102             dz20             = _mm256_sub_pd(iz2,jz0);
1103
1104             /* Calculate squared distance and things based on it */
1105             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1106             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1107             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1108
1109             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1110             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1111             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1112
1113             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1114             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1115             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1116
1117             /* Load parameters for j particles */
1118             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1119                                                                  charge+jnrC+0,charge+jnrD+0);
1120             vdwjidx0A        = 2*vdwtype[jnrA+0];
1121             vdwjidx0B        = 2*vdwtype[jnrB+0];
1122             vdwjidx0C        = 2*vdwtype[jnrC+0];
1123             vdwjidx0D        = 2*vdwtype[jnrD+0];
1124
1125             fjx0             = _mm256_setzero_pd();
1126             fjy0             = _mm256_setzero_pd();
1127             fjz0             = _mm256_setzero_pd();
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             r00              = _mm256_mul_pd(rsq00,rinv00);
1134             r00              = _mm256_andnot_pd(dummy_mask,r00);
1135
1136             /* Compute parameters for interactions between i and j atoms */
1137             qq00             = _mm256_mul_pd(iq0,jq0);
1138             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1139                                             vdwioffsetptr0+vdwjidx0B,
1140                                             vdwioffsetptr0+vdwjidx0C,
1141                                             vdwioffsetptr0+vdwjidx0D,
1142                                             &c6_00,&c12_00);
1143
1144             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1145                                                                   vdwgridioffsetptr0+vdwjidx0B,
1146                                                                   vdwgridioffsetptr0+vdwjidx0C,
1147                                                                   vdwgridioffsetptr0+vdwjidx0D);
1148
1149             /* EWALD ELECTROSTATICS */
1150
1151             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1152             ewrt             = _mm256_mul_pd(r00,ewtabscale);
1153             ewitab           = _mm256_cvttpd_epi32(ewrt);
1154             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1155             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1156                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1157                                             &ewtabF,&ewtabFn);
1158             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1159             felec            = _mm256_mul_pd(_mm256_mul_pd(qq00,rinv00),_mm256_sub_pd(rinvsq00,felec));
1160
1161             /* Analytical LJ-PME */
1162             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1163             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1164             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1165             exponent         = gmx_simd_exp_d(ewcljrsq);
1166             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1167             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1168             /* f6A = 6 * C6grid * (1 - poly) */
1169             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1170             /* f6B = C6grid * exponent * beta^6 */
1171             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1172             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1173             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);
1174
1175             fscal            = _mm256_add_pd(felec,fvdw);
1176
1177             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1178
1179             /* Calculate temporary vectorial force */
1180             tx               = _mm256_mul_pd(fscal,dx00);
1181             ty               = _mm256_mul_pd(fscal,dy00);
1182             tz               = _mm256_mul_pd(fscal,dz00);
1183
1184             /* Update vectorial force */
1185             fix0             = _mm256_add_pd(fix0,tx);
1186             fiy0             = _mm256_add_pd(fiy0,ty);
1187             fiz0             = _mm256_add_pd(fiz0,tz);
1188
1189             fjx0             = _mm256_add_pd(fjx0,tx);
1190             fjy0             = _mm256_add_pd(fjy0,ty);
1191             fjz0             = _mm256_add_pd(fjz0,tz);
1192
1193             /**************************
1194              * CALCULATE INTERACTIONS *
1195              **************************/
1196
1197             r10              = _mm256_mul_pd(rsq10,rinv10);
1198             r10              = _mm256_andnot_pd(dummy_mask,r10);
1199
1200             /* Compute parameters for interactions between i and j atoms */
1201             qq10             = _mm256_mul_pd(iq1,jq0);
1202
1203             /* EWALD ELECTROSTATICS */
1204
1205             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1206             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1207             ewitab           = _mm256_cvttpd_epi32(ewrt);
1208             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1209             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1210                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1211                                             &ewtabF,&ewtabFn);
1212             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1213             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1214
1215             fscal            = felec;
1216
1217             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1218
1219             /* Calculate temporary vectorial force */
1220             tx               = _mm256_mul_pd(fscal,dx10);
1221             ty               = _mm256_mul_pd(fscal,dy10);
1222             tz               = _mm256_mul_pd(fscal,dz10);
1223
1224             /* Update vectorial force */
1225             fix1             = _mm256_add_pd(fix1,tx);
1226             fiy1             = _mm256_add_pd(fiy1,ty);
1227             fiz1             = _mm256_add_pd(fiz1,tz);
1228
1229             fjx0             = _mm256_add_pd(fjx0,tx);
1230             fjy0             = _mm256_add_pd(fjy0,ty);
1231             fjz0             = _mm256_add_pd(fjz0,tz);
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             r20              = _mm256_mul_pd(rsq20,rinv20);
1238             r20              = _mm256_andnot_pd(dummy_mask,r20);
1239
1240             /* Compute parameters for interactions between i and j atoms */
1241             qq20             = _mm256_mul_pd(iq2,jq0);
1242
1243             /* EWALD ELECTROSTATICS */
1244
1245             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1246             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1247             ewitab           = _mm256_cvttpd_epi32(ewrt);
1248             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1249             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1250                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1251                                             &ewtabF,&ewtabFn);
1252             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1253             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1254
1255             fscal            = felec;
1256
1257             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1258
1259             /* Calculate temporary vectorial force */
1260             tx               = _mm256_mul_pd(fscal,dx20);
1261             ty               = _mm256_mul_pd(fscal,dy20);
1262             tz               = _mm256_mul_pd(fscal,dz20);
1263
1264             /* Update vectorial force */
1265             fix2             = _mm256_add_pd(fix2,tx);
1266             fiy2             = _mm256_add_pd(fiy2,ty);
1267             fiz2             = _mm256_add_pd(fiz2,tz);
1268
1269             fjx0             = _mm256_add_pd(fjx0,tx);
1270             fjy0             = _mm256_add_pd(fjy0,ty);
1271             fjz0             = _mm256_add_pd(fjz0,tz);
1272
1273             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1274             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1275             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1276             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1277
1278             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1279
1280             /* Inner loop uses 137 flops */
1281         }
1282
1283         /* End of innermost loop */
1284
1285         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1286                                                  f+i_coord_offset,fshift+i_shift_offset);
1287
1288         /* Increment number of inner iterations */
1289         inneriter                  += j_index_end - j_index_start;
1290
1291         /* Outer loop uses 18 flops */
1292     }
1293
1294     /* Increment number of outer iterations */
1295     outeriter        += nri;
1296
1297     /* Update outer/inner flops */
1298
1299     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1300 }