src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEw_VdwLJEw_GeomW4P1_avx_256_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_avx_256_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_avx_256_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  79     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  84     real             scratch[4*DIM];
  85     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     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     real *           vdwioffsetptr3;
  96     real *           vdwgridioffsetptr3;
  97     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  98     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  99     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 100     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 101     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 102     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 103     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 104     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 105     real             *charge;
 106     int              nvdwtype;
 107     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 108     int              *vdwtype;
 109     real             *vdwparam;
 110     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 111     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 112     __m256d           c6grid_00;
 113     __m256d           c6grid_10;
 114     __m256d           c6grid_20;
 115     __m256d           c6grid_30;
 116     real             *vdwgridparam;
 117     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 118     __m256d           one_half  = _mm256_set1_pd(0.5);
 119     __m256d           minus_one = _mm256_set1_pd(-1.0);
 120     __m128i          ewitab;
 121     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 122     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 123     real             *ewtab;
 124     __m256d          dummy_mask,cutoff_mask;
 125     __m128           tmpmask0,tmpmask1;
 126     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 127     __m256d          one     = _mm256_set1_pd(1.0);
 128     __m256d          two     = _mm256_set1_pd(2.0);
 129     x                = xx[0];
 130     f                = ff[0];
 131
 132     nri              = nlist->nri;
 133     iinr             = nlist->iinr;
 134     jindex           = nlist->jindex;
 135     jjnr             = nlist->jjnr;
 136     shiftidx         = nlist->shift;
 137     gid              = nlist->gid;
 138     shiftvec         = fr->shift_vec[0];
 139     fshift           = fr->fshift[0];
 140     facel            = _mm256_set1_pd(fr->epsfac);
 141     charge           = mdatoms->chargeA;
 142     nvdwtype         = fr->ntype;
 143     vdwparam         = fr->nbfp;
 144     vdwtype          = mdatoms->typeA;
 145     vdwgridparam     = fr->ljpme_c6grid;
 146     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 147     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 148     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 149
 150     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 151     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 152     beta2            = _mm256_mul_pd(beta,beta);
 153     beta3            = _mm256_mul_pd(beta,beta2);
 154
 155     ewtab            = fr->ic->tabq_coul_FDV0;
 156     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 157     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 158
 159     /* Setup water-specific parameters */
 160     inr              = nlist->iinr[0];
 161     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 162     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 163     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 164     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 165     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 166
 167     /* Avoid stupid compiler warnings */
 168     jnrA = jnrB = jnrC = jnrD = 0;
 169     j_coord_offsetA = 0;
 170     j_coord_offsetB = 0;
 171     j_coord_offsetC = 0;
 172     j_coord_offsetD = 0;
 173
 174     outeriter        = 0;
 175     inneriter        = 0;
 176
 177     for(iidx=0;iidx<4*DIM;iidx++)
 178     {
 179         scratch[iidx] = 0.0;
 180     }
 181
 182     /* Start outer loop over neighborlists */
 183     for(iidx=0; iidx<nri; iidx++)
 184     {
 185         /* Load shift vector for this list */
 186         i_shift_offset   = DIM*shiftidx[iidx];
 187
 188         /* Load limits for loop over neighbors */
 189         j_index_start    = jindex[iidx];
 190         j_index_end      = jindex[iidx+1];
 191
 192         /* Get outer coordinate index */
 193         inr              = iinr[iidx];
 194         i_coord_offset   = DIM*inr;
 195
 196         /* Load i particle coords and add shift vector */
 197         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 198                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 199
 200         fix0             = _mm256_setzero_pd();
 201         fiy0             = _mm256_setzero_pd();
 202         fiz0             = _mm256_setzero_pd();
 203         fix1             = _mm256_setzero_pd();
 204         fiy1             = _mm256_setzero_pd();
 205         fiz1             = _mm256_setzero_pd();
 206         fix2             = _mm256_setzero_pd();
 207         fiy2             = _mm256_setzero_pd();
 208         fiz2             = _mm256_setzero_pd();
 209         fix3             = _mm256_setzero_pd();
 210         fiy3             = _mm256_setzero_pd();
 211         fiz3             = _mm256_setzero_pd();
 212
 213         /* Reset potential sums */
 214         velecsum         = _mm256_setzero_pd();
 215         vvdwsum          = _mm256_setzero_pd();
 216
 217         /* Start inner kernel loop */
 218         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 219         {
 220
 221             /* Get j neighbor index, and coordinate index */
 222             jnrA             = jjnr[jidx];
 223             jnrB             = jjnr[jidx+1];
 224             jnrC             = jjnr[jidx+2];
 225             jnrD             = jjnr[jidx+3];
 226             j_coord_offsetA  = DIM*jnrA;
 227             j_coord_offsetB  = DIM*jnrB;
 228             j_coord_offsetC  = DIM*jnrC;
 229             j_coord_offsetD  = DIM*jnrD;
 230
 231             /* load j atom coordinates */
 232             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 233                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 234                                                  &jx0,&jy0,&jz0);
 235
 236             /* Calculate displacement vector */
 237             dx00             = _mm256_sub_pd(ix0,jx0);
 238             dy00             = _mm256_sub_pd(iy0,jy0);
 239             dz00             = _mm256_sub_pd(iz0,jz0);
 240             dx10             = _mm256_sub_pd(ix1,jx0);
 241             dy10             = _mm256_sub_pd(iy1,jy0);
 242             dz10             = _mm256_sub_pd(iz1,jz0);
 243             dx20             = _mm256_sub_pd(ix2,jx0);
 244             dy20             = _mm256_sub_pd(iy2,jy0);
 245             dz20             = _mm256_sub_pd(iz2,jz0);
 246             dx30             = _mm256_sub_pd(ix3,jx0);
 247             dy30             = _mm256_sub_pd(iy3,jy0);
 248             dz30             = _mm256_sub_pd(iz3,jz0);
 249
 250             /* Calculate squared distance and things based on it */
 251             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 252             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 253             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 254             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 255
 256             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 257             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 258             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 259             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 260
 261             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 262             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 263             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 264             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 265
 266             /* Load parameters for j particles */
 267             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 268                                                                  charge+jnrC+0,charge+jnrD+0);
 269             vdwjidx0A        = 2*vdwtype[jnrA+0];
 270             vdwjidx0B        = 2*vdwtype[jnrB+0];
 271             vdwjidx0C        = 2*vdwtype[jnrC+0];
 272             vdwjidx0D        = 2*vdwtype[jnrD+0];
 273
 274             fjx0             = _mm256_setzero_pd();
 275             fjy0             = _mm256_setzero_pd();
 276             fjz0             = _mm256_setzero_pd();
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             r00              = _mm256_mul_pd(rsq00,rinv00);
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 286                                             vdwioffsetptr0+vdwjidx0B,
 287                                             vdwioffsetptr0+vdwjidx0C,
 288                                             vdwioffsetptr0+vdwjidx0D,
 289                                             &c6_00,&c12_00);
 290
 291             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 292                                                                   vdwgridioffsetptr0+vdwjidx0B,
 293                                                                   vdwgridioffsetptr0+vdwjidx0C,
 294                                                                   vdwgridioffsetptr0+vdwjidx0D);
 295
 296             /* Analytical LJ-PME */
 297             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 298             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 299             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 300             exponent         = gmx_simd_exp_d(ewcljrsq);
 301             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 302             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 303             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 304             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 305             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 306             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 307             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 308             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);
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 312
 313             fscal            = fvdw;
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm256_mul_pd(fscal,dx00);
 317             ty               = _mm256_mul_pd(fscal,dy00);
 318             tz               = _mm256_mul_pd(fscal,dz00);
 319
 320             /* Update vectorial force */
 321             fix0             = _mm256_add_pd(fix0,tx);
 322             fiy0             = _mm256_add_pd(fiy0,ty);
 323             fiz0             = _mm256_add_pd(fiz0,tz);
 324
 325             fjx0             = _mm256_add_pd(fjx0,tx);
 326             fjy0             = _mm256_add_pd(fjy0,ty);
 327             fjz0             = _mm256_add_pd(fjz0,tz);
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             r10              = _mm256_mul_pd(rsq10,rinv10);
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq10             = _mm256_mul_pd(iq1,jq0);
 337
 338             /* EWALD ELECTROSTATICS */
 339
 340             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 341             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 342             ewitab           = _mm256_cvttpd_epi32(ewrt);
 343             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 344             ewitab           = _mm_slli_epi32(ewitab,2);
 345             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 346             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 347             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 348             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 349             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 350             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 351             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 352             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 353             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm256_add_pd(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm256_mul_pd(fscal,dx10);
 362             ty               = _mm256_mul_pd(fscal,dy10);
 363             tz               = _mm256_mul_pd(fscal,dz10);
 364
 365             /* Update vectorial force */
 366             fix1             = _mm256_add_pd(fix1,tx);
 367             fiy1             = _mm256_add_pd(fiy1,ty);
 368             fiz1             = _mm256_add_pd(fiz1,tz);
 369
 370             fjx0             = _mm256_add_pd(fjx0,tx);
 371             fjy0             = _mm256_add_pd(fjy0,ty);
 372             fjz0             = _mm256_add_pd(fjz0,tz);
 373
 374             /**************************
 375              * CALCULATE INTERACTIONS *
 376              **************************/
 377
 378             r20              = _mm256_mul_pd(rsq20,rinv20);
 379
 380             /* Compute parameters for interactions between i and j atoms */
 381             qq20             = _mm256_mul_pd(iq2,jq0);
 382
 383             /* EWALD ELECTROSTATICS */
 384
 385             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 386             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 387             ewitab           = _mm256_cvttpd_epi32(ewrt);
 388             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 389             ewitab           = _mm_slli_epi32(ewitab,2);
 390             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 391             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 392             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 393             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 394             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 395             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 396             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 397             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 398             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velecsum         = _mm256_add_pd(velecsum,velec);
 402
 403             fscal            = felec;
 404
 405             /* Calculate temporary vectorial force */
 406             tx               = _mm256_mul_pd(fscal,dx20);
 407             ty               = _mm256_mul_pd(fscal,dy20);
 408             tz               = _mm256_mul_pd(fscal,dz20);
 409
 410             /* Update vectorial force */
 411             fix2             = _mm256_add_pd(fix2,tx);
 412             fiy2             = _mm256_add_pd(fiy2,ty);
 413             fiz2             = _mm256_add_pd(fiz2,tz);
 414
 415             fjx0             = _mm256_add_pd(fjx0,tx);
 416             fjy0             = _mm256_add_pd(fjy0,ty);
 417             fjz0             = _mm256_add_pd(fjz0,tz);
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             r30              = _mm256_mul_pd(rsq30,rinv30);
 424
 425             /* Compute parameters for interactions between i and j atoms */
 426             qq30             = _mm256_mul_pd(iq3,jq0);
 427
 428             /* EWALD ELECTROSTATICS */
 429
 430             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 431             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 432             ewitab           = _mm256_cvttpd_epi32(ewrt);
 433             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 434             ewitab           = _mm_slli_epi32(ewitab,2);
 435             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 436             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 437             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 438             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 439             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 440             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 441             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 442             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 443             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 444
 445             /* Update potential sum for this i atom from the interaction with this j atom. */
 446             velecsum         = _mm256_add_pd(velecsum,velec);
 447
 448             fscal            = felec;
 449
 450             /* Calculate temporary vectorial force */
 451             tx               = _mm256_mul_pd(fscal,dx30);
 452             ty               = _mm256_mul_pd(fscal,dy30);
 453             tz               = _mm256_mul_pd(fscal,dz30);
 454
 455             /* Update vectorial force */
 456             fix3             = _mm256_add_pd(fix3,tx);
 457             fiy3             = _mm256_add_pd(fiy3,ty);
 458             fiz3             = _mm256_add_pd(fiz3,tz);
 459
 460             fjx0             = _mm256_add_pd(fjx0,tx);
 461             fjy0             = _mm256_add_pd(fjy0,ty);
 462             fjz0             = _mm256_add_pd(fjz0,tz);
 463
 464             fjptrA             = f+j_coord_offsetA;
 465             fjptrB             = f+j_coord_offsetB;
 466             fjptrC             = f+j_coord_offsetC;
 467             fjptrD             = f+j_coord_offsetD;
 468
 469             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 470
 471             /* Inner loop uses 180 flops */
 472         }
 473
 474         if(jidx<j_index_end)
 475         {
 476
 477             /* Get j neighbor index, and coordinate index */
 478             jnrlistA         = jjnr[jidx];
 479             jnrlistB         = jjnr[jidx+1];
 480             jnrlistC         = jjnr[jidx+2];
 481             jnrlistD         = jjnr[jidx+3];
 482             /* Sign of each element will be negative for non-real atoms.
 483              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 484              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 485              */
 486             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 487
 488             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 489             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 490             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 491
 492             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 493             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 494             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 495             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 496             j_coord_offsetA  = DIM*jnrA;
 497             j_coord_offsetB  = DIM*jnrB;
 498             j_coord_offsetC  = DIM*jnrC;
 499             j_coord_offsetD  = DIM*jnrD;
 500
 501             /* load j atom coordinates */
 502             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 503                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 504                                                  &jx0,&jy0,&jz0);
 505
 506             /* Calculate displacement vector */
 507             dx00             = _mm256_sub_pd(ix0,jx0);
 508             dy00             = _mm256_sub_pd(iy0,jy0);
 509             dz00             = _mm256_sub_pd(iz0,jz0);
 510             dx10             = _mm256_sub_pd(ix1,jx0);
 511             dy10             = _mm256_sub_pd(iy1,jy0);
 512             dz10             = _mm256_sub_pd(iz1,jz0);
 513             dx20             = _mm256_sub_pd(ix2,jx0);
 514             dy20             = _mm256_sub_pd(iy2,jy0);
 515             dz20             = _mm256_sub_pd(iz2,jz0);
 516             dx30             = _mm256_sub_pd(ix3,jx0);
 517             dy30             = _mm256_sub_pd(iy3,jy0);
 518             dz30             = _mm256_sub_pd(iz3,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 522             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 523             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 524             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 525
 526             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 527             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 528             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 529             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 530
 531             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 532             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 533             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 534             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 535
 536             /* Load parameters for j particles */
 537             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 538                                                                  charge+jnrC+0,charge+jnrD+0);
 539             vdwjidx0A        = 2*vdwtype[jnrA+0];
 540             vdwjidx0B        = 2*vdwtype[jnrB+0];
 541             vdwjidx0C        = 2*vdwtype[jnrC+0];
 542             vdwjidx0D        = 2*vdwtype[jnrD+0];
 543
 544             fjx0             = _mm256_setzero_pd();
 545             fjy0             = _mm256_setzero_pd();
 546             fjz0             = _mm256_setzero_pd();
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             r00              = _mm256_mul_pd(rsq00,rinv00);
 553             r00              = _mm256_andnot_pd(dummy_mask,r00);
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 557                                             vdwioffsetptr0+vdwjidx0B,
 558                                             vdwioffsetptr0+vdwjidx0C,
 559                                             vdwioffsetptr0+vdwjidx0D,
 560                                             &c6_00,&c12_00);
 561
 562             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 563                                                                   vdwgridioffsetptr0+vdwjidx0B,
 564                                                                   vdwgridioffsetptr0+vdwjidx0C,
 565                                                                   vdwgridioffsetptr0+vdwjidx0D);
 566
 567             /* Analytical LJ-PME */
 568             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 569             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 570             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 571             exponent         = gmx_simd_exp_d(ewcljrsq);
 572             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 573             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 574             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 575             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 576             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 577             vvdw             = _mm256_sub_pd(_mm256_mul_pd(vvdw12,one_twelfth),_mm256_mul_pd(vvdw6,one_sixth));
 578             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 579             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);
 580
 581             /* Update potential sum for this i atom from the interaction with this j atom. */
 582             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 583             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 584
 585             fscal            = fvdw;
 586
 587             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 588
 589             /* Calculate temporary vectorial force */
 590             tx               = _mm256_mul_pd(fscal,dx00);
 591             ty               = _mm256_mul_pd(fscal,dy00);
 592             tz               = _mm256_mul_pd(fscal,dz00);
 593
 594             /* Update vectorial force */
 595             fix0             = _mm256_add_pd(fix0,tx);
 596             fiy0             = _mm256_add_pd(fiy0,ty);
 597             fiz0             = _mm256_add_pd(fiz0,tz);
 598
 599             fjx0             = _mm256_add_pd(fjx0,tx);
 600             fjy0             = _mm256_add_pd(fjy0,ty);
 601             fjz0             = _mm256_add_pd(fjz0,tz);
 602
 603             /**************************
 604              * CALCULATE INTERACTIONS *
 605              **************************/
 606
 607             r10              = _mm256_mul_pd(rsq10,rinv10);
 608             r10              = _mm256_andnot_pd(dummy_mask,r10);
 609
 610             /* Compute parameters for interactions between i and j atoms */
 611             qq10             = _mm256_mul_pd(iq1,jq0);
 612
 613             /* EWALD ELECTROSTATICS */
 614
 615             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 616             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 617             ewitab           = _mm256_cvttpd_epi32(ewrt);
 618             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 619             ewitab           = _mm_slli_epi32(ewitab,2);
 620             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 621             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 622             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 623             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 624             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 625             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 626             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 627             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(rinv10,velec));
 628             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 629
 630             /* Update potential sum for this i atom from the interaction with this j atom. */
 631             velec            = _mm256_andnot_pd(dummy_mask,velec);
 632             velecsum         = _mm256_add_pd(velecsum,velec);
 633
 634             fscal            = felec;
 635
 636             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 637
 638             /* Calculate temporary vectorial force */
 639             tx               = _mm256_mul_pd(fscal,dx10);
 640             ty               = _mm256_mul_pd(fscal,dy10);
 641             tz               = _mm256_mul_pd(fscal,dz10);
 642
 643             /* Update vectorial force */
 644             fix1             = _mm256_add_pd(fix1,tx);
 645             fiy1             = _mm256_add_pd(fiy1,ty);
 646             fiz1             = _mm256_add_pd(fiz1,tz);
 647
 648             fjx0             = _mm256_add_pd(fjx0,tx);
 649             fjy0             = _mm256_add_pd(fjy0,ty);
 650             fjz0             = _mm256_add_pd(fjz0,tz);
 651
 652             /**************************
 653              * CALCULATE INTERACTIONS *
 654              **************************/
 655
 656             r20              = _mm256_mul_pd(rsq20,rinv20);
 657             r20              = _mm256_andnot_pd(dummy_mask,r20);
 658
 659             /* Compute parameters for interactions between i and j atoms */
 660             qq20             = _mm256_mul_pd(iq2,jq0);
 661
 662             /* EWALD ELECTROSTATICS */
 663
 664             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 665             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 666             ewitab           = _mm256_cvttpd_epi32(ewrt);
 667             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 668             ewitab           = _mm_slli_epi32(ewitab,2);
 669             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 670             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 671             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 672             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 673             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 674             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 675             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 676             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(rinv20,velec));
 677             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 678
 679             /* Update potential sum for this i atom from the interaction with this j atom. */
 680             velec            = _mm256_andnot_pd(dummy_mask,velec);
 681             velecsum         = _mm256_add_pd(velecsum,velec);
 682
 683             fscal            = felec;
 684
 685             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 686
 687             /* Calculate temporary vectorial force */
 688             tx               = _mm256_mul_pd(fscal,dx20);
 689             ty               = _mm256_mul_pd(fscal,dy20);
 690             tz               = _mm256_mul_pd(fscal,dz20);
 691
 692             /* Update vectorial force */
 693             fix2             = _mm256_add_pd(fix2,tx);
 694             fiy2             = _mm256_add_pd(fiy2,ty);
 695             fiz2             = _mm256_add_pd(fiz2,tz);
 696
 697             fjx0             = _mm256_add_pd(fjx0,tx);
 698             fjy0             = _mm256_add_pd(fjy0,ty);
 699             fjz0             = _mm256_add_pd(fjz0,tz);
 700
 701             /**************************
 702              * CALCULATE INTERACTIONS *
 703              **************************/
 704
 705             r30              = _mm256_mul_pd(rsq30,rinv30);
 706             r30              = _mm256_andnot_pd(dummy_mask,r30);
 707
 708             /* Compute parameters for interactions between i and j atoms */
 709             qq30             = _mm256_mul_pd(iq3,jq0);
 710
 711             /* EWALD ELECTROSTATICS */
 712
 713             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 714             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 715             ewitab           = _mm256_cvttpd_epi32(ewrt);
 716             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 717             ewitab           = _mm_slli_epi32(ewitab,2);
 718             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 719             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 720             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 721             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 722             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 723             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 724             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 725             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(rinv30,velec));
 726             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 727
 728             /* Update potential sum for this i atom from the interaction with this j atom. */
 729             velec            = _mm256_andnot_pd(dummy_mask,velec);
 730             velecsum         = _mm256_add_pd(velecsum,velec);
 731
 732             fscal            = felec;
 733
 734             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 735
 736             /* Calculate temporary vectorial force */
 737             tx               = _mm256_mul_pd(fscal,dx30);
 738             ty               = _mm256_mul_pd(fscal,dy30);
 739             tz               = _mm256_mul_pd(fscal,dz30);
 740
 741             /* Update vectorial force */
 742             fix3             = _mm256_add_pd(fix3,tx);
 743             fiy3             = _mm256_add_pd(fiy3,ty);
 744             fiz3             = _mm256_add_pd(fiz3,tz);
 745
 746             fjx0             = _mm256_add_pd(fjx0,tx);
 747             fjy0             = _mm256_add_pd(fjy0,ty);
 748             fjz0             = _mm256_add_pd(fjz0,tz);
 749
 750             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 751             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 752             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 753             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 754
 755             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 756
 757             /* Inner loop uses 184 flops */
 758         }
 759
 760         /* End of innermost loop */
 761
 762         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 763                                                  f+i_coord_offset,fshift+i_shift_offset);
 764
 765         ggid                        = gid[iidx];
 766         /* Update potential energies */
 767         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 768         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 769
 770         /* Increment number of inner iterations */
 771         inneriter                  += j_index_end - j_index_start;
 772
 773         /* Outer loop uses 26 flops */
 774     }
 775
 776     /* Increment number of outer iterations */
 777     outeriter        += nri;
 778
 779     /* Update outer/inner flops */
 780
 781     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*184);
 782 }
 783 /*
 784  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_256_double
 785  * Electrostatics interaction: Ewald
 786  * VdW interaction:            LJEwald
 787  * Geometry:                   Water4-Particle
 788  * Calculate force/pot:        Force
 789  */
 790 void
 791 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_256_double
 792                     (t_nblist                    * gmx_restrict       nlist,
 793                      rvec                        * gmx_restrict          xx,
 794                      rvec                        * gmx_restrict          ff,
 795                      t_forcerec                  * gmx_restrict          fr,
 796                      t_mdatoms                   * gmx_restrict     mdatoms,
 797                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 798                      t_nrnb                      * gmx_restrict        nrnb)
 799 {
 800     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 801      * just 0 for non-waters.
 802      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 803      * jnr indices corresponding to data put in the four positions in the SIMD register.
 804      */
 805     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 806     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 807     int              jnrA,jnrB,jnrC,jnrD;
 808     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 809     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 810     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 811     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 812     real             rcutoff_scalar;
 813     real             *shiftvec,*fshift,*x,*f;
 814     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 815     real             scratch[4*DIM];
 816     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 817     real *           vdwioffsetptr0;
 818     real *           vdwgridioffsetptr0;
 819     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 820     real *           vdwioffsetptr1;
 821     real *           vdwgridioffsetptr1;
 822     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 823     real *           vdwioffsetptr2;
 824     real *           vdwgridioffsetptr2;
 825     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 826     real *           vdwioffsetptr3;
 827     real *           vdwgridioffsetptr3;
 828     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 829     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 830     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 831     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 832     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 833     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 834     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 835     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 836     real             *charge;
 837     int              nvdwtype;
 838     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 839     int              *vdwtype;
 840     real             *vdwparam;
 841     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 842     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 843     __m256d           c6grid_00;
 844     __m256d           c6grid_10;
 845     __m256d           c6grid_20;
 846     __m256d           c6grid_30;
 847     real             *vdwgridparam;
 848     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 849     __m256d           one_half  = _mm256_set1_pd(0.5);
 850     __m256d           minus_one = _mm256_set1_pd(-1.0);
 851     __m128i          ewitab;
 852     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 853     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 854     real             *ewtab;
 855     __m256d          dummy_mask,cutoff_mask;
 856     __m128           tmpmask0,tmpmask1;
 857     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 858     __m256d          one     = _mm256_set1_pd(1.0);
 859     __m256d          two     = _mm256_set1_pd(2.0);
 860     x                = xx[0];
 861     f                = ff[0];
 862
 863     nri              = nlist->nri;
 864     iinr             = nlist->iinr;
 865     jindex           = nlist->jindex;
 866     jjnr             = nlist->jjnr;
 867     shiftidx         = nlist->shift;
 868     gid              = nlist->gid;
 869     shiftvec         = fr->shift_vec[0];
 870     fshift           = fr->fshift[0];
 871     facel            = _mm256_set1_pd(fr->epsfac);
 872     charge           = mdatoms->chargeA;
 873     nvdwtype         = fr->ntype;
 874     vdwparam         = fr->nbfp;
 875     vdwtype          = mdatoms->typeA;
 876     vdwgridparam     = fr->ljpme_c6grid;
 877     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 878     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 879     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 880
 881     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 882     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 883     beta2            = _mm256_mul_pd(beta,beta);
 884     beta3            = _mm256_mul_pd(beta,beta2);
 885
 886     ewtab            = fr->ic->tabq_coul_F;
 887     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 888     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 889
 890     /* Setup water-specific parameters */
 891     inr              = nlist->iinr[0];
 892     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 893     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 894     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 895     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 896     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 897
 898     /* Avoid stupid compiler warnings */
 899     jnrA = jnrB = jnrC = jnrD = 0;
 900     j_coord_offsetA = 0;
 901     j_coord_offsetB = 0;
 902     j_coord_offsetC = 0;
 903     j_coord_offsetD = 0;
 904
 905     outeriter        = 0;
 906     inneriter        = 0;
 907
 908     for(iidx=0;iidx<4*DIM;iidx++)
 909     {
 910         scratch[iidx] = 0.0;
 911     }
 912
 913     /* Start outer loop over neighborlists */
 914     for(iidx=0; iidx<nri; iidx++)
 915     {
 916         /* Load shift vector for this list */
 917         i_shift_offset   = DIM*shiftidx[iidx];
 918
 919         /* Load limits for loop over neighbors */
 920         j_index_start    = jindex[iidx];
 921         j_index_end      = jindex[iidx+1];
 922
 923         /* Get outer coordinate index */
 924         inr              = iinr[iidx];
 925         i_coord_offset   = DIM*inr;
 926
 927         /* Load i particle coords and add shift vector */
 928         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 929                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 930
 931         fix0             = _mm256_setzero_pd();
 932         fiy0             = _mm256_setzero_pd();
 933         fiz0             = _mm256_setzero_pd();
 934         fix1             = _mm256_setzero_pd();
 935         fiy1             = _mm256_setzero_pd();
 936         fiz1             = _mm256_setzero_pd();
 937         fix2             = _mm256_setzero_pd();
 938         fiy2             = _mm256_setzero_pd();
 939         fiz2             = _mm256_setzero_pd();
 940         fix3             = _mm256_setzero_pd();
 941         fiy3             = _mm256_setzero_pd();
 942         fiz3             = _mm256_setzero_pd();
 943
 944         /* Start inner kernel loop */
 945         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 946         {
 947
 948             /* Get j neighbor index, and coordinate index */
 949             jnrA             = jjnr[jidx];
 950             jnrB             = jjnr[jidx+1];
 951             jnrC             = jjnr[jidx+2];
 952             jnrD             = jjnr[jidx+3];
 953             j_coord_offsetA  = DIM*jnrA;
 954             j_coord_offsetB  = DIM*jnrB;
 955             j_coord_offsetC  = DIM*jnrC;
 956             j_coord_offsetD  = DIM*jnrD;
 957
 958             /* load j atom coordinates */
 959             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 960                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 961                                                  &jx0,&jy0,&jz0);
 962
 963             /* Calculate displacement vector */
 964             dx00             = _mm256_sub_pd(ix0,jx0);
 965             dy00             = _mm256_sub_pd(iy0,jy0);
 966             dz00             = _mm256_sub_pd(iz0,jz0);
 967             dx10             = _mm256_sub_pd(ix1,jx0);
 968             dy10             = _mm256_sub_pd(iy1,jy0);
 969             dz10             = _mm256_sub_pd(iz1,jz0);
 970             dx20             = _mm256_sub_pd(ix2,jx0);
 971             dy20             = _mm256_sub_pd(iy2,jy0);
 972             dz20             = _mm256_sub_pd(iz2,jz0);
 973             dx30             = _mm256_sub_pd(ix3,jx0);
 974             dy30             = _mm256_sub_pd(iy3,jy0);
 975             dz30             = _mm256_sub_pd(iz3,jz0);
 976
 977             /* Calculate squared distance and things based on it */
 978             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 979             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 980             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 981             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 982
 983             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 984             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 985             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 986             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 987
 988             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 989             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 990             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 991             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 992
 993             /* Load parameters for j particles */
 994             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 995                                                                  charge+jnrC+0,charge+jnrD+0);
 996             vdwjidx0A        = 2*vdwtype[jnrA+0];
 997             vdwjidx0B        = 2*vdwtype[jnrB+0];
 998             vdwjidx0C        = 2*vdwtype[jnrC+0];
 999             vdwjidx0D        = 2*vdwtype[jnrD+0];
1000
1001             fjx0             = _mm256_setzero_pd();
1002             fjy0             = _mm256_setzero_pd();
1003             fjz0             = _mm256_setzero_pd();
1004
1005             /**************************
1006              * CALCULATE INTERACTIONS *
1007              **************************/
1008
1009             r00              = _mm256_mul_pd(rsq00,rinv00);
1010
1011             /* Compute parameters for interactions between i and j atoms */
1012             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1013                                             vdwioffsetptr0+vdwjidx0B,
1014                                             vdwioffsetptr0+vdwjidx0C,
1015                                             vdwioffsetptr0+vdwjidx0D,
1016                                             &c6_00,&c12_00);
1017
1018             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1019                                                                   vdwgridioffsetptr0+vdwjidx0B,
1020                                                                   vdwgridioffsetptr0+vdwjidx0C,
1021                                                                   vdwgridioffsetptr0+vdwjidx0D);
1022
1023             /* Analytical LJ-PME */
1024             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1025             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1026             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1027             exponent         = gmx_simd_exp_d(ewcljrsq);
1028             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1029             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1030             /* f6A = 6 * C6grid * (1 - poly) */
1031             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1032             /* f6B = C6grid * exponent * beta^6 */
1033             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1034             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1035             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);
1036
1037             fscal            = fvdw;
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm256_mul_pd(fscal,dx00);
1041             ty               = _mm256_mul_pd(fscal,dy00);
1042             tz               = _mm256_mul_pd(fscal,dz00);
1043
1044             /* Update vectorial force */
1045             fix0             = _mm256_add_pd(fix0,tx);
1046             fiy0             = _mm256_add_pd(fiy0,ty);
1047             fiz0             = _mm256_add_pd(fiz0,tz);
1048
1049             fjx0             = _mm256_add_pd(fjx0,tx);
1050             fjy0             = _mm256_add_pd(fjy0,ty);
1051             fjz0             = _mm256_add_pd(fjz0,tz);
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             r10              = _mm256_mul_pd(rsq10,rinv10);
1058
1059             /* Compute parameters for interactions between i and j atoms */
1060             qq10             = _mm256_mul_pd(iq1,jq0);
1061
1062             /* EWALD ELECTROSTATICS */
1063
1064             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1065             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1066             ewitab           = _mm256_cvttpd_epi32(ewrt);
1067             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1068             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1069                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1070                                             &ewtabF,&ewtabFn);
1071             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1072             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1073
1074             fscal            = felec;
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm256_mul_pd(fscal,dx10);
1078             ty               = _mm256_mul_pd(fscal,dy10);
1079             tz               = _mm256_mul_pd(fscal,dz10);
1080
1081             /* Update vectorial force */
1082             fix1             = _mm256_add_pd(fix1,tx);
1083             fiy1             = _mm256_add_pd(fiy1,ty);
1084             fiz1             = _mm256_add_pd(fiz1,tz);
1085
1086             fjx0             = _mm256_add_pd(fjx0,tx);
1087             fjy0             = _mm256_add_pd(fjy0,ty);
1088             fjz0             = _mm256_add_pd(fjz0,tz);
1089
1090             /**************************
1091              * CALCULATE INTERACTIONS *
1092              **************************/
1093
1094             r20              = _mm256_mul_pd(rsq20,rinv20);
1095
1096             /* Compute parameters for interactions between i and j atoms */
1097             qq20             = _mm256_mul_pd(iq2,jq0);
1098
1099             /* EWALD ELECTROSTATICS */
1100
1101             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1102             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1103             ewitab           = _mm256_cvttpd_epi32(ewrt);
1104             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1105             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1106                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1107                                             &ewtabF,&ewtabFn);
1108             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1109             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1110
1111             fscal            = felec;
1112
1113             /* Calculate temporary vectorial force */
1114             tx               = _mm256_mul_pd(fscal,dx20);
1115             ty               = _mm256_mul_pd(fscal,dy20);
1116             tz               = _mm256_mul_pd(fscal,dz20);
1117
1118             /* Update vectorial force */
1119             fix2             = _mm256_add_pd(fix2,tx);
1120             fiy2             = _mm256_add_pd(fiy2,ty);
1121             fiz2             = _mm256_add_pd(fiz2,tz);
1122
1123             fjx0             = _mm256_add_pd(fjx0,tx);
1124             fjy0             = _mm256_add_pd(fjy0,ty);
1125             fjz0             = _mm256_add_pd(fjz0,tz);
1126
1127             /**************************
1128              * CALCULATE INTERACTIONS *
1129              **************************/
1130
1131             r30              = _mm256_mul_pd(rsq30,rinv30);
1132
1133             /* Compute parameters for interactions between i and j atoms */
1134             qq30             = _mm256_mul_pd(iq3,jq0);
1135
1136             /* EWALD ELECTROSTATICS */
1137
1138             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1139             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1140             ewitab           = _mm256_cvttpd_epi32(ewrt);
1141             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1142             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1143                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1144                                             &ewtabF,&ewtabFn);
1145             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1146             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1147
1148             fscal            = felec;
1149
1150             /* Calculate temporary vectorial force */
1151             tx               = _mm256_mul_pd(fscal,dx30);
1152             ty               = _mm256_mul_pd(fscal,dy30);
1153             tz               = _mm256_mul_pd(fscal,dz30);
1154
1155             /* Update vectorial force */
1156             fix3             = _mm256_add_pd(fix3,tx);
1157             fiy3             = _mm256_add_pd(fiy3,ty);
1158             fiz3             = _mm256_add_pd(fiz3,tz);
1159
1160             fjx0             = _mm256_add_pd(fjx0,tx);
1161             fjy0             = _mm256_add_pd(fjy0,ty);
1162             fjz0             = _mm256_add_pd(fjz0,tz);
1163
1164             fjptrA             = f+j_coord_offsetA;
1165             fjptrB             = f+j_coord_offsetB;
1166             fjptrC             = f+j_coord_offsetC;
1167             fjptrD             = f+j_coord_offsetD;
1168
1169             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1170
1171             /* Inner loop uses 157 flops */
1172         }
1173
1174         if(jidx<j_index_end)
1175         {
1176
1177             /* Get j neighbor index, and coordinate index */
1178             jnrlistA         = jjnr[jidx];
1179             jnrlistB         = jjnr[jidx+1];
1180             jnrlistC         = jjnr[jidx+2];
1181             jnrlistD         = jjnr[jidx+3];
1182             /* Sign of each element will be negative for non-real atoms.
1183              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1184              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1185              */
1186             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1187
1188             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1189             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1190             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1191
1192             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1193             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1194             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1195             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1196             j_coord_offsetA  = DIM*jnrA;
1197             j_coord_offsetB  = DIM*jnrB;
1198             j_coord_offsetC  = DIM*jnrC;
1199             j_coord_offsetD  = DIM*jnrD;
1200
1201             /* load j atom coordinates */
1202             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1203                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1204                                                  &jx0,&jy0,&jz0);
1205
1206             /* Calculate displacement vector */
1207             dx00             = _mm256_sub_pd(ix0,jx0);
1208             dy00             = _mm256_sub_pd(iy0,jy0);
1209             dz00             = _mm256_sub_pd(iz0,jz0);
1210             dx10             = _mm256_sub_pd(ix1,jx0);
1211             dy10             = _mm256_sub_pd(iy1,jy0);
1212             dz10             = _mm256_sub_pd(iz1,jz0);
1213             dx20             = _mm256_sub_pd(ix2,jx0);
1214             dy20             = _mm256_sub_pd(iy2,jy0);
1215             dz20             = _mm256_sub_pd(iz2,jz0);
1216             dx30             = _mm256_sub_pd(ix3,jx0);
1217             dy30             = _mm256_sub_pd(iy3,jy0);
1218             dz30             = _mm256_sub_pd(iz3,jz0);
1219
1220             /* Calculate squared distance and things based on it */
1221             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1222             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1223             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1224             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1225
1226             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1227             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1228             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1229             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1230
1231             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1232             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1233             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1234             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1235
1236             /* Load parameters for j particles */
1237             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1238                                                                  charge+jnrC+0,charge+jnrD+0);
1239             vdwjidx0A        = 2*vdwtype[jnrA+0];
1240             vdwjidx0B        = 2*vdwtype[jnrB+0];
1241             vdwjidx0C        = 2*vdwtype[jnrC+0];
1242             vdwjidx0D        = 2*vdwtype[jnrD+0];
1243
1244             fjx0             = _mm256_setzero_pd();
1245             fjy0             = _mm256_setzero_pd();
1246             fjz0             = _mm256_setzero_pd();
1247
1248             /**************************
1249              * CALCULATE INTERACTIONS *
1250              **************************/
1251
1252             r00              = _mm256_mul_pd(rsq00,rinv00);
1253             r00              = _mm256_andnot_pd(dummy_mask,r00);
1254
1255             /* Compute parameters for interactions between i and j atoms */
1256             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1257                                             vdwioffsetptr0+vdwjidx0B,
1258                                             vdwioffsetptr0+vdwjidx0C,
1259                                             vdwioffsetptr0+vdwjidx0D,
1260                                             &c6_00,&c12_00);
1261
1262             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1263                                                                   vdwgridioffsetptr0+vdwjidx0B,
1264                                                                   vdwgridioffsetptr0+vdwjidx0C,
1265                                                                   vdwgridioffsetptr0+vdwjidx0D);
1266
1267             /* Analytical LJ-PME */
1268             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1269             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1270             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1271             exponent         = gmx_simd_exp_d(ewcljrsq);
1272             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1273             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1274             /* f6A = 6 * C6grid * (1 - poly) */
1275             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1276             /* f6B = C6grid * exponent * beta^6 */
1277             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1278             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1279             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);
1280
1281             fscal            = fvdw;
1282
1283             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1284
1285             /* Calculate temporary vectorial force */
1286             tx               = _mm256_mul_pd(fscal,dx00);
1287             ty               = _mm256_mul_pd(fscal,dy00);
1288             tz               = _mm256_mul_pd(fscal,dz00);
1289
1290             /* Update vectorial force */
1291             fix0             = _mm256_add_pd(fix0,tx);
1292             fiy0             = _mm256_add_pd(fiy0,ty);
1293             fiz0             = _mm256_add_pd(fiz0,tz);
1294
1295             fjx0             = _mm256_add_pd(fjx0,tx);
1296             fjy0             = _mm256_add_pd(fjy0,ty);
1297             fjz0             = _mm256_add_pd(fjz0,tz);
1298
1299             /**************************
1300              * CALCULATE INTERACTIONS *
1301              **************************/
1302
1303             r10              = _mm256_mul_pd(rsq10,rinv10);
1304             r10              = _mm256_andnot_pd(dummy_mask,r10);
1305
1306             /* Compute parameters for interactions between i and j atoms */
1307             qq10             = _mm256_mul_pd(iq1,jq0);
1308
1309             /* EWALD ELECTROSTATICS */
1310
1311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1312             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1313             ewitab           = _mm256_cvttpd_epi32(ewrt);
1314             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1315             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1316                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1317                                             &ewtabF,&ewtabFn);
1318             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1319             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1320
1321             fscal            = felec;
1322
1323             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1324
1325             /* Calculate temporary vectorial force */
1326             tx               = _mm256_mul_pd(fscal,dx10);
1327             ty               = _mm256_mul_pd(fscal,dy10);
1328             tz               = _mm256_mul_pd(fscal,dz10);
1329
1330             /* Update vectorial force */
1331             fix1             = _mm256_add_pd(fix1,tx);
1332             fiy1             = _mm256_add_pd(fiy1,ty);
1333             fiz1             = _mm256_add_pd(fiz1,tz);
1334
1335             fjx0             = _mm256_add_pd(fjx0,tx);
1336             fjy0             = _mm256_add_pd(fjy0,ty);
1337             fjz0             = _mm256_add_pd(fjz0,tz);
1338
1339             /**************************
1340              * CALCULATE INTERACTIONS *
1341              **************************/
1342
1343             r20              = _mm256_mul_pd(rsq20,rinv20);
1344             r20              = _mm256_andnot_pd(dummy_mask,r20);
1345
1346             /* Compute parameters for interactions between i and j atoms */
1347             qq20             = _mm256_mul_pd(iq2,jq0);
1348
1349             /* EWALD ELECTROSTATICS */
1350
1351             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1352             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1353             ewitab           = _mm256_cvttpd_epi32(ewrt);
1354             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1355             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1356                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1357                                             &ewtabF,&ewtabFn);
1358             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1359             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1360
1361             fscal            = felec;
1362
1363             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1364
1365             /* Calculate temporary vectorial force */
1366             tx               = _mm256_mul_pd(fscal,dx20);
1367             ty               = _mm256_mul_pd(fscal,dy20);
1368             tz               = _mm256_mul_pd(fscal,dz20);
1369
1370             /* Update vectorial force */
1371             fix2             = _mm256_add_pd(fix2,tx);
1372             fiy2             = _mm256_add_pd(fiy2,ty);
1373             fiz2             = _mm256_add_pd(fiz2,tz);
1374
1375             fjx0             = _mm256_add_pd(fjx0,tx);
1376             fjy0             = _mm256_add_pd(fjy0,ty);
1377             fjz0             = _mm256_add_pd(fjz0,tz);
1378
1379             /**************************
1380              * CALCULATE INTERACTIONS *
1381              **************************/
1382
1383             r30              = _mm256_mul_pd(rsq30,rinv30);
1384             r30              = _mm256_andnot_pd(dummy_mask,r30);
1385
1386             /* Compute parameters for interactions between i and j atoms */
1387             qq30             = _mm256_mul_pd(iq3,jq0);
1388
1389             /* EWALD ELECTROSTATICS */
1390
1391             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1392             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1393             ewitab           = _mm256_cvttpd_epi32(ewrt);
1394             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1395             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1396                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1397                                             &ewtabF,&ewtabFn);
1398             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1399             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1400
1401             fscal            = felec;
1402
1403             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1404
1405             /* Calculate temporary vectorial force */
1406             tx               = _mm256_mul_pd(fscal,dx30);
1407             ty               = _mm256_mul_pd(fscal,dy30);
1408             tz               = _mm256_mul_pd(fscal,dz30);
1409
1410             /* Update vectorial force */
1411             fix3             = _mm256_add_pd(fix3,tx);
1412             fiy3             = _mm256_add_pd(fiy3,ty);
1413             fiz3             = _mm256_add_pd(fiz3,tz);
1414
1415             fjx0             = _mm256_add_pd(fjx0,tx);
1416             fjy0             = _mm256_add_pd(fjy0,ty);
1417             fjz0             = _mm256_add_pd(fjz0,tz);
1418
1419             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1420             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1421             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1422             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1423
1424             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1425
1426             /* Inner loop uses 161 flops */
1427         }
1428
1429         /* End of innermost loop */
1430
1431         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1432                                                  f+i_coord_offset,fshift+i_shift_offset);
1433
1434         /* Increment number of inner iterations */
1435         inneriter                  += j_index_end - j_index_start;
1436
1437         /* Outer loop uses 24 flops */
1438     }
1439
1440     /* Increment number of outer iterations */
1441     outeriter        += nri;
1442
1443     /* Update outer/inner flops */
1444
1445     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*161);
1446 }