src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJEwSh_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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_VF_avx_256_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  82     real             scratch[4*DIM];
  83     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  84     real *           vdwioffsetptr0;
  85     real *           vdwgridioffsetptr0;
  86     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     real *           vdwgridioffsetptr1;
  89     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  90     real *           vdwioffsetptr2;
  91     real *           vdwgridioffsetptr2;
  92     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  93     real *           vdwioffsetptr3;
  94     real *           vdwgridioffsetptr3;
  95     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  96     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  97     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  98     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  99     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 100     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 101     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 102     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 103     real             *charge;
 104     int              nvdwtype;
 105     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 106     int              *vdwtype;
 107     real             *vdwparam;
 108     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 109     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 110     __m256d           c6grid_00;
 111     __m256d           c6grid_10;
 112     __m256d           c6grid_20;
 113     __m256d           c6grid_30;
 114     real             *vdwgridparam;
 115     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 116     __m256d           one_half  = _mm256_set1_pd(0.5);
 117     __m256d           minus_one = _mm256_set1_pd(-1.0);
 118     __m128i          ewitab;
 119     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 120     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 121     real             *ewtab;
 122     __m256d          dummy_mask,cutoff_mask;
 123     __m128           tmpmask0,tmpmask1;
 124     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 125     __m256d          one     = _mm256_set1_pd(1.0);
 126     __m256d          two     = _mm256_set1_pd(2.0);
 127     x                = xx[0];
 128     f                = ff[0];
 129
 130     nri              = nlist->nri;
 131     iinr             = nlist->iinr;
 132     jindex           = nlist->jindex;
 133     jjnr             = nlist->jjnr;
 134     shiftidx         = nlist->shift;
 135     gid              = nlist->gid;
 136     shiftvec         = fr->shift_vec[0];
 137     fshift           = fr->fshift[0];
 138     facel            = _mm256_set1_pd(fr->epsfac);
 139     charge           = mdatoms->chargeA;
 140     nvdwtype         = fr->ntype;
 141     vdwparam         = fr->nbfp;
 142     vdwtype          = mdatoms->typeA;
 143     vdwgridparam     = fr->ljpme_c6grid;
 144     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 145     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 146     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 147
 148     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 149     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 150     beta2            = _mm256_mul_pd(beta,beta);
 151     beta3            = _mm256_mul_pd(beta,beta2);
 152
 153     ewtab            = fr->ic->tabq_coul_FDV0;
 154     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 155     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 156
 157     /* Setup water-specific parameters */
 158     inr              = nlist->iinr[0];
 159     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 160     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 161     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 162     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 163     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 164
 165     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 166     rcutoff_scalar   = fr->rcoulomb;
 167     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 168     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 169
 170     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 171     rvdw             = _mm256_set1_pd(fr->rvdw);
 172
 173     /* Avoid stupid compiler warnings */
 174     jnrA = jnrB = jnrC = jnrD = 0;
 175     j_coord_offsetA = 0;
 176     j_coord_offsetB = 0;
 177     j_coord_offsetC = 0;
 178     j_coord_offsetD = 0;
 179
 180     outeriter        = 0;
 181     inneriter        = 0;
 182
 183     for(iidx=0;iidx<4*DIM;iidx++)
 184     {
 185         scratch[iidx] = 0.0;
 186     }
 187
 188     /* Start outer loop over neighborlists */
 189     for(iidx=0; iidx<nri; iidx++)
 190     {
 191         /* Load shift vector for this list */
 192         i_shift_offset   = DIM*shiftidx[iidx];
 193
 194         /* Load limits for loop over neighbors */
 195         j_index_start    = jindex[iidx];
 196         j_index_end      = jindex[iidx+1];
 197
 198         /* Get outer coordinate index */
 199         inr              = iinr[iidx];
 200         i_coord_offset   = DIM*inr;
 201
 202         /* Load i particle coords and add shift vector */
 203         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 204                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 205
 206         fix0             = _mm256_setzero_pd();
 207         fiy0             = _mm256_setzero_pd();
 208         fiz0             = _mm256_setzero_pd();
 209         fix1             = _mm256_setzero_pd();
 210         fiy1             = _mm256_setzero_pd();
 211         fiz1             = _mm256_setzero_pd();
 212         fix2             = _mm256_setzero_pd();
 213         fiy2             = _mm256_setzero_pd();
 214         fiz2             = _mm256_setzero_pd();
 215         fix3             = _mm256_setzero_pd();
 216         fiy3             = _mm256_setzero_pd();
 217         fiz3             = _mm256_setzero_pd();
 218
 219         /* Reset potential sums */
 220         velecsum         = _mm256_setzero_pd();
 221         vvdwsum          = _mm256_setzero_pd();
 222
 223         /* Start inner kernel loop */
 224         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 225         {
 226
 227             /* Get j neighbor index, and coordinate index */
 228             jnrA             = jjnr[jidx];
 229             jnrB             = jjnr[jidx+1];
 230             jnrC             = jjnr[jidx+2];
 231             jnrD             = jjnr[jidx+3];
 232             j_coord_offsetA  = DIM*jnrA;
 233             j_coord_offsetB  = DIM*jnrB;
 234             j_coord_offsetC  = DIM*jnrC;
 235             j_coord_offsetD  = DIM*jnrD;
 236
 237             /* load j atom coordinates */
 238             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 239                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 240                                                  &jx0,&jy0,&jz0);
 241
 242             /* Calculate displacement vector */
 243             dx00             = _mm256_sub_pd(ix0,jx0);
 244             dy00             = _mm256_sub_pd(iy0,jy0);
 245             dz00             = _mm256_sub_pd(iz0,jz0);
 246             dx10             = _mm256_sub_pd(ix1,jx0);
 247             dy10             = _mm256_sub_pd(iy1,jy0);
 248             dz10             = _mm256_sub_pd(iz1,jz0);
 249             dx20             = _mm256_sub_pd(ix2,jx0);
 250             dy20             = _mm256_sub_pd(iy2,jy0);
 251             dz20             = _mm256_sub_pd(iz2,jz0);
 252             dx30             = _mm256_sub_pd(ix3,jx0);
 253             dy30             = _mm256_sub_pd(iy3,jy0);
 254             dz30             = _mm256_sub_pd(iz3,jz0);
 255
 256             /* Calculate squared distance and things based on it */
 257             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 258             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 259             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 260             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 261
 262             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 263             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 264             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 265             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 266
 267             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 268             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 269             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 270             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 271
 272             /* Load parameters for j particles */
 273             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 274                                                                  charge+jnrC+0,charge+jnrD+0);
 275             vdwjidx0A        = 2*vdwtype[jnrA+0];
 276             vdwjidx0B        = 2*vdwtype[jnrB+0];
 277             vdwjidx0C        = 2*vdwtype[jnrC+0];
 278             vdwjidx0D        = 2*vdwtype[jnrD+0];
 279
 280             fjx0             = _mm256_setzero_pd();
 281             fjy0             = _mm256_setzero_pd();
 282             fjz0             = _mm256_setzero_pd();
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 289             {
 290
 291             r00              = _mm256_mul_pd(rsq00,rinv00);
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 295                                             vdwioffsetptr0+vdwjidx0B,
 296                                             vdwioffsetptr0+vdwjidx0C,
 297                                             vdwioffsetptr0+vdwjidx0D,
 298                                             &c6_00,&c12_00);
 299
 300             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 301                                                                   vdwgridioffsetptr0+vdwjidx0B,
 302                                                                   vdwgridioffsetptr0+vdwjidx0C,
 303                                                                   vdwgridioffsetptr0+vdwjidx0D);
 304
 305             /* Analytical LJ-PME */
 306             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 307             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 308             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 309             exponent         = gmx_simd_exp_d(ewcljrsq);
 310             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 311             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 312             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 313             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 314             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 315             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 316                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_add_pd(_mm256_mul_pd(c6_00,sh_vdw_invrcut6),_mm256_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 317             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 318             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);
 319
 320             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 321
 322             /* Update potential sum for this i atom from the interaction with this j atom. */
 323             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 324             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 325
 326             fscal            = fvdw;
 327
 328             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 329
 330             /* Calculate temporary vectorial force */
 331             tx               = _mm256_mul_pd(fscal,dx00);
 332             ty               = _mm256_mul_pd(fscal,dy00);
 333             tz               = _mm256_mul_pd(fscal,dz00);
 334
 335             /* Update vectorial force */
 336             fix0             = _mm256_add_pd(fix0,tx);
 337             fiy0             = _mm256_add_pd(fiy0,ty);
 338             fiz0             = _mm256_add_pd(fiz0,tz);
 339
 340             fjx0             = _mm256_add_pd(fjx0,tx);
 341             fjy0             = _mm256_add_pd(fjy0,ty);
 342             fjz0             = _mm256_add_pd(fjz0,tz);
 343
 344             }
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 351             {
 352
 353             r10              = _mm256_mul_pd(rsq10,rinv10);
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq10             = _mm256_mul_pd(iq1,jq0);
 357
 358             /* EWALD ELECTROSTATICS */
 359
 360             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 361             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 362             ewitab           = _mm256_cvttpd_epi32(ewrt);
 363             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 364             ewitab           = _mm_slli_epi32(ewitab,2);
 365             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 366             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 367             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 368             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 369             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 370             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 371             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 372             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 373             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 374
 375             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm256_and_pd(velec,cutoff_mask);
 379             velecsum         = _mm256_add_pd(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm256_mul_pd(fscal,dx10);
 387             ty               = _mm256_mul_pd(fscal,dy10);
 388             tz               = _mm256_mul_pd(fscal,dz10);
 389
 390             /* Update vectorial force */
 391             fix1             = _mm256_add_pd(fix1,tx);
 392             fiy1             = _mm256_add_pd(fiy1,ty);
 393             fiz1             = _mm256_add_pd(fiz1,tz);
 394
 395             fjx0             = _mm256_add_pd(fjx0,tx);
 396             fjy0             = _mm256_add_pd(fjy0,ty);
 397             fjz0             = _mm256_add_pd(fjz0,tz);
 398
 399             }
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 406             {
 407
 408             r20              = _mm256_mul_pd(rsq20,rinv20);
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq20             = _mm256_mul_pd(iq2,jq0);
 412
 413             /* EWALD ELECTROSTATICS */
 414
 415             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 416             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 417             ewitab           = _mm256_cvttpd_epi32(ewrt);
 418             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 419             ewitab           = _mm_slli_epi32(ewitab,2);
 420             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 421             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 422             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 423             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 424             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 425             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 426             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 427             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 428             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 429
 430             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 431
 432             /* Update potential sum for this i atom from the interaction with this j atom. */
 433             velec            = _mm256_and_pd(velec,cutoff_mask);
 434             velecsum         = _mm256_add_pd(velecsum,velec);
 435
 436             fscal            = felec;
 437
 438             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 439
 440             /* Calculate temporary vectorial force */
 441             tx               = _mm256_mul_pd(fscal,dx20);
 442             ty               = _mm256_mul_pd(fscal,dy20);
 443             tz               = _mm256_mul_pd(fscal,dz20);
 444
 445             /* Update vectorial force */
 446             fix2             = _mm256_add_pd(fix2,tx);
 447             fiy2             = _mm256_add_pd(fiy2,ty);
 448             fiz2             = _mm256_add_pd(fiz2,tz);
 449
 450             fjx0             = _mm256_add_pd(fjx0,tx);
 451             fjy0             = _mm256_add_pd(fjy0,ty);
 452             fjz0             = _mm256_add_pd(fjz0,tz);
 453
 454             }
 455
 456             /**************************
 457              * CALCULATE INTERACTIONS *
 458              **************************/
 459
 460             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 461             {
 462
 463             r30              = _mm256_mul_pd(rsq30,rinv30);
 464
 465             /* Compute parameters for interactions between i and j atoms */
 466             qq30             = _mm256_mul_pd(iq3,jq0);
 467
 468             /* EWALD ELECTROSTATICS */
 469
 470             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 471             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 472             ewitab           = _mm256_cvttpd_epi32(ewrt);
 473             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 474             ewitab           = _mm_slli_epi32(ewitab,2);
 475             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 476             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 477             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 478             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 479             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 480             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 481             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 482             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 483             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 484
 485             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 486
 487             /* Update potential sum for this i atom from the interaction with this j atom. */
 488             velec            = _mm256_and_pd(velec,cutoff_mask);
 489             velecsum         = _mm256_add_pd(velecsum,velec);
 490
 491             fscal            = felec;
 492
 493             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 494
 495             /* Calculate temporary vectorial force */
 496             tx               = _mm256_mul_pd(fscal,dx30);
 497             ty               = _mm256_mul_pd(fscal,dy30);
 498             tz               = _mm256_mul_pd(fscal,dz30);
 499
 500             /* Update vectorial force */
 501             fix3             = _mm256_add_pd(fix3,tx);
 502             fiy3             = _mm256_add_pd(fiy3,ty);
 503             fiz3             = _mm256_add_pd(fiz3,tz);
 504
 505             fjx0             = _mm256_add_pd(fjx0,tx);
 506             fjy0             = _mm256_add_pd(fjy0,ty);
 507             fjz0             = _mm256_add_pd(fjz0,tz);
 508
 509             }
 510
 511             fjptrA             = f+j_coord_offsetA;
 512             fjptrB             = f+j_coord_offsetB;
 513             fjptrC             = f+j_coord_offsetC;
 514             fjptrD             = f+j_coord_offsetD;
 515
 516             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 517
 518             /* Inner loop uses 203 flops */
 519         }
 520
 521         if(jidx<j_index_end)
 522         {
 523
 524             /* Get j neighbor index, and coordinate index */
 525             jnrlistA         = jjnr[jidx];
 526             jnrlistB         = jjnr[jidx+1];
 527             jnrlistC         = jjnr[jidx+2];
 528             jnrlistD         = jjnr[jidx+3];
 529             /* Sign of each element will be negative for non-real atoms.
 530              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 531              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 532              */
 533             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 534
 535             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 536             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 537             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 538
 539             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 540             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 541             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 542             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 543             j_coord_offsetA  = DIM*jnrA;
 544             j_coord_offsetB  = DIM*jnrB;
 545             j_coord_offsetC  = DIM*jnrC;
 546             j_coord_offsetD  = DIM*jnrD;
 547
 548             /* load j atom coordinates */
 549             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 550                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 551                                                  &jx0,&jy0,&jz0);
 552
 553             /* Calculate displacement vector */
 554             dx00             = _mm256_sub_pd(ix0,jx0);
 555             dy00             = _mm256_sub_pd(iy0,jy0);
 556             dz00             = _mm256_sub_pd(iz0,jz0);
 557             dx10             = _mm256_sub_pd(ix1,jx0);
 558             dy10             = _mm256_sub_pd(iy1,jy0);
 559             dz10             = _mm256_sub_pd(iz1,jz0);
 560             dx20             = _mm256_sub_pd(ix2,jx0);
 561             dy20             = _mm256_sub_pd(iy2,jy0);
 562             dz20             = _mm256_sub_pd(iz2,jz0);
 563             dx30             = _mm256_sub_pd(ix3,jx0);
 564             dy30             = _mm256_sub_pd(iy3,jy0);
 565             dz30             = _mm256_sub_pd(iz3,jz0);
 566
 567             /* Calculate squared distance and things based on it */
 568             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 569             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 570             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 571             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 572
 573             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 574             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 575             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 576             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 577
 578             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 579             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 580             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 581             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 582
 583             /* Load parameters for j particles */
 584             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 585                                                                  charge+jnrC+0,charge+jnrD+0);
 586             vdwjidx0A        = 2*vdwtype[jnrA+0];
 587             vdwjidx0B        = 2*vdwtype[jnrB+0];
 588             vdwjidx0C        = 2*vdwtype[jnrC+0];
 589             vdwjidx0D        = 2*vdwtype[jnrD+0];
 590
 591             fjx0             = _mm256_setzero_pd();
 592             fjy0             = _mm256_setzero_pd();
 593             fjz0             = _mm256_setzero_pd();
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 600             {
 601
 602             r00              = _mm256_mul_pd(rsq00,rinv00);
 603             r00              = _mm256_andnot_pd(dummy_mask,r00);
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 607                                             vdwioffsetptr0+vdwjidx0B,
 608                                             vdwioffsetptr0+vdwjidx0C,
 609                                             vdwioffsetptr0+vdwjidx0D,
 610                                             &c6_00,&c12_00);
 611
 612             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 613                                                                   vdwgridioffsetptr0+vdwjidx0B,
 614                                                                   vdwgridioffsetptr0+vdwjidx0C,
 615                                                                   vdwgridioffsetptr0+vdwjidx0D);
 616
 617             /* Analytical LJ-PME */
 618             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 619             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 620             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 621             exponent         = gmx_simd_exp_d(ewcljrsq);
 622             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 623             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 624             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 625             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 626             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 627             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_00,_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 628                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_add_pd(_mm256_mul_pd(c6_00,sh_vdw_invrcut6),_mm256_mul_pd(c6grid_00,sh_lj_ewald))),one_sixth));
 629             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 630             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);
 631
 632             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 636             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 637             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 638
 639             fscal            = fvdw;
 640
 641             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 642
 643             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 644
 645             /* Calculate temporary vectorial force */
 646             tx               = _mm256_mul_pd(fscal,dx00);
 647             ty               = _mm256_mul_pd(fscal,dy00);
 648             tz               = _mm256_mul_pd(fscal,dz00);
 649
 650             /* Update vectorial force */
 651             fix0             = _mm256_add_pd(fix0,tx);
 652             fiy0             = _mm256_add_pd(fiy0,ty);
 653             fiz0             = _mm256_add_pd(fiz0,tz);
 654
 655             fjx0             = _mm256_add_pd(fjx0,tx);
 656             fjy0             = _mm256_add_pd(fjy0,ty);
 657             fjz0             = _mm256_add_pd(fjz0,tz);
 658
 659             }
 660
 661             /**************************
 662              * CALCULATE INTERACTIONS *
 663              **************************/
 664
 665             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 666             {
 667
 668             r10              = _mm256_mul_pd(rsq10,rinv10);
 669             r10              = _mm256_andnot_pd(dummy_mask,r10);
 670
 671             /* Compute parameters for interactions between i and j atoms */
 672             qq10             = _mm256_mul_pd(iq1,jq0);
 673
 674             /* EWALD ELECTROSTATICS */
 675
 676             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 677             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 678             ewitab           = _mm256_cvttpd_epi32(ewrt);
 679             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 680             ewitab           = _mm_slli_epi32(ewitab,2);
 681             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 682             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 683             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 684             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 685             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 686             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 687             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 688             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 689             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 690
 691             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 692
 693             /* Update potential sum for this i atom from the interaction with this j atom. */
 694             velec            = _mm256_and_pd(velec,cutoff_mask);
 695             velec            = _mm256_andnot_pd(dummy_mask,velec);
 696             velecsum         = _mm256_add_pd(velecsum,velec);
 697
 698             fscal            = felec;
 699
 700             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 701
 702             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 703
 704             /* Calculate temporary vectorial force */
 705             tx               = _mm256_mul_pd(fscal,dx10);
 706             ty               = _mm256_mul_pd(fscal,dy10);
 707             tz               = _mm256_mul_pd(fscal,dz10);
 708
 709             /* Update vectorial force */
 710             fix1             = _mm256_add_pd(fix1,tx);
 711             fiy1             = _mm256_add_pd(fiy1,ty);
 712             fiz1             = _mm256_add_pd(fiz1,tz);
 713
 714             fjx0             = _mm256_add_pd(fjx0,tx);
 715             fjy0             = _mm256_add_pd(fjy0,ty);
 716             fjz0             = _mm256_add_pd(fjz0,tz);
 717
 718             }
 719
 720             /**************************
 721              * CALCULATE INTERACTIONS *
 722              **************************/
 723
 724             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 725             {
 726
 727             r20              = _mm256_mul_pd(rsq20,rinv20);
 728             r20              = _mm256_andnot_pd(dummy_mask,r20);
 729
 730             /* Compute parameters for interactions between i and j atoms */
 731             qq20             = _mm256_mul_pd(iq2,jq0);
 732
 733             /* EWALD ELECTROSTATICS */
 734
 735             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 736             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 737             ewitab           = _mm256_cvttpd_epi32(ewrt);
 738             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 739             ewitab           = _mm_slli_epi32(ewitab,2);
 740             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 741             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 742             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 743             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 744             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 745             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 746             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 747             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 748             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 749
 750             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 751
 752             /* Update potential sum for this i atom from the interaction with this j atom. */
 753             velec            = _mm256_and_pd(velec,cutoff_mask);
 754             velec            = _mm256_andnot_pd(dummy_mask,velec);
 755             velecsum         = _mm256_add_pd(velecsum,velec);
 756
 757             fscal            = felec;
 758
 759             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 760
 761             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 762
 763             /* Calculate temporary vectorial force */
 764             tx               = _mm256_mul_pd(fscal,dx20);
 765             ty               = _mm256_mul_pd(fscal,dy20);
 766             tz               = _mm256_mul_pd(fscal,dz20);
 767
 768             /* Update vectorial force */
 769             fix2             = _mm256_add_pd(fix2,tx);
 770             fiy2             = _mm256_add_pd(fiy2,ty);
 771             fiz2             = _mm256_add_pd(fiz2,tz);
 772
 773             fjx0             = _mm256_add_pd(fjx0,tx);
 774             fjy0             = _mm256_add_pd(fjy0,ty);
 775             fjz0             = _mm256_add_pd(fjz0,tz);
 776
 777             }
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 784             {
 785
 786             r30              = _mm256_mul_pd(rsq30,rinv30);
 787             r30              = _mm256_andnot_pd(dummy_mask,r30);
 788
 789             /* Compute parameters for interactions between i and j atoms */
 790             qq30             = _mm256_mul_pd(iq3,jq0);
 791
 792             /* EWALD ELECTROSTATICS */
 793
 794             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 795             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 796             ewitab           = _mm256_cvttpd_epi32(ewrt);
 797             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 798             ewitab           = _mm_slli_epi32(ewitab,2);
 799             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 800             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 801             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 802             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 803             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 804             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 805             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 806             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 807             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 808
 809             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 810
 811             /* Update potential sum for this i atom from the interaction with this j atom. */
 812             velec            = _mm256_and_pd(velec,cutoff_mask);
 813             velec            = _mm256_andnot_pd(dummy_mask,velec);
 814             velecsum         = _mm256_add_pd(velecsum,velec);
 815
 816             fscal            = felec;
 817
 818             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 819
 820             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 821
 822             /* Calculate temporary vectorial force */
 823             tx               = _mm256_mul_pd(fscal,dx30);
 824             ty               = _mm256_mul_pd(fscal,dy30);
 825             tz               = _mm256_mul_pd(fscal,dz30);
 826
 827             /* Update vectorial force */
 828             fix3             = _mm256_add_pd(fix3,tx);
 829             fiy3             = _mm256_add_pd(fiy3,ty);
 830             fiz3             = _mm256_add_pd(fiz3,tz);
 831
 832             fjx0             = _mm256_add_pd(fjx0,tx);
 833             fjy0             = _mm256_add_pd(fjy0,ty);
 834             fjz0             = _mm256_add_pd(fjz0,tz);
 835
 836             }
 837
 838             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 839             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 840             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 841             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 842
 843             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 844
 845             /* Inner loop uses 207 flops */
 846         }
 847
 848         /* End of innermost loop */
 849
 850         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 851                                                  f+i_coord_offset,fshift+i_shift_offset);
 852
 853         ggid                        = gid[iidx];
 854         /* Update potential energies */
 855         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 856         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 857
 858         /* Increment number of inner iterations */
 859         inneriter                  += j_index_end - j_index_start;
 860
 861         /* Outer loop uses 26 flops */
 862     }
 863
 864     /* Increment number of outer iterations */
 865     outeriter        += nri;
 866
 867     /* Update outer/inner flops */
 868
 869     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*207);
 870 }
 871 /*
 872  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_256_double
 873  * Electrostatics interaction: Ewald
 874  * VdW interaction:            LJEwald
 875  * Geometry:                   Water4-Particle
 876  * Calculate force/pot:        Force
 877  */
 878 void
 879 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_256_double
 880                     (t_nblist                    * gmx_restrict       nlist,
 881                      rvec                        * gmx_restrict          xx,
 882                      rvec                        * gmx_restrict          ff,
 883                      t_forcerec                  * gmx_restrict          fr,
 884                      t_mdatoms                   * gmx_restrict     mdatoms,
 885                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 886                      t_nrnb                      * gmx_restrict        nrnb)
 887 {
 888     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 889      * just 0 for non-waters.
 890      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 891      * jnr indices corresponding to data put in the four positions in the SIMD register.
 892      */
 893     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 894     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 895     int              jnrA,jnrB,jnrC,jnrD;
 896     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 897     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 898     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 899     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 900     real             rcutoff_scalar;
 901     real             *shiftvec,*fshift,*x,*f;
 902     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 903     real             scratch[4*DIM];
 904     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 905     real *           vdwioffsetptr0;
 906     real *           vdwgridioffsetptr0;
 907     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 908     real *           vdwioffsetptr1;
 909     real *           vdwgridioffsetptr1;
 910     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 911     real *           vdwioffsetptr2;
 912     real *           vdwgridioffsetptr2;
 913     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 914     real *           vdwioffsetptr3;
 915     real *           vdwgridioffsetptr3;
 916     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 917     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 918     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 919     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 920     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 921     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 922     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 923     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 924     real             *charge;
 925     int              nvdwtype;
 926     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 927     int              *vdwtype;
 928     real             *vdwparam;
 929     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 930     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 931     __m256d           c6grid_00;
 932     __m256d           c6grid_10;
 933     __m256d           c6grid_20;
 934     __m256d           c6grid_30;
 935     real             *vdwgridparam;
 936     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 937     __m256d           one_half  = _mm256_set1_pd(0.5);
 938     __m256d           minus_one = _mm256_set1_pd(-1.0);
 939     __m128i          ewitab;
 940     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 941     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 942     real             *ewtab;
 943     __m256d          dummy_mask,cutoff_mask;
 944     __m128           tmpmask0,tmpmask1;
 945     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 946     __m256d          one     = _mm256_set1_pd(1.0);
 947     __m256d          two     = _mm256_set1_pd(2.0);
 948     x                = xx[0];
 949     f                = ff[0];
 950
 951     nri              = nlist->nri;
 952     iinr             = nlist->iinr;
 953     jindex           = nlist->jindex;
 954     jjnr             = nlist->jjnr;
 955     shiftidx         = nlist->shift;
 956     gid              = nlist->gid;
 957     shiftvec         = fr->shift_vec[0];
 958     fshift           = fr->fshift[0];
 959     facel            = _mm256_set1_pd(fr->epsfac);
 960     charge           = mdatoms->chargeA;
 961     nvdwtype         = fr->ntype;
 962     vdwparam         = fr->nbfp;
 963     vdwtype          = mdatoms->typeA;
 964     vdwgridparam     = fr->ljpme_c6grid;
 965     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 966     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 967     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 968
 969     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 970     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 971     beta2            = _mm256_mul_pd(beta,beta);
 972     beta3            = _mm256_mul_pd(beta,beta2);
 973
 974     ewtab            = fr->ic->tabq_coul_F;
 975     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 976     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 977
 978     /* Setup water-specific parameters */
 979     inr              = nlist->iinr[0];
 980     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 981     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 982     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 983     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 984     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 985
 986     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 987     rcutoff_scalar   = fr->rcoulomb;
 988     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 989     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 990
 991     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 992     rvdw             = _mm256_set1_pd(fr->rvdw);
 993
 994     /* Avoid stupid compiler warnings */
 995     jnrA = jnrB = jnrC = jnrD = 0;
 996     j_coord_offsetA = 0;
 997     j_coord_offsetB = 0;
 998     j_coord_offsetC = 0;
 999     j_coord_offsetD = 0;
1000
1001     outeriter        = 0;
1002     inneriter        = 0;
1003
1004     for(iidx=0;iidx<4*DIM;iidx++)
1005     {
1006         scratch[iidx] = 0.0;
1007     }
1008
1009     /* Start outer loop over neighborlists */
1010     for(iidx=0; iidx<nri; iidx++)
1011     {
1012         /* Load shift vector for this list */
1013         i_shift_offset   = DIM*shiftidx[iidx];
1014
1015         /* Load limits for loop over neighbors */
1016         j_index_start    = jindex[iidx];
1017         j_index_end      = jindex[iidx+1];
1018
1019         /* Get outer coordinate index */
1020         inr              = iinr[iidx];
1021         i_coord_offset   = DIM*inr;
1022
1023         /* Load i particle coords and add shift vector */
1024         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1025                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1026
1027         fix0             = _mm256_setzero_pd();
1028         fiy0             = _mm256_setzero_pd();
1029         fiz0             = _mm256_setzero_pd();
1030         fix1             = _mm256_setzero_pd();
1031         fiy1             = _mm256_setzero_pd();
1032         fiz1             = _mm256_setzero_pd();
1033         fix2             = _mm256_setzero_pd();
1034         fiy2             = _mm256_setzero_pd();
1035         fiz2             = _mm256_setzero_pd();
1036         fix3             = _mm256_setzero_pd();
1037         fiy3             = _mm256_setzero_pd();
1038         fiz3             = _mm256_setzero_pd();
1039
1040         /* Start inner kernel loop */
1041         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1042         {
1043
1044             /* Get j neighbor index, and coordinate index */
1045             jnrA             = jjnr[jidx];
1046             jnrB             = jjnr[jidx+1];
1047             jnrC             = jjnr[jidx+2];
1048             jnrD             = jjnr[jidx+3];
1049             j_coord_offsetA  = DIM*jnrA;
1050             j_coord_offsetB  = DIM*jnrB;
1051             j_coord_offsetC  = DIM*jnrC;
1052             j_coord_offsetD  = DIM*jnrD;
1053
1054             /* load j atom coordinates */
1055             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1056                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1057                                                  &jx0,&jy0,&jz0);
1058
1059             /* Calculate displacement vector */
1060             dx00             = _mm256_sub_pd(ix0,jx0);
1061             dy00             = _mm256_sub_pd(iy0,jy0);
1062             dz00             = _mm256_sub_pd(iz0,jz0);
1063             dx10             = _mm256_sub_pd(ix1,jx0);
1064             dy10             = _mm256_sub_pd(iy1,jy0);
1065             dz10             = _mm256_sub_pd(iz1,jz0);
1066             dx20             = _mm256_sub_pd(ix2,jx0);
1067             dy20             = _mm256_sub_pd(iy2,jy0);
1068             dz20             = _mm256_sub_pd(iz2,jz0);
1069             dx30             = _mm256_sub_pd(ix3,jx0);
1070             dy30             = _mm256_sub_pd(iy3,jy0);
1071             dz30             = _mm256_sub_pd(iz3,jz0);
1072
1073             /* Calculate squared distance and things based on it */
1074             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1075             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1076             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1077             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1078
1079             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1080             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1081             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1082             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1083
1084             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1085             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1086             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1087             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1088
1089             /* Load parameters for j particles */
1090             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1091                                                                  charge+jnrC+0,charge+jnrD+0);
1092             vdwjidx0A        = 2*vdwtype[jnrA+0];
1093             vdwjidx0B        = 2*vdwtype[jnrB+0];
1094             vdwjidx0C        = 2*vdwtype[jnrC+0];
1095             vdwjidx0D        = 2*vdwtype[jnrD+0];
1096
1097             fjx0             = _mm256_setzero_pd();
1098             fjy0             = _mm256_setzero_pd();
1099             fjz0             = _mm256_setzero_pd();
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1106             {
1107
1108             r00              = _mm256_mul_pd(rsq00,rinv00);
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1112                                             vdwioffsetptr0+vdwjidx0B,
1113                                             vdwioffsetptr0+vdwjidx0C,
1114                                             vdwioffsetptr0+vdwjidx0D,
1115                                             &c6_00,&c12_00);
1116
1117             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1118                                                                   vdwgridioffsetptr0+vdwjidx0B,
1119                                                                   vdwgridioffsetptr0+vdwjidx0C,
1120                                                                   vdwgridioffsetptr0+vdwjidx0D);
1121
1122             /* Analytical LJ-PME */
1123             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1124             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1125             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1126             exponent         = gmx_simd_exp_d(ewcljrsq);
1127             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1128             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1129             /* f6A = 6 * C6grid * (1 - poly) */
1130             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1131             /* f6B = C6grid * exponent * beta^6 */
1132             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1133             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1134             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);
1135
1136             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1137
1138             fscal            = fvdw;
1139
1140             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1141
1142             /* Calculate temporary vectorial force */
1143             tx               = _mm256_mul_pd(fscal,dx00);
1144             ty               = _mm256_mul_pd(fscal,dy00);
1145             tz               = _mm256_mul_pd(fscal,dz00);
1146
1147             /* Update vectorial force */
1148             fix0             = _mm256_add_pd(fix0,tx);
1149             fiy0             = _mm256_add_pd(fiy0,ty);
1150             fiz0             = _mm256_add_pd(fiz0,tz);
1151
1152             fjx0             = _mm256_add_pd(fjx0,tx);
1153             fjy0             = _mm256_add_pd(fjy0,ty);
1154             fjz0             = _mm256_add_pd(fjz0,tz);
1155
1156             }
1157
1158             /**************************
1159              * CALCULATE INTERACTIONS *
1160              **************************/
1161
1162             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1163             {
1164
1165             r10              = _mm256_mul_pd(rsq10,rinv10);
1166
1167             /* Compute parameters for interactions between i and j atoms */
1168             qq10             = _mm256_mul_pd(iq1,jq0);
1169
1170             /* EWALD ELECTROSTATICS */
1171
1172             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1173             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1174             ewitab           = _mm256_cvttpd_epi32(ewrt);
1175             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1176             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1177                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1178                                             &ewtabF,&ewtabFn);
1179             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1180             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1181
1182             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1183
1184             fscal            = felec;
1185
1186             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1187
1188             /* Calculate temporary vectorial force */
1189             tx               = _mm256_mul_pd(fscal,dx10);
1190             ty               = _mm256_mul_pd(fscal,dy10);
1191             tz               = _mm256_mul_pd(fscal,dz10);
1192
1193             /* Update vectorial force */
1194             fix1             = _mm256_add_pd(fix1,tx);
1195             fiy1             = _mm256_add_pd(fiy1,ty);
1196             fiz1             = _mm256_add_pd(fiz1,tz);
1197
1198             fjx0             = _mm256_add_pd(fjx0,tx);
1199             fjy0             = _mm256_add_pd(fjy0,ty);
1200             fjz0             = _mm256_add_pd(fjz0,tz);
1201
1202             }
1203
1204             /**************************
1205              * CALCULATE INTERACTIONS *
1206              **************************/
1207
1208             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1209             {
1210
1211             r20              = _mm256_mul_pd(rsq20,rinv20);
1212
1213             /* Compute parameters for interactions between i and j atoms */
1214             qq20             = _mm256_mul_pd(iq2,jq0);
1215
1216             /* EWALD ELECTROSTATICS */
1217
1218             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1219             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1220             ewitab           = _mm256_cvttpd_epi32(ewrt);
1221             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1222             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1223                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1224                                             &ewtabF,&ewtabFn);
1225             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1226             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1227
1228             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1229
1230             fscal            = felec;
1231
1232             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1233
1234             /* Calculate temporary vectorial force */
1235             tx               = _mm256_mul_pd(fscal,dx20);
1236             ty               = _mm256_mul_pd(fscal,dy20);
1237             tz               = _mm256_mul_pd(fscal,dz20);
1238
1239             /* Update vectorial force */
1240             fix2             = _mm256_add_pd(fix2,tx);
1241             fiy2             = _mm256_add_pd(fiy2,ty);
1242             fiz2             = _mm256_add_pd(fiz2,tz);
1243
1244             fjx0             = _mm256_add_pd(fjx0,tx);
1245             fjy0             = _mm256_add_pd(fjy0,ty);
1246             fjz0             = _mm256_add_pd(fjz0,tz);
1247
1248             }
1249
1250             /**************************
1251              * CALCULATE INTERACTIONS *
1252              **************************/
1253
1254             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1255             {
1256
1257             r30              = _mm256_mul_pd(rsq30,rinv30);
1258
1259             /* Compute parameters for interactions between i and j atoms */
1260             qq30             = _mm256_mul_pd(iq3,jq0);
1261
1262             /* EWALD ELECTROSTATICS */
1263
1264             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1265             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1266             ewitab           = _mm256_cvttpd_epi32(ewrt);
1267             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1268             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1269                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1270                                             &ewtabF,&ewtabFn);
1271             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1272             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1273
1274             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1275
1276             fscal            = felec;
1277
1278             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1279
1280             /* Calculate temporary vectorial force */
1281             tx               = _mm256_mul_pd(fscal,dx30);
1282             ty               = _mm256_mul_pd(fscal,dy30);
1283             tz               = _mm256_mul_pd(fscal,dz30);
1284
1285             /* Update vectorial force */
1286             fix3             = _mm256_add_pd(fix3,tx);
1287             fiy3             = _mm256_add_pd(fiy3,ty);
1288             fiz3             = _mm256_add_pd(fiz3,tz);
1289
1290             fjx0             = _mm256_add_pd(fjx0,tx);
1291             fjy0             = _mm256_add_pd(fjy0,ty);
1292             fjz0             = _mm256_add_pd(fjz0,tz);
1293
1294             }
1295
1296             fjptrA             = f+j_coord_offsetA;
1297             fjptrB             = f+j_coord_offsetB;
1298             fjptrC             = f+j_coord_offsetC;
1299             fjptrD             = f+j_coord_offsetD;
1300
1301             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1302
1303             /* Inner loop uses 169 flops */
1304         }
1305
1306         if(jidx<j_index_end)
1307         {
1308
1309             /* Get j neighbor index, and coordinate index */
1310             jnrlistA         = jjnr[jidx];
1311             jnrlistB         = jjnr[jidx+1];
1312             jnrlistC         = jjnr[jidx+2];
1313             jnrlistD         = jjnr[jidx+3];
1314             /* Sign of each element will be negative for non-real atoms.
1315              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1316              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1317              */
1318             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1319
1320             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1321             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1322             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1323
1324             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1325             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1326             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1327             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1328             j_coord_offsetA  = DIM*jnrA;
1329             j_coord_offsetB  = DIM*jnrB;
1330             j_coord_offsetC  = DIM*jnrC;
1331             j_coord_offsetD  = DIM*jnrD;
1332
1333             /* load j atom coordinates */
1334             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1335                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1336                                                  &jx0,&jy0,&jz0);
1337
1338             /* Calculate displacement vector */
1339             dx00             = _mm256_sub_pd(ix0,jx0);
1340             dy00             = _mm256_sub_pd(iy0,jy0);
1341             dz00             = _mm256_sub_pd(iz0,jz0);
1342             dx10             = _mm256_sub_pd(ix1,jx0);
1343             dy10             = _mm256_sub_pd(iy1,jy0);
1344             dz10             = _mm256_sub_pd(iz1,jz0);
1345             dx20             = _mm256_sub_pd(ix2,jx0);
1346             dy20             = _mm256_sub_pd(iy2,jy0);
1347             dz20             = _mm256_sub_pd(iz2,jz0);
1348             dx30             = _mm256_sub_pd(ix3,jx0);
1349             dy30             = _mm256_sub_pd(iy3,jy0);
1350             dz30             = _mm256_sub_pd(iz3,jz0);
1351
1352             /* Calculate squared distance and things based on it */
1353             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1354             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1355             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1356             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1357
1358             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1359             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1360             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1361             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1362
1363             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1364             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1365             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1366             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1367
1368             /* Load parameters for j particles */
1369             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1370                                                                  charge+jnrC+0,charge+jnrD+0);
1371             vdwjidx0A        = 2*vdwtype[jnrA+0];
1372             vdwjidx0B        = 2*vdwtype[jnrB+0];
1373             vdwjidx0C        = 2*vdwtype[jnrC+0];
1374             vdwjidx0D        = 2*vdwtype[jnrD+0];
1375
1376             fjx0             = _mm256_setzero_pd();
1377             fjy0             = _mm256_setzero_pd();
1378             fjz0             = _mm256_setzero_pd();
1379
1380             /**************************
1381              * CALCULATE INTERACTIONS *
1382              **************************/
1383
1384             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1385             {
1386
1387             r00              = _mm256_mul_pd(rsq00,rinv00);
1388             r00              = _mm256_andnot_pd(dummy_mask,r00);
1389
1390             /* Compute parameters for interactions between i and j atoms */
1391             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1392                                             vdwioffsetptr0+vdwjidx0B,
1393                                             vdwioffsetptr0+vdwjidx0C,
1394                                             vdwioffsetptr0+vdwjidx0D,
1395                                             &c6_00,&c12_00);
1396
1397             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1398                                                                   vdwgridioffsetptr0+vdwjidx0B,
1399                                                                   vdwgridioffsetptr0+vdwjidx0C,
1400                                                                   vdwgridioffsetptr0+vdwjidx0D);
1401
1402             /* Analytical LJ-PME */
1403             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1404             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1405             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1406             exponent         = gmx_simd_exp_d(ewcljrsq);
1407             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1408             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1409             /* f6A = 6 * C6grid * (1 - poly) */
1410             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1411             /* f6B = C6grid * exponent * beta^6 */
1412             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1413             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1414             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);
1415
1416             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1417
1418             fscal            = fvdw;
1419
1420             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1421
1422             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1423
1424             /* Calculate temporary vectorial force */
1425             tx               = _mm256_mul_pd(fscal,dx00);
1426             ty               = _mm256_mul_pd(fscal,dy00);
1427             tz               = _mm256_mul_pd(fscal,dz00);
1428
1429             /* Update vectorial force */
1430             fix0             = _mm256_add_pd(fix0,tx);
1431             fiy0             = _mm256_add_pd(fiy0,ty);
1432             fiz0             = _mm256_add_pd(fiz0,tz);
1433
1434             fjx0             = _mm256_add_pd(fjx0,tx);
1435             fjy0             = _mm256_add_pd(fjy0,ty);
1436             fjz0             = _mm256_add_pd(fjz0,tz);
1437
1438             }
1439
1440             /**************************
1441              * CALCULATE INTERACTIONS *
1442              **************************/
1443
1444             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1445             {
1446
1447             r10              = _mm256_mul_pd(rsq10,rinv10);
1448             r10              = _mm256_andnot_pd(dummy_mask,r10);
1449
1450             /* Compute parameters for interactions between i and j atoms */
1451             qq10             = _mm256_mul_pd(iq1,jq0);
1452
1453             /* EWALD ELECTROSTATICS */
1454
1455             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1456             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1457             ewitab           = _mm256_cvttpd_epi32(ewrt);
1458             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1459             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1460                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1461                                             &ewtabF,&ewtabFn);
1462             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1463             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1464
1465             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1466
1467             fscal            = felec;
1468
1469             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1470
1471             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1472
1473             /* Calculate temporary vectorial force */
1474             tx               = _mm256_mul_pd(fscal,dx10);
1475             ty               = _mm256_mul_pd(fscal,dy10);
1476             tz               = _mm256_mul_pd(fscal,dz10);
1477
1478             /* Update vectorial force */
1479             fix1             = _mm256_add_pd(fix1,tx);
1480             fiy1             = _mm256_add_pd(fiy1,ty);
1481             fiz1             = _mm256_add_pd(fiz1,tz);
1482
1483             fjx0             = _mm256_add_pd(fjx0,tx);
1484             fjy0             = _mm256_add_pd(fjy0,ty);
1485             fjz0             = _mm256_add_pd(fjz0,tz);
1486
1487             }
1488
1489             /**************************
1490              * CALCULATE INTERACTIONS *
1491              **************************/
1492
1493             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1494             {
1495
1496             r20              = _mm256_mul_pd(rsq20,rinv20);
1497             r20              = _mm256_andnot_pd(dummy_mask,r20);
1498
1499             /* Compute parameters for interactions between i and j atoms */
1500             qq20             = _mm256_mul_pd(iq2,jq0);
1501
1502             /* EWALD ELECTROSTATICS */
1503
1504             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1505             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1506             ewitab           = _mm256_cvttpd_epi32(ewrt);
1507             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1508             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1509                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1510                                             &ewtabF,&ewtabFn);
1511             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1512             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1513
1514             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1515
1516             fscal            = felec;
1517
1518             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1519
1520             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1521
1522             /* Calculate temporary vectorial force */
1523             tx               = _mm256_mul_pd(fscal,dx20);
1524             ty               = _mm256_mul_pd(fscal,dy20);
1525             tz               = _mm256_mul_pd(fscal,dz20);
1526
1527             /* Update vectorial force */
1528             fix2             = _mm256_add_pd(fix2,tx);
1529             fiy2             = _mm256_add_pd(fiy2,ty);
1530             fiz2             = _mm256_add_pd(fiz2,tz);
1531
1532             fjx0             = _mm256_add_pd(fjx0,tx);
1533             fjy0             = _mm256_add_pd(fjy0,ty);
1534             fjz0             = _mm256_add_pd(fjz0,tz);
1535
1536             }
1537
1538             /**************************
1539              * CALCULATE INTERACTIONS *
1540              **************************/
1541
1542             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1543             {
1544
1545             r30              = _mm256_mul_pd(rsq30,rinv30);
1546             r30              = _mm256_andnot_pd(dummy_mask,r30);
1547
1548             /* Compute parameters for interactions between i and j atoms */
1549             qq30             = _mm256_mul_pd(iq3,jq0);
1550
1551             /* EWALD ELECTROSTATICS */
1552
1553             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1554             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1555             ewitab           = _mm256_cvttpd_epi32(ewrt);
1556             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1557             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1558                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1559                                             &ewtabF,&ewtabFn);
1560             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1561             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1562
1563             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1564
1565             fscal            = felec;
1566
1567             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1568
1569             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1570
1571             /* Calculate temporary vectorial force */
1572             tx               = _mm256_mul_pd(fscal,dx30);
1573             ty               = _mm256_mul_pd(fscal,dy30);
1574             tz               = _mm256_mul_pd(fscal,dz30);
1575
1576             /* Update vectorial force */
1577             fix3             = _mm256_add_pd(fix3,tx);
1578             fiy3             = _mm256_add_pd(fiy3,ty);
1579             fiz3             = _mm256_add_pd(fiz3,tz);
1580
1581             fjx0             = _mm256_add_pd(fjx0,tx);
1582             fjy0             = _mm256_add_pd(fjy0,ty);
1583             fjz0             = _mm256_add_pd(fjz0,tz);
1584
1585             }
1586
1587             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1588             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1589             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1590             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1591
1592             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1593
1594             /* Inner loop uses 173 flops */
1595         }
1596
1597         /* End of innermost loop */
1598
1599         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1600                                                  f+i_coord_offset,fshift+i_shift_offset);
1601
1602         /* Increment number of inner iterations */
1603         inneriter                  += j_index_end - j_index_start;
1604
1605         /* Outer loop uses 24 flops */
1606     }
1607
1608     /* Increment number of outer iterations */
1609     outeriter        += nri;
1610
1611     /* Update outer/inner flops */
1612
1613     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*173);
1614 }