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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_double.h"
  50 #include "kernelutil_x86_avx_256_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_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_ElecEwSh_VdwLJEwSh_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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 168     rcutoff_scalar   = fr->rcoulomb;
 169     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 170     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 171
 172     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 173     rvdw             = _mm256_set1_pd(fr->rvdw);
 174
 175     /* Avoid stupid compiler warnings */
 176     jnrA = jnrB = jnrC = jnrD = 0;
 177     j_coord_offsetA = 0;
 178     j_coord_offsetB = 0;
 179     j_coord_offsetC = 0;
 180     j_coord_offsetD = 0;
 181
 182     outeriter        = 0;
 183     inneriter        = 0;
 184
 185     for(iidx=0;iidx<4*DIM;iidx++)
 186     {
 187         scratch[iidx] = 0.0;
 188     }
 189
 190     /* Start outer loop over neighborlists */
 191     for(iidx=0; iidx<nri; iidx++)
 192     {
 193         /* Load shift vector for this list */
 194         i_shift_offset   = DIM*shiftidx[iidx];
 195
 196         /* Load limits for loop over neighbors */
 197         j_index_start    = jindex[iidx];
 198         j_index_end      = jindex[iidx+1];
 199
 200         /* Get outer coordinate index */
 201         inr              = iinr[iidx];
 202         i_coord_offset   = DIM*inr;
 203
 204         /* Load i particle coords and add shift vector */
 205         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 206                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 207
 208         fix0             = _mm256_setzero_pd();
 209         fiy0             = _mm256_setzero_pd();
 210         fiz0             = _mm256_setzero_pd();
 211         fix1             = _mm256_setzero_pd();
 212         fiy1             = _mm256_setzero_pd();
 213         fiz1             = _mm256_setzero_pd();
 214         fix2             = _mm256_setzero_pd();
 215         fiy2             = _mm256_setzero_pd();
 216         fiz2             = _mm256_setzero_pd();
 217         fix3             = _mm256_setzero_pd();
 218         fiy3             = _mm256_setzero_pd();
 219         fiz3             = _mm256_setzero_pd();
 220
 221         /* Reset potential sums */
 222         velecsum         = _mm256_setzero_pd();
 223         vvdwsum          = _mm256_setzero_pd();
 224
 225         /* Start inner kernel loop */
 226         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 227         {
 228
 229             /* Get j neighbor index, and coordinate index */
 230             jnrA             = jjnr[jidx];
 231             jnrB             = jjnr[jidx+1];
 232             jnrC             = jjnr[jidx+2];
 233             jnrD             = jjnr[jidx+3];
 234             j_coord_offsetA  = DIM*jnrA;
 235             j_coord_offsetB  = DIM*jnrB;
 236             j_coord_offsetC  = DIM*jnrC;
 237             j_coord_offsetD  = DIM*jnrD;
 238
 239             /* load j atom coordinates */
 240             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 241                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 242                                                  &jx0,&jy0,&jz0);
 243
 244             /* Calculate displacement vector */
 245             dx00             = _mm256_sub_pd(ix0,jx0);
 246             dy00             = _mm256_sub_pd(iy0,jy0);
 247             dz00             = _mm256_sub_pd(iz0,jz0);
 248             dx10             = _mm256_sub_pd(ix1,jx0);
 249             dy10             = _mm256_sub_pd(iy1,jy0);
 250             dz10             = _mm256_sub_pd(iz1,jz0);
 251             dx20             = _mm256_sub_pd(ix2,jx0);
 252             dy20             = _mm256_sub_pd(iy2,jy0);
 253             dz20             = _mm256_sub_pd(iz2,jz0);
 254             dx30             = _mm256_sub_pd(ix3,jx0);
 255             dy30             = _mm256_sub_pd(iy3,jy0);
 256             dz30             = _mm256_sub_pd(iz3,jz0);
 257
 258             /* Calculate squared distance and things based on it */
 259             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 260             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 261             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 262             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 263
 264             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 265             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 266             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 267             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 268
 269             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 270             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 271             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 272             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 273
 274             /* Load parameters for j particles */
 275             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 276                                                                  charge+jnrC+0,charge+jnrD+0);
 277             vdwjidx0A        = 2*vdwtype[jnrA+0];
 278             vdwjidx0B        = 2*vdwtype[jnrB+0];
 279             vdwjidx0C        = 2*vdwtype[jnrC+0];
 280             vdwjidx0D        = 2*vdwtype[jnrD+0];
 281
 282             fjx0             = _mm256_setzero_pd();
 283             fjy0             = _mm256_setzero_pd();
 284             fjz0             = _mm256_setzero_pd();
 285
 286             /**************************
 287              * CALCULATE INTERACTIONS *
 288              **************************/
 289
 290             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 291             {
 292
 293             r00              = _mm256_mul_pd(rsq00,rinv00);
 294
 295             /* Compute parameters for interactions between i and j atoms */
 296             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 297                                             vdwioffsetptr0+vdwjidx0B,
 298                                             vdwioffsetptr0+vdwjidx0C,
 299                                             vdwioffsetptr0+vdwjidx0D,
 300                                             &c6_00,&c12_00);
 301
 302             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 303                                                                   vdwgridioffsetptr0+vdwjidx0B,
 304                                                                   vdwgridioffsetptr0+vdwjidx0C,
 305                                                                   vdwgridioffsetptr0+vdwjidx0D);
 306
 307             /* Analytical LJ-PME */
 308             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 309             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 310             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 311             exponent         = gmx_simd_exp_d(ewcljrsq);
 312             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 313             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 314             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 315             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 316             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 317             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) ,
 318                                           _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));
 319             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 320             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);
 321
 322             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 326             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 327
 328             fscal            = fvdw;
 329
 330             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 331
 332             /* Calculate temporary vectorial force */
 333             tx               = _mm256_mul_pd(fscal,dx00);
 334             ty               = _mm256_mul_pd(fscal,dy00);
 335             tz               = _mm256_mul_pd(fscal,dz00);
 336
 337             /* Update vectorial force */
 338             fix0             = _mm256_add_pd(fix0,tx);
 339             fiy0             = _mm256_add_pd(fiy0,ty);
 340             fiz0             = _mm256_add_pd(fiz0,tz);
 341
 342             fjx0             = _mm256_add_pd(fjx0,tx);
 343             fjy0             = _mm256_add_pd(fjy0,ty);
 344             fjz0             = _mm256_add_pd(fjz0,tz);
 345
 346             }
 347
 348             /**************************
 349              * CALCULATE INTERACTIONS *
 350              **************************/
 351
 352             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 353             {
 354
 355             r10              = _mm256_mul_pd(rsq10,rinv10);
 356
 357             /* Compute parameters for interactions between i and j atoms */
 358             qq10             = _mm256_mul_pd(iq1,jq0);
 359
 360             /* EWALD ELECTROSTATICS */
 361
 362             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 363             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 364             ewitab           = _mm256_cvttpd_epi32(ewrt);
 365             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 366             ewitab           = _mm_slli_epi32(ewitab,2);
 367             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 368             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 369             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 370             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 371             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 372             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 373             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 374             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 375             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 376
 377             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velec            = _mm256_and_pd(velec,cutoff_mask);
 381             velecsum         = _mm256_add_pd(velecsum,velec);
 382
 383             fscal            = felec;
 384
 385             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 386
 387             /* Calculate temporary vectorial force */
 388             tx               = _mm256_mul_pd(fscal,dx10);
 389             ty               = _mm256_mul_pd(fscal,dy10);
 390             tz               = _mm256_mul_pd(fscal,dz10);
 391
 392             /* Update vectorial force */
 393             fix1             = _mm256_add_pd(fix1,tx);
 394             fiy1             = _mm256_add_pd(fiy1,ty);
 395             fiz1             = _mm256_add_pd(fiz1,tz);
 396
 397             fjx0             = _mm256_add_pd(fjx0,tx);
 398             fjy0             = _mm256_add_pd(fjy0,ty);
 399             fjz0             = _mm256_add_pd(fjz0,tz);
 400
 401             }
 402
 403             /**************************
 404              * CALCULATE INTERACTIONS *
 405              **************************/
 406
 407             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 408             {
 409
 410             r20              = _mm256_mul_pd(rsq20,rinv20);
 411
 412             /* Compute parameters for interactions between i and j atoms */
 413             qq20             = _mm256_mul_pd(iq2,jq0);
 414
 415             /* EWALD ELECTROSTATICS */
 416
 417             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 418             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 419             ewitab           = _mm256_cvttpd_epi32(ewrt);
 420             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 421             ewitab           = _mm_slli_epi32(ewitab,2);
 422             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 423             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 424             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 425             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 426             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 427             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 428             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 429             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 430             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 431
 432             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 433
 434             /* Update potential sum for this i atom from the interaction with this j atom. */
 435             velec            = _mm256_and_pd(velec,cutoff_mask);
 436             velecsum         = _mm256_add_pd(velecsum,velec);
 437
 438             fscal            = felec;
 439
 440             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 441
 442             /* Calculate temporary vectorial force */
 443             tx               = _mm256_mul_pd(fscal,dx20);
 444             ty               = _mm256_mul_pd(fscal,dy20);
 445             tz               = _mm256_mul_pd(fscal,dz20);
 446
 447             /* Update vectorial force */
 448             fix2             = _mm256_add_pd(fix2,tx);
 449             fiy2             = _mm256_add_pd(fiy2,ty);
 450             fiz2             = _mm256_add_pd(fiz2,tz);
 451
 452             fjx0             = _mm256_add_pd(fjx0,tx);
 453             fjy0             = _mm256_add_pd(fjy0,ty);
 454             fjz0             = _mm256_add_pd(fjz0,tz);
 455
 456             }
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 463             {
 464
 465             r30              = _mm256_mul_pd(rsq30,rinv30);
 466
 467             /* Compute parameters for interactions between i and j atoms */
 468             qq30             = _mm256_mul_pd(iq3,jq0);
 469
 470             /* EWALD ELECTROSTATICS */
 471
 472             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 473             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 474             ewitab           = _mm256_cvttpd_epi32(ewrt);
 475             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 476             ewitab           = _mm_slli_epi32(ewitab,2);
 477             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 478             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 479             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 480             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 481             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 482             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 483             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 484             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 485             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 486
 487             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velec            = _mm256_and_pd(velec,cutoff_mask);
 491             velecsum         = _mm256_add_pd(velecsum,velec);
 492
 493             fscal            = felec;
 494
 495             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 496
 497             /* Calculate temporary vectorial force */
 498             tx               = _mm256_mul_pd(fscal,dx30);
 499             ty               = _mm256_mul_pd(fscal,dy30);
 500             tz               = _mm256_mul_pd(fscal,dz30);
 501
 502             /* Update vectorial force */
 503             fix3             = _mm256_add_pd(fix3,tx);
 504             fiy3             = _mm256_add_pd(fiy3,ty);
 505             fiz3             = _mm256_add_pd(fiz3,tz);
 506
 507             fjx0             = _mm256_add_pd(fjx0,tx);
 508             fjy0             = _mm256_add_pd(fjy0,ty);
 509             fjz0             = _mm256_add_pd(fjz0,tz);
 510
 511             }
 512
 513             fjptrA             = f+j_coord_offsetA;
 514             fjptrB             = f+j_coord_offsetB;
 515             fjptrC             = f+j_coord_offsetC;
 516             fjptrD             = f+j_coord_offsetD;
 517
 518             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 519
 520             /* Inner loop uses 203 flops */
 521         }
 522
 523         if(jidx<j_index_end)
 524         {
 525
 526             /* Get j neighbor index, and coordinate index */
 527             jnrlistA         = jjnr[jidx];
 528             jnrlistB         = jjnr[jidx+1];
 529             jnrlistC         = jjnr[jidx+2];
 530             jnrlistD         = jjnr[jidx+3];
 531             /* Sign of each element will be negative for non-real atoms.
 532              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 533              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 534              */
 535             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 536
 537             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 538             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 539             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 540
 541             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 542             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 543             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 544             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 545             j_coord_offsetA  = DIM*jnrA;
 546             j_coord_offsetB  = DIM*jnrB;
 547             j_coord_offsetC  = DIM*jnrC;
 548             j_coord_offsetD  = DIM*jnrD;
 549
 550             /* load j atom coordinates */
 551             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 552                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 553                                                  &jx0,&jy0,&jz0);
 554
 555             /* Calculate displacement vector */
 556             dx00             = _mm256_sub_pd(ix0,jx0);
 557             dy00             = _mm256_sub_pd(iy0,jy0);
 558             dz00             = _mm256_sub_pd(iz0,jz0);
 559             dx10             = _mm256_sub_pd(ix1,jx0);
 560             dy10             = _mm256_sub_pd(iy1,jy0);
 561             dz10             = _mm256_sub_pd(iz1,jz0);
 562             dx20             = _mm256_sub_pd(ix2,jx0);
 563             dy20             = _mm256_sub_pd(iy2,jy0);
 564             dz20             = _mm256_sub_pd(iz2,jz0);
 565             dx30             = _mm256_sub_pd(ix3,jx0);
 566             dy30             = _mm256_sub_pd(iy3,jy0);
 567             dz30             = _mm256_sub_pd(iz3,jz0);
 568
 569             /* Calculate squared distance and things based on it */
 570             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 571             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 572             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 573             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 574
 575             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
 576             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 577             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 578             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 579
 580             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
 581             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 582             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 583             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 584
 585             /* Load parameters for j particles */
 586             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 587                                                                  charge+jnrC+0,charge+jnrD+0);
 588             vdwjidx0A        = 2*vdwtype[jnrA+0];
 589             vdwjidx0B        = 2*vdwtype[jnrB+0];
 590             vdwjidx0C        = 2*vdwtype[jnrC+0];
 591             vdwjidx0D        = 2*vdwtype[jnrD+0];
 592
 593             fjx0             = _mm256_setzero_pd();
 594             fjy0             = _mm256_setzero_pd();
 595             fjz0             = _mm256_setzero_pd();
 596
 597             /**************************
 598              * CALCULATE INTERACTIONS *
 599              **************************/
 600
 601             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 602             {
 603
 604             r00              = _mm256_mul_pd(rsq00,rinv00);
 605             r00              = _mm256_andnot_pd(dummy_mask,r00);
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 609                                             vdwioffsetptr0+vdwjidx0B,
 610                                             vdwioffsetptr0+vdwjidx0C,
 611                                             vdwioffsetptr0+vdwjidx0D,
 612                                             &c6_00,&c12_00);
 613
 614             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
 615                                                                   vdwgridioffsetptr0+vdwjidx0B,
 616                                                                   vdwgridioffsetptr0+vdwjidx0C,
 617                                                                   vdwgridioffsetptr0+vdwjidx0D);
 618
 619             /* Analytical LJ-PME */
 620             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 621             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
 622             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
 623             exponent         = gmx_simd_exp_d(ewcljrsq);
 624             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 625             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
 626             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 627             vvdw6            = _mm256_mul_pd(_mm256_sub_pd(c6_00,_mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly))),rinvsix);
 628             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 629             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) ,
 630                                           _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));
 631             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 632             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);
 633
 634             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 635
 636             /* Update potential sum for this i atom from the interaction with this j atom. */
 637             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 638             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 639             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 640
 641             fscal            = fvdw;
 642
 643             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 644
 645             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm256_mul_pd(fscal,dx00);
 649             ty               = _mm256_mul_pd(fscal,dy00);
 650             tz               = _mm256_mul_pd(fscal,dz00);
 651
 652             /* Update vectorial force */
 653             fix0             = _mm256_add_pd(fix0,tx);
 654             fiy0             = _mm256_add_pd(fiy0,ty);
 655             fiz0             = _mm256_add_pd(fiz0,tz);
 656
 657             fjx0             = _mm256_add_pd(fjx0,tx);
 658             fjy0             = _mm256_add_pd(fjy0,ty);
 659             fjz0             = _mm256_add_pd(fjz0,tz);
 660
 661             }
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 668             {
 669
 670             r10              = _mm256_mul_pd(rsq10,rinv10);
 671             r10              = _mm256_andnot_pd(dummy_mask,r10);
 672
 673             /* Compute parameters for interactions between i and j atoms */
 674             qq10             = _mm256_mul_pd(iq1,jq0);
 675
 676             /* EWALD ELECTROSTATICS */
 677
 678             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 679             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 680             ewitab           = _mm256_cvttpd_epi32(ewrt);
 681             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 682             ewitab           = _mm_slli_epi32(ewitab,2);
 683             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 684             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 685             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 686             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 687             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 688             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 689             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 690             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 691             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 692
 693             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 694
 695             /* Update potential sum for this i atom from the interaction with this j atom. */
 696             velec            = _mm256_and_pd(velec,cutoff_mask);
 697             velec            = _mm256_andnot_pd(dummy_mask,velec);
 698             velecsum         = _mm256_add_pd(velecsum,velec);
 699
 700             fscal            = felec;
 701
 702             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 703
 704             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 705
 706             /* Calculate temporary vectorial force */
 707             tx               = _mm256_mul_pd(fscal,dx10);
 708             ty               = _mm256_mul_pd(fscal,dy10);
 709             tz               = _mm256_mul_pd(fscal,dz10);
 710
 711             /* Update vectorial force */
 712             fix1             = _mm256_add_pd(fix1,tx);
 713             fiy1             = _mm256_add_pd(fiy1,ty);
 714             fiz1             = _mm256_add_pd(fiz1,tz);
 715
 716             fjx0             = _mm256_add_pd(fjx0,tx);
 717             fjy0             = _mm256_add_pd(fjy0,ty);
 718             fjz0             = _mm256_add_pd(fjz0,tz);
 719
 720             }
 721
 722             /**************************
 723              * CALCULATE INTERACTIONS *
 724              **************************/
 725
 726             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 727             {
 728
 729             r20              = _mm256_mul_pd(rsq20,rinv20);
 730             r20              = _mm256_andnot_pd(dummy_mask,r20);
 731
 732             /* Compute parameters for interactions between i and j atoms */
 733             qq20             = _mm256_mul_pd(iq2,jq0);
 734
 735             /* EWALD ELECTROSTATICS */
 736
 737             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 738             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 739             ewitab           = _mm256_cvttpd_epi32(ewrt);
 740             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 741             ewitab           = _mm_slli_epi32(ewitab,2);
 742             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 743             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 744             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 745             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 746             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 747             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 748             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 749             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 750             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 751
 752             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 753
 754             /* Update potential sum for this i atom from the interaction with this j atom. */
 755             velec            = _mm256_and_pd(velec,cutoff_mask);
 756             velec            = _mm256_andnot_pd(dummy_mask,velec);
 757             velecsum         = _mm256_add_pd(velecsum,velec);
 758
 759             fscal            = felec;
 760
 761             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 762
 763             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 764
 765             /* Calculate temporary vectorial force */
 766             tx               = _mm256_mul_pd(fscal,dx20);
 767             ty               = _mm256_mul_pd(fscal,dy20);
 768             tz               = _mm256_mul_pd(fscal,dz20);
 769
 770             /* Update vectorial force */
 771             fix2             = _mm256_add_pd(fix2,tx);
 772             fiy2             = _mm256_add_pd(fiy2,ty);
 773             fiz2             = _mm256_add_pd(fiz2,tz);
 774
 775             fjx0             = _mm256_add_pd(fjx0,tx);
 776             fjy0             = _mm256_add_pd(fjy0,ty);
 777             fjz0             = _mm256_add_pd(fjz0,tz);
 778
 779             }
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 786             {
 787
 788             r30              = _mm256_mul_pd(rsq30,rinv30);
 789             r30              = _mm256_andnot_pd(dummy_mask,r30);
 790
 791             /* Compute parameters for interactions between i and j atoms */
 792             qq30             = _mm256_mul_pd(iq3,jq0);
 793
 794             /* EWALD ELECTROSTATICS */
 795
 796             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 797             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 798             ewitab           = _mm256_cvttpd_epi32(ewrt);
 799             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 800             ewitab           = _mm_slli_epi32(ewitab,2);
 801             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 802             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 803             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 804             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 805             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 806             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 807             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 808             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 809             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 810
 811             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 812
 813             /* Update potential sum for this i atom from the interaction with this j atom. */
 814             velec            = _mm256_and_pd(velec,cutoff_mask);
 815             velec            = _mm256_andnot_pd(dummy_mask,velec);
 816             velecsum         = _mm256_add_pd(velecsum,velec);
 817
 818             fscal            = felec;
 819
 820             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 821
 822             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 823
 824             /* Calculate temporary vectorial force */
 825             tx               = _mm256_mul_pd(fscal,dx30);
 826             ty               = _mm256_mul_pd(fscal,dy30);
 827             tz               = _mm256_mul_pd(fscal,dz30);
 828
 829             /* Update vectorial force */
 830             fix3             = _mm256_add_pd(fix3,tx);
 831             fiy3             = _mm256_add_pd(fiy3,ty);
 832             fiz3             = _mm256_add_pd(fiz3,tz);
 833
 834             fjx0             = _mm256_add_pd(fjx0,tx);
 835             fjy0             = _mm256_add_pd(fjy0,ty);
 836             fjz0             = _mm256_add_pd(fjz0,tz);
 837
 838             }
 839
 840             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 841             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 842             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 843             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 844
 845             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 846
 847             /* Inner loop uses 207 flops */
 848         }
 849
 850         /* End of innermost loop */
 851
 852         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 853                                                  f+i_coord_offset,fshift+i_shift_offset);
 854
 855         ggid                        = gid[iidx];
 856         /* Update potential energies */
 857         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 858         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 859
 860         /* Increment number of inner iterations */
 861         inneriter                  += j_index_end - j_index_start;
 862
 863         /* Outer loop uses 26 flops */
 864     }
 865
 866     /* Increment number of outer iterations */
 867     outeriter        += nri;
 868
 869     /* Update outer/inner flops */
 870
 871     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*207);
 872 }
 873 /*
 874  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_256_double
 875  * Electrostatics interaction: Ewald
 876  * VdW interaction:            LJEwald
 877  * Geometry:                   Water4-Particle
 878  * Calculate force/pot:        Force
 879  */
 880 void
 881 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4P1_F_avx_256_double
 882                     (t_nblist                    * gmx_restrict       nlist,
 883                      rvec                        * gmx_restrict          xx,
 884                      rvec                        * gmx_restrict          ff,
 885                      t_forcerec                  * gmx_restrict          fr,
 886                      t_mdatoms                   * gmx_restrict     mdatoms,
 887                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 888                      t_nrnb                      * gmx_restrict        nrnb)
 889 {
 890     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 891      * just 0 for non-waters.
 892      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 893      * jnr indices corresponding to data put in the four positions in the SIMD register.
 894      */
 895     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 896     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 897     int              jnrA,jnrB,jnrC,jnrD;
 898     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 899     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 900     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 901     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 902     real             rcutoff_scalar;
 903     real             *shiftvec,*fshift,*x,*f;
 904     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 905     real             scratch[4*DIM];
 906     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 907     real *           vdwioffsetptr0;
 908     real *           vdwgridioffsetptr0;
 909     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 910     real *           vdwioffsetptr1;
 911     real *           vdwgridioffsetptr1;
 912     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 913     real *           vdwioffsetptr2;
 914     real *           vdwgridioffsetptr2;
 915     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 916     real *           vdwioffsetptr3;
 917     real *           vdwgridioffsetptr3;
 918     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 919     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 920     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 921     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 922     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 923     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 924     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 925     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 926     real             *charge;
 927     int              nvdwtype;
 928     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 929     int              *vdwtype;
 930     real             *vdwparam;
 931     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 932     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 933     __m256d           c6grid_00;
 934     __m256d           c6grid_10;
 935     __m256d           c6grid_20;
 936     __m256d           c6grid_30;
 937     real             *vdwgridparam;
 938     __m256d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 939     __m256d           one_half  = _mm256_set1_pd(0.5);
 940     __m256d           minus_one = _mm256_set1_pd(-1.0);
 941     __m128i          ewitab;
 942     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 943     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 944     real             *ewtab;
 945     __m256d          dummy_mask,cutoff_mask;
 946     __m128           tmpmask0,tmpmask1;
 947     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 948     __m256d          one     = _mm256_set1_pd(1.0);
 949     __m256d          two     = _mm256_set1_pd(2.0);
 950     x                = xx[0];
 951     f                = ff[0];
 952
 953     nri              = nlist->nri;
 954     iinr             = nlist->iinr;
 955     jindex           = nlist->jindex;
 956     jjnr             = nlist->jjnr;
 957     shiftidx         = nlist->shift;
 958     gid              = nlist->gid;
 959     shiftvec         = fr->shift_vec[0];
 960     fshift           = fr->fshift[0];
 961     facel            = _mm256_set1_pd(fr->epsfac);
 962     charge           = mdatoms->chargeA;
 963     nvdwtype         = fr->ntype;
 964     vdwparam         = fr->nbfp;
 965     vdwtype          = mdatoms->typeA;
 966     vdwgridparam     = fr->ljpme_c6grid;
 967     sh_lj_ewald      = _mm256_set1_pd(fr->ic->sh_lj_ewald);
 968     ewclj            = _mm256_set1_pd(fr->ewaldcoeff_lj);
 969     ewclj2           = _mm256_mul_pd(minus_one,_mm256_mul_pd(ewclj,ewclj));
 970
 971     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 972     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 973     beta2            = _mm256_mul_pd(beta,beta);
 974     beta3            = _mm256_mul_pd(beta,beta2);
 975
 976     ewtab            = fr->ic->tabq_coul_F;
 977     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 978     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 979
 980     /* Setup water-specific parameters */
 981     inr              = nlist->iinr[0];
 982     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 983     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 984     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 985     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 986     vdwgridioffsetptr0 = vdwgridparam+2*nvdwtype*vdwtype[inr+0];
 987
 988     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 989     rcutoff_scalar   = fr->rcoulomb;
 990     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 991     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 992
 993     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 994     rvdw             = _mm256_set1_pd(fr->rvdw);
 995
 996     /* Avoid stupid compiler warnings */
 997     jnrA = jnrB = jnrC = jnrD = 0;
 998     j_coord_offsetA = 0;
 999     j_coord_offsetB = 0;
1000     j_coord_offsetC = 0;
1001     j_coord_offsetD = 0;
1002
1003     outeriter        = 0;
1004     inneriter        = 0;
1005
1006     for(iidx=0;iidx<4*DIM;iidx++)
1007     {
1008         scratch[iidx] = 0.0;
1009     }
1010
1011     /* Start outer loop over neighborlists */
1012     for(iidx=0; iidx<nri; iidx++)
1013     {
1014         /* Load shift vector for this list */
1015         i_shift_offset   = DIM*shiftidx[iidx];
1016
1017         /* Load limits for loop over neighbors */
1018         j_index_start    = jindex[iidx];
1019         j_index_end      = jindex[iidx+1];
1020
1021         /* Get outer coordinate index */
1022         inr              = iinr[iidx];
1023         i_coord_offset   = DIM*inr;
1024
1025         /* Load i particle coords and add shift vector */
1026         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
1027                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1028
1029         fix0             = _mm256_setzero_pd();
1030         fiy0             = _mm256_setzero_pd();
1031         fiz0             = _mm256_setzero_pd();
1032         fix1             = _mm256_setzero_pd();
1033         fiy1             = _mm256_setzero_pd();
1034         fiz1             = _mm256_setzero_pd();
1035         fix2             = _mm256_setzero_pd();
1036         fiy2             = _mm256_setzero_pd();
1037         fiz2             = _mm256_setzero_pd();
1038         fix3             = _mm256_setzero_pd();
1039         fiy3             = _mm256_setzero_pd();
1040         fiz3             = _mm256_setzero_pd();
1041
1042         /* Start inner kernel loop */
1043         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1044         {
1045
1046             /* Get j neighbor index, and coordinate index */
1047             jnrA             = jjnr[jidx];
1048             jnrB             = jjnr[jidx+1];
1049             jnrC             = jjnr[jidx+2];
1050             jnrD             = jjnr[jidx+3];
1051             j_coord_offsetA  = DIM*jnrA;
1052             j_coord_offsetB  = DIM*jnrB;
1053             j_coord_offsetC  = DIM*jnrC;
1054             j_coord_offsetD  = DIM*jnrD;
1055
1056             /* load j atom coordinates */
1057             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1058                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1059                                                  &jx0,&jy0,&jz0);
1060
1061             /* Calculate displacement vector */
1062             dx00             = _mm256_sub_pd(ix0,jx0);
1063             dy00             = _mm256_sub_pd(iy0,jy0);
1064             dz00             = _mm256_sub_pd(iz0,jz0);
1065             dx10             = _mm256_sub_pd(ix1,jx0);
1066             dy10             = _mm256_sub_pd(iy1,jy0);
1067             dz10             = _mm256_sub_pd(iz1,jz0);
1068             dx20             = _mm256_sub_pd(ix2,jx0);
1069             dy20             = _mm256_sub_pd(iy2,jy0);
1070             dz20             = _mm256_sub_pd(iz2,jz0);
1071             dx30             = _mm256_sub_pd(ix3,jx0);
1072             dy30             = _mm256_sub_pd(iy3,jy0);
1073             dz30             = _mm256_sub_pd(iz3,jz0);
1074
1075             /* Calculate squared distance and things based on it */
1076             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1077             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1078             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1079             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1080
1081             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1082             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1083             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1084             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1085
1086             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1087             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1088             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1089             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1090
1091             /* Load parameters for j particles */
1092             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1093                                                                  charge+jnrC+0,charge+jnrD+0);
1094             vdwjidx0A        = 2*vdwtype[jnrA+0];
1095             vdwjidx0B        = 2*vdwtype[jnrB+0];
1096             vdwjidx0C        = 2*vdwtype[jnrC+0];
1097             vdwjidx0D        = 2*vdwtype[jnrD+0];
1098
1099             fjx0             = _mm256_setzero_pd();
1100             fjy0             = _mm256_setzero_pd();
1101             fjz0             = _mm256_setzero_pd();
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1108             {
1109
1110             r00              = _mm256_mul_pd(rsq00,rinv00);
1111
1112             /* Compute parameters for interactions between i and j atoms */
1113             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1114                                             vdwioffsetptr0+vdwjidx0B,
1115                                             vdwioffsetptr0+vdwjidx0C,
1116                                             vdwioffsetptr0+vdwjidx0D,
1117                                             &c6_00,&c12_00);
1118
1119             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1120                                                                   vdwgridioffsetptr0+vdwjidx0B,
1121                                                                   vdwgridioffsetptr0+vdwjidx0C,
1122                                                                   vdwgridioffsetptr0+vdwjidx0D);
1123
1124             /* Analytical LJ-PME */
1125             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1126             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1127             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1128             exponent         = gmx_simd_exp_d(ewcljrsq);
1129             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1130             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1131             /* f6A = 6 * C6grid * (1 - poly) */
1132             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1133             /* f6B = C6grid * exponent * beta^6 */
1134             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1135             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1136             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);
1137
1138             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1139
1140             fscal            = fvdw;
1141
1142             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1143
1144             /* Calculate temporary vectorial force */
1145             tx               = _mm256_mul_pd(fscal,dx00);
1146             ty               = _mm256_mul_pd(fscal,dy00);
1147             tz               = _mm256_mul_pd(fscal,dz00);
1148
1149             /* Update vectorial force */
1150             fix0             = _mm256_add_pd(fix0,tx);
1151             fiy0             = _mm256_add_pd(fiy0,ty);
1152             fiz0             = _mm256_add_pd(fiz0,tz);
1153
1154             fjx0             = _mm256_add_pd(fjx0,tx);
1155             fjy0             = _mm256_add_pd(fjy0,ty);
1156             fjz0             = _mm256_add_pd(fjz0,tz);
1157
1158             }
1159
1160             /**************************
1161              * CALCULATE INTERACTIONS *
1162              **************************/
1163
1164             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1165             {
1166
1167             r10              = _mm256_mul_pd(rsq10,rinv10);
1168
1169             /* Compute parameters for interactions between i and j atoms */
1170             qq10             = _mm256_mul_pd(iq1,jq0);
1171
1172             /* EWALD ELECTROSTATICS */
1173
1174             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1175             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1176             ewitab           = _mm256_cvttpd_epi32(ewrt);
1177             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1178             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1179                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1180                                             &ewtabF,&ewtabFn);
1181             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1182             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1183
1184             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1185
1186             fscal            = felec;
1187
1188             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1189
1190             /* Calculate temporary vectorial force */
1191             tx               = _mm256_mul_pd(fscal,dx10);
1192             ty               = _mm256_mul_pd(fscal,dy10);
1193             tz               = _mm256_mul_pd(fscal,dz10);
1194
1195             /* Update vectorial force */
1196             fix1             = _mm256_add_pd(fix1,tx);
1197             fiy1             = _mm256_add_pd(fiy1,ty);
1198             fiz1             = _mm256_add_pd(fiz1,tz);
1199
1200             fjx0             = _mm256_add_pd(fjx0,tx);
1201             fjy0             = _mm256_add_pd(fjy0,ty);
1202             fjz0             = _mm256_add_pd(fjz0,tz);
1203
1204             }
1205
1206             /**************************
1207              * CALCULATE INTERACTIONS *
1208              **************************/
1209
1210             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1211             {
1212
1213             r20              = _mm256_mul_pd(rsq20,rinv20);
1214
1215             /* Compute parameters for interactions between i and j atoms */
1216             qq20             = _mm256_mul_pd(iq2,jq0);
1217
1218             /* EWALD ELECTROSTATICS */
1219
1220             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1221             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1222             ewitab           = _mm256_cvttpd_epi32(ewrt);
1223             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1224             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1225                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1226                                             &ewtabF,&ewtabFn);
1227             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1228             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1229
1230             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1231
1232             fscal            = felec;
1233
1234             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1235
1236             /* Calculate temporary vectorial force */
1237             tx               = _mm256_mul_pd(fscal,dx20);
1238             ty               = _mm256_mul_pd(fscal,dy20);
1239             tz               = _mm256_mul_pd(fscal,dz20);
1240
1241             /* Update vectorial force */
1242             fix2             = _mm256_add_pd(fix2,tx);
1243             fiy2             = _mm256_add_pd(fiy2,ty);
1244             fiz2             = _mm256_add_pd(fiz2,tz);
1245
1246             fjx0             = _mm256_add_pd(fjx0,tx);
1247             fjy0             = _mm256_add_pd(fjy0,ty);
1248             fjz0             = _mm256_add_pd(fjz0,tz);
1249
1250             }
1251
1252             /**************************
1253              * CALCULATE INTERACTIONS *
1254              **************************/
1255
1256             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1257             {
1258
1259             r30              = _mm256_mul_pd(rsq30,rinv30);
1260
1261             /* Compute parameters for interactions between i and j atoms */
1262             qq30             = _mm256_mul_pd(iq3,jq0);
1263
1264             /* EWALD ELECTROSTATICS */
1265
1266             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1267             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1268             ewitab           = _mm256_cvttpd_epi32(ewrt);
1269             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1270             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1271                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1272                                             &ewtabF,&ewtabFn);
1273             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1274             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1275
1276             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1277
1278             fscal            = felec;
1279
1280             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1281
1282             /* Calculate temporary vectorial force */
1283             tx               = _mm256_mul_pd(fscal,dx30);
1284             ty               = _mm256_mul_pd(fscal,dy30);
1285             tz               = _mm256_mul_pd(fscal,dz30);
1286
1287             /* Update vectorial force */
1288             fix3             = _mm256_add_pd(fix3,tx);
1289             fiy3             = _mm256_add_pd(fiy3,ty);
1290             fiz3             = _mm256_add_pd(fiz3,tz);
1291
1292             fjx0             = _mm256_add_pd(fjx0,tx);
1293             fjy0             = _mm256_add_pd(fjy0,ty);
1294             fjz0             = _mm256_add_pd(fjz0,tz);
1295
1296             }
1297
1298             fjptrA             = f+j_coord_offsetA;
1299             fjptrB             = f+j_coord_offsetB;
1300             fjptrC             = f+j_coord_offsetC;
1301             fjptrD             = f+j_coord_offsetD;
1302
1303             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1304
1305             /* Inner loop uses 169 flops */
1306         }
1307
1308         if(jidx<j_index_end)
1309         {
1310
1311             /* Get j neighbor index, and coordinate index */
1312             jnrlistA         = jjnr[jidx];
1313             jnrlistB         = jjnr[jidx+1];
1314             jnrlistC         = jjnr[jidx+2];
1315             jnrlistD         = jjnr[jidx+3];
1316             /* Sign of each element will be negative for non-real atoms.
1317              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1318              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1319              */
1320             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1321
1322             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1323             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1324             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1325
1326             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1327             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1328             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1329             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1330             j_coord_offsetA  = DIM*jnrA;
1331             j_coord_offsetB  = DIM*jnrB;
1332             j_coord_offsetC  = DIM*jnrC;
1333             j_coord_offsetD  = DIM*jnrD;
1334
1335             /* load j atom coordinates */
1336             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1337                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1338                                                  &jx0,&jy0,&jz0);
1339
1340             /* Calculate displacement vector */
1341             dx00             = _mm256_sub_pd(ix0,jx0);
1342             dy00             = _mm256_sub_pd(iy0,jy0);
1343             dz00             = _mm256_sub_pd(iz0,jz0);
1344             dx10             = _mm256_sub_pd(ix1,jx0);
1345             dy10             = _mm256_sub_pd(iy1,jy0);
1346             dz10             = _mm256_sub_pd(iz1,jz0);
1347             dx20             = _mm256_sub_pd(ix2,jx0);
1348             dy20             = _mm256_sub_pd(iy2,jy0);
1349             dz20             = _mm256_sub_pd(iz2,jz0);
1350             dx30             = _mm256_sub_pd(ix3,jx0);
1351             dy30             = _mm256_sub_pd(iy3,jy0);
1352             dz30             = _mm256_sub_pd(iz3,jz0);
1353
1354             /* Calculate squared distance and things based on it */
1355             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1356             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1357             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1358             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1359
1360             rinv00           = gmx_mm256_invsqrt_pd(rsq00);
1361             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1362             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1363             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1364
1365             rinvsq00         = _mm256_mul_pd(rinv00,rinv00);
1366             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1367             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1368             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1369
1370             /* Load parameters for j particles */
1371             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1372                                                                  charge+jnrC+0,charge+jnrD+0);
1373             vdwjidx0A        = 2*vdwtype[jnrA+0];
1374             vdwjidx0B        = 2*vdwtype[jnrB+0];
1375             vdwjidx0C        = 2*vdwtype[jnrC+0];
1376             vdwjidx0D        = 2*vdwtype[jnrD+0];
1377
1378             fjx0             = _mm256_setzero_pd();
1379             fjy0             = _mm256_setzero_pd();
1380             fjz0             = _mm256_setzero_pd();
1381
1382             /**************************
1383              * CALCULATE INTERACTIONS *
1384              **************************/
1385
1386             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1387             {
1388
1389             r00              = _mm256_mul_pd(rsq00,rinv00);
1390             r00              = _mm256_andnot_pd(dummy_mask,r00);
1391
1392             /* Compute parameters for interactions between i and j atoms */
1393             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1394                                             vdwioffsetptr0+vdwjidx0B,
1395                                             vdwioffsetptr0+vdwjidx0C,
1396                                             vdwioffsetptr0+vdwjidx0D,
1397                                             &c6_00,&c12_00);
1398
1399             c6grid_00       = gmx_mm256_load_4real_swizzle_pd(vdwgridioffsetptr0+vdwjidx0A,
1400                                                                   vdwgridioffsetptr0+vdwjidx0B,
1401                                                                   vdwgridioffsetptr0+vdwjidx0C,
1402                                                                   vdwgridioffsetptr0+vdwjidx0D);
1403
1404             /* Analytical LJ-PME */
1405             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1406             ewcljrsq         = _mm256_mul_pd(ewclj2,rsq00);
1407             ewclj6           = _mm256_mul_pd(ewclj2,_mm256_mul_pd(ewclj2,ewclj2));
1408             exponent         = gmx_simd_exp_d(ewcljrsq);
1409             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1410             poly             = _mm256_mul_pd(exponent,_mm256_add_pd(_mm256_sub_pd(one,ewcljrsq),_mm256_mul_pd(_mm256_mul_pd(ewcljrsq,ewcljrsq),one_half)));
1411             /* f6A = 6 * C6grid * (1 - poly) */
1412             f6A              = _mm256_mul_pd(c6grid_00,_mm256_sub_pd(one,poly));
1413             /* f6B = C6grid * exponent * beta^6 */
1414             f6B              = _mm256_mul_pd(_mm256_mul_pd(c6grid_00,one_sixth),_mm256_mul_pd(exponent,ewclj6));
1415             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1416             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);
1417
1418             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1419
1420             fscal            = fvdw;
1421
1422             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1423
1424             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1425
1426             /* Calculate temporary vectorial force */
1427             tx               = _mm256_mul_pd(fscal,dx00);
1428             ty               = _mm256_mul_pd(fscal,dy00);
1429             tz               = _mm256_mul_pd(fscal,dz00);
1430
1431             /* Update vectorial force */
1432             fix0             = _mm256_add_pd(fix0,tx);
1433             fiy0             = _mm256_add_pd(fiy0,ty);
1434             fiz0             = _mm256_add_pd(fiz0,tz);
1435
1436             fjx0             = _mm256_add_pd(fjx0,tx);
1437             fjy0             = _mm256_add_pd(fjy0,ty);
1438             fjz0             = _mm256_add_pd(fjz0,tz);
1439
1440             }
1441
1442             /**************************
1443              * CALCULATE INTERACTIONS *
1444              **************************/
1445
1446             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1447             {
1448
1449             r10              = _mm256_mul_pd(rsq10,rinv10);
1450             r10              = _mm256_andnot_pd(dummy_mask,r10);
1451
1452             /* Compute parameters for interactions between i and j atoms */
1453             qq10             = _mm256_mul_pd(iq1,jq0);
1454
1455             /* EWALD ELECTROSTATICS */
1456
1457             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1458             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1459             ewitab           = _mm256_cvttpd_epi32(ewrt);
1460             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1461             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1462                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1463                                             &ewtabF,&ewtabFn);
1464             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1465             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1466
1467             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1468
1469             fscal            = felec;
1470
1471             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1472
1473             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1474
1475             /* Calculate temporary vectorial force */
1476             tx               = _mm256_mul_pd(fscal,dx10);
1477             ty               = _mm256_mul_pd(fscal,dy10);
1478             tz               = _mm256_mul_pd(fscal,dz10);
1479
1480             /* Update vectorial force */
1481             fix1             = _mm256_add_pd(fix1,tx);
1482             fiy1             = _mm256_add_pd(fiy1,ty);
1483             fiz1             = _mm256_add_pd(fiz1,tz);
1484
1485             fjx0             = _mm256_add_pd(fjx0,tx);
1486             fjy0             = _mm256_add_pd(fjy0,ty);
1487             fjz0             = _mm256_add_pd(fjz0,tz);
1488
1489             }
1490
1491             /**************************
1492              * CALCULATE INTERACTIONS *
1493              **************************/
1494
1495             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1496             {
1497
1498             r20              = _mm256_mul_pd(rsq20,rinv20);
1499             r20              = _mm256_andnot_pd(dummy_mask,r20);
1500
1501             /* Compute parameters for interactions between i and j atoms */
1502             qq20             = _mm256_mul_pd(iq2,jq0);
1503
1504             /* EWALD ELECTROSTATICS */
1505
1506             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1507             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1508             ewitab           = _mm256_cvttpd_epi32(ewrt);
1509             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1510             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1511                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1512                                             &ewtabF,&ewtabFn);
1513             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1514             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1515
1516             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1517
1518             fscal            = felec;
1519
1520             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1521
1522             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1523
1524             /* Calculate temporary vectorial force */
1525             tx               = _mm256_mul_pd(fscal,dx20);
1526             ty               = _mm256_mul_pd(fscal,dy20);
1527             tz               = _mm256_mul_pd(fscal,dz20);
1528
1529             /* Update vectorial force */
1530             fix2             = _mm256_add_pd(fix2,tx);
1531             fiy2             = _mm256_add_pd(fiy2,ty);
1532             fiz2             = _mm256_add_pd(fiz2,tz);
1533
1534             fjx0             = _mm256_add_pd(fjx0,tx);
1535             fjy0             = _mm256_add_pd(fjy0,ty);
1536             fjz0             = _mm256_add_pd(fjz0,tz);
1537
1538             }
1539
1540             /**************************
1541              * CALCULATE INTERACTIONS *
1542              **************************/
1543
1544             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1545             {
1546
1547             r30              = _mm256_mul_pd(rsq30,rinv30);
1548             r30              = _mm256_andnot_pd(dummy_mask,r30);
1549
1550             /* Compute parameters for interactions between i and j atoms */
1551             qq30             = _mm256_mul_pd(iq3,jq0);
1552
1553             /* EWALD ELECTROSTATICS */
1554
1555             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1556             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1557             ewitab           = _mm256_cvttpd_epi32(ewrt);
1558             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1559             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1560                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1561                                             &ewtabF,&ewtabFn);
1562             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1563             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1564
1565             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1566
1567             fscal            = felec;
1568
1569             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1570
1571             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1572
1573             /* Calculate temporary vectorial force */
1574             tx               = _mm256_mul_pd(fscal,dx30);
1575             ty               = _mm256_mul_pd(fscal,dy30);
1576             tz               = _mm256_mul_pd(fscal,dz30);
1577
1578             /* Update vectorial force */
1579             fix3             = _mm256_add_pd(fix3,tx);
1580             fiy3             = _mm256_add_pd(fiy3,ty);
1581             fiz3             = _mm256_add_pd(fiz3,tz);
1582
1583             fjx0             = _mm256_add_pd(fjx0,tx);
1584             fjy0             = _mm256_add_pd(fjy0,ty);
1585             fjz0             = _mm256_add_pd(fjz0,tz);
1586
1587             }
1588
1589             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1590             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1591             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1592             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1593
1594             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1595
1596             /* Inner loop uses 173 flops */
1597         }
1598
1599         /* End of innermost loop */
1600
1601         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1602                                                  f+i_coord_offset,fshift+i_shift_offset);
1603
1604         /* Increment number of inner iterations */
1605         inneriter                  += j_index_end - j_index_start;
1606
1607         /* Outer loop uses 24 flops */
1608     }
1609
1610     /* Increment number of outer iterations */
1611     outeriter        += nri;
1612
1613     /* Update outer/inner flops */
1614
1615     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*173);
1616 }