src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecEwSh_VdwLJSh_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 "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_double.h"
  48 #include "kernelutil_x86_avx_256_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_256_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_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     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  86     real *           vdwioffsetptr1;
  87     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  88     real *           vdwioffsetptr2;
  89     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  90     real *           vdwioffsetptr3;
  91     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  92     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  93     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  94     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  95     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  96     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  97     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  98     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  99     real             *charge;
 100     int              nvdwtype;
 101     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 102     int              *vdwtype;
 103     real             *vdwparam;
 104     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 105     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 106     __m128i          ewitab;
 107     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 109     real             *ewtab;
 110     __m256d          dummy_mask,cutoff_mask;
 111     __m128           tmpmask0,tmpmask1;
 112     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 113     __m256d          one     = _mm256_set1_pd(1.0);
 114     __m256d          two     = _mm256_set1_pd(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm256_set1_pd(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 133     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 134     beta2            = _mm256_mul_pd(beta,beta);
 135     beta3            = _mm256_mul_pd(beta,beta2);
 136
 137     ewtab            = fr->ic->tabq_coul_FDV0;
 138     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 139     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 140
 141     /* Setup water-specific parameters */
 142     inr              = nlist->iinr[0];
 143     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 144     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 145     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 146     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 147
 148     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 149     rcutoff_scalar   = fr->rcoulomb;
 150     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 151     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 152
 153     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 154     rvdw             = _mm256_set1_pd(fr->rvdw);
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = jnrC = jnrD = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160     j_coord_offsetC = 0;
 161     j_coord_offsetD = 0;
 162
 163     outeriter        = 0;
 164     inneriter        = 0;
 165
 166     for(iidx=0;iidx<4*DIM;iidx++)
 167     {
 168         scratch[iidx] = 0.0;
 169     }
 170
 171     /* Start outer loop over neighborlists */
 172     for(iidx=0; iidx<nri; iidx++)
 173     {
 174         /* Load shift vector for this list */
 175         i_shift_offset   = DIM*shiftidx[iidx];
 176
 177         /* Load limits for loop over neighbors */
 178         j_index_start    = jindex[iidx];
 179         j_index_end      = jindex[iidx+1];
 180
 181         /* Get outer coordinate index */
 182         inr              = iinr[iidx];
 183         i_coord_offset   = DIM*inr;
 184
 185         /* Load i particle coords and add shift vector */
 186         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 187                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 188
 189         fix0             = _mm256_setzero_pd();
 190         fiy0             = _mm256_setzero_pd();
 191         fiz0             = _mm256_setzero_pd();
 192         fix1             = _mm256_setzero_pd();
 193         fiy1             = _mm256_setzero_pd();
 194         fiz1             = _mm256_setzero_pd();
 195         fix2             = _mm256_setzero_pd();
 196         fiy2             = _mm256_setzero_pd();
 197         fiz2             = _mm256_setzero_pd();
 198         fix3             = _mm256_setzero_pd();
 199         fiy3             = _mm256_setzero_pd();
 200         fiz3             = _mm256_setzero_pd();
 201
 202         /* Reset potential sums */
 203         velecsum         = _mm256_setzero_pd();
 204         vvdwsum          = _mm256_setzero_pd();
 205
 206         /* Start inner kernel loop */
 207         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 208         {
 209
 210             /* Get j neighbor index, and coordinate index */
 211             jnrA             = jjnr[jidx];
 212             jnrB             = jjnr[jidx+1];
 213             jnrC             = jjnr[jidx+2];
 214             jnrD             = jjnr[jidx+3];
 215             j_coord_offsetA  = DIM*jnrA;
 216             j_coord_offsetB  = DIM*jnrB;
 217             j_coord_offsetC  = DIM*jnrC;
 218             j_coord_offsetD  = DIM*jnrD;
 219
 220             /* load j atom coordinates */
 221             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 222                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 223                                                  &jx0,&jy0,&jz0);
 224
 225             /* Calculate displacement vector */
 226             dx00             = _mm256_sub_pd(ix0,jx0);
 227             dy00             = _mm256_sub_pd(iy0,jy0);
 228             dz00             = _mm256_sub_pd(iz0,jz0);
 229             dx10             = _mm256_sub_pd(ix1,jx0);
 230             dy10             = _mm256_sub_pd(iy1,jy0);
 231             dz10             = _mm256_sub_pd(iz1,jz0);
 232             dx20             = _mm256_sub_pd(ix2,jx0);
 233             dy20             = _mm256_sub_pd(iy2,jy0);
 234             dz20             = _mm256_sub_pd(iz2,jz0);
 235             dx30             = _mm256_sub_pd(ix3,jx0);
 236             dy30             = _mm256_sub_pd(iy3,jy0);
 237             dz30             = _mm256_sub_pd(iz3,jz0);
 238
 239             /* Calculate squared distance and things based on it */
 240             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 241             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 242             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 243             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 244
 245             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 246             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 247             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 248
 249             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 250             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 251             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 252             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 253
 254             /* Load parameters for j particles */
 255             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 256                                                                  charge+jnrC+0,charge+jnrD+0);
 257             vdwjidx0A        = 2*vdwtype[jnrA+0];
 258             vdwjidx0B        = 2*vdwtype[jnrB+0];
 259             vdwjidx0C        = 2*vdwtype[jnrC+0];
 260             vdwjidx0D        = 2*vdwtype[jnrD+0];
 261
 262             fjx0             = _mm256_setzero_pd();
 263             fjy0             = _mm256_setzero_pd();
 264             fjz0             = _mm256_setzero_pd();
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 271             {
 272
 273             /* Compute parameters for interactions between i and j atoms */
 274             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 275                                             vdwioffsetptr0+vdwjidx0B,
 276                                             vdwioffsetptr0+vdwjidx0C,
 277                                             vdwioffsetptr0+vdwjidx0D,
 278                                             &c6_00,&c12_00);
 279
 280             /* LENNARD-JONES DISPERSION/REPULSION */
 281
 282             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 283             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 284             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 285             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) ,
 286                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 287             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 288
 289             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 290
 291             /* Update potential sum for this i atom from the interaction with this j atom. */
 292             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 293             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 294
 295             fscal            = fvdw;
 296
 297             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm256_mul_pd(fscal,dx00);
 301             ty               = _mm256_mul_pd(fscal,dy00);
 302             tz               = _mm256_mul_pd(fscal,dz00);
 303
 304             /* Update vectorial force */
 305             fix0             = _mm256_add_pd(fix0,tx);
 306             fiy0             = _mm256_add_pd(fiy0,ty);
 307             fiz0             = _mm256_add_pd(fiz0,tz);
 308
 309             fjx0             = _mm256_add_pd(fjx0,tx);
 310             fjy0             = _mm256_add_pd(fjy0,ty);
 311             fjz0             = _mm256_add_pd(fjz0,tz);
 312
 313             }
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 320             {
 321
 322             r10              = _mm256_mul_pd(rsq10,rinv10);
 323
 324             /* Compute parameters for interactions between i and j atoms */
 325             qq10             = _mm256_mul_pd(iq1,jq0);
 326
 327             /* EWALD ELECTROSTATICS */
 328
 329             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 330             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 331             ewitab           = _mm256_cvttpd_epi32(ewrt);
 332             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 333             ewitab           = _mm_slli_epi32(ewitab,2);
 334             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 335             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 336             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 337             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 338             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 339             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 340             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 341             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 342             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 343
 344             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velec            = _mm256_and_pd(velec,cutoff_mask);
 348             velecsum         = _mm256_add_pd(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm256_mul_pd(fscal,dx10);
 356             ty               = _mm256_mul_pd(fscal,dy10);
 357             tz               = _mm256_mul_pd(fscal,dz10);
 358
 359             /* Update vectorial force */
 360             fix1             = _mm256_add_pd(fix1,tx);
 361             fiy1             = _mm256_add_pd(fiy1,ty);
 362             fiz1             = _mm256_add_pd(fiz1,tz);
 363
 364             fjx0             = _mm256_add_pd(fjx0,tx);
 365             fjy0             = _mm256_add_pd(fjy0,ty);
 366             fjz0             = _mm256_add_pd(fjz0,tz);
 367
 368             }
 369
 370             /**************************
 371              * CALCULATE INTERACTIONS *
 372              **************************/
 373
 374             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 375             {
 376
 377             r20              = _mm256_mul_pd(rsq20,rinv20);
 378
 379             /* Compute parameters for interactions between i and j atoms */
 380             qq20             = _mm256_mul_pd(iq2,jq0);
 381
 382             /* EWALD ELECTROSTATICS */
 383
 384             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 385             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 386             ewitab           = _mm256_cvttpd_epi32(ewrt);
 387             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 388             ewitab           = _mm_slli_epi32(ewitab,2);
 389             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 390             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 391             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 392             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 393             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 394             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 395             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 396             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 397             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 398
 399             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 400
 401             /* Update potential sum for this i atom from the interaction with this j atom. */
 402             velec            = _mm256_and_pd(velec,cutoff_mask);
 403             velecsum         = _mm256_add_pd(velecsum,velec);
 404
 405             fscal            = felec;
 406
 407             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 408
 409             /* Calculate temporary vectorial force */
 410             tx               = _mm256_mul_pd(fscal,dx20);
 411             ty               = _mm256_mul_pd(fscal,dy20);
 412             tz               = _mm256_mul_pd(fscal,dz20);
 413
 414             /* Update vectorial force */
 415             fix2             = _mm256_add_pd(fix2,tx);
 416             fiy2             = _mm256_add_pd(fiy2,ty);
 417             fiz2             = _mm256_add_pd(fiz2,tz);
 418
 419             fjx0             = _mm256_add_pd(fjx0,tx);
 420             fjy0             = _mm256_add_pd(fjy0,ty);
 421             fjz0             = _mm256_add_pd(fjz0,tz);
 422
 423             }
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 430             {
 431
 432             r30              = _mm256_mul_pd(rsq30,rinv30);
 433
 434             /* Compute parameters for interactions between i and j atoms */
 435             qq30             = _mm256_mul_pd(iq3,jq0);
 436
 437             /* EWALD ELECTROSTATICS */
 438
 439             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 440             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 441             ewitab           = _mm256_cvttpd_epi32(ewrt);
 442             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 443             ewitab           = _mm_slli_epi32(ewitab,2);
 444             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 445             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 446             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 447             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 448             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 449             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 450             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 451             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 452             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 453
 454             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm256_and_pd(velec,cutoff_mask);
 458             velecsum         = _mm256_add_pd(velecsum,velec);
 459
 460             fscal            = felec;
 461
 462             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm256_mul_pd(fscal,dx30);
 466             ty               = _mm256_mul_pd(fscal,dy30);
 467             tz               = _mm256_mul_pd(fscal,dz30);
 468
 469             /* Update vectorial force */
 470             fix3             = _mm256_add_pd(fix3,tx);
 471             fiy3             = _mm256_add_pd(fiy3,ty);
 472             fiz3             = _mm256_add_pd(fiz3,tz);
 473
 474             fjx0             = _mm256_add_pd(fjx0,tx);
 475             fjy0             = _mm256_add_pd(fjy0,ty);
 476             fjz0             = _mm256_add_pd(fjz0,tz);
 477
 478             }
 479
 480             fjptrA             = f+j_coord_offsetA;
 481             fjptrB             = f+j_coord_offsetB;
 482             fjptrC             = f+j_coord_offsetC;
 483             fjptrD             = f+j_coord_offsetD;
 484
 485             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 486
 487             /* Inner loop uses 182 flops */
 488         }
 489
 490         if(jidx<j_index_end)
 491         {
 492
 493             /* Get j neighbor index, and coordinate index */
 494             jnrlistA         = jjnr[jidx];
 495             jnrlistB         = jjnr[jidx+1];
 496             jnrlistC         = jjnr[jidx+2];
 497             jnrlistD         = jjnr[jidx+3];
 498             /* Sign of each element will be negative for non-real atoms.
 499              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 500              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 501              */
 502             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 503
 504             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 505             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 506             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 507
 508             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 509             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 510             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 511             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 512             j_coord_offsetA  = DIM*jnrA;
 513             j_coord_offsetB  = DIM*jnrB;
 514             j_coord_offsetC  = DIM*jnrC;
 515             j_coord_offsetD  = DIM*jnrD;
 516
 517             /* load j atom coordinates */
 518             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 519                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 520                                                  &jx0,&jy0,&jz0);
 521
 522             /* Calculate displacement vector */
 523             dx00             = _mm256_sub_pd(ix0,jx0);
 524             dy00             = _mm256_sub_pd(iy0,jy0);
 525             dz00             = _mm256_sub_pd(iz0,jz0);
 526             dx10             = _mm256_sub_pd(ix1,jx0);
 527             dy10             = _mm256_sub_pd(iy1,jy0);
 528             dz10             = _mm256_sub_pd(iz1,jz0);
 529             dx20             = _mm256_sub_pd(ix2,jx0);
 530             dy20             = _mm256_sub_pd(iy2,jy0);
 531             dz20             = _mm256_sub_pd(iz2,jz0);
 532             dx30             = _mm256_sub_pd(ix3,jx0);
 533             dy30             = _mm256_sub_pd(iy3,jy0);
 534             dz30             = _mm256_sub_pd(iz3,jz0);
 535
 536             /* Calculate squared distance and things based on it */
 537             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 538             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 539             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 540             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 541
 542             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
 543             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
 544             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
 545
 546             rinvsq00         = gmx_mm256_inv_pd(rsq00);
 547             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 548             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 549             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 550
 551             /* Load parameters for j particles */
 552             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 553                                                                  charge+jnrC+0,charge+jnrD+0);
 554             vdwjidx0A        = 2*vdwtype[jnrA+0];
 555             vdwjidx0B        = 2*vdwtype[jnrB+0];
 556             vdwjidx0C        = 2*vdwtype[jnrC+0];
 557             vdwjidx0D        = 2*vdwtype[jnrD+0];
 558
 559             fjx0             = _mm256_setzero_pd();
 560             fjy0             = _mm256_setzero_pd();
 561             fjz0             = _mm256_setzero_pd();
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 568             {
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 572                                             vdwioffsetptr0+vdwjidx0B,
 573                                             vdwioffsetptr0+vdwjidx0C,
 574                                             vdwioffsetptr0+vdwjidx0D,
 575                                             &c6_00,&c12_00);
 576
 577             /* LENNARD-JONES DISPERSION/REPULSION */
 578
 579             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 580             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 581             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 582             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) ,
 583                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 584             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 585
 586             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 590             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 591             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 592
 593             fscal            = fvdw;
 594
 595             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 596
 597             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 598
 599             /* Calculate temporary vectorial force */
 600             tx               = _mm256_mul_pd(fscal,dx00);
 601             ty               = _mm256_mul_pd(fscal,dy00);
 602             tz               = _mm256_mul_pd(fscal,dz00);
 603
 604             /* Update vectorial force */
 605             fix0             = _mm256_add_pd(fix0,tx);
 606             fiy0             = _mm256_add_pd(fiy0,ty);
 607             fiz0             = _mm256_add_pd(fiz0,tz);
 608
 609             fjx0             = _mm256_add_pd(fjx0,tx);
 610             fjy0             = _mm256_add_pd(fjy0,ty);
 611             fjz0             = _mm256_add_pd(fjz0,tz);
 612
 613             }
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 620             {
 621
 622             r10              = _mm256_mul_pd(rsq10,rinv10);
 623             r10              = _mm256_andnot_pd(dummy_mask,r10);
 624
 625             /* Compute parameters for interactions between i and j atoms */
 626             qq10             = _mm256_mul_pd(iq1,jq0);
 627
 628             /* EWALD ELECTROSTATICS */
 629
 630             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 631             ewrt             = _mm256_mul_pd(r10,ewtabscale);
 632             ewitab           = _mm256_cvttpd_epi32(ewrt);
 633             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 634             ewitab           = _mm_slli_epi32(ewitab,2);
 635             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 636             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 637             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 638             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 639             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 640             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 641             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 642             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_sub_pd(rinv10,sh_ewald),velec));
 643             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
 644
 645             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 646
 647             /* Update potential sum for this i atom from the interaction with this j atom. */
 648             velec            = _mm256_and_pd(velec,cutoff_mask);
 649             velec            = _mm256_andnot_pd(dummy_mask,velec);
 650             velecsum         = _mm256_add_pd(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 655
 656             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 657
 658             /* Calculate temporary vectorial force */
 659             tx               = _mm256_mul_pd(fscal,dx10);
 660             ty               = _mm256_mul_pd(fscal,dy10);
 661             tz               = _mm256_mul_pd(fscal,dz10);
 662
 663             /* Update vectorial force */
 664             fix1             = _mm256_add_pd(fix1,tx);
 665             fiy1             = _mm256_add_pd(fiy1,ty);
 666             fiz1             = _mm256_add_pd(fiz1,tz);
 667
 668             fjx0             = _mm256_add_pd(fjx0,tx);
 669             fjy0             = _mm256_add_pd(fjy0,ty);
 670             fjz0             = _mm256_add_pd(fjz0,tz);
 671
 672             }
 673
 674             /**************************
 675              * CALCULATE INTERACTIONS *
 676              **************************/
 677
 678             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 679             {
 680
 681             r20              = _mm256_mul_pd(rsq20,rinv20);
 682             r20              = _mm256_andnot_pd(dummy_mask,r20);
 683
 684             /* Compute parameters for interactions between i and j atoms */
 685             qq20             = _mm256_mul_pd(iq2,jq0);
 686
 687             /* EWALD ELECTROSTATICS */
 688
 689             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 690             ewrt             = _mm256_mul_pd(r20,ewtabscale);
 691             ewitab           = _mm256_cvttpd_epi32(ewrt);
 692             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 693             ewitab           = _mm_slli_epi32(ewitab,2);
 694             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 695             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 696             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 697             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 698             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 699             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 700             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 701             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_sub_pd(rinv20,sh_ewald),velec));
 702             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
 703
 704             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 705
 706             /* Update potential sum for this i atom from the interaction with this j atom. */
 707             velec            = _mm256_and_pd(velec,cutoff_mask);
 708             velec            = _mm256_andnot_pd(dummy_mask,velec);
 709             velecsum         = _mm256_add_pd(velecsum,velec);
 710
 711             fscal            = felec;
 712
 713             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 714
 715             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 716
 717             /* Calculate temporary vectorial force */
 718             tx               = _mm256_mul_pd(fscal,dx20);
 719             ty               = _mm256_mul_pd(fscal,dy20);
 720             tz               = _mm256_mul_pd(fscal,dz20);
 721
 722             /* Update vectorial force */
 723             fix2             = _mm256_add_pd(fix2,tx);
 724             fiy2             = _mm256_add_pd(fiy2,ty);
 725             fiz2             = _mm256_add_pd(fiz2,tz);
 726
 727             fjx0             = _mm256_add_pd(fjx0,tx);
 728             fjy0             = _mm256_add_pd(fjy0,ty);
 729             fjz0             = _mm256_add_pd(fjz0,tz);
 730
 731             }
 732
 733             /**************************
 734              * CALCULATE INTERACTIONS *
 735              **************************/
 736
 737             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 738             {
 739
 740             r30              = _mm256_mul_pd(rsq30,rinv30);
 741             r30              = _mm256_andnot_pd(dummy_mask,r30);
 742
 743             /* Compute parameters for interactions between i and j atoms */
 744             qq30             = _mm256_mul_pd(iq3,jq0);
 745
 746             /* EWALD ELECTROSTATICS */
 747
 748             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 749             ewrt             = _mm256_mul_pd(r30,ewtabscale);
 750             ewitab           = _mm256_cvttpd_epi32(ewrt);
 751             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 752             ewitab           = _mm_slli_epi32(ewitab,2);
 753             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 754             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 755             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 756             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 757             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 758             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 759             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 760             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_sub_pd(rinv30,sh_ewald),velec));
 761             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
 762
 763             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 764
 765             /* Update potential sum for this i atom from the interaction with this j atom. */
 766             velec            = _mm256_and_pd(velec,cutoff_mask);
 767             velec            = _mm256_andnot_pd(dummy_mask,velec);
 768             velecsum         = _mm256_add_pd(velecsum,velec);
 769
 770             fscal            = felec;
 771
 772             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 773
 774             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 775
 776             /* Calculate temporary vectorial force */
 777             tx               = _mm256_mul_pd(fscal,dx30);
 778             ty               = _mm256_mul_pd(fscal,dy30);
 779             tz               = _mm256_mul_pd(fscal,dz30);
 780
 781             /* Update vectorial force */
 782             fix3             = _mm256_add_pd(fix3,tx);
 783             fiy3             = _mm256_add_pd(fiy3,ty);
 784             fiz3             = _mm256_add_pd(fiz3,tz);
 785
 786             fjx0             = _mm256_add_pd(fjx0,tx);
 787             fjy0             = _mm256_add_pd(fjy0,ty);
 788             fjz0             = _mm256_add_pd(fjz0,tz);
 789
 790             }
 791
 792             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 793             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 794             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 795             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 796
 797             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 798
 799             /* Inner loop uses 185 flops */
 800         }
 801
 802         /* End of innermost loop */
 803
 804         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 805                                                  f+i_coord_offset,fshift+i_shift_offset);
 806
 807         ggid                        = gid[iidx];
 808         /* Update potential energies */
 809         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 810         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 811
 812         /* Increment number of inner iterations */
 813         inneriter                  += j_index_end - j_index_start;
 814
 815         /* Outer loop uses 26 flops */
 816     }
 817
 818     /* Increment number of outer iterations */
 819     outeriter        += nri;
 820
 821     /* Update outer/inner flops */
 822
 823     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*185);
 824 }
 825 /*
 826  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_double
 827  * Electrostatics interaction: Ewald
 828  * VdW interaction:            LennardJones
 829  * Geometry:                   Water4-Particle
 830  * Calculate force/pot:        Force
 831  */
 832 void
 833 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_256_double
 834                     (t_nblist                    * gmx_restrict       nlist,
 835                      rvec                        * gmx_restrict          xx,
 836                      rvec                        * gmx_restrict          ff,
 837                      t_forcerec                  * gmx_restrict          fr,
 838                      t_mdatoms                   * gmx_restrict     mdatoms,
 839                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 840                      t_nrnb                      * gmx_restrict        nrnb)
 841 {
 842     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 843      * just 0 for non-waters.
 844      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 845      * jnr indices corresponding to data put in the four positions in the SIMD register.
 846      */
 847     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 848     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 849     int              jnrA,jnrB,jnrC,jnrD;
 850     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 851     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 852     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 853     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 854     real             rcutoff_scalar;
 855     real             *shiftvec,*fshift,*x,*f;
 856     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 857     real             scratch[4*DIM];
 858     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 859     real *           vdwioffsetptr0;
 860     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 861     real *           vdwioffsetptr1;
 862     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 863     real *           vdwioffsetptr2;
 864     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 865     real *           vdwioffsetptr3;
 866     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 867     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 868     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 869     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 870     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 871     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 872     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 873     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 874     real             *charge;
 875     int              nvdwtype;
 876     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 877     int              *vdwtype;
 878     real             *vdwparam;
 879     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 880     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 881     __m128i          ewitab;
 882     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 883     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 884     real             *ewtab;
 885     __m256d          dummy_mask,cutoff_mask;
 886     __m128           tmpmask0,tmpmask1;
 887     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 888     __m256d          one     = _mm256_set1_pd(1.0);
 889     __m256d          two     = _mm256_set1_pd(2.0);
 890     x                = xx[0];
 891     f                = ff[0];
 892
 893     nri              = nlist->nri;
 894     iinr             = nlist->iinr;
 895     jindex           = nlist->jindex;
 896     jjnr             = nlist->jjnr;
 897     shiftidx         = nlist->shift;
 898     gid              = nlist->gid;
 899     shiftvec         = fr->shift_vec[0];
 900     fshift           = fr->fshift[0];
 901     facel            = _mm256_set1_pd(fr->epsfac);
 902     charge           = mdatoms->chargeA;
 903     nvdwtype         = fr->ntype;
 904     vdwparam         = fr->nbfp;
 905     vdwtype          = mdatoms->typeA;
 906
 907     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 908     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff_q);
 909     beta2            = _mm256_mul_pd(beta,beta);
 910     beta3            = _mm256_mul_pd(beta,beta2);
 911
 912     ewtab            = fr->ic->tabq_coul_F;
 913     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 914     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 915
 916     /* Setup water-specific parameters */
 917     inr              = nlist->iinr[0];
 918     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 919     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 920     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 921     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 922
 923     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 924     rcutoff_scalar   = fr->rcoulomb;
 925     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 926     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 927
 928     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 929     rvdw             = _mm256_set1_pd(fr->rvdw);
 930
 931     /* Avoid stupid compiler warnings */
 932     jnrA = jnrB = jnrC = jnrD = 0;
 933     j_coord_offsetA = 0;
 934     j_coord_offsetB = 0;
 935     j_coord_offsetC = 0;
 936     j_coord_offsetD = 0;
 937
 938     outeriter        = 0;
 939     inneriter        = 0;
 940
 941     for(iidx=0;iidx<4*DIM;iidx++)
 942     {
 943         scratch[iidx] = 0.0;
 944     }
 945
 946     /* Start outer loop over neighborlists */
 947     for(iidx=0; iidx<nri; iidx++)
 948     {
 949         /* Load shift vector for this list */
 950         i_shift_offset   = DIM*shiftidx[iidx];
 951
 952         /* Load limits for loop over neighbors */
 953         j_index_start    = jindex[iidx];
 954         j_index_end      = jindex[iidx+1];
 955
 956         /* Get outer coordinate index */
 957         inr              = iinr[iidx];
 958         i_coord_offset   = DIM*inr;
 959
 960         /* Load i particle coords and add shift vector */
 961         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 962                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 963
 964         fix0             = _mm256_setzero_pd();
 965         fiy0             = _mm256_setzero_pd();
 966         fiz0             = _mm256_setzero_pd();
 967         fix1             = _mm256_setzero_pd();
 968         fiy1             = _mm256_setzero_pd();
 969         fiz1             = _mm256_setzero_pd();
 970         fix2             = _mm256_setzero_pd();
 971         fiy2             = _mm256_setzero_pd();
 972         fiz2             = _mm256_setzero_pd();
 973         fix3             = _mm256_setzero_pd();
 974         fiy3             = _mm256_setzero_pd();
 975         fiz3             = _mm256_setzero_pd();
 976
 977         /* Start inner kernel loop */
 978         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 979         {
 980
 981             /* Get j neighbor index, and coordinate index */
 982             jnrA             = jjnr[jidx];
 983             jnrB             = jjnr[jidx+1];
 984             jnrC             = jjnr[jidx+2];
 985             jnrD             = jjnr[jidx+3];
 986             j_coord_offsetA  = DIM*jnrA;
 987             j_coord_offsetB  = DIM*jnrB;
 988             j_coord_offsetC  = DIM*jnrC;
 989             j_coord_offsetD  = DIM*jnrD;
 990
 991             /* load j atom coordinates */
 992             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 993                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 994                                                  &jx0,&jy0,&jz0);
 995
 996             /* Calculate displacement vector */
 997             dx00             = _mm256_sub_pd(ix0,jx0);
 998             dy00             = _mm256_sub_pd(iy0,jy0);
 999             dz00             = _mm256_sub_pd(iz0,jz0);
1000             dx10             = _mm256_sub_pd(ix1,jx0);
1001             dy10             = _mm256_sub_pd(iy1,jy0);
1002             dz10             = _mm256_sub_pd(iz1,jz0);
1003             dx20             = _mm256_sub_pd(ix2,jx0);
1004             dy20             = _mm256_sub_pd(iy2,jy0);
1005             dz20             = _mm256_sub_pd(iz2,jz0);
1006             dx30             = _mm256_sub_pd(ix3,jx0);
1007             dy30             = _mm256_sub_pd(iy3,jy0);
1008             dz30             = _mm256_sub_pd(iz3,jz0);
1009
1010             /* Calculate squared distance and things based on it */
1011             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1012             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1013             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1014             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1015
1016             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1017             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1018             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1019
1020             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1021             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1022             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1023             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1024
1025             /* Load parameters for j particles */
1026             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1027                                                                  charge+jnrC+0,charge+jnrD+0);
1028             vdwjidx0A        = 2*vdwtype[jnrA+0];
1029             vdwjidx0B        = 2*vdwtype[jnrB+0];
1030             vdwjidx0C        = 2*vdwtype[jnrC+0];
1031             vdwjidx0D        = 2*vdwtype[jnrD+0];
1032
1033             fjx0             = _mm256_setzero_pd();
1034             fjy0             = _mm256_setzero_pd();
1035             fjz0             = _mm256_setzero_pd();
1036
1037             /**************************
1038              * CALCULATE INTERACTIONS *
1039              **************************/
1040
1041             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1042             {
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1046                                             vdwioffsetptr0+vdwjidx0B,
1047                                             vdwioffsetptr0+vdwjidx0C,
1048                                             vdwioffsetptr0+vdwjidx0D,
1049                                             &c6_00,&c12_00);
1050
1051             /* LENNARD-JONES DISPERSION/REPULSION */
1052
1053             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1054             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1055
1056             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1057
1058             fscal            = fvdw;
1059
1060             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1061
1062             /* Calculate temporary vectorial force */
1063             tx               = _mm256_mul_pd(fscal,dx00);
1064             ty               = _mm256_mul_pd(fscal,dy00);
1065             tz               = _mm256_mul_pd(fscal,dz00);
1066
1067             /* Update vectorial force */
1068             fix0             = _mm256_add_pd(fix0,tx);
1069             fiy0             = _mm256_add_pd(fiy0,ty);
1070             fiz0             = _mm256_add_pd(fiz0,tz);
1071
1072             fjx0             = _mm256_add_pd(fjx0,tx);
1073             fjy0             = _mm256_add_pd(fjy0,ty);
1074             fjz0             = _mm256_add_pd(fjz0,tz);
1075
1076             }
1077
1078             /**************************
1079              * CALCULATE INTERACTIONS *
1080              **************************/
1081
1082             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1083             {
1084
1085             r10              = _mm256_mul_pd(rsq10,rinv10);
1086
1087             /* Compute parameters for interactions between i and j atoms */
1088             qq10             = _mm256_mul_pd(iq1,jq0);
1089
1090             /* EWALD ELECTROSTATICS */
1091
1092             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1093             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1094             ewitab           = _mm256_cvttpd_epi32(ewrt);
1095             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1096             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1097                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1098                                             &ewtabF,&ewtabFn);
1099             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1100             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1101
1102             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1103
1104             fscal            = felec;
1105
1106             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1107
1108             /* Calculate temporary vectorial force */
1109             tx               = _mm256_mul_pd(fscal,dx10);
1110             ty               = _mm256_mul_pd(fscal,dy10);
1111             tz               = _mm256_mul_pd(fscal,dz10);
1112
1113             /* Update vectorial force */
1114             fix1             = _mm256_add_pd(fix1,tx);
1115             fiy1             = _mm256_add_pd(fiy1,ty);
1116             fiz1             = _mm256_add_pd(fiz1,tz);
1117
1118             fjx0             = _mm256_add_pd(fjx0,tx);
1119             fjy0             = _mm256_add_pd(fjy0,ty);
1120             fjz0             = _mm256_add_pd(fjz0,tz);
1121
1122             }
1123
1124             /**************************
1125              * CALCULATE INTERACTIONS *
1126              **************************/
1127
1128             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1129             {
1130
1131             r20              = _mm256_mul_pd(rsq20,rinv20);
1132
1133             /* Compute parameters for interactions between i and j atoms */
1134             qq20             = _mm256_mul_pd(iq2,jq0);
1135
1136             /* EWALD ELECTROSTATICS */
1137
1138             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1139             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1140             ewitab           = _mm256_cvttpd_epi32(ewrt);
1141             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1142             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1143                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1144                                             &ewtabF,&ewtabFn);
1145             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1146             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1147
1148             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1149
1150             fscal            = felec;
1151
1152             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1153
1154             /* Calculate temporary vectorial force */
1155             tx               = _mm256_mul_pd(fscal,dx20);
1156             ty               = _mm256_mul_pd(fscal,dy20);
1157             tz               = _mm256_mul_pd(fscal,dz20);
1158
1159             /* Update vectorial force */
1160             fix2             = _mm256_add_pd(fix2,tx);
1161             fiy2             = _mm256_add_pd(fiy2,ty);
1162             fiz2             = _mm256_add_pd(fiz2,tz);
1163
1164             fjx0             = _mm256_add_pd(fjx0,tx);
1165             fjy0             = _mm256_add_pd(fjy0,ty);
1166             fjz0             = _mm256_add_pd(fjz0,tz);
1167
1168             }
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1175             {
1176
1177             r30              = _mm256_mul_pd(rsq30,rinv30);
1178
1179             /* Compute parameters for interactions between i and j atoms */
1180             qq30             = _mm256_mul_pd(iq3,jq0);
1181
1182             /* EWALD ELECTROSTATICS */
1183
1184             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1185             ewrt             = _mm256_mul_pd(r30,ewtabscale);
1186             ewitab           = _mm256_cvttpd_epi32(ewrt);
1187             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1188             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1189                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1190                                             &ewtabF,&ewtabFn);
1191             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1192             felec            = _mm256_mul_pd(_mm256_mul_pd(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1193
1194             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1195
1196             fscal            = felec;
1197
1198             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1199
1200             /* Calculate temporary vectorial force */
1201             tx               = _mm256_mul_pd(fscal,dx30);
1202             ty               = _mm256_mul_pd(fscal,dy30);
1203             tz               = _mm256_mul_pd(fscal,dz30);
1204
1205             /* Update vectorial force */
1206             fix3             = _mm256_add_pd(fix3,tx);
1207             fiy3             = _mm256_add_pd(fiy3,ty);
1208             fiz3             = _mm256_add_pd(fiz3,tz);
1209
1210             fjx0             = _mm256_add_pd(fjx0,tx);
1211             fjy0             = _mm256_add_pd(fjy0,ty);
1212             fjz0             = _mm256_add_pd(fjz0,tz);
1213
1214             }
1215
1216             fjptrA             = f+j_coord_offsetA;
1217             fjptrB             = f+j_coord_offsetB;
1218             fjptrC             = f+j_coord_offsetC;
1219             fjptrD             = f+j_coord_offsetD;
1220
1221             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1222
1223             /* Inner loop uses 150 flops */
1224         }
1225
1226         if(jidx<j_index_end)
1227         {
1228
1229             /* Get j neighbor index, and coordinate index */
1230             jnrlistA         = jjnr[jidx];
1231             jnrlistB         = jjnr[jidx+1];
1232             jnrlistC         = jjnr[jidx+2];
1233             jnrlistD         = jjnr[jidx+3];
1234             /* Sign of each element will be negative for non-real atoms.
1235              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1236              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1237              */
1238             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1239
1240             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1241             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1242             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1243
1244             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1245             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1246             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1247             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1248             j_coord_offsetA  = DIM*jnrA;
1249             j_coord_offsetB  = DIM*jnrB;
1250             j_coord_offsetC  = DIM*jnrC;
1251             j_coord_offsetD  = DIM*jnrD;
1252
1253             /* load j atom coordinates */
1254             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1255                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1256                                                  &jx0,&jy0,&jz0);
1257
1258             /* Calculate displacement vector */
1259             dx00             = _mm256_sub_pd(ix0,jx0);
1260             dy00             = _mm256_sub_pd(iy0,jy0);
1261             dz00             = _mm256_sub_pd(iz0,jz0);
1262             dx10             = _mm256_sub_pd(ix1,jx0);
1263             dy10             = _mm256_sub_pd(iy1,jy0);
1264             dz10             = _mm256_sub_pd(iz1,jz0);
1265             dx20             = _mm256_sub_pd(ix2,jx0);
1266             dy20             = _mm256_sub_pd(iy2,jy0);
1267             dz20             = _mm256_sub_pd(iz2,jz0);
1268             dx30             = _mm256_sub_pd(ix3,jx0);
1269             dy30             = _mm256_sub_pd(iy3,jy0);
1270             dz30             = _mm256_sub_pd(iz3,jz0);
1271
1272             /* Calculate squared distance and things based on it */
1273             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1274             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1275             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1276             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1277
1278             rinv10           = gmx_mm256_invsqrt_pd(rsq10);
1279             rinv20           = gmx_mm256_invsqrt_pd(rsq20);
1280             rinv30           = gmx_mm256_invsqrt_pd(rsq30);
1281
1282             rinvsq00         = gmx_mm256_inv_pd(rsq00);
1283             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1284             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1285             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1286
1287             /* Load parameters for j particles */
1288             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1289                                                                  charge+jnrC+0,charge+jnrD+0);
1290             vdwjidx0A        = 2*vdwtype[jnrA+0];
1291             vdwjidx0B        = 2*vdwtype[jnrB+0];
1292             vdwjidx0C        = 2*vdwtype[jnrC+0];
1293             vdwjidx0D        = 2*vdwtype[jnrD+0];
1294
1295             fjx0             = _mm256_setzero_pd();
1296             fjy0             = _mm256_setzero_pd();
1297             fjz0             = _mm256_setzero_pd();
1298
1299             /**************************
1300              * CALCULATE INTERACTIONS *
1301              **************************/
1302
1303             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1304             {
1305
1306             /* Compute parameters for interactions between i and j atoms */
1307             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1308                                             vdwioffsetptr0+vdwjidx0B,
1309                                             vdwioffsetptr0+vdwjidx0C,
1310                                             vdwioffsetptr0+vdwjidx0D,
1311                                             &c6_00,&c12_00);
1312
1313             /* LENNARD-JONES DISPERSION/REPULSION */
1314
1315             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1316             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1317
1318             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1319
1320             fscal            = fvdw;
1321
1322             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1323
1324             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1325
1326             /* Calculate temporary vectorial force */
1327             tx               = _mm256_mul_pd(fscal,dx00);
1328             ty               = _mm256_mul_pd(fscal,dy00);
1329             tz               = _mm256_mul_pd(fscal,dz00);
1330
1331             /* Update vectorial force */
1332             fix0             = _mm256_add_pd(fix0,tx);
1333             fiy0             = _mm256_add_pd(fiy0,ty);
1334             fiz0             = _mm256_add_pd(fiz0,tz);
1335
1336             fjx0             = _mm256_add_pd(fjx0,tx);
1337             fjy0             = _mm256_add_pd(fjy0,ty);
1338             fjz0             = _mm256_add_pd(fjz0,tz);
1339
1340             }
1341
1342             /**************************
1343              * CALCULATE INTERACTIONS *
1344              **************************/
1345
1346             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1347             {
1348
1349             r10              = _mm256_mul_pd(rsq10,rinv10);
1350             r10              = _mm256_andnot_pd(dummy_mask,r10);
1351
1352             /* Compute parameters for interactions between i and j atoms */
1353             qq10             = _mm256_mul_pd(iq1,jq0);
1354
1355             /* EWALD ELECTROSTATICS */
1356
1357             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1358             ewrt             = _mm256_mul_pd(r10,ewtabscale);
1359             ewitab           = _mm256_cvttpd_epi32(ewrt);
1360             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1361             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1362                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1363                                             &ewtabF,&ewtabFn);
1364             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1365             felec            = _mm256_mul_pd(_mm256_mul_pd(qq10,rinv10),_mm256_sub_pd(rinvsq10,felec));
1366
1367             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1368
1369             fscal            = felec;
1370
1371             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1372
1373             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1374
1375             /* Calculate temporary vectorial force */
1376             tx               = _mm256_mul_pd(fscal,dx10);
1377             ty               = _mm256_mul_pd(fscal,dy10);
1378             tz               = _mm256_mul_pd(fscal,dz10);
1379
1380             /* Update vectorial force */
1381             fix1             = _mm256_add_pd(fix1,tx);
1382             fiy1             = _mm256_add_pd(fiy1,ty);
1383             fiz1             = _mm256_add_pd(fiz1,tz);
1384
1385             fjx0             = _mm256_add_pd(fjx0,tx);
1386             fjy0             = _mm256_add_pd(fjy0,ty);
1387             fjz0             = _mm256_add_pd(fjz0,tz);
1388
1389             }
1390
1391             /**************************
1392              * CALCULATE INTERACTIONS *
1393              **************************/
1394
1395             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1396             {
1397
1398             r20              = _mm256_mul_pd(rsq20,rinv20);
1399             r20              = _mm256_andnot_pd(dummy_mask,r20);
1400
1401             /* Compute parameters for interactions between i and j atoms */
1402             qq20             = _mm256_mul_pd(iq2,jq0);
1403
1404             /* EWALD ELECTROSTATICS */
1405
1406             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1407             ewrt             = _mm256_mul_pd(r20,ewtabscale);
1408             ewitab           = _mm256_cvttpd_epi32(ewrt);
1409             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
1410             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
1411                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
1412                                             &ewtabF,&ewtabFn);
1413             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
1414             felec            = _mm256_mul_pd(_mm256_mul_pd(qq20,rinv20),_mm256_sub_pd(rinvsq20,felec));
1415
1416             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1417
1418             fscal            = felec;
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,dx20);
1426             ty               = _mm256_mul_pd(fscal,dy20);
1427             tz               = _mm256_mul_pd(fscal,dz20);
1428
1429             /* Update vectorial force */
1430             fix2             = _mm256_add_pd(fix2,tx);
1431             fiy2             = _mm256_add_pd(fiy2,ty);
1432             fiz2             = _mm256_add_pd(fiz2,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(rsq30,rcutoff2))
1445             {
1446
1447             r30              = _mm256_mul_pd(rsq30,rinv30);
1448             r30              = _mm256_andnot_pd(dummy_mask,r30);
1449
1450             /* Compute parameters for interactions between i and j atoms */
1451             qq30             = _mm256_mul_pd(iq3,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(r30,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(qq30,rinv30),_mm256_sub_pd(rinvsq30,felec));
1464
1465             cutoff_mask      = _mm256_cmp_pd(rsq30,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,dx30);
1475             ty               = _mm256_mul_pd(fscal,dy30);
1476             tz               = _mm256_mul_pd(fscal,dz30);
1477
1478             /* Update vectorial force */
1479             fix3             = _mm256_add_pd(fix3,tx);
1480             fiy3             = _mm256_add_pd(fiy3,ty);
1481             fiz3             = _mm256_add_pd(fiz3,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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1490             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1491             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1492             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1493
1494             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1495
1496             /* Inner loop uses 153 flops */
1497         }
1498
1499         /* End of innermost loop */
1500
1501         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1502                                                  f+i_coord_offset,fshift+i_shift_offset);
1503
1504         /* Increment number of inner iterations */
1505         inneriter                  += j_index_end - j_index_start;
1506
1507         /* Outer loop uses 24 flops */
1508     }
1509
1510     /* Increment number of outer iterations */
1511     outeriter        += nri;
1512
1513     /* Update outer/inner flops */
1514
1515     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*153);
1516 }