src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_avx_256_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_256_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_256_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     real *           vdwioffsetptr0;
  84     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     real *           vdwioffsetptr1;
  86     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     real *           vdwioffsetptr2;
  88     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     real *           vdwioffsetptr3;
  90     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  97     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     int              nvdwtype;
 100     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 101     int              *vdwtype;
 102     real             *vdwparam;
 103     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 104     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 105     __m256d          dummy_mask,cutoff_mask;
 106     __m128           tmpmask0,tmpmask1;
 107     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 108     __m256d          one     = _mm256_set1_pd(1.0);
 109     __m256d          two     = _mm256_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_pd(fr->ic->epsfac);
 122     charge           = mdatoms->chargeA;
 123     krf              = _mm256_set1_pd(fr->ic->k_rf);
 124     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 125     crf              = _mm256_set1_pd(fr->ic->c_rf);
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 133     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 134     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 135     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->ic->rcoulomb;
 139     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 140     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 141
 142     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 143     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm256_setzero_pd();
 179         fiy0             = _mm256_setzero_pd();
 180         fiz0             = _mm256_setzero_pd();
 181         fix1             = _mm256_setzero_pd();
 182         fiy1             = _mm256_setzero_pd();
 183         fiz1             = _mm256_setzero_pd();
 184         fix2             = _mm256_setzero_pd();
 185         fiy2             = _mm256_setzero_pd();
 186         fiz2             = _mm256_setzero_pd();
 187         fix3             = _mm256_setzero_pd();
 188         fiy3             = _mm256_setzero_pd();
 189         fiz3             = _mm256_setzero_pd();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm256_setzero_pd();
 193         vvdwsum          = _mm256_setzero_pd();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208
 209             /* load j atom coordinates */
 210             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 212                                                  &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm256_sub_pd(ix0,jx0);
 216             dy00             = _mm256_sub_pd(iy0,jy0);
 217             dz00             = _mm256_sub_pd(iz0,jz0);
 218             dx10             = _mm256_sub_pd(ix1,jx0);
 219             dy10             = _mm256_sub_pd(iy1,jy0);
 220             dz10             = _mm256_sub_pd(iz1,jz0);
 221             dx20             = _mm256_sub_pd(ix2,jx0);
 222             dy20             = _mm256_sub_pd(iy2,jy0);
 223             dz20             = _mm256_sub_pd(iz2,jz0);
 224             dx30             = _mm256_sub_pd(ix3,jx0);
 225             dy30             = _mm256_sub_pd(iy3,jy0);
 226             dz30             = _mm256_sub_pd(iz3,jz0);
 227
 228             /* Calculate squared distance and things based on it */
 229             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 230             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 231             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 232             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 233
 234             rinv10           = avx256_invsqrt_d(rsq10);
 235             rinv20           = avx256_invsqrt_d(rsq20);
 236             rinv30           = avx256_invsqrt_d(rsq30);
 237
 238             rinvsq00         = avx256_inv_d(rsq00);
 239             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 240             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 241             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 245                                                                  charge+jnrC+0,charge+jnrD+0);
 246             vdwjidx0A        = 2*vdwtype[jnrA+0];
 247             vdwjidx0B        = 2*vdwtype[jnrB+0];
 248             vdwjidx0C        = 2*vdwtype[jnrC+0];
 249             vdwjidx0D        = 2*vdwtype[jnrD+0];
 250
 251             fjx0             = _mm256_setzero_pd();
 252             fjy0             = _mm256_setzero_pd();
 253             fjz0             = _mm256_setzero_pd();
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 260             {
 261
 262             /* Compute parameters for interactions between i and j atoms */
 263             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 264                                             vdwioffsetptr0+vdwjidx0B,
 265                                             vdwioffsetptr0+vdwjidx0C,
 266                                             vdwioffsetptr0+vdwjidx0D,
 267                                             &c6_00,&c12_00);
 268
 269             /* LENNARD-JONES DISPERSION/REPULSION */
 270
 271             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 272             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 273             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 274             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) ,
 275                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 276             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 277
 278             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 282             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 283
 284             fscal            = fvdw;
 285
 286             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm256_mul_pd(fscal,dx00);
 290             ty               = _mm256_mul_pd(fscal,dy00);
 291             tz               = _mm256_mul_pd(fscal,dz00);
 292
 293             /* Update vectorial force */
 294             fix0             = _mm256_add_pd(fix0,tx);
 295             fiy0             = _mm256_add_pd(fiy0,ty);
 296             fiz0             = _mm256_add_pd(fiz0,tz);
 297
 298             fjx0             = _mm256_add_pd(fjx0,tx);
 299             fjy0             = _mm256_add_pd(fjy0,ty);
 300             fjz0             = _mm256_add_pd(fjz0,tz);
 301
 302             }
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 309             {
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq10             = _mm256_mul_pd(iq1,jq0);
 313
 314             /* REACTION-FIELD ELECTROSTATICS */
 315             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 316             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 317
 318             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velec            = _mm256_and_pd(velec,cutoff_mask);
 322             velecsum         = _mm256_add_pd(velecsum,velec);
 323
 324             fscal            = felec;
 325
 326             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 327
 328             /* Calculate temporary vectorial force */
 329             tx               = _mm256_mul_pd(fscal,dx10);
 330             ty               = _mm256_mul_pd(fscal,dy10);
 331             tz               = _mm256_mul_pd(fscal,dz10);
 332
 333             /* Update vectorial force */
 334             fix1             = _mm256_add_pd(fix1,tx);
 335             fiy1             = _mm256_add_pd(fiy1,ty);
 336             fiz1             = _mm256_add_pd(fiz1,tz);
 337
 338             fjx0             = _mm256_add_pd(fjx0,tx);
 339             fjy0             = _mm256_add_pd(fjy0,ty);
 340             fjz0             = _mm256_add_pd(fjz0,tz);
 341
 342             }
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 349             {
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq20             = _mm256_mul_pd(iq2,jq0);
 353
 354             /* REACTION-FIELD ELECTROSTATICS */
 355             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 356             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 357
 358             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velec            = _mm256_and_pd(velec,cutoff_mask);
 362             velecsum         = _mm256_add_pd(velecsum,velec);
 363
 364             fscal            = felec;
 365
 366             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 367
 368             /* Calculate temporary vectorial force */
 369             tx               = _mm256_mul_pd(fscal,dx20);
 370             ty               = _mm256_mul_pd(fscal,dy20);
 371             tz               = _mm256_mul_pd(fscal,dz20);
 372
 373             /* Update vectorial force */
 374             fix2             = _mm256_add_pd(fix2,tx);
 375             fiy2             = _mm256_add_pd(fiy2,ty);
 376             fiz2             = _mm256_add_pd(fiz2,tz);
 377
 378             fjx0             = _mm256_add_pd(fjx0,tx);
 379             fjy0             = _mm256_add_pd(fjy0,ty);
 380             fjz0             = _mm256_add_pd(fjz0,tz);
 381
 382             }
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 389             {
 390
 391             /* Compute parameters for interactions between i and j atoms */
 392             qq30             = _mm256_mul_pd(iq3,jq0);
 393
 394             /* REACTION-FIELD ELECTROSTATICS */
 395             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 396             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 397
 398             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velec            = _mm256_and_pd(velec,cutoff_mask);
 402             velecsum         = _mm256_add_pd(velecsum,velec);
 403
 404             fscal            = felec;
 405
 406             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 407
 408             /* Calculate temporary vectorial force */
 409             tx               = _mm256_mul_pd(fscal,dx30);
 410             ty               = _mm256_mul_pd(fscal,dy30);
 411             tz               = _mm256_mul_pd(fscal,dz30);
 412
 413             /* Update vectorial force */
 414             fix3             = _mm256_add_pd(fix3,tx);
 415             fiy3             = _mm256_add_pd(fiy3,ty);
 416             fiz3             = _mm256_add_pd(fiz3,tz);
 417
 418             fjx0             = _mm256_add_pd(fjx0,tx);
 419             fjy0             = _mm256_add_pd(fjy0,ty);
 420             fjz0             = _mm256_add_pd(fjz0,tz);
 421
 422             }
 423
 424             fjptrA             = f+j_coord_offsetA;
 425             fjptrB             = f+j_coord_offsetB;
 426             fjptrC             = f+j_coord_offsetC;
 427             fjptrD             = f+j_coord_offsetD;
 428
 429             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 430
 431             /* Inner loop uses 152 flops */
 432         }
 433
 434         if(jidx<j_index_end)
 435         {
 436
 437             /* Get j neighbor index, and coordinate index */
 438             jnrlistA         = jjnr[jidx];
 439             jnrlistB         = jjnr[jidx+1];
 440             jnrlistC         = jjnr[jidx+2];
 441             jnrlistD         = jjnr[jidx+3];
 442             /* Sign of each element will be negative for non-real atoms.
 443              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 444              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 445              */
 446             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 447
 448             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 449             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 450             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 451
 452             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 453             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 454             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 455             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 456             j_coord_offsetA  = DIM*jnrA;
 457             j_coord_offsetB  = DIM*jnrB;
 458             j_coord_offsetC  = DIM*jnrC;
 459             j_coord_offsetD  = DIM*jnrD;
 460
 461             /* load j atom coordinates */
 462             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 463                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 464                                                  &jx0,&jy0,&jz0);
 465
 466             /* Calculate displacement vector */
 467             dx00             = _mm256_sub_pd(ix0,jx0);
 468             dy00             = _mm256_sub_pd(iy0,jy0);
 469             dz00             = _mm256_sub_pd(iz0,jz0);
 470             dx10             = _mm256_sub_pd(ix1,jx0);
 471             dy10             = _mm256_sub_pd(iy1,jy0);
 472             dz10             = _mm256_sub_pd(iz1,jz0);
 473             dx20             = _mm256_sub_pd(ix2,jx0);
 474             dy20             = _mm256_sub_pd(iy2,jy0);
 475             dz20             = _mm256_sub_pd(iz2,jz0);
 476             dx30             = _mm256_sub_pd(ix3,jx0);
 477             dy30             = _mm256_sub_pd(iy3,jy0);
 478             dz30             = _mm256_sub_pd(iz3,jz0);
 479
 480             /* Calculate squared distance and things based on it */
 481             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 482             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 483             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 484             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 485
 486             rinv10           = avx256_invsqrt_d(rsq10);
 487             rinv20           = avx256_invsqrt_d(rsq20);
 488             rinv30           = avx256_invsqrt_d(rsq30);
 489
 490             rinvsq00         = avx256_inv_d(rsq00);
 491             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 492             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 493             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 494
 495             /* Load parameters for j particles */
 496             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 497                                                                  charge+jnrC+0,charge+jnrD+0);
 498             vdwjidx0A        = 2*vdwtype[jnrA+0];
 499             vdwjidx0B        = 2*vdwtype[jnrB+0];
 500             vdwjidx0C        = 2*vdwtype[jnrC+0];
 501             vdwjidx0D        = 2*vdwtype[jnrD+0];
 502
 503             fjx0             = _mm256_setzero_pd();
 504             fjy0             = _mm256_setzero_pd();
 505             fjz0             = _mm256_setzero_pd();
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 512             {
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 516                                             vdwioffsetptr0+vdwjidx0B,
 517                                             vdwioffsetptr0+vdwjidx0C,
 518                                             vdwioffsetptr0+vdwjidx0D,
 519                                             &c6_00,&c12_00);
 520
 521             /* LENNARD-JONES DISPERSION/REPULSION */
 522
 523             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 524             vvdw6            = _mm256_mul_pd(c6_00,rinvsix);
 525             vvdw12           = _mm256_mul_pd(c12_00,_mm256_mul_pd(rinvsix,rinvsix));
 526             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) ,
 527                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_00,sh_vdw_invrcut6)),one_sixth));
 528             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq00);
 529
 530             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 534             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 535             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 536
 537             fscal            = fvdw;
 538
 539             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 540
 541             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 542
 543             /* Calculate temporary vectorial force */
 544             tx               = _mm256_mul_pd(fscal,dx00);
 545             ty               = _mm256_mul_pd(fscal,dy00);
 546             tz               = _mm256_mul_pd(fscal,dz00);
 547
 548             /* Update vectorial force */
 549             fix0             = _mm256_add_pd(fix0,tx);
 550             fiy0             = _mm256_add_pd(fiy0,ty);
 551             fiz0             = _mm256_add_pd(fiz0,tz);
 552
 553             fjx0             = _mm256_add_pd(fjx0,tx);
 554             fjy0             = _mm256_add_pd(fjy0,ty);
 555             fjz0             = _mm256_add_pd(fjz0,tz);
 556
 557             }
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 564             {
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq10             = _mm256_mul_pd(iq1,jq0);
 568
 569             /* REACTION-FIELD ELECTROSTATICS */
 570             velec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_add_pd(rinv10,_mm256_mul_pd(krf,rsq10)),crf));
 571             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 572
 573             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 574
 575             /* Update potential sum for this i atom from the interaction with this j atom. */
 576             velec            = _mm256_and_pd(velec,cutoff_mask);
 577             velec            = _mm256_andnot_pd(dummy_mask,velec);
 578             velecsum         = _mm256_add_pd(velecsum,velec);
 579
 580             fscal            = felec;
 581
 582             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 583
 584             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 585
 586             /* Calculate temporary vectorial force */
 587             tx               = _mm256_mul_pd(fscal,dx10);
 588             ty               = _mm256_mul_pd(fscal,dy10);
 589             tz               = _mm256_mul_pd(fscal,dz10);
 590
 591             /* Update vectorial force */
 592             fix1             = _mm256_add_pd(fix1,tx);
 593             fiy1             = _mm256_add_pd(fiy1,ty);
 594             fiz1             = _mm256_add_pd(fiz1,tz);
 595
 596             fjx0             = _mm256_add_pd(fjx0,tx);
 597             fjy0             = _mm256_add_pd(fjy0,ty);
 598             fjz0             = _mm256_add_pd(fjz0,tz);
 599
 600             }
 601
 602             /**************************
 603              * CALCULATE INTERACTIONS *
 604              **************************/
 605
 606             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 607             {
 608
 609             /* Compute parameters for interactions between i and j atoms */
 610             qq20             = _mm256_mul_pd(iq2,jq0);
 611
 612             /* REACTION-FIELD ELECTROSTATICS */
 613             velec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_add_pd(rinv20,_mm256_mul_pd(krf,rsq20)),crf));
 614             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
 615
 616             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
 617
 618             /* Update potential sum for this i atom from the interaction with this j atom. */
 619             velec            = _mm256_and_pd(velec,cutoff_mask);
 620             velec            = _mm256_andnot_pd(dummy_mask,velec);
 621             velecsum         = _mm256_add_pd(velecsum,velec);
 622
 623             fscal            = felec;
 624
 625             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 626
 627             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 628
 629             /* Calculate temporary vectorial force */
 630             tx               = _mm256_mul_pd(fscal,dx20);
 631             ty               = _mm256_mul_pd(fscal,dy20);
 632             tz               = _mm256_mul_pd(fscal,dz20);
 633
 634             /* Update vectorial force */
 635             fix2             = _mm256_add_pd(fix2,tx);
 636             fiy2             = _mm256_add_pd(fiy2,ty);
 637             fiz2             = _mm256_add_pd(fiz2,tz);
 638
 639             fjx0             = _mm256_add_pd(fjx0,tx);
 640             fjy0             = _mm256_add_pd(fjy0,ty);
 641             fjz0             = _mm256_add_pd(fjz0,tz);
 642
 643             }
 644
 645             /**************************
 646              * CALCULATE INTERACTIONS *
 647              **************************/
 648
 649             if (gmx_mm256_any_lt(rsq30,rcutoff2))
 650             {
 651
 652             /* Compute parameters for interactions between i and j atoms */
 653             qq30             = _mm256_mul_pd(iq3,jq0);
 654
 655             /* REACTION-FIELD ELECTROSTATICS */
 656             velec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_add_pd(rinv30,_mm256_mul_pd(krf,rsq30)),crf));
 657             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
 658
 659             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
 660
 661             /* Update potential sum for this i atom from the interaction with this j atom. */
 662             velec            = _mm256_and_pd(velec,cutoff_mask);
 663             velec            = _mm256_andnot_pd(dummy_mask,velec);
 664             velecsum         = _mm256_add_pd(velecsum,velec);
 665
 666             fscal            = felec;
 667
 668             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 669
 670             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 671
 672             /* Calculate temporary vectorial force */
 673             tx               = _mm256_mul_pd(fscal,dx30);
 674             ty               = _mm256_mul_pd(fscal,dy30);
 675             tz               = _mm256_mul_pd(fscal,dz30);
 676
 677             /* Update vectorial force */
 678             fix3             = _mm256_add_pd(fix3,tx);
 679             fiy3             = _mm256_add_pd(fiy3,ty);
 680             fiz3             = _mm256_add_pd(fiz3,tz);
 681
 682             fjx0             = _mm256_add_pd(fjx0,tx);
 683             fjy0             = _mm256_add_pd(fjy0,ty);
 684             fjz0             = _mm256_add_pd(fjz0,tz);
 685
 686             }
 687
 688             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 689             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 690             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 691             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 692
 693             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 694
 695             /* Inner loop uses 152 flops */
 696         }
 697
 698         /* End of innermost loop */
 699
 700         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 701                                                  f+i_coord_offset,fshift+i_shift_offset);
 702
 703         ggid                        = gid[iidx];
 704         /* Update potential energies */
 705         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 706         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 707
 708         /* Increment number of inner iterations */
 709         inneriter                  += j_index_end - j_index_start;
 710
 711         /* Outer loop uses 26 flops */
 712     }
 713
 714     /* Increment number of outer iterations */
 715     outeriter        += nri;
 716
 717     /* Update outer/inner flops */
 718
 719     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*152);
 720 }
 721 /*
 722  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_double
 723  * Electrostatics interaction: ReactionField
 724  * VdW interaction:            LennardJones
 725  * Geometry:                   Water4-Particle
 726  * Calculate force/pot:        Force
 727  */
 728 void
 729 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_256_double
 730                     (t_nblist                    * gmx_restrict       nlist,
 731                      rvec                        * gmx_restrict          xx,
 732                      rvec                        * gmx_restrict          ff,
 733                      struct t_forcerec           * gmx_restrict          fr,
 734                      t_mdatoms                   * gmx_restrict     mdatoms,
 735                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 736                      t_nrnb                      * gmx_restrict        nrnb)
 737 {
 738     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 739      * just 0 for non-waters.
 740      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
 741      * jnr indices corresponding to data put in the four positions in the SIMD register.
 742      */
 743     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 744     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 745     int              jnrA,jnrB,jnrC,jnrD;
 746     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 747     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 748     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 749     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 750     real             rcutoff_scalar;
 751     real             *shiftvec,*fshift,*x,*f;
 752     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 753     real             scratch[4*DIM];
 754     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 755     real *           vdwioffsetptr0;
 756     __m256d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 757     real *           vdwioffsetptr1;
 758     __m256d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 759     real *           vdwioffsetptr2;
 760     __m256d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 761     real *           vdwioffsetptr3;
 762     __m256d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 763     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 764     __m256d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 765     __m256d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 766     __m256d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 767     __m256d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 768     __m256d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 769     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 770     real             *charge;
 771     int              nvdwtype;
 772     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 773     int              *vdwtype;
 774     real             *vdwparam;
 775     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 776     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 777     __m256d          dummy_mask,cutoff_mask;
 778     __m128           tmpmask0,tmpmask1;
 779     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 780     __m256d          one     = _mm256_set1_pd(1.0);
 781     __m256d          two     = _mm256_set1_pd(2.0);
 782     x                = xx[0];
 783     f                = ff[0];
 784
 785     nri              = nlist->nri;
 786     iinr             = nlist->iinr;
 787     jindex           = nlist->jindex;
 788     jjnr             = nlist->jjnr;
 789     shiftidx         = nlist->shift;
 790     gid              = nlist->gid;
 791     shiftvec         = fr->shift_vec[0];
 792     fshift           = fr->fshift[0];
 793     facel            = _mm256_set1_pd(fr->ic->epsfac);
 794     charge           = mdatoms->chargeA;
 795     krf              = _mm256_set1_pd(fr->ic->k_rf);
 796     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 797     crf              = _mm256_set1_pd(fr->ic->c_rf);
 798     nvdwtype         = fr->ntype;
 799     vdwparam         = fr->nbfp;
 800     vdwtype          = mdatoms->typeA;
 801
 802     /* Setup water-specific parameters */
 803     inr              = nlist->iinr[0];
 804     iq1              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+1]));
 805     iq2              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+2]));
 806     iq3              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+3]));
 807     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 808
 809     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 810     rcutoff_scalar   = fr->ic->rcoulomb;
 811     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 812     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 813
 814     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 815     rvdw             = _mm256_set1_pd(fr->ic->rvdw);
 816
 817     /* Avoid stupid compiler warnings */
 818     jnrA = jnrB = jnrC = jnrD = 0;
 819     j_coord_offsetA = 0;
 820     j_coord_offsetB = 0;
 821     j_coord_offsetC = 0;
 822     j_coord_offsetD = 0;
 823
 824     outeriter        = 0;
 825     inneriter        = 0;
 826
 827     for(iidx=0;iidx<4*DIM;iidx++)
 828     {
 829         scratch[iidx] = 0.0;
 830     }
 831
 832     /* Start outer loop over neighborlists */
 833     for(iidx=0; iidx<nri; iidx++)
 834     {
 835         /* Load shift vector for this list */
 836         i_shift_offset   = DIM*shiftidx[iidx];
 837
 838         /* Load limits for loop over neighbors */
 839         j_index_start    = jindex[iidx];
 840         j_index_end      = jindex[iidx+1];
 841
 842         /* Get outer coordinate index */
 843         inr              = iinr[iidx];
 844         i_coord_offset   = DIM*inr;
 845
 846         /* Load i particle coords and add shift vector */
 847         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 848                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 849
 850         fix0             = _mm256_setzero_pd();
 851         fiy0             = _mm256_setzero_pd();
 852         fiz0             = _mm256_setzero_pd();
 853         fix1             = _mm256_setzero_pd();
 854         fiy1             = _mm256_setzero_pd();
 855         fiz1             = _mm256_setzero_pd();
 856         fix2             = _mm256_setzero_pd();
 857         fiy2             = _mm256_setzero_pd();
 858         fiz2             = _mm256_setzero_pd();
 859         fix3             = _mm256_setzero_pd();
 860         fiy3             = _mm256_setzero_pd();
 861         fiz3             = _mm256_setzero_pd();
 862
 863         /* Start inner kernel loop */
 864         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 865         {
 866
 867             /* Get j neighbor index, and coordinate index */
 868             jnrA             = jjnr[jidx];
 869             jnrB             = jjnr[jidx+1];
 870             jnrC             = jjnr[jidx+2];
 871             jnrD             = jjnr[jidx+3];
 872             j_coord_offsetA  = DIM*jnrA;
 873             j_coord_offsetB  = DIM*jnrB;
 874             j_coord_offsetC  = DIM*jnrC;
 875             j_coord_offsetD  = DIM*jnrD;
 876
 877             /* load j atom coordinates */
 878             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 879                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 880                                                  &jx0,&jy0,&jz0);
 881
 882             /* Calculate displacement vector */
 883             dx00             = _mm256_sub_pd(ix0,jx0);
 884             dy00             = _mm256_sub_pd(iy0,jy0);
 885             dz00             = _mm256_sub_pd(iz0,jz0);
 886             dx10             = _mm256_sub_pd(ix1,jx0);
 887             dy10             = _mm256_sub_pd(iy1,jy0);
 888             dz10             = _mm256_sub_pd(iz1,jz0);
 889             dx20             = _mm256_sub_pd(ix2,jx0);
 890             dy20             = _mm256_sub_pd(iy2,jy0);
 891             dz20             = _mm256_sub_pd(iz2,jz0);
 892             dx30             = _mm256_sub_pd(ix3,jx0);
 893             dy30             = _mm256_sub_pd(iy3,jy0);
 894             dz30             = _mm256_sub_pd(iz3,jz0);
 895
 896             /* Calculate squared distance and things based on it */
 897             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
 898             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
 899             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
 900             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
 901
 902             rinv10           = avx256_invsqrt_d(rsq10);
 903             rinv20           = avx256_invsqrt_d(rsq20);
 904             rinv30           = avx256_invsqrt_d(rsq30);
 905
 906             rinvsq00         = avx256_inv_d(rsq00);
 907             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
 908             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
 909             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
 910
 911             /* Load parameters for j particles */
 912             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
 913                                                                  charge+jnrC+0,charge+jnrD+0);
 914             vdwjidx0A        = 2*vdwtype[jnrA+0];
 915             vdwjidx0B        = 2*vdwtype[jnrB+0];
 916             vdwjidx0C        = 2*vdwtype[jnrC+0];
 917             vdwjidx0D        = 2*vdwtype[jnrD+0];
 918
 919             fjx0             = _mm256_setzero_pd();
 920             fjy0             = _mm256_setzero_pd();
 921             fjz0             = _mm256_setzero_pd();
 922
 923             /**************************
 924              * CALCULATE INTERACTIONS *
 925              **************************/
 926
 927             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 928             {
 929
 930             /* Compute parameters for interactions between i and j atoms */
 931             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
 932                                             vdwioffsetptr0+vdwjidx0B,
 933                                             vdwioffsetptr0+vdwjidx0C,
 934                                             vdwioffsetptr0+vdwjidx0D,
 935                                             &c6_00,&c12_00);
 936
 937             /* LENNARD-JONES DISPERSION/REPULSION */
 938
 939             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 940             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
 941
 942             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
 943
 944             fscal            = fvdw;
 945
 946             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 947
 948             /* Calculate temporary vectorial force */
 949             tx               = _mm256_mul_pd(fscal,dx00);
 950             ty               = _mm256_mul_pd(fscal,dy00);
 951             tz               = _mm256_mul_pd(fscal,dz00);
 952
 953             /* Update vectorial force */
 954             fix0             = _mm256_add_pd(fix0,tx);
 955             fiy0             = _mm256_add_pd(fiy0,ty);
 956             fiz0             = _mm256_add_pd(fiz0,tz);
 957
 958             fjx0             = _mm256_add_pd(fjx0,tx);
 959             fjy0             = _mm256_add_pd(fjy0,ty);
 960             fjz0             = _mm256_add_pd(fjz0,tz);
 961
 962             }
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 969             {
 970
 971             /* Compute parameters for interactions between i and j atoms */
 972             qq10             = _mm256_mul_pd(iq1,jq0);
 973
 974             /* REACTION-FIELD ELECTROSTATICS */
 975             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
 976
 977             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
 978
 979             fscal            = felec;
 980
 981             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 982
 983             /* Calculate temporary vectorial force */
 984             tx               = _mm256_mul_pd(fscal,dx10);
 985             ty               = _mm256_mul_pd(fscal,dy10);
 986             tz               = _mm256_mul_pd(fscal,dz10);
 987
 988             /* Update vectorial force */
 989             fix1             = _mm256_add_pd(fix1,tx);
 990             fiy1             = _mm256_add_pd(fiy1,ty);
 991             fiz1             = _mm256_add_pd(fiz1,tz);
 992
 993             fjx0             = _mm256_add_pd(fjx0,tx);
 994             fjy0             = _mm256_add_pd(fjy0,ty);
 995             fjz0             = _mm256_add_pd(fjz0,tz);
 996
 997             }
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1004             {
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             qq20             = _mm256_mul_pd(iq2,jq0);
1008
1009             /* REACTION-FIELD ELECTROSTATICS */
1010             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1011
1012             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1013
1014             fscal            = felec;
1015
1016             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1017
1018             /* Calculate temporary vectorial force */
1019             tx               = _mm256_mul_pd(fscal,dx20);
1020             ty               = _mm256_mul_pd(fscal,dy20);
1021             tz               = _mm256_mul_pd(fscal,dz20);
1022
1023             /* Update vectorial force */
1024             fix2             = _mm256_add_pd(fix2,tx);
1025             fiy2             = _mm256_add_pd(fiy2,ty);
1026             fiz2             = _mm256_add_pd(fiz2,tz);
1027
1028             fjx0             = _mm256_add_pd(fjx0,tx);
1029             fjy0             = _mm256_add_pd(fjy0,ty);
1030             fjz0             = _mm256_add_pd(fjz0,tz);
1031
1032             }
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1039             {
1040
1041             /* Compute parameters for interactions between i and j atoms */
1042             qq30             = _mm256_mul_pd(iq3,jq0);
1043
1044             /* REACTION-FIELD ELECTROSTATICS */
1045             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1046
1047             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1048
1049             fscal            = felec;
1050
1051             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm256_mul_pd(fscal,dx30);
1055             ty               = _mm256_mul_pd(fscal,dy30);
1056             tz               = _mm256_mul_pd(fscal,dz30);
1057
1058             /* Update vectorial force */
1059             fix3             = _mm256_add_pd(fix3,tx);
1060             fiy3             = _mm256_add_pd(fiy3,ty);
1061             fiz3             = _mm256_add_pd(fiz3,tz);
1062
1063             fjx0             = _mm256_add_pd(fjx0,tx);
1064             fjy0             = _mm256_add_pd(fjy0,ty);
1065             fjz0             = _mm256_add_pd(fjz0,tz);
1066
1067             }
1068
1069             fjptrA             = f+j_coord_offsetA;
1070             fjptrB             = f+j_coord_offsetB;
1071             fjptrC             = f+j_coord_offsetC;
1072             fjptrD             = f+j_coord_offsetD;
1073
1074             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1075
1076             /* Inner loop uses 123 flops */
1077         }
1078
1079         if(jidx<j_index_end)
1080         {
1081
1082             /* Get j neighbor index, and coordinate index */
1083             jnrlistA         = jjnr[jidx];
1084             jnrlistB         = jjnr[jidx+1];
1085             jnrlistC         = jjnr[jidx+2];
1086             jnrlistD         = jjnr[jidx+3];
1087             /* Sign of each element will be negative for non-real atoms.
1088              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1089              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
1090              */
1091             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1092
1093             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
1094             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
1095             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
1096
1097             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1098             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1099             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1100             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1101             j_coord_offsetA  = DIM*jnrA;
1102             j_coord_offsetB  = DIM*jnrB;
1103             j_coord_offsetC  = DIM*jnrC;
1104             j_coord_offsetD  = DIM*jnrD;
1105
1106             /* load j atom coordinates */
1107             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1108                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1109                                                  &jx0,&jy0,&jz0);
1110
1111             /* Calculate displacement vector */
1112             dx00             = _mm256_sub_pd(ix0,jx0);
1113             dy00             = _mm256_sub_pd(iy0,jy0);
1114             dz00             = _mm256_sub_pd(iz0,jz0);
1115             dx10             = _mm256_sub_pd(ix1,jx0);
1116             dy10             = _mm256_sub_pd(iy1,jy0);
1117             dz10             = _mm256_sub_pd(iz1,jz0);
1118             dx20             = _mm256_sub_pd(ix2,jx0);
1119             dy20             = _mm256_sub_pd(iy2,jy0);
1120             dz20             = _mm256_sub_pd(iz2,jz0);
1121             dx30             = _mm256_sub_pd(ix3,jx0);
1122             dy30             = _mm256_sub_pd(iy3,jy0);
1123             dz30             = _mm256_sub_pd(iz3,jz0);
1124
1125             /* Calculate squared distance and things based on it */
1126             rsq00            = gmx_mm256_calc_rsq_pd(dx00,dy00,dz00);
1127             rsq10            = gmx_mm256_calc_rsq_pd(dx10,dy10,dz10);
1128             rsq20            = gmx_mm256_calc_rsq_pd(dx20,dy20,dz20);
1129             rsq30            = gmx_mm256_calc_rsq_pd(dx30,dy30,dz30);
1130
1131             rinv10           = avx256_invsqrt_d(rsq10);
1132             rinv20           = avx256_invsqrt_d(rsq20);
1133             rinv30           = avx256_invsqrt_d(rsq30);
1134
1135             rinvsq00         = avx256_inv_d(rsq00);
1136             rinvsq10         = _mm256_mul_pd(rinv10,rinv10);
1137             rinvsq20         = _mm256_mul_pd(rinv20,rinv20);
1138             rinvsq30         = _mm256_mul_pd(rinv30,rinv30);
1139
1140             /* Load parameters for j particles */
1141             jq0              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+0,charge+jnrB+0,
1142                                                                  charge+jnrC+0,charge+jnrD+0);
1143             vdwjidx0A        = 2*vdwtype[jnrA+0];
1144             vdwjidx0B        = 2*vdwtype[jnrB+0];
1145             vdwjidx0C        = 2*vdwtype[jnrC+0];
1146             vdwjidx0D        = 2*vdwtype[jnrD+0];
1147
1148             fjx0             = _mm256_setzero_pd();
1149             fjy0             = _mm256_setzero_pd();
1150             fjz0             = _mm256_setzero_pd();
1151
1152             /**************************
1153              * CALCULATE INTERACTIONS *
1154              **************************/
1155
1156             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1157             {
1158
1159             /* Compute parameters for interactions between i and j atoms */
1160             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr0+vdwjidx0A,
1161                                             vdwioffsetptr0+vdwjidx0B,
1162                                             vdwioffsetptr0+vdwjidx0C,
1163                                             vdwioffsetptr0+vdwjidx0D,
1164                                             &c6_00,&c12_00);
1165
1166             /* LENNARD-JONES DISPERSION/REPULSION */
1167
1168             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1169             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_00,rinvsix),c6_00),_mm256_mul_pd(rinvsix,rinvsq00));
1170
1171             cutoff_mask      = _mm256_cmp_pd(rsq00,rcutoff2,_CMP_LT_OQ);
1172
1173             fscal            = fvdw;
1174
1175             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1176
1177             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1178
1179             /* Calculate temporary vectorial force */
1180             tx               = _mm256_mul_pd(fscal,dx00);
1181             ty               = _mm256_mul_pd(fscal,dy00);
1182             tz               = _mm256_mul_pd(fscal,dz00);
1183
1184             /* Update vectorial force */
1185             fix0             = _mm256_add_pd(fix0,tx);
1186             fiy0             = _mm256_add_pd(fiy0,ty);
1187             fiz0             = _mm256_add_pd(fiz0,tz);
1188
1189             fjx0             = _mm256_add_pd(fjx0,tx);
1190             fjy0             = _mm256_add_pd(fjy0,ty);
1191             fjz0             = _mm256_add_pd(fjz0,tz);
1192
1193             }
1194
1195             /**************************
1196              * CALCULATE INTERACTIONS *
1197              **************************/
1198
1199             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1200             {
1201
1202             /* Compute parameters for interactions between i and j atoms */
1203             qq10             = _mm256_mul_pd(iq1,jq0);
1204
1205             /* REACTION-FIELD ELECTROSTATICS */
1206             felec            = _mm256_mul_pd(qq10,_mm256_sub_pd(_mm256_mul_pd(rinv10,rinvsq10),krf2));
1207
1208             cutoff_mask      = _mm256_cmp_pd(rsq10,rcutoff2,_CMP_LT_OQ);
1209
1210             fscal            = felec;
1211
1212             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1213
1214             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1215
1216             /* Calculate temporary vectorial force */
1217             tx               = _mm256_mul_pd(fscal,dx10);
1218             ty               = _mm256_mul_pd(fscal,dy10);
1219             tz               = _mm256_mul_pd(fscal,dz10);
1220
1221             /* Update vectorial force */
1222             fix1             = _mm256_add_pd(fix1,tx);
1223             fiy1             = _mm256_add_pd(fiy1,ty);
1224             fiz1             = _mm256_add_pd(fiz1,tz);
1225
1226             fjx0             = _mm256_add_pd(fjx0,tx);
1227             fjy0             = _mm256_add_pd(fjy0,ty);
1228             fjz0             = _mm256_add_pd(fjz0,tz);
1229
1230             }
1231
1232             /**************************
1233              * CALCULATE INTERACTIONS *
1234              **************************/
1235
1236             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1237             {
1238
1239             /* Compute parameters for interactions between i and j atoms */
1240             qq20             = _mm256_mul_pd(iq2,jq0);
1241
1242             /* REACTION-FIELD ELECTROSTATICS */
1243             felec            = _mm256_mul_pd(qq20,_mm256_sub_pd(_mm256_mul_pd(rinv20,rinvsq20),krf2));
1244
1245             cutoff_mask      = _mm256_cmp_pd(rsq20,rcutoff2,_CMP_LT_OQ);
1246
1247             fscal            = felec;
1248
1249             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1250
1251             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1252
1253             /* Calculate temporary vectorial force */
1254             tx               = _mm256_mul_pd(fscal,dx20);
1255             ty               = _mm256_mul_pd(fscal,dy20);
1256             tz               = _mm256_mul_pd(fscal,dz20);
1257
1258             /* Update vectorial force */
1259             fix2             = _mm256_add_pd(fix2,tx);
1260             fiy2             = _mm256_add_pd(fiy2,ty);
1261             fiz2             = _mm256_add_pd(fiz2,tz);
1262
1263             fjx0             = _mm256_add_pd(fjx0,tx);
1264             fjy0             = _mm256_add_pd(fjy0,ty);
1265             fjz0             = _mm256_add_pd(fjz0,tz);
1266
1267             }
1268
1269             /**************************
1270              * CALCULATE INTERACTIONS *
1271              **************************/
1272
1273             if (gmx_mm256_any_lt(rsq30,rcutoff2))
1274             {
1275
1276             /* Compute parameters for interactions between i and j atoms */
1277             qq30             = _mm256_mul_pd(iq3,jq0);
1278
1279             /* REACTION-FIELD ELECTROSTATICS */
1280             felec            = _mm256_mul_pd(qq30,_mm256_sub_pd(_mm256_mul_pd(rinv30,rinvsq30),krf2));
1281
1282             cutoff_mask      = _mm256_cmp_pd(rsq30,rcutoff2,_CMP_LT_OQ);
1283
1284             fscal            = felec;
1285
1286             fscal            = _mm256_and_pd(fscal,cutoff_mask);
1287
1288             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
1289
1290             /* Calculate temporary vectorial force */
1291             tx               = _mm256_mul_pd(fscal,dx30);
1292             ty               = _mm256_mul_pd(fscal,dy30);
1293             tz               = _mm256_mul_pd(fscal,dz30);
1294
1295             /* Update vectorial force */
1296             fix3             = _mm256_add_pd(fix3,tx);
1297             fiy3             = _mm256_add_pd(fiy3,ty);
1298             fiz3             = _mm256_add_pd(fiz3,tz);
1299
1300             fjx0             = _mm256_add_pd(fjx0,tx);
1301             fjy0             = _mm256_add_pd(fjy0,ty);
1302             fjz0             = _mm256_add_pd(fjz0,tz);
1303
1304             }
1305
1306             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1307             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1308             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1309             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1310
1311             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1312
1313             /* Inner loop uses 123 flops */
1314         }
1315
1316         /* End of innermost loop */
1317
1318         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1319                                                  f+i_coord_offset,fshift+i_shift_offset);
1320
1321         /* Increment number of inner iterations */
1322         inneriter                  += j_index_end - j_index_start;
1323
1324         /* Outer loop uses 24 flops */
1325     }
1326
1327     /* Increment number of outer iterations */
1328     outeriter        += nri;
1329
1330     /* Update outer/inner flops */
1331
1332     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*123);
1333 }