src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     real *           vdwioffsetptr1;
  91     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  92     real *           vdwioffsetptr2;
  93     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  94     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  95     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  96     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  97     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  98     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  99     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 106     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 107     __m256i          ewitab;
 108     __m128i          ewitab_lo,ewitab_hi;
 109     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 111     real             *ewtab;
 112     __m256           dummy_mask,cutoff_mask;
 113     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 114     __m256           one     = _mm256_set1_ps(1.0);
 115     __m256           two     = _mm256_set1_ps(2.0);
 116     x                = xx[0];
 117     f                = ff[0];
 118
 119     nri              = nlist->nri;
 120     iinr             = nlist->iinr;
 121     jindex           = nlist->jindex;
 122     jjnr             = nlist->jjnr;
 123     shiftidx         = nlist->shift;
 124     gid              = nlist->gid;
 125     shiftvec         = fr->shift_vec[0];
 126     fshift           = fr->fshift[0];
 127     facel            = _mm256_set1_ps(fr->epsfac);
 128     charge           = mdatoms->chargeA;
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 134     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 135     beta2            = _mm256_mul_ps(beta,beta);
 136     beta3            = _mm256_mul_ps(beta,beta2);
 137
 138     ewtab            = fr->ic->tabq_coul_FDV0;
 139     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 140     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 141
 142     /* Setup water-specific parameters */
 143     inr              = nlist->iinr[0];
 144     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 145     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 146     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 147     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 148
 149     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 150     rcutoff_scalar   = fr->rcoulomb;
 151     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 152     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 153
 154     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 155     rvdw             = _mm256_set1_ps(fr->rvdw);
 156
 157     /* Avoid stupid compiler warnings */
 158     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 159     j_coord_offsetA = 0;
 160     j_coord_offsetB = 0;
 161     j_coord_offsetC = 0;
 162     j_coord_offsetD = 0;
 163     j_coord_offsetE = 0;
 164     j_coord_offsetF = 0;
 165     j_coord_offsetG = 0;
 166     j_coord_offsetH = 0;
 167
 168     outeriter        = 0;
 169     inneriter        = 0;
 170
 171     for(iidx=0;iidx<4*DIM;iidx++)
 172     {
 173         scratch[iidx] = 0.0;
 174     }
 175
 176     /* Start outer loop over neighborlists */
 177     for(iidx=0; iidx<nri; iidx++)
 178     {
 179         /* Load shift vector for this list */
 180         i_shift_offset   = DIM*shiftidx[iidx];
 181
 182         /* Load limits for loop over neighbors */
 183         j_index_start    = jindex[iidx];
 184         j_index_end      = jindex[iidx+1];
 185
 186         /* Get outer coordinate index */
 187         inr              = iinr[iidx];
 188         i_coord_offset   = DIM*inr;
 189
 190         /* Load i particle coords and add shift vector */
 191         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 192                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 193
 194         fix0             = _mm256_setzero_ps();
 195         fiy0             = _mm256_setzero_ps();
 196         fiz0             = _mm256_setzero_ps();
 197         fix1             = _mm256_setzero_ps();
 198         fiy1             = _mm256_setzero_ps();
 199         fiz1             = _mm256_setzero_ps();
 200         fix2             = _mm256_setzero_ps();
 201         fiy2             = _mm256_setzero_ps();
 202         fiz2             = _mm256_setzero_ps();
 203
 204         /* Reset potential sums */
 205         velecsum         = _mm256_setzero_ps();
 206         vvdwsum          = _mm256_setzero_ps();
 207
 208         /* Start inner kernel loop */
 209         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 210         {
 211
 212             /* Get j neighbor index, and coordinate index */
 213             jnrA             = jjnr[jidx];
 214             jnrB             = jjnr[jidx+1];
 215             jnrC             = jjnr[jidx+2];
 216             jnrD             = jjnr[jidx+3];
 217             jnrE             = jjnr[jidx+4];
 218             jnrF             = jjnr[jidx+5];
 219             jnrG             = jjnr[jidx+6];
 220             jnrH             = jjnr[jidx+7];
 221             j_coord_offsetA  = DIM*jnrA;
 222             j_coord_offsetB  = DIM*jnrB;
 223             j_coord_offsetC  = DIM*jnrC;
 224             j_coord_offsetD  = DIM*jnrD;
 225             j_coord_offsetE  = DIM*jnrE;
 226             j_coord_offsetF  = DIM*jnrF;
 227             j_coord_offsetG  = DIM*jnrG;
 228             j_coord_offsetH  = DIM*jnrH;
 229
 230             /* load j atom coordinates */
 231             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 232                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 233                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 234                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 235                                                  &jx0,&jy0,&jz0);
 236
 237             /* Calculate displacement vector */
 238             dx00             = _mm256_sub_ps(ix0,jx0);
 239             dy00             = _mm256_sub_ps(iy0,jy0);
 240             dz00             = _mm256_sub_ps(iz0,jz0);
 241             dx10             = _mm256_sub_ps(ix1,jx0);
 242             dy10             = _mm256_sub_ps(iy1,jy0);
 243             dz10             = _mm256_sub_ps(iz1,jz0);
 244             dx20             = _mm256_sub_ps(ix2,jx0);
 245             dy20             = _mm256_sub_ps(iy2,jy0);
 246             dz20             = _mm256_sub_ps(iz2,jz0);
 247
 248             /* Calculate squared distance and things based on it */
 249             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 250             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 251             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 252
 253             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 254             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 255             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 256
 257             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 258             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 259             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 260
 261             /* Load parameters for j particles */
 262             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 263                                                                  charge+jnrC+0,charge+jnrD+0,
 264                                                                  charge+jnrE+0,charge+jnrF+0,
 265                                                                  charge+jnrG+0,charge+jnrH+0);
 266             vdwjidx0A        = 2*vdwtype[jnrA+0];
 267             vdwjidx0B        = 2*vdwtype[jnrB+0];
 268             vdwjidx0C        = 2*vdwtype[jnrC+0];
 269             vdwjidx0D        = 2*vdwtype[jnrD+0];
 270             vdwjidx0E        = 2*vdwtype[jnrE+0];
 271             vdwjidx0F        = 2*vdwtype[jnrF+0];
 272             vdwjidx0G        = 2*vdwtype[jnrG+0];
 273             vdwjidx0H        = 2*vdwtype[jnrH+0];
 274
 275             fjx0             = _mm256_setzero_ps();
 276             fjy0             = _mm256_setzero_ps();
 277             fjz0             = _mm256_setzero_ps();
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 284             {
 285
 286             r00              = _mm256_mul_ps(rsq00,rinv00);
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             qq00             = _mm256_mul_ps(iq0,jq0);
 290             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 291                                             vdwioffsetptr0+vdwjidx0B,
 292                                             vdwioffsetptr0+vdwjidx0C,
 293                                             vdwioffsetptr0+vdwjidx0D,
 294                                             vdwioffsetptr0+vdwjidx0E,
 295                                             vdwioffsetptr0+vdwjidx0F,
 296                                             vdwioffsetptr0+vdwjidx0G,
 297                                             vdwioffsetptr0+vdwjidx0H,
 298                                             &c6_00,&c12_00);
 299
 300             /* EWALD ELECTROSTATICS */
 301
 302             /* Analytical PME correction */
 303             zeta2            = _mm256_mul_ps(beta2,rsq00);
 304             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 305             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 306             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 307             felec            = _mm256_mul_ps(qq00,felec);
 308             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 309             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 310             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 311             velec            = _mm256_mul_ps(qq00,velec);
 312
 313             /* LENNARD-JONES DISPERSION/REPULSION */
 314
 315             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 316             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 317             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 318             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 319                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 320             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 321
 322             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 323
 324             /* Update potential sum for this i atom from the interaction with this j atom. */
 325             velec            = _mm256_and_ps(velec,cutoff_mask);
 326             velecsum         = _mm256_add_ps(velecsum,velec);
 327             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 328             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 329
 330             fscal            = _mm256_add_ps(felec,fvdw);
 331
 332             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 333
 334             /* Calculate temporary vectorial force */
 335             tx               = _mm256_mul_ps(fscal,dx00);
 336             ty               = _mm256_mul_ps(fscal,dy00);
 337             tz               = _mm256_mul_ps(fscal,dz00);
 338
 339             /* Update vectorial force */
 340             fix0             = _mm256_add_ps(fix0,tx);
 341             fiy0             = _mm256_add_ps(fiy0,ty);
 342             fiz0             = _mm256_add_ps(fiz0,tz);
 343
 344             fjx0             = _mm256_add_ps(fjx0,tx);
 345             fjy0             = _mm256_add_ps(fjy0,ty);
 346             fjz0             = _mm256_add_ps(fjz0,tz);
 347
 348             }
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 355             {
 356
 357             r10              = _mm256_mul_ps(rsq10,rinv10);
 358
 359             /* Compute parameters for interactions between i and j atoms */
 360             qq10             = _mm256_mul_ps(iq1,jq0);
 361
 362             /* EWALD ELECTROSTATICS */
 363
 364             /* Analytical PME correction */
 365             zeta2            = _mm256_mul_ps(beta2,rsq10);
 366             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 367             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 368             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 369             felec            = _mm256_mul_ps(qq10,felec);
 370             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 371             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 372             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 373             velec            = _mm256_mul_ps(qq10,velec);
 374
 375             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velec            = _mm256_and_ps(velec,cutoff_mask);
 379             velecsum         = _mm256_add_ps(velecsum,velec);
 380
 381             fscal            = felec;
 382
 383             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 384
 385             /* Calculate temporary vectorial force */
 386             tx               = _mm256_mul_ps(fscal,dx10);
 387             ty               = _mm256_mul_ps(fscal,dy10);
 388             tz               = _mm256_mul_ps(fscal,dz10);
 389
 390             /* Update vectorial force */
 391             fix1             = _mm256_add_ps(fix1,tx);
 392             fiy1             = _mm256_add_ps(fiy1,ty);
 393             fiz1             = _mm256_add_ps(fiz1,tz);
 394
 395             fjx0             = _mm256_add_ps(fjx0,tx);
 396             fjy0             = _mm256_add_ps(fjy0,ty);
 397             fjz0             = _mm256_add_ps(fjz0,tz);
 398
 399             }
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 406             {
 407
 408             r20              = _mm256_mul_ps(rsq20,rinv20);
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq20             = _mm256_mul_ps(iq2,jq0);
 412
 413             /* EWALD ELECTROSTATICS */
 414
 415             /* Analytical PME correction */
 416             zeta2            = _mm256_mul_ps(beta2,rsq20);
 417             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 418             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 419             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 420             felec            = _mm256_mul_ps(qq20,felec);
 421             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 422             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 423             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 424             velec            = _mm256_mul_ps(qq20,velec);
 425
 426             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velec            = _mm256_and_ps(velec,cutoff_mask);
 430             velecsum         = _mm256_add_ps(velecsum,velec);
 431
 432             fscal            = felec;
 433
 434             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 435
 436             /* Calculate temporary vectorial force */
 437             tx               = _mm256_mul_ps(fscal,dx20);
 438             ty               = _mm256_mul_ps(fscal,dy20);
 439             tz               = _mm256_mul_ps(fscal,dz20);
 440
 441             /* Update vectorial force */
 442             fix2             = _mm256_add_ps(fix2,tx);
 443             fiy2             = _mm256_add_ps(fiy2,ty);
 444             fiz2             = _mm256_add_ps(fiz2,tz);
 445
 446             fjx0             = _mm256_add_ps(fjx0,tx);
 447             fjy0             = _mm256_add_ps(fjy0,ty);
 448             fjz0             = _mm256_add_ps(fjz0,tz);
 449
 450             }
 451
 452             fjptrA             = f+j_coord_offsetA;
 453             fjptrB             = f+j_coord_offsetB;
 454             fjptrC             = f+j_coord_offsetC;
 455             fjptrD             = f+j_coord_offsetD;
 456             fjptrE             = f+j_coord_offsetE;
 457             fjptrF             = f+j_coord_offsetF;
 458             fjptrG             = f+j_coord_offsetG;
 459             fjptrH             = f+j_coord_offsetH;
 460
 461             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 462
 463             /* Inner loop uses 348 flops */
 464         }
 465
 466         if(jidx<j_index_end)
 467         {
 468
 469             /* Get j neighbor index, and coordinate index */
 470             jnrlistA         = jjnr[jidx];
 471             jnrlistB         = jjnr[jidx+1];
 472             jnrlistC         = jjnr[jidx+2];
 473             jnrlistD         = jjnr[jidx+3];
 474             jnrlistE         = jjnr[jidx+4];
 475             jnrlistF         = jjnr[jidx+5];
 476             jnrlistG         = jjnr[jidx+6];
 477             jnrlistH         = jjnr[jidx+7];
 478             /* Sign of each element will be negative for non-real atoms.
 479              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 480              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 481              */
 482             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 483                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 484
 485             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 486             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 487             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 488             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 489             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 490             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 491             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 492             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 493             j_coord_offsetA  = DIM*jnrA;
 494             j_coord_offsetB  = DIM*jnrB;
 495             j_coord_offsetC  = DIM*jnrC;
 496             j_coord_offsetD  = DIM*jnrD;
 497             j_coord_offsetE  = DIM*jnrE;
 498             j_coord_offsetF  = DIM*jnrF;
 499             j_coord_offsetG  = DIM*jnrG;
 500             j_coord_offsetH  = DIM*jnrH;
 501
 502             /* load j atom coordinates */
 503             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 504                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 505                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 506                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 507                                                  &jx0,&jy0,&jz0);
 508
 509             /* Calculate displacement vector */
 510             dx00             = _mm256_sub_ps(ix0,jx0);
 511             dy00             = _mm256_sub_ps(iy0,jy0);
 512             dz00             = _mm256_sub_ps(iz0,jz0);
 513             dx10             = _mm256_sub_ps(ix1,jx0);
 514             dy10             = _mm256_sub_ps(iy1,jy0);
 515             dz10             = _mm256_sub_ps(iz1,jz0);
 516             dx20             = _mm256_sub_ps(ix2,jx0);
 517             dy20             = _mm256_sub_ps(iy2,jy0);
 518             dz20             = _mm256_sub_ps(iz2,jz0);
 519
 520             /* Calculate squared distance and things based on it */
 521             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 522             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 523             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 524
 525             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 526             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 527             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 528
 529             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 530             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 531             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 532
 533             /* Load parameters for j particles */
 534             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 535                                                                  charge+jnrC+0,charge+jnrD+0,
 536                                                                  charge+jnrE+0,charge+jnrF+0,
 537                                                                  charge+jnrG+0,charge+jnrH+0);
 538             vdwjidx0A        = 2*vdwtype[jnrA+0];
 539             vdwjidx0B        = 2*vdwtype[jnrB+0];
 540             vdwjidx0C        = 2*vdwtype[jnrC+0];
 541             vdwjidx0D        = 2*vdwtype[jnrD+0];
 542             vdwjidx0E        = 2*vdwtype[jnrE+0];
 543             vdwjidx0F        = 2*vdwtype[jnrF+0];
 544             vdwjidx0G        = 2*vdwtype[jnrG+0];
 545             vdwjidx0H        = 2*vdwtype[jnrH+0];
 546
 547             fjx0             = _mm256_setzero_ps();
 548             fjy0             = _mm256_setzero_ps();
 549             fjz0             = _mm256_setzero_ps();
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 556             {
 557
 558             r00              = _mm256_mul_ps(rsq00,rinv00);
 559             r00              = _mm256_andnot_ps(dummy_mask,r00);
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq00             = _mm256_mul_ps(iq0,jq0);
 563             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 564                                             vdwioffsetptr0+vdwjidx0B,
 565                                             vdwioffsetptr0+vdwjidx0C,
 566                                             vdwioffsetptr0+vdwjidx0D,
 567                                             vdwioffsetptr0+vdwjidx0E,
 568                                             vdwioffsetptr0+vdwjidx0F,
 569                                             vdwioffsetptr0+vdwjidx0G,
 570                                             vdwioffsetptr0+vdwjidx0H,
 571                                             &c6_00,&c12_00);
 572
 573             /* EWALD ELECTROSTATICS */
 574
 575             /* Analytical PME correction */
 576             zeta2            = _mm256_mul_ps(beta2,rsq00);
 577             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 578             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 579             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 580             felec            = _mm256_mul_ps(qq00,felec);
 581             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 582             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 583             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 584             velec            = _mm256_mul_ps(qq00,velec);
 585
 586             /* LENNARD-JONES DISPERSION/REPULSION */
 587
 588             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 589             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 590             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 591             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_00,_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 592                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 593             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 594
 595             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 596
 597             /* Update potential sum for this i atom from the interaction with this j atom. */
 598             velec            = _mm256_and_ps(velec,cutoff_mask);
 599             velec            = _mm256_andnot_ps(dummy_mask,velec);
 600             velecsum         = _mm256_add_ps(velecsum,velec);
 601             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 602             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 603             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 604
 605             fscal            = _mm256_add_ps(felec,fvdw);
 606
 607             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 608
 609             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm256_mul_ps(fscal,dx00);
 613             ty               = _mm256_mul_ps(fscal,dy00);
 614             tz               = _mm256_mul_ps(fscal,dz00);
 615
 616             /* Update vectorial force */
 617             fix0             = _mm256_add_ps(fix0,tx);
 618             fiy0             = _mm256_add_ps(fiy0,ty);
 619             fiz0             = _mm256_add_ps(fiz0,tz);
 620
 621             fjx0             = _mm256_add_ps(fjx0,tx);
 622             fjy0             = _mm256_add_ps(fjy0,ty);
 623             fjz0             = _mm256_add_ps(fjz0,tz);
 624
 625             }
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 632             {
 633
 634             r10              = _mm256_mul_ps(rsq10,rinv10);
 635             r10              = _mm256_andnot_ps(dummy_mask,r10);
 636
 637             /* Compute parameters for interactions between i and j atoms */
 638             qq10             = _mm256_mul_ps(iq1,jq0);
 639
 640             /* EWALD ELECTROSTATICS */
 641
 642             /* Analytical PME correction */
 643             zeta2            = _mm256_mul_ps(beta2,rsq10);
 644             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 645             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 646             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 647             felec            = _mm256_mul_ps(qq10,felec);
 648             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 649             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 650             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 651             velec            = _mm256_mul_ps(qq10,velec);
 652
 653             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 654
 655             /* Update potential sum for this i atom from the interaction with this j atom. */
 656             velec            = _mm256_and_ps(velec,cutoff_mask);
 657             velec            = _mm256_andnot_ps(dummy_mask,velec);
 658             velecsum         = _mm256_add_ps(velecsum,velec);
 659
 660             fscal            = felec;
 661
 662             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 663
 664             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 665
 666             /* Calculate temporary vectorial force */
 667             tx               = _mm256_mul_ps(fscal,dx10);
 668             ty               = _mm256_mul_ps(fscal,dy10);
 669             tz               = _mm256_mul_ps(fscal,dz10);
 670
 671             /* Update vectorial force */
 672             fix1             = _mm256_add_ps(fix1,tx);
 673             fiy1             = _mm256_add_ps(fiy1,ty);
 674             fiz1             = _mm256_add_ps(fiz1,tz);
 675
 676             fjx0             = _mm256_add_ps(fjx0,tx);
 677             fjy0             = _mm256_add_ps(fjy0,ty);
 678             fjz0             = _mm256_add_ps(fjz0,tz);
 679
 680             }
 681
 682             /**************************
 683              * CALCULATE INTERACTIONS *
 684              **************************/
 685
 686             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 687             {
 688
 689             r20              = _mm256_mul_ps(rsq20,rinv20);
 690             r20              = _mm256_andnot_ps(dummy_mask,r20);
 691
 692             /* Compute parameters for interactions between i and j atoms */
 693             qq20             = _mm256_mul_ps(iq2,jq0);
 694
 695             /* EWALD ELECTROSTATICS */
 696
 697             /* Analytical PME correction */
 698             zeta2            = _mm256_mul_ps(beta2,rsq20);
 699             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 700             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 701             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 702             felec            = _mm256_mul_ps(qq20,felec);
 703             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 704             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 705             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 706             velec            = _mm256_mul_ps(qq20,velec);
 707
 708             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 709
 710             /* Update potential sum for this i atom from the interaction with this j atom. */
 711             velec            = _mm256_and_ps(velec,cutoff_mask);
 712             velec            = _mm256_andnot_ps(dummy_mask,velec);
 713             velecsum         = _mm256_add_ps(velecsum,velec);
 714
 715             fscal            = felec;
 716
 717             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 718
 719             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 720
 721             /* Calculate temporary vectorial force */
 722             tx               = _mm256_mul_ps(fscal,dx20);
 723             ty               = _mm256_mul_ps(fscal,dy20);
 724             tz               = _mm256_mul_ps(fscal,dz20);
 725
 726             /* Update vectorial force */
 727             fix2             = _mm256_add_ps(fix2,tx);
 728             fiy2             = _mm256_add_ps(fiy2,ty);
 729             fiz2             = _mm256_add_ps(fiz2,tz);
 730
 731             fjx0             = _mm256_add_ps(fjx0,tx);
 732             fjy0             = _mm256_add_ps(fjy0,ty);
 733             fjz0             = _mm256_add_ps(fjz0,tz);
 734
 735             }
 736
 737             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 738             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 739             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 740             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 741             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 742             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 743             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 744             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 745
 746             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 747
 748             /* Inner loop uses 351 flops */
 749         }
 750
 751         /* End of innermost loop */
 752
 753         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 754                                                  f+i_coord_offset,fshift+i_shift_offset);
 755
 756         ggid                        = gid[iidx];
 757         /* Update potential energies */
 758         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 759         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 760
 761         /* Increment number of inner iterations */
 762         inneriter                  += j_index_end - j_index_start;
 763
 764         /* Outer loop uses 20 flops */
 765     }
 766
 767     /* Increment number of outer iterations */
 768     outeriter        += nri;
 769
 770     /* Update outer/inner flops */
 771
 772     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*351);
 773 }
 774 /*
 775  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_256_single
 776  * Electrostatics interaction: Ewald
 777  * VdW interaction:            LennardJones
 778  * Geometry:                   Water3-Particle
 779  * Calculate force/pot:        Force
 780  */
 781 void
 782 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_256_single
 783                     (t_nblist                    * gmx_restrict       nlist,
 784                      rvec                        * gmx_restrict          xx,
 785                      rvec                        * gmx_restrict          ff,
 786                      t_forcerec                  * gmx_restrict          fr,
 787                      t_mdatoms                   * gmx_restrict     mdatoms,
 788                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 789                      t_nrnb                      * gmx_restrict        nrnb)
 790 {
 791     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 792      * just 0 for non-waters.
 793      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 794      * jnr indices corresponding to data put in the four positions in the SIMD register.
 795      */
 796     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 797     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 798     int              jnrA,jnrB,jnrC,jnrD;
 799     int              jnrE,jnrF,jnrG,jnrH;
 800     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 801     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 802     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 803     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 804     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 805     real             rcutoff_scalar;
 806     real             *shiftvec,*fshift,*x,*f;
 807     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 808     real             scratch[4*DIM];
 809     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 810     real *           vdwioffsetptr0;
 811     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 812     real *           vdwioffsetptr1;
 813     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 814     real *           vdwioffsetptr2;
 815     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 816     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 817     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 818     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 819     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 820     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 821     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 822     real             *charge;
 823     int              nvdwtype;
 824     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 825     int              *vdwtype;
 826     real             *vdwparam;
 827     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 828     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 829     __m256i          ewitab;
 830     __m128i          ewitab_lo,ewitab_hi;
 831     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 832     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 833     real             *ewtab;
 834     __m256           dummy_mask,cutoff_mask;
 835     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 836     __m256           one     = _mm256_set1_ps(1.0);
 837     __m256           two     = _mm256_set1_ps(2.0);
 838     x                = xx[0];
 839     f                = ff[0];
 840
 841     nri              = nlist->nri;
 842     iinr             = nlist->iinr;
 843     jindex           = nlist->jindex;
 844     jjnr             = nlist->jjnr;
 845     shiftidx         = nlist->shift;
 846     gid              = nlist->gid;
 847     shiftvec         = fr->shift_vec[0];
 848     fshift           = fr->fshift[0];
 849     facel            = _mm256_set1_ps(fr->epsfac);
 850     charge           = mdatoms->chargeA;
 851     nvdwtype         = fr->ntype;
 852     vdwparam         = fr->nbfp;
 853     vdwtype          = mdatoms->typeA;
 854
 855     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 856     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 857     beta2            = _mm256_mul_ps(beta,beta);
 858     beta3            = _mm256_mul_ps(beta,beta2);
 859
 860     ewtab            = fr->ic->tabq_coul_F;
 861     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 862     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 863
 864     /* Setup water-specific parameters */
 865     inr              = nlist->iinr[0];
 866     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 867     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 868     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 869     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 870
 871     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 872     rcutoff_scalar   = fr->rcoulomb;
 873     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 874     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 875
 876     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 877     rvdw             = _mm256_set1_ps(fr->rvdw);
 878
 879     /* Avoid stupid compiler warnings */
 880     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 881     j_coord_offsetA = 0;
 882     j_coord_offsetB = 0;
 883     j_coord_offsetC = 0;
 884     j_coord_offsetD = 0;
 885     j_coord_offsetE = 0;
 886     j_coord_offsetF = 0;
 887     j_coord_offsetG = 0;
 888     j_coord_offsetH = 0;
 889
 890     outeriter        = 0;
 891     inneriter        = 0;
 892
 893     for(iidx=0;iidx<4*DIM;iidx++)
 894     {
 895         scratch[iidx] = 0.0;
 896     }
 897
 898     /* Start outer loop over neighborlists */
 899     for(iidx=0; iidx<nri; iidx++)
 900     {
 901         /* Load shift vector for this list */
 902         i_shift_offset   = DIM*shiftidx[iidx];
 903
 904         /* Load limits for loop over neighbors */
 905         j_index_start    = jindex[iidx];
 906         j_index_end      = jindex[iidx+1];
 907
 908         /* Get outer coordinate index */
 909         inr              = iinr[iidx];
 910         i_coord_offset   = DIM*inr;
 911
 912         /* Load i particle coords and add shift vector */
 913         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 914                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 915
 916         fix0             = _mm256_setzero_ps();
 917         fiy0             = _mm256_setzero_ps();
 918         fiz0             = _mm256_setzero_ps();
 919         fix1             = _mm256_setzero_ps();
 920         fiy1             = _mm256_setzero_ps();
 921         fiz1             = _mm256_setzero_ps();
 922         fix2             = _mm256_setzero_ps();
 923         fiy2             = _mm256_setzero_ps();
 924         fiz2             = _mm256_setzero_ps();
 925
 926         /* Start inner kernel loop */
 927         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 928         {
 929
 930             /* Get j neighbor index, and coordinate index */
 931             jnrA             = jjnr[jidx];
 932             jnrB             = jjnr[jidx+1];
 933             jnrC             = jjnr[jidx+2];
 934             jnrD             = jjnr[jidx+3];
 935             jnrE             = jjnr[jidx+4];
 936             jnrF             = jjnr[jidx+5];
 937             jnrG             = jjnr[jidx+6];
 938             jnrH             = jjnr[jidx+7];
 939             j_coord_offsetA  = DIM*jnrA;
 940             j_coord_offsetB  = DIM*jnrB;
 941             j_coord_offsetC  = DIM*jnrC;
 942             j_coord_offsetD  = DIM*jnrD;
 943             j_coord_offsetE  = DIM*jnrE;
 944             j_coord_offsetF  = DIM*jnrF;
 945             j_coord_offsetG  = DIM*jnrG;
 946             j_coord_offsetH  = DIM*jnrH;
 947
 948             /* load j atom coordinates */
 949             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 950                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 951                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 952                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 953                                                  &jx0,&jy0,&jz0);
 954
 955             /* Calculate displacement vector */
 956             dx00             = _mm256_sub_ps(ix0,jx0);
 957             dy00             = _mm256_sub_ps(iy0,jy0);
 958             dz00             = _mm256_sub_ps(iz0,jz0);
 959             dx10             = _mm256_sub_ps(ix1,jx0);
 960             dy10             = _mm256_sub_ps(iy1,jy0);
 961             dz10             = _mm256_sub_ps(iz1,jz0);
 962             dx20             = _mm256_sub_ps(ix2,jx0);
 963             dy20             = _mm256_sub_ps(iy2,jy0);
 964             dz20             = _mm256_sub_ps(iz2,jz0);
 965
 966             /* Calculate squared distance and things based on it */
 967             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 968             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 969             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 970
 971             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 972             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 973             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 974
 975             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 976             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 977             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 978
 979             /* Load parameters for j particles */
 980             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 981                                                                  charge+jnrC+0,charge+jnrD+0,
 982                                                                  charge+jnrE+0,charge+jnrF+0,
 983                                                                  charge+jnrG+0,charge+jnrH+0);
 984             vdwjidx0A        = 2*vdwtype[jnrA+0];
 985             vdwjidx0B        = 2*vdwtype[jnrB+0];
 986             vdwjidx0C        = 2*vdwtype[jnrC+0];
 987             vdwjidx0D        = 2*vdwtype[jnrD+0];
 988             vdwjidx0E        = 2*vdwtype[jnrE+0];
 989             vdwjidx0F        = 2*vdwtype[jnrF+0];
 990             vdwjidx0G        = 2*vdwtype[jnrG+0];
 991             vdwjidx0H        = 2*vdwtype[jnrH+0];
 992
 993             fjx0             = _mm256_setzero_ps();
 994             fjy0             = _mm256_setzero_ps();
 995             fjz0             = _mm256_setzero_ps();
 996
 997             /**************************
 998              * CALCULATE INTERACTIONS *
 999              **************************/
1000
1001             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1002             {
1003
1004             r00              = _mm256_mul_ps(rsq00,rinv00);
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             qq00             = _mm256_mul_ps(iq0,jq0);
1008             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1009                                             vdwioffsetptr0+vdwjidx0B,
1010                                             vdwioffsetptr0+vdwjidx0C,
1011                                             vdwioffsetptr0+vdwjidx0D,
1012                                             vdwioffsetptr0+vdwjidx0E,
1013                                             vdwioffsetptr0+vdwjidx0F,
1014                                             vdwioffsetptr0+vdwjidx0G,
1015                                             vdwioffsetptr0+vdwjidx0H,
1016                                             &c6_00,&c12_00);
1017
1018             /* EWALD ELECTROSTATICS */
1019
1020             /* Analytical PME correction */
1021             zeta2            = _mm256_mul_ps(beta2,rsq00);
1022             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1023             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1024             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1025             felec            = _mm256_mul_ps(qq00,felec);
1026
1027             /* LENNARD-JONES DISPERSION/REPULSION */
1028
1029             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1030             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1031
1032             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1033
1034             fscal            = _mm256_add_ps(felec,fvdw);
1035
1036             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1037
1038             /* Calculate temporary vectorial force */
1039             tx               = _mm256_mul_ps(fscal,dx00);
1040             ty               = _mm256_mul_ps(fscal,dy00);
1041             tz               = _mm256_mul_ps(fscal,dz00);
1042
1043             /* Update vectorial force */
1044             fix0             = _mm256_add_ps(fix0,tx);
1045             fiy0             = _mm256_add_ps(fiy0,ty);
1046             fiz0             = _mm256_add_ps(fiz0,tz);
1047
1048             fjx0             = _mm256_add_ps(fjx0,tx);
1049             fjy0             = _mm256_add_ps(fjy0,ty);
1050             fjz0             = _mm256_add_ps(fjz0,tz);
1051
1052             }
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1059             {
1060
1061             r10              = _mm256_mul_ps(rsq10,rinv10);
1062
1063             /* Compute parameters for interactions between i and j atoms */
1064             qq10             = _mm256_mul_ps(iq1,jq0);
1065
1066             /* EWALD ELECTROSTATICS */
1067
1068             /* Analytical PME correction */
1069             zeta2            = _mm256_mul_ps(beta2,rsq10);
1070             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1071             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1072             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1073             felec            = _mm256_mul_ps(qq10,felec);
1074
1075             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1076
1077             fscal            = felec;
1078
1079             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm256_mul_ps(fscal,dx10);
1083             ty               = _mm256_mul_ps(fscal,dy10);
1084             tz               = _mm256_mul_ps(fscal,dz10);
1085
1086             /* Update vectorial force */
1087             fix1             = _mm256_add_ps(fix1,tx);
1088             fiy1             = _mm256_add_ps(fiy1,ty);
1089             fiz1             = _mm256_add_ps(fiz1,tz);
1090
1091             fjx0             = _mm256_add_ps(fjx0,tx);
1092             fjy0             = _mm256_add_ps(fjy0,ty);
1093             fjz0             = _mm256_add_ps(fjz0,tz);
1094
1095             }
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1102             {
1103
1104             r20              = _mm256_mul_ps(rsq20,rinv20);
1105
1106             /* Compute parameters for interactions between i and j atoms */
1107             qq20             = _mm256_mul_ps(iq2,jq0);
1108
1109             /* EWALD ELECTROSTATICS */
1110
1111             /* Analytical PME correction */
1112             zeta2            = _mm256_mul_ps(beta2,rsq20);
1113             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1114             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1115             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1116             felec            = _mm256_mul_ps(qq20,felec);
1117
1118             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1119
1120             fscal            = felec;
1121
1122             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1123
1124             /* Calculate temporary vectorial force */
1125             tx               = _mm256_mul_ps(fscal,dx20);
1126             ty               = _mm256_mul_ps(fscal,dy20);
1127             tz               = _mm256_mul_ps(fscal,dz20);
1128
1129             /* Update vectorial force */
1130             fix2             = _mm256_add_ps(fix2,tx);
1131             fiy2             = _mm256_add_ps(fiy2,ty);
1132             fiz2             = _mm256_add_ps(fiz2,tz);
1133
1134             fjx0             = _mm256_add_ps(fjx0,tx);
1135             fjy0             = _mm256_add_ps(fjy0,ty);
1136             fjz0             = _mm256_add_ps(fjz0,tz);
1137
1138             }
1139
1140             fjptrA             = f+j_coord_offsetA;
1141             fjptrB             = f+j_coord_offsetB;
1142             fjptrC             = f+j_coord_offsetC;
1143             fjptrD             = f+j_coord_offsetD;
1144             fjptrE             = f+j_coord_offsetE;
1145             fjptrF             = f+j_coord_offsetF;
1146             fjptrG             = f+j_coord_offsetG;
1147             fjptrH             = f+j_coord_offsetH;
1148
1149             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1150
1151             /* Inner loop uses 187 flops */
1152         }
1153
1154         if(jidx<j_index_end)
1155         {
1156
1157             /* Get j neighbor index, and coordinate index */
1158             jnrlistA         = jjnr[jidx];
1159             jnrlistB         = jjnr[jidx+1];
1160             jnrlistC         = jjnr[jidx+2];
1161             jnrlistD         = jjnr[jidx+3];
1162             jnrlistE         = jjnr[jidx+4];
1163             jnrlistF         = jjnr[jidx+5];
1164             jnrlistG         = jjnr[jidx+6];
1165             jnrlistH         = jjnr[jidx+7];
1166             /* Sign of each element will be negative for non-real atoms.
1167              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1168              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1169              */
1170             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1171                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1172
1173             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1174             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1175             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1176             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1177             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1178             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1179             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1180             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1181             j_coord_offsetA  = DIM*jnrA;
1182             j_coord_offsetB  = DIM*jnrB;
1183             j_coord_offsetC  = DIM*jnrC;
1184             j_coord_offsetD  = DIM*jnrD;
1185             j_coord_offsetE  = DIM*jnrE;
1186             j_coord_offsetF  = DIM*jnrF;
1187             j_coord_offsetG  = DIM*jnrG;
1188             j_coord_offsetH  = DIM*jnrH;
1189
1190             /* load j atom coordinates */
1191             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1192                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1193                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1194                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1195                                                  &jx0,&jy0,&jz0);
1196
1197             /* Calculate displacement vector */
1198             dx00             = _mm256_sub_ps(ix0,jx0);
1199             dy00             = _mm256_sub_ps(iy0,jy0);
1200             dz00             = _mm256_sub_ps(iz0,jz0);
1201             dx10             = _mm256_sub_ps(ix1,jx0);
1202             dy10             = _mm256_sub_ps(iy1,jy0);
1203             dz10             = _mm256_sub_ps(iz1,jz0);
1204             dx20             = _mm256_sub_ps(ix2,jx0);
1205             dy20             = _mm256_sub_ps(iy2,jy0);
1206             dz20             = _mm256_sub_ps(iz2,jz0);
1207
1208             /* Calculate squared distance and things based on it */
1209             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1210             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1211             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1212
1213             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1214             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1215             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1216
1217             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1218             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1219             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1220
1221             /* Load parameters for j particles */
1222             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1223                                                                  charge+jnrC+0,charge+jnrD+0,
1224                                                                  charge+jnrE+0,charge+jnrF+0,
1225                                                                  charge+jnrG+0,charge+jnrH+0);
1226             vdwjidx0A        = 2*vdwtype[jnrA+0];
1227             vdwjidx0B        = 2*vdwtype[jnrB+0];
1228             vdwjidx0C        = 2*vdwtype[jnrC+0];
1229             vdwjidx0D        = 2*vdwtype[jnrD+0];
1230             vdwjidx0E        = 2*vdwtype[jnrE+0];
1231             vdwjidx0F        = 2*vdwtype[jnrF+0];
1232             vdwjidx0G        = 2*vdwtype[jnrG+0];
1233             vdwjidx0H        = 2*vdwtype[jnrH+0];
1234
1235             fjx0             = _mm256_setzero_ps();
1236             fjy0             = _mm256_setzero_ps();
1237             fjz0             = _mm256_setzero_ps();
1238
1239             /**************************
1240              * CALCULATE INTERACTIONS *
1241              **************************/
1242
1243             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1244             {
1245
1246             r00              = _mm256_mul_ps(rsq00,rinv00);
1247             r00              = _mm256_andnot_ps(dummy_mask,r00);
1248
1249             /* Compute parameters for interactions between i and j atoms */
1250             qq00             = _mm256_mul_ps(iq0,jq0);
1251             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1252                                             vdwioffsetptr0+vdwjidx0B,
1253                                             vdwioffsetptr0+vdwjidx0C,
1254                                             vdwioffsetptr0+vdwjidx0D,
1255                                             vdwioffsetptr0+vdwjidx0E,
1256                                             vdwioffsetptr0+vdwjidx0F,
1257                                             vdwioffsetptr0+vdwjidx0G,
1258                                             vdwioffsetptr0+vdwjidx0H,
1259                                             &c6_00,&c12_00);
1260
1261             /* EWALD ELECTROSTATICS */
1262
1263             /* Analytical PME correction */
1264             zeta2            = _mm256_mul_ps(beta2,rsq00);
1265             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1266             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1267             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1268             felec            = _mm256_mul_ps(qq00,felec);
1269
1270             /* LENNARD-JONES DISPERSION/REPULSION */
1271
1272             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1273             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1274
1275             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1276
1277             fscal            = _mm256_add_ps(felec,fvdw);
1278
1279             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1280
1281             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1282
1283             /* Calculate temporary vectorial force */
1284             tx               = _mm256_mul_ps(fscal,dx00);
1285             ty               = _mm256_mul_ps(fscal,dy00);
1286             tz               = _mm256_mul_ps(fscal,dz00);
1287
1288             /* Update vectorial force */
1289             fix0             = _mm256_add_ps(fix0,tx);
1290             fiy0             = _mm256_add_ps(fiy0,ty);
1291             fiz0             = _mm256_add_ps(fiz0,tz);
1292
1293             fjx0             = _mm256_add_ps(fjx0,tx);
1294             fjy0             = _mm256_add_ps(fjy0,ty);
1295             fjz0             = _mm256_add_ps(fjz0,tz);
1296
1297             }
1298
1299             /**************************
1300              * CALCULATE INTERACTIONS *
1301              **************************/
1302
1303             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1304             {
1305
1306             r10              = _mm256_mul_ps(rsq10,rinv10);
1307             r10              = _mm256_andnot_ps(dummy_mask,r10);
1308
1309             /* Compute parameters for interactions between i and j atoms */
1310             qq10             = _mm256_mul_ps(iq1,jq0);
1311
1312             /* EWALD ELECTROSTATICS */
1313
1314             /* Analytical PME correction */
1315             zeta2            = _mm256_mul_ps(beta2,rsq10);
1316             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1317             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1318             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1319             felec            = _mm256_mul_ps(qq10,felec);
1320
1321             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1322
1323             fscal            = felec;
1324
1325             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1326
1327             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1328
1329             /* Calculate temporary vectorial force */
1330             tx               = _mm256_mul_ps(fscal,dx10);
1331             ty               = _mm256_mul_ps(fscal,dy10);
1332             tz               = _mm256_mul_ps(fscal,dz10);
1333
1334             /* Update vectorial force */
1335             fix1             = _mm256_add_ps(fix1,tx);
1336             fiy1             = _mm256_add_ps(fiy1,ty);
1337             fiz1             = _mm256_add_ps(fiz1,tz);
1338
1339             fjx0             = _mm256_add_ps(fjx0,tx);
1340             fjy0             = _mm256_add_ps(fjy0,ty);
1341             fjz0             = _mm256_add_ps(fjz0,tz);
1342
1343             }
1344
1345             /**************************
1346              * CALCULATE INTERACTIONS *
1347              **************************/
1348
1349             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1350             {
1351
1352             r20              = _mm256_mul_ps(rsq20,rinv20);
1353             r20              = _mm256_andnot_ps(dummy_mask,r20);
1354
1355             /* Compute parameters for interactions between i and j atoms */
1356             qq20             = _mm256_mul_ps(iq2,jq0);
1357
1358             /* EWALD ELECTROSTATICS */
1359
1360             /* Analytical PME correction */
1361             zeta2            = _mm256_mul_ps(beta2,rsq20);
1362             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1363             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
1364             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1365             felec            = _mm256_mul_ps(qq20,felec);
1366
1367             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1368
1369             fscal            = felec;
1370
1371             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1372
1373             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1374
1375             /* Calculate temporary vectorial force */
1376             tx               = _mm256_mul_ps(fscal,dx20);
1377             ty               = _mm256_mul_ps(fscal,dy20);
1378             tz               = _mm256_mul_ps(fscal,dz20);
1379
1380             /* Update vectorial force */
1381             fix2             = _mm256_add_ps(fix2,tx);
1382             fiy2             = _mm256_add_ps(fiy2,ty);
1383             fiz2             = _mm256_add_ps(fiz2,tz);
1384
1385             fjx0             = _mm256_add_ps(fjx0,tx);
1386             fjy0             = _mm256_add_ps(fjy0,ty);
1387             fjz0             = _mm256_add_ps(fjz0,tz);
1388
1389             }
1390
1391             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1392             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1393             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1394             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1395             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1396             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1397             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1398             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1399
1400             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1401
1402             /* Inner loop uses 190 flops */
1403         }
1404
1405         /* End of innermost loop */
1406
1407         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1408                                                  f+i_coord_offset,fshift+i_shift_offset);
1409
1410         /* Increment number of inner iterations */
1411         inneriter                  += j_index_end - j_index_start;
1412
1413         /* Outer loop uses 18 flops */
1414     }
1415
1416     /* Increment number of outer iterations */
1417     outeriter        += nri;
1418
1419     /* Update outer/inner flops */
1420
1421     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*190);
1422 }