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) 2012,2013,2014,2015,2017, 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 #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_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_256_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_avx_256_single
  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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  77     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  78     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  83     real             scratch[4*DIM];
  84     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     real *           vdwioffsetptr0;
  86     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     real *           vdwioffsetptr1;
  88     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     real *           vdwioffsetptr2;
  90     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  92     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 103     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 104     __m256i          ewitab;
 105     __m128i          ewitab_lo,ewitab_hi;
 106     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 107     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 108     real             *ewtab;
 109     __m256           dummy_mask,cutoff_mask;
 110     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 111     __m256           one     = _mm256_set1_ps(1.0);
 112     __m256           two     = _mm256_set1_ps(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm256_set1_ps(fr->ic->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129
 130     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 131     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 132     beta2            = _mm256_mul_ps(beta,beta);
 133     beta3            = _mm256_mul_ps(beta,beta2);
 134
 135     ewtab            = fr->ic->tabq_coul_FDV0;
 136     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 137     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 142     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 143     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 144     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 145
 146     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 147     rcutoff_scalar   = fr->ic->rcoulomb;
 148     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 149     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 150
 151     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 152     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 153
 154     /* Avoid stupid compiler warnings */
 155     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 156     j_coord_offsetA = 0;
 157     j_coord_offsetB = 0;
 158     j_coord_offsetC = 0;
 159     j_coord_offsetD = 0;
 160     j_coord_offsetE = 0;
 161     j_coord_offsetF = 0;
 162     j_coord_offsetG = 0;
 163     j_coord_offsetH = 0;
 164
 165     outeriter        = 0;
 166     inneriter        = 0;
 167
 168     for(iidx=0;iidx<4*DIM;iidx++)
 169     {
 170         scratch[iidx] = 0.0;
 171     }
 172
 173     /* Start outer loop over neighborlists */
 174     for(iidx=0; iidx<nri; iidx++)
 175     {
 176         /* Load shift vector for this list */
 177         i_shift_offset   = DIM*shiftidx[iidx];
 178
 179         /* Load limits for loop over neighbors */
 180         j_index_start    = jindex[iidx];
 181         j_index_end      = jindex[iidx+1];
 182
 183         /* Get outer coordinate index */
 184         inr              = iinr[iidx];
 185         i_coord_offset   = DIM*inr;
 186
 187         /* Load i particle coords and add shift vector */
 188         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 189                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 190
 191         fix0             = _mm256_setzero_ps();
 192         fiy0             = _mm256_setzero_ps();
 193         fiz0             = _mm256_setzero_ps();
 194         fix1             = _mm256_setzero_ps();
 195         fiy1             = _mm256_setzero_ps();
 196         fiz1             = _mm256_setzero_ps();
 197         fix2             = _mm256_setzero_ps();
 198         fiy2             = _mm256_setzero_ps();
 199         fiz2             = _mm256_setzero_ps();
 200
 201         /* Reset potential sums */
 202         velecsum         = _mm256_setzero_ps();
 203         vvdwsum          = _mm256_setzero_ps();
 204
 205         /* Start inner kernel loop */
 206         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 207         {
 208
 209             /* Get j neighbor index, and coordinate index */
 210             jnrA             = jjnr[jidx];
 211             jnrB             = jjnr[jidx+1];
 212             jnrC             = jjnr[jidx+2];
 213             jnrD             = jjnr[jidx+3];
 214             jnrE             = jjnr[jidx+4];
 215             jnrF             = jjnr[jidx+5];
 216             jnrG             = jjnr[jidx+6];
 217             jnrH             = jjnr[jidx+7];
 218             j_coord_offsetA  = DIM*jnrA;
 219             j_coord_offsetB  = DIM*jnrB;
 220             j_coord_offsetC  = DIM*jnrC;
 221             j_coord_offsetD  = DIM*jnrD;
 222             j_coord_offsetE  = DIM*jnrE;
 223             j_coord_offsetF  = DIM*jnrF;
 224             j_coord_offsetG  = DIM*jnrG;
 225             j_coord_offsetH  = DIM*jnrH;
 226
 227             /* load j atom coordinates */
 228             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 229                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 230                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 231                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 232                                                  &jx0,&jy0,&jz0);
 233
 234             /* Calculate displacement vector */
 235             dx00             = _mm256_sub_ps(ix0,jx0);
 236             dy00             = _mm256_sub_ps(iy0,jy0);
 237             dz00             = _mm256_sub_ps(iz0,jz0);
 238             dx10             = _mm256_sub_ps(ix1,jx0);
 239             dy10             = _mm256_sub_ps(iy1,jy0);
 240             dz10             = _mm256_sub_ps(iz1,jz0);
 241             dx20             = _mm256_sub_ps(ix2,jx0);
 242             dy20             = _mm256_sub_ps(iy2,jy0);
 243             dz20             = _mm256_sub_ps(iz2,jz0);
 244
 245             /* Calculate squared distance and things based on it */
 246             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 247             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 248             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 249
 250             rinv00           = avx256_invsqrt_f(rsq00);
 251             rinv10           = avx256_invsqrt_f(rsq10);
 252             rinv20           = avx256_invsqrt_f(rsq20);
 253
 254             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 255             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 256             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 257
 258             /* Load parameters for j particles */
 259             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 260                                                                  charge+jnrC+0,charge+jnrD+0,
 261                                                                  charge+jnrE+0,charge+jnrF+0,
 262                                                                  charge+jnrG+0,charge+jnrH+0);
 263             vdwjidx0A        = 2*vdwtype[jnrA+0];
 264             vdwjidx0B        = 2*vdwtype[jnrB+0];
 265             vdwjidx0C        = 2*vdwtype[jnrC+0];
 266             vdwjidx0D        = 2*vdwtype[jnrD+0];
 267             vdwjidx0E        = 2*vdwtype[jnrE+0];
 268             vdwjidx0F        = 2*vdwtype[jnrF+0];
 269             vdwjidx0G        = 2*vdwtype[jnrG+0];
 270             vdwjidx0H        = 2*vdwtype[jnrH+0];
 271
 272             fjx0             = _mm256_setzero_ps();
 273             fjy0             = _mm256_setzero_ps();
 274             fjz0             = _mm256_setzero_ps();
 275
 276             /**************************
 277              * CALCULATE INTERACTIONS *
 278              **************************/
 279
 280             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 281             {
 282
 283             r00              = _mm256_mul_ps(rsq00,rinv00);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq00             = _mm256_mul_ps(iq0,jq0);
 287             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 288                                             vdwioffsetptr0+vdwjidx0B,
 289                                             vdwioffsetptr0+vdwjidx0C,
 290                                             vdwioffsetptr0+vdwjidx0D,
 291                                             vdwioffsetptr0+vdwjidx0E,
 292                                             vdwioffsetptr0+vdwjidx0F,
 293                                             vdwioffsetptr0+vdwjidx0G,
 294                                             vdwioffsetptr0+vdwjidx0H,
 295                                             &c6_00,&c12_00);
 296
 297             /* EWALD ELECTROSTATICS */
 298
 299             /* Analytical PME correction */
 300             zeta2            = _mm256_mul_ps(beta2,rsq00);
 301             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 302             pmecorrF         = avx256_pmecorrF_f(zeta2);
 303             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 304             felec            = _mm256_mul_ps(qq00,felec);
 305             pmecorrV         = avx256_pmecorrV_f(zeta2);
 306             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 307             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 308             velec            = _mm256_mul_ps(qq00,velec);
 309
 310             /* LENNARD-JONES DISPERSION/REPULSION */
 311
 312             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 313             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 314             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 315             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) ,
 316                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 317             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 318
 319             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velec            = _mm256_and_ps(velec,cutoff_mask);
 323             velecsum         = _mm256_add_ps(velecsum,velec);
 324             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 325             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 326
 327             fscal            = _mm256_add_ps(felec,fvdw);
 328
 329             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm256_mul_ps(fscal,dx00);
 333             ty               = _mm256_mul_ps(fscal,dy00);
 334             tz               = _mm256_mul_ps(fscal,dz00);
 335
 336             /* Update vectorial force */
 337             fix0             = _mm256_add_ps(fix0,tx);
 338             fiy0             = _mm256_add_ps(fiy0,ty);
 339             fiz0             = _mm256_add_ps(fiz0,tz);
 340
 341             fjx0             = _mm256_add_ps(fjx0,tx);
 342             fjy0             = _mm256_add_ps(fjy0,ty);
 343             fjz0             = _mm256_add_ps(fjz0,tz);
 344
 345             }
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 352             {
 353
 354             r10              = _mm256_mul_ps(rsq10,rinv10);
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq10             = _mm256_mul_ps(iq1,jq0);
 358
 359             /* EWALD ELECTROSTATICS */
 360
 361             /* Analytical PME correction */
 362             zeta2            = _mm256_mul_ps(beta2,rsq10);
 363             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 364             pmecorrF         = avx256_pmecorrF_f(zeta2);
 365             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 366             felec            = _mm256_mul_ps(qq10,felec);
 367             pmecorrV         = avx256_pmecorrV_f(zeta2);
 368             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 369             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 370             velec            = _mm256_mul_ps(qq10,velec);
 371
 372             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velec            = _mm256_and_ps(velec,cutoff_mask);
 376             velecsum         = _mm256_add_ps(velecsum,velec);
 377
 378             fscal            = felec;
 379
 380             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm256_mul_ps(fscal,dx10);
 384             ty               = _mm256_mul_ps(fscal,dy10);
 385             tz               = _mm256_mul_ps(fscal,dz10);
 386
 387             /* Update vectorial force */
 388             fix1             = _mm256_add_ps(fix1,tx);
 389             fiy1             = _mm256_add_ps(fiy1,ty);
 390             fiz1             = _mm256_add_ps(fiz1,tz);
 391
 392             fjx0             = _mm256_add_ps(fjx0,tx);
 393             fjy0             = _mm256_add_ps(fjy0,ty);
 394             fjz0             = _mm256_add_ps(fjz0,tz);
 395
 396             }
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 403             {
 404
 405             r20              = _mm256_mul_ps(rsq20,rinv20);
 406
 407             /* Compute parameters for interactions between i and j atoms */
 408             qq20             = _mm256_mul_ps(iq2,jq0);
 409
 410             /* EWALD ELECTROSTATICS */
 411
 412             /* Analytical PME correction */
 413             zeta2            = _mm256_mul_ps(beta2,rsq20);
 414             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 415             pmecorrF         = avx256_pmecorrF_f(zeta2);
 416             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 417             felec            = _mm256_mul_ps(qq20,felec);
 418             pmecorrV         = avx256_pmecorrV_f(zeta2);
 419             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 420             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 421             velec            = _mm256_mul_ps(qq20,velec);
 422
 423             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 424
 425             /* Update potential sum for this i atom from the interaction with this j atom. */
 426             velec            = _mm256_and_ps(velec,cutoff_mask);
 427             velecsum         = _mm256_add_ps(velecsum,velec);
 428
 429             fscal            = felec;
 430
 431             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 432
 433             /* Calculate temporary vectorial force */
 434             tx               = _mm256_mul_ps(fscal,dx20);
 435             ty               = _mm256_mul_ps(fscal,dy20);
 436             tz               = _mm256_mul_ps(fscal,dz20);
 437
 438             /* Update vectorial force */
 439             fix2             = _mm256_add_ps(fix2,tx);
 440             fiy2             = _mm256_add_ps(fiy2,ty);
 441             fiz2             = _mm256_add_ps(fiz2,tz);
 442
 443             fjx0             = _mm256_add_ps(fjx0,tx);
 444             fjy0             = _mm256_add_ps(fjy0,ty);
 445             fjz0             = _mm256_add_ps(fjz0,tz);
 446
 447             }
 448
 449             fjptrA             = f+j_coord_offsetA;
 450             fjptrB             = f+j_coord_offsetB;
 451             fjptrC             = f+j_coord_offsetC;
 452             fjptrD             = f+j_coord_offsetD;
 453             fjptrE             = f+j_coord_offsetE;
 454             fjptrF             = f+j_coord_offsetF;
 455             fjptrG             = f+j_coord_offsetG;
 456             fjptrH             = f+j_coord_offsetH;
 457
 458             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 459
 460             /* Inner loop uses 348 flops */
 461         }
 462
 463         if(jidx<j_index_end)
 464         {
 465
 466             /* Get j neighbor index, and coordinate index */
 467             jnrlistA         = jjnr[jidx];
 468             jnrlistB         = jjnr[jidx+1];
 469             jnrlistC         = jjnr[jidx+2];
 470             jnrlistD         = jjnr[jidx+3];
 471             jnrlistE         = jjnr[jidx+4];
 472             jnrlistF         = jjnr[jidx+5];
 473             jnrlistG         = jjnr[jidx+6];
 474             jnrlistH         = jjnr[jidx+7];
 475             /* Sign of each element will be negative for non-real atoms.
 476              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 477              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 478              */
 479             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 480                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 481
 482             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 483             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 484             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 485             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 486             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 487             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 488             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 489             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 490             j_coord_offsetA  = DIM*jnrA;
 491             j_coord_offsetB  = DIM*jnrB;
 492             j_coord_offsetC  = DIM*jnrC;
 493             j_coord_offsetD  = DIM*jnrD;
 494             j_coord_offsetE  = DIM*jnrE;
 495             j_coord_offsetF  = DIM*jnrF;
 496             j_coord_offsetG  = DIM*jnrG;
 497             j_coord_offsetH  = DIM*jnrH;
 498
 499             /* load j atom coordinates */
 500             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 501                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 502                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 503                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 504                                                  &jx0,&jy0,&jz0);
 505
 506             /* Calculate displacement vector */
 507             dx00             = _mm256_sub_ps(ix0,jx0);
 508             dy00             = _mm256_sub_ps(iy0,jy0);
 509             dz00             = _mm256_sub_ps(iz0,jz0);
 510             dx10             = _mm256_sub_ps(ix1,jx0);
 511             dy10             = _mm256_sub_ps(iy1,jy0);
 512             dz10             = _mm256_sub_ps(iz1,jz0);
 513             dx20             = _mm256_sub_ps(ix2,jx0);
 514             dy20             = _mm256_sub_ps(iy2,jy0);
 515             dz20             = _mm256_sub_ps(iz2,jz0);
 516
 517             /* Calculate squared distance and things based on it */
 518             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 519             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 520             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 521
 522             rinv00           = avx256_invsqrt_f(rsq00);
 523             rinv10           = avx256_invsqrt_f(rsq10);
 524             rinv20           = avx256_invsqrt_f(rsq20);
 525
 526             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 527             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 528             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 529
 530             /* Load parameters for j particles */
 531             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 532                                                                  charge+jnrC+0,charge+jnrD+0,
 533                                                                  charge+jnrE+0,charge+jnrF+0,
 534                                                                  charge+jnrG+0,charge+jnrH+0);
 535             vdwjidx0A        = 2*vdwtype[jnrA+0];
 536             vdwjidx0B        = 2*vdwtype[jnrB+0];
 537             vdwjidx0C        = 2*vdwtype[jnrC+0];
 538             vdwjidx0D        = 2*vdwtype[jnrD+0];
 539             vdwjidx0E        = 2*vdwtype[jnrE+0];
 540             vdwjidx0F        = 2*vdwtype[jnrF+0];
 541             vdwjidx0G        = 2*vdwtype[jnrG+0];
 542             vdwjidx0H        = 2*vdwtype[jnrH+0];
 543
 544             fjx0             = _mm256_setzero_ps();
 545             fjy0             = _mm256_setzero_ps();
 546             fjz0             = _mm256_setzero_ps();
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 553             {
 554
 555             r00              = _mm256_mul_ps(rsq00,rinv00);
 556             r00              = _mm256_andnot_ps(dummy_mask,r00);
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             qq00             = _mm256_mul_ps(iq0,jq0);
 560             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 561                                             vdwioffsetptr0+vdwjidx0B,
 562                                             vdwioffsetptr0+vdwjidx0C,
 563                                             vdwioffsetptr0+vdwjidx0D,
 564                                             vdwioffsetptr0+vdwjidx0E,
 565                                             vdwioffsetptr0+vdwjidx0F,
 566                                             vdwioffsetptr0+vdwjidx0G,
 567                                             vdwioffsetptr0+vdwjidx0H,
 568                                             &c6_00,&c12_00);
 569
 570             /* EWALD ELECTROSTATICS */
 571
 572             /* Analytical PME correction */
 573             zeta2            = _mm256_mul_ps(beta2,rsq00);
 574             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 575             pmecorrF         = avx256_pmecorrF_f(zeta2);
 576             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 577             felec            = _mm256_mul_ps(qq00,felec);
 578             pmecorrV         = avx256_pmecorrV_f(zeta2);
 579             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 580             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv00,sh_ewald),pmecorrV);
 581             velec            = _mm256_mul_ps(qq00,velec);
 582
 583             /* LENNARD-JONES DISPERSION/REPULSION */
 584
 585             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 586             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 587             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 588             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) ,
 589                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 590             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 591
 592             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 593
 594             /* Update potential sum for this i atom from the interaction with this j atom. */
 595             velec            = _mm256_and_ps(velec,cutoff_mask);
 596             velec            = _mm256_andnot_ps(dummy_mask,velec);
 597             velecsum         = _mm256_add_ps(velecsum,velec);
 598             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 599             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 600             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 601
 602             fscal            = _mm256_add_ps(felec,fvdw);
 603
 604             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 605
 606             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 607
 608             /* Calculate temporary vectorial force */
 609             tx               = _mm256_mul_ps(fscal,dx00);
 610             ty               = _mm256_mul_ps(fscal,dy00);
 611             tz               = _mm256_mul_ps(fscal,dz00);
 612
 613             /* Update vectorial force */
 614             fix0             = _mm256_add_ps(fix0,tx);
 615             fiy0             = _mm256_add_ps(fiy0,ty);
 616             fiz0             = _mm256_add_ps(fiz0,tz);
 617
 618             fjx0             = _mm256_add_ps(fjx0,tx);
 619             fjy0             = _mm256_add_ps(fjy0,ty);
 620             fjz0             = _mm256_add_ps(fjz0,tz);
 621
 622             }
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 629             {
 630
 631             r10              = _mm256_mul_ps(rsq10,rinv10);
 632             r10              = _mm256_andnot_ps(dummy_mask,r10);
 633
 634             /* Compute parameters for interactions between i and j atoms */
 635             qq10             = _mm256_mul_ps(iq1,jq0);
 636
 637             /* EWALD ELECTROSTATICS */
 638
 639             /* Analytical PME correction */
 640             zeta2            = _mm256_mul_ps(beta2,rsq10);
 641             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
 642             pmecorrF         = avx256_pmecorrF_f(zeta2);
 643             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 644             felec            = _mm256_mul_ps(qq10,felec);
 645             pmecorrV         = avx256_pmecorrV_f(zeta2);
 646             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 647             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv10,sh_ewald),pmecorrV);
 648             velec            = _mm256_mul_ps(qq10,velec);
 649
 650             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 651
 652             /* Update potential sum for this i atom from the interaction with this j atom. */
 653             velec            = _mm256_and_ps(velec,cutoff_mask);
 654             velec            = _mm256_andnot_ps(dummy_mask,velec);
 655             velecsum         = _mm256_add_ps(velecsum,velec);
 656
 657             fscal            = felec;
 658
 659             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 660
 661             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 662
 663             /* Calculate temporary vectorial force */
 664             tx               = _mm256_mul_ps(fscal,dx10);
 665             ty               = _mm256_mul_ps(fscal,dy10);
 666             tz               = _mm256_mul_ps(fscal,dz10);
 667
 668             /* Update vectorial force */
 669             fix1             = _mm256_add_ps(fix1,tx);
 670             fiy1             = _mm256_add_ps(fiy1,ty);
 671             fiz1             = _mm256_add_ps(fiz1,tz);
 672
 673             fjx0             = _mm256_add_ps(fjx0,tx);
 674             fjy0             = _mm256_add_ps(fjy0,ty);
 675             fjz0             = _mm256_add_ps(fjz0,tz);
 676
 677             }
 678
 679             /**************************
 680              * CALCULATE INTERACTIONS *
 681              **************************/
 682
 683             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 684             {
 685
 686             r20              = _mm256_mul_ps(rsq20,rinv20);
 687             r20              = _mm256_andnot_ps(dummy_mask,r20);
 688
 689             /* Compute parameters for interactions between i and j atoms */
 690             qq20             = _mm256_mul_ps(iq2,jq0);
 691
 692             /* EWALD ELECTROSTATICS */
 693
 694             /* Analytical PME correction */
 695             zeta2            = _mm256_mul_ps(beta2,rsq20);
 696             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
 697             pmecorrF         = avx256_pmecorrF_f(zeta2);
 698             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 699             felec            = _mm256_mul_ps(qq20,felec);
 700             pmecorrV         = avx256_pmecorrV_f(zeta2);
 701             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 702             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv20,sh_ewald),pmecorrV);
 703             velec            = _mm256_mul_ps(qq20,velec);
 704
 705             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 706
 707             /* Update potential sum for this i atom from the interaction with this j atom. */
 708             velec            = _mm256_and_ps(velec,cutoff_mask);
 709             velec            = _mm256_andnot_ps(dummy_mask,velec);
 710             velecsum         = _mm256_add_ps(velecsum,velec);
 711
 712             fscal            = felec;
 713
 714             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 715
 716             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 717
 718             /* Calculate temporary vectorial force */
 719             tx               = _mm256_mul_ps(fscal,dx20);
 720             ty               = _mm256_mul_ps(fscal,dy20);
 721             tz               = _mm256_mul_ps(fscal,dz20);
 722
 723             /* Update vectorial force */
 724             fix2             = _mm256_add_ps(fix2,tx);
 725             fiy2             = _mm256_add_ps(fiy2,ty);
 726             fiz2             = _mm256_add_ps(fiz2,tz);
 727
 728             fjx0             = _mm256_add_ps(fjx0,tx);
 729             fjy0             = _mm256_add_ps(fjy0,ty);
 730             fjz0             = _mm256_add_ps(fjz0,tz);
 731
 732             }
 733
 734             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 735             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 736             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 737             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 738             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 739             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 740             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 741             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 742
 743             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 744
 745             /* Inner loop uses 351 flops */
 746         }
 747
 748         /* End of innermost loop */
 749
 750         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 751                                                  f+i_coord_offset,fshift+i_shift_offset);
 752
 753         ggid                        = gid[iidx];
 754         /* Update potential energies */
 755         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 756         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 757
 758         /* Increment number of inner iterations */
 759         inneriter                  += j_index_end - j_index_start;
 760
 761         /* Outer loop uses 20 flops */
 762     }
 763
 764     /* Increment number of outer iterations */
 765     outeriter        += nri;
 766
 767     /* Update outer/inner flops */
 768
 769     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*351);
 770 }
 771 /*
 772  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_256_single
 773  * Electrostatics interaction: Ewald
 774  * VdW interaction:            LennardJones
 775  * Geometry:                   Water3-Particle
 776  * Calculate force/pot:        Force
 777  */
 778 void
 779 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_avx_256_single
 780                     (t_nblist                    * gmx_restrict       nlist,
 781                      rvec                        * gmx_restrict          xx,
 782                      rvec                        * gmx_restrict          ff,
 783                      struct t_forcerec           * gmx_restrict          fr,
 784                      t_mdatoms                   * gmx_restrict     mdatoms,
 785                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 786                      t_nrnb                      * gmx_restrict        nrnb)
 787 {
 788     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 789      * just 0 for non-waters.
 790      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 791      * jnr indices corresponding to data put in the four positions in the SIMD register.
 792      */
 793     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 794     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 795     int              jnrA,jnrB,jnrC,jnrD;
 796     int              jnrE,jnrF,jnrG,jnrH;
 797     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 798     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 799     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 800     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 801     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 802     real             rcutoff_scalar;
 803     real             *shiftvec,*fshift,*x,*f;
 804     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 805     real             scratch[4*DIM];
 806     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 807     real *           vdwioffsetptr0;
 808     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 809     real *           vdwioffsetptr1;
 810     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 811     real *           vdwioffsetptr2;
 812     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 813     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 814     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 815     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 816     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 817     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 818     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 819     real             *charge;
 820     int              nvdwtype;
 821     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 822     int              *vdwtype;
 823     real             *vdwparam;
 824     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 825     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 826     __m256i          ewitab;
 827     __m128i          ewitab_lo,ewitab_hi;
 828     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 829     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 830     real             *ewtab;
 831     __m256           dummy_mask,cutoff_mask;
 832     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 833     __m256           one     = _mm256_set1_ps(1.0);
 834     __m256           two     = _mm256_set1_ps(2.0);
 835     x                = xx[0];
 836     f                = ff[0];
 837
 838     nri              = nlist->nri;
 839     iinr             = nlist->iinr;
 840     jindex           = nlist->jindex;
 841     jjnr             = nlist->jjnr;
 842     shiftidx         = nlist->shift;
 843     gid              = nlist->gid;
 844     shiftvec         = fr->shift_vec[0];
 845     fshift           = fr->fshift[0];
 846     facel            = _mm256_set1_ps(fr->ic->epsfac);
 847     charge           = mdatoms->chargeA;
 848     nvdwtype         = fr->ntype;
 849     vdwparam         = fr->nbfp;
 850     vdwtype          = mdatoms->typeA;
 851
 852     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 853     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 854     beta2            = _mm256_mul_ps(beta,beta);
 855     beta3            = _mm256_mul_ps(beta,beta2);
 856
 857     ewtab            = fr->ic->tabq_coul_F;
 858     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 859     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 860
 861     /* Setup water-specific parameters */
 862     inr              = nlist->iinr[0];
 863     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 864     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 865     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 866     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 867
 868     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 869     rcutoff_scalar   = fr->ic->rcoulomb;
 870     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 871     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 872
 873     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 874     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 875
 876     /* Avoid stupid compiler warnings */
 877     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 878     j_coord_offsetA = 0;
 879     j_coord_offsetB = 0;
 880     j_coord_offsetC = 0;
 881     j_coord_offsetD = 0;
 882     j_coord_offsetE = 0;
 883     j_coord_offsetF = 0;
 884     j_coord_offsetG = 0;
 885     j_coord_offsetH = 0;
 886
 887     outeriter        = 0;
 888     inneriter        = 0;
 889
 890     for(iidx=0;iidx<4*DIM;iidx++)
 891     {
 892         scratch[iidx] = 0.0;
 893     }
 894
 895     /* Start outer loop over neighborlists */
 896     for(iidx=0; iidx<nri; iidx++)
 897     {
 898         /* Load shift vector for this list */
 899         i_shift_offset   = DIM*shiftidx[iidx];
 900
 901         /* Load limits for loop over neighbors */
 902         j_index_start    = jindex[iidx];
 903         j_index_end      = jindex[iidx+1];
 904
 905         /* Get outer coordinate index */
 906         inr              = iinr[iidx];
 907         i_coord_offset   = DIM*inr;
 908
 909         /* Load i particle coords and add shift vector */
 910         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 911                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 912
 913         fix0             = _mm256_setzero_ps();
 914         fiy0             = _mm256_setzero_ps();
 915         fiz0             = _mm256_setzero_ps();
 916         fix1             = _mm256_setzero_ps();
 917         fiy1             = _mm256_setzero_ps();
 918         fiz1             = _mm256_setzero_ps();
 919         fix2             = _mm256_setzero_ps();
 920         fiy2             = _mm256_setzero_ps();
 921         fiz2             = _mm256_setzero_ps();
 922
 923         /* Start inner kernel loop */
 924         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 925         {
 926
 927             /* Get j neighbor index, and coordinate index */
 928             jnrA             = jjnr[jidx];
 929             jnrB             = jjnr[jidx+1];
 930             jnrC             = jjnr[jidx+2];
 931             jnrD             = jjnr[jidx+3];
 932             jnrE             = jjnr[jidx+4];
 933             jnrF             = jjnr[jidx+5];
 934             jnrG             = jjnr[jidx+6];
 935             jnrH             = jjnr[jidx+7];
 936             j_coord_offsetA  = DIM*jnrA;
 937             j_coord_offsetB  = DIM*jnrB;
 938             j_coord_offsetC  = DIM*jnrC;
 939             j_coord_offsetD  = DIM*jnrD;
 940             j_coord_offsetE  = DIM*jnrE;
 941             j_coord_offsetF  = DIM*jnrF;
 942             j_coord_offsetG  = DIM*jnrG;
 943             j_coord_offsetH  = DIM*jnrH;
 944
 945             /* load j atom coordinates */
 946             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 947                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 948                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 949                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 950                                                  &jx0,&jy0,&jz0);
 951
 952             /* Calculate displacement vector */
 953             dx00             = _mm256_sub_ps(ix0,jx0);
 954             dy00             = _mm256_sub_ps(iy0,jy0);
 955             dz00             = _mm256_sub_ps(iz0,jz0);
 956             dx10             = _mm256_sub_ps(ix1,jx0);
 957             dy10             = _mm256_sub_ps(iy1,jy0);
 958             dz10             = _mm256_sub_ps(iz1,jz0);
 959             dx20             = _mm256_sub_ps(ix2,jx0);
 960             dy20             = _mm256_sub_ps(iy2,jy0);
 961             dz20             = _mm256_sub_ps(iz2,jz0);
 962
 963             /* Calculate squared distance and things based on it */
 964             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 965             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 966             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 967
 968             rinv00           = avx256_invsqrt_f(rsq00);
 969             rinv10           = avx256_invsqrt_f(rsq10);
 970             rinv20           = avx256_invsqrt_f(rsq20);
 971
 972             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 973             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 974             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 975
 976             /* Load parameters for j particles */
 977             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 978                                                                  charge+jnrC+0,charge+jnrD+0,
 979                                                                  charge+jnrE+0,charge+jnrF+0,
 980                                                                  charge+jnrG+0,charge+jnrH+0);
 981             vdwjidx0A        = 2*vdwtype[jnrA+0];
 982             vdwjidx0B        = 2*vdwtype[jnrB+0];
 983             vdwjidx0C        = 2*vdwtype[jnrC+0];
 984             vdwjidx0D        = 2*vdwtype[jnrD+0];
 985             vdwjidx0E        = 2*vdwtype[jnrE+0];
 986             vdwjidx0F        = 2*vdwtype[jnrF+0];
 987             vdwjidx0G        = 2*vdwtype[jnrG+0];
 988             vdwjidx0H        = 2*vdwtype[jnrH+0];
 989
 990             fjx0             = _mm256_setzero_ps();
 991             fjy0             = _mm256_setzero_ps();
 992             fjz0             = _mm256_setzero_ps();
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 999             {
1000
1001             r00              = _mm256_mul_ps(rsq00,rinv00);
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq00             = _mm256_mul_ps(iq0,jq0);
1005             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1006                                             vdwioffsetptr0+vdwjidx0B,
1007                                             vdwioffsetptr0+vdwjidx0C,
1008                                             vdwioffsetptr0+vdwjidx0D,
1009                                             vdwioffsetptr0+vdwjidx0E,
1010                                             vdwioffsetptr0+vdwjidx0F,
1011                                             vdwioffsetptr0+vdwjidx0G,
1012                                             vdwioffsetptr0+vdwjidx0H,
1013                                             &c6_00,&c12_00);
1014
1015             /* EWALD ELECTROSTATICS */
1016
1017             /* Analytical PME correction */
1018             zeta2            = _mm256_mul_ps(beta2,rsq00);
1019             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1020             pmecorrF         = avx256_pmecorrF_f(zeta2);
1021             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1022             felec            = _mm256_mul_ps(qq00,felec);
1023
1024             /* LENNARD-JONES DISPERSION/REPULSION */
1025
1026             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1027             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1028
1029             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1030
1031             fscal            = _mm256_add_ps(felec,fvdw);
1032
1033             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm256_mul_ps(fscal,dx00);
1037             ty               = _mm256_mul_ps(fscal,dy00);
1038             tz               = _mm256_mul_ps(fscal,dz00);
1039
1040             /* Update vectorial force */
1041             fix0             = _mm256_add_ps(fix0,tx);
1042             fiy0             = _mm256_add_ps(fiy0,ty);
1043             fiz0             = _mm256_add_ps(fiz0,tz);
1044
1045             fjx0             = _mm256_add_ps(fjx0,tx);
1046             fjy0             = _mm256_add_ps(fjy0,ty);
1047             fjz0             = _mm256_add_ps(fjz0,tz);
1048
1049             }
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1056             {
1057
1058             r10              = _mm256_mul_ps(rsq10,rinv10);
1059
1060             /* Compute parameters for interactions between i and j atoms */
1061             qq10             = _mm256_mul_ps(iq1,jq0);
1062
1063             /* EWALD ELECTROSTATICS */
1064
1065             /* Analytical PME correction */
1066             zeta2            = _mm256_mul_ps(beta2,rsq10);
1067             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1068             pmecorrF         = avx256_pmecorrF_f(zeta2);
1069             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1070             felec            = _mm256_mul_ps(qq10,felec);
1071
1072             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1073
1074             fscal            = felec;
1075
1076             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1077
1078             /* Calculate temporary vectorial force */
1079             tx               = _mm256_mul_ps(fscal,dx10);
1080             ty               = _mm256_mul_ps(fscal,dy10);
1081             tz               = _mm256_mul_ps(fscal,dz10);
1082
1083             /* Update vectorial force */
1084             fix1             = _mm256_add_ps(fix1,tx);
1085             fiy1             = _mm256_add_ps(fiy1,ty);
1086             fiz1             = _mm256_add_ps(fiz1,tz);
1087
1088             fjx0             = _mm256_add_ps(fjx0,tx);
1089             fjy0             = _mm256_add_ps(fjy0,ty);
1090             fjz0             = _mm256_add_ps(fjz0,tz);
1091
1092             }
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1099             {
1100
1101             r20              = _mm256_mul_ps(rsq20,rinv20);
1102
1103             /* Compute parameters for interactions between i and j atoms */
1104             qq20             = _mm256_mul_ps(iq2,jq0);
1105
1106             /* EWALD ELECTROSTATICS */
1107
1108             /* Analytical PME correction */
1109             zeta2            = _mm256_mul_ps(beta2,rsq20);
1110             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1111             pmecorrF         = avx256_pmecorrF_f(zeta2);
1112             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1113             felec            = _mm256_mul_ps(qq20,felec);
1114
1115             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1116
1117             fscal            = felec;
1118
1119             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1120
1121             /* Calculate temporary vectorial force */
1122             tx               = _mm256_mul_ps(fscal,dx20);
1123             ty               = _mm256_mul_ps(fscal,dy20);
1124             tz               = _mm256_mul_ps(fscal,dz20);
1125
1126             /* Update vectorial force */
1127             fix2             = _mm256_add_ps(fix2,tx);
1128             fiy2             = _mm256_add_ps(fiy2,ty);
1129             fiz2             = _mm256_add_ps(fiz2,tz);
1130
1131             fjx0             = _mm256_add_ps(fjx0,tx);
1132             fjy0             = _mm256_add_ps(fjy0,ty);
1133             fjz0             = _mm256_add_ps(fjz0,tz);
1134
1135             }
1136
1137             fjptrA             = f+j_coord_offsetA;
1138             fjptrB             = f+j_coord_offsetB;
1139             fjptrC             = f+j_coord_offsetC;
1140             fjptrD             = f+j_coord_offsetD;
1141             fjptrE             = f+j_coord_offsetE;
1142             fjptrF             = f+j_coord_offsetF;
1143             fjptrG             = f+j_coord_offsetG;
1144             fjptrH             = f+j_coord_offsetH;
1145
1146             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1147
1148             /* Inner loop uses 187 flops */
1149         }
1150
1151         if(jidx<j_index_end)
1152         {
1153
1154             /* Get j neighbor index, and coordinate index */
1155             jnrlistA         = jjnr[jidx];
1156             jnrlistB         = jjnr[jidx+1];
1157             jnrlistC         = jjnr[jidx+2];
1158             jnrlistD         = jjnr[jidx+3];
1159             jnrlistE         = jjnr[jidx+4];
1160             jnrlistF         = jjnr[jidx+5];
1161             jnrlistG         = jjnr[jidx+6];
1162             jnrlistH         = jjnr[jidx+7];
1163             /* Sign of each element will be negative for non-real atoms.
1164              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1165              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1166              */
1167             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1168                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1169
1170             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1171             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1172             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1173             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1174             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1175             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1176             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1177             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1178             j_coord_offsetA  = DIM*jnrA;
1179             j_coord_offsetB  = DIM*jnrB;
1180             j_coord_offsetC  = DIM*jnrC;
1181             j_coord_offsetD  = DIM*jnrD;
1182             j_coord_offsetE  = DIM*jnrE;
1183             j_coord_offsetF  = DIM*jnrF;
1184             j_coord_offsetG  = DIM*jnrG;
1185             j_coord_offsetH  = DIM*jnrH;
1186
1187             /* load j atom coordinates */
1188             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1189                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1190                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1191                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1192                                                  &jx0,&jy0,&jz0);
1193
1194             /* Calculate displacement vector */
1195             dx00             = _mm256_sub_ps(ix0,jx0);
1196             dy00             = _mm256_sub_ps(iy0,jy0);
1197             dz00             = _mm256_sub_ps(iz0,jz0);
1198             dx10             = _mm256_sub_ps(ix1,jx0);
1199             dy10             = _mm256_sub_ps(iy1,jy0);
1200             dz10             = _mm256_sub_ps(iz1,jz0);
1201             dx20             = _mm256_sub_ps(ix2,jx0);
1202             dy20             = _mm256_sub_ps(iy2,jy0);
1203             dz20             = _mm256_sub_ps(iz2,jz0);
1204
1205             /* Calculate squared distance and things based on it */
1206             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1207             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1208             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1209
1210             rinv00           = avx256_invsqrt_f(rsq00);
1211             rinv10           = avx256_invsqrt_f(rsq10);
1212             rinv20           = avx256_invsqrt_f(rsq20);
1213
1214             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1215             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1216             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1217
1218             /* Load parameters for j particles */
1219             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1220                                                                  charge+jnrC+0,charge+jnrD+0,
1221                                                                  charge+jnrE+0,charge+jnrF+0,
1222                                                                  charge+jnrG+0,charge+jnrH+0);
1223             vdwjidx0A        = 2*vdwtype[jnrA+0];
1224             vdwjidx0B        = 2*vdwtype[jnrB+0];
1225             vdwjidx0C        = 2*vdwtype[jnrC+0];
1226             vdwjidx0D        = 2*vdwtype[jnrD+0];
1227             vdwjidx0E        = 2*vdwtype[jnrE+0];
1228             vdwjidx0F        = 2*vdwtype[jnrF+0];
1229             vdwjidx0G        = 2*vdwtype[jnrG+0];
1230             vdwjidx0H        = 2*vdwtype[jnrH+0];
1231
1232             fjx0             = _mm256_setzero_ps();
1233             fjy0             = _mm256_setzero_ps();
1234             fjz0             = _mm256_setzero_ps();
1235
1236             /**************************
1237              * CALCULATE INTERACTIONS *
1238              **************************/
1239
1240             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1241             {
1242
1243             r00              = _mm256_mul_ps(rsq00,rinv00);
1244             r00              = _mm256_andnot_ps(dummy_mask,r00);
1245
1246             /* Compute parameters for interactions between i and j atoms */
1247             qq00             = _mm256_mul_ps(iq0,jq0);
1248             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1249                                             vdwioffsetptr0+vdwjidx0B,
1250                                             vdwioffsetptr0+vdwjidx0C,
1251                                             vdwioffsetptr0+vdwjidx0D,
1252                                             vdwioffsetptr0+vdwjidx0E,
1253                                             vdwioffsetptr0+vdwjidx0F,
1254                                             vdwioffsetptr0+vdwjidx0G,
1255                                             vdwioffsetptr0+vdwjidx0H,
1256                                             &c6_00,&c12_00);
1257
1258             /* EWALD ELECTROSTATICS */
1259
1260             /* Analytical PME correction */
1261             zeta2            = _mm256_mul_ps(beta2,rsq00);
1262             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
1263             pmecorrF         = avx256_pmecorrF_f(zeta2);
1264             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1265             felec            = _mm256_mul_ps(qq00,felec);
1266
1267             /* LENNARD-JONES DISPERSION/REPULSION */
1268
1269             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1270             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_00,rinvsix),c6_00),_mm256_mul_ps(rinvsix,rinvsq00));
1271
1272             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1273
1274             fscal            = _mm256_add_ps(felec,fvdw);
1275
1276             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1277
1278             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1279
1280             /* Calculate temporary vectorial force */
1281             tx               = _mm256_mul_ps(fscal,dx00);
1282             ty               = _mm256_mul_ps(fscal,dy00);
1283             tz               = _mm256_mul_ps(fscal,dz00);
1284
1285             /* Update vectorial force */
1286             fix0             = _mm256_add_ps(fix0,tx);
1287             fiy0             = _mm256_add_ps(fiy0,ty);
1288             fiz0             = _mm256_add_ps(fiz0,tz);
1289
1290             fjx0             = _mm256_add_ps(fjx0,tx);
1291             fjy0             = _mm256_add_ps(fjy0,ty);
1292             fjz0             = _mm256_add_ps(fjz0,tz);
1293
1294             }
1295
1296             /**************************
1297              * CALCULATE INTERACTIONS *
1298              **************************/
1299
1300             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1301             {
1302
1303             r10              = _mm256_mul_ps(rsq10,rinv10);
1304             r10              = _mm256_andnot_ps(dummy_mask,r10);
1305
1306             /* Compute parameters for interactions between i and j atoms */
1307             qq10             = _mm256_mul_ps(iq1,jq0);
1308
1309             /* EWALD ELECTROSTATICS */
1310
1311             /* Analytical PME correction */
1312             zeta2            = _mm256_mul_ps(beta2,rsq10);
1313             rinv3            = _mm256_mul_ps(rinvsq10,rinv10);
1314             pmecorrF         = avx256_pmecorrF_f(zeta2);
1315             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1316             felec            = _mm256_mul_ps(qq10,felec);
1317
1318             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1319
1320             fscal            = felec;
1321
1322             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1323
1324             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1325
1326             /* Calculate temporary vectorial force */
1327             tx               = _mm256_mul_ps(fscal,dx10);
1328             ty               = _mm256_mul_ps(fscal,dy10);
1329             tz               = _mm256_mul_ps(fscal,dz10);
1330
1331             /* Update vectorial force */
1332             fix1             = _mm256_add_ps(fix1,tx);
1333             fiy1             = _mm256_add_ps(fiy1,ty);
1334             fiz1             = _mm256_add_ps(fiz1,tz);
1335
1336             fjx0             = _mm256_add_ps(fjx0,tx);
1337             fjy0             = _mm256_add_ps(fjy0,ty);
1338             fjz0             = _mm256_add_ps(fjz0,tz);
1339
1340             }
1341
1342             /**************************
1343              * CALCULATE INTERACTIONS *
1344              **************************/
1345
1346             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1347             {
1348
1349             r20              = _mm256_mul_ps(rsq20,rinv20);
1350             r20              = _mm256_andnot_ps(dummy_mask,r20);
1351
1352             /* Compute parameters for interactions between i and j atoms */
1353             qq20             = _mm256_mul_ps(iq2,jq0);
1354
1355             /* EWALD ELECTROSTATICS */
1356
1357             /* Analytical PME correction */
1358             zeta2            = _mm256_mul_ps(beta2,rsq20);
1359             rinv3            = _mm256_mul_ps(rinvsq20,rinv20);
1360             pmecorrF         = avx256_pmecorrF_f(zeta2);
1361             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
1362             felec            = _mm256_mul_ps(qq20,felec);
1363
1364             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1365
1366             fscal            = felec;
1367
1368             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1369
1370             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1371
1372             /* Calculate temporary vectorial force */
1373             tx               = _mm256_mul_ps(fscal,dx20);
1374             ty               = _mm256_mul_ps(fscal,dy20);
1375             tz               = _mm256_mul_ps(fscal,dz20);
1376
1377             /* Update vectorial force */
1378             fix2             = _mm256_add_ps(fix2,tx);
1379             fiy2             = _mm256_add_ps(fiy2,ty);
1380             fiz2             = _mm256_add_ps(fiz2,tz);
1381
1382             fjx0             = _mm256_add_ps(fjx0,tx);
1383             fjy0             = _mm256_add_ps(fjy0,ty);
1384             fjz0             = _mm256_add_ps(fjz0,tz);
1385
1386             }
1387
1388             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1389             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1390             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1391             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1392             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1393             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1394             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1395             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1396
1397             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1398
1399             /* Inner loop uses 190 flops */
1400         }
1401
1402         /* End of innermost loop */
1403
1404         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1405                                                  f+i_coord_offset,fshift+i_shift_offset);
1406
1407         /* Increment number of inner iterations */
1408         inneriter                  += j_index_end - j_index_start;
1409
1410         /* Outer loop uses 18 flops */
1411     }
1412
1413     /* Increment number of outer iterations */
1414     outeriter        += nri;
1415
1416     /* Update outer/inner flops */
1417
1418     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*190);
1419 }