src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/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_ElecRFCut_VdwLJSw_GeomW3P1_VF_avx_256_single
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwLJSw_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     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 108     real             rswitch_scalar,d_scalar;
 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->epsfac);
 125     charge           = mdatoms->chargeA;
 126     krf              = _mm256_set1_ps(fr->ic->k_rf);
 127     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 128     crf              = _mm256_set1_ps(fr->ic->c_rf);
 129     nvdwtype         = fr->ntype;
 130     vdwparam         = fr->nbfp;
 131     vdwtype          = mdatoms->typeA;
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 136     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 137     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 138     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 139
 140     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 141     rcutoff_scalar   = fr->rcoulomb;
 142     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 143     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 144
 145     rswitch_scalar   = fr->rvdw_switch;
 146     rswitch          = _mm256_set1_ps(rswitch_scalar);
 147     /* Setup switch parameters */
 148     d_scalar         = rcutoff_scalar-rswitch_scalar;
 149     d                = _mm256_set1_ps(d_scalar);
 150     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 151     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 152     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 153     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 154     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 155     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 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             /* REACTION-FIELD ELECTROSTATICS */
 301             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 302             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 303
 304             /* LENNARD-JONES DISPERSION/REPULSION */
 305
 306             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 307             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 308             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 309             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 310             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 311
 312             d                = _mm256_sub_ps(r00,rswitch);
 313             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 314             d2               = _mm256_mul_ps(d,d);
 315             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 316
 317             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 318
 319             /* Evaluate switch function */
 320             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 321             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 322             vvdw             = _mm256_mul_ps(vvdw,sw);
 323             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 324
 325             /* Update potential sum for this i atom from the interaction with this j atom. */
 326             velec            = _mm256_and_ps(velec,cutoff_mask);
 327             velecsum         = _mm256_add_ps(velecsum,velec);
 328             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 329             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 330
 331             fscal            = _mm256_add_ps(felec,fvdw);
 332
 333             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm256_mul_ps(fscal,dx00);
 337             ty               = _mm256_mul_ps(fscal,dy00);
 338             tz               = _mm256_mul_ps(fscal,dz00);
 339
 340             /* Update vectorial force */
 341             fix0             = _mm256_add_ps(fix0,tx);
 342             fiy0             = _mm256_add_ps(fiy0,ty);
 343             fiz0             = _mm256_add_ps(fiz0,tz);
 344
 345             fjx0             = _mm256_add_ps(fjx0,tx);
 346             fjy0             = _mm256_add_ps(fjy0,ty);
 347             fjz0             = _mm256_add_ps(fjz0,tz);
 348
 349             }
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 356             {
 357
 358             /* Compute parameters for interactions between i and j atoms */
 359             qq10             = _mm256_mul_ps(iq1,jq0);
 360
 361             /* REACTION-FIELD ELECTROSTATICS */
 362             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 363             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 364
 365             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velec            = _mm256_and_ps(velec,cutoff_mask);
 369             velecsum         = _mm256_add_ps(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 374
 375             /* Calculate temporary vectorial force */
 376             tx               = _mm256_mul_ps(fscal,dx10);
 377             ty               = _mm256_mul_ps(fscal,dy10);
 378             tz               = _mm256_mul_ps(fscal,dz10);
 379
 380             /* Update vectorial force */
 381             fix1             = _mm256_add_ps(fix1,tx);
 382             fiy1             = _mm256_add_ps(fiy1,ty);
 383             fiz1             = _mm256_add_ps(fiz1,tz);
 384
 385             fjx0             = _mm256_add_ps(fjx0,tx);
 386             fjy0             = _mm256_add_ps(fjy0,ty);
 387             fjz0             = _mm256_add_ps(fjz0,tz);
 388
 389             }
 390
 391             /**************************
 392              * CALCULATE INTERACTIONS *
 393              **************************/
 394
 395             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 396             {
 397
 398             /* Compute parameters for interactions between i and j atoms */
 399             qq20             = _mm256_mul_ps(iq2,jq0);
 400
 401             /* REACTION-FIELD ELECTROSTATICS */
 402             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 403             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 404
 405             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velec            = _mm256_and_ps(velec,cutoff_mask);
 409             velecsum         = _mm256_add_ps(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 414
 415             /* Calculate temporary vectorial force */
 416             tx               = _mm256_mul_ps(fscal,dx20);
 417             ty               = _mm256_mul_ps(fscal,dy20);
 418             tz               = _mm256_mul_ps(fscal,dz20);
 419
 420             /* Update vectorial force */
 421             fix2             = _mm256_add_ps(fix2,tx);
 422             fiy2             = _mm256_add_ps(fiy2,ty);
 423             fiz2             = _mm256_add_ps(fiz2,tz);
 424
 425             fjx0             = _mm256_add_ps(fjx0,tx);
 426             fjy0             = _mm256_add_ps(fjy0,ty);
 427             fjz0             = _mm256_add_ps(fjz0,tz);
 428
 429             }
 430
 431             fjptrA             = f+j_coord_offsetA;
 432             fjptrB             = f+j_coord_offsetB;
 433             fjptrC             = f+j_coord_offsetC;
 434             fjptrD             = f+j_coord_offsetD;
 435             fjptrE             = f+j_coord_offsetE;
 436             fjptrF             = f+j_coord_offsetF;
 437             fjptrG             = f+j_coord_offsetG;
 438             fjptrH             = f+j_coord_offsetH;
 439
 440             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 441
 442             /* Inner loop uses 145 flops */
 443         }
 444
 445         if(jidx<j_index_end)
 446         {
 447
 448             /* Get j neighbor index, and coordinate index */
 449             jnrlistA         = jjnr[jidx];
 450             jnrlistB         = jjnr[jidx+1];
 451             jnrlistC         = jjnr[jidx+2];
 452             jnrlistD         = jjnr[jidx+3];
 453             jnrlistE         = jjnr[jidx+4];
 454             jnrlistF         = jjnr[jidx+5];
 455             jnrlistG         = jjnr[jidx+6];
 456             jnrlistH         = jjnr[jidx+7];
 457             /* Sign of each element will be negative for non-real atoms.
 458              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 459              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 460              */
 461             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 462                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 463
 464             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 465             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 466             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 467             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 468             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 469             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 470             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 471             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 472             j_coord_offsetA  = DIM*jnrA;
 473             j_coord_offsetB  = DIM*jnrB;
 474             j_coord_offsetC  = DIM*jnrC;
 475             j_coord_offsetD  = DIM*jnrD;
 476             j_coord_offsetE  = DIM*jnrE;
 477             j_coord_offsetF  = DIM*jnrF;
 478             j_coord_offsetG  = DIM*jnrG;
 479             j_coord_offsetH  = DIM*jnrH;
 480
 481             /* load j atom coordinates */
 482             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 483                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 484                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 485                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 486                                                  &jx0,&jy0,&jz0);
 487
 488             /* Calculate displacement vector */
 489             dx00             = _mm256_sub_ps(ix0,jx0);
 490             dy00             = _mm256_sub_ps(iy0,jy0);
 491             dz00             = _mm256_sub_ps(iz0,jz0);
 492             dx10             = _mm256_sub_ps(ix1,jx0);
 493             dy10             = _mm256_sub_ps(iy1,jy0);
 494             dz10             = _mm256_sub_ps(iz1,jz0);
 495             dx20             = _mm256_sub_ps(ix2,jx0);
 496             dy20             = _mm256_sub_ps(iy2,jy0);
 497             dz20             = _mm256_sub_ps(iz2,jz0);
 498
 499             /* Calculate squared distance and things based on it */
 500             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 501             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 502             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 503
 504             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 505             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 506             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 507
 508             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 509             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 510             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 511
 512             /* Load parameters for j particles */
 513             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 514                                                                  charge+jnrC+0,charge+jnrD+0,
 515                                                                  charge+jnrE+0,charge+jnrF+0,
 516                                                                  charge+jnrG+0,charge+jnrH+0);
 517             vdwjidx0A        = 2*vdwtype[jnrA+0];
 518             vdwjidx0B        = 2*vdwtype[jnrB+0];
 519             vdwjidx0C        = 2*vdwtype[jnrC+0];
 520             vdwjidx0D        = 2*vdwtype[jnrD+0];
 521             vdwjidx0E        = 2*vdwtype[jnrE+0];
 522             vdwjidx0F        = 2*vdwtype[jnrF+0];
 523             vdwjidx0G        = 2*vdwtype[jnrG+0];
 524             vdwjidx0H        = 2*vdwtype[jnrH+0];
 525
 526             fjx0             = _mm256_setzero_ps();
 527             fjy0             = _mm256_setzero_ps();
 528             fjz0             = _mm256_setzero_ps();
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 535             {
 536
 537             r00              = _mm256_mul_ps(rsq00,rinv00);
 538             r00              = _mm256_andnot_ps(dummy_mask,r00);
 539
 540             /* Compute parameters for interactions between i and j atoms */
 541             qq00             = _mm256_mul_ps(iq0,jq0);
 542             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 543                                             vdwioffsetptr0+vdwjidx0B,
 544                                             vdwioffsetptr0+vdwjidx0C,
 545                                             vdwioffsetptr0+vdwjidx0D,
 546                                             vdwioffsetptr0+vdwjidx0E,
 547                                             vdwioffsetptr0+vdwjidx0F,
 548                                             vdwioffsetptr0+vdwjidx0G,
 549                                             vdwioffsetptr0+vdwjidx0H,
 550                                             &c6_00,&c12_00);
 551
 552             /* REACTION-FIELD ELECTROSTATICS */
 553             velec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_add_ps(rinv00,_mm256_mul_ps(krf,rsq00)),crf));
 554             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 555
 556             /* LENNARD-JONES DISPERSION/REPULSION */
 557
 558             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 559             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 560             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 561             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 562             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 563
 564             d                = _mm256_sub_ps(r00,rswitch);
 565             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 566             d2               = _mm256_mul_ps(d,d);
 567             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 568
 569             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 570
 571             /* Evaluate switch function */
 572             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 573             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 574             vvdw             = _mm256_mul_ps(vvdw,sw);
 575             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm256_and_ps(velec,cutoff_mask);
 579             velec            = _mm256_andnot_ps(dummy_mask,velec);
 580             velecsum         = _mm256_add_ps(velecsum,velec);
 581             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 582             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 583             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 584
 585             fscal            = _mm256_add_ps(felec,fvdw);
 586
 587             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 588
 589             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 590
 591             /* Calculate temporary vectorial force */
 592             tx               = _mm256_mul_ps(fscal,dx00);
 593             ty               = _mm256_mul_ps(fscal,dy00);
 594             tz               = _mm256_mul_ps(fscal,dz00);
 595
 596             /* Update vectorial force */
 597             fix0             = _mm256_add_ps(fix0,tx);
 598             fiy0             = _mm256_add_ps(fiy0,ty);
 599             fiz0             = _mm256_add_ps(fiz0,tz);
 600
 601             fjx0             = _mm256_add_ps(fjx0,tx);
 602             fjy0             = _mm256_add_ps(fjy0,ty);
 603             fjz0             = _mm256_add_ps(fjz0,tz);
 604
 605             }
 606
 607             /**************************
 608              * CALCULATE INTERACTIONS *
 609              **************************/
 610
 611             if (gmx_mm256_any_lt(rsq10,rcutoff2))
 612             {
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             qq10             = _mm256_mul_ps(iq1,jq0);
 616
 617             /* REACTION-FIELD ELECTROSTATICS */
 618             velec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_add_ps(rinv10,_mm256_mul_ps(krf,rsq10)),crf));
 619             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
 620
 621             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
 622
 623             /* Update potential sum for this i atom from the interaction with this j atom. */
 624             velec            = _mm256_and_ps(velec,cutoff_mask);
 625             velec            = _mm256_andnot_ps(dummy_mask,velec);
 626             velecsum         = _mm256_add_ps(velecsum,velec);
 627
 628             fscal            = felec;
 629
 630             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 631
 632             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 633
 634             /* Calculate temporary vectorial force */
 635             tx               = _mm256_mul_ps(fscal,dx10);
 636             ty               = _mm256_mul_ps(fscal,dy10);
 637             tz               = _mm256_mul_ps(fscal,dz10);
 638
 639             /* Update vectorial force */
 640             fix1             = _mm256_add_ps(fix1,tx);
 641             fiy1             = _mm256_add_ps(fiy1,ty);
 642             fiz1             = _mm256_add_ps(fiz1,tz);
 643
 644             fjx0             = _mm256_add_ps(fjx0,tx);
 645             fjy0             = _mm256_add_ps(fjy0,ty);
 646             fjz0             = _mm256_add_ps(fjz0,tz);
 647
 648             }
 649
 650             /**************************
 651              * CALCULATE INTERACTIONS *
 652              **************************/
 653
 654             if (gmx_mm256_any_lt(rsq20,rcutoff2))
 655             {
 656
 657             /* Compute parameters for interactions between i and j atoms */
 658             qq20             = _mm256_mul_ps(iq2,jq0);
 659
 660             /* REACTION-FIELD ELECTROSTATICS */
 661             velec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_add_ps(rinv20,_mm256_mul_ps(krf,rsq20)),crf));
 662             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
 663
 664             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
 665
 666             /* Update potential sum for this i atom from the interaction with this j atom. */
 667             velec            = _mm256_and_ps(velec,cutoff_mask);
 668             velec            = _mm256_andnot_ps(dummy_mask,velec);
 669             velecsum         = _mm256_add_ps(velecsum,velec);
 670
 671             fscal            = felec;
 672
 673             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 674
 675             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 676
 677             /* Calculate temporary vectorial force */
 678             tx               = _mm256_mul_ps(fscal,dx20);
 679             ty               = _mm256_mul_ps(fscal,dy20);
 680             tz               = _mm256_mul_ps(fscal,dz20);
 681
 682             /* Update vectorial force */
 683             fix2             = _mm256_add_ps(fix2,tx);
 684             fiy2             = _mm256_add_ps(fiy2,ty);
 685             fiz2             = _mm256_add_ps(fiz2,tz);
 686
 687             fjx0             = _mm256_add_ps(fjx0,tx);
 688             fjy0             = _mm256_add_ps(fjy0,ty);
 689             fjz0             = _mm256_add_ps(fjz0,tz);
 690
 691             }
 692
 693             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 694             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 695             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 696             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 697             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 698             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 699             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 700             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 701
 702             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
 703
 704             /* Inner loop uses 146 flops */
 705         }
 706
 707         /* End of innermost loop */
 708
 709         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 710                                                  f+i_coord_offset,fshift+i_shift_offset);
 711
 712         ggid                        = gid[iidx];
 713         /* Update potential energies */
 714         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 715         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 716
 717         /* Increment number of inner iterations */
 718         inneriter                  += j_index_end - j_index_start;
 719
 720         /* Outer loop uses 20 flops */
 721     }
 722
 723     /* Increment number of outer iterations */
 724     outeriter        += nri;
 725
 726     /* Update outer/inner flops */
 727
 728     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*146);
 729 }
 730 /*
 731  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_single
 732  * Electrostatics interaction: ReactionField
 733  * VdW interaction:            LennardJones
 734  * Geometry:                   Water3-Particle
 735  * Calculate force/pot:        Force
 736  */
 737 void
 738 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_avx_256_single
 739                     (t_nblist                    * gmx_restrict       nlist,
 740                      rvec                        * gmx_restrict          xx,
 741                      rvec                        * gmx_restrict          ff,
 742                      t_forcerec                  * gmx_restrict          fr,
 743                      t_mdatoms                   * gmx_restrict     mdatoms,
 744                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 745                      t_nrnb                      * gmx_restrict        nrnb)
 746 {
 747     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 748      * just 0 for non-waters.
 749      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 750      * jnr indices corresponding to data put in the four positions in the SIMD register.
 751      */
 752     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 753     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 754     int              jnrA,jnrB,jnrC,jnrD;
 755     int              jnrE,jnrF,jnrG,jnrH;
 756     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 757     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 758     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 759     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 760     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 761     real             rcutoff_scalar;
 762     real             *shiftvec,*fshift,*x,*f;
 763     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 764     real             scratch[4*DIM];
 765     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 766     real *           vdwioffsetptr0;
 767     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 768     real *           vdwioffsetptr1;
 769     __m256           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 770     real *           vdwioffsetptr2;
 771     __m256           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 772     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 773     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 774     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 775     __m256           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 776     __m256           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 777     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 778     real             *charge;
 779     int              nvdwtype;
 780     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 781     int              *vdwtype;
 782     real             *vdwparam;
 783     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 784     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 785     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 786     real             rswitch_scalar,d_scalar;
 787     __m256           dummy_mask,cutoff_mask;
 788     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 789     __m256           one     = _mm256_set1_ps(1.0);
 790     __m256           two     = _mm256_set1_ps(2.0);
 791     x                = xx[0];
 792     f                = ff[0];
 793
 794     nri              = nlist->nri;
 795     iinr             = nlist->iinr;
 796     jindex           = nlist->jindex;
 797     jjnr             = nlist->jjnr;
 798     shiftidx         = nlist->shift;
 799     gid              = nlist->gid;
 800     shiftvec         = fr->shift_vec[0];
 801     fshift           = fr->fshift[0];
 802     facel            = _mm256_set1_ps(fr->epsfac);
 803     charge           = mdatoms->chargeA;
 804     krf              = _mm256_set1_ps(fr->ic->k_rf);
 805     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 806     crf              = _mm256_set1_ps(fr->ic->c_rf);
 807     nvdwtype         = fr->ntype;
 808     vdwparam         = fr->nbfp;
 809     vdwtype          = mdatoms->typeA;
 810
 811     /* Setup water-specific parameters */
 812     inr              = nlist->iinr[0];
 813     iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 814     iq1              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+1]));
 815     iq2              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+2]));
 816     vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 817
 818     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 819     rcutoff_scalar   = fr->rcoulomb;
 820     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 821     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 822
 823     rswitch_scalar   = fr->rvdw_switch;
 824     rswitch          = _mm256_set1_ps(rswitch_scalar);
 825     /* Setup switch parameters */
 826     d_scalar         = rcutoff_scalar-rswitch_scalar;
 827     d                = _mm256_set1_ps(d_scalar);
 828     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 829     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 830     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 831     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 832     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 833     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 834
 835     /* Avoid stupid compiler warnings */
 836     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 837     j_coord_offsetA = 0;
 838     j_coord_offsetB = 0;
 839     j_coord_offsetC = 0;
 840     j_coord_offsetD = 0;
 841     j_coord_offsetE = 0;
 842     j_coord_offsetF = 0;
 843     j_coord_offsetG = 0;
 844     j_coord_offsetH = 0;
 845
 846     outeriter        = 0;
 847     inneriter        = 0;
 848
 849     for(iidx=0;iidx<4*DIM;iidx++)
 850     {
 851         scratch[iidx] = 0.0;
 852     }
 853
 854     /* Start outer loop over neighborlists */
 855     for(iidx=0; iidx<nri; iidx++)
 856     {
 857         /* Load shift vector for this list */
 858         i_shift_offset   = DIM*shiftidx[iidx];
 859
 860         /* Load limits for loop over neighbors */
 861         j_index_start    = jindex[iidx];
 862         j_index_end      = jindex[iidx+1];
 863
 864         /* Get outer coordinate index */
 865         inr              = iinr[iidx];
 866         i_coord_offset   = DIM*inr;
 867
 868         /* Load i particle coords and add shift vector */
 869         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 870                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 871
 872         fix0             = _mm256_setzero_ps();
 873         fiy0             = _mm256_setzero_ps();
 874         fiz0             = _mm256_setzero_ps();
 875         fix1             = _mm256_setzero_ps();
 876         fiy1             = _mm256_setzero_ps();
 877         fiz1             = _mm256_setzero_ps();
 878         fix2             = _mm256_setzero_ps();
 879         fiy2             = _mm256_setzero_ps();
 880         fiz2             = _mm256_setzero_ps();
 881
 882         /* Start inner kernel loop */
 883         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 884         {
 885
 886             /* Get j neighbor index, and coordinate index */
 887             jnrA             = jjnr[jidx];
 888             jnrB             = jjnr[jidx+1];
 889             jnrC             = jjnr[jidx+2];
 890             jnrD             = jjnr[jidx+3];
 891             jnrE             = jjnr[jidx+4];
 892             jnrF             = jjnr[jidx+5];
 893             jnrG             = jjnr[jidx+6];
 894             jnrH             = jjnr[jidx+7];
 895             j_coord_offsetA  = DIM*jnrA;
 896             j_coord_offsetB  = DIM*jnrB;
 897             j_coord_offsetC  = DIM*jnrC;
 898             j_coord_offsetD  = DIM*jnrD;
 899             j_coord_offsetE  = DIM*jnrE;
 900             j_coord_offsetF  = DIM*jnrF;
 901             j_coord_offsetG  = DIM*jnrG;
 902             j_coord_offsetH  = DIM*jnrH;
 903
 904             /* load j atom coordinates */
 905             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 906                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 907                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 908                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 909                                                  &jx0,&jy0,&jz0);
 910
 911             /* Calculate displacement vector */
 912             dx00             = _mm256_sub_ps(ix0,jx0);
 913             dy00             = _mm256_sub_ps(iy0,jy0);
 914             dz00             = _mm256_sub_ps(iz0,jz0);
 915             dx10             = _mm256_sub_ps(ix1,jx0);
 916             dy10             = _mm256_sub_ps(iy1,jy0);
 917             dz10             = _mm256_sub_ps(iz1,jz0);
 918             dx20             = _mm256_sub_ps(ix2,jx0);
 919             dy20             = _mm256_sub_ps(iy2,jy0);
 920             dz20             = _mm256_sub_ps(iz2,jz0);
 921
 922             /* Calculate squared distance and things based on it */
 923             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 924             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
 925             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
 926
 927             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 928             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
 929             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
 930
 931             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 932             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
 933             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
 934
 935             /* Load parameters for j particles */
 936             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 937                                                                  charge+jnrC+0,charge+jnrD+0,
 938                                                                  charge+jnrE+0,charge+jnrF+0,
 939                                                                  charge+jnrG+0,charge+jnrH+0);
 940             vdwjidx0A        = 2*vdwtype[jnrA+0];
 941             vdwjidx0B        = 2*vdwtype[jnrB+0];
 942             vdwjidx0C        = 2*vdwtype[jnrC+0];
 943             vdwjidx0D        = 2*vdwtype[jnrD+0];
 944             vdwjidx0E        = 2*vdwtype[jnrE+0];
 945             vdwjidx0F        = 2*vdwtype[jnrF+0];
 946             vdwjidx0G        = 2*vdwtype[jnrG+0];
 947             vdwjidx0H        = 2*vdwtype[jnrH+0];
 948
 949             fjx0             = _mm256_setzero_ps();
 950             fjy0             = _mm256_setzero_ps();
 951             fjz0             = _mm256_setzero_ps();
 952
 953             /**************************
 954              * CALCULATE INTERACTIONS *
 955              **************************/
 956
 957             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 958             {
 959
 960             r00              = _mm256_mul_ps(rsq00,rinv00);
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             qq00             = _mm256_mul_ps(iq0,jq0);
 964             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 965                                             vdwioffsetptr0+vdwjidx0B,
 966                                             vdwioffsetptr0+vdwjidx0C,
 967                                             vdwioffsetptr0+vdwjidx0D,
 968                                             vdwioffsetptr0+vdwjidx0E,
 969                                             vdwioffsetptr0+vdwjidx0F,
 970                                             vdwioffsetptr0+vdwjidx0G,
 971                                             vdwioffsetptr0+vdwjidx0H,
 972                                             &c6_00,&c12_00);
 973
 974             /* REACTION-FIELD ELECTROSTATICS */
 975             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
 976
 977             /* LENNARD-JONES DISPERSION/REPULSION */
 978
 979             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 980             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 981             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 982             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 983             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 984
 985             d                = _mm256_sub_ps(r00,rswitch);
 986             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 987             d2               = _mm256_mul_ps(d,d);
 988             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 989
 990             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 991
 992             /* Evaluate switch function */
 993             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 994             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 995             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 996
 997             fscal            = _mm256_add_ps(felec,fvdw);
 998
 999             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1000
1001             /* Calculate temporary vectorial force */
1002             tx               = _mm256_mul_ps(fscal,dx00);
1003             ty               = _mm256_mul_ps(fscal,dy00);
1004             tz               = _mm256_mul_ps(fscal,dz00);
1005
1006             /* Update vectorial force */
1007             fix0             = _mm256_add_ps(fix0,tx);
1008             fiy0             = _mm256_add_ps(fiy0,ty);
1009             fiz0             = _mm256_add_ps(fiz0,tz);
1010
1011             fjx0             = _mm256_add_ps(fjx0,tx);
1012             fjy0             = _mm256_add_ps(fjy0,ty);
1013             fjz0             = _mm256_add_ps(fjz0,tz);
1014
1015             }
1016
1017             /**************************
1018              * CALCULATE INTERACTIONS *
1019              **************************/
1020
1021             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1022             {
1023
1024             /* Compute parameters for interactions between i and j atoms */
1025             qq10             = _mm256_mul_ps(iq1,jq0);
1026
1027             /* REACTION-FIELD ELECTROSTATICS */
1028             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1029
1030             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1035
1036             /* Calculate temporary vectorial force */
1037             tx               = _mm256_mul_ps(fscal,dx10);
1038             ty               = _mm256_mul_ps(fscal,dy10);
1039             tz               = _mm256_mul_ps(fscal,dz10);
1040
1041             /* Update vectorial force */
1042             fix1             = _mm256_add_ps(fix1,tx);
1043             fiy1             = _mm256_add_ps(fiy1,ty);
1044             fiz1             = _mm256_add_ps(fiz1,tz);
1045
1046             fjx0             = _mm256_add_ps(fjx0,tx);
1047             fjy0             = _mm256_add_ps(fjy0,ty);
1048             fjz0             = _mm256_add_ps(fjz0,tz);
1049
1050             }
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1057             {
1058
1059             /* Compute parameters for interactions between i and j atoms */
1060             qq20             = _mm256_mul_ps(iq2,jq0);
1061
1062             /* REACTION-FIELD ELECTROSTATICS */
1063             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1064
1065             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1066
1067             fscal            = felec;
1068
1069             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1070
1071             /* Calculate temporary vectorial force */
1072             tx               = _mm256_mul_ps(fscal,dx20);
1073             ty               = _mm256_mul_ps(fscal,dy20);
1074             tz               = _mm256_mul_ps(fscal,dz20);
1075
1076             /* Update vectorial force */
1077             fix2             = _mm256_add_ps(fix2,tx);
1078             fiy2             = _mm256_add_ps(fiy2,ty);
1079             fiz2             = _mm256_add_ps(fiz2,tz);
1080
1081             fjx0             = _mm256_add_ps(fjx0,tx);
1082             fjy0             = _mm256_add_ps(fjy0,ty);
1083             fjz0             = _mm256_add_ps(fjz0,tz);
1084
1085             }
1086
1087             fjptrA             = f+j_coord_offsetA;
1088             fjptrB             = f+j_coord_offsetB;
1089             fjptrC             = f+j_coord_offsetC;
1090             fjptrD             = f+j_coord_offsetD;
1091             fjptrE             = f+j_coord_offsetE;
1092             fjptrF             = f+j_coord_offsetF;
1093             fjptrG             = f+j_coord_offsetG;
1094             fjptrH             = f+j_coord_offsetH;
1095
1096             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1097
1098             /* Inner loop uses 124 flops */
1099         }
1100
1101         if(jidx<j_index_end)
1102         {
1103
1104             /* Get j neighbor index, and coordinate index */
1105             jnrlistA         = jjnr[jidx];
1106             jnrlistB         = jjnr[jidx+1];
1107             jnrlistC         = jjnr[jidx+2];
1108             jnrlistD         = jjnr[jidx+3];
1109             jnrlistE         = jjnr[jidx+4];
1110             jnrlistF         = jjnr[jidx+5];
1111             jnrlistG         = jjnr[jidx+6];
1112             jnrlistH         = jjnr[jidx+7];
1113             /* Sign of each element will be negative for non-real atoms.
1114              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1115              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1116              */
1117             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
1118                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
1119
1120             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1121             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1122             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1123             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1124             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
1125             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
1126             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
1127             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
1128             j_coord_offsetA  = DIM*jnrA;
1129             j_coord_offsetB  = DIM*jnrB;
1130             j_coord_offsetC  = DIM*jnrC;
1131             j_coord_offsetD  = DIM*jnrD;
1132             j_coord_offsetE  = DIM*jnrE;
1133             j_coord_offsetF  = DIM*jnrF;
1134             j_coord_offsetG  = DIM*jnrG;
1135             j_coord_offsetH  = DIM*jnrH;
1136
1137             /* load j atom coordinates */
1138             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1139                                                  x+j_coord_offsetC,x+j_coord_offsetD,
1140                                                  x+j_coord_offsetE,x+j_coord_offsetF,
1141                                                  x+j_coord_offsetG,x+j_coord_offsetH,
1142                                                  &jx0,&jy0,&jz0);
1143
1144             /* Calculate displacement vector */
1145             dx00             = _mm256_sub_ps(ix0,jx0);
1146             dy00             = _mm256_sub_ps(iy0,jy0);
1147             dz00             = _mm256_sub_ps(iz0,jz0);
1148             dx10             = _mm256_sub_ps(ix1,jx0);
1149             dy10             = _mm256_sub_ps(iy1,jy0);
1150             dz10             = _mm256_sub_ps(iz1,jz0);
1151             dx20             = _mm256_sub_ps(ix2,jx0);
1152             dy20             = _mm256_sub_ps(iy2,jy0);
1153             dz20             = _mm256_sub_ps(iz2,jz0);
1154
1155             /* Calculate squared distance and things based on it */
1156             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
1157             rsq10            = gmx_mm256_calc_rsq_ps(dx10,dy10,dz10);
1158             rsq20            = gmx_mm256_calc_rsq_ps(dx20,dy20,dz20);
1159
1160             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
1161             rinv10           = gmx_mm256_invsqrt_ps(rsq10);
1162             rinv20           = gmx_mm256_invsqrt_ps(rsq20);
1163
1164             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
1165             rinvsq10         = _mm256_mul_ps(rinv10,rinv10);
1166             rinvsq20         = _mm256_mul_ps(rinv20,rinv20);
1167
1168             /* Load parameters for j particles */
1169             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1170                                                                  charge+jnrC+0,charge+jnrD+0,
1171                                                                  charge+jnrE+0,charge+jnrF+0,
1172                                                                  charge+jnrG+0,charge+jnrH+0);
1173             vdwjidx0A        = 2*vdwtype[jnrA+0];
1174             vdwjidx0B        = 2*vdwtype[jnrB+0];
1175             vdwjidx0C        = 2*vdwtype[jnrC+0];
1176             vdwjidx0D        = 2*vdwtype[jnrD+0];
1177             vdwjidx0E        = 2*vdwtype[jnrE+0];
1178             vdwjidx0F        = 2*vdwtype[jnrF+0];
1179             vdwjidx0G        = 2*vdwtype[jnrG+0];
1180             vdwjidx0H        = 2*vdwtype[jnrH+0];
1181
1182             fjx0             = _mm256_setzero_ps();
1183             fjy0             = _mm256_setzero_ps();
1184             fjz0             = _mm256_setzero_ps();
1185
1186             /**************************
1187              * CALCULATE INTERACTIONS *
1188              **************************/
1189
1190             if (gmx_mm256_any_lt(rsq00,rcutoff2))
1191             {
1192
1193             r00              = _mm256_mul_ps(rsq00,rinv00);
1194             r00              = _mm256_andnot_ps(dummy_mask,r00);
1195
1196             /* Compute parameters for interactions between i and j atoms */
1197             qq00             = _mm256_mul_ps(iq0,jq0);
1198             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
1199                                             vdwioffsetptr0+vdwjidx0B,
1200                                             vdwioffsetptr0+vdwjidx0C,
1201                                             vdwioffsetptr0+vdwjidx0D,
1202                                             vdwioffsetptr0+vdwjidx0E,
1203                                             vdwioffsetptr0+vdwjidx0F,
1204                                             vdwioffsetptr0+vdwjidx0G,
1205                                             vdwioffsetptr0+vdwjidx0H,
1206                                             &c6_00,&c12_00);
1207
1208             /* REACTION-FIELD ELECTROSTATICS */
1209             felec            = _mm256_mul_ps(qq00,_mm256_sub_ps(_mm256_mul_ps(rinv00,rinvsq00),krf2));
1210
1211             /* LENNARD-JONES DISPERSION/REPULSION */
1212
1213             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1214             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
1215             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
1216             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
1217             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
1218
1219             d                = _mm256_sub_ps(r00,rswitch);
1220             d                = _mm256_max_ps(d,_mm256_setzero_ps());
1221             d2               = _mm256_mul_ps(d,d);
1222             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
1223
1224             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
1225
1226             /* Evaluate switch function */
1227             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1228             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
1229             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
1230
1231             fscal            = _mm256_add_ps(felec,fvdw);
1232
1233             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1234
1235             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1236
1237             /* Calculate temporary vectorial force */
1238             tx               = _mm256_mul_ps(fscal,dx00);
1239             ty               = _mm256_mul_ps(fscal,dy00);
1240             tz               = _mm256_mul_ps(fscal,dz00);
1241
1242             /* Update vectorial force */
1243             fix0             = _mm256_add_ps(fix0,tx);
1244             fiy0             = _mm256_add_ps(fiy0,ty);
1245             fiz0             = _mm256_add_ps(fiz0,tz);
1246
1247             fjx0             = _mm256_add_ps(fjx0,tx);
1248             fjy0             = _mm256_add_ps(fjy0,ty);
1249             fjz0             = _mm256_add_ps(fjz0,tz);
1250
1251             }
1252
1253             /**************************
1254              * CALCULATE INTERACTIONS *
1255              **************************/
1256
1257             if (gmx_mm256_any_lt(rsq10,rcutoff2))
1258             {
1259
1260             /* Compute parameters for interactions between i and j atoms */
1261             qq10             = _mm256_mul_ps(iq1,jq0);
1262
1263             /* REACTION-FIELD ELECTROSTATICS */
1264             felec            = _mm256_mul_ps(qq10,_mm256_sub_ps(_mm256_mul_ps(rinv10,rinvsq10),krf2));
1265
1266             cutoff_mask      = _mm256_cmp_ps(rsq10,rcutoff2,_CMP_LT_OQ);
1267
1268             fscal            = felec;
1269
1270             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1271
1272             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1273
1274             /* Calculate temporary vectorial force */
1275             tx               = _mm256_mul_ps(fscal,dx10);
1276             ty               = _mm256_mul_ps(fscal,dy10);
1277             tz               = _mm256_mul_ps(fscal,dz10);
1278
1279             /* Update vectorial force */
1280             fix1             = _mm256_add_ps(fix1,tx);
1281             fiy1             = _mm256_add_ps(fiy1,ty);
1282             fiz1             = _mm256_add_ps(fiz1,tz);
1283
1284             fjx0             = _mm256_add_ps(fjx0,tx);
1285             fjy0             = _mm256_add_ps(fjy0,ty);
1286             fjz0             = _mm256_add_ps(fjz0,tz);
1287
1288             }
1289
1290             /**************************
1291              * CALCULATE INTERACTIONS *
1292              **************************/
1293
1294             if (gmx_mm256_any_lt(rsq20,rcutoff2))
1295             {
1296
1297             /* Compute parameters for interactions between i and j atoms */
1298             qq20             = _mm256_mul_ps(iq2,jq0);
1299
1300             /* REACTION-FIELD ELECTROSTATICS */
1301             felec            = _mm256_mul_ps(qq20,_mm256_sub_ps(_mm256_mul_ps(rinv20,rinvsq20),krf2));
1302
1303             cutoff_mask      = _mm256_cmp_ps(rsq20,rcutoff2,_CMP_LT_OQ);
1304
1305             fscal            = felec;
1306
1307             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1308
1309             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1310
1311             /* Calculate temporary vectorial force */
1312             tx               = _mm256_mul_ps(fscal,dx20);
1313             ty               = _mm256_mul_ps(fscal,dy20);
1314             tz               = _mm256_mul_ps(fscal,dz20);
1315
1316             /* Update vectorial force */
1317             fix2             = _mm256_add_ps(fix2,tx);
1318             fiy2             = _mm256_add_ps(fiy2,ty);
1319             fiz2             = _mm256_add_ps(fiz2,tz);
1320
1321             fjx0             = _mm256_add_ps(fjx0,tx);
1322             fjy0             = _mm256_add_ps(fjy0,ty);
1323             fjz0             = _mm256_add_ps(fjz0,tz);
1324
1325             }
1326
1327             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1328             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1329             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1330             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1331             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1332             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1333             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1334             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1335
1336             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1337
1338             /* Inner loop uses 125 flops */
1339         }
1340
1341         /* End of innermost loop */
1342
1343         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1344                                                  f+i_coord_offset,fshift+i_shift_offset);
1345
1346         /* Increment number of inner iterations */
1347         inneriter                  += j_index_end - j_index_start;
1348
1349         /* Outer loop uses 18 flops */
1350     }
1351
1352     /* Increment number of outer iterations */
1353     outeriter        += nri;
1354
1355     /* Update outer/inner flops */
1356
1357     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*125);
1358 }