src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_sse4_1_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 sse4_1_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_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse4_1_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_sse4_1_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 refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128i          ewitab;
 102     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 103     real             *ewtab;
 104     __m128           dummy_mask,cutoff_mask;
 105     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 106     __m128           one     = _mm_set1_ps(1.0);
 107     __m128           two     = _mm_set1_ps(2.0);
 108     x                = xx[0];
 109     f                = ff[0];
 110
 111     nri              = nlist->nri;
 112     iinr             = nlist->iinr;
 113     jindex           = nlist->jindex;
 114     jjnr             = nlist->jjnr;
 115     shiftidx         = nlist->shift;
 116     gid              = nlist->gid;
 117     shiftvec         = fr->shift_vec[0];
 118     fshift           = fr->fshift[0];
 119     facel            = _mm_set1_ps(fr->ic->epsfac);
 120     charge           = mdatoms->chargeA;
 121     nvdwtype         = fr->ntype;
 122     vdwparam         = fr->nbfp;
 123     vdwtype          = mdatoms->typeA;
 124
 125     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 126     ewtab            = fr->ic->tabq_coul_FDV0;
 127     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 128     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 129
 130     /* Setup water-specific parameters */
 131     inr              = nlist->iinr[0];
 132     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 133     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 134     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 135     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 136
 137     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 138     rcutoff_scalar   = fr->ic->rcoulomb;
 139     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 140     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 141
 142     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 143     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm_setzero_ps();
 179         fiy0             = _mm_setzero_ps();
 180         fiz0             = _mm_setzero_ps();
 181         fix1             = _mm_setzero_ps();
 182         fiy1             = _mm_setzero_ps();
 183         fiz1             = _mm_setzero_ps();
 184         fix2             = _mm_setzero_ps();
 185         fiy2             = _mm_setzero_ps();
 186         fiz2             = _mm_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm_setzero_ps();
 190         vvdwsum          = _mm_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               x+j_coord_offsetC,x+j_coord_offsetD,
 209                                               &jx0,&jy0,&jz0);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm_sub_ps(ix0,jx0);
 213             dy00             = _mm_sub_ps(iy0,jy0);
 214             dz00             = _mm_sub_ps(iz0,jz0);
 215             dx10             = _mm_sub_ps(ix1,jx0);
 216             dy10             = _mm_sub_ps(iy1,jy0);
 217             dz10             = _mm_sub_ps(iz1,jz0);
 218             dx20             = _mm_sub_ps(ix2,jx0);
 219             dy20             = _mm_sub_ps(iy2,jy0);
 220             dz20             = _mm_sub_ps(iz2,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 224             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 226
 227             rinv00           = sse41_invsqrt_f(rsq00);
 228             rinv10           = sse41_invsqrt_f(rsq10);
 229             rinv20           = sse41_invsqrt_f(rsq20);
 230
 231             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 232             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 233             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 234
 235             /* Load parameters for j particles */
 236             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 237                                                               charge+jnrC+0,charge+jnrD+0);
 238             vdwjidx0A        = 2*vdwtype[jnrA+0];
 239             vdwjidx0B        = 2*vdwtype[jnrB+0];
 240             vdwjidx0C        = 2*vdwtype[jnrC+0];
 241             vdwjidx0D        = 2*vdwtype[jnrD+0];
 242
 243             fjx0             = _mm_setzero_ps();
 244             fjy0             = _mm_setzero_ps();
 245             fjz0             = _mm_setzero_ps();
 246
 247             /**************************
 248              * CALCULATE INTERACTIONS *
 249              **************************/
 250
 251             if (gmx_mm_any_lt(rsq00,rcutoff2))
 252             {
 253
 254             r00              = _mm_mul_ps(rsq00,rinv00);
 255
 256             /* Compute parameters for interactions between i and j atoms */
 257             qq00             = _mm_mul_ps(iq0,jq0);
 258             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 259                                          vdwparam+vdwioffset0+vdwjidx0B,
 260                                          vdwparam+vdwioffset0+vdwjidx0C,
 261                                          vdwparam+vdwioffset0+vdwjidx0D,
 262                                          &c6_00,&c12_00);
 263
 264             /* EWALD ELECTROSTATICS */
 265
 266             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 267             ewrt             = _mm_mul_ps(r00,ewtabscale);
 268             ewitab           = _mm_cvttps_epi32(ewrt);
 269             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 270             ewitab           = _mm_slli_epi32(ewitab,2);
 271             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 272             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 273             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 274             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 275             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 276             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 277             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 278             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 279             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 280
 281             /* LENNARD-JONES DISPERSION/REPULSION */
 282
 283             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 284             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 285             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 286             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 287                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 288             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 289
 290             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             velec            = _mm_and_ps(velec,cutoff_mask);
 294             velecsum         = _mm_add_ps(velecsum,velec);
 295             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 296             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 297
 298             fscal            = _mm_add_ps(felec,fvdw);
 299
 300             fscal            = _mm_and_ps(fscal,cutoff_mask);
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_ps(fscal,dx00);
 304             ty               = _mm_mul_ps(fscal,dy00);
 305             tz               = _mm_mul_ps(fscal,dz00);
 306
 307             /* Update vectorial force */
 308             fix0             = _mm_add_ps(fix0,tx);
 309             fiy0             = _mm_add_ps(fiy0,ty);
 310             fiz0             = _mm_add_ps(fiz0,tz);
 311
 312             fjx0             = _mm_add_ps(fjx0,tx);
 313             fjy0             = _mm_add_ps(fjy0,ty);
 314             fjz0             = _mm_add_ps(fjz0,tz);
 315
 316             }
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             if (gmx_mm_any_lt(rsq10,rcutoff2))
 323             {
 324
 325             r10              = _mm_mul_ps(rsq10,rinv10);
 326
 327             /* Compute parameters for interactions between i and j atoms */
 328             qq10             = _mm_mul_ps(iq1,jq0);
 329
 330             /* EWALD ELECTROSTATICS */
 331
 332             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 333             ewrt             = _mm_mul_ps(r10,ewtabscale);
 334             ewitab           = _mm_cvttps_epi32(ewrt);
 335             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 336             ewitab           = _mm_slli_epi32(ewitab,2);
 337             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 338             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 339             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 340             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 341             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 342             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 343             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 344             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 345             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 346
 347             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velec            = _mm_and_ps(velec,cutoff_mask);
 351             velecsum         = _mm_add_ps(velecsum,velec);
 352
 353             fscal            = felec;
 354
 355             fscal            = _mm_and_ps(fscal,cutoff_mask);
 356
 357             /* Calculate temporary vectorial force */
 358             tx               = _mm_mul_ps(fscal,dx10);
 359             ty               = _mm_mul_ps(fscal,dy10);
 360             tz               = _mm_mul_ps(fscal,dz10);
 361
 362             /* Update vectorial force */
 363             fix1             = _mm_add_ps(fix1,tx);
 364             fiy1             = _mm_add_ps(fiy1,ty);
 365             fiz1             = _mm_add_ps(fiz1,tz);
 366
 367             fjx0             = _mm_add_ps(fjx0,tx);
 368             fjy0             = _mm_add_ps(fjy0,ty);
 369             fjz0             = _mm_add_ps(fjz0,tz);
 370
 371             }
 372
 373             /**************************
 374              * CALCULATE INTERACTIONS *
 375              **************************/
 376
 377             if (gmx_mm_any_lt(rsq20,rcutoff2))
 378             {
 379
 380             r20              = _mm_mul_ps(rsq20,rinv20);
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq20             = _mm_mul_ps(iq2,jq0);
 384
 385             /* EWALD ELECTROSTATICS */
 386
 387             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 388             ewrt             = _mm_mul_ps(r20,ewtabscale);
 389             ewitab           = _mm_cvttps_epi32(ewrt);
 390             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 391             ewitab           = _mm_slli_epi32(ewitab,2);
 392             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 393             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 394             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 395             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 396             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 397             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 398             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 399             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 400             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 401
 402             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm_and_ps(velec,cutoff_mask);
 406             velecsum         = _mm_add_ps(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             fscal            = _mm_and_ps(fscal,cutoff_mask);
 411
 412             /* Calculate temporary vectorial force */
 413             tx               = _mm_mul_ps(fscal,dx20);
 414             ty               = _mm_mul_ps(fscal,dy20);
 415             tz               = _mm_mul_ps(fscal,dz20);
 416
 417             /* Update vectorial force */
 418             fix2             = _mm_add_ps(fix2,tx);
 419             fiy2             = _mm_add_ps(fiy2,ty);
 420             fiz2             = _mm_add_ps(fiz2,tz);
 421
 422             fjx0             = _mm_add_ps(fjx0,tx);
 423             fjy0             = _mm_add_ps(fjy0,ty);
 424             fjz0             = _mm_add_ps(fjz0,tz);
 425
 426             }
 427
 428             fjptrA             = f+j_coord_offsetA;
 429             fjptrB             = f+j_coord_offsetB;
 430             fjptrC             = f+j_coord_offsetC;
 431             fjptrD             = f+j_coord_offsetD;
 432
 433             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 434
 435             /* Inner loop uses 156 flops */
 436         }
 437
 438         if(jidx<j_index_end)
 439         {
 440
 441             /* Get j neighbor index, and coordinate index */
 442             jnrlistA         = jjnr[jidx];
 443             jnrlistB         = jjnr[jidx+1];
 444             jnrlistC         = jjnr[jidx+2];
 445             jnrlistD         = jjnr[jidx+3];
 446             /* Sign of each element will be negative for non-real atoms.
 447              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 448              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 449              */
 450             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 451             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 452             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 453             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 454             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 455             j_coord_offsetA  = DIM*jnrA;
 456             j_coord_offsetB  = DIM*jnrB;
 457             j_coord_offsetC  = DIM*jnrC;
 458             j_coord_offsetD  = DIM*jnrD;
 459
 460             /* load j atom coordinates */
 461             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 462                                               x+j_coord_offsetC,x+j_coord_offsetD,
 463                                               &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm_sub_ps(ix0,jx0);
 467             dy00             = _mm_sub_ps(iy0,jy0);
 468             dz00             = _mm_sub_ps(iz0,jz0);
 469             dx10             = _mm_sub_ps(ix1,jx0);
 470             dy10             = _mm_sub_ps(iy1,jy0);
 471             dz10             = _mm_sub_ps(iz1,jz0);
 472             dx20             = _mm_sub_ps(ix2,jx0);
 473             dy20             = _mm_sub_ps(iy2,jy0);
 474             dz20             = _mm_sub_ps(iz2,jz0);
 475
 476             /* Calculate squared distance and things based on it */
 477             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 478             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 479             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 480
 481             rinv00           = sse41_invsqrt_f(rsq00);
 482             rinv10           = sse41_invsqrt_f(rsq10);
 483             rinv20           = sse41_invsqrt_f(rsq20);
 484
 485             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 486             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 487             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 488
 489             /* Load parameters for j particles */
 490             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 491                                                               charge+jnrC+0,charge+jnrD+0);
 492             vdwjidx0A        = 2*vdwtype[jnrA+0];
 493             vdwjidx0B        = 2*vdwtype[jnrB+0];
 494             vdwjidx0C        = 2*vdwtype[jnrC+0];
 495             vdwjidx0D        = 2*vdwtype[jnrD+0];
 496
 497             fjx0             = _mm_setzero_ps();
 498             fjy0             = _mm_setzero_ps();
 499             fjz0             = _mm_setzero_ps();
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             if (gmx_mm_any_lt(rsq00,rcutoff2))
 506             {
 507
 508             r00              = _mm_mul_ps(rsq00,rinv00);
 509             r00              = _mm_andnot_ps(dummy_mask,r00);
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             qq00             = _mm_mul_ps(iq0,jq0);
 513             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 514                                          vdwparam+vdwioffset0+vdwjidx0B,
 515                                          vdwparam+vdwioffset0+vdwjidx0C,
 516                                          vdwparam+vdwioffset0+vdwjidx0D,
 517                                          &c6_00,&c12_00);
 518
 519             /* EWALD ELECTROSTATICS */
 520
 521             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 522             ewrt             = _mm_mul_ps(r00,ewtabscale);
 523             ewitab           = _mm_cvttps_epi32(ewrt);
 524             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 525             ewitab           = _mm_slli_epi32(ewitab,2);
 526             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 527             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 528             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 529             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 530             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 531             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 532             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 533             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 534             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 535
 536             /* LENNARD-JONES DISPERSION/REPULSION */
 537
 538             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 539             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 540             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 541             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 542                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 543             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 544
 545             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 546
 547             /* Update potential sum for this i atom from the interaction with this j atom. */
 548             velec            = _mm_and_ps(velec,cutoff_mask);
 549             velec            = _mm_andnot_ps(dummy_mask,velec);
 550             velecsum         = _mm_add_ps(velecsum,velec);
 551             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 552             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 553             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 554
 555             fscal            = _mm_add_ps(felec,fvdw);
 556
 557             fscal            = _mm_and_ps(fscal,cutoff_mask);
 558
 559             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm_mul_ps(fscal,dx00);
 563             ty               = _mm_mul_ps(fscal,dy00);
 564             tz               = _mm_mul_ps(fscal,dz00);
 565
 566             /* Update vectorial force */
 567             fix0             = _mm_add_ps(fix0,tx);
 568             fiy0             = _mm_add_ps(fiy0,ty);
 569             fiz0             = _mm_add_ps(fiz0,tz);
 570
 571             fjx0             = _mm_add_ps(fjx0,tx);
 572             fjy0             = _mm_add_ps(fjy0,ty);
 573             fjz0             = _mm_add_ps(fjz0,tz);
 574
 575             }
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             if (gmx_mm_any_lt(rsq10,rcutoff2))
 582             {
 583
 584             r10              = _mm_mul_ps(rsq10,rinv10);
 585             r10              = _mm_andnot_ps(dummy_mask,r10);
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             qq10             = _mm_mul_ps(iq1,jq0);
 589
 590             /* EWALD ELECTROSTATICS */
 591
 592             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 593             ewrt             = _mm_mul_ps(r10,ewtabscale);
 594             ewitab           = _mm_cvttps_epi32(ewrt);
 595             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 596             ewitab           = _mm_slli_epi32(ewitab,2);
 597             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 598             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 599             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 600             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 601             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 602             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 603             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 604             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 605             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 606
 607             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 608
 609             /* Update potential sum for this i atom from the interaction with this j atom. */
 610             velec            = _mm_and_ps(velec,cutoff_mask);
 611             velec            = _mm_andnot_ps(dummy_mask,velec);
 612             velecsum         = _mm_add_ps(velecsum,velec);
 613
 614             fscal            = felec;
 615
 616             fscal            = _mm_and_ps(fscal,cutoff_mask);
 617
 618             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 619
 620             /* Calculate temporary vectorial force */
 621             tx               = _mm_mul_ps(fscal,dx10);
 622             ty               = _mm_mul_ps(fscal,dy10);
 623             tz               = _mm_mul_ps(fscal,dz10);
 624
 625             /* Update vectorial force */
 626             fix1             = _mm_add_ps(fix1,tx);
 627             fiy1             = _mm_add_ps(fiy1,ty);
 628             fiz1             = _mm_add_ps(fiz1,tz);
 629
 630             fjx0             = _mm_add_ps(fjx0,tx);
 631             fjy0             = _mm_add_ps(fjy0,ty);
 632             fjz0             = _mm_add_ps(fjz0,tz);
 633
 634             }
 635
 636             /**************************
 637              * CALCULATE INTERACTIONS *
 638              **************************/
 639
 640             if (gmx_mm_any_lt(rsq20,rcutoff2))
 641             {
 642
 643             r20              = _mm_mul_ps(rsq20,rinv20);
 644             r20              = _mm_andnot_ps(dummy_mask,r20);
 645
 646             /* Compute parameters for interactions between i and j atoms */
 647             qq20             = _mm_mul_ps(iq2,jq0);
 648
 649             /* EWALD ELECTROSTATICS */
 650
 651             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 652             ewrt             = _mm_mul_ps(r20,ewtabscale);
 653             ewitab           = _mm_cvttps_epi32(ewrt);
 654             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 655             ewitab           = _mm_slli_epi32(ewitab,2);
 656             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 657             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 658             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 659             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 660             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 661             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 662             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 663             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 664             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 665
 666             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 667
 668             /* Update potential sum for this i atom from the interaction with this j atom. */
 669             velec            = _mm_and_ps(velec,cutoff_mask);
 670             velec            = _mm_andnot_ps(dummy_mask,velec);
 671             velecsum         = _mm_add_ps(velecsum,velec);
 672
 673             fscal            = felec;
 674
 675             fscal            = _mm_and_ps(fscal,cutoff_mask);
 676
 677             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 678
 679             /* Calculate temporary vectorial force */
 680             tx               = _mm_mul_ps(fscal,dx20);
 681             ty               = _mm_mul_ps(fscal,dy20);
 682             tz               = _mm_mul_ps(fscal,dz20);
 683
 684             /* Update vectorial force */
 685             fix2             = _mm_add_ps(fix2,tx);
 686             fiy2             = _mm_add_ps(fiy2,ty);
 687             fiz2             = _mm_add_ps(fiz2,tz);
 688
 689             fjx0             = _mm_add_ps(fjx0,tx);
 690             fjy0             = _mm_add_ps(fjy0,ty);
 691             fjz0             = _mm_add_ps(fjz0,tz);
 692
 693             }
 694
 695             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 696             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 697             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 698             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 699
 700             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 701
 702             /* Inner loop uses 159 flops */
 703         }
 704
 705         /* End of innermost loop */
 706
 707         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 708                                               f+i_coord_offset,fshift+i_shift_offset);
 709
 710         ggid                        = gid[iidx];
 711         /* Update potential energies */
 712         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 713         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 714
 715         /* Increment number of inner iterations */
 716         inneriter                  += j_index_end - j_index_start;
 717
 718         /* Outer loop uses 20 flops */
 719     }
 720
 721     /* Increment number of outer iterations */
 722     outeriter        += nri;
 723
 724     /* Update outer/inner flops */
 725
 726     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
 727 }
 728 /*
 729  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_single
 730  * Electrostatics interaction: Ewald
 731  * VdW interaction:            LennardJones
 732  * Geometry:                   Water3-Particle
 733  * Calculate force/pot:        Force
 734  */
 735 void
 736 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_single
 737                     (t_nblist                    * gmx_restrict       nlist,
 738                      rvec                        * gmx_restrict          xx,
 739                      rvec                        * gmx_restrict          ff,
 740                      struct t_forcerec           * gmx_restrict          fr,
 741                      t_mdatoms                   * gmx_restrict     mdatoms,
 742                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 743                      t_nrnb                      * gmx_restrict        nrnb)
 744 {
 745     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 746      * just 0 for non-waters.
 747      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 748      * jnr indices corresponding to data put in the four positions in the SIMD register.
 749      */
 750     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 751     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 752     int              jnrA,jnrB,jnrC,jnrD;
 753     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 754     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 755     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 756     real             rcutoff_scalar;
 757     real             *shiftvec,*fshift,*x,*f;
 758     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 759     real             scratch[4*DIM];
 760     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 761     int              vdwioffset0;
 762     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 763     int              vdwioffset1;
 764     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 765     int              vdwioffset2;
 766     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 767     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 768     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 769     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 770     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 771     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 772     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 773     real             *charge;
 774     int              nvdwtype;
 775     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 776     int              *vdwtype;
 777     real             *vdwparam;
 778     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 779     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 780     __m128i          ewitab;
 781     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 782     real             *ewtab;
 783     __m128           dummy_mask,cutoff_mask;
 784     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 785     __m128           one     = _mm_set1_ps(1.0);
 786     __m128           two     = _mm_set1_ps(2.0);
 787     x                = xx[0];
 788     f                = ff[0];
 789
 790     nri              = nlist->nri;
 791     iinr             = nlist->iinr;
 792     jindex           = nlist->jindex;
 793     jjnr             = nlist->jjnr;
 794     shiftidx         = nlist->shift;
 795     gid              = nlist->gid;
 796     shiftvec         = fr->shift_vec[0];
 797     fshift           = fr->fshift[0];
 798     facel            = _mm_set1_ps(fr->ic->epsfac);
 799     charge           = mdatoms->chargeA;
 800     nvdwtype         = fr->ntype;
 801     vdwparam         = fr->nbfp;
 802     vdwtype          = mdatoms->typeA;
 803
 804     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 805     ewtab            = fr->ic->tabq_coul_F;
 806     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 807     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 808
 809     /* Setup water-specific parameters */
 810     inr              = nlist->iinr[0];
 811     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 812     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 813     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 814     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 815
 816     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 817     rcutoff_scalar   = fr->ic->rcoulomb;
 818     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 819     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 820
 821     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 822     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 823
 824     /* Avoid stupid compiler warnings */
 825     jnrA = jnrB = jnrC = jnrD = 0;
 826     j_coord_offsetA = 0;
 827     j_coord_offsetB = 0;
 828     j_coord_offsetC = 0;
 829     j_coord_offsetD = 0;
 830
 831     outeriter        = 0;
 832     inneriter        = 0;
 833
 834     for(iidx=0;iidx<4*DIM;iidx++)
 835     {
 836         scratch[iidx] = 0.0;
 837     }
 838
 839     /* Start outer loop over neighborlists */
 840     for(iidx=0; iidx<nri; iidx++)
 841     {
 842         /* Load shift vector for this list */
 843         i_shift_offset   = DIM*shiftidx[iidx];
 844
 845         /* Load limits for loop over neighbors */
 846         j_index_start    = jindex[iidx];
 847         j_index_end      = jindex[iidx+1];
 848
 849         /* Get outer coordinate index */
 850         inr              = iinr[iidx];
 851         i_coord_offset   = DIM*inr;
 852
 853         /* Load i particle coords and add shift vector */
 854         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 855                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 856
 857         fix0             = _mm_setzero_ps();
 858         fiy0             = _mm_setzero_ps();
 859         fiz0             = _mm_setzero_ps();
 860         fix1             = _mm_setzero_ps();
 861         fiy1             = _mm_setzero_ps();
 862         fiz1             = _mm_setzero_ps();
 863         fix2             = _mm_setzero_ps();
 864         fiy2             = _mm_setzero_ps();
 865         fiz2             = _mm_setzero_ps();
 866
 867         /* Start inner kernel loop */
 868         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 869         {
 870
 871             /* Get j neighbor index, and coordinate index */
 872             jnrA             = jjnr[jidx];
 873             jnrB             = jjnr[jidx+1];
 874             jnrC             = jjnr[jidx+2];
 875             jnrD             = jjnr[jidx+3];
 876             j_coord_offsetA  = DIM*jnrA;
 877             j_coord_offsetB  = DIM*jnrB;
 878             j_coord_offsetC  = DIM*jnrC;
 879             j_coord_offsetD  = DIM*jnrD;
 880
 881             /* load j atom coordinates */
 882             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 883                                               x+j_coord_offsetC,x+j_coord_offsetD,
 884                                               &jx0,&jy0,&jz0);
 885
 886             /* Calculate displacement vector */
 887             dx00             = _mm_sub_ps(ix0,jx0);
 888             dy00             = _mm_sub_ps(iy0,jy0);
 889             dz00             = _mm_sub_ps(iz0,jz0);
 890             dx10             = _mm_sub_ps(ix1,jx0);
 891             dy10             = _mm_sub_ps(iy1,jy0);
 892             dz10             = _mm_sub_ps(iz1,jz0);
 893             dx20             = _mm_sub_ps(ix2,jx0);
 894             dy20             = _mm_sub_ps(iy2,jy0);
 895             dz20             = _mm_sub_ps(iz2,jz0);
 896
 897             /* Calculate squared distance and things based on it */
 898             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 899             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 900             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 901
 902             rinv00           = sse41_invsqrt_f(rsq00);
 903             rinv10           = sse41_invsqrt_f(rsq10);
 904             rinv20           = sse41_invsqrt_f(rsq20);
 905
 906             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 907             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 908             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 909
 910             /* Load parameters for j particles */
 911             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 912                                                               charge+jnrC+0,charge+jnrD+0);
 913             vdwjidx0A        = 2*vdwtype[jnrA+0];
 914             vdwjidx0B        = 2*vdwtype[jnrB+0];
 915             vdwjidx0C        = 2*vdwtype[jnrC+0];
 916             vdwjidx0D        = 2*vdwtype[jnrD+0];
 917
 918             fjx0             = _mm_setzero_ps();
 919             fjy0             = _mm_setzero_ps();
 920             fjz0             = _mm_setzero_ps();
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             if (gmx_mm_any_lt(rsq00,rcutoff2))
 927             {
 928
 929             r00              = _mm_mul_ps(rsq00,rinv00);
 930
 931             /* Compute parameters for interactions between i and j atoms */
 932             qq00             = _mm_mul_ps(iq0,jq0);
 933             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 934                                          vdwparam+vdwioffset0+vdwjidx0B,
 935                                          vdwparam+vdwioffset0+vdwjidx0C,
 936                                          vdwparam+vdwioffset0+vdwjidx0D,
 937                                          &c6_00,&c12_00);
 938
 939             /* EWALD ELECTROSTATICS */
 940
 941             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 942             ewrt             = _mm_mul_ps(r00,ewtabscale);
 943             ewitab           = _mm_cvttps_epi32(ewrt);
 944             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 945             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 946                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 947                                          &ewtabF,&ewtabFn);
 948             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 949             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 950
 951             /* LENNARD-JONES DISPERSION/REPULSION */
 952
 953             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 954             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 955
 956             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 957
 958             fscal            = _mm_add_ps(felec,fvdw);
 959
 960             fscal            = _mm_and_ps(fscal,cutoff_mask);
 961
 962             /* Calculate temporary vectorial force */
 963             tx               = _mm_mul_ps(fscal,dx00);
 964             ty               = _mm_mul_ps(fscal,dy00);
 965             tz               = _mm_mul_ps(fscal,dz00);
 966
 967             /* Update vectorial force */
 968             fix0             = _mm_add_ps(fix0,tx);
 969             fiy0             = _mm_add_ps(fiy0,ty);
 970             fiz0             = _mm_add_ps(fiz0,tz);
 971
 972             fjx0             = _mm_add_ps(fjx0,tx);
 973             fjy0             = _mm_add_ps(fjy0,ty);
 974             fjz0             = _mm_add_ps(fjz0,tz);
 975
 976             }
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             if (gmx_mm_any_lt(rsq10,rcutoff2))
 983             {
 984
 985             r10              = _mm_mul_ps(rsq10,rinv10);
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             qq10             = _mm_mul_ps(iq1,jq0);
 989
 990             /* EWALD ELECTROSTATICS */
 991
 992             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 993             ewrt             = _mm_mul_ps(r10,ewtabscale);
 994             ewitab           = _mm_cvttps_epi32(ewrt);
 995             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 996             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 997                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 998                                          &ewtabF,&ewtabFn);
 999             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1000             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1001
1002             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1003
1004             fscal            = felec;
1005
1006             fscal            = _mm_and_ps(fscal,cutoff_mask);
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm_mul_ps(fscal,dx10);
1010             ty               = _mm_mul_ps(fscal,dy10);
1011             tz               = _mm_mul_ps(fscal,dz10);
1012
1013             /* Update vectorial force */
1014             fix1             = _mm_add_ps(fix1,tx);
1015             fiy1             = _mm_add_ps(fiy1,ty);
1016             fiz1             = _mm_add_ps(fiz1,tz);
1017
1018             fjx0             = _mm_add_ps(fjx0,tx);
1019             fjy0             = _mm_add_ps(fjy0,ty);
1020             fjz0             = _mm_add_ps(fjz0,tz);
1021
1022             }
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             if (gmx_mm_any_lt(rsq20,rcutoff2))
1029             {
1030
1031             r20              = _mm_mul_ps(rsq20,rinv20);
1032
1033             /* Compute parameters for interactions between i and j atoms */
1034             qq20             = _mm_mul_ps(iq2,jq0);
1035
1036             /* EWALD ELECTROSTATICS */
1037
1038             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1039             ewrt             = _mm_mul_ps(r20,ewtabscale);
1040             ewitab           = _mm_cvttps_epi32(ewrt);
1041             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1042             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1043                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1044                                          &ewtabF,&ewtabFn);
1045             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1046             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1047
1048             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1049
1050             fscal            = felec;
1051
1052             fscal            = _mm_and_ps(fscal,cutoff_mask);
1053
1054             /* Calculate temporary vectorial force */
1055             tx               = _mm_mul_ps(fscal,dx20);
1056             ty               = _mm_mul_ps(fscal,dy20);
1057             tz               = _mm_mul_ps(fscal,dz20);
1058
1059             /* Update vectorial force */
1060             fix2             = _mm_add_ps(fix2,tx);
1061             fiy2             = _mm_add_ps(fiy2,ty);
1062             fiz2             = _mm_add_ps(fiz2,tz);
1063
1064             fjx0             = _mm_add_ps(fjx0,tx);
1065             fjy0             = _mm_add_ps(fjy0,ty);
1066             fjz0             = _mm_add_ps(fjz0,tz);
1067
1068             }
1069
1070             fjptrA             = f+j_coord_offsetA;
1071             fjptrB             = f+j_coord_offsetB;
1072             fjptrC             = f+j_coord_offsetC;
1073             fjptrD             = f+j_coord_offsetD;
1074
1075             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1076
1077             /* Inner loop uses 124 flops */
1078         }
1079
1080         if(jidx<j_index_end)
1081         {
1082
1083             /* Get j neighbor index, and coordinate index */
1084             jnrlistA         = jjnr[jidx];
1085             jnrlistB         = jjnr[jidx+1];
1086             jnrlistC         = jjnr[jidx+2];
1087             jnrlistD         = jjnr[jidx+3];
1088             /* Sign of each element will be negative for non-real atoms.
1089              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1090              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1091              */
1092             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1093             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1094             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1095             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1096             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1097             j_coord_offsetA  = DIM*jnrA;
1098             j_coord_offsetB  = DIM*jnrB;
1099             j_coord_offsetC  = DIM*jnrC;
1100             j_coord_offsetD  = DIM*jnrD;
1101
1102             /* load j atom coordinates */
1103             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1104                                               x+j_coord_offsetC,x+j_coord_offsetD,
1105                                               &jx0,&jy0,&jz0);
1106
1107             /* Calculate displacement vector */
1108             dx00             = _mm_sub_ps(ix0,jx0);
1109             dy00             = _mm_sub_ps(iy0,jy0);
1110             dz00             = _mm_sub_ps(iz0,jz0);
1111             dx10             = _mm_sub_ps(ix1,jx0);
1112             dy10             = _mm_sub_ps(iy1,jy0);
1113             dz10             = _mm_sub_ps(iz1,jz0);
1114             dx20             = _mm_sub_ps(ix2,jx0);
1115             dy20             = _mm_sub_ps(iy2,jy0);
1116             dz20             = _mm_sub_ps(iz2,jz0);
1117
1118             /* Calculate squared distance and things based on it */
1119             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1120             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1121             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1122
1123             rinv00           = sse41_invsqrt_f(rsq00);
1124             rinv10           = sse41_invsqrt_f(rsq10);
1125             rinv20           = sse41_invsqrt_f(rsq20);
1126
1127             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1128             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1129             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1130
1131             /* Load parameters for j particles */
1132             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1133                                                               charge+jnrC+0,charge+jnrD+0);
1134             vdwjidx0A        = 2*vdwtype[jnrA+0];
1135             vdwjidx0B        = 2*vdwtype[jnrB+0];
1136             vdwjidx0C        = 2*vdwtype[jnrC+0];
1137             vdwjidx0D        = 2*vdwtype[jnrD+0];
1138
1139             fjx0             = _mm_setzero_ps();
1140             fjy0             = _mm_setzero_ps();
1141             fjz0             = _mm_setzero_ps();
1142
1143             /**************************
1144              * CALCULATE INTERACTIONS *
1145              **************************/
1146
1147             if (gmx_mm_any_lt(rsq00,rcutoff2))
1148             {
1149
1150             r00              = _mm_mul_ps(rsq00,rinv00);
1151             r00              = _mm_andnot_ps(dummy_mask,r00);
1152
1153             /* Compute parameters for interactions between i and j atoms */
1154             qq00             = _mm_mul_ps(iq0,jq0);
1155             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1156                                          vdwparam+vdwioffset0+vdwjidx0B,
1157                                          vdwparam+vdwioffset0+vdwjidx0C,
1158                                          vdwparam+vdwioffset0+vdwjidx0D,
1159                                          &c6_00,&c12_00);
1160
1161             /* EWALD ELECTROSTATICS */
1162
1163             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1164             ewrt             = _mm_mul_ps(r00,ewtabscale);
1165             ewitab           = _mm_cvttps_epi32(ewrt);
1166             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1167             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1168                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1169                                          &ewtabF,&ewtabFn);
1170             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1171             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1172
1173             /* LENNARD-JONES DISPERSION/REPULSION */
1174
1175             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1176             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1177
1178             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1179
1180             fscal            = _mm_add_ps(felec,fvdw);
1181
1182             fscal            = _mm_and_ps(fscal,cutoff_mask);
1183
1184             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1185
1186             /* Calculate temporary vectorial force */
1187             tx               = _mm_mul_ps(fscal,dx00);
1188             ty               = _mm_mul_ps(fscal,dy00);
1189             tz               = _mm_mul_ps(fscal,dz00);
1190
1191             /* Update vectorial force */
1192             fix0             = _mm_add_ps(fix0,tx);
1193             fiy0             = _mm_add_ps(fiy0,ty);
1194             fiz0             = _mm_add_ps(fiz0,tz);
1195
1196             fjx0             = _mm_add_ps(fjx0,tx);
1197             fjy0             = _mm_add_ps(fjy0,ty);
1198             fjz0             = _mm_add_ps(fjz0,tz);
1199
1200             }
1201
1202             /**************************
1203              * CALCULATE INTERACTIONS *
1204              **************************/
1205
1206             if (gmx_mm_any_lt(rsq10,rcutoff2))
1207             {
1208
1209             r10              = _mm_mul_ps(rsq10,rinv10);
1210             r10              = _mm_andnot_ps(dummy_mask,r10);
1211
1212             /* Compute parameters for interactions between i and j atoms */
1213             qq10             = _mm_mul_ps(iq1,jq0);
1214
1215             /* EWALD ELECTROSTATICS */
1216
1217             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1218             ewrt             = _mm_mul_ps(r10,ewtabscale);
1219             ewitab           = _mm_cvttps_epi32(ewrt);
1220             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1221             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1222                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1223                                          &ewtabF,&ewtabFn);
1224             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1225             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1226
1227             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1228
1229             fscal            = felec;
1230
1231             fscal            = _mm_and_ps(fscal,cutoff_mask);
1232
1233             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1234
1235             /* Calculate temporary vectorial force */
1236             tx               = _mm_mul_ps(fscal,dx10);
1237             ty               = _mm_mul_ps(fscal,dy10);
1238             tz               = _mm_mul_ps(fscal,dz10);
1239
1240             /* Update vectorial force */
1241             fix1             = _mm_add_ps(fix1,tx);
1242             fiy1             = _mm_add_ps(fiy1,ty);
1243             fiz1             = _mm_add_ps(fiz1,tz);
1244
1245             fjx0             = _mm_add_ps(fjx0,tx);
1246             fjy0             = _mm_add_ps(fjy0,ty);
1247             fjz0             = _mm_add_ps(fjz0,tz);
1248
1249             }
1250
1251             /**************************
1252              * CALCULATE INTERACTIONS *
1253              **************************/
1254
1255             if (gmx_mm_any_lt(rsq20,rcutoff2))
1256             {
1257
1258             r20              = _mm_mul_ps(rsq20,rinv20);
1259             r20              = _mm_andnot_ps(dummy_mask,r20);
1260
1261             /* Compute parameters for interactions between i and j atoms */
1262             qq20             = _mm_mul_ps(iq2,jq0);
1263
1264             /* EWALD ELECTROSTATICS */
1265
1266             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1267             ewrt             = _mm_mul_ps(r20,ewtabscale);
1268             ewitab           = _mm_cvttps_epi32(ewrt);
1269             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1270             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1271                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1272                                          &ewtabF,&ewtabFn);
1273             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1274             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1275
1276             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1277
1278             fscal            = felec;
1279
1280             fscal            = _mm_and_ps(fscal,cutoff_mask);
1281
1282             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1283
1284             /* Calculate temporary vectorial force */
1285             tx               = _mm_mul_ps(fscal,dx20);
1286             ty               = _mm_mul_ps(fscal,dy20);
1287             tz               = _mm_mul_ps(fscal,dz20);
1288
1289             /* Update vectorial force */
1290             fix2             = _mm_add_ps(fix2,tx);
1291             fiy2             = _mm_add_ps(fiy2,ty);
1292             fiz2             = _mm_add_ps(fiz2,tz);
1293
1294             fjx0             = _mm_add_ps(fjx0,tx);
1295             fjy0             = _mm_add_ps(fjy0,ty);
1296             fjz0             = _mm_add_ps(fjz0,tz);
1297
1298             }
1299
1300             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1301             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1302             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1303             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1304
1305             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1306
1307             /* Inner loop uses 127 flops */
1308         }
1309
1310         /* End of innermost loop */
1311
1312         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1313                                               f+i_coord_offset,fshift+i_shift_offset);
1314
1315         /* Increment number of inner iterations */
1316         inneriter                  += j_index_end - j_index_start;
1317
1318         /* Outer loop uses 18 flops */
1319     }
1320
1321     /* Increment number of outer iterations */
1322     outeriter        += nri;
1323
1324     /* Update outer/inner flops */
1325
1326     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);
1327 }