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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LennardJones
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_VF_sse4_1_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 refer to j loop unrolling done with SSE, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          ewitab;
 105     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 106     real             *ewtab;
 107     __m128           dummy_mask,cutoff_mask;
 108     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 109     __m128           one     = _mm_set1_ps(1.0);
 110     __m128           two     = _mm_set1_ps(2.0);
 111     x                = xx[0];
 112     f                = ff[0];
 113
 114     nri              = nlist->nri;
 115     iinr             = nlist->iinr;
 116     jindex           = nlist->jindex;
 117     jjnr             = nlist->jjnr;
 118     shiftidx         = nlist->shift;
 119     gid              = nlist->gid;
 120     shiftvec         = fr->shift_vec[0];
 121     fshift           = fr->fshift[0];
 122     facel            = _mm_set1_ps(fr->epsfac);
 123     charge           = mdatoms->chargeA;
 124     nvdwtype         = fr->ntype;
 125     vdwparam         = fr->nbfp;
 126     vdwtype          = mdatoms->typeA;
 127
 128     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 129     ewtab            = fr->ic->tabq_coul_FDV0;
 130     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 131     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 132
 133     /* Setup water-specific parameters */
 134     inr              = nlist->iinr[0];
 135     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 136     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 137     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 138     vdwioffset0      = 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          = _mm_set1_ps(rcutoff_scalar);
 143     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 144
 145     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 146     rvdw             = _mm_set1_ps(fr->rvdw);
 147
 148     /* Avoid stupid compiler warnings */
 149     jnrA = jnrB = jnrC = jnrD = 0;
 150     j_coord_offsetA = 0;
 151     j_coord_offsetB = 0;
 152     j_coord_offsetC = 0;
 153     j_coord_offsetD = 0;
 154
 155     outeriter        = 0;
 156     inneriter        = 0;
 157
 158     for(iidx=0;iidx<4*DIM;iidx++)
 159     {
 160         scratch[iidx] = 0.0;
 161     }
 162
 163     /* Start outer loop over neighborlists */
 164     for(iidx=0; iidx<nri; iidx++)
 165     {
 166         /* Load shift vector for this list */
 167         i_shift_offset   = DIM*shiftidx[iidx];
 168
 169         /* Load limits for loop over neighbors */
 170         j_index_start    = jindex[iidx];
 171         j_index_end      = jindex[iidx+1];
 172
 173         /* Get outer coordinate index */
 174         inr              = iinr[iidx];
 175         i_coord_offset   = DIM*inr;
 176
 177         /* Load i particle coords and add shift vector */
 178         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 179                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 180
 181         fix0             = _mm_setzero_ps();
 182         fiy0             = _mm_setzero_ps();
 183         fiz0             = _mm_setzero_ps();
 184         fix1             = _mm_setzero_ps();
 185         fiy1             = _mm_setzero_ps();
 186         fiz1             = _mm_setzero_ps();
 187         fix2             = _mm_setzero_ps();
 188         fiy2             = _mm_setzero_ps();
 189         fiz2             = _mm_setzero_ps();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_ps();
 193         vvdwsum          = _mm_setzero_ps();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             jnrC             = jjnr[jidx+2];
 203             jnrD             = jjnr[jidx+3];
 204             j_coord_offsetA  = DIM*jnrA;
 205             j_coord_offsetB  = DIM*jnrB;
 206             j_coord_offsetC  = DIM*jnrC;
 207             j_coord_offsetD  = DIM*jnrD;
 208
 209             /* load j atom coordinates */
 210             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                               x+j_coord_offsetC,x+j_coord_offsetD,
 212                                               &jx0,&jy0,&jz0);
 213
 214             /* Calculate displacement vector */
 215             dx00             = _mm_sub_ps(ix0,jx0);
 216             dy00             = _mm_sub_ps(iy0,jy0);
 217             dz00             = _mm_sub_ps(iz0,jz0);
 218             dx10             = _mm_sub_ps(ix1,jx0);
 219             dy10             = _mm_sub_ps(iy1,jy0);
 220             dz10             = _mm_sub_ps(iz1,jz0);
 221             dx20             = _mm_sub_ps(ix2,jx0);
 222             dy20             = _mm_sub_ps(iy2,jy0);
 223             dz20             = _mm_sub_ps(iz2,jz0);
 224
 225             /* Calculate squared distance and things based on it */
 226             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 227             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 228             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 229
 230             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 231             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 232             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 233
 234             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 235             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 236             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 237
 238             /* Load parameters for j particles */
 239             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 240                                                               charge+jnrC+0,charge+jnrD+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243             vdwjidx0C        = 2*vdwtype[jnrC+0];
 244             vdwjidx0D        = 2*vdwtype[jnrD+0];
 245
 246             fjx0             = _mm_setzero_ps();
 247             fjy0             = _mm_setzero_ps();
 248             fjz0             = _mm_setzero_ps();
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             if (gmx_mm_any_lt(rsq00,rcutoff2))
 255             {
 256
 257             r00              = _mm_mul_ps(rsq00,rinv00);
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq00             = _mm_mul_ps(iq0,jq0);
 261             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 262                                          vdwparam+vdwioffset0+vdwjidx0B,
 263                                          vdwparam+vdwioffset0+vdwjidx0C,
 264                                          vdwparam+vdwioffset0+vdwjidx0D,
 265                                          &c6_00,&c12_00);
 266
 267             /* EWALD ELECTROSTATICS */
 268
 269             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 270             ewrt             = _mm_mul_ps(r00,ewtabscale);
 271             ewitab           = _mm_cvttps_epi32(ewrt);
 272             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 273             ewitab           = _mm_slli_epi32(ewitab,2);
 274             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 275             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 276             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 277             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 278             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 279             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 280             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 281             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 282             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 283
 284             /* LENNARD-JONES DISPERSION/REPULSION */
 285
 286             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 287             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 288             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 289             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) ,
 290                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 291             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 292
 293             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velec            = _mm_and_ps(velec,cutoff_mask);
 297             velecsum         = _mm_add_ps(velecsum,velec);
 298             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = _mm_add_ps(felec,fvdw);
 302
 303             fscal            = _mm_and_ps(fscal,cutoff_mask);
 304
 305             /* Calculate temporary vectorial force */
 306             tx               = _mm_mul_ps(fscal,dx00);
 307             ty               = _mm_mul_ps(fscal,dy00);
 308             tz               = _mm_mul_ps(fscal,dz00);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm_add_ps(fix0,tx);
 312             fiy0             = _mm_add_ps(fiy0,ty);
 313             fiz0             = _mm_add_ps(fiz0,tz);
 314
 315             fjx0             = _mm_add_ps(fjx0,tx);
 316             fjy0             = _mm_add_ps(fjy0,ty);
 317             fjz0             = _mm_add_ps(fjz0,tz);
 318
 319             }
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             if (gmx_mm_any_lt(rsq10,rcutoff2))
 326             {
 327
 328             r10              = _mm_mul_ps(rsq10,rinv10);
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq10             = _mm_mul_ps(iq1,jq0);
 332
 333             /* EWALD ELECTROSTATICS */
 334
 335             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 336             ewrt             = _mm_mul_ps(r10,ewtabscale);
 337             ewitab           = _mm_cvttps_epi32(ewrt);
 338             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 339             ewitab           = _mm_slli_epi32(ewitab,2);
 340             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 341             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 342             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 343             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 344             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 345             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 346             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 347             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 348             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 349
 350             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 351
 352             /* Update potential sum for this i atom from the interaction with this j atom. */
 353             velec            = _mm_and_ps(velec,cutoff_mask);
 354             velecsum         = _mm_add_ps(velecsum,velec);
 355
 356             fscal            = felec;
 357
 358             fscal            = _mm_and_ps(fscal,cutoff_mask);
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm_mul_ps(fscal,dx10);
 362             ty               = _mm_mul_ps(fscal,dy10);
 363             tz               = _mm_mul_ps(fscal,dz10);
 364
 365             /* Update vectorial force */
 366             fix1             = _mm_add_ps(fix1,tx);
 367             fiy1             = _mm_add_ps(fiy1,ty);
 368             fiz1             = _mm_add_ps(fiz1,tz);
 369
 370             fjx0             = _mm_add_ps(fjx0,tx);
 371             fjy0             = _mm_add_ps(fjy0,ty);
 372             fjz0             = _mm_add_ps(fjz0,tz);
 373
 374             }
 375
 376             /**************************
 377              * CALCULATE INTERACTIONS *
 378              **************************/
 379
 380             if (gmx_mm_any_lt(rsq20,rcutoff2))
 381             {
 382
 383             r20              = _mm_mul_ps(rsq20,rinv20);
 384
 385             /* Compute parameters for interactions between i and j atoms */
 386             qq20             = _mm_mul_ps(iq2,jq0);
 387
 388             /* EWALD ELECTROSTATICS */
 389
 390             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 391             ewrt             = _mm_mul_ps(r20,ewtabscale);
 392             ewitab           = _mm_cvttps_epi32(ewrt);
 393             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 394             ewitab           = _mm_slli_epi32(ewitab,2);
 395             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 396             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 397             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 398             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 399             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 400             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 401             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 402             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 403             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 404
 405             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velec            = _mm_and_ps(velec,cutoff_mask);
 409             velecsum         = _mm_add_ps(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413             fscal            = _mm_and_ps(fscal,cutoff_mask);
 414
 415             /* Calculate temporary vectorial force */
 416             tx               = _mm_mul_ps(fscal,dx20);
 417             ty               = _mm_mul_ps(fscal,dy20);
 418             tz               = _mm_mul_ps(fscal,dz20);
 419
 420             /* Update vectorial force */
 421             fix2             = _mm_add_ps(fix2,tx);
 422             fiy2             = _mm_add_ps(fiy2,ty);
 423             fiz2             = _mm_add_ps(fiz2,tz);
 424
 425             fjx0             = _mm_add_ps(fjx0,tx);
 426             fjy0             = _mm_add_ps(fjy0,ty);
 427             fjz0             = _mm_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
 436             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 437
 438             /* Inner loop uses 156 flops */
 439         }
 440
 441         if(jidx<j_index_end)
 442         {
 443
 444             /* Get j neighbor index, and coordinate index */
 445             jnrlistA         = jjnr[jidx];
 446             jnrlistB         = jjnr[jidx+1];
 447             jnrlistC         = jjnr[jidx+2];
 448             jnrlistD         = jjnr[jidx+3];
 449             /* Sign of each element will be negative for non-real atoms.
 450              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 451              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 452              */
 453             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 454             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 455             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 456             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 457             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 458             j_coord_offsetA  = DIM*jnrA;
 459             j_coord_offsetB  = DIM*jnrB;
 460             j_coord_offsetC  = DIM*jnrC;
 461             j_coord_offsetD  = DIM*jnrD;
 462
 463             /* load j atom coordinates */
 464             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 465                                               x+j_coord_offsetC,x+j_coord_offsetD,
 466                                               &jx0,&jy0,&jz0);
 467
 468             /* Calculate displacement vector */
 469             dx00             = _mm_sub_ps(ix0,jx0);
 470             dy00             = _mm_sub_ps(iy0,jy0);
 471             dz00             = _mm_sub_ps(iz0,jz0);
 472             dx10             = _mm_sub_ps(ix1,jx0);
 473             dy10             = _mm_sub_ps(iy1,jy0);
 474             dz10             = _mm_sub_ps(iz1,jz0);
 475             dx20             = _mm_sub_ps(ix2,jx0);
 476             dy20             = _mm_sub_ps(iy2,jy0);
 477             dz20             = _mm_sub_ps(iz2,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 481             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 482             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 483
 484             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 485             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 486             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 487
 488             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 489             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 490             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 491
 492             /* Load parameters for j particles */
 493             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 494                                                               charge+jnrC+0,charge+jnrD+0);
 495             vdwjidx0A        = 2*vdwtype[jnrA+0];
 496             vdwjidx0B        = 2*vdwtype[jnrB+0];
 497             vdwjidx0C        = 2*vdwtype[jnrC+0];
 498             vdwjidx0D        = 2*vdwtype[jnrD+0];
 499
 500             fjx0             = _mm_setzero_ps();
 501             fjy0             = _mm_setzero_ps();
 502             fjz0             = _mm_setzero_ps();
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             if (gmx_mm_any_lt(rsq00,rcutoff2))
 509             {
 510
 511             r00              = _mm_mul_ps(rsq00,rinv00);
 512             r00              = _mm_andnot_ps(dummy_mask,r00);
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq00             = _mm_mul_ps(iq0,jq0);
 516             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 517                                          vdwparam+vdwioffset0+vdwjidx0B,
 518                                          vdwparam+vdwioffset0+vdwjidx0C,
 519                                          vdwparam+vdwioffset0+vdwjidx0D,
 520                                          &c6_00,&c12_00);
 521
 522             /* EWALD ELECTROSTATICS */
 523
 524             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 525             ewrt             = _mm_mul_ps(r00,ewtabscale);
 526             ewitab           = _mm_cvttps_epi32(ewrt);
 527             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 528             ewitab           = _mm_slli_epi32(ewitab,2);
 529             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 530             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 531             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 532             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 533             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 534             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 535             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 536             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 537             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 538
 539             /* LENNARD-JONES DISPERSION/REPULSION */
 540
 541             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 542             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 543             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 544             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) ,
 545                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 546             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 547
 548             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm_and_ps(velec,cutoff_mask);
 552             velec            = _mm_andnot_ps(dummy_mask,velec);
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 555             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 556             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 557
 558             fscal            = _mm_add_ps(felec,fvdw);
 559
 560             fscal            = _mm_and_ps(fscal,cutoff_mask);
 561
 562             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_ps(fscal,dx00);
 566             ty               = _mm_mul_ps(fscal,dy00);
 567             tz               = _mm_mul_ps(fscal,dz00);
 568
 569             /* Update vectorial force */
 570             fix0             = _mm_add_ps(fix0,tx);
 571             fiy0             = _mm_add_ps(fiy0,ty);
 572             fiz0             = _mm_add_ps(fiz0,tz);
 573
 574             fjx0             = _mm_add_ps(fjx0,tx);
 575             fjy0             = _mm_add_ps(fjy0,ty);
 576             fjz0             = _mm_add_ps(fjz0,tz);
 577
 578             }
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             if (gmx_mm_any_lt(rsq10,rcutoff2))
 585             {
 586
 587             r10              = _mm_mul_ps(rsq10,rinv10);
 588             r10              = _mm_andnot_ps(dummy_mask,r10);
 589
 590             /* Compute parameters for interactions between i and j atoms */
 591             qq10             = _mm_mul_ps(iq1,jq0);
 592
 593             /* EWALD ELECTROSTATICS */
 594
 595             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 596             ewrt             = _mm_mul_ps(r10,ewtabscale);
 597             ewitab           = _mm_cvttps_epi32(ewrt);
 598             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 599             ewitab           = _mm_slli_epi32(ewitab,2);
 600             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 601             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 602             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 603             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 604             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 605             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 606             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 607             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 608             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 609
 610             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velec            = _mm_and_ps(velec,cutoff_mask);
 614             velec            = _mm_andnot_ps(dummy_mask,velec);
 615             velecsum         = _mm_add_ps(velecsum,velec);
 616
 617             fscal            = felec;
 618
 619             fscal            = _mm_and_ps(fscal,cutoff_mask);
 620
 621             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm_mul_ps(fscal,dx10);
 625             ty               = _mm_mul_ps(fscal,dy10);
 626             tz               = _mm_mul_ps(fscal,dz10);
 627
 628             /* Update vectorial force */
 629             fix1             = _mm_add_ps(fix1,tx);
 630             fiy1             = _mm_add_ps(fiy1,ty);
 631             fiz1             = _mm_add_ps(fiz1,tz);
 632
 633             fjx0             = _mm_add_ps(fjx0,tx);
 634             fjy0             = _mm_add_ps(fjy0,ty);
 635             fjz0             = _mm_add_ps(fjz0,tz);
 636
 637             }
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             if (gmx_mm_any_lt(rsq20,rcutoff2))
 644             {
 645
 646             r20              = _mm_mul_ps(rsq20,rinv20);
 647             r20              = _mm_andnot_ps(dummy_mask,r20);
 648
 649             /* Compute parameters for interactions between i and j atoms */
 650             qq20             = _mm_mul_ps(iq2,jq0);
 651
 652             /* EWALD ELECTROSTATICS */
 653
 654             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 655             ewrt             = _mm_mul_ps(r20,ewtabscale);
 656             ewitab           = _mm_cvttps_epi32(ewrt);
 657             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 658             ewitab           = _mm_slli_epi32(ewitab,2);
 659             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 660             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 661             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 662             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 663             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 664             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 665             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 666             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 667             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 668
 669             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 670
 671             /* Update potential sum for this i atom from the interaction with this j atom. */
 672             velec            = _mm_and_ps(velec,cutoff_mask);
 673             velec            = _mm_andnot_ps(dummy_mask,velec);
 674             velecsum         = _mm_add_ps(velecsum,velec);
 675
 676             fscal            = felec;
 677
 678             fscal            = _mm_and_ps(fscal,cutoff_mask);
 679
 680             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 681
 682             /* Calculate temporary vectorial force */
 683             tx               = _mm_mul_ps(fscal,dx20);
 684             ty               = _mm_mul_ps(fscal,dy20);
 685             tz               = _mm_mul_ps(fscal,dz20);
 686
 687             /* Update vectorial force */
 688             fix2             = _mm_add_ps(fix2,tx);
 689             fiy2             = _mm_add_ps(fiy2,ty);
 690             fiz2             = _mm_add_ps(fiz2,tz);
 691
 692             fjx0             = _mm_add_ps(fjx0,tx);
 693             fjy0             = _mm_add_ps(fjy0,ty);
 694             fjz0             = _mm_add_ps(fjz0,tz);
 695
 696             }
 697
 698             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 699             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 700             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 701             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 702
 703             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 704
 705             /* Inner loop uses 159 flops */
 706         }
 707
 708         /* End of innermost loop */
 709
 710         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 711                                               f+i_coord_offset,fshift+i_shift_offset);
 712
 713         ggid                        = gid[iidx];
 714         /* Update potential energies */
 715         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 716         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 717
 718         /* Increment number of inner iterations */
 719         inneriter                  += j_index_end - j_index_start;
 720
 721         /* Outer loop uses 20 flops */
 722     }
 723
 724     /* Increment number of outer iterations */
 725     outeriter        += nri;
 726
 727     /* Update outer/inner flops */
 728
 729     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
 730 }
 731 /*
 732  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_single
 733  * Electrostatics interaction: Ewald
 734  * VdW interaction:            LennardJones
 735  * Geometry:                   Water3-Particle
 736  * Calculate force/pot:        Force
 737  */
 738 void
 739 nb_kernel_ElecEwSh_VdwLJSh_GeomW3P1_F_sse4_1_single
 740                     (t_nblist                    * gmx_restrict       nlist,
 741                      rvec                        * gmx_restrict          xx,
 742                      rvec                        * gmx_restrict          ff,
 743                      t_forcerec                  * gmx_restrict          fr,
 744                      t_mdatoms                   * gmx_restrict     mdatoms,
 745                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 746                      t_nrnb                      * gmx_restrict        nrnb)
 747 {
 748     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 749      * just 0 for non-waters.
 750      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 751      * jnr indices corresponding to data put in the four positions in the SIMD register.
 752      */
 753     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 754     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 755     int              jnrA,jnrB,jnrC,jnrD;
 756     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 757     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 758     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 759     real             rcutoff_scalar;
 760     real             *shiftvec,*fshift,*x,*f;
 761     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 762     real             scratch[4*DIM];
 763     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 764     int              vdwioffset0;
 765     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 766     int              vdwioffset1;
 767     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 768     int              vdwioffset2;
 769     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 770     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 771     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 772     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 773     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 774     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 775     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 776     real             *charge;
 777     int              nvdwtype;
 778     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 779     int              *vdwtype;
 780     real             *vdwparam;
 781     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 782     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 783     __m128i          ewitab;
 784     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 785     real             *ewtab;
 786     __m128           dummy_mask,cutoff_mask;
 787     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 788     __m128           one     = _mm_set1_ps(1.0);
 789     __m128           two     = _mm_set1_ps(2.0);
 790     x                = xx[0];
 791     f                = ff[0];
 792
 793     nri              = nlist->nri;
 794     iinr             = nlist->iinr;
 795     jindex           = nlist->jindex;
 796     jjnr             = nlist->jjnr;
 797     shiftidx         = nlist->shift;
 798     gid              = nlist->gid;
 799     shiftvec         = fr->shift_vec[0];
 800     fshift           = fr->fshift[0];
 801     facel            = _mm_set1_ps(fr->epsfac);
 802     charge           = mdatoms->chargeA;
 803     nvdwtype         = fr->ntype;
 804     vdwparam         = fr->nbfp;
 805     vdwtype          = mdatoms->typeA;
 806
 807     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 808     ewtab            = fr->ic->tabq_coul_F;
 809     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 810     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 811
 812     /* Setup water-specific parameters */
 813     inr              = nlist->iinr[0];
 814     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 815     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 816     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 817     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 818
 819     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 820     rcutoff_scalar   = fr->rcoulomb;
 821     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 822     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 823
 824     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 825     rvdw             = _mm_set1_ps(fr->rvdw);
 826
 827     /* Avoid stupid compiler warnings */
 828     jnrA = jnrB = jnrC = jnrD = 0;
 829     j_coord_offsetA = 0;
 830     j_coord_offsetB = 0;
 831     j_coord_offsetC = 0;
 832     j_coord_offsetD = 0;
 833
 834     outeriter        = 0;
 835     inneriter        = 0;
 836
 837     for(iidx=0;iidx<4*DIM;iidx++)
 838     {
 839         scratch[iidx] = 0.0;
 840     }
 841
 842     /* Start outer loop over neighborlists */
 843     for(iidx=0; iidx<nri; iidx++)
 844     {
 845         /* Load shift vector for this list */
 846         i_shift_offset   = DIM*shiftidx[iidx];
 847
 848         /* Load limits for loop over neighbors */
 849         j_index_start    = jindex[iidx];
 850         j_index_end      = jindex[iidx+1];
 851
 852         /* Get outer coordinate index */
 853         inr              = iinr[iidx];
 854         i_coord_offset   = DIM*inr;
 855
 856         /* Load i particle coords and add shift vector */
 857         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 858                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 859
 860         fix0             = _mm_setzero_ps();
 861         fiy0             = _mm_setzero_ps();
 862         fiz0             = _mm_setzero_ps();
 863         fix1             = _mm_setzero_ps();
 864         fiy1             = _mm_setzero_ps();
 865         fiz1             = _mm_setzero_ps();
 866         fix2             = _mm_setzero_ps();
 867         fiy2             = _mm_setzero_ps();
 868         fiz2             = _mm_setzero_ps();
 869
 870         /* Start inner kernel loop */
 871         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 872         {
 873
 874             /* Get j neighbor index, and coordinate index */
 875             jnrA             = jjnr[jidx];
 876             jnrB             = jjnr[jidx+1];
 877             jnrC             = jjnr[jidx+2];
 878             jnrD             = jjnr[jidx+3];
 879             j_coord_offsetA  = DIM*jnrA;
 880             j_coord_offsetB  = DIM*jnrB;
 881             j_coord_offsetC  = DIM*jnrC;
 882             j_coord_offsetD  = DIM*jnrD;
 883
 884             /* load j atom coordinates */
 885             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 886                                               x+j_coord_offsetC,x+j_coord_offsetD,
 887                                               &jx0,&jy0,&jz0);
 888
 889             /* Calculate displacement vector */
 890             dx00             = _mm_sub_ps(ix0,jx0);
 891             dy00             = _mm_sub_ps(iy0,jy0);
 892             dz00             = _mm_sub_ps(iz0,jz0);
 893             dx10             = _mm_sub_ps(ix1,jx0);
 894             dy10             = _mm_sub_ps(iy1,jy0);
 895             dz10             = _mm_sub_ps(iz1,jz0);
 896             dx20             = _mm_sub_ps(ix2,jx0);
 897             dy20             = _mm_sub_ps(iy2,jy0);
 898             dz20             = _mm_sub_ps(iz2,jz0);
 899
 900             /* Calculate squared distance and things based on it */
 901             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 902             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 903             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 904
 905             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 906             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 907             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 908
 909             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 910             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 911             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 912
 913             /* Load parameters for j particles */
 914             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 915                                                               charge+jnrC+0,charge+jnrD+0);
 916             vdwjidx0A        = 2*vdwtype[jnrA+0];
 917             vdwjidx0B        = 2*vdwtype[jnrB+0];
 918             vdwjidx0C        = 2*vdwtype[jnrC+0];
 919             vdwjidx0D        = 2*vdwtype[jnrD+0];
 920
 921             fjx0             = _mm_setzero_ps();
 922             fjy0             = _mm_setzero_ps();
 923             fjz0             = _mm_setzero_ps();
 924
 925             /**************************
 926              * CALCULATE INTERACTIONS *
 927              **************************/
 928
 929             if (gmx_mm_any_lt(rsq00,rcutoff2))
 930             {
 931
 932             r00              = _mm_mul_ps(rsq00,rinv00);
 933
 934             /* Compute parameters for interactions between i and j atoms */
 935             qq00             = _mm_mul_ps(iq0,jq0);
 936             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 937                                          vdwparam+vdwioffset0+vdwjidx0B,
 938                                          vdwparam+vdwioffset0+vdwjidx0C,
 939                                          vdwparam+vdwioffset0+vdwjidx0D,
 940                                          &c6_00,&c12_00);
 941
 942             /* EWALD ELECTROSTATICS */
 943
 944             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 945             ewrt             = _mm_mul_ps(r00,ewtabscale);
 946             ewitab           = _mm_cvttps_epi32(ewrt);
 947             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 948             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 949                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 950                                          &ewtabF,&ewtabFn);
 951             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 952             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 953
 954             /* LENNARD-JONES DISPERSION/REPULSION */
 955
 956             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 957             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 958
 959             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 960
 961             fscal            = _mm_add_ps(felec,fvdw);
 962
 963             fscal            = _mm_and_ps(fscal,cutoff_mask);
 964
 965             /* Calculate temporary vectorial force */
 966             tx               = _mm_mul_ps(fscal,dx00);
 967             ty               = _mm_mul_ps(fscal,dy00);
 968             tz               = _mm_mul_ps(fscal,dz00);
 969
 970             /* Update vectorial force */
 971             fix0             = _mm_add_ps(fix0,tx);
 972             fiy0             = _mm_add_ps(fiy0,ty);
 973             fiz0             = _mm_add_ps(fiz0,tz);
 974
 975             fjx0             = _mm_add_ps(fjx0,tx);
 976             fjy0             = _mm_add_ps(fjy0,ty);
 977             fjz0             = _mm_add_ps(fjz0,tz);
 978
 979             }
 980
 981             /**************************
 982              * CALCULATE INTERACTIONS *
 983              **************************/
 984
 985             if (gmx_mm_any_lt(rsq10,rcutoff2))
 986             {
 987
 988             r10              = _mm_mul_ps(rsq10,rinv10);
 989
 990             /* Compute parameters for interactions between i and j atoms */
 991             qq10             = _mm_mul_ps(iq1,jq0);
 992
 993             /* EWALD ELECTROSTATICS */
 994
 995             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 996             ewrt             = _mm_mul_ps(r10,ewtabscale);
 997             ewitab           = _mm_cvttps_epi32(ewrt);
 998             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 999             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1000                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1001                                          &ewtabF,&ewtabFn);
1002             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1003             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1004
1005             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1006
1007             fscal            = felec;
1008
1009             fscal            = _mm_and_ps(fscal,cutoff_mask);
1010
1011             /* Calculate temporary vectorial force */
1012             tx               = _mm_mul_ps(fscal,dx10);
1013             ty               = _mm_mul_ps(fscal,dy10);
1014             tz               = _mm_mul_ps(fscal,dz10);
1015
1016             /* Update vectorial force */
1017             fix1             = _mm_add_ps(fix1,tx);
1018             fiy1             = _mm_add_ps(fiy1,ty);
1019             fiz1             = _mm_add_ps(fiz1,tz);
1020
1021             fjx0             = _mm_add_ps(fjx0,tx);
1022             fjy0             = _mm_add_ps(fjy0,ty);
1023             fjz0             = _mm_add_ps(fjz0,tz);
1024
1025             }
1026
1027             /**************************
1028              * CALCULATE INTERACTIONS *
1029              **************************/
1030
1031             if (gmx_mm_any_lt(rsq20,rcutoff2))
1032             {
1033
1034             r20              = _mm_mul_ps(rsq20,rinv20);
1035
1036             /* Compute parameters for interactions between i and j atoms */
1037             qq20             = _mm_mul_ps(iq2,jq0);
1038
1039             /* EWALD ELECTROSTATICS */
1040
1041             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1042             ewrt             = _mm_mul_ps(r20,ewtabscale);
1043             ewitab           = _mm_cvttps_epi32(ewrt);
1044             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1045             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1046                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1047                                          &ewtabF,&ewtabFn);
1048             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1049             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1050
1051             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1052
1053             fscal            = felec;
1054
1055             fscal            = _mm_and_ps(fscal,cutoff_mask);
1056
1057             /* Calculate temporary vectorial force */
1058             tx               = _mm_mul_ps(fscal,dx20);
1059             ty               = _mm_mul_ps(fscal,dy20);
1060             tz               = _mm_mul_ps(fscal,dz20);
1061
1062             /* Update vectorial force */
1063             fix2             = _mm_add_ps(fix2,tx);
1064             fiy2             = _mm_add_ps(fiy2,ty);
1065             fiz2             = _mm_add_ps(fiz2,tz);
1066
1067             fjx0             = _mm_add_ps(fjx0,tx);
1068             fjy0             = _mm_add_ps(fjy0,ty);
1069             fjz0             = _mm_add_ps(fjz0,tz);
1070
1071             }
1072
1073             fjptrA             = f+j_coord_offsetA;
1074             fjptrB             = f+j_coord_offsetB;
1075             fjptrC             = f+j_coord_offsetC;
1076             fjptrD             = f+j_coord_offsetD;
1077
1078             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1079
1080             /* Inner loop uses 124 flops */
1081         }
1082
1083         if(jidx<j_index_end)
1084         {
1085
1086             /* Get j neighbor index, and coordinate index */
1087             jnrlistA         = jjnr[jidx];
1088             jnrlistB         = jjnr[jidx+1];
1089             jnrlistC         = jjnr[jidx+2];
1090             jnrlistD         = jjnr[jidx+3];
1091             /* Sign of each element will be negative for non-real atoms.
1092              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1093              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1094              */
1095             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1096             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1097             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1098             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1099             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1100             j_coord_offsetA  = DIM*jnrA;
1101             j_coord_offsetB  = DIM*jnrB;
1102             j_coord_offsetC  = DIM*jnrC;
1103             j_coord_offsetD  = DIM*jnrD;
1104
1105             /* load j atom coordinates */
1106             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1107                                               x+j_coord_offsetC,x+j_coord_offsetD,
1108                                               &jx0,&jy0,&jz0);
1109
1110             /* Calculate displacement vector */
1111             dx00             = _mm_sub_ps(ix0,jx0);
1112             dy00             = _mm_sub_ps(iy0,jy0);
1113             dz00             = _mm_sub_ps(iz0,jz0);
1114             dx10             = _mm_sub_ps(ix1,jx0);
1115             dy10             = _mm_sub_ps(iy1,jy0);
1116             dz10             = _mm_sub_ps(iz1,jz0);
1117             dx20             = _mm_sub_ps(ix2,jx0);
1118             dy20             = _mm_sub_ps(iy2,jy0);
1119             dz20             = _mm_sub_ps(iz2,jz0);
1120
1121             /* Calculate squared distance and things based on it */
1122             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1123             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1124             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1125
1126             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1127             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1128             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1129
1130             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1131             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1132             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1133
1134             /* Load parameters for j particles */
1135             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1136                                                               charge+jnrC+0,charge+jnrD+0);
1137             vdwjidx0A        = 2*vdwtype[jnrA+0];
1138             vdwjidx0B        = 2*vdwtype[jnrB+0];
1139             vdwjidx0C        = 2*vdwtype[jnrC+0];
1140             vdwjidx0D        = 2*vdwtype[jnrD+0];
1141
1142             fjx0             = _mm_setzero_ps();
1143             fjy0             = _mm_setzero_ps();
1144             fjz0             = _mm_setzero_ps();
1145
1146             /**************************
1147              * CALCULATE INTERACTIONS *
1148              **************************/
1149
1150             if (gmx_mm_any_lt(rsq00,rcutoff2))
1151             {
1152
1153             r00              = _mm_mul_ps(rsq00,rinv00);
1154             r00              = _mm_andnot_ps(dummy_mask,r00);
1155
1156             /* Compute parameters for interactions between i and j atoms */
1157             qq00             = _mm_mul_ps(iq0,jq0);
1158             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1159                                          vdwparam+vdwioffset0+vdwjidx0B,
1160                                          vdwparam+vdwioffset0+vdwjidx0C,
1161                                          vdwparam+vdwioffset0+vdwjidx0D,
1162                                          &c6_00,&c12_00);
1163
1164             /* EWALD ELECTROSTATICS */
1165
1166             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1167             ewrt             = _mm_mul_ps(r00,ewtabscale);
1168             ewitab           = _mm_cvttps_epi32(ewrt);
1169             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1170             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1171                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1172                                          &ewtabF,&ewtabFn);
1173             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1174             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1175
1176             /* LENNARD-JONES DISPERSION/REPULSION */
1177
1178             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1179             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1180
1181             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1182
1183             fscal            = _mm_add_ps(felec,fvdw);
1184
1185             fscal            = _mm_and_ps(fscal,cutoff_mask);
1186
1187             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1188
1189             /* Calculate temporary vectorial force */
1190             tx               = _mm_mul_ps(fscal,dx00);
1191             ty               = _mm_mul_ps(fscal,dy00);
1192             tz               = _mm_mul_ps(fscal,dz00);
1193
1194             /* Update vectorial force */
1195             fix0             = _mm_add_ps(fix0,tx);
1196             fiy0             = _mm_add_ps(fiy0,ty);
1197             fiz0             = _mm_add_ps(fiz0,tz);
1198
1199             fjx0             = _mm_add_ps(fjx0,tx);
1200             fjy0             = _mm_add_ps(fjy0,ty);
1201             fjz0             = _mm_add_ps(fjz0,tz);
1202
1203             }
1204
1205             /**************************
1206              * CALCULATE INTERACTIONS *
1207              **************************/
1208
1209             if (gmx_mm_any_lt(rsq10,rcutoff2))
1210             {
1211
1212             r10              = _mm_mul_ps(rsq10,rinv10);
1213             r10              = _mm_andnot_ps(dummy_mask,r10);
1214
1215             /* Compute parameters for interactions between i and j atoms */
1216             qq10             = _mm_mul_ps(iq1,jq0);
1217
1218             /* EWALD ELECTROSTATICS */
1219
1220             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1221             ewrt             = _mm_mul_ps(r10,ewtabscale);
1222             ewitab           = _mm_cvttps_epi32(ewrt);
1223             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1224             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1225                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1226                                          &ewtabF,&ewtabFn);
1227             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1228             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1229
1230             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1231
1232             fscal            = felec;
1233
1234             fscal            = _mm_and_ps(fscal,cutoff_mask);
1235
1236             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1237
1238             /* Calculate temporary vectorial force */
1239             tx               = _mm_mul_ps(fscal,dx10);
1240             ty               = _mm_mul_ps(fscal,dy10);
1241             tz               = _mm_mul_ps(fscal,dz10);
1242
1243             /* Update vectorial force */
1244             fix1             = _mm_add_ps(fix1,tx);
1245             fiy1             = _mm_add_ps(fiy1,ty);
1246             fiz1             = _mm_add_ps(fiz1,tz);
1247
1248             fjx0             = _mm_add_ps(fjx0,tx);
1249             fjy0             = _mm_add_ps(fjy0,ty);
1250             fjz0             = _mm_add_ps(fjz0,tz);
1251
1252             }
1253
1254             /**************************
1255              * CALCULATE INTERACTIONS *
1256              **************************/
1257
1258             if (gmx_mm_any_lt(rsq20,rcutoff2))
1259             {
1260
1261             r20              = _mm_mul_ps(rsq20,rinv20);
1262             r20              = _mm_andnot_ps(dummy_mask,r20);
1263
1264             /* Compute parameters for interactions between i and j atoms */
1265             qq20             = _mm_mul_ps(iq2,jq0);
1266
1267             /* EWALD ELECTROSTATICS */
1268
1269             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1270             ewrt             = _mm_mul_ps(r20,ewtabscale);
1271             ewitab           = _mm_cvttps_epi32(ewrt);
1272             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1273             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1274                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1275                                          &ewtabF,&ewtabFn);
1276             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1277             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1278
1279             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1280
1281             fscal            = felec;
1282
1283             fscal            = _mm_and_ps(fscal,cutoff_mask);
1284
1285             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1286
1287             /* Calculate temporary vectorial force */
1288             tx               = _mm_mul_ps(fscal,dx20);
1289             ty               = _mm_mul_ps(fscal,dy20);
1290             tz               = _mm_mul_ps(fscal,dz20);
1291
1292             /* Update vectorial force */
1293             fix2             = _mm_add_ps(fix2,tx);
1294             fiy2             = _mm_add_ps(fiy2,ty);
1295             fiz2             = _mm_add_ps(fiz2,tz);
1296
1297             fjx0             = _mm_add_ps(fjx0,tx);
1298             fjy0             = _mm_add_ps(fjy0,ty);
1299             fjz0             = _mm_add_ps(fjz0,tz);
1300
1301             }
1302
1303             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1304             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1305             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1306             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1307
1308             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1309
1310             /* Inner loop uses 127 flops */
1311         }
1312
1313         /* End of innermost loop */
1314
1315         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1316                                               f+i_coord_offset,fshift+i_shift_offset);
1317
1318         /* Increment number of inner iterations */
1319         inneriter                  += j_index_end - j_index_start;
1320
1321         /* Outer loop uses 18 flops */
1322     }
1323
1324     /* Increment number of outer iterations */
1325     outeriter        += nri;
1326
1327     /* Update outer/inner flops */
1328
1329     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*127);
1330 }