src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_sse4_1_single.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_ElecRFCut_VdwCSTab_GeomW4P1_VF_sse4_1_single
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_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              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128           dummy_mask,cutoff_mask;
 109     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 110     __m128           one     = _mm_set1_ps(1.0);
 111     __m128           two     = _mm_set1_ps(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_ps(fr->ic->epsfac);
 124     charge           = mdatoms->chargeA;
 125     krf              = _mm_set1_ps(fr->ic->k_rf);
 126     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 127     crf              = _mm_set1_ps(fr->ic->c_rf);
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 134
 135     /* Setup water-specific parameters */
 136     inr              = nlist->iinr[0];
 137     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 138     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 139     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 140     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 141
 142     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 143     rcutoff_scalar   = fr->ic->rcoulomb;
 144     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 145     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 179
 180         fix0             = _mm_setzero_ps();
 181         fiy0             = _mm_setzero_ps();
 182         fiz0             = _mm_setzero_ps();
 183         fix1             = _mm_setzero_ps();
 184         fiy1             = _mm_setzero_ps();
 185         fiz1             = _mm_setzero_ps();
 186         fix2             = _mm_setzero_ps();
 187         fiy2             = _mm_setzero_ps();
 188         fiz2             = _mm_setzero_ps();
 189         fix3             = _mm_setzero_ps();
 190         fiy3             = _mm_setzero_ps();
 191         fiz3             = _mm_setzero_ps();
 192
 193         /* Reset potential sums */
 194         velecsum         = _mm_setzero_ps();
 195         vvdwsum          = _mm_setzero_ps();
 196
 197         /* Start inner kernel loop */
 198         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 199         {
 200
 201             /* Get j neighbor index, and coordinate index */
 202             jnrA             = jjnr[jidx];
 203             jnrB             = jjnr[jidx+1];
 204             jnrC             = jjnr[jidx+2];
 205             jnrD             = jjnr[jidx+3];
 206             j_coord_offsetA  = DIM*jnrA;
 207             j_coord_offsetB  = DIM*jnrB;
 208             j_coord_offsetC  = DIM*jnrC;
 209             j_coord_offsetD  = DIM*jnrD;
 210
 211             /* load j atom coordinates */
 212             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 213                                               x+j_coord_offsetC,x+j_coord_offsetD,
 214                                               &jx0,&jy0,&jz0);
 215
 216             /* Calculate displacement vector */
 217             dx00             = _mm_sub_ps(ix0,jx0);
 218             dy00             = _mm_sub_ps(iy0,jy0);
 219             dz00             = _mm_sub_ps(iz0,jz0);
 220             dx10             = _mm_sub_ps(ix1,jx0);
 221             dy10             = _mm_sub_ps(iy1,jy0);
 222             dz10             = _mm_sub_ps(iz1,jz0);
 223             dx20             = _mm_sub_ps(ix2,jx0);
 224             dy20             = _mm_sub_ps(iy2,jy0);
 225             dz20             = _mm_sub_ps(iz2,jz0);
 226             dx30             = _mm_sub_ps(ix3,jx0);
 227             dy30             = _mm_sub_ps(iy3,jy0);
 228             dz30             = _mm_sub_ps(iz3,jz0);
 229
 230             /* Calculate squared distance and things based on it */
 231             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 232             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 233             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 234             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 235
 236             rinv00           = sse41_invsqrt_f(rsq00);
 237             rinv10           = sse41_invsqrt_f(rsq10);
 238             rinv20           = sse41_invsqrt_f(rsq20);
 239             rinv30           = sse41_invsqrt_f(rsq30);
 240
 241             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 242             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 243             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 244
 245             /* Load parameters for j particles */
 246             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 247                                                               charge+jnrC+0,charge+jnrD+0);
 248             vdwjidx0A        = 2*vdwtype[jnrA+0];
 249             vdwjidx0B        = 2*vdwtype[jnrB+0];
 250             vdwjidx0C        = 2*vdwtype[jnrC+0];
 251             vdwjidx0D        = 2*vdwtype[jnrD+0];
 252
 253             fjx0             = _mm_setzero_ps();
 254             fjy0             = _mm_setzero_ps();
 255             fjz0             = _mm_setzero_ps();
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             r00              = _mm_mul_ps(rsq00,rinv00);
 262
 263             /* Compute parameters for interactions between i and j atoms */
 264             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 265                                          vdwparam+vdwioffset0+vdwjidx0B,
 266                                          vdwparam+vdwioffset0+vdwjidx0C,
 267                                          vdwparam+vdwioffset0+vdwjidx0D,
 268                                          &c6_00,&c12_00);
 269
 270             /* Calculate table index by multiplying r with table scale and truncate to integer */
 271             rt               = _mm_mul_ps(r00,vftabscale);
 272             vfitab           = _mm_cvttps_epi32(rt);
 273             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 274             vfitab           = _mm_slli_epi32(vfitab,3);
 275
 276             /* CUBIC SPLINE TABLE DISPERSION */
 277             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 278             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 279             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 280             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 281             _MM_TRANSPOSE4_PS(Y,F,G,H);
 282             Heps             = _mm_mul_ps(vfeps,H);
 283             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 284             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 285             vvdw6            = _mm_mul_ps(c6_00,VV);
 286             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 287             fvdw6            = _mm_mul_ps(c6_00,FF);
 288
 289             /* CUBIC SPLINE TABLE REPULSION */
 290             vfitab           = _mm_add_epi32(vfitab,ifour);
 291             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 292             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 293             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 294             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 295             _MM_TRANSPOSE4_PS(Y,F,G,H);
 296             Heps             = _mm_mul_ps(vfeps,H);
 297             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 298             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 299             vvdw12           = _mm_mul_ps(c12_00,VV);
 300             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 301             fvdw12           = _mm_mul_ps(c12_00,FF);
 302             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 303             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 307
 308             fscal            = fvdw;
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm_mul_ps(fscal,dx00);
 312             ty               = _mm_mul_ps(fscal,dy00);
 313             tz               = _mm_mul_ps(fscal,dz00);
 314
 315             /* Update vectorial force */
 316             fix0             = _mm_add_ps(fix0,tx);
 317             fiy0             = _mm_add_ps(fiy0,ty);
 318             fiz0             = _mm_add_ps(fiz0,tz);
 319
 320             fjx0             = _mm_add_ps(fjx0,tx);
 321             fjy0             = _mm_add_ps(fjy0,ty);
 322             fjz0             = _mm_add_ps(fjz0,tz);
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             if (gmx_mm_any_lt(rsq10,rcutoff2))
 329             {
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq10             = _mm_mul_ps(iq1,jq0);
 333
 334             /* REACTION-FIELD ELECTROSTATICS */
 335             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 336             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 337
 338             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velec            = _mm_and_ps(velec,cutoff_mask);
 342             velecsum         = _mm_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             fscal            = _mm_and_ps(fscal,cutoff_mask);
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_ps(fscal,dx10);
 350             ty               = _mm_mul_ps(fscal,dy10);
 351             tz               = _mm_mul_ps(fscal,dz10);
 352
 353             /* Update vectorial force */
 354             fix1             = _mm_add_ps(fix1,tx);
 355             fiy1             = _mm_add_ps(fiy1,ty);
 356             fiz1             = _mm_add_ps(fiz1,tz);
 357
 358             fjx0             = _mm_add_ps(fjx0,tx);
 359             fjy0             = _mm_add_ps(fjy0,ty);
 360             fjz0             = _mm_add_ps(fjz0,tz);
 361
 362             }
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             if (gmx_mm_any_lt(rsq20,rcutoff2))
 369             {
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm_mul_ps(iq2,jq0);
 373
 374             /* REACTION-FIELD ELECTROSTATICS */
 375             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 376             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 377
 378             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 379
 380             /* Update potential sum for this i atom from the interaction with this j atom. */
 381             velec            = _mm_and_ps(velec,cutoff_mask);
 382             velecsum         = _mm_add_ps(velecsum,velec);
 383
 384             fscal            = felec;
 385
 386             fscal            = _mm_and_ps(fscal,cutoff_mask);
 387
 388             /* Calculate temporary vectorial force */
 389             tx               = _mm_mul_ps(fscal,dx20);
 390             ty               = _mm_mul_ps(fscal,dy20);
 391             tz               = _mm_mul_ps(fscal,dz20);
 392
 393             /* Update vectorial force */
 394             fix2             = _mm_add_ps(fix2,tx);
 395             fiy2             = _mm_add_ps(fiy2,ty);
 396             fiz2             = _mm_add_ps(fiz2,tz);
 397
 398             fjx0             = _mm_add_ps(fjx0,tx);
 399             fjy0             = _mm_add_ps(fjy0,ty);
 400             fjz0             = _mm_add_ps(fjz0,tz);
 401
 402             }
 403
 404             /**************************
 405              * CALCULATE INTERACTIONS *
 406              **************************/
 407
 408             if (gmx_mm_any_lt(rsq30,rcutoff2))
 409             {
 410
 411             /* Compute parameters for interactions between i and j atoms */
 412             qq30             = _mm_mul_ps(iq3,jq0);
 413
 414             /* REACTION-FIELD ELECTROSTATICS */
 415             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 416             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 417
 418             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velec            = _mm_and_ps(velec,cutoff_mask);
 422             velecsum         = _mm_add_ps(velecsum,velec);
 423
 424             fscal            = felec;
 425
 426             fscal            = _mm_and_ps(fscal,cutoff_mask);
 427
 428             /* Calculate temporary vectorial force */
 429             tx               = _mm_mul_ps(fscal,dx30);
 430             ty               = _mm_mul_ps(fscal,dy30);
 431             tz               = _mm_mul_ps(fscal,dz30);
 432
 433             /* Update vectorial force */
 434             fix3             = _mm_add_ps(fix3,tx);
 435             fiy3             = _mm_add_ps(fiy3,ty);
 436             fiz3             = _mm_add_ps(fiz3,tz);
 437
 438             fjx0             = _mm_add_ps(fjx0,tx);
 439             fjy0             = _mm_add_ps(fjy0,ty);
 440             fjz0             = _mm_add_ps(fjz0,tz);
 441
 442             }
 443
 444             fjptrA             = f+j_coord_offsetA;
 445             fjptrB             = f+j_coord_offsetB;
 446             fjptrC             = f+j_coord_offsetC;
 447             fjptrD             = f+j_coord_offsetD;
 448
 449             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 450
 451             /* Inner loop uses 164 flops */
 452         }
 453
 454         if(jidx<j_index_end)
 455         {
 456
 457             /* Get j neighbor index, and coordinate index */
 458             jnrlistA         = jjnr[jidx];
 459             jnrlistB         = jjnr[jidx+1];
 460             jnrlistC         = jjnr[jidx+2];
 461             jnrlistD         = jjnr[jidx+3];
 462             /* Sign of each element will be negative for non-real atoms.
 463              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 464              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 465              */
 466             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 467             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 468             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 469             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 470             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 471             j_coord_offsetA  = DIM*jnrA;
 472             j_coord_offsetB  = DIM*jnrB;
 473             j_coord_offsetC  = DIM*jnrC;
 474             j_coord_offsetD  = DIM*jnrD;
 475
 476             /* load j atom coordinates */
 477             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 478                                               x+j_coord_offsetC,x+j_coord_offsetD,
 479                                               &jx0,&jy0,&jz0);
 480
 481             /* Calculate displacement vector */
 482             dx00             = _mm_sub_ps(ix0,jx0);
 483             dy00             = _mm_sub_ps(iy0,jy0);
 484             dz00             = _mm_sub_ps(iz0,jz0);
 485             dx10             = _mm_sub_ps(ix1,jx0);
 486             dy10             = _mm_sub_ps(iy1,jy0);
 487             dz10             = _mm_sub_ps(iz1,jz0);
 488             dx20             = _mm_sub_ps(ix2,jx0);
 489             dy20             = _mm_sub_ps(iy2,jy0);
 490             dz20             = _mm_sub_ps(iz2,jz0);
 491             dx30             = _mm_sub_ps(ix3,jx0);
 492             dy30             = _mm_sub_ps(iy3,jy0);
 493             dz30             = _mm_sub_ps(iz3,jz0);
 494
 495             /* Calculate squared distance and things based on it */
 496             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 497             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 498             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 499             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 500
 501             rinv00           = sse41_invsqrt_f(rsq00);
 502             rinv10           = sse41_invsqrt_f(rsq10);
 503             rinv20           = sse41_invsqrt_f(rsq20);
 504             rinv30           = sse41_invsqrt_f(rsq30);
 505
 506             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 507             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 508             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 509
 510             /* Load parameters for j particles */
 511             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 512                                                               charge+jnrC+0,charge+jnrD+0);
 513             vdwjidx0A        = 2*vdwtype[jnrA+0];
 514             vdwjidx0B        = 2*vdwtype[jnrB+0];
 515             vdwjidx0C        = 2*vdwtype[jnrC+0];
 516             vdwjidx0D        = 2*vdwtype[jnrD+0];
 517
 518             fjx0             = _mm_setzero_ps();
 519             fjy0             = _mm_setzero_ps();
 520             fjz0             = _mm_setzero_ps();
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             r00              = _mm_mul_ps(rsq00,rinv00);
 527             r00              = _mm_andnot_ps(dummy_mask,r00);
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 531                                          vdwparam+vdwioffset0+vdwjidx0B,
 532                                          vdwparam+vdwioffset0+vdwjidx0C,
 533                                          vdwparam+vdwioffset0+vdwjidx0D,
 534                                          &c6_00,&c12_00);
 535
 536             /* Calculate table index by multiplying r with table scale and truncate to integer */
 537             rt               = _mm_mul_ps(r00,vftabscale);
 538             vfitab           = _mm_cvttps_epi32(rt);
 539             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 540             vfitab           = _mm_slli_epi32(vfitab,3);
 541
 542             /* CUBIC SPLINE TABLE DISPERSION */
 543             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 544             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 545             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 546             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 547             _MM_TRANSPOSE4_PS(Y,F,G,H);
 548             Heps             = _mm_mul_ps(vfeps,H);
 549             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 550             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 551             vvdw6            = _mm_mul_ps(c6_00,VV);
 552             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 553             fvdw6            = _mm_mul_ps(c6_00,FF);
 554
 555             /* CUBIC SPLINE TABLE REPULSION */
 556             vfitab           = _mm_add_epi32(vfitab,ifour);
 557             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 558             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 559             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 560             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 561             _MM_TRANSPOSE4_PS(Y,F,G,H);
 562             Heps             = _mm_mul_ps(vfeps,H);
 563             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 564             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 565             vvdw12           = _mm_mul_ps(c12_00,VV);
 566             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 567             fvdw12           = _mm_mul_ps(c12_00,FF);
 568             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 569             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 573             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 574
 575             fscal            = fvdw;
 576
 577             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm_mul_ps(fscal,dx00);
 581             ty               = _mm_mul_ps(fscal,dy00);
 582             tz               = _mm_mul_ps(fscal,dz00);
 583
 584             /* Update vectorial force */
 585             fix0             = _mm_add_ps(fix0,tx);
 586             fiy0             = _mm_add_ps(fiy0,ty);
 587             fiz0             = _mm_add_ps(fiz0,tz);
 588
 589             fjx0             = _mm_add_ps(fjx0,tx);
 590             fjy0             = _mm_add_ps(fjy0,ty);
 591             fjz0             = _mm_add_ps(fjz0,tz);
 592
 593             /**************************
 594              * CALCULATE INTERACTIONS *
 595              **************************/
 596
 597             if (gmx_mm_any_lt(rsq10,rcutoff2))
 598             {
 599
 600             /* Compute parameters for interactions between i and j atoms */
 601             qq10             = _mm_mul_ps(iq1,jq0);
 602
 603             /* REACTION-FIELD ELECTROSTATICS */
 604             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 605             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 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             /* Compute parameters for interactions between i and j atoms */
 644             qq20             = _mm_mul_ps(iq2,jq0);
 645
 646             /* REACTION-FIELD ELECTROSTATICS */
 647             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 648             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 649
 650             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 651
 652             /* Update potential sum for this i atom from the interaction with this j atom. */
 653             velec            = _mm_and_ps(velec,cutoff_mask);
 654             velec            = _mm_andnot_ps(dummy_mask,velec);
 655             velecsum         = _mm_add_ps(velecsum,velec);
 656
 657             fscal            = felec;
 658
 659             fscal            = _mm_and_ps(fscal,cutoff_mask);
 660
 661             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 662
 663             /* Calculate temporary vectorial force */
 664             tx               = _mm_mul_ps(fscal,dx20);
 665             ty               = _mm_mul_ps(fscal,dy20);
 666             tz               = _mm_mul_ps(fscal,dz20);
 667
 668             /* Update vectorial force */
 669             fix2             = _mm_add_ps(fix2,tx);
 670             fiy2             = _mm_add_ps(fiy2,ty);
 671             fiz2             = _mm_add_ps(fiz2,tz);
 672
 673             fjx0             = _mm_add_ps(fjx0,tx);
 674             fjy0             = _mm_add_ps(fjy0,ty);
 675             fjz0             = _mm_add_ps(fjz0,tz);
 676
 677             }
 678
 679             /**************************
 680              * CALCULATE INTERACTIONS *
 681              **************************/
 682
 683             if (gmx_mm_any_lt(rsq30,rcutoff2))
 684             {
 685
 686             /* Compute parameters for interactions between i and j atoms */
 687             qq30             = _mm_mul_ps(iq3,jq0);
 688
 689             /* REACTION-FIELD ELECTROSTATICS */
 690             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 691             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 692
 693             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 694
 695             /* Update potential sum for this i atom from the interaction with this j atom. */
 696             velec            = _mm_and_ps(velec,cutoff_mask);
 697             velec            = _mm_andnot_ps(dummy_mask,velec);
 698             velecsum         = _mm_add_ps(velecsum,velec);
 699
 700             fscal            = felec;
 701
 702             fscal            = _mm_and_ps(fscal,cutoff_mask);
 703
 704             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 705
 706             /* Calculate temporary vectorial force */
 707             tx               = _mm_mul_ps(fscal,dx30);
 708             ty               = _mm_mul_ps(fscal,dy30);
 709             tz               = _mm_mul_ps(fscal,dz30);
 710
 711             /* Update vectorial force */
 712             fix3             = _mm_add_ps(fix3,tx);
 713             fiy3             = _mm_add_ps(fiy3,ty);
 714             fiz3             = _mm_add_ps(fiz3,tz);
 715
 716             fjx0             = _mm_add_ps(fjx0,tx);
 717             fjy0             = _mm_add_ps(fjy0,ty);
 718             fjz0             = _mm_add_ps(fjz0,tz);
 719
 720             }
 721
 722             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 723             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 724             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 725             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 726
 727             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 728
 729             /* Inner loop uses 165 flops */
 730         }
 731
 732         /* End of innermost loop */
 733
 734         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 735                                               f+i_coord_offset,fshift+i_shift_offset);
 736
 737         ggid                        = gid[iidx];
 738         /* Update potential energies */
 739         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 740         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 741
 742         /* Increment number of inner iterations */
 743         inneriter                  += j_index_end - j_index_start;
 744
 745         /* Outer loop uses 26 flops */
 746     }
 747
 748     /* Increment number of outer iterations */
 749     outeriter        += nri;
 750
 751     /* Update outer/inner flops */
 752
 753     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*165);
 754 }
 755 /*
 756  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_single
 757  * Electrostatics interaction: ReactionField
 758  * VdW interaction:            CubicSplineTable
 759  * Geometry:                   Water4-Particle
 760  * Calculate force/pot:        Force
 761  */
 762 void
 763 nb_kernel_ElecRFCut_VdwCSTab_GeomW4P1_F_sse4_1_single
 764                     (t_nblist                    * gmx_restrict       nlist,
 765                      rvec                        * gmx_restrict          xx,
 766                      rvec                        * gmx_restrict          ff,
 767                      struct t_forcerec           * gmx_restrict          fr,
 768                      t_mdatoms                   * gmx_restrict     mdatoms,
 769                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 770                      t_nrnb                      * gmx_restrict        nrnb)
 771 {
 772     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 773      * just 0 for non-waters.
 774      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 775      * jnr indices corresponding to data put in the four positions in the SIMD register.
 776      */
 777     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 778     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 779     int              jnrA,jnrB,jnrC,jnrD;
 780     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 781     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 782     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 783     real             rcutoff_scalar;
 784     real             *shiftvec,*fshift,*x,*f;
 785     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 786     real             scratch[4*DIM];
 787     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 788     int              vdwioffset0;
 789     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 790     int              vdwioffset1;
 791     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 792     int              vdwioffset2;
 793     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 794     int              vdwioffset3;
 795     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 796     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 797     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 798     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 799     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 800     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 801     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 802     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 803     real             *charge;
 804     int              nvdwtype;
 805     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 806     int              *vdwtype;
 807     real             *vdwparam;
 808     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 809     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 810     __m128i          vfitab;
 811     __m128i          ifour       = _mm_set1_epi32(4);
 812     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 813     real             *vftab;
 814     __m128           dummy_mask,cutoff_mask;
 815     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 816     __m128           one     = _mm_set1_ps(1.0);
 817     __m128           two     = _mm_set1_ps(2.0);
 818     x                = xx[0];
 819     f                = ff[0];
 820
 821     nri              = nlist->nri;
 822     iinr             = nlist->iinr;
 823     jindex           = nlist->jindex;
 824     jjnr             = nlist->jjnr;
 825     shiftidx         = nlist->shift;
 826     gid              = nlist->gid;
 827     shiftvec         = fr->shift_vec[0];
 828     fshift           = fr->fshift[0];
 829     facel            = _mm_set1_ps(fr->ic->epsfac);
 830     charge           = mdatoms->chargeA;
 831     krf              = _mm_set1_ps(fr->ic->k_rf);
 832     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 833     crf              = _mm_set1_ps(fr->ic->c_rf);
 834     nvdwtype         = fr->ntype;
 835     vdwparam         = fr->nbfp;
 836     vdwtype          = mdatoms->typeA;
 837
 838     vftab            = kernel_data->table_vdw->data;
 839     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 840
 841     /* Setup water-specific parameters */
 842     inr              = nlist->iinr[0];
 843     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 844     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 845     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 846     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 847
 848     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 849     rcutoff_scalar   = fr->ic->rcoulomb;
 850     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 851     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 852
 853     /* Avoid stupid compiler warnings */
 854     jnrA = jnrB = jnrC = jnrD = 0;
 855     j_coord_offsetA = 0;
 856     j_coord_offsetB = 0;
 857     j_coord_offsetC = 0;
 858     j_coord_offsetD = 0;
 859
 860     outeriter        = 0;
 861     inneriter        = 0;
 862
 863     for(iidx=0;iidx<4*DIM;iidx++)
 864     {
 865         scratch[iidx] = 0.0;
 866     }
 867
 868     /* Start outer loop over neighborlists */
 869     for(iidx=0; iidx<nri; iidx++)
 870     {
 871         /* Load shift vector for this list */
 872         i_shift_offset   = DIM*shiftidx[iidx];
 873
 874         /* Load limits for loop over neighbors */
 875         j_index_start    = jindex[iidx];
 876         j_index_end      = jindex[iidx+1];
 877
 878         /* Get outer coordinate index */
 879         inr              = iinr[iidx];
 880         i_coord_offset   = DIM*inr;
 881
 882         /* Load i particle coords and add shift vector */
 883         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 884                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 885
 886         fix0             = _mm_setzero_ps();
 887         fiy0             = _mm_setzero_ps();
 888         fiz0             = _mm_setzero_ps();
 889         fix1             = _mm_setzero_ps();
 890         fiy1             = _mm_setzero_ps();
 891         fiz1             = _mm_setzero_ps();
 892         fix2             = _mm_setzero_ps();
 893         fiy2             = _mm_setzero_ps();
 894         fiz2             = _mm_setzero_ps();
 895         fix3             = _mm_setzero_ps();
 896         fiy3             = _mm_setzero_ps();
 897         fiz3             = _mm_setzero_ps();
 898
 899         /* Start inner kernel loop */
 900         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 901         {
 902
 903             /* Get j neighbor index, and coordinate index */
 904             jnrA             = jjnr[jidx];
 905             jnrB             = jjnr[jidx+1];
 906             jnrC             = jjnr[jidx+2];
 907             jnrD             = jjnr[jidx+3];
 908             j_coord_offsetA  = DIM*jnrA;
 909             j_coord_offsetB  = DIM*jnrB;
 910             j_coord_offsetC  = DIM*jnrC;
 911             j_coord_offsetD  = DIM*jnrD;
 912
 913             /* load j atom coordinates */
 914             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 915                                               x+j_coord_offsetC,x+j_coord_offsetD,
 916                                               &jx0,&jy0,&jz0);
 917
 918             /* Calculate displacement vector */
 919             dx00             = _mm_sub_ps(ix0,jx0);
 920             dy00             = _mm_sub_ps(iy0,jy0);
 921             dz00             = _mm_sub_ps(iz0,jz0);
 922             dx10             = _mm_sub_ps(ix1,jx0);
 923             dy10             = _mm_sub_ps(iy1,jy0);
 924             dz10             = _mm_sub_ps(iz1,jz0);
 925             dx20             = _mm_sub_ps(ix2,jx0);
 926             dy20             = _mm_sub_ps(iy2,jy0);
 927             dz20             = _mm_sub_ps(iz2,jz0);
 928             dx30             = _mm_sub_ps(ix3,jx0);
 929             dy30             = _mm_sub_ps(iy3,jy0);
 930             dz30             = _mm_sub_ps(iz3,jz0);
 931
 932             /* Calculate squared distance and things based on it */
 933             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 934             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 935             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 936             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 937
 938             rinv00           = sse41_invsqrt_f(rsq00);
 939             rinv10           = sse41_invsqrt_f(rsq10);
 940             rinv20           = sse41_invsqrt_f(rsq20);
 941             rinv30           = sse41_invsqrt_f(rsq30);
 942
 943             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 944             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 945             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 946
 947             /* Load parameters for j particles */
 948             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 949                                                               charge+jnrC+0,charge+jnrD+0);
 950             vdwjidx0A        = 2*vdwtype[jnrA+0];
 951             vdwjidx0B        = 2*vdwtype[jnrB+0];
 952             vdwjidx0C        = 2*vdwtype[jnrC+0];
 953             vdwjidx0D        = 2*vdwtype[jnrD+0];
 954
 955             fjx0             = _mm_setzero_ps();
 956             fjy0             = _mm_setzero_ps();
 957             fjz0             = _mm_setzero_ps();
 958
 959             /**************************
 960              * CALCULATE INTERACTIONS *
 961              **************************/
 962
 963             r00              = _mm_mul_ps(rsq00,rinv00);
 964
 965             /* Compute parameters for interactions between i and j atoms */
 966             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 967                                          vdwparam+vdwioffset0+vdwjidx0B,
 968                                          vdwparam+vdwioffset0+vdwjidx0C,
 969                                          vdwparam+vdwioffset0+vdwjidx0D,
 970                                          &c6_00,&c12_00);
 971
 972             /* Calculate table index by multiplying r with table scale and truncate to integer */
 973             rt               = _mm_mul_ps(r00,vftabscale);
 974             vfitab           = _mm_cvttps_epi32(rt);
 975             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 976             vfitab           = _mm_slli_epi32(vfitab,3);
 977
 978             /* CUBIC SPLINE TABLE DISPERSION */
 979             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 980             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 981             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 982             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 983             _MM_TRANSPOSE4_PS(Y,F,G,H);
 984             Heps             = _mm_mul_ps(vfeps,H);
 985             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 986             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 987             fvdw6            = _mm_mul_ps(c6_00,FF);
 988
 989             /* CUBIC SPLINE TABLE REPULSION */
 990             vfitab           = _mm_add_epi32(vfitab,ifour);
 991             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 992             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 993             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 994             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 995             _MM_TRANSPOSE4_PS(Y,F,G,H);
 996             Heps             = _mm_mul_ps(vfeps,H);
 997             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 998             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 999             fvdw12           = _mm_mul_ps(c12_00,FF);
1000             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1001
1002             fscal            = fvdw;
1003
1004             /* Calculate temporary vectorial force */
1005             tx               = _mm_mul_ps(fscal,dx00);
1006             ty               = _mm_mul_ps(fscal,dy00);
1007             tz               = _mm_mul_ps(fscal,dz00);
1008
1009             /* Update vectorial force */
1010             fix0             = _mm_add_ps(fix0,tx);
1011             fiy0             = _mm_add_ps(fiy0,ty);
1012             fiz0             = _mm_add_ps(fiz0,tz);
1013
1014             fjx0             = _mm_add_ps(fjx0,tx);
1015             fjy0             = _mm_add_ps(fjy0,ty);
1016             fjz0             = _mm_add_ps(fjz0,tz);
1017
1018             /**************************
1019              * CALCULATE INTERACTIONS *
1020              **************************/
1021
1022             if (gmx_mm_any_lt(rsq10,rcutoff2))
1023             {
1024
1025             /* Compute parameters for interactions between i and j atoms */
1026             qq10             = _mm_mul_ps(iq1,jq0);
1027
1028             /* REACTION-FIELD ELECTROSTATICS */
1029             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1030
1031             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1032
1033             fscal            = felec;
1034
1035             fscal            = _mm_and_ps(fscal,cutoff_mask);
1036
1037             /* Calculate temporary vectorial force */
1038             tx               = _mm_mul_ps(fscal,dx10);
1039             ty               = _mm_mul_ps(fscal,dy10);
1040             tz               = _mm_mul_ps(fscal,dz10);
1041
1042             /* Update vectorial force */
1043             fix1             = _mm_add_ps(fix1,tx);
1044             fiy1             = _mm_add_ps(fiy1,ty);
1045             fiz1             = _mm_add_ps(fiz1,tz);
1046
1047             fjx0             = _mm_add_ps(fjx0,tx);
1048             fjy0             = _mm_add_ps(fjy0,ty);
1049             fjz0             = _mm_add_ps(fjz0,tz);
1050
1051             }
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             if (gmx_mm_any_lt(rsq20,rcutoff2))
1058             {
1059
1060             /* Compute parameters for interactions between i and j atoms */
1061             qq20             = _mm_mul_ps(iq2,jq0);
1062
1063             /* REACTION-FIELD ELECTROSTATICS */
1064             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1065
1066             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1067
1068             fscal            = felec;
1069
1070             fscal            = _mm_and_ps(fscal,cutoff_mask);
1071
1072             /* Calculate temporary vectorial force */
1073             tx               = _mm_mul_ps(fscal,dx20);
1074             ty               = _mm_mul_ps(fscal,dy20);
1075             tz               = _mm_mul_ps(fscal,dz20);
1076
1077             /* Update vectorial force */
1078             fix2             = _mm_add_ps(fix2,tx);
1079             fiy2             = _mm_add_ps(fiy2,ty);
1080             fiz2             = _mm_add_ps(fiz2,tz);
1081
1082             fjx0             = _mm_add_ps(fjx0,tx);
1083             fjy0             = _mm_add_ps(fjy0,ty);
1084             fjz0             = _mm_add_ps(fjz0,tz);
1085
1086             }
1087
1088             /**************************
1089              * CALCULATE INTERACTIONS *
1090              **************************/
1091
1092             if (gmx_mm_any_lt(rsq30,rcutoff2))
1093             {
1094
1095             /* Compute parameters for interactions between i and j atoms */
1096             qq30             = _mm_mul_ps(iq3,jq0);
1097
1098             /* REACTION-FIELD ELECTROSTATICS */
1099             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1100
1101             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1102
1103             fscal            = felec;
1104
1105             fscal            = _mm_and_ps(fscal,cutoff_mask);
1106
1107             /* Calculate temporary vectorial force */
1108             tx               = _mm_mul_ps(fscal,dx30);
1109             ty               = _mm_mul_ps(fscal,dy30);
1110             tz               = _mm_mul_ps(fscal,dz30);
1111
1112             /* Update vectorial force */
1113             fix3             = _mm_add_ps(fix3,tx);
1114             fiy3             = _mm_add_ps(fiy3,ty);
1115             fiz3             = _mm_add_ps(fiz3,tz);
1116
1117             fjx0             = _mm_add_ps(fjx0,tx);
1118             fjy0             = _mm_add_ps(fjy0,ty);
1119             fjz0             = _mm_add_ps(fjz0,tz);
1120
1121             }
1122
1123             fjptrA             = f+j_coord_offsetA;
1124             fjptrB             = f+j_coord_offsetB;
1125             fjptrC             = f+j_coord_offsetC;
1126             fjptrD             = f+j_coord_offsetD;
1127
1128             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1129
1130             /* Inner loop uses 138 flops */
1131         }
1132
1133         if(jidx<j_index_end)
1134         {
1135
1136             /* Get j neighbor index, and coordinate index */
1137             jnrlistA         = jjnr[jidx];
1138             jnrlistB         = jjnr[jidx+1];
1139             jnrlistC         = jjnr[jidx+2];
1140             jnrlistD         = jjnr[jidx+3];
1141             /* Sign of each element will be negative for non-real atoms.
1142              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1143              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1144              */
1145             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1146             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1147             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1148             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1149             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1150             j_coord_offsetA  = DIM*jnrA;
1151             j_coord_offsetB  = DIM*jnrB;
1152             j_coord_offsetC  = DIM*jnrC;
1153             j_coord_offsetD  = DIM*jnrD;
1154
1155             /* load j atom coordinates */
1156             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1157                                               x+j_coord_offsetC,x+j_coord_offsetD,
1158                                               &jx0,&jy0,&jz0);
1159
1160             /* Calculate displacement vector */
1161             dx00             = _mm_sub_ps(ix0,jx0);
1162             dy00             = _mm_sub_ps(iy0,jy0);
1163             dz00             = _mm_sub_ps(iz0,jz0);
1164             dx10             = _mm_sub_ps(ix1,jx0);
1165             dy10             = _mm_sub_ps(iy1,jy0);
1166             dz10             = _mm_sub_ps(iz1,jz0);
1167             dx20             = _mm_sub_ps(ix2,jx0);
1168             dy20             = _mm_sub_ps(iy2,jy0);
1169             dz20             = _mm_sub_ps(iz2,jz0);
1170             dx30             = _mm_sub_ps(ix3,jx0);
1171             dy30             = _mm_sub_ps(iy3,jy0);
1172             dz30             = _mm_sub_ps(iz3,jz0);
1173
1174             /* Calculate squared distance and things based on it */
1175             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1176             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1177             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1178             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1179
1180             rinv00           = sse41_invsqrt_f(rsq00);
1181             rinv10           = sse41_invsqrt_f(rsq10);
1182             rinv20           = sse41_invsqrt_f(rsq20);
1183             rinv30           = sse41_invsqrt_f(rsq30);
1184
1185             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1186             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1187             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1188
1189             /* Load parameters for j particles */
1190             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1191                                                               charge+jnrC+0,charge+jnrD+0);
1192             vdwjidx0A        = 2*vdwtype[jnrA+0];
1193             vdwjidx0B        = 2*vdwtype[jnrB+0];
1194             vdwjidx0C        = 2*vdwtype[jnrC+0];
1195             vdwjidx0D        = 2*vdwtype[jnrD+0];
1196
1197             fjx0             = _mm_setzero_ps();
1198             fjy0             = _mm_setzero_ps();
1199             fjz0             = _mm_setzero_ps();
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             r00              = _mm_mul_ps(rsq00,rinv00);
1206             r00              = _mm_andnot_ps(dummy_mask,r00);
1207
1208             /* Compute parameters for interactions between i and j atoms */
1209             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1210                                          vdwparam+vdwioffset0+vdwjidx0B,
1211                                          vdwparam+vdwioffset0+vdwjidx0C,
1212                                          vdwparam+vdwioffset0+vdwjidx0D,
1213                                          &c6_00,&c12_00);
1214
1215             /* Calculate table index by multiplying r with table scale and truncate to integer */
1216             rt               = _mm_mul_ps(r00,vftabscale);
1217             vfitab           = _mm_cvttps_epi32(rt);
1218             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1219             vfitab           = _mm_slli_epi32(vfitab,3);
1220
1221             /* CUBIC SPLINE TABLE DISPERSION */
1222             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1223             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1224             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1225             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1226             _MM_TRANSPOSE4_PS(Y,F,G,H);
1227             Heps             = _mm_mul_ps(vfeps,H);
1228             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1229             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1230             fvdw6            = _mm_mul_ps(c6_00,FF);
1231
1232             /* CUBIC SPLINE TABLE REPULSION */
1233             vfitab           = _mm_add_epi32(vfitab,ifour);
1234             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1235             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1236             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1237             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1238             _MM_TRANSPOSE4_PS(Y,F,G,H);
1239             Heps             = _mm_mul_ps(vfeps,H);
1240             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1241             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1242             fvdw12           = _mm_mul_ps(c12_00,FF);
1243             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1244
1245             fscal            = fvdw;
1246
1247             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1248
1249             /* Calculate temporary vectorial force */
1250             tx               = _mm_mul_ps(fscal,dx00);
1251             ty               = _mm_mul_ps(fscal,dy00);
1252             tz               = _mm_mul_ps(fscal,dz00);
1253
1254             /* Update vectorial force */
1255             fix0             = _mm_add_ps(fix0,tx);
1256             fiy0             = _mm_add_ps(fiy0,ty);
1257             fiz0             = _mm_add_ps(fiz0,tz);
1258
1259             fjx0             = _mm_add_ps(fjx0,tx);
1260             fjy0             = _mm_add_ps(fjy0,ty);
1261             fjz0             = _mm_add_ps(fjz0,tz);
1262
1263             /**************************
1264              * CALCULATE INTERACTIONS *
1265              **************************/
1266
1267             if (gmx_mm_any_lt(rsq10,rcutoff2))
1268             {
1269
1270             /* Compute parameters for interactions between i and j atoms */
1271             qq10             = _mm_mul_ps(iq1,jq0);
1272
1273             /* REACTION-FIELD ELECTROSTATICS */
1274             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1275
1276             cutoff_mask      = _mm_cmplt_ps(rsq10,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,dx10);
1286             ty               = _mm_mul_ps(fscal,dy10);
1287             tz               = _mm_mul_ps(fscal,dz10);
1288
1289             /* Update vectorial force */
1290             fix1             = _mm_add_ps(fix1,tx);
1291             fiy1             = _mm_add_ps(fiy1,ty);
1292             fiz1             = _mm_add_ps(fiz1,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             /**************************
1301              * CALCULATE INTERACTIONS *
1302              **************************/
1303
1304             if (gmx_mm_any_lt(rsq20,rcutoff2))
1305             {
1306
1307             /* Compute parameters for interactions between i and j atoms */
1308             qq20             = _mm_mul_ps(iq2,jq0);
1309
1310             /* REACTION-FIELD ELECTROSTATICS */
1311             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1312
1313             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1314
1315             fscal            = felec;
1316
1317             fscal            = _mm_and_ps(fscal,cutoff_mask);
1318
1319             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1320
1321             /* Calculate temporary vectorial force */
1322             tx               = _mm_mul_ps(fscal,dx20);
1323             ty               = _mm_mul_ps(fscal,dy20);
1324             tz               = _mm_mul_ps(fscal,dz20);
1325
1326             /* Update vectorial force */
1327             fix2             = _mm_add_ps(fix2,tx);
1328             fiy2             = _mm_add_ps(fiy2,ty);
1329             fiz2             = _mm_add_ps(fiz2,tz);
1330
1331             fjx0             = _mm_add_ps(fjx0,tx);
1332             fjy0             = _mm_add_ps(fjy0,ty);
1333             fjz0             = _mm_add_ps(fjz0,tz);
1334
1335             }
1336
1337             /**************************
1338              * CALCULATE INTERACTIONS *
1339              **************************/
1340
1341             if (gmx_mm_any_lt(rsq30,rcutoff2))
1342             {
1343
1344             /* Compute parameters for interactions between i and j atoms */
1345             qq30             = _mm_mul_ps(iq3,jq0);
1346
1347             /* REACTION-FIELD ELECTROSTATICS */
1348             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1349
1350             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1351
1352             fscal            = felec;
1353
1354             fscal            = _mm_and_ps(fscal,cutoff_mask);
1355
1356             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1357
1358             /* Calculate temporary vectorial force */
1359             tx               = _mm_mul_ps(fscal,dx30);
1360             ty               = _mm_mul_ps(fscal,dy30);
1361             tz               = _mm_mul_ps(fscal,dz30);
1362
1363             /* Update vectorial force */
1364             fix3             = _mm_add_ps(fix3,tx);
1365             fiy3             = _mm_add_ps(fiy3,ty);
1366             fiz3             = _mm_add_ps(fiz3,tz);
1367
1368             fjx0             = _mm_add_ps(fjx0,tx);
1369             fjy0             = _mm_add_ps(fjy0,ty);
1370             fjz0             = _mm_add_ps(fjz0,tz);
1371
1372             }
1373
1374             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1375             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1376             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1377             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1378
1379             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1380
1381             /* Inner loop uses 139 flops */
1382         }
1383
1384         /* End of innermost loop */
1385
1386         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1387                                               f+i_coord_offset,fshift+i_shift_offset);
1388
1389         /* Increment number of inner iterations */
1390         inneriter                  += j_index_end - j_index_start;
1391
1392         /* Outer loop uses 24 flops */
1393     }
1394
1395     /* Increment number of outer iterations */
1396     outeriter        += nri;
1397
1398     /* Update outer/inner flops */
1399
1400     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*139);
1401 }