src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_VF_sse4_1_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_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     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 135     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 136     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = jnrC = jnrD = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143     j_coord_offsetC = 0;
 144     j_coord_offsetD = 0;
 145
 146     outeriter        = 0;
 147     inneriter        = 0;
 148
 149     for(iidx=0;iidx<4*DIM;iidx++)
 150     {
 151         scratch[iidx] = 0.0;
 152     }
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 171
 172         fix0             = _mm_setzero_ps();
 173         fiy0             = _mm_setzero_ps();
 174         fiz0             = _mm_setzero_ps();
 175         fix1             = _mm_setzero_ps();
 176         fiy1             = _mm_setzero_ps();
 177         fiz1             = _mm_setzero_ps();
 178         fix2             = _mm_setzero_ps();
 179         fiy2             = _mm_setzero_ps();
 180         fiz2             = _mm_setzero_ps();
 181         fix3             = _mm_setzero_ps();
 182         fiy3             = _mm_setzero_ps();
 183         fiz3             = _mm_setzero_ps();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_ps();
 187         vvdwsum          = _mm_setzero_ps();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             jnrC             = jjnr[jidx+2];
 197             jnrD             = jjnr[jidx+3];
 198             j_coord_offsetA  = DIM*jnrA;
 199             j_coord_offsetB  = DIM*jnrB;
 200             j_coord_offsetC  = DIM*jnrC;
 201             j_coord_offsetD  = DIM*jnrD;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               x+j_coord_offsetC,x+j_coord_offsetD,
 206                                               &jx0,&jy0,&jz0);
 207
 208             /* Calculate displacement vector */
 209             dx00             = _mm_sub_ps(ix0,jx0);
 210             dy00             = _mm_sub_ps(iy0,jy0);
 211             dz00             = _mm_sub_ps(iz0,jz0);
 212             dx10             = _mm_sub_ps(ix1,jx0);
 213             dy10             = _mm_sub_ps(iy1,jy0);
 214             dz10             = _mm_sub_ps(iz1,jz0);
 215             dx20             = _mm_sub_ps(ix2,jx0);
 216             dy20             = _mm_sub_ps(iy2,jy0);
 217             dz20             = _mm_sub_ps(iz2,jz0);
 218             dx30             = _mm_sub_ps(ix3,jx0);
 219             dy30             = _mm_sub_ps(iy3,jy0);
 220             dz30             = _mm_sub_ps(iz3,jz0);
 221
 222             /* Calculate squared distance and things based on it */
 223             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 224             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 225             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 226             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 227
 228             rinv10           = sse41_invsqrt_f(rsq10);
 229             rinv20           = sse41_invsqrt_f(rsq20);
 230             rinv30           = sse41_invsqrt_f(rsq30);
 231
 232             rinvsq00         = sse41_inv_f(rsq00);
 233
 234             /* Load parameters for j particles */
 235             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 236                                                               charge+jnrC+0,charge+jnrD+0);
 237             vdwjidx0A        = 2*vdwtype[jnrA+0];
 238             vdwjidx0B        = 2*vdwtype[jnrB+0];
 239             vdwjidx0C        = 2*vdwtype[jnrC+0];
 240             vdwjidx0D        = 2*vdwtype[jnrD+0];
 241
 242             fjx0             = _mm_setzero_ps();
 243             fjy0             = _mm_setzero_ps();
 244             fjz0             = _mm_setzero_ps();
 245
 246             /**************************
 247              * CALCULATE INTERACTIONS *
 248              **************************/
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,
 253                                          vdwparam+vdwioffset0+vdwjidx0C,
 254                                          vdwparam+vdwioffset0+vdwjidx0D,
 255                                          &c6_00,&c12_00);
 256
 257             /* LENNARD-JONES DISPERSION/REPULSION */
 258
 259             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 260             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 261             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 262             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 263             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 267
 268             fscal            = fvdw;
 269
 270             /* Calculate temporary vectorial force */
 271             tx               = _mm_mul_ps(fscal,dx00);
 272             ty               = _mm_mul_ps(fscal,dy00);
 273             tz               = _mm_mul_ps(fscal,dz00);
 274
 275             /* Update vectorial force */
 276             fix0             = _mm_add_ps(fix0,tx);
 277             fiy0             = _mm_add_ps(fiy0,ty);
 278             fiz0             = _mm_add_ps(fiz0,tz);
 279
 280             fjx0             = _mm_add_ps(fjx0,tx);
 281             fjy0             = _mm_add_ps(fjy0,ty);
 282             fjz0             = _mm_add_ps(fjz0,tz);
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             r10              = _mm_mul_ps(rsq10,rinv10);
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq10             = _mm_mul_ps(iq1,jq0);
 292
 293             /* Calculate table index by multiplying r with table scale and truncate to integer */
 294             rt               = _mm_mul_ps(r10,vftabscale);
 295             vfitab           = _mm_cvttps_epi32(rt);
 296             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 297             vfitab           = _mm_slli_epi32(vfitab,2);
 298
 299             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 300             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 301             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 302             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 303             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 304             _MM_TRANSPOSE4_PS(Y,F,G,H);
 305             Heps             = _mm_mul_ps(vfeps,H);
 306             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 307             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 308             velec            = _mm_mul_ps(qq10,VV);
 309             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 310             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_ps(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm_mul_ps(fscal,dx10);
 319             ty               = _mm_mul_ps(fscal,dy10);
 320             tz               = _mm_mul_ps(fscal,dz10);
 321
 322             /* Update vectorial force */
 323             fix1             = _mm_add_ps(fix1,tx);
 324             fiy1             = _mm_add_ps(fiy1,ty);
 325             fiz1             = _mm_add_ps(fiz1,tz);
 326
 327             fjx0             = _mm_add_ps(fjx0,tx);
 328             fjy0             = _mm_add_ps(fjy0,ty);
 329             fjz0             = _mm_add_ps(fjz0,tz);
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r20              = _mm_mul_ps(rsq20,rinv20);
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq20             = _mm_mul_ps(iq2,jq0);
 339
 340             /* Calculate table index by multiplying r with table scale and truncate to integer */
 341             rt               = _mm_mul_ps(r20,vftabscale);
 342             vfitab           = _mm_cvttps_epi32(rt);
 343             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 344             vfitab           = _mm_slli_epi32(vfitab,2);
 345
 346             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 347             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 348             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 349             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 350             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 351             _MM_TRANSPOSE4_PS(Y,F,G,H);
 352             Heps             = _mm_mul_ps(vfeps,H);
 353             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 354             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 355             velec            = _mm_mul_ps(qq20,VV);
 356             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 357             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm_add_ps(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm_mul_ps(fscal,dx20);
 366             ty               = _mm_mul_ps(fscal,dy20);
 367             tz               = _mm_mul_ps(fscal,dz20);
 368
 369             /* Update vectorial force */
 370             fix2             = _mm_add_ps(fix2,tx);
 371             fiy2             = _mm_add_ps(fiy2,ty);
 372             fiz2             = _mm_add_ps(fiz2,tz);
 373
 374             fjx0             = _mm_add_ps(fjx0,tx);
 375             fjy0             = _mm_add_ps(fjy0,ty);
 376             fjz0             = _mm_add_ps(fjz0,tz);
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             r30              = _mm_mul_ps(rsq30,rinv30);
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq30             = _mm_mul_ps(iq3,jq0);
 386
 387             /* Calculate table index by multiplying r with table scale and truncate to integer */
 388             rt               = _mm_mul_ps(r30,vftabscale);
 389             vfitab           = _mm_cvttps_epi32(rt);
 390             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 391             vfitab           = _mm_slli_epi32(vfitab,2);
 392
 393             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 394             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 395             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 396             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 397             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 398             _MM_TRANSPOSE4_PS(Y,F,G,H);
 399             Heps             = _mm_mul_ps(vfeps,H);
 400             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 401             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 402             velec            = _mm_mul_ps(qq30,VV);
 403             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 404             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velecsum         = _mm_add_ps(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm_mul_ps(fscal,dx30);
 413             ty               = _mm_mul_ps(fscal,dy30);
 414             tz               = _mm_mul_ps(fscal,dz30);
 415
 416             /* Update vectorial force */
 417             fix3             = _mm_add_ps(fix3,tx);
 418             fiy3             = _mm_add_ps(fiy3,ty);
 419             fiz3             = _mm_add_ps(fiz3,tz);
 420
 421             fjx0             = _mm_add_ps(fjx0,tx);
 422             fjy0             = _mm_add_ps(fjy0,ty);
 423             fjz0             = _mm_add_ps(fjz0,tz);
 424
 425             fjptrA             = f+j_coord_offsetA;
 426             fjptrB             = f+j_coord_offsetB;
 427             fjptrC             = f+j_coord_offsetC;
 428             fjptrD             = f+j_coord_offsetD;
 429
 430             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 431
 432             /* Inner loop uses 161 flops */
 433         }
 434
 435         if(jidx<j_index_end)
 436         {
 437
 438             /* Get j neighbor index, and coordinate index */
 439             jnrlistA         = jjnr[jidx];
 440             jnrlistB         = jjnr[jidx+1];
 441             jnrlistC         = jjnr[jidx+2];
 442             jnrlistD         = jjnr[jidx+3];
 443             /* Sign of each element will be negative for non-real atoms.
 444              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 445              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 446              */
 447             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 448             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 449             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 450             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 451             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 452             j_coord_offsetA  = DIM*jnrA;
 453             j_coord_offsetB  = DIM*jnrB;
 454             j_coord_offsetC  = DIM*jnrC;
 455             j_coord_offsetD  = DIM*jnrD;
 456
 457             /* load j atom coordinates */
 458             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 459                                               x+j_coord_offsetC,x+j_coord_offsetD,
 460                                               &jx0,&jy0,&jz0);
 461
 462             /* Calculate displacement vector */
 463             dx00             = _mm_sub_ps(ix0,jx0);
 464             dy00             = _mm_sub_ps(iy0,jy0);
 465             dz00             = _mm_sub_ps(iz0,jz0);
 466             dx10             = _mm_sub_ps(ix1,jx0);
 467             dy10             = _mm_sub_ps(iy1,jy0);
 468             dz10             = _mm_sub_ps(iz1,jz0);
 469             dx20             = _mm_sub_ps(ix2,jx0);
 470             dy20             = _mm_sub_ps(iy2,jy0);
 471             dz20             = _mm_sub_ps(iz2,jz0);
 472             dx30             = _mm_sub_ps(ix3,jx0);
 473             dy30             = _mm_sub_ps(iy3,jy0);
 474             dz30             = _mm_sub_ps(iz3,jz0);
 475
 476             /* Calculate squared distance and things based on it */
 477             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 478             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 479             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 480             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 481
 482             rinv10           = sse41_invsqrt_f(rsq10);
 483             rinv20           = sse41_invsqrt_f(rsq20);
 484             rinv30           = sse41_invsqrt_f(rsq30);
 485
 486             rinvsq00         = sse41_inv_f(rsq00);
 487
 488             /* Load parameters for j particles */
 489             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 490                                                               charge+jnrC+0,charge+jnrD+0);
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493             vdwjidx0C        = 2*vdwtype[jnrC+0];
 494             vdwjidx0D        = 2*vdwtype[jnrD+0];
 495
 496             fjx0             = _mm_setzero_ps();
 497             fjy0             = _mm_setzero_ps();
 498             fjz0             = _mm_setzero_ps();
 499
 500             /**************************
 501              * CALCULATE INTERACTIONS *
 502              **************************/
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 506                                          vdwparam+vdwioffset0+vdwjidx0B,
 507                                          vdwparam+vdwioffset0+vdwjidx0C,
 508                                          vdwparam+vdwioffset0+vdwjidx0D,
 509                                          &c6_00,&c12_00);
 510
 511             /* LENNARD-JONES DISPERSION/REPULSION */
 512
 513             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 514             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 515             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 516             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 517             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 521             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 522
 523             fscal            = fvdw;
 524
 525             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 526
 527             /* Calculate temporary vectorial force */
 528             tx               = _mm_mul_ps(fscal,dx00);
 529             ty               = _mm_mul_ps(fscal,dy00);
 530             tz               = _mm_mul_ps(fscal,dz00);
 531
 532             /* Update vectorial force */
 533             fix0             = _mm_add_ps(fix0,tx);
 534             fiy0             = _mm_add_ps(fiy0,ty);
 535             fiz0             = _mm_add_ps(fiz0,tz);
 536
 537             fjx0             = _mm_add_ps(fjx0,tx);
 538             fjy0             = _mm_add_ps(fjy0,ty);
 539             fjz0             = _mm_add_ps(fjz0,tz);
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             r10              = _mm_mul_ps(rsq10,rinv10);
 546             r10              = _mm_andnot_ps(dummy_mask,r10);
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq10             = _mm_mul_ps(iq1,jq0);
 550
 551             /* Calculate table index by multiplying r with table scale and truncate to integer */
 552             rt               = _mm_mul_ps(r10,vftabscale);
 553             vfitab           = _mm_cvttps_epi32(rt);
 554             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 555             vfitab           = _mm_slli_epi32(vfitab,2);
 556
 557             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 558             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 559             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 560             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 561             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 562             _MM_TRANSPOSE4_PS(Y,F,G,H);
 563             Heps             = _mm_mul_ps(vfeps,H);
 564             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 565             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 566             velec            = _mm_mul_ps(qq10,VV);
 567             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 568             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 569
 570             /* Update potential sum for this i atom from the interaction with this j atom. */
 571             velec            = _mm_andnot_ps(dummy_mask,velec);
 572             velecsum         = _mm_add_ps(velecsum,velec);
 573
 574             fscal            = felec;
 575
 576             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 577
 578             /* Calculate temporary vectorial force */
 579             tx               = _mm_mul_ps(fscal,dx10);
 580             ty               = _mm_mul_ps(fscal,dy10);
 581             tz               = _mm_mul_ps(fscal,dz10);
 582
 583             /* Update vectorial force */
 584             fix1             = _mm_add_ps(fix1,tx);
 585             fiy1             = _mm_add_ps(fiy1,ty);
 586             fiz1             = _mm_add_ps(fiz1,tz);
 587
 588             fjx0             = _mm_add_ps(fjx0,tx);
 589             fjy0             = _mm_add_ps(fjy0,ty);
 590             fjz0             = _mm_add_ps(fjz0,tz);
 591
 592             /**************************
 593              * CALCULATE INTERACTIONS *
 594              **************************/
 595
 596             r20              = _mm_mul_ps(rsq20,rinv20);
 597             r20              = _mm_andnot_ps(dummy_mask,r20);
 598
 599             /* Compute parameters for interactions between i and j atoms */
 600             qq20             = _mm_mul_ps(iq2,jq0);
 601
 602             /* Calculate table index by multiplying r with table scale and truncate to integer */
 603             rt               = _mm_mul_ps(r20,vftabscale);
 604             vfitab           = _mm_cvttps_epi32(rt);
 605             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 606             vfitab           = _mm_slli_epi32(vfitab,2);
 607
 608             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 609             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 610             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 611             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 612             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 613             _MM_TRANSPOSE4_PS(Y,F,G,H);
 614             Heps             = _mm_mul_ps(vfeps,H);
 615             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 616             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 617             velec            = _mm_mul_ps(qq20,VV);
 618             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 619             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velec            = _mm_andnot_ps(dummy_mask,velec);
 623             velecsum         = _mm_add_ps(velecsum,velec);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 628
 629             /* Calculate temporary vectorial force */
 630             tx               = _mm_mul_ps(fscal,dx20);
 631             ty               = _mm_mul_ps(fscal,dy20);
 632             tz               = _mm_mul_ps(fscal,dz20);
 633
 634             /* Update vectorial force */
 635             fix2             = _mm_add_ps(fix2,tx);
 636             fiy2             = _mm_add_ps(fiy2,ty);
 637             fiz2             = _mm_add_ps(fiz2,tz);
 638
 639             fjx0             = _mm_add_ps(fjx0,tx);
 640             fjy0             = _mm_add_ps(fjy0,ty);
 641             fjz0             = _mm_add_ps(fjz0,tz);
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             r30              = _mm_mul_ps(rsq30,rinv30);
 648             r30              = _mm_andnot_ps(dummy_mask,r30);
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq30             = _mm_mul_ps(iq3,jq0);
 652
 653             /* Calculate table index by multiplying r with table scale and truncate to integer */
 654             rt               = _mm_mul_ps(r30,vftabscale);
 655             vfitab           = _mm_cvttps_epi32(rt);
 656             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 657             vfitab           = _mm_slli_epi32(vfitab,2);
 658
 659             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 660             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 661             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 662             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 663             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 664             _MM_TRANSPOSE4_PS(Y,F,G,H);
 665             Heps             = _mm_mul_ps(vfeps,H);
 666             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 667             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 668             velec            = _mm_mul_ps(qq30,VV);
 669             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 670             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 671
 672             /* Update potential sum for this i atom from the interaction with this j atom. */
 673             velec            = _mm_andnot_ps(dummy_mask,velec);
 674             velecsum         = _mm_add_ps(velecsum,velec);
 675
 676             fscal            = felec;
 677
 678             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 679
 680             /* Calculate temporary vectorial force */
 681             tx               = _mm_mul_ps(fscal,dx30);
 682             ty               = _mm_mul_ps(fscal,dy30);
 683             tz               = _mm_mul_ps(fscal,dz30);
 684
 685             /* Update vectorial force */
 686             fix3             = _mm_add_ps(fix3,tx);
 687             fiy3             = _mm_add_ps(fiy3,ty);
 688             fiz3             = _mm_add_ps(fiz3,tz);
 689
 690             fjx0             = _mm_add_ps(fjx0,tx);
 691             fjy0             = _mm_add_ps(fjy0,ty);
 692             fjz0             = _mm_add_ps(fjz0,tz);
 693
 694             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 695             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 696             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 697             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 698
 699             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 700
 701             /* Inner loop uses 164 flops */
 702         }
 703
 704         /* End of innermost loop */
 705
 706         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 707                                               f+i_coord_offset,fshift+i_shift_offset);
 708
 709         ggid                        = gid[iidx];
 710         /* Update potential energies */
 711         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 712         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 713
 714         /* Increment number of inner iterations */
 715         inneriter                  += j_index_end - j_index_start;
 716
 717         /* Outer loop uses 26 flops */
 718     }
 719
 720     /* Increment number of outer iterations */
 721     outeriter        += nri;
 722
 723     /* Update outer/inner flops */
 724
 725     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*164);
 726 }
 727 /*
 728  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse4_1_single
 729  * Electrostatics interaction: CubicSplineTable
 730  * VdW interaction:            LennardJones
 731  * Geometry:                   Water4-Particle
 732  * Calculate force/pot:        Force
 733  */
 734 void
 735 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_sse4_1_single
 736                     (t_nblist                    * gmx_restrict       nlist,
 737                      rvec                        * gmx_restrict          xx,
 738                      rvec                        * gmx_restrict          ff,
 739                      struct t_forcerec           * gmx_restrict          fr,
 740                      t_mdatoms                   * gmx_restrict     mdatoms,
 741                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 742                      t_nrnb                      * gmx_restrict        nrnb)
 743 {
 744     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 745      * just 0 for non-waters.
 746      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 747      * jnr indices corresponding to data put in the four positions in the SIMD register.
 748      */
 749     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 750     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 751     int              jnrA,jnrB,jnrC,jnrD;
 752     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 753     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 754     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 755     real             rcutoff_scalar;
 756     real             *shiftvec,*fshift,*x,*f;
 757     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 758     real             scratch[4*DIM];
 759     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 760     int              vdwioffset0;
 761     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 762     int              vdwioffset1;
 763     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 764     int              vdwioffset2;
 765     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 766     int              vdwioffset3;
 767     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 768     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 769     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 770     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 771     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 772     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 773     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 774     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 775     real             *charge;
 776     int              nvdwtype;
 777     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 778     int              *vdwtype;
 779     real             *vdwparam;
 780     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 781     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 782     __m128i          vfitab;
 783     __m128i          ifour       = _mm_set1_epi32(4);
 784     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 785     real             *vftab;
 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->ic->epsfac);
 802     charge           = mdatoms->chargeA;
 803     nvdwtype         = fr->ntype;
 804     vdwparam         = fr->nbfp;
 805     vdwtype          = mdatoms->typeA;
 806
 807     vftab            = kernel_data->table_elec->data;
 808     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 809
 810     /* Setup water-specific parameters */
 811     inr              = nlist->iinr[0];
 812     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 813     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 814     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 815     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 816
 817     /* Avoid stupid compiler warnings */
 818     jnrA = jnrB = jnrC = jnrD = 0;
 819     j_coord_offsetA = 0;
 820     j_coord_offsetB = 0;
 821     j_coord_offsetC = 0;
 822     j_coord_offsetD = 0;
 823
 824     outeriter        = 0;
 825     inneriter        = 0;
 826
 827     for(iidx=0;iidx<4*DIM;iidx++)
 828     {
 829         scratch[iidx] = 0.0;
 830     }
 831
 832     /* Start outer loop over neighborlists */
 833     for(iidx=0; iidx<nri; iidx++)
 834     {
 835         /* Load shift vector for this list */
 836         i_shift_offset   = DIM*shiftidx[iidx];
 837
 838         /* Load limits for loop over neighbors */
 839         j_index_start    = jindex[iidx];
 840         j_index_end      = jindex[iidx+1];
 841
 842         /* Get outer coordinate index */
 843         inr              = iinr[iidx];
 844         i_coord_offset   = DIM*inr;
 845
 846         /* Load i particle coords and add shift vector */
 847         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 848                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 849
 850         fix0             = _mm_setzero_ps();
 851         fiy0             = _mm_setzero_ps();
 852         fiz0             = _mm_setzero_ps();
 853         fix1             = _mm_setzero_ps();
 854         fiy1             = _mm_setzero_ps();
 855         fiz1             = _mm_setzero_ps();
 856         fix2             = _mm_setzero_ps();
 857         fiy2             = _mm_setzero_ps();
 858         fiz2             = _mm_setzero_ps();
 859         fix3             = _mm_setzero_ps();
 860         fiy3             = _mm_setzero_ps();
 861         fiz3             = _mm_setzero_ps();
 862
 863         /* Start inner kernel loop */
 864         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 865         {
 866
 867             /* Get j neighbor index, and coordinate index */
 868             jnrA             = jjnr[jidx];
 869             jnrB             = jjnr[jidx+1];
 870             jnrC             = jjnr[jidx+2];
 871             jnrD             = jjnr[jidx+3];
 872             j_coord_offsetA  = DIM*jnrA;
 873             j_coord_offsetB  = DIM*jnrB;
 874             j_coord_offsetC  = DIM*jnrC;
 875             j_coord_offsetD  = DIM*jnrD;
 876
 877             /* load j atom coordinates */
 878             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 879                                               x+j_coord_offsetC,x+j_coord_offsetD,
 880                                               &jx0,&jy0,&jz0);
 881
 882             /* Calculate displacement vector */
 883             dx00             = _mm_sub_ps(ix0,jx0);
 884             dy00             = _mm_sub_ps(iy0,jy0);
 885             dz00             = _mm_sub_ps(iz0,jz0);
 886             dx10             = _mm_sub_ps(ix1,jx0);
 887             dy10             = _mm_sub_ps(iy1,jy0);
 888             dz10             = _mm_sub_ps(iz1,jz0);
 889             dx20             = _mm_sub_ps(ix2,jx0);
 890             dy20             = _mm_sub_ps(iy2,jy0);
 891             dz20             = _mm_sub_ps(iz2,jz0);
 892             dx30             = _mm_sub_ps(ix3,jx0);
 893             dy30             = _mm_sub_ps(iy3,jy0);
 894             dz30             = _mm_sub_ps(iz3,jz0);
 895
 896             /* Calculate squared distance and things based on it */
 897             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 898             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 899             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 900             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 901
 902             rinv10           = sse41_invsqrt_f(rsq10);
 903             rinv20           = sse41_invsqrt_f(rsq20);
 904             rinv30           = sse41_invsqrt_f(rsq30);
 905
 906             rinvsq00         = sse41_inv_f(rsq00);
 907
 908             /* Load parameters for j particles */
 909             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 910                                                               charge+jnrC+0,charge+jnrD+0);
 911             vdwjidx0A        = 2*vdwtype[jnrA+0];
 912             vdwjidx0B        = 2*vdwtype[jnrB+0];
 913             vdwjidx0C        = 2*vdwtype[jnrC+0];
 914             vdwjidx0D        = 2*vdwtype[jnrD+0];
 915
 916             fjx0             = _mm_setzero_ps();
 917             fjy0             = _mm_setzero_ps();
 918             fjz0             = _mm_setzero_ps();
 919
 920             /**************************
 921              * CALCULATE INTERACTIONS *
 922              **************************/
 923
 924             /* Compute parameters for interactions between i and j atoms */
 925             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 926                                          vdwparam+vdwioffset0+vdwjidx0B,
 927                                          vdwparam+vdwioffset0+vdwjidx0C,
 928                                          vdwparam+vdwioffset0+vdwjidx0D,
 929                                          &c6_00,&c12_00);
 930
 931             /* LENNARD-JONES DISPERSION/REPULSION */
 932
 933             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 934             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 935
 936             fscal            = fvdw;
 937
 938             /* Calculate temporary vectorial force */
 939             tx               = _mm_mul_ps(fscal,dx00);
 940             ty               = _mm_mul_ps(fscal,dy00);
 941             tz               = _mm_mul_ps(fscal,dz00);
 942
 943             /* Update vectorial force */
 944             fix0             = _mm_add_ps(fix0,tx);
 945             fiy0             = _mm_add_ps(fiy0,ty);
 946             fiz0             = _mm_add_ps(fiz0,tz);
 947
 948             fjx0             = _mm_add_ps(fjx0,tx);
 949             fjy0             = _mm_add_ps(fjy0,ty);
 950             fjz0             = _mm_add_ps(fjz0,tz);
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             r10              = _mm_mul_ps(rsq10,rinv10);
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq10             = _mm_mul_ps(iq1,jq0);
 960
 961             /* Calculate table index by multiplying r with table scale and truncate to integer */
 962             rt               = _mm_mul_ps(r10,vftabscale);
 963             vfitab           = _mm_cvttps_epi32(rt);
 964             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 965             vfitab           = _mm_slli_epi32(vfitab,2);
 966
 967             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 968             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 969             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 970             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 971             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 972             _MM_TRANSPOSE4_PS(Y,F,G,H);
 973             Heps             = _mm_mul_ps(vfeps,H);
 974             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 975             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 976             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 977
 978             fscal            = felec;
 979
 980             /* Calculate temporary vectorial force */
 981             tx               = _mm_mul_ps(fscal,dx10);
 982             ty               = _mm_mul_ps(fscal,dy10);
 983             tz               = _mm_mul_ps(fscal,dz10);
 984
 985             /* Update vectorial force */
 986             fix1             = _mm_add_ps(fix1,tx);
 987             fiy1             = _mm_add_ps(fiy1,ty);
 988             fiz1             = _mm_add_ps(fiz1,tz);
 989
 990             fjx0             = _mm_add_ps(fjx0,tx);
 991             fjy0             = _mm_add_ps(fjy0,ty);
 992             fjz0             = _mm_add_ps(fjz0,tz);
 993
 994             /**************************
 995              * CALCULATE INTERACTIONS *
 996              **************************/
 997
 998             r20              = _mm_mul_ps(rsq20,rinv20);
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq20             = _mm_mul_ps(iq2,jq0);
1002
1003             /* Calculate table index by multiplying r with table scale and truncate to integer */
1004             rt               = _mm_mul_ps(r20,vftabscale);
1005             vfitab           = _mm_cvttps_epi32(rt);
1006             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1007             vfitab           = _mm_slli_epi32(vfitab,2);
1008
1009             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1010             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1011             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1012             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1013             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1014             _MM_TRANSPOSE4_PS(Y,F,G,H);
1015             Heps             = _mm_mul_ps(vfeps,H);
1016             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1017             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1018             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1019
1020             fscal            = felec;
1021
1022             /* Calculate temporary vectorial force */
1023             tx               = _mm_mul_ps(fscal,dx20);
1024             ty               = _mm_mul_ps(fscal,dy20);
1025             tz               = _mm_mul_ps(fscal,dz20);
1026
1027             /* Update vectorial force */
1028             fix2             = _mm_add_ps(fix2,tx);
1029             fiy2             = _mm_add_ps(fiy2,ty);
1030             fiz2             = _mm_add_ps(fiz2,tz);
1031
1032             fjx0             = _mm_add_ps(fjx0,tx);
1033             fjy0             = _mm_add_ps(fjy0,ty);
1034             fjz0             = _mm_add_ps(fjz0,tz);
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             r30              = _mm_mul_ps(rsq30,rinv30);
1041
1042             /* Compute parameters for interactions between i and j atoms */
1043             qq30             = _mm_mul_ps(iq3,jq0);
1044
1045             /* Calculate table index by multiplying r with table scale and truncate to integer */
1046             rt               = _mm_mul_ps(r30,vftabscale);
1047             vfitab           = _mm_cvttps_epi32(rt);
1048             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1049             vfitab           = _mm_slli_epi32(vfitab,2);
1050
1051             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1052             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1053             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1054             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1055             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1056             _MM_TRANSPOSE4_PS(Y,F,G,H);
1057             Heps             = _mm_mul_ps(vfeps,H);
1058             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1059             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1060             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1061
1062             fscal            = felec;
1063
1064             /* Calculate temporary vectorial force */
1065             tx               = _mm_mul_ps(fscal,dx30);
1066             ty               = _mm_mul_ps(fscal,dy30);
1067             tz               = _mm_mul_ps(fscal,dz30);
1068
1069             /* Update vectorial force */
1070             fix3             = _mm_add_ps(fix3,tx);
1071             fiy3             = _mm_add_ps(fiy3,ty);
1072             fiz3             = _mm_add_ps(fiz3,tz);
1073
1074             fjx0             = _mm_add_ps(fjx0,tx);
1075             fjy0             = _mm_add_ps(fjy0,ty);
1076             fjz0             = _mm_add_ps(fjz0,tz);
1077
1078             fjptrA             = f+j_coord_offsetA;
1079             fjptrB             = f+j_coord_offsetB;
1080             fjptrC             = f+j_coord_offsetC;
1081             fjptrD             = f+j_coord_offsetD;
1082
1083             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1084
1085             /* Inner loop uses 144 flops */
1086         }
1087
1088         if(jidx<j_index_end)
1089         {
1090
1091             /* Get j neighbor index, and coordinate index */
1092             jnrlistA         = jjnr[jidx];
1093             jnrlistB         = jjnr[jidx+1];
1094             jnrlistC         = jjnr[jidx+2];
1095             jnrlistD         = jjnr[jidx+3];
1096             /* Sign of each element will be negative for non-real atoms.
1097              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1098              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1099              */
1100             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1101             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1102             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1103             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1104             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1105             j_coord_offsetA  = DIM*jnrA;
1106             j_coord_offsetB  = DIM*jnrB;
1107             j_coord_offsetC  = DIM*jnrC;
1108             j_coord_offsetD  = DIM*jnrD;
1109
1110             /* load j atom coordinates */
1111             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1112                                               x+j_coord_offsetC,x+j_coord_offsetD,
1113                                               &jx0,&jy0,&jz0);
1114
1115             /* Calculate displacement vector */
1116             dx00             = _mm_sub_ps(ix0,jx0);
1117             dy00             = _mm_sub_ps(iy0,jy0);
1118             dz00             = _mm_sub_ps(iz0,jz0);
1119             dx10             = _mm_sub_ps(ix1,jx0);
1120             dy10             = _mm_sub_ps(iy1,jy0);
1121             dz10             = _mm_sub_ps(iz1,jz0);
1122             dx20             = _mm_sub_ps(ix2,jx0);
1123             dy20             = _mm_sub_ps(iy2,jy0);
1124             dz20             = _mm_sub_ps(iz2,jz0);
1125             dx30             = _mm_sub_ps(ix3,jx0);
1126             dy30             = _mm_sub_ps(iy3,jy0);
1127             dz30             = _mm_sub_ps(iz3,jz0);
1128
1129             /* Calculate squared distance and things based on it */
1130             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1131             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1132             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1133             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1134
1135             rinv10           = sse41_invsqrt_f(rsq10);
1136             rinv20           = sse41_invsqrt_f(rsq20);
1137             rinv30           = sse41_invsqrt_f(rsq30);
1138
1139             rinvsq00         = sse41_inv_f(rsq00);
1140
1141             /* Load parameters for j particles */
1142             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1143                                                               charge+jnrC+0,charge+jnrD+0);
1144             vdwjidx0A        = 2*vdwtype[jnrA+0];
1145             vdwjidx0B        = 2*vdwtype[jnrB+0];
1146             vdwjidx0C        = 2*vdwtype[jnrC+0];
1147             vdwjidx0D        = 2*vdwtype[jnrD+0];
1148
1149             fjx0             = _mm_setzero_ps();
1150             fjy0             = _mm_setzero_ps();
1151             fjz0             = _mm_setzero_ps();
1152
1153             /**************************
1154              * CALCULATE INTERACTIONS *
1155              **************************/
1156
1157             /* Compute parameters for interactions between i and j atoms */
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             /* LENNARD-JONES DISPERSION/REPULSION */
1165
1166             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1167             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1168
1169             fscal            = fvdw;
1170
1171             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1172
1173             /* Calculate temporary vectorial force */
1174             tx               = _mm_mul_ps(fscal,dx00);
1175             ty               = _mm_mul_ps(fscal,dy00);
1176             tz               = _mm_mul_ps(fscal,dz00);
1177
1178             /* Update vectorial force */
1179             fix0             = _mm_add_ps(fix0,tx);
1180             fiy0             = _mm_add_ps(fiy0,ty);
1181             fiz0             = _mm_add_ps(fiz0,tz);
1182
1183             fjx0             = _mm_add_ps(fjx0,tx);
1184             fjy0             = _mm_add_ps(fjy0,ty);
1185             fjz0             = _mm_add_ps(fjz0,tz);
1186
1187             /**************************
1188              * CALCULATE INTERACTIONS *
1189              **************************/
1190
1191             r10              = _mm_mul_ps(rsq10,rinv10);
1192             r10              = _mm_andnot_ps(dummy_mask,r10);
1193
1194             /* Compute parameters for interactions between i and j atoms */
1195             qq10             = _mm_mul_ps(iq1,jq0);
1196
1197             /* Calculate table index by multiplying r with table scale and truncate to integer */
1198             rt               = _mm_mul_ps(r10,vftabscale);
1199             vfitab           = _mm_cvttps_epi32(rt);
1200             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1201             vfitab           = _mm_slli_epi32(vfitab,2);
1202
1203             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1204             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1205             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1206             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1207             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1208             _MM_TRANSPOSE4_PS(Y,F,G,H);
1209             Heps             = _mm_mul_ps(vfeps,H);
1210             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1211             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1212             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1213
1214             fscal            = felec;
1215
1216             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1217
1218             /* Calculate temporary vectorial force */
1219             tx               = _mm_mul_ps(fscal,dx10);
1220             ty               = _mm_mul_ps(fscal,dy10);
1221             tz               = _mm_mul_ps(fscal,dz10);
1222
1223             /* Update vectorial force */
1224             fix1             = _mm_add_ps(fix1,tx);
1225             fiy1             = _mm_add_ps(fiy1,ty);
1226             fiz1             = _mm_add_ps(fiz1,tz);
1227
1228             fjx0             = _mm_add_ps(fjx0,tx);
1229             fjy0             = _mm_add_ps(fjy0,ty);
1230             fjz0             = _mm_add_ps(fjz0,tz);
1231
1232             /**************************
1233              * CALCULATE INTERACTIONS *
1234              **************************/
1235
1236             r20              = _mm_mul_ps(rsq20,rinv20);
1237             r20              = _mm_andnot_ps(dummy_mask,r20);
1238
1239             /* Compute parameters for interactions between i and j atoms */
1240             qq20             = _mm_mul_ps(iq2,jq0);
1241
1242             /* Calculate table index by multiplying r with table scale and truncate to integer */
1243             rt               = _mm_mul_ps(r20,vftabscale);
1244             vfitab           = _mm_cvttps_epi32(rt);
1245             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1246             vfitab           = _mm_slli_epi32(vfitab,2);
1247
1248             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1249             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1250             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1251             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1252             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1253             _MM_TRANSPOSE4_PS(Y,F,G,H);
1254             Heps             = _mm_mul_ps(vfeps,H);
1255             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1256             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1257             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1258
1259             fscal            = felec;
1260
1261             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1262
1263             /* Calculate temporary vectorial force */
1264             tx               = _mm_mul_ps(fscal,dx20);
1265             ty               = _mm_mul_ps(fscal,dy20);
1266             tz               = _mm_mul_ps(fscal,dz20);
1267
1268             /* Update vectorial force */
1269             fix2             = _mm_add_ps(fix2,tx);
1270             fiy2             = _mm_add_ps(fiy2,ty);
1271             fiz2             = _mm_add_ps(fiz2,tz);
1272
1273             fjx0             = _mm_add_ps(fjx0,tx);
1274             fjy0             = _mm_add_ps(fjy0,ty);
1275             fjz0             = _mm_add_ps(fjz0,tz);
1276
1277             /**************************
1278              * CALCULATE INTERACTIONS *
1279              **************************/
1280
1281             r30              = _mm_mul_ps(rsq30,rinv30);
1282             r30              = _mm_andnot_ps(dummy_mask,r30);
1283
1284             /* Compute parameters for interactions between i and j atoms */
1285             qq30             = _mm_mul_ps(iq3,jq0);
1286
1287             /* Calculate table index by multiplying r with table scale and truncate to integer */
1288             rt               = _mm_mul_ps(r30,vftabscale);
1289             vfitab           = _mm_cvttps_epi32(rt);
1290             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1291             vfitab           = _mm_slli_epi32(vfitab,2);
1292
1293             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1294             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1295             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1296             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1297             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1298             _MM_TRANSPOSE4_PS(Y,F,G,H);
1299             Heps             = _mm_mul_ps(vfeps,H);
1300             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1301             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1302             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1303
1304             fscal            = felec;
1305
1306             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1307
1308             /* Calculate temporary vectorial force */
1309             tx               = _mm_mul_ps(fscal,dx30);
1310             ty               = _mm_mul_ps(fscal,dy30);
1311             tz               = _mm_mul_ps(fscal,dz30);
1312
1313             /* Update vectorial force */
1314             fix3             = _mm_add_ps(fix3,tx);
1315             fiy3             = _mm_add_ps(fiy3,ty);
1316             fiz3             = _mm_add_ps(fiz3,tz);
1317
1318             fjx0             = _mm_add_ps(fjx0,tx);
1319             fjy0             = _mm_add_ps(fjy0,ty);
1320             fjz0             = _mm_add_ps(fjz0,tz);
1321
1322             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1323             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1324             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1325             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1326
1327             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1328
1329             /* Inner loop uses 147 flops */
1330         }
1331
1332         /* End of innermost loop */
1333
1334         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1335                                               f+i_coord_offset,fshift+i_shift_offset);
1336
1337         /* Increment number of inner iterations */
1338         inneriter                  += j_index_end - j_index_start;
1339
1340         /* Outer loop uses 24 flops */
1341     }
1342
1343     /* Increment number of outer iterations */
1344     outeriter        += nri;
1345
1346     /* Update outer/inner flops */
1347
1348     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1349 }