src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_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_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_single
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse4_1_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 100     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m128           dummy_mask,cutoff_mask;
 106     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 107     __m128           one     = _mm_set1_ps(1.0);
 108     __m128           two     = _mm_set1_ps(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_ps(fr->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_elec_vdw->data;
 127     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 132     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 133     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = jnrC = jnrD = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140     j_coord_offsetC = 0;
 141     j_coord_offsetD = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     for(iidx=0;iidx<4*DIM;iidx++)
 147     {
 148         scratch[iidx] = 0.0;
 149     }
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 167                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 168
 169         fix0             = _mm_setzero_ps();
 170         fiy0             = _mm_setzero_ps();
 171         fiz0             = _mm_setzero_ps();
 172         fix1             = _mm_setzero_ps();
 173         fiy1             = _mm_setzero_ps();
 174         fiz1             = _mm_setzero_ps();
 175         fix2             = _mm_setzero_ps();
 176         fiy2             = _mm_setzero_ps();
 177         fiz2             = _mm_setzero_ps();
 178
 179         /* Reset potential sums */
 180         velecsum         = _mm_setzero_ps();
 181         vvdwsum          = _mm_setzero_ps();
 182
 183         /* Start inner kernel loop */
 184         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 185         {
 186
 187             /* Get j neighbor index, and coordinate index */
 188             jnrA             = jjnr[jidx];
 189             jnrB             = jjnr[jidx+1];
 190             jnrC             = jjnr[jidx+2];
 191             jnrD             = jjnr[jidx+3];
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               x+j_coord_offsetC,x+j_coord_offsetD,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_ps(ix0,jx0);
 204             dy00             = _mm_sub_ps(iy0,jy0);
 205             dz00             = _mm_sub_ps(iz0,jz0);
 206             dx10             = _mm_sub_ps(ix1,jx0);
 207             dy10             = _mm_sub_ps(iy1,jy0);
 208             dz10             = _mm_sub_ps(iz1,jz0);
 209             dx20             = _mm_sub_ps(ix2,jx0);
 210             dy20             = _mm_sub_ps(iy2,jy0);
 211             dz20             = _mm_sub_ps(iz2,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 217
 218             rinv00           = sse41_invsqrt_f(rsq00);
 219             rinv10           = sse41_invsqrt_f(rsq10);
 220             rinv20           = sse41_invsqrt_f(rsq20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 224                                                               charge+jnrC+0,charge+jnrD+0);
 225             vdwjidx0A        = 2*vdwtype[jnrA+0];
 226             vdwjidx0B        = 2*vdwtype[jnrB+0];
 227             vdwjidx0C        = 2*vdwtype[jnrC+0];
 228             vdwjidx0D        = 2*vdwtype[jnrD+0];
 229
 230             fjx0             = _mm_setzero_ps();
 231             fjy0             = _mm_setzero_ps();
 232             fjz0             = _mm_setzero_ps();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             r00              = _mm_mul_ps(rsq00,rinv00);
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             qq00             = _mm_mul_ps(iq0,jq0);
 242             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 243                                          vdwparam+vdwioffset0+vdwjidx0B,
 244                                          vdwparam+vdwioffset0+vdwjidx0C,
 245                                          vdwparam+vdwioffset0+vdwjidx0D,
 246                                          &c6_00,&c12_00);
 247
 248             /* Calculate table index by multiplying r with table scale and truncate to integer */
 249             rt               = _mm_mul_ps(r00,vftabscale);
 250             vfitab           = _mm_cvttps_epi32(rt);
 251             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 252             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 253
 254             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 255             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 256             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 257             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 258             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 259             _MM_TRANSPOSE4_PS(Y,F,G,H);
 260             Heps             = _mm_mul_ps(vfeps,H);
 261             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 262             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 263             velec            = _mm_mul_ps(qq00,VV);
 264             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 265             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 266
 267             /* CUBIC SPLINE TABLE DISPERSION */
 268             vfitab           = _mm_add_epi32(vfitab,ifour);
 269             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 271             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 272             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 273             _MM_TRANSPOSE4_PS(Y,F,G,H);
 274             Heps             = _mm_mul_ps(vfeps,H);
 275             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 276             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 277             vvdw6            = _mm_mul_ps(c6_00,VV);
 278             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 279             fvdw6            = _mm_mul_ps(c6_00,FF);
 280
 281             /* CUBIC SPLINE TABLE REPULSION */
 282             vfitab           = _mm_add_epi32(vfitab,ifour);
 283             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 284             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 285             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 286             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 287             _MM_TRANSPOSE4_PS(Y,F,G,H);
 288             Heps             = _mm_mul_ps(vfeps,H);
 289             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 290             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 291             vvdw12           = _mm_mul_ps(c12_00,VV);
 292             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 293             fvdw12           = _mm_mul_ps(c12_00,FF);
 294             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 295             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = _mm_add_ps(felec,fvdw);
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm_mul_ps(fscal,dx00);
 305             ty               = _mm_mul_ps(fscal,dy00);
 306             tz               = _mm_mul_ps(fscal,dz00);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm_add_ps(fix0,tx);
 310             fiy0             = _mm_add_ps(fiy0,ty);
 311             fiz0             = _mm_add_ps(fiz0,tz);
 312
 313             fjx0             = _mm_add_ps(fjx0,tx);
 314             fjy0             = _mm_add_ps(fjy0,ty);
 315             fjz0             = _mm_add_ps(fjz0,tz);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             r10              = _mm_mul_ps(rsq10,rinv10);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm_mul_ps(iq1,jq0);
 325
 326             /* Calculate table index by multiplying r with table scale and truncate to integer */
 327             rt               = _mm_mul_ps(r10,vftabscale);
 328             vfitab           = _mm_cvttps_epi32(rt);
 329             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 330             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 331
 332             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 333             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 334             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 335             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 336             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 337             _MM_TRANSPOSE4_PS(Y,F,G,H);
 338             Heps             = _mm_mul_ps(vfeps,H);
 339             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 340             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 341             velec            = _mm_mul_ps(qq10,VV);
 342             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 343             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velecsum         = _mm_add_ps(velecsum,velec);
 347
 348             fscal            = felec;
 349
 350             /* Calculate temporary vectorial force */
 351             tx               = _mm_mul_ps(fscal,dx10);
 352             ty               = _mm_mul_ps(fscal,dy10);
 353             tz               = _mm_mul_ps(fscal,dz10);
 354
 355             /* Update vectorial force */
 356             fix1             = _mm_add_ps(fix1,tx);
 357             fiy1             = _mm_add_ps(fiy1,ty);
 358             fiz1             = _mm_add_ps(fiz1,tz);
 359
 360             fjx0             = _mm_add_ps(fjx0,tx);
 361             fjy0             = _mm_add_ps(fjy0,ty);
 362             fjz0             = _mm_add_ps(fjz0,tz);
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             r20              = _mm_mul_ps(rsq20,rinv20);
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq20             = _mm_mul_ps(iq2,jq0);
 372
 373             /* Calculate table index by multiplying r with table scale and truncate to integer */
 374             rt               = _mm_mul_ps(r20,vftabscale);
 375             vfitab           = _mm_cvttps_epi32(rt);
 376             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 377             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 378
 379             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 380             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 381             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 382             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 383             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 384             _MM_TRANSPOSE4_PS(Y,F,G,H);
 385             Heps             = _mm_mul_ps(vfeps,H);
 386             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 387             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 388             velec            = _mm_mul_ps(qq20,VV);
 389             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 390             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 391
 392             /* Update potential sum for this i atom from the interaction with this j atom. */
 393             velecsum         = _mm_add_ps(velecsum,velec);
 394
 395             fscal            = felec;
 396
 397             /* Calculate temporary vectorial force */
 398             tx               = _mm_mul_ps(fscal,dx20);
 399             ty               = _mm_mul_ps(fscal,dy20);
 400             tz               = _mm_mul_ps(fscal,dz20);
 401
 402             /* Update vectorial force */
 403             fix2             = _mm_add_ps(fix2,tx);
 404             fiy2             = _mm_add_ps(fiy2,ty);
 405             fiz2             = _mm_add_ps(fiz2,tz);
 406
 407             fjx0             = _mm_add_ps(fjx0,tx);
 408             fjy0             = _mm_add_ps(fjy0,ty);
 409             fjz0             = _mm_add_ps(fjz0,tz);
 410
 411             fjptrA             = f+j_coord_offsetA;
 412             fjptrB             = f+j_coord_offsetB;
 413             fjptrC             = f+j_coord_offsetC;
 414             fjptrD             = f+j_coord_offsetD;
 415
 416             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 417
 418             /* Inner loop uses 159 flops */
 419         }
 420
 421         if(jidx<j_index_end)
 422         {
 423
 424             /* Get j neighbor index, and coordinate index */
 425             jnrlistA         = jjnr[jidx];
 426             jnrlistB         = jjnr[jidx+1];
 427             jnrlistC         = jjnr[jidx+2];
 428             jnrlistD         = jjnr[jidx+3];
 429             /* Sign of each element will be negative for non-real atoms.
 430              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 431              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 432              */
 433             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 434             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 435             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 436             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 437             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 438             j_coord_offsetA  = DIM*jnrA;
 439             j_coord_offsetB  = DIM*jnrB;
 440             j_coord_offsetC  = DIM*jnrC;
 441             j_coord_offsetD  = DIM*jnrD;
 442
 443             /* load j atom coordinates */
 444             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 445                                               x+j_coord_offsetC,x+j_coord_offsetD,
 446                                               &jx0,&jy0,&jz0);
 447
 448             /* Calculate displacement vector */
 449             dx00             = _mm_sub_ps(ix0,jx0);
 450             dy00             = _mm_sub_ps(iy0,jy0);
 451             dz00             = _mm_sub_ps(iz0,jz0);
 452             dx10             = _mm_sub_ps(ix1,jx0);
 453             dy10             = _mm_sub_ps(iy1,jy0);
 454             dz10             = _mm_sub_ps(iz1,jz0);
 455             dx20             = _mm_sub_ps(ix2,jx0);
 456             dy20             = _mm_sub_ps(iy2,jy0);
 457             dz20             = _mm_sub_ps(iz2,jz0);
 458
 459             /* Calculate squared distance and things based on it */
 460             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 461             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 462             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 463
 464             rinv00           = sse41_invsqrt_f(rsq00);
 465             rinv10           = sse41_invsqrt_f(rsq10);
 466             rinv20           = sse41_invsqrt_f(rsq20);
 467
 468             /* Load parameters for j particles */
 469             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 470                                                               charge+jnrC+0,charge+jnrD+0);
 471             vdwjidx0A        = 2*vdwtype[jnrA+0];
 472             vdwjidx0B        = 2*vdwtype[jnrB+0];
 473             vdwjidx0C        = 2*vdwtype[jnrC+0];
 474             vdwjidx0D        = 2*vdwtype[jnrD+0];
 475
 476             fjx0             = _mm_setzero_ps();
 477             fjy0             = _mm_setzero_ps();
 478             fjz0             = _mm_setzero_ps();
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r00              = _mm_mul_ps(rsq00,rinv00);
 485             r00              = _mm_andnot_ps(dummy_mask,r00);
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq00             = _mm_mul_ps(iq0,jq0);
 489             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 490                                          vdwparam+vdwioffset0+vdwjidx0B,
 491                                          vdwparam+vdwioffset0+vdwjidx0C,
 492                                          vdwparam+vdwioffset0+vdwjidx0D,
 493                                          &c6_00,&c12_00);
 494
 495             /* Calculate table index by multiplying r with table scale and truncate to integer */
 496             rt               = _mm_mul_ps(r00,vftabscale);
 497             vfitab           = _mm_cvttps_epi32(rt);
 498             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 499             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 500
 501             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 502             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 503             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 504             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 505             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 506             _MM_TRANSPOSE4_PS(Y,F,G,H);
 507             Heps             = _mm_mul_ps(vfeps,H);
 508             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 509             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 510             velec            = _mm_mul_ps(qq00,VV);
 511             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 512             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 513
 514             /* CUBIC SPLINE TABLE DISPERSION */
 515             vfitab           = _mm_add_epi32(vfitab,ifour);
 516             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 517             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 518             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 519             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 520             _MM_TRANSPOSE4_PS(Y,F,G,H);
 521             Heps             = _mm_mul_ps(vfeps,H);
 522             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 523             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 524             vvdw6            = _mm_mul_ps(c6_00,VV);
 525             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 526             fvdw6            = _mm_mul_ps(c6_00,FF);
 527
 528             /* CUBIC SPLINE TABLE REPULSION */
 529             vfitab           = _mm_add_epi32(vfitab,ifour);
 530             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 531             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 532             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 533             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 534             _MM_TRANSPOSE4_PS(Y,F,G,H);
 535             Heps             = _mm_mul_ps(vfeps,H);
 536             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 537             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 538             vvdw12           = _mm_mul_ps(c12_00,VV);
 539             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 540             fvdw12           = _mm_mul_ps(c12_00,FF);
 541             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 542             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 543
 544             /* Update potential sum for this i atom from the interaction with this j atom. */
 545             velec            = _mm_andnot_ps(dummy_mask,velec);
 546             velecsum         = _mm_add_ps(velecsum,velec);
 547             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 548             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 549
 550             fscal            = _mm_add_ps(felec,fvdw);
 551
 552             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm_mul_ps(fscal,dx00);
 556             ty               = _mm_mul_ps(fscal,dy00);
 557             tz               = _mm_mul_ps(fscal,dz00);
 558
 559             /* Update vectorial force */
 560             fix0             = _mm_add_ps(fix0,tx);
 561             fiy0             = _mm_add_ps(fiy0,ty);
 562             fiz0             = _mm_add_ps(fiz0,tz);
 563
 564             fjx0             = _mm_add_ps(fjx0,tx);
 565             fjy0             = _mm_add_ps(fjy0,ty);
 566             fjz0             = _mm_add_ps(fjz0,tz);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             r10              = _mm_mul_ps(rsq10,rinv10);
 573             r10              = _mm_andnot_ps(dummy_mask,r10);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq10             = _mm_mul_ps(iq1,jq0);
 577
 578             /* Calculate table index by multiplying r with table scale and truncate to integer */
 579             rt               = _mm_mul_ps(r10,vftabscale);
 580             vfitab           = _mm_cvttps_epi32(rt);
 581             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 582             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 583
 584             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 585             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 586             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 587             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 588             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 589             _MM_TRANSPOSE4_PS(Y,F,G,H);
 590             Heps             = _mm_mul_ps(vfeps,H);
 591             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 592             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 593             velec            = _mm_mul_ps(qq10,VV);
 594             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 595             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 596
 597             /* Update potential sum for this i atom from the interaction with this j atom. */
 598             velec            = _mm_andnot_ps(dummy_mask,velec);
 599             velecsum         = _mm_add_ps(velecsum,velec);
 600
 601             fscal            = felec;
 602
 603             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 604
 605             /* Calculate temporary vectorial force */
 606             tx               = _mm_mul_ps(fscal,dx10);
 607             ty               = _mm_mul_ps(fscal,dy10);
 608             tz               = _mm_mul_ps(fscal,dz10);
 609
 610             /* Update vectorial force */
 611             fix1             = _mm_add_ps(fix1,tx);
 612             fiy1             = _mm_add_ps(fiy1,ty);
 613             fiz1             = _mm_add_ps(fiz1,tz);
 614
 615             fjx0             = _mm_add_ps(fjx0,tx);
 616             fjy0             = _mm_add_ps(fjy0,ty);
 617             fjz0             = _mm_add_ps(fjz0,tz);
 618
 619             /**************************
 620              * CALCULATE INTERACTIONS *
 621              **************************/
 622
 623             r20              = _mm_mul_ps(rsq20,rinv20);
 624             r20              = _mm_andnot_ps(dummy_mask,r20);
 625
 626             /* Compute parameters for interactions between i and j atoms */
 627             qq20             = _mm_mul_ps(iq2,jq0);
 628
 629             /* Calculate table index by multiplying r with table scale and truncate to integer */
 630             rt               = _mm_mul_ps(r20,vftabscale);
 631             vfitab           = _mm_cvttps_epi32(rt);
 632             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 633             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 634
 635             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 636             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 637             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 638             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 639             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 640             _MM_TRANSPOSE4_PS(Y,F,G,H);
 641             Heps             = _mm_mul_ps(vfeps,H);
 642             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 643             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 644             velec            = _mm_mul_ps(qq20,VV);
 645             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 646             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velec            = _mm_andnot_ps(dummy_mask,velec);
 650             velecsum         = _mm_add_ps(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 655
 656             /* Calculate temporary vectorial force */
 657             tx               = _mm_mul_ps(fscal,dx20);
 658             ty               = _mm_mul_ps(fscal,dy20);
 659             tz               = _mm_mul_ps(fscal,dz20);
 660
 661             /* Update vectorial force */
 662             fix2             = _mm_add_ps(fix2,tx);
 663             fiy2             = _mm_add_ps(fiy2,ty);
 664             fiz2             = _mm_add_ps(fiz2,tz);
 665
 666             fjx0             = _mm_add_ps(fjx0,tx);
 667             fjy0             = _mm_add_ps(fjy0,ty);
 668             fjz0             = _mm_add_ps(fjz0,tz);
 669
 670             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 671             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 672             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 673             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 674
 675             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 676
 677             /* Inner loop uses 162 flops */
 678         }
 679
 680         /* End of innermost loop */
 681
 682         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 683                                               f+i_coord_offset,fshift+i_shift_offset);
 684
 685         ggid                        = gid[iidx];
 686         /* Update potential energies */
 687         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 688         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 689
 690         /* Increment number of inner iterations */
 691         inneriter                  += j_index_end - j_index_start;
 692
 693         /* Outer loop uses 20 flops */
 694     }
 695
 696     /* Increment number of outer iterations */
 697     outeriter        += nri;
 698
 699     /* Update outer/inner flops */
 700
 701     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*162);
 702 }
 703 /*
 704  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_single
 705  * Electrostatics interaction: CubicSplineTable
 706  * VdW interaction:            CubicSplineTable
 707  * Geometry:                   Water3-Particle
 708  * Calculate force/pot:        Force
 709  */
 710 void
 711 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse4_1_single
 712                     (t_nblist                    * gmx_restrict       nlist,
 713                      rvec                        * gmx_restrict          xx,
 714                      rvec                        * gmx_restrict          ff,
 715                      struct t_forcerec           * gmx_restrict          fr,
 716                      t_mdatoms                   * gmx_restrict     mdatoms,
 717                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 718                      t_nrnb                      * gmx_restrict        nrnb)
 719 {
 720     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 721      * just 0 for non-waters.
 722      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 723      * jnr indices corresponding to data put in the four positions in the SIMD register.
 724      */
 725     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 726     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 727     int              jnrA,jnrB,jnrC,jnrD;
 728     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 729     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 730     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 731     real             rcutoff_scalar;
 732     real             *shiftvec,*fshift,*x,*f;
 733     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 734     real             scratch[4*DIM];
 735     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 736     int              vdwioffset0;
 737     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 738     int              vdwioffset1;
 739     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 740     int              vdwioffset2;
 741     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 742     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 743     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 744     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 745     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 746     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 747     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 748     real             *charge;
 749     int              nvdwtype;
 750     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 751     int              *vdwtype;
 752     real             *vdwparam;
 753     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 754     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 755     __m128i          vfitab;
 756     __m128i          ifour       = _mm_set1_epi32(4);
 757     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 758     real             *vftab;
 759     __m128           dummy_mask,cutoff_mask;
 760     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 761     __m128           one     = _mm_set1_ps(1.0);
 762     __m128           two     = _mm_set1_ps(2.0);
 763     x                = xx[0];
 764     f                = ff[0];
 765
 766     nri              = nlist->nri;
 767     iinr             = nlist->iinr;
 768     jindex           = nlist->jindex;
 769     jjnr             = nlist->jjnr;
 770     shiftidx         = nlist->shift;
 771     gid              = nlist->gid;
 772     shiftvec         = fr->shift_vec[0];
 773     fshift           = fr->fshift[0];
 774     facel            = _mm_set1_ps(fr->ic->epsfac);
 775     charge           = mdatoms->chargeA;
 776     nvdwtype         = fr->ntype;
 777     vdwparam         = fr->nbfp;
 778     vdwtype          = mdatoms->typeA;
 779
 780     vftab            = kernel_data->table_elec_vdw->data;
 781     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 782
 783     /* Setup water-specific parameters */
 784     inr              = nlist->iinr[0];
 785     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 786     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 787     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 788     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 789
 790     /* Avoid stupid compiler warnings */
 791     jnrA = jnrB = jnrC = jnrD = 0;
 792     j_coord_offsetA = 0;
 793     j_coord_offsetB = 0;
 794     j_coord_offsetC = 0;
 795     j_coord_offsetD = 0;
 796
 797     outeriter        = 0;
 798     inneriter        = 0;
 799
 800     for(iidx=0;iidx<4*DIM;iidx++)
 801     {
 802         scratch[iidx] = 0.0;
 803     }
 804
 805     /* Start outer loop over neighborlists */
 806     for(iidx=0; iidx<nri; iidx++)
 807     {
 808         /* Load shift vector for this list */
 809         i_shift_offset   = DIM*shiftidx[iidx];
 810
 811         /* Load limits for loop over neighbors */
 812         j_index_start    = jindex[iidx];
 813         j_index_end      = jindex[iidx+1];
 814
 815         /* Get outer coordinate index */
 816         inr              = iinr[iidx];
 817         i_coord_offset   = DIM*inr;
 818
 819         /* Load i particle coords and add shift vector */
 820         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 821                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 822
 823         fix0             = _mm_setzero_ps();
 824         fiy0             = _mm_setzero_ps();
 825         fiz0             = _mm_setzero_ps();
 826         fix1             = _mm_setzero_ps();
 827         fiy1             = _mm_setzero_ps();
 828         fiz1             = _mm_setzero_ps();
 829         fix2             = _mm_setzero_ps();
 830         fiy2             = _mm_setzero_ps();
 831         fiz2             = _mm_setzero_ps();
 832
 833         /* Start inner kernel loop */
 834         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 835         {
 836
 837             /* Get j neighbor index, and coordinate index */
 838             jnrA             = jjnr[jidx];
 839             jnrB             = jjnr[jidx+1];
 840             jnrC             = jjnr[jidx+2];
 841             jnrD             = jjnr[jidx+3];
 842             j_coord_offsetA  = DIM*jnrA;
 843             j_coord_offsetB  = DIM*jnrB;
 844             j_coord_offsetC  = DIM*jnrC;
 845             j_coord_offsetD  = DIM*jnrD;
 846
 847             /* load j atom coordinates */
 848             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 849                                               x+j_coord_offsetC,x+j_coord_offsetD,
 850                                               &jx0,&jy0,&jz0);
 851
 852             /* Calculate displacement vector */
 853             dx00             = _mm_sub_ps(ix0,jx0);
 854             dy00             = _mm_sub_ps(iy0,jy0);
 855             dz00             = _mm_sub_ps(iz0,jz0);
 856             dx10             = _mm_sub_ps(ix1,jx0);
 857             dy10             = _mm_sub_ps(iy1,jy0);
 858             dz10             = _mm_sub_ps(iz1,jz0);
 859             dx20             = _mm_sub_ps(ix2,jx0);
 860             dy20             = _mm_sub_ps(iy2,jy0);
 861             dz20             = _mm_sub_ps(iz2,jz0);
 862
 863             /* Calculate squared distance and things based on it */
 864             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 865             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 866             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 867
 868             rinv00           = sse41_invsqrt_f(rsq00);
 869             rinv10           = sse41_invsqrt_f(rsq10);
 870             rinv20           = sse41_invsqrt_f(rsq20);
 871
 872             /* Load parameters for j particles */
 873             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 874                                                               charge+jnrC+0,charge+jnrD+0);
 875             vdwjidx0A        = 2*vdwtype[jnrA+0];
 876             vdwjidx0B        = 2*vdwtype[jnrB+0];
 877             vdwjidx0C        = 2*vdwtype[jnrC+0];
 878             vdwjidx0D        = 2*vdwtype[jnrD+0];
 879
 880             fjx0             = _mm_setzero_ps();
 881             fjy0             = _mm_setzero_ps();
 882             fjz0             = _mm_setzero_ps();
 883
 884             /**************************
 885              * CALCULATE INTERACTIONS *
 886              **************************/
 887
 888             r00              = _mm_mul_ps(rsq00,rinv00);
 889
 890             /* Compute parameters for interactions between i and j atoms */
 891             qq00             = _mm_mul_ps(iq0,jq0);
 892             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 893                                          vdwparam+vdwioffset0+vdwjidx0B,
 894                                          vdwparam+vdwioffset0+vdwjidx0C,
 895                                          vdwparam+vdwioffset0+vdwjidx0D,
 896                                          &c6_00,&c12_00);
 897
 898             /* Calculate table index by multiplying r with table scale and truncate to integer */
 899             rt               = _mm_mul_ps(r00,vftabscale);
 900             vfitab           = _mm_cvttps_epi32(rt);
 901             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 902             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 903
 904             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 905             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 906             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 907             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 908             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 909             _MM_TRANSPOSE4_PS(Y,F,G,H);
 910             Heps             = _mm_mul_ps(vfeps,H);
 911             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 912             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 913             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 914
 915             /* CUBIC SPLINE TABLE DISPERSION */
 916             vfitab           = _mm_add_epi32(vfitab,ifour);
 917             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 918             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 919             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 920             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 921             _MM_TRANSPOSE4_PS(Y,F,G,H);
 922             Heps             = _mm_mul_ps(vfeps,H);
 923             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 924             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 925             fvdw6            = _mm_mul_ps(c6_00,FF);
 926
 927             /* CUBIC SPLINE TABLE REPULSION */
 928             vfitab           = _mm_add_epi32(vfitab,ifour);
 929             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 930             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 931             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 932             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 933             _MM_TRANSPOSE4_PS(Y,F,G,H);
 934             Heps             = _mm_mul_ps(vfeps,H);
 935             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 936             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 937             fvdw12           = _mm_mul_ps(c12_00,FF);
 938             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 939
 940             fscal            = _mm_add_ps(felec,fvdw);
 941
 942             /* Calculate temporary vectorial force */
 943             tx               = _mm_mul_ps(fscal,dx00);
 944             ty               = _mm_mul_ps(fscal,dy00);
 945             tz               = _mm_mul_ps(fscal,dz00);
 946
 947             /* Update vectorial force */
 948             fix0             = _mm_add_ps(fix0,tx);
 949             fiy0             = _mm_add_ps(fiy0,ty);
 950             fiz0             = _mm_add_ps(fiz0,tz);
 951
 952             fjx0             = _mm_add_ps(fjx0,tx);
 953             fjy0             = _mm_add_ps(fjy0,ty);
 954             fjz0             = _mm_add_ps(fjz0,tz);
 955
 956             /**************************
 957              * CALCULATE INTERACTIONS *
 958              **************************/
 959
 960             r10              = _mm_mul_ps(rsq10,rinv10);
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             qq10             = _mm_mul_ps(iq1,jq0);
 964
 965             /* Calculate table index by multiplying r with table scale and truncate to integer */
 966             rt               = _mm_mul_ps(r10,vftabscale);
 967             vfitab           = _mm_cvttps_epi32(rt);
 968             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 969             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 970
 971             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 972             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 973             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 974             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 975             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 976             _MM_TRANSPOSE4_PS(Y,F,G,H);
 977             Heps             = _mm_mul_ps(vfeps,H);
 978             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 979             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 980             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 981
 982             fscal            = felec;
 983
 984             /* Calculate temporary vectorial force */
 985             tx               = _mm_mul_ps(fscal,dx10);
 986             ty               = _mm_mul_ps(fscal,dy10);
 987             tz               = _mm_mul_ps(fscal,dz10);
 988
 989             /* Update vectorial force */
 990             fix1             = _mm_add_ps(fix1,tx);
 991             fiy1             = _mm_add_ps(fiy1,ty);
 992             fiz1             = _mm_add_ps(fiz1,tz);
 993
 994             fjx0             = _mm_add_ps(fjx0,tx);
 995             fjy0             = _mm_add_ps(fjy0,ty);
 996             fjz0             = _mm_add_ps(fjz0,tz);
 997
 998             /**************************
 999              * CALCULATE INTERACTIONS *
1000              **************************/
1001
1002             r20              = _mm_mul_ps(rsq20,rinv20);
1003
1004             /* Compute parameters for interactions between i and j atoms */
1005             qq20             = _mm_mul_ps(iq2,jq0);
1006
1007             /* Calculate table index by multiplying r with table scale and truncate to integer */
1008             rt               = _mm_mul_ps(r20,vftabscale);
1009             vfitab           = _mm_cvttps_epi32(rt);
1010             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1011             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1012
1013             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1014             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1015             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1016             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1017             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1018             _MM_TRANSPOSE4_PS(Y,F,G,H);
1019             Heps             = _mm_mul_ps(vfeps,H);
1020             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1021             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1022             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1023
1024             fscal            = felec;
1025
1026             /* Calculate temporary vectorial force */
1027             tx               = _mm_mul_ps(fscal,dx20);
1028             ty               = _mm_mul_ps(fscal,dy20);
1029             tz               = _mm_mul_ps(fscal,dz20);
1030
1031             /* Update vectorial force */
1032             fix2             = _mm_add_ps(fix2,tx);
1033             fiy2             = _mm_add_ps(fiy2,ty);
1034             fiz2             = _mm_add_ps(fiz2,tz);
1035
1036             fjx0             = _mm_add_ps(fjx0,tx);
1037             fjy0             = _mm_add_ps(fjy0,ty);
1038             fjz0             = _mm_add_ps(fjz0,tz);
1039
1040             fjptrA             = f+j_coord_offsetA;
1041             fjptrB             = f+j_coord_offsetB;
1042             fjptrC             = f+j_coord_offsetC;
1043             fjptrD             = f+j_coord_offsetD;
1044
1045             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1046
1047             /* Inner loop uses 139 flops */
1048         }
1049
1050         if(jidx<j_index_end)
1051         {
1052
1053             /* Get j neighbor index, and coordinate index */
1054             jnrlistA         = jjnr[jidx];
1055             jnrlistB         = jjnr[jidx+1];
1056             jnrlistC         = jjnr[jidx+2];
1057             jnrlistD         = jjnr[jidx+3];
1058             /* Sign of each element will be negative for non-real atoms.
1059              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1060              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1061              */
1062             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1063             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1064             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1065             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1066             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1067             j_coord_offsetA  = DIM*jnrA;
1068             j_coord_offsetB  = DIM*jnrB;
1069             j_coord_offsetC  = DIM*jnrC;
1070             j_coord_offsetD  = DIM*jnrD;
1071
1072             /* load j atom coordinates */
1073             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1074                                               x+j_coord_offsetC,x+j_coord_offsetD,
1075                                               &jx0,&jy0,&jz0);
1076
1077             /* Calculate displacement vector */
1078             dx00             = _mm_sub_ps(ix0,jx0);
1079             dy00             = _mm_sub_ps(iy0,jy0);
1080             dz00             = _mm_sub_ps(iz0,jz0);
1081             dx10             = _mm_sub_ps(ix1,jx0);
1082             dy10             = _mm_sub_ps(iy1,jy0);
1083             dz10             = _mm_sub_ps(iz1,jz0);
1084             dx20             = _mm_sub_ps(ix2,jx0);
1085             dy20             = _mm_sub_ps(iy2,jy0);
1086             dz20             = _mm_sub_ps(iz2,jz0);
1087
1088             /* Calculate squared distance and things based on it */
1089             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1090             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1091             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1092
1093             rinv00           = sse41_invsqrt_f(rsq00);
1094             rinv10           = sse41_invsqrt_f(rsq10);
1095             rinv20           = sse41_invsqrt_f(rsq20);
1096
1097             /* Load parameters for j particles */
1098             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1099                                                               charge+jnrC+0,charge+jnrD+0);
1100             vdwjidx0A        = 2*vdwtype[jnrA+0];
1101             vdwjidx0B        = 2*vdwtype[jnrB+0];
1102             vdwjidx0C        = 2*vdwtype[jnrC+0];
1103             vdwjidx0D        = 2*vdwtype[jnrD+0];
1104
1105             fjx0             = _mm_setzero_ps();
1106             fjy0             = _mm_setzero_ps();
1107             fjz0             = _mm_setzero_ps();
1108
1109             /**************************
1110              * CALCULATE INTERACTIONS *
1111              **************************/
1112
1113             r00              = _mm_mul_ps(rsq00,rinv00);
1114             r00              = _mm_andnot_ps(dummy_mask,r00);
1115
1116             /* Compute parameters for interactions between i and j atoms */
1117             qq00             = _mm_mul_ps(iq0,jq0);
1118             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1119                                          vdwparam+vdwioffset0+vdwjidx0B,
1120                                          vdwparam+vdwioffset0+vdwjidx0C,
1121                                          vdwparam+vdwioffset0+vdwjidx0D,
1122                                          &c6_00,&c12_00);
1123
1124             /* Calculate table index by multiplying r with table scale and truncate to integer */
1125             rt               = _mm_mul_ps(r00,vftabscale);
1126             vfitab           = _mm_cvttps_epi32(rt);
1127             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1128             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1129
1130             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1131             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1132             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1133             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1134             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1135             _MM_TRANSPOSE4_PS(Y,F,G,H);
1136             Heps             = _mm_mul_ps(vfeps,H);
1137             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1138             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1139             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1140
1141             /* CUBIC SPLINE TABLE DISPERSION */
1142             vfitab           = _mm_add_epi32(vfitab,ifour);
1143             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1144             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1145             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1146             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1147             _MM_TRANSPOSE4_PS(Y,F,G,H);
1148             Heps             = _mm_mul_ps(vfeps,H);
1149             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1150             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1151             fvdw6            = _mm_mul_ps(c6_00,FF);
1152
1153             /* CUBIC SPLINE TABLE REPULSION */
1154             vfitab           = _mm_add_epi32(vfitab,ifour);
1155             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1156             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1157             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1158             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1159             _MM_TRANSPOSE4_PS(Y,F,G,H);
1160             Heps             = _mm_mul_ps(vfeps,H);
1161             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1162             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1163             fvdw12           = _mm_mul_ps(c12_00,FF);
1164             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1165
1166             fscal            = _mm_add_ps(felec,fvdw);
1167
1168             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1169
1170             /* Calculate temporary vectorial force */
1171             tx               = _mm_mul_ps(fscal,dx00);
1172             ty               = _mm_mul_ps(fscal,dy00);
1173             tz               = _mm_mul_ps(fscal,dz00);
1174
1175             /* Update vectorial force */
1176             fix0             = _mm_add_ps(fix0,tx);
1177             fiy0             = _mm_add_ps(fiy0,ty);
1178             fiz0             = _mm_add_ps(fiz0,tz);
1179
1180             fjx0             = _mm_add_ps(fjx0,tx);
1181             fjy0             = _mm_add_ps(fjy0,ty);
1182             fjz0             = _mm_add_ps(fjz0,tz);
1183
1184             /**************************
1185              * CALCULATE INTERACTIONS *
1186              **************************/
1187
1188             r10              = _mm_mul_ps(rsq10,rinv10);
1189             r10              = _mm_andnot_ps(dummy_mask,r10);
1190
1191             /* Compute parameters for interactions between i and j atoms */
1192             qq10             = _mm_mul_ps(iq1,jq0);
1193
1194             /* Calculate table index by multiplying r with table scale and truncate to integer */
1195             rt               = _mm_mul_ps(r10,vftabscale);
1196             vfitab           = _mm_cvttps_epi32(rt);
1197             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1198             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1199
1200             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1201             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1202             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1203             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1204             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1205             _MM_TRANSPOSE4_PS(Y,F,G,H);
1206             Heps             = _mm_mul_ps(vfeps,H);
1207             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1208             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1209             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1210
1211             fscal            = felec;
1212
1213             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1214
1215             /* Calculate temporary vectorial force */
1216             tx               = _mm_mul_ps(fscal,dx10);
1217             ty               = _mm_mul_ps(fscal,dy10);
1218             tz               = _mm_mul_ps(fscal,dz10);
1219
1220             /* Update vectorial force */
1221             fix1             = _mm_add_ps(fix1,tx);
1222             fiy1             = _mm_add_ps(fiy1,ty);
1223             fiz1             = _mm_add_ps(fiz1,tz);
1224
1225             fjx0             = _mm_add_ps(fjx0,tx);
1226             fjy0             = _mm_add_ps(fjy0,ty);
1227             fjz0             = _mm_add_ps(fjz0,tz);
1228
1229             /**************************
1230              * CALCULATE INTERACTIONS *
1231              **************************/
1232
1233             r20              = _mm_mul_ps(rsq20,rinv20);
1234             r20              = _mm_andnot_ps(dummy_mask,r20);
1235
1236             /* Compute parameters for interactions between i and j atoms */
1237             qq20             = _mm_mul_ps(iq2,jq0);
1238
1239             /* Calculate table index by multiplying r with table scale and truncate to integer */
1240             rt               = _mm_mul_ps(r20,vftabscale);
1241             vfitab           = _mm_cvttps_epi32(rt);
1242             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1243             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1244
1245             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1246             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1247             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1248             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1249             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1250             _MM_TRANSPOSE4_PS(Y,F,G,H);
1251             Heps             = _mm_mul_ps(vfeps,H);
1252             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1253             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1254             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1255
1256             fscal            = felec;
1257
1258             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1259
1260             /* Calculate temporary vectorial force */
1261             tx               = _mm_mul_ps(fscal,dx20);
1262             ty               = _mm_mul_ps(fscal,dy20);
1263             tz               = _mm_mul_ps(fscal,dz20);
1264
1265             /* Update vectorial force */
1266             fix2             = _mm_add_ps(fix2,tx);
1267             fiy2             = _mm_add_ps(fiy2,ty);
1268             fiz2             = _mm_add_ps(fiz2,tz);
1269
1270             fjx0             = _mm_add_ps(fjx0,tx);
1271             fjy0             = _mm_add_ps(fjy0,ty);
1272             fjz0             = _mm_add_ps(fjz0,tz);
1273
1274             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1275             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1276             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1277             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1278
1279             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1280
1281             /* Inner loop uses 142 flops */
1282         }
1283
1284         /* End of innermost loop */
1285
1286         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1287                                               f+i_coord_offset,fshift+i_shift_offset);
1288
1289         /* Increment number of inner iterations */
1290         inneriter                  += j_index_end - j_index_start;
1291
1292         /* Outer loop uses 18 flops */
1293     }
1294
1295     /* Increment number of outer iterations */
1296     outeriter        += nri;
1297
1298     /* Update outer/inner flops */
1299
1300     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1301 }