src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse2_single
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          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->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec_vdw->data;
 130     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 135     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 136     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 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_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 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
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_ps();
 184         vvdwsum          = _mm_setzero_ps();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             jnrC             = jjnr[jidx+2];
 194             jnrD             = jjnr[jidx+3];
 195             j_coord_offsetA  = DIM*jnrA;
 196             j_coord_offsetB  = DIM*jnrB;
 197             j_coord_offsetC  = DIM*jnrC;
 198             j_coord_offsetD  = DIM*jnrD;
 199
 200             /* load j atom coordinates */
 201             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 202                                               x+j_coord_offsetC,x+j_coord_offsetD,
 203                                               &jx0,&jy0,&jz0);
 204
 205             /* Calculate displacement vector */
 206             dx00             = _mm_sub_ps(ix0,jx0);
 207             dy00             = _mm_sub_ps(iy0,jy0);
 208             dz00             = _mm_sub_ps(iz0,jz0);
 209             dx10             = _mm_sub_ps(ix1,jx0);
 210             dy10             = _mm_sub_ps(iy1,jy0);
 211             dz10             = _mm_sub_ps(iz1,jz0);
 212             dx20             = _mm_sub_ps(ix2,jx0);
 213             dy20             = _mm_sub_ps(iy2,jy0);
 214             dz20             = _mm_sub_ps(iz2,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 220
 221             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 222             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 223             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                               charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             fjx0             = _mm_setzero_ps();
 234             fjy0             = _mm_setzero_ps();
 235             fjz0             = _mm_setzero_ps();
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             r00              = _mm_mul_ps(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm_mul_ps(iq0,jq0);
 245             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 246                                          vdwparam+vdwioffset0+vdwjidx0B,
 247                                          vdwparam+vdwioffset0+vdwjidx0C,
 248                                          vdwparam+vdwioffset0+vdwjidx0D,
 249                                          &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm_mul_ps(r00,vftabscale);
 253             vfitab           = _mm_cvttps_epi32(rt);
 254             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 255             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 256
 257             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 258             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 259             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 260             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 261             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 262             _MM_TRANSPOSE4_PS(Y,F,G,H);
 263             Heps             = _mm_mul_ps(vfeps,H);
 264             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 265             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 266             velec            = _mm_mul_ps(qq00,VV);
 267             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 268             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 269
 270             /* CUBIC SPLINE TABLE DISPERSION */
 271             vfitab           = _mm_add_epi32(vfitab,ifour);
 272             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 273             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 274             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 275             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 276             _MM_TRANSPOSE4_PS(Y,F,G,H);
 277             Heps             = _mm_mul_ps(vfeps,H);
 278             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 279             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 280             vvdw6            = _mm_mul_ps(c6_00,VV);
 281             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 282             fvdw6            = _mm_mul_ps(c6_00,FF);
 283
 284             /* CUBIC SPLINE TABLE REPULSION */
 285             vfitab           = _mm_add_epi32(vfitab,ifour);
 286             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 287             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 288             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 289             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 290             _MM_TRANSPOSE4_PS(Y,F,G,H);
 291             Heps             = _mm_mul_ps(vfeps,H);
 292             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 293             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 294             vvdw12           = _mm_mul_ps(c12_00,VV);
 295             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 296             fvdw12           = _mm_mul_ps(c12_00,FF);
 297             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 298             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm_add_ps(velecsum,velec);
 302             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 303
 304             fscal            = _mm_add_ps(felec,fvdw);
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm_mul_ps(fscal,dx00);
 308             ty               = _mm_mul_ps(fscal,dy00);
 309             tz               = _mm_mul_ps(fscal,dz00);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm_add_ps(fix0,tx);
 313             fiy0             = _mm_add_ps(fiy0,ty);
 314             fiz0             = _mm_add_ps(fiz0,tz);
 315
 316             fjx0             = _mm_add_ps(fjx0,tx);
 317             fjy0             = _mm_add_ps(fjy0,ty);
 318             fjz0             = _mm_add_ps(fjz0,tz);
 319
 320             /**************************
 321              * CALCULATE INTERACTIONS *
 322              **************************/
 323
 324             r10              = _mm_mul_ps(rsq10,rinv10);
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq10             = _mm_mul_ps(iq1,jq0);
 328
 329             /* Calculate table index by multiplying r with table scale and truncate to integer */
 330             rt               = _mm_mul_ps(r10,vftabscale);
 331             vfitab           = _mm_cvttps_epi32(rt);
 332             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 333             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 334
 335             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 336             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 337             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 338             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 339             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 340             _MM_TRANSPOSE4_PS(Y,F,G,H);
 341             Heps             = _mm_mul_ps(vfeps,H);
 342             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 343             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 344             velec            = _mm_mul_ps(qq10,VV);
 345             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 346             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm_mul_ps(fscal,dx10);
 355             ty               = _mm_mul_ps(fscal,dy10);
 356             tz               = _mm_mul_ps(fscal,dz10);
 357
 358             /* Update vectorial force */
 359             fix1             = _mm_add_ps(fix1,tx);
 360             fiy1             = _mm_add_ps(fiy1,ty);
 361             fiz1             = _mm_add_ps(fiz1,tz);
 362
 363             fjx0             = _mm_add_ps(fjx0,tx);
 364             fjy0             = _mm_add_ps(fjy0,ty);
 365             fjz0             = _mm_add_ps(fjz0,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r20              = _mm_mul_ps(rsq20,rinv20);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq20             = _mm_mul_ps(iq2,jq0);
 375
 376             /* Calculate table index by multiplying r with table scale and truncate to integer */
 377             rt               = _mm_mul_ps(r20,vftabscale);
 378             vfitab           = _mm_cvttps_epi32(rt);
 379             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 380             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 381
 382             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 383             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 384             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 385             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 386             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 387             _MM_TRANSPOSE4_PS(Y,F,G,H);
 388             Heps             = _mm_mul_ps(vfeps,H);
 389             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 390             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 391             velec            = _mm_mul_ps(qq20,VV);
 392             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 393             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 394
 395             /* Update potential sum for this i atom from the interaction with this j atom. */
 396             velecsum         = _mm_add_ps(velecsum,velec);
 397
 398             fscal            = felec;
 399
 400             /* Calculate temporary vectorial force */
 401             tx               = _mm_mul_ps(fscal,dx20);
 402             ty               = _mm_mul_ps(fscal,dy20);
 403             tz               = _mm_mul_ps(fscal,dz20);
 404
 405             /* Update vectorial force */
 406             fix2             = _mm_add_ps(fix2,tx);
 407             fiy2             = _mm_add_ps(fiy2,ty);
 408             fiz2             = _mm_add_ps(fiz2,tz);
 409
 410             fjx0             = _mm_add_ps(fjx0,tx);
 411             fjy0             = _mm_add_ps(fjy0,ty);
 412             fjz0             = _mm_add_ps(fjz0,tz);
 413
 414             fjptrA             = f+j_coord_offsetA;
 415             fjptrB             = f+j_coord_offsetB;
 416             fjptrC             = f+j_coord_offsetC;
 417             fjptrD             = f+j_coord_offsetD;
 418
 419             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 420
 421             /* Inner loop uses 159 flops */
 422         }
 423
 424         if(jidx<j_index_end)
 425         {
 426
 427             /* Get j neighbor index, and coordinate index */
 428             jnrlistA         = jjnr[jidx];
 429             jnrlistB         = jjnr[jidx+1];
 430             jnrlistC         = jjnr[jidx+2];
 431             jnrlistD         = jjnr[jidx+3];
 432             /* Sign of each element will be negative for non-real atoms.
 433              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 434              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 435              */
 436             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 437             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 438             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 439             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 440             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 441             j_coord_offsetA  = DIM*jnrA;
 442             j_coord_offsetB  = DIM*jnrB;
 443             j_coord_offsetC  = DIM*jnrC;
 444             j_coord_offsetD  = DIM*jnrD;
 445
 446             /* load j atom coordinates */
 447             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 448                                               x+j_coord_offsetC,x+j_coord_offsetD,
 449                                               &jx0,&jy0,&jz0);
 450
 451             /* Calculate displacement vector */
 452             dx00             = _mm_sub_ps(ix0,jx0);
 453             dy00             = _mm_sub_ps(iy0,jy0);
 454             dz00             = _mm_sub_ps(iz0,jz0);
 455             dx10             = _mm_sub_ps(ix1,jx0);
 456             dy10             = _mm_sub_ps(iy1,jy0);
 457             dz10             = _mm_sub_ps(iz1,jz0);
 458             dx20             = _mm_sub_ps(ix2,jx0);
 459             dy20             = _mm_sub_ps(iy2,jy0);
 460             dz20             = _mm_sub_ps(iz2,jz0);
 461
 462             /* Calculate squared distance and things based on it */
 463             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 464             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 465             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 466
 467             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 468             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 469             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 470
 471             /* Load parameters for j particles */
 472             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 473                                                               charge+jnrC+0,charge+jnrD+0);
 474             vdwjidx0A        = 2*vdwtype[jnrA+0];
 475             vdwjidx0B        = 2*vdwtype[jnrB+0];
 476             vdwjidx0C        = 2*vdwtype[jnrC+0];
 477             vdwjidx0D        = 2*vdwtype[jnrD+0];
 478
 479             fjx0             = _mm_setzero_ps();
 480             fjy0             = _mm_setzero_ps();
 481             fjz0             = _mm_setzero_ps();
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             r00              = _mm_mul_ps(rsq00,rinv00);
 488             r00              = _mm_andnot_ps(dummy_mask,r00);
 489
 490             /* Compute parameters for interactions between i and j atoms */
 491             qq00             = _mm_mul_ps(iq0,jq0);
 492             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 493                                          vdwparam+vdwioffset0+vdwjidx0B,
 494                                          vdwparam+vdwioffset0+vdwjidx0C,
 495                                          vdwparam+vdwioffset0+vdwjidx0D,
 496                                          &c6_00,&c12_00);
 497
 498             /* Calculate table index by multiplying r with table scale and truncate to integer */
 499             rt               = _mm_mul_ps(r00,vftabscale);
 500             vfitab           = _mm_cvttps_epi32(rt);
 501             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 502             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 503
 504             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 505             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 506             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 507             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 508             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 509             _MM_TRANSPOSE4_PS(Y,F,G,H);
 510             Heps             = _mm_mul_ps(vfeps,H);
 511             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 512             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 513             velec            = _mm_mul_ps(qq00,VV);
 514             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 515             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 516
 517             /* CUBIC SPLINE TABLE DISPERSION */
 518             vfitab           = _mm_add_epi32(vfitab,ifour);
 519             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 520             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 521             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 522             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 523             _MM_TRANSPOSE4_PS(Y,F,G,H);
 524             Heps             = _mm_mul_ps(vfeps,H);
 525             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 526             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 527             vvdw6            = _mm_mul_ps(c6_00,VV);
 528             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 529             fvdw6            = _mm_mul_ps(c6_00,FF);
 530
 531             /* CUBIC SPLINE TABLE REPULSION */
 532             vfitab           = _mm_add_epi32(vfitab,ifour);
 533             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 534             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 535             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 536             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 537             _MM_TRANSPOSE4_PS(Y,F,G,H);
 538             Heps             = _mm_mul_ps(vfeps,H);
 539             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 540             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 541             vvdw12           = _mm_mul_ps(c12_00,VV);
 542             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 543             fvdw12           = _mm_mul_ps(c12_00,FF);
 544             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 545             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 546
 547             /* Update potential sum for this i atom from the interaction with this j atom. */
 548             velec            = _mm_andnot_ps(dummy_mask,velec);
 549             velecsum         = _mm_add_ps(velecsum,velec);
 550             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 551             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 552
 553             fscal            = _mm_add_ps(felec,fvdw);
 554
 555             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm_mul_ps(fscal,dx00);
 559             ty               = _mm_mul_ps(fscal,dy00);
 560             tz               = _mm_mul_ps(fscal,dz00);
 561
 562             /* Update vectorial force */
 563             fix0             = _mm_add_ps(fix0,tx);
 564             fiy0             = _mm_add_ps(fiy0,ty);
 565             fiz0             = _mm_add_ps(fiz0,tz);
 566
 567             fjx0             = _mm_add_ps(fjx0,tx);
 568             fjy0             = _mm_add_ps(fjy0,ty);
 569             fjz0             = _mm_add_ps(fjz0,tz);
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             r10              = _mm_mul_ps(rsq10,rinv10);
 576             r10              = _mm_andnot_ps(dummy_mask,r10);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             qq10             = _mm_mul_ps(iq1,jq0);
 580
 581             /* Calculate table index by multiplying r with table scale and truncate to integer */
 582             rt               = _mm_mul_ps(r10,vftabscale);
 583             vfitab           = _mm_cvttps_epi32(rt);
 584             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 585             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 586
 587             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 588             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 589             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 590             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 591             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 592             _MM_TRANSPOSE4_PS(Y,F,G,H);
 593             Heps             = _mm_mul_ps(vfeps,H);
 594             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 595             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 596             velec            = _mm_mul_ps(qq10,VV);
 597             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 598             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             velec            = _mm_andnot_ps(dummy_mask,velec);
 602             velecsum         = _mm_add_ps(velecsum,velec);
 603
 604             fscal            = felec;
 605
 606             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 607
 608             /* Calculate temporary vectorial force */
 609             tx               = _mm_mul_ps(fscal,dx10);
 610             ty               = _mm_mul_ps(fscal,dy10);
 611             tz               = _mm_mul_ps(fscal,dz10);
 612
 613             /* Update vectorial force */
 614             fix1             = _mm_add_ps(fix1,tx);
 615             fiy1             = _mm_add_ps(fiy1,ty);
 616             fiz1             = _mm_add_ps(fiz1,tz);
 617
 618             fjx0             = _mm_add_ps(fjx0,tx);
 619             fjy0             = _mm_add_ps(fjy0,ty);
 620             fjz0             = _mm_add_ps(fjz0,tz);
 621
 622             /**************************
 623              * CALCULATE INTERACTIONS *
 624              **************************/
 625
 626             r20              = _mm_mul_ps(rsq20,rinv20);
 627             r20              = _mm_andnot_ps(dummy_mask,r20);
 628
 629             /* Compute parameters for interactions between i and j atoms */
 630             qq20             = _mm_mul_ps(iq2,jq0);
 631
 632             /* Calculate table index by multiplying r with table scale and truncate to integer */
 633             rt               = _mm_mul_ps(r20,vftabscale);
 634             vfitab           = _mm_cvttps_epi32(rt);
 635             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 636             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 637
 638             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 639             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 640             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 641             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 642             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 643             _MM_TRANSPOSE4_PS(Y,F,G,H);
 644             Heps             = _mm_mul_ps(vfeps,H);
 645             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 646             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 647             velec            = _mm_mul_ps(qq20,VV);
 648             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 649             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 650
 651             /* Update potential sum for this i atom from the interaction with this j atom. */
 652             velec            = _mm_andnot_ps(dummy_mask,velec);
 653             velecsum         = _mm_add_ps(velecsum,velec);
 654
 655             fscal            = felec;
 656
 657             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 658
 659             /* Calculate temporary vectorial force */
 660             tx               = _mm_mul_ps(fscal,dx20);
 661             ty               = _mm_mul_ps(fscal,dy20);
 662             tz               = _mm_mul_ps(fscal,dz20);
 663
 664             /* Update vectorial force */
 665             fix2             = _mm_add_ps(fix2,tx);
 666             fiy2             = _mm_add_ps(fiy2,ty);
 667             fiz2             = _mm_add_ps(fiz2,tz);
 668
 669             fjx0             = _mm_add_ps(fjx0,tx);
 670             fjy0             = _mm_add_ps(fjy0,ty);
 671             fjz0             = _mm_add_ps(fjz0,tz);
 672
 673             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 674             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 675             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 676             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 677
 678             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 679
 680             /* Inner loop uses 162 flops */
 681         }
 682
 683         /* End of innermost loop */
 684
 685         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 686                                               f+i_coord_offset,fshift+i_shift_offset);
 687
 688         ggid                        = gid[iidx];
 689         /* Update potential energies */
 690         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 691         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 692
 693         /* Increment number of inner iterations */
 694         inneriter                  += j_index_end - j_index_start;
 695
 696         /* Outer loop uses 20 flops */
 697     }
 698
 699     /* Increment number of outer iterations */
 700     outeriter        += nri;
 701
 702     /* Update outer/inner flops */
 703
 704     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*162);
 705 }
 706 /*
 707  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse2_single
 708  * Electrostatics interaction: CubicSplineTable
 709  * VdW interaction:            CubicSplineTable
 710  * Geometry:                   Water3-Particle
 711  * Calculate force/pot:        Force
 712  */
 713 void
 714 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse2_single
 715                     (t_nblist                    * gmx_restrict       nlist,
 716                      rvec                        * gmx_restrict          xx,
 717                      rvec                        * gmx_restrict          ff,
 718                      t_forcerec                  * gmx_restrict          fr,
 719                      t_mdatoms                   * gmx_restrict     mdatoms,
 720                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 721                      t_nrnb                      * gmx_restrict        nrnb)
 722 {
 723     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 724      * just 0 for non-waters.
 725      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 726      * jnr indices corresponding to data put in the four positions in the SIMD register.
 727      */
 728     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 729     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 730     int              jnrA,jnrB,jnrC,jnrD;
 731     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 732     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 733     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 734     real             rcutoff_scalar;
 735     real             *shiftvec,*fshift,*x,*f;
 736     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 737     real             scratch[4*DIM];
 738     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 739     int              vdwioffset0;
 740     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 741     int              vdwioffset1;
 742     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 743     int              vdwioffset2;
 744     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 745     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 746     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 747     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 748     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 749     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 750     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 751     real             *charge;
 752     int              nvdwtype;
 753     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 754     int              *vdwtype;
 755     real             *vdwparam;
 756     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 757     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 758     __m128i          vfitab;
 759     __m128i          ifour       = _mm_set1_epi32(4);
 760     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 761     real             *vftab;
 762     __m128           dummy_mask,cutoff_mask;
 763     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 764     __m128           one     = _mm_set1_ps(1.0);
 765     __m128           two     = _mm_set1_ps(2.0);
 766     x                = xx[0];
 767     f                = ff[0];
 768
 769     nri              = nlist->nri;
 770     iinr             = nlist->iinr;
 771     jindex           = nlist->jindex;
 772     jjnr             = nlist->jjnr;
 773     shiftidx         = nlist->shift;
 774     gid              = nlist->gid;
 775     shiftvec         = fr->shift_vec[0];
 776     fshift           = fr->fshift[0];
 777     facel            = _mm_set1_ps(fr->epsfac);
 778     charge           = mdatoms->chargeA;
 779     nvdwtype         = fr->ntype;
 780     vdwparam         = fr->nbfp;
 781     vdwtype          = mdatoms->typeA;
 782
 783     vftab            = kernel_data->table_elec_vdw->data;
 784     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 785
 786     /* Setup water-specific parameters */
 787     inr              = nlist->iinr[0];
 788     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 789     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 790     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 791     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 792
 793     /* Avoid stupid compiler warnings */
 794     jnrA = jnrB = jnrC = jnrD = 0;
 795     j_coord_offsetA = 0;
 796     j_coord_offsetB = 0;
 797     j_coord_offsetC = 0;
 798     j_coord_offsetD = 0;
 799
 800     outeriter        = 0;
 801     inneriter        = 0;
 802
 803     for(iidx=0;iidx<4*DIM;iidx++)
 804     {
 805         scratch[iidx] = 0.0;
 806     }
 807
 808     /* Start outer loop over neighborlists */
 809     for(iidx=0; iidx<nri; iidx++)
 810     {
 811         /* Load shift vector for this list */
 812         i_shift_offset   = DIM*shiftidx[iidx];
 813
 814         /* Load limits for loop over neighbors */
 815         j_index_start    = jindex[iidx];
 816         j_index_end      = jindex[iidx+1];
 817
 818         /* Get outer coordinate index */
 819         inr              = iinr[iidx];
 820         i_coord_offset   = DIM*inr;
 821
 822         /* Load i particle coords and add shift vector */
 823         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 824                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 825
 826         fix0             = _mm_setzero_ps();
 827         fiy0             = _mm_setzero_ps();
 828         fiz0             = _mm_setzero_ps();
 829         fix1             = _mm_setzero_ps();
 830         fiy1             = _mm_setzero_ps();
 831         fiz1             = _mm_setzero_ps();
 832         fix2             = _mm_setzero_ps();
 833         fiy2             = _mm_setzero_ps();
 834         fiz2             = _mm_setzero_ps();
 835
 836         /* Start inner kernel loop */
 837         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 838         {
 839
 840             /* Get j neighbor index, and coordinate index */
 841             jnrA             = jjnr[jidx];
 842             jnrB             = jjnr[jidx+1];
 843             jnrC             = jjnr[jidx+2];
 844             jnrD             = jjnr[jidx+3];
 845             j_coord_offsetA  = DIM*jnrA;
 846             j_coord_offsetB  = DIM*jnrB;
 847             j_coord_offsetC  = DIM*jnrC;
 848             j_coord_offsetD  = DIM*jnrD;
 849
 850             /* load j atom coordinates */
 851             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 852                                               x+j_coord_offsetC,x+j_coord_offsetD,
 853                                               &jx0,&jy0,&jz0);
 854
 855             /* Calculate displacement vector */
 856             dx00             = _mm_sub_ps(ix0,jx0);
 857             dy00             = _mm_sub_ps(iy0,jy0);
 858             dz00             = _mm_sub_ps(iz0,jz0);
 859             dx10             = _mm_sub_ps(ix1,jx0);
 860             dy10             = _mm_sub_ps(iy1,jy0);
 861             dz10             = _mm_sub_ps(iz1,jz0);
 862             dx20             = _mm_sub_ps(ix2,jx0);
 863             dy20             = _mm_sub_ps(iy2,jy0);
 864             dz20             = _mm_sub_ps(iz2,jz0);
 865
 866             /* Calculate squared distance and things based on it */
 867             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 868             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 869             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 870
 871             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 872             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 873             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 874
 875             /* Load parameters for j particles */
 876             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 877                                                               charge+jnrC+0,charge+jnrD+0);
 878             vdwjidx0A        = 2*vdwtype[jnrA+0];
 879             vdwjidx0B        = 2*vdwtype[jnrB+0];
 880             vdwjidx0C        = 2*vdwtype[jnrC+0];
 881             vdwjidx0D        = 2*vdwtype[jnrD+0];
 882
 883             fjx0             = _mm_setzero_ps();
 884             fjy0             = _mm_setzero_ps();
 885             fjz0             = _mm_setzero_ps();
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             r00              = _mm_mul_ps(rsq00,rinv00);
 892
 893             /* Compute parameters for interactions between i and j atoms */
 894             qq00             = _mm_mul_ps(iq0,jq0);
 895             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 896                                          vdwparam+vdwioffset0+vdwjidx0B,
 897                                          vdwparam+vdwioffset0+vdwjidx0C,
 898                                          vdwparam+vdwioffset0+vdwjidx0D,
 899                                          &c6_00,&c12_00);
 900
 901             /* Calculate table index by multiplying r with table scale and truncate to integer */
 902             rt               = _mm_mul_ps(r00,vftabscale);
 903             vfitab           = _mm_cvttps_epi32(rt);
 904             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 905             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 906
 907             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 908             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 909             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 910             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 911             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 912             _MM_TRANSPOSE4_PS(Y,F,G,H);
 913             Heps             = _mm_mul_ps(vfeps,H);
 914             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 915             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 916             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 917
 918             /* CUBIC SPLINE TABLE DISPERSION */
 919             vfitab           = _mm_add_epi32(vfitab,ifour);
 920             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 921             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 922             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 923             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 924             _MM_TRANSPOSE4_PS(Y,F,G,H);
 925             Heps             = _mm_mul_ps(vfeps,H);
 926             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 927             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 928             fvdw6            = _mm_mul_ps(c6_00,FF);
 929
 930             /* CUBIC SPLINE TABLE REPULSION */
 931             vfitab           = _mm_add_epi32(vfitab,ifour);
 932             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 933             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 934             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 935             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 936             _MM_TRANSPOSE4_PS(Y,F,G,H);
 937             Heps             = _mm_mul_ps(vfeps,H);
 938             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 939             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 940             fvdw12           = _mm_mul_ps(c12_00,FF);
 941             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 942
 943             fscal            = _mm_add_ps(felec,fvdw);
 944
 945             /* Calculate temporary vectorial force */
 946             tx               = _mm_mul_ps(fscal,dx00);
 947             ty               = _mm_mul_ps(fscal,dy00);
 948             tz               = _mm_mul_ps(fscal,dz00);
 949
 950             /* Update vectorial force */
 951             fix0             = _mm_add_ps(fix0,tx);
 952             fiy0             = _mm_add_ps(fiy0,ty);
 953             fiz0             = _mm_add_ps(fiz0,tz);
 954
 955             fjx0             = _mm_add_ps(fjx0,tx);
 956             fjy0             = _mm_add_ps(fjy0,ty);
 957             fjz0             = _mm_add_ps(fjz0,tz);
 958
 959             /**************************
 960              * CALCULATE INTERACTIONS *
 961              **************************/
 962
 963             r10              = _mm_mul_ps(rsq10,rinv10);
 964
 965             /* Compute parameters for interactions between i and j atoms */
 966             qq10             = _mm_mul_ps(iq1,jq0);
 967
 968             /* Calculate table index by multiplying r with table scale and truncate to integer */
 969             rt               = _mm_mul_ps(r10,vftabscale);
 970             vfitab           = _mm_cvttps_epi32(rt);
 971             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 972             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 973
 974             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 975             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 976             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 977             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 978             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 979             _MM_TRANSPOSE4_PS(Y,F,G,H);
 980             Heps             = _mm_mul_ps(vfeps,H);
 981             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 982             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 983             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 984
 985             fscal            = felec;
 986
 987             /* Calculate temporary vectorial force */
 988             tx               = _mm_mul_ps(fscal,dx10);
 989             ty               = _mm_mul_ps(fscal,dy10);
 990             tz               = _mm_mul_ps(fscal,dz10);
 991
 992             /* Update vectorial force */
 993             fix1             = _mm_add_ps(fix1,tx);
 994             fiy1             = _mm_add_ps(fiy1,ty);
 995             fiz1             = _mm_add_ps(fiz1,tz);
 996
 997             fjx0             = _mm_add_ps(fjx0,tx);
 998             fjy0             = _mm_add_ps(fjy0,ty);
 999             fjz0             = _mm_add_ps(fjz0,tz);
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             r20              = _mm_mul_ps(rsq20,rinv20);
1006
1007             /* Compute parameters for interactions between i and j atoms */
1008             qq20             = _mm_mul_ps(iq2,jq0);
1009
1010             /* Calculate table index by multiplying r with table scale and truncate to integer */
1011             rt               = _mm_mul_ps(r20,vftabscale);
1012             vfitab           = _mm_cvttps_epi32(rt);
1013             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1014             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1015
1016             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1017             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1018             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1019             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1020             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1021             _MM_TRANSPOSE4_PS(Y,F,G,H);
1022             Heps             = _mm_mul_ps(vfeps,H);
1023             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1024             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1025             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1026
1027             fscal            = felec;
1028
1029             /* Calculate temporary vectorial force */
1030             tx               = _mm_mul_ps(fscal,dx20);
1031             ty               = _mm_mul_ps(fscal,dy20);
1032             tz               = _mm_mul_ps(fscal,dz20);
1033
1034             /* Update vectorial force */
1035             fix2             = _mm_add_ps(fix2,tx);
1036             fiy2             = _mm_add_ps(fiy2,ty);
1037             fiz2             = _mm_add_ps(fiz2,tz);
1038
1039             fjx0             = _mm_add_ps(fjx0,tx);
1040             fjy0             = _mm_add_ps(fjy0,ty);
1041             fjz0             = _mm_add_ps(fjz0,tz);
1042
1043             fjptrA             = f+j_coord_offsetA;
1044             fjptrB             = f+j_coord_offsetB;
1045             fjptrC             = f+j_coord_offsetC;
1046             fjptrD             = f+j_coord_offsetD;
1047
1048             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1049
1050             /* Inner loop uses 139 flops */
1051         }
1052
1053         if(jidx<j_index_end)
1054         {
1055
1056             /* Get j neighbor index, and coordinate index */
1057             jnrlistA         = jjnr[jidx];
1058             jnrlistB         = jjnr[jidx+1];
1059             jnrlistC         = jjnr[jidx+2];
1060             jnrlistD         = jjnr[jidx+3];
1061             /* Sign of each element will be negative for non-real atoms.
1062              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1063              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1064              */
1065             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1066             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1067             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1068             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1069             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1070             j_coord_offsetA  = DIM*jnrA;
1071             j_coord_offsetB  = DIM*jnrB;
1072             j_coord_offsetC  = DIM*jnrC;
1073             j_coord_offsetD  = DIM*jnrD;
1074
1075             /* load j atom coordinates */
1076             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1077                                               x+j_coord_offsetC,x+j_coord_offsetD,
1078                                               &jx0,&jy0,&jz0);
1079
1080             /* Calculate displacement vector */
1081             dx00             = _mm_sub_ps(ix0,jx0);
1082             dy00             = _mm_sub_ps(iy0,jy0);
1083             dz00             = _mm_sub_ps(iz0,jz0);
1084             dx10             = _mm_sub_ps(ix1,jx0);
1085             dy10             = _mm_sub_ps(iy1,jy0);
1086             dz10             = _mm_sub_ps(iz1,jz0);
1087             dx20             = _mm_sub_ps(ix2,jx0);
1088             dy20             = _mm_sub_ps(iy2,jy0);
1089             dz20             = _mm_sub_ps(iz2,jz0);
1090
1091             /* Calculate squared distance and things based on it */
1092             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1093             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1094             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1095
1096             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1097             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1098             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1099
1100             /* Load parameters for j particles */
1101             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1102                                                               charge+jnrC+0,charge+jnrD+0);
1103             vdwjidx0A        = 2*vdwtype[jnrA+0];
1104             vdwjidx0B        = 2*vdwtype[jnrB+0];
1105             vdwjidx0C        = 2*vdwtype[jnrC+0];
1106             vdwjidx0D        = 2*vdwtype[jnrD+0];
1107
1108             fjx0             = _mm_setzero_ps();
1109             fjy0             = _mm_setzero_ps();
1110             fjz0             = _mm_setzero_ps();
1111
1112             /**************************
1113              * CALCULATE INTERACTIONS *
1114              **************************/
1115
1116             r00              = _mm_mul_ps(rsq00,rinv00);
1117             r00              = _mm_andnot_ps(dummy_mask,r00);
1118
1119             /* Compute parameters for interactions between i and j atoms */
1120             qq00             = _mm_mul_ps(iq0,jq0);
1121             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1122                                          vdwparam+vdwioffset0+vdwjidx0B,
1123                                          vdwparam+vdwioffset0+vdwjidx0C,
1124                                          vdwparam+vdwioffset0+vdwjidx0D,
1125                                          &c6_00,&c12_00);
1126
1127             /* Calculate table index by multiplying r with table scale and truncate to integer */
1128             rt               = _mm_mul_ps(r00,vftabscale);
1129             vfitab           = _mm_cvttps_epi32(rt);
1130             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1131             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1132
1133             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1134             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1135             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1136             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1137             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1138             _MM_TRANSPOSE4_PS(Y,F,G,H);
1139             Heps             = _mm_mul_ps(vfeps,H);
1140             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1141             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1142             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1143
1144             /* CUBIC SPLINE TABLE DISPERSION */
1145             vfitab           = _mm_add_epi32(vfitab,ifour);
1146             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1147             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1148             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1149             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1150             _MM_TRANSPOSE4_PS(Y,F,G,H);
1151             Heps             = _mm_mul_ps(vfeps,H);
1152             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1153             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1154             fvdw6            = _mm_mul_ps(c6_00,FF);
1155
1156             /* CUBIC SPLINE TABLE REPULSION */
1157             vfitab           = _mm_add_epi32(vfitab,ifour);
1158             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1159             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1160             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1161             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1162             _MM_TRANSPOSE4_PS(Y,F,G,H);
1163             Heps             = _mm_mul_ps(vfeps,H);
1164             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1165             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1166             fvdw12           = _mm_mul_ps(c12_00,FF);
1167             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1168
1169             fscal            = _mm_add_ps(felec,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_cvtepi32_ps(vfitab));
1201             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),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_cvtepi32_ps(vfitab));
1246             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1278             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1279             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1280             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1281
1282             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1283
1284             /* Inner loop uses 142 flops */
1285         }
1286
1287         /* End of innermost loop */
1288
1289         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1290                                               f+i_coord_offset,fshift+i_shift_offset);
1291
1292         /* Increment number of inner iterations */
1293         inneriter                  += j_index_end - j_index_start;
1294
1295         /* Outer loop uses 18 flops */
1296     }
1297
1298     /* Increment number of outer iterations */
1299     outeriter        += nri;
1300
1301     /* Update outer/inner flops */
1302
1303     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1304 }