src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_sse2_double.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 sse2_double 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_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse2_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse2_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(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_pd(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_pd(kernel_data->table_elec_vdw->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 132     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 133     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 161
 162         fix0             = _mm_setzero_pd();
 163         fiy0             = _mm_setzero_pd();
 164         fiz0             = _mm_setzero_pd();
 165         fix1             = _mm_setzero_pd();
 166         fiy1             = _mm_setzero_pd();
 167         fiz1             = _mm_setzero_pd();
 168         fix2             = _mm_setzero_pd();
 169         fiy2             = _mm_setzero_pd();
 170         fiz2             = _mm_setzero_pd();
 171         fix3             = _mm_setzero_pd();
 172         fiy3             = _mm_setzero_pd();
 173         fiz3             = _mm_setzero_pd();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_pd();
 177         vvdwsum          = _mm_setzero_pd();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188
 189             /* load j atom coordinates */
 190             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                               &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm_sub_pd(ix0,jx0);
 195             dy00             = _mm_sub_pd(iy0,jy0);
 196             dz00             = _mm_sub_pd(iz0,jz0);
 197             dx10             = _mm_sub_pd(ix1,jx0);
 198             dy10             = _mm_sub_pd(iy1,jy0);
 199             dz10             = _mm_sub_pd(iz1,jz0);
 200             dx20             = _mm_sub_pd(ix2,jx0);
 201             dy20             = _mm_sub_pd(iy2,jy0);
 202             dz20             = _mm_sub_pd(iz2,jz0);
 203             dx30             = _mm_sub_pd(ix3,jx0);
 204             dy30             = _mm_sub_pd(iy3,jy0);
 205             dz30             = _mm_sub_pd(iz3,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 209             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 210             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 211             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 212
 213             rinv00           = sse2_invsqrt_d(rsq00);
 214             rinv10           = sse2_invsqrt_d(rsq10);
 215             rinv20           = sse2_invsqrt_d(rsq20);
 216             rinv30           = sse2_invsqrt_d(rsq30);
 217
 218             /* Load parameters for j particles */
 219             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 220             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222
 223             fjx0             = _mm_setzero_pd();
 224             fjy0             = _mm_setzero_pd();
 225             fjz0             = _mm_setzero_pd();
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             r00              = _mm_mul_pd(rsq00,rinv00);
 232
 233             /* Compute parameters for interactions between i and j atoms */
 234             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 235                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 236
 237             /* Calculate table index by multiplying r with table scale and truncate to integer */
 238             rt               = _mm_mul_pd(r00,vftabscale);
 239             vfitab           = _mm_cvttpd_epi32(rt);
 240             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 241             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 242
 243             /* CUBIC SPLINE TABLE DISPERSION */
 244             vfitab           = _mm_add_epi32(vfitab,ifour);
 245             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 246             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 247             GMX_MM_TRANSPOSE2_PD(Y,F);
 248             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 249             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 250             GMX_MM_TRANSPOSE2_PD(G,H);
 251             Heps             = _mm_mul_pd(vfeps,H);
 252             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 253             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 254             vvdw6            = _mm_mul_pd(c6_00,VV);
 255             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 256             fvdw6            = _mm_mul_pd(c6_00,FF);
 257
 258             /* CUBIC SPLINE TABLE REPULSION */
 259             vfitab           = _mm_add_epi32(vfitab,ifour);
 260             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 261             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 262             GMX_MM_TRANSPOSE2_PD(Y,F);
 263             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 264             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 265             GMX_MM_TRANSPOSE2_PD(G,H);
 266             Heps             = _mm_mul_pd(vfeps,H);
 267             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 268             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 269             vvdw12           = _mm_mul_pd(c12_00,VV);
 270             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 271             fvdw12           = _mm_mul_pd(c12_00,FF);
 272             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 273             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 277
 278             fscal            = fvdw;
 279
 280             /* Calculate temporary vectorial force */
 281             tx               = _mm_mul_pd(fscal,dx00);
 282             ty               = _mm_mul_pd(fscal,dy00);
 283             tz               = _mm_mul_pd(fscal,dz00);
 284
 285             /* Update vectorial force */
 286             fix0             = _mm_add_pd(fix0,tx);
 287             fiy0             = _mm_add_pd(fiy0,ty);
 288             fiz0             = _mm_add_pd(fiz0,tz);
 289
 290             fjx0             = _mm_add_pd(fjx0,tx);
 291             fjy0             = _mm_add_pd(fjy0,ty);
 292             fjz0             = _mm_add_pd(fjz0,tz);
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             r10              = _mm_mul_pd(rsq10,rinv10);
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq10             = _mm_mul_pd(iq1,jq0);
 302
 303             /* Calculate table index by multiplying r with table scale and truncate to integer */
 304             rt               = _mm_mul_pd(r10,vftabscale);
 305             vfitab           = _mm_cvttpd_epi32(rt);
 306             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 307             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 308
 309             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 310             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 311             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 312             GMX_MM_TRANSPOSE2_PD(Y,F);
 313             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 314             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 315             GMX_MM_TRANSPOSE2_PD(G,H);
 316             Heps             = _mm_mul_pd(vfeps,H);
 317             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 318             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 319             velec            = _mm_mul_pd(qq10,VV);
 320             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 321             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velecsum         = _mm_add_pd(velecsum,velec);
 325
 326             fscal            = felec;
 327
 328             /* Calculate temporary vectorial force */
 329             tx               = _mm_mul_pd(fscal,dx10);
 330             ty               = _mm_mul_pd(fscal,dy10);
 331             tz               = _mm_mul_pd(fscal,dz10);
 332
 333             /* Update vectorial force */
 334             fix1             = _mm_add_pd(fix1,tx);
 335             fiy1             = _mm_add_pd(fiy1,ty);
 336             fiz1             = _mm_add_pd(fiz1,tz);
 337
 338             fjx0             = _mm_add_pd(fjx0,tx);
 339             fjy0             = _mm_add_pd(fjy0,ty);
 340             fjz0             = _mm_add_pd(fjz0,tz);
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             r20              = _mm_mul_pd(rsq20,rinv20);
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq20             = _mm_mul_pd(iq2,jq0);
 350
 351             /* Calculate table index by multiplying r with table scale and truncate to integer */
 352             rt               = _mm_mul_pd(r20,vftabscale);
 353             vfitab           = _mm_cvttpd_epi32(rt);
 354             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 355             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 356
 357             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 358             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 359             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 360             GMX_MM_TRANSPOSE2_PD(Y,F);
 361             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 362             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 363             GMX_MM_TRANSPOSE2_PD(G,H);
 364             Heps             = _mm_mul_pd(vfeps,H);
 365             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 366             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 367             velec            = _mm_mul_pd(qq20,VV);
 368             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 369             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velecsum         = _mm_add_pd(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm_mul_pd(fscal,dx20);
 378             ty               = _mm_mul_pd(fscal,dy20);
 379             tz               = _mm_mul_pd(fscal,dz20);
 380
 381             /* Update vectorial force */
 382             fix2             = _mm_add_pd(fix2,tx);
 383             fiy2             = _mm_add_pd(fiy2,ty);
 384             fiz2             = _mm_add_pd(fiz2,tz);
 385
 386             fjx0             = _mm_add_pd(fjx0,tx);
 387             fjy0             = _mm_add_pd(fjy0,ty);
 388             fjz0             = _mm_add_pd(fjz0,tz);
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             r30              = _mm_mul_pd(rsq30,rinv30);
 395
 396             /* Compute parameters for interactions between i and j atoms */
 397             qq30             = _mm_mul_pd(iq3,jq0);
 398
 399             /* Calculate table index by multiplying r with table scale and truncate to integer */
 400             rt               = _mm_mul_pd(r30,vftabscale);
 401             vfitab           = _mm_cvttpd_epi32(rt);
 402             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 403             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 404
 405             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 406             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 407             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 408             GMX_MM_TRANSPOSE2_PD(Y,F);
 409             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 410             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 411             GMX_MM_TRANSPOSE2_PD(G,H);
 412             Heps             = _mm_mul_pd(vfeps,H);
 413             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 414             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 415             velec            = _mm_mul_pd(qq30,VV);
 416             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 417             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 418
 419             /* Update potential sum for this i atom from the interaction with this j atom. */
 420             velecsum         = _mm_add_pd(velecsum,velec);
 421
 422             fscal            = felec;
 423
 424             /* Calculate temporary vectorial force */
 425             tx               = _mm_mul_pd(fscal,dx30);
 426             ty               = _mm_mul_pd(fscal,dy30);
 427             tz               = _mm_mul_pd(fscal,dz30);
 428
 429             /* Update vectorial force */
 430             fix3             = _mm_add_pd(fix3,tx);
 431             fiy3             = _mm_add_pd(fiy3,ty);
 432             fiz3             = _mm_add_pd(fiz3,tz);
 433
 434             fjx0             = _mm_add_pd(fjx0,tx);
 435             fjy0             = _mm_add_pd(fjy0,ty);
 436             fjz0             = _mm_add_pd(fjz0,tz);
 437
 438             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 439
 440             /* Inner loop uses 188 flops */
 441         }
 442
 443         if(jidx<j_index_end)
 444         {
 445
 446             jnrA             = jjnr[jidx];
 447             j_coord_offsetA  = DIM*jnrA;
 448
 449             /* load j atom coordinates */
 450             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 451                                               &jx0,&jy0,&jz0);
 452
 453             /* Calculate displacement vector */
 454             dx00             = _mm_sub_pd(ix0,jx0);
 455             dy00             = _mm_sub_pd(iy0,jy0);
 456             dz00             = _mm_sub_pd(iz0,jz0);
 457             dx10             = _mm_sub_pd(ix1,jx0);
 458             dy10             = _mm_sub_pd(iy1,jy0);
 459             dz10             = _mm_sub_pd(iz1,jz0);
 460             dx20             = _mm_sub_pd(ix2,jx0);
 461             dy20             = _mm_sub_pd(iy2,jy0);
 462             dz20             = _mm_sub_pd(iz2,jz0);
 463             dx30             = _mm_sub_pd(ix3,jx0);
 464             dy30             = _mm_sub_pd(iy3,jy0);
 465             dz30             = _mm_sub_pd(iz3,jz0);
 466
 467             /* Calculate squared distance and things based on it */
 468             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 469             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 470             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 471             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 472
 473             rinv00           = sse2_invsqrt_d(rsq00);
 474             rinv10           = sse2_invsqrt_d(rsq10);
 475             rinv20           = sse2_invsqrt_d(rsq20);
 476             rinv30           = sse2_invsqrt_d(rsq30);
 477
 478             /* Load parameters for j particles */
 479             jq0              = _mm_load_sd(charge+jnrA+0);
 480             vdwjidx0A        = 2*vdwtype[jnrA+0];
 481
 482             fjx0             = _mm_setzero_pd();
 483             fjy0             = _mm_setzero_pd();
 484             fjz0             = _mm_setzero_pd();
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             r00              = _mm_mul_pd(rsq00,rinv00);
 491
 492             /* Compute parameters for interactions between i and j atoms */
 493             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 494
 495             /* Calculate table index by multiplying r with table scale and truncate to integer */
 496             rt               = _mm_mul_pd(r00,vftabscale);
 497             vfitab           = _mm_cvttpd_epi32(rt);
 498             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 499             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 500
 501             /* CUBIC SPLINE TABLE DISPERSION */
 502             vfitab           = _mm_add_epi32(vfitab,ifour);
 503             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 504             F                = _mm_setzero_pd();
 505             GMX_MM_TRANSPOSE2_PD(Y,F);
 506             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 507             H                = _mm_setzero_pd();
 508             GMX_MM_TRANSPOSE2_PD(G,H);
 509             Heps             = _mm_mul_pd(vfeps,H);
 510             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 511             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 512             vvdw6            = _mm_mul_pd(c6_00,VV);
 513             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 514             fvdw6            = _mm_mul_pd(c6_00,FF);
 515
 516             /* CUBIC SPLINE TABLE REPULSION */
 517             vfitab           = _mm_add_epi32(vfitab,ifour);
 518             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 519             F                = _mm_setzero_pd();
 520             GMX_MM_TRANSPOSE2_PD(Y,F);
 521             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 522             H                = _mm_setzero_pd();
 523             GMX_MM_TRANSPOSE2_PD(G,H);
 524             Heps             = _mm_mul_pd(vfeps,H);
 525             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 526             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 527             vvdw12           = _mm_mul_pd(c12_00,VV);
 528             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 529             fvdw12           = _mm_mul_pd(c12_00,FF);
 530             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 531             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 532
 533             /* Update potential sum for this i atom from the interaction with this j atom. */
 534             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 535             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 536
 537             fscal            = fvdw;
 538
 539             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm_mul_pd(fscal,dx00);
 543             ty               = _mm_mul_pd(fscal,dy00);
 544             tz               = _mm_mul_pd(fscal,dz00);
 545
 546             /* Update vectorial force */
 547             fix0             = _mm_add_pd(fix0,tx);
 548             fiy0             = _mm_add_pd(fiy0,ty);
 549             fiz0             = _mm_add_pd(fiz0,tz);
 550
 551             fjx0             = _mm_add_pd(fjx0,tx);
 552             fjy0             = _mm_add_pd(fjy0,ty);
 553             fjz0             = _mm_add_pd(fjz0,tz);
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             r10              = _mm_mul_pd(rsq10,rinv10);
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq10             = _mm_mul_pd(iq1,jq0);
 563
 564             /* Calculate table index by multiplying r with table scale and truncate to integer */
 565             rt               = _mm_mul_pd(r10,vftabscale);
 566             vfitab           = _mm_cvttpd_epi32(rt);
 567             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 568             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 569
 570             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 571             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 572             F                = _mm_setzero_pd();
 573             GMX_MM_TRANSPOSE2_PD(Y,F);
 574             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 575             H                = _mm_setzero_pd();
 576             GMX_MM_TRANSPOSE2_PD(G,H);
 577             Heps             = _mm_mul_pd(vfeps,H);
 578             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 579             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 580             velec            = _mm_mul_pd(qq10,VV);
 581             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 582             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 586             velecsum         = _mm_add_pd(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 591
 592             /* Calculate temporary vectorial force */
 593             tx               = _mm_mul_pd(fscal,dx10);
 594             ty               = _mm_mul_pd(fscal,dy10);
 595             tz               = _mm_mul_pd(fscal,dz10);
 596
 597             /* Update vectorial force */
 598             fix1             = _mm_add_pd(fix1,tx);
 599             fiy1             = _mm_add_pd(fiy1,ty);
 600             fiz1             = _mm_add_pd(fiz1,tz);
 601
 602             fjx0             = _mm_add_pd(fjx0,tx);
 603             fjy0             = _mm_add_pd(fjy0,ty);
 604             fjz0             = _mm_add_pd(fjz0,tz);
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             r20              = _mm_mul_pd(rsq20,rinv20);
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             qq20             = _mm_mul_pd(iq2,jq0);
 614
 615             /* Calculate table index by multiplying r with table scale and truncate to integer */
 616             rt               = _mm_mul_pd(r20,vftabscale);
 617             vfitab           = _mm_cvttpd_epi32(rt);
 618             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 619             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 620
 621             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 622             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 623             F                = _mm_setzero_pd();
 624             GMX_MM_TRANSPOSE2_PD(Y,F);
 625             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 626             H                = _mm_setzero_pd();
 627             GMX_MM_TRANSPOSE2_PD(G,H);
 628             Heps             = _mm_mul_pd(vfeps,H);
 629             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 630             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 631             velec            = _mm_mul_pd(qq20,VV);
 632             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 633             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 634
 635             /* Update potential sum for this i atom from the interaction with this j atom. */
 636             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 637             velecsum         = _mm_add_pd(velecsum,velec);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 642
 643             /* Calculate temporary vectorial force */
 644             tx               = _mm_mul_pd(fscal,dx20);
 645             ty               = _mm_mul_pd(fscal,dy20);
 646             tz               = _mm_mul_pd(fscal,dz20);
 647
 648             /* Update vectorial force */
 649             fix2             = _mm_add_pd(fix2,tx);
 650             fiy2             = _mm_add_pd(fiy2,ty);
 651             fiz2             = _mm_add_pd(fiz2,tz);
 652
 653             fjx0             = _mm_add_pd(fjx0,tx);
 654             fjy0             = _mm_add_pd(fjy0,ty);
 655             fjz0             = _mm_add_pd(fjz0,tz);
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             r30              = _mm_mul_pd(rsq30,rinv30);
 662
 663             /* Compute parameters for interactions between i and j atoms */
 664             qq30             = _mm_mul_pd(iq3,jq0);
 665
 666             /* Calculate table index by multiplying r with table scale and truncate to integer */
 667             rt               = _mm_mul_pd(r30,vftabscale);
 668             vfitab           = _mm_cvttpd_epi32(rt);
 669             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 670             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 671
 672             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 673             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 674             F                = _mm_setzero_pd();
 675             GMX_MM_TRANSPOSE2_PD(Y,F);
 676             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 677             H                = _mm_setzero_pd();
 678             GMX_MM_TRANSPOSE2_PD(G,H);
 679             Heps             = _mm_mul_pd(vfeps,H);
 680             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 681             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 682             velec            = _mm_mul_pd(qq30,VV);
 683             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 684             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 685
 686             /* Update potential sum for this i atom from the interaction with this j atom. */
 687             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 688             velecsum         = _mm_add_pd(velecsum,velec);
 689
 690             fscal            = felec;
 691
 692             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 693
 694             /* Calculate temporary vectorial force */
 695             tx               = _mm_mul_pd(fscal,dx30);
 696             ty               = _mm_mul_pd(fscal,dy30);
 697             tz               = _mm_mul_pd(fscal,dz30);
 698
 699             /* Update vectorial force */
 700             fix3             = _mm_add_pd(fix3,tx);
 701             fiy3             = _mm_add_pd(fiy3,ty);
 702             fiz3             = _mm_add_pd(fiz3,tz);
 703
 704             fjx0             = _mm_add_pd(fjx0,tx);
 705             fjy0             = _mm_add_pd(fjy0,ty);
 706             fjz0             = _mm_add_pd(fjz0,tz);
 707
 708             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 709
 710             /* Inner loop uses 188 flops */
 711         }
 712
 713         /* End of innermost loop */
 714
 715         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 716                                               f+i_coord_offset,fshift+i_shift_offset);
 717
 718         ggid                        = gid[iidx];
 719         /* Update potential energies */
 720         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 721         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 722
 723         /* Increment number of inner iterations */
 724         inneriter                  += j_index_end - j_index_start;
 725
 726         /* Outer loop uses 26 flops */
 727     }
 728
 729     /* Increment number of outer iterations */
 730     outeriter        += nri;
 731
 732     /* Update outer/inner flops */
 733
 734     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*188);
 735 }
 736 /*
 737  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse2_double
 738  * Electrostatics interaction: CubicSplineTable
 739  * VdW interaction:            CubicSplineTable
 740  * Geometry:                   Water4-Particle
 741  * Calculate force/pot:        Force
 742  */
 743 void
 744 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse2_double
 745                     (t_nblist                    * gmx_restrict       nlist,
 746                      rvec                        * gmx_restrict          xx,
 747                      rvec                        * gmx_restrict          ff,
 748                      struct t_forcerec           * gmx_restrict          fr,
 749                      t_mdatoms                   * gmx_restrict     mdatoms,
 750                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 751                      t_nrnb                      * gmx_restrict        nrnb)
 752 {
 753     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 754      * just 0 for non-waters.
 755      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 756      * jnr indices corresponding to data put in the four positions in the SIMD register.
 757      */
 758     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 759     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 760     int              jnrA,jnrB;
 761     int              j_coord_offsetA,j_coord_offsetB;
 762     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 763     real             rcutoff_scalar;
 764     real             *shiftvec,*fshift,*x,*f;
 765     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 766     int              vdwioffset0;
 767     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 768     int              vdwioffset1;
 769     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 770     int              vdwioffset2;
 771     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 772     int              vdwioffset3;
 773     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 774     int              vdwjidx0A,vdwjidx0B;
 775     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 776     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 777     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 778     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 779     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 780     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 781     real             *charge;
 782     int              nvdwtype;
 783     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 784     int              *vdwtype;
 785     real             *vdwparam;
 786     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 787     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 788     __m128i          vfitab;
 789     __m128i          ifour       = _mm_set1_epi32(4);
 790     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 791     real             *vftab;
 792     __m128d          dummy_mask,cutoff_mask;
 793     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 794     __m128d          one     = _mm_set1_pd(1.0);
 795     __m128d          two     = _mm_set1_pd(2.0);
 796     x                = xx[0];
 797     f                = ff[0];
 798
 799     nri              = nlist->nri;
 800     iinr             = nlist->iinr;
 801     jindex           = nlist->jindex;
 802     jjnr             = nlist->jjnr;
 803     shiftidx         = nlist->shift;
 804     gid              = nlist->gid;
 805     shiftvec         = fr->shift_vec[0];
 806     fshift           = fr->fshift[0];
 807     facel            = _mm_set1_pd(fr->ic->epsfac);
 808     charge           = mdatoms->chargeA;
 809     nvdwtype         = fr->ntype;
 810     vdwparam         = fr->nbfp;
 811     vdwtype          = mdatoms->typeA;
 812
 813     vftab            = kernel_data->table_elec_vdw->data;
 814     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 815
 816     /* Setup water-specific parameters */
 817     inr              = nlist->iinr[0];
 818     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 819     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 820     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 821     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 822
 823     /* Avoid stupid compiler warnings */
 824     jnrA = jnrB = 0;
 825     j_coord_offsetA = 0;
 826     j_coord_offsetB = 0;
 827
 828     outeriter        = 0;
 829     inneriter        = 0;
 830
 831     /* Start outer loop over neighborlists */
 832     for(iidx=0; iidx<nri; iidx++)
 833     {
 834         /* Load shift vector for this list */
 835         i_shift_offset   = DIM*shiftidx[iidx];
 836
 837         /* Load limits for loop over neighbors */
 838         j_index_start    = jindex[iidx];
 839         j_index_end      = jindex[iidx+1];
 840
 841         /* Get outer coordinate index */
 842         inr              = iinr[iidx];
 843         i_coord_offset   = DIM*inr;
 844
 845         /* Load i particle coords and add shift vector */
 846         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 847                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 848
 849         fix0             = _mm_setzero_pd();
 850         fiy0             = _mm_setzero_pd();
 851         fiz0             = _mm_setzero_pd();
 852         fix1             = _mm_setzero_pd();
 853         fiy1             = _mm_setzero_pd();
 854         fiz1             = _mm_setzero_pd();
 855         fix2             = _mm_setzero_pd();
 856         fiy2             = _mm_setzero_pd();
 857         fiz2             = _mm_setzero_pd();
 858         fix3             = _mm_setzero_pd();
 859         fiy3             = _mm_setzero_pd();
 860         fiz3             = _mm_setzero_pd();
 861
 862         /* Start inner kernel loop */
 863         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 864         {
 865
 866             /* Get j neighbor index, and coordinate index */
 867             jnrA             = jjnr[jidx];
 868             jnrB             = jjnr[jidx+1];
 869             j_coord_offsetA  = DIM*jnrA;
 870             j_coord_offsetB  = DIM*jnrB;
 871
 872             /* load j atom coordinates */
 873             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 874                                               &jx0,&jy0,&jz0);
 875
 876             /* Calculate displacement vector */
 877             dx00             = _mm_sub_pd(ix0,jx0);
 878             dy00             = _mm_sub_pd(iy0,jy0);
 879             dz00             = _mm_sub_pd(iz0,jz0);
 880             dx10             = _mm_sub_pd(ix1,jx0);
 881             dy10             = _mm_sub_pd(iy1,jy0);
 882             dz10             = _mm_sub_pd(iz1,jz0);
 883             dx20             = _mm_sub_pd(ix2,jx0);
 884             dy20             = _mm_sub_pd(iy2,jy0);
 885             dz20             = _mm_sub_pd(iz2,jz0);
 886             dx30             = _mm_sub_pd(ix3,jx0);
 887             dy30             = _mm_sub_pd(iy3,jy0);
 888             dz30             = _mm_sub_pd(iz3,jz0);
 889
 890             /* Calculate squared distance and things based on it */
 891             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 892             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 893             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 894             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 895
 896             rinv00           = sse2_invsqrt_d(rsq00);
 897             rinv10           = sse2_invsqrt_d(rsq10);
 898             rinv20           = sse2_invsqrt_d(rsq20);
 899             rinv30           = sse2_invsqrt_d(rsq30);
 900
 901             /* Load parameters for j particles */
 902             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 903             vdwjidx0A        = 2*vdwtype[jnrA+0];
 904             vdwjidx0B        = 2*vdwtype[jnrB+0];
 905
 906             fjx0             = _mm_setzero_pd();
 907             fjy0             = _mm_setzero_pd();
 908             fjz0             = _mm_setzero_pd();
 909
 910             /**************************
 911              * CALCULATE INTERACTIONS *
 912              **************************/
 913
 914             r00              = _mm_mul_pd(rsq00,rinv00);
 915
 916             /* Compute parameters for interactions between i and j atoms */
 917             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 918                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 919
 920             /* Calculate table index by multiplying r with table scale and truncate to integer */
 921             rt               = _mm_mul_pd(r00,vftabscale);
 922             vfitab           = _mm_cvttpd_epi32(rt);
 923             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 924             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 925
 926             /* CUBIC SPLINE TABLE DISPERSION */
 927             vfitab           = _mm_add_epi32(vfitab,ifour);
 928             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 929             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 930             GMX_MM_TRANSPOSE2_PD(Y,F);
 931             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 932             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 933             GMX_MM_TRANSPOSE2_PD(G,H);
 934             Heps             = _mm_mul_pd(vfeps,H);
 935             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 936             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 937             fvdw6            = _mm_mul_pd(c6_00,FF);
 938
 939             /* CUBIC SPLINE TABLE REPULSION */
 940             vfitab           = _mm_add_epi32(vfitab,ifour);
 941             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 942             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 943             GMX_MM_TRANSPOSE2_PD(Y,F);
 944             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 945             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 946             GMX_MM_TRANSPOSE2_PD(G,H);
 947             Heps             = _mm_mul_pd(vfeps,H);
 948             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 949             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 950             fvdw12           = _mm_mul_pd(c12_00,FF);
 951             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 952
 953             fscal            = fvdw;
 954
 955             /* Calculate temporary vectorial force */
 956             tx               = _mm_mul_pd(fscal,dx00);
 957             ty               = _mm_mul_pd(fscal,dy00);
 958             tz               = _mm_mul_pd(fscal,dz00);
 959
 960             /* Update vectorial force */
 961             fix0             = _mm_add_pd(fix0,tx);
 962             fiy0             = _mm_add_pd(fiy0,ty);
 963             fiz0             = _mm_add_pd(fiz0,tz);
 964
 965             fjx0             = _mm_add_pd(fjx0,tx);
 966             fjy0             = _mm_add_pd(fjy0,ty);
 967             fjz0             = _mm_add_pd(fjz0,tz);
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             r10              = _mm_mul_pd(rsq10,rinv10);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq10             = _mm_mul_pd(iq1,jq0);
 977
 978             /* Calculate table index by multiplying r with table scale and truncate to integer */
 979             rt               = _mm_mul_pd(r10,vftabscale);
 980             vfitab           = _mm_cvttpd_epi32(rt);
 981             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 982             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 983
 984             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 985             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 986             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 987             GMX_MM_TRANSPOSE2_PD(Y,F);
 988             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 989             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 990             GMX_MM_TRANSPOSE2_PD(G,H);
 991             Heps             = _mm_mul_pd(vfeps,H);
 992             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 993             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 994             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 995
 996             fscal            = felec;
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm_mul_pd(fscal,dx10);
1000             ty               = _mm_mul_pd(fscal,dy10);
1001             tz               = _mm_mul_pd(fscal,dz10);
1002
1003             /* Update vectorial force */
1004             fix1             = _mm_add_pd(fix1,tx);
1005             fiy1             = _mm_add_pd(fiy1,ty);
1006             fiz1             = _mm_add_pd(fiz1,tz);
1007
1008             fjx0             = _mm_add_pd(fjx0,tx);
1009             fjy0             = _mm_add_pd(fjy0,ty);
1010             fjz0             = _mm_add_pd(fjz0,tz);
1011
1012             /**************************
1013              * CALCULATE INTERACTIONS *
1014              **************************/
1015
1016             r20              = _mm_mul_pd(rsq20,rinv20);
1017
1018             /* Compute parameters for interactions between i and j atoms */
1019             qq20             = _mm_mul_pd(iq2,jq0);
1020
1021             /* Calculate table index by multiplying r with table scale and truncate to integer */
1022             rt               = _mm_mul_pd(r20,vftabscale);
1023             vfitab           = _mm_cvttpd_epi32(rt);
1024             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1025             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1026
1027             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1028             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1029             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
1030             GMX_MM_TRANSPOSE2_PD(Y,F);
1031             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1032             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
1033             GMX_MM_TRANSPOSE2_PD(G,H);
1034             Heps             = _mm_mul_pd(vfeps,H);
1035             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1036             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1037             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1038
1039             fscal            = felec;
1040
1041             /* Calculate temporary vectorial force */
1042             tx               = _mm_mul_pd(fscal,dx20);
1043             ty               = _mm_mul_pd(fscal,dy20);
1044             tz               = _mm_mul_pd(fscal,dz20);
1045
1046             /* Update vectorial force */
1047             fix2             = _mm_add_pd(fix2,tx);
1048             fiy2             = _mm_add_pd(fiy2,ty);
1049             fiz2             = _mm_add_pd(fiz2,tz);
1050
1051             fjx0             = _mm_add_pd(fjx0,tx);
1052             fjy0             = _mm_add_pd(fjy0,ty);
1053             fjz0             = _mm_add_pd(fjz0,tz);
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             r30              = _mm_mul_pd(rsq30,rinv30);
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq30             = _mm_mul_pd(iq3,jq0);
1063
1064             /* Calculate table index by multiplying r with table scale and truncate to integer */
1065             rt               = _mm_mul_pd(r30,vftabscale);
1066             vfitab           = _mm_cvttpd_epi32(rt);
1067             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1068             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1069
1070             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1071             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1072             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
1073             GMX_MM_TRANSPOSE2_PD(Y,F);
1074             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1075             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
1076             GMX_MM_TRANSPOSE2_PD(G,H);
1077             Heps             = _mm_mul_pd(vfeps,H);
1078             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1079             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1080             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1081
1082             fscal            = felec;
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm_mul_pd(fscal,dx30);
1086             ty               = _mm_mul_pd(fscal,dy30);
1087             tz               = _mm_mul_pd(fscal,dz30);
1088
1089             /* Update vectorial force */
1090             fix3             = _mm_add_pd(fix3,tx);
1091             fiy3             = _mm_add_pd(fiy3,ty);
1092             fiz3             = _mm_add_pd(fiz3,tz);
1093
1094             fjx0             = _mm_add_pd(fjx0,tx);
1095             fjy0             = _mm_add_pd(fjy0,ty);
1096             fjz0             = _mm_add_pd(fjz0,tz);
1097
1098             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1099
1100             /* Inner loop uses 168 flops */
1101         }
1102
1103         if(jidx<j_index_end)
1104         {
1105
1106             jnrA             = jjnr[jidx];
1107             j_coord_offsetA  = DIM*jnrA;
1108
1109             /* load j atom coordinates */
1110             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1111                                               &jx0,&jy0,&jz0);
1112
1113             /* Calculate displacement vector */
1114             dx00             = _mm_sub_pd(ix0,jx0);
1115             dy00             = _mm_sub_pd(iy0,jy0);
1116             dz00             = _mm_sub_pd(iz0,jz0);
1117             dx10             = _mm_sub_pd(ix1,jx0);
1118             dy10             = _mm_sub_pd(iy1,jy0);
1119             dz10             = _mm_sub_pd(iz1,jz0);
1120             dx20             = _mm_sub_pd(ix2,jx0);
1121             dy20             = _mm_sub_pd(iy2,jy0);
1122             dz20             = _mm_sub_pd(iz2,jz0);
1123             dx30             = _mm_sub_pd(ix3,jx0);
1124             dy30             = _mm_sub_pd(iy3,jy0);
1125             dz30             = _mm_sub_pd(iz3,jz0);
1126
1127             /* Calculate squared distance and things based on it */
1128             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1129             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1130             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1131             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1132
1133             rinv00           = sse2_invsqrt_d(rsq00);
1134             rinv10           = sse2_invsqrt_d(rsq10);
1135             rinv20           = sse2_invsqrt_d(rsq20);
1136             rinv30           = sse2_invsqrt_d(rsq30);
1137
1138             /* Load parameters for j particles */
1139             jq0              = _mm_load_sd(charge+jnrA+0);
1140             vdwjidx0A        = 2*vdwtype[jnrA+0];
1141
1142             fjx0             = _mm_setzero_pd();
1143             fjy0             = _mm_setzero_pd();
1144             fjz0             = _mm_setzero_pd();
1145
1146             /**************************
1147              * CALCULATE INTERACTIONS *
1148              **************************/
1149
1150             r00              = _mm_mul_pd(rsq00,rinv00);
1151
1152             /* Compute parameters for interactions between i and j atoms */
1153             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1154
1155             /* Calculate table index by multiplying r with table scale and truncate to integer */
1156             rt               = _mm_mul_pd(r00,vftabscale);
1157             vfitab           = _mm_cvttpd_epi32(rt);
1158             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1159             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1160
1161             /* CUBIC SPLINE TABLE DISPERSION */
1162             vfitab           = _mm_add_epi32(vfitab,ifour);
1163             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1164             F                = _mm_setzero_pd();
1165             GMX_MM_TRANSPOSE2_PD(Y,F);
1166             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1167             H                = _mm_setzero_pd();
1168             GMX_MM_TRANSPOSE2_PD(G,H);
1169             Heps             = _mm_mul_pd(vfeps,H);
1170             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1171             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1172             fvdw6            = _mm_mul_pd(c6_00,FF);
1173
1174             /* CUBIC SPLINE TABLE REPULSION */
1175             vfitab           = _mm_add_epi32(vfitab,ifour);
1176             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1177             F                = _mm_setzero_pd();
1178             GMX_MM_TRANSPOSE2_PD(Y,F);
1179             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1180             H                = _mm_setzero_pd();
1181             GMX_MM_TRANSPOSE2_PD(G,H);
1182             Heps             = _mm_mul_pd(vfeps,H);
1183             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1184             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1185             fvdw12           = _mm_mul_pd(c12_00,FF);
1186             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1187
1188             fscal            = fvdw;
1189
1190             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1191
1192             /* Calculate temporary vectorial force */
1193             tx               = _mm_mul_pd(fscal,dx00);
1194             ty               = _mm_mul_pd(fscal,dy00);
1195             tz               = _mm_mul_pd(fscal,dz00);
1196
1197             /* Update vectorial force */
1198             fix0             = _mm_add_pd(fix0,tx);
1199             fiy0             = _mm_add_pd(fiy0,ty);
1200             fiz0             = _mm_add_pd(fiz0,tz);
1201
1202             fjx0             = _mm_add_pd(fjx0,tx);
1203             fjy0             = _mm_add_pd(fjy0,ty);
1204             fjz0             = _mm_add_pd(fjz0,tz);
1205
1206             /**************************
1207              * CALCULATE INTERACTIONS *
1208              **************************/
1209
1210             r10              = _mm_mul_pd(rsq10,rinv10);
1211
1212             /* Compute parameters for interactions between i and j atoms */
1213             qq10             = _mm_mul_pd(iq1,jq0);
1214
1215             /* Calculate table index by multiplying r with table scale and truncate to integer */
1216             rt               = _mm_mul_pd(r10,vftabscale);
1217             vfitab           = _mm_cvttpd_epi32(rt);
1218             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1219             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1220
1221             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1222             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1223             F                = _mm_setzero_pd();
1224             GMX_MM_TRANSPOSE2_PD(Y,F);
1225             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1226             H                = _mm_setzero_pd();
1227             GMX_MM_TRANSPOSE2_PD(G,H);
1228             Heps             = _mm_mul_pd(vfeps,H);
1229             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1230             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1231             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1232
1233             fscal            = felec;
1234
1235             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1236
1237             /* Calculate temporary vectorial force */
1238             tx               = _mm_mul_pd(fscal,dx10);
1239             ty               = _mm_mul_pd(fscal,dy10);
1240             tz               = _mm_mul_pd(fscal,dz10);
1241
1242             /* Update vectorial force */
1243             fix1             = _mm_add_pd(fix1,tx);
1244             fiy1             = _mm_add_pd(fiy1,ty);
1245             fiz1             = _mm_add_pd(fiz1,tz);
1246
1247             fjx0             = _mm_add_pd(fjx0,tx);
1248             fjy0             = _mm_add_pd(fjy0,ty);
1249             fjz0             = _mm_add_pd(fjz0,tz);
1250
1251             /**************************
1252              * CALCULATE INTERACTIONS *
1253              **************************/
1254
1255             r20              = _mm_mul_pd(rsq20,rinv20);
1256
1257             /* Compute parameters for interactions between i and j atoms */
1258             qq20             = _mm_mul_pd(iq2,jq0);
1259
1260             /* Calculate table index by multiplying r with table scale and truncate to integer */
1261             rt               = _mm_mul_pd(r20,vftabscale);
1262             vfitab           = _mm_cvttpd_epi32(rt);
1263             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1264             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1265
1266             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1267             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1268             F                = _mm_setzero_pd();
1269             GMX_MM_TRANSPOSE2_PD(Y,F);
1270             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1271             H                = _mm_setzero_pd();
1272             GMX_MM_TRANSPOSE2_PD(G,H);
1273             Heps             = _mm_mul_pd(vfeps,H);
1274             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1275             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1276             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1277
1278             fscal            = felec;
1279
1280             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1281
1282             /* Calculate temporary vectorial force */
1283             tx               = _mm_mul_pd(fscal,dx20);
1284             ty               = _mm_mul_pd(fscal,dy20);
1285             tz               = _mm_mul_pd(fscal,dz20);
1286
1287             /* Update vectorial force */
1288             fix2             = _mm_add_pd(fix2,tx);
1289             fiy2             = _mm_add_pd(fiy2,ty);
1290             fiz2             = _mm_add_pd(fiz2,tz);
1291
1292             fjx0             = _mm_add_pd(fjx0,tx);
1293             fjy0             = _mm_add_pd(fjy0,ty);
1294             fjz0             = _mm_add_pd(fjz0,tz);
1295
1296             /**************************
1297              * CALCULATE INTERACTIONS *
1298              **************************/
1299
1300             r30              = _mm_mul_pd(rsq30,rinv30);
1301
1302             /* Compute parameters for interactions between i and j atoms */
1303             qq30             = _mm_mul_pd(iq3,jq0);
1304
1305             /* Calculate table index by multiplying r with table scale and truncate to integer */
1306             rt               = _mm_mul_pd(r30,vftabscale);
1307             vfitab           = _mm_cvttpd_epi32(rt);
1308             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1309             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1310
1311             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1312             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1313             F                = _mm_setzero_pd();
1314             GMX_MM_TRANSPOSE2_PD(Y,F);
1315             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1316             H                = _mm_setzero_pd();
1317             GMX_MM_TRANSPOSE2_PD(G,H);
1318             Heps             = _mm_mul_pd(vfeps,H);
1319             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1320             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1321             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1322
1323             fscal            = felec;
1324
1325             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1326
1327             /* Calculate temporary vectorial force */
1328             tx               = _mm_mul_pd(fscal,dx30);
1329             ty               = _mm_mul_pd(fscal,dy30);
1330             tz               = _mm_mul_pd(fscal,dz30);
1331
1332             /* Update vectorial force */
1333             fix3             = _mm_add_pd(fix3,tx);
1334             fiy3             = _mm_add_pd(fiy3,ty);
1335             fiz3             = _mm_add_pd(fiz3,tz);
1336
1337             fjx0             = _mm_add_pd(fjx0,tx);
1338             fjy0             = _mm_add_pd(fjy0,ty);
1339             fjz0             = _mm_add_pd(fjz0,tz);
1340
1341             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1342
1343             /* Inner loop uses 168 flops */
1344         }
1345
1346         /* End of innermost loop */
1347
1348         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1349                                               f+i_coord_offset,fshift+i_shift_offset);
1350
1351         /* Increment number of inner iterations */
1352         inneriter                  += j_index_end - j_index_start;
1353
1354         /* Outer loop uses 24 flops */
1355     }
1356
1357     /* Increment number of outer iterations */
1358     outeriter        += nri;
1359
1360     /* Update outer/inner flops */
1361
1362     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*168);
1363 }