src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS 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_ElecCoul_VdwCSTab_GeomW3P1_VF_sse2_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_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              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 101     real             *vftab;
 102     __m128d          dummy_mask,cutoff_mask;
 103     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 104     __m128d          one     = _mm_set1_pd(1.0);
 105     __m128d          two     = _mm_set1_pd(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_pd(fr->ic->epsfac);
 118     charge           = mdatoms->chargeA;
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_vdw->data;
 124     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 129     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 130     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 131     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 158
 159         fix0             = _mm_setzero_pd();
 160         fiy0             = _mm_setzero_pd();
 161         fiz0             = _mm_setzero_pd();
 162         fix1             = _mm_setzero_pd();
 163         fiy1             = _mm_setzero_pd();
 164         fiz1             = _mm_setzero_pd();
 165         fix2             = _mm_setzero_pd();
 166         fiy2             = _mm_setzero_pd();
 167         fiz2             = _mm_setzero_pd();
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm_setzero_pd();
 171         vvdwsum          = _mm_setzero_pd();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                               &jx0,&jy0,&jz0);
 186
 187             /* Calculate displacement vector */
 188             dx00             = _mm_sub_pd(ix0,jx0);
 189             dy00             = _mm_sub_pd(iy0,jy0);
 190             dz00             = _mm_sub_pd(iz0,jz0);
 191             dx10             = _mm_sub_pd(ix1,jx0);
 192             dy10             = _mm_sub_pd(iy1,jy0);
 193             dz10             = _mm_sub_pd(iz1,jz0);
 194             dx20             = _mm_sub_pd(ix2,jx0);
 195             dy20             = _mm_sub_pd(iy2,jy0);
 196             dz20             = _mm_sub_pd(iz2,jz0);
 197
 198             /* Calculate squared distance and things based on it */
 199             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 200             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 201             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 202
 203             rinv00           = sse2_invsqrt_d(rsq00);
 204             rinv10           = sse2_invsqrt_d(rsq10);
 205             rinv20           = sse2_invsqrt_d(rsq20);
 206
 207             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 208             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 209             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 210
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 213             vdwjidx0A        = 2*vdwtype[jnrA+0];
 214             vdwjidx0B        = 2*vdwtype[jnrB+0];
 215
 216             fjx0             = _mm_setzero_pd();
 217             fjy0             = _mm_setzero_pd();
 218             fjz0             = _mm_setzero_pd();
 219
 220             /**************************
 221              * CALCULATE INTERACTIONS *
 222              **************************/
 223
 224             r00              = _mm_mul_pd(rsq00,rinv00);
 225
 226             /* Compute parameters for interactions between i and j atoms */
 227             qq00             = _mm_mul_pd(iq0,jq0);
 228             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 229                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 230
 231             /* Calculate table index by multiplying r with table scale and truncate to integer */
 232             rt               = _mm_mul_pd(r00,vftabscale);
 233             vfitab           = _mm_cvttpd_epi32(rt);
 234             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 235             vfitab           = _mm_slli_epi32(vfitab,3);
 236
 237             /* COULOMB ELECTROSTATICS */
 238             velec            = _mm_mul_pd(qq00,rinv00);
 239             felec            = _mm_mul_pd(velec,rinvsq00);
 240
 241             /* CUBIC SPLINE TABLE DISPERSION */
 242             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 243             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 244             GMX_MM_TRANSPOSE2_PD(Y,F);
 245             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 246             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 247             GMX_MM_TRANSPOSE2_PD(G,H);
 248             Heps             = _mm_mul_pd(vfeps,H);
 249             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 250             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 251             vvdw6            = _mm_mul_pd(c6_00,VV);
 252             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 253             fvdw6            = _mm_mul_pd(c6_00,FF);
 254
 255             /* CUBIC SPLINE TABLE REPULSION */
 256             vfitab           = _mm_add_epi32(vfitab,ifour);
 257             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 258             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 259             GMX_MM_TRANSPOSE2_PD(Y,F);
 260             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 261             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 262             GMX_MM_TRANSPOSE2_PD(G,H);
 263             Heps             = _mm_mul_pd(vfeps,H);
 264             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 265             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 266             vvdw12           = _mm_mul_pd(c12_00,VV);
 267             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 268             fvdw12           = _mm_mul_pd(c12_00,FF);
 269             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 270             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velecsum         = _mm_add_pd(velecsum,velec);
 274             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 275
 276             fscal            = _mm_add_pd(felec,fvdw);
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm_mul_pd(fscal,dx00);
 280             ty               = _mm_mul_pd(fscal,dy00);
 281             tz               = _mm_mul_pd(fscal,dz00);
 282
 283             /* Update vectorial force */
 284             fix0             = _mm_add_pd(fix0,tx);
 285             fiy0             = _mm_add_pd(fiy0,ty);
 286             fiz0             = _mm_add_pd(fiz0,tz);
 287
 288             fjx0             = _mm_add_pd(fjx0,tx);
 289             fjy0             = _mm_add_pd(fjy0,ty);
 290             fjz0             = _mm_add_pd(fjz0,tz);
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq10             = _mm_mul_pd(iq1,jq0);
 298
 299             /* COULOMB ELECTROSTATICS */
 300             velec            = _mm_mul_pd(qq10,rinv10);
 301             felec            = _mm_mul_pd(velec,rinvsq10);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velecsum         = _mm_add_pd(velecsum,velec);
 305
 306             fscal            = felec;
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm_mul_pd(fscal,dx10);
 310             ty               = _mm_mul_pd(fscal,dy10);
 311             tz               = _mm_mul_pd(fscal,dz10);
 312
 313             /* Update vectorial force */
 314             fix1             = _mm_add_pd(fix1,tx);
 315             fiy1             = _mm_add_pd(fiy1,ty);
 316             fiz1             = _mm_add_pd(fiz1,tz);
 317
 318             fjx0             = _mm_add_pd(fjx0,tx);
 319             fjy0             = _mm_add_pd(fjy0,ty);
 320             fjz0             = _mm_add_pd(fjz0,tz);
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq20             = _mm_mul_pd(iq2,jq0);
 328
 329             /* COULOMB ELECTROSTATICS */
 330             velec            = _mm_mul_pd(qq20,rinv20);
 331             felec            = _mm_mul_pd(velec,rinvsq20);
 332
 333             /* Update potential sum for this i atom from the interaction with this j atom. */
 334             velecsum         = _mm_add_pd(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm_mul_pd(fscal,dx20);
 340             ty               = _mm_mul_pd(fscal,dy20);
 341             tz               = _mm_mul_pd(fscal,dz20);
 342
 343             /* Update vectorial force */
 344             fix2             = _mm_add_pd(fix2,tx);
 345             fiy2             = _mm_add_pd(fiy2,ty);
 346             fiz2             = _mm_add_pd(fiz2,tz);
 347
 348             fjx0             = _mm_add_pd(fjx0,tx);
 349             fjy0             = _mm_add_pd(fjy0,ty);
 350             fjz0             = _mm_add_pd(fjz0,tz);
 351
 352             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 353
 354             /* Inner loop uses 122 flops */
 355         }
 356
 357         if(jidx<j_index_end)
 358         {
 359
 360             jnrA             = jjnr[jidx];
 361             j_coord_offsetA  = DIM*jnrA;
 362
 363             /* load j atom coordinates */
 364             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 365                                               &jx0,&jy0,&jz0);
 366
 367             /* Calculate displacement vector */
 368             dx00             = _mm_sub_pd(ix0,jx0);
 369             dy00             = _mm_sub_pd(iy0,jy0);
 370             dz00             = _mm_sub_pd(iz0,jz0);
 371             dx10             = _mm_sub_pd(ix1,jx0);
 372             dy10             = _mm_sub_pd(iy1,jy0);
 373             dz10             = _mm_sub_pd(iz1,jz0);
 374             dx20             = _mm_sub_pd(ix2,jx0);
 375             dy20             = _mm_sub_pd(iy2,jy0);
 376             dz20             = _mm_sub_pd(iz2,jz0);
 377
 378             /* Calculate squared distance and things based on it */
 379             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 380             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 381             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 382
 383             rinv00           = sse2_invsqrt_d(rsq00);
 384             rinv10           = sse2_invsqrt_d(rsq10);
 385             rinv20           = sse2_invsqrt_d(rsq20);
 386
 387             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 388             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 389             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 390
 391             /* Load parameters for j particles */
 392             jq0              = _mm_load_sd(charge+jnrA+0);
 393             vdwjidx0A        = 2*vdwtype[jnrA+0];
 394
 395             fjx0             = _mm_setzero_pd();
 396             fjy0             = _mm_setzero_pd();
 397             fjz0             = _mm_setzero_pd();
 398
 399             /**************************
 400              * CALCULATE INTERACTIONS *
 401              **************************/
 402
 403             r00              = _mm_mul_pd(rsq00,rinv00);
 404
 405             /* Compute parameters for interactions between i and j atoms */
 406             qq00             = _mm_mul_pd(iq0,jq0);
 407             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 408
 409             /* Calculate table index by multiplying r with table scale and truncate to integer */
 410             rt               = _mm_mul_pd(r00,vftabscale);
 411             vfitab           = _mm_cvttpd_epi32(rt);
 412             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 413             vfitab           = _mm_slli_epi32(vfitab,3);
 414
 415             /* COULOMB ELECTROSTATICS */
 416             velec            = _mm_mul_pd(qq00,rinv00);
 417             felec            = _mm_mul_pd(velec,rinvsq00);
 418
 419             /* CUBIC SPLINE TABLE DISPERSION */
 420             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 421             F                = _mm_setzero_pd();
 422             GMX_MM_TRANSPOSE2_PD(Y,F);
 423             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 424             H                = _mm_setzero_pd();
 425             GMX_MM_TRANSPOSE2_PD(G,H);
 426             Heps             = _mm_mul_pd(vfeps,H);
 427             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 428             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 429             vvdw6            = _mm_mul_pd(c6_00,VV);
 430             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 431             fvdw6            = _mm_mul_pd(c6_00,FF);
 432
 433             /* CUBIC SPLINE TABLE REPULSION */
 434             vfitab           = _mm_add_epi32(vfitab,ifour);
 435             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 436             F                = _mm_setzero_pd();
 437             GMX_MM_TRANSPOSE2_PD(Y,F);
 438             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 439             H                = _mm_setzero_pd();
 440             GMX_MM_TRANSPOSE2_PD(G,H);
 441             Heps             = _mm_mul_pd(vfeps,H);
 442             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 443             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 444             vvdw12           = _mm_mul_pd(c12_00,VV);
 445             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 446             fvdw12           = _mm_mul_pd(c12_00,FF);
 447             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 448             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 452             velecsum         = _mm_add_pd(velecsum,velec);
 453             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 454             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 455
 456             fscal            = _mm_add_pd(felec,fvdw);
 457
 458             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 459
 460             /* Calculate temporary vectorial force */
 461             tx               = _mm_mul_pd(fscal,dx00);
 462             ty               = _mm_mul_pd(fscal,dy00);
 463             tz               = _mm_mul_pd(fscal,dz00);
 464
 465             /* Update vectorial force */
 466             fix0             = _mm_add_pd(fix0,tx);
 467             fiy0             = _mm_add_pd(fiy0,ty);
 468             fiz0             = _mm_add_pd(fiz0,tz);
 469
 470             fjx0             = _mm_add_pd(fjx0,tx);
 471             fjy0             = _mm_add_pd(fjy0,ty);
 472             fjz0             = _mm_add_pd(fjz0,tz);
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             /* Compute parameters for interactions between i and j atoms */
 479             qq10             = _mm_mul_pd(iq1,jq0);
 480
 481             /* COULOMB ELECTROSTATICS */
 482             velec            = _mm_mul_pd(qq10,rinv10);
 483             felec            = _mm_mul_pd(velec,rinvsq10);
 484
 485             /* Update potential sum for this i atom from the interaction with this j atom. */
 486             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 487             velecsum         = _mm_add_pd(velecsum,velec);
 488
 489             fscal            = felec;
 490
 491             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 492
 493             /* Calculate temporary vectorial force */
 494             tx               = _mm_mul_pd(fscal,dx10);
 495             ty               = _mm_mul_pd(fscal,dy10);
 496             tz               = _mm_mul_pd(fscal,dz10);
 497
 498             /* Update vectorial force */
 499             fix1             = _mm_add_pd(fix1,tx);
 500             fiy1             = _mm_add_pd(fiy1,ty);
 501             fiz1             = _mm_add_pd(fiz1,tz);
 502
 503             fjx0             = _mm_add_pd(fjx0,tx);
 504             fjy0             = _mm_add_pd(fjy0,ty);
 505             fjz0             = _mm_add_pd(fjz0,tz);
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             qq20             = _mm_mul_pd(iq2,jq0);
 513
 514             /* COULOMB ELECTROSTATICS */
 515             velec            = _mm_mul_pd(qq20,rinv20);
 516             felec            = _mm_mul_pd(velec,rinvsq20);
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 520             velecsum         = _mm_add_pd(velecsum,velec);
 521
 522             fscal            = felec;
 523
 524             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 525
 526             /* Calculate temporary vectorial force */
 527             tx               = _mm_mul_pd(fscal,dx20);
 528             ty               = _mm_mul_pd(fscal,dy20);
 529             tz               = _mm_mul_pd(fscal,dz20);
 530
 531             /* Update vectorial force */
 532             fix2             = _mm_add_pd(fix2,tx);
 533             fiy2             = _mm_add_pd(fiy2,ty);
 534             fiz2             = _mm_add_pd(fiz2,tz);
 535
 536             fjx0             = _mm_add_pd(fjx0,tx);
 537             fjy0             = _mm_add_pd(fjy0,ty);
 538             fjz0             = _mm_add_pd(fjz0,tz);
 539
 540             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 541
 542             /* Inner loop uses 122 flops */
 543         }
 544
 545         /* End of innermost loop */
 546
 547         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 548                                               f+i_coord_offset,fshift+i_shift_offset);
 549
 550         ggid                        = gid[iidx];
 551         /* Update potential energies */
 552         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 553         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 554
 555         /* Increment number of inner iterations */
 556         inneriter                  += j_index_end - j_index_start;
 557
 558         /* Outer loop uses 20 flops */
 559     }
 560
 561     /* Increment number of outer iterations */
 562     outeriter        += nri;
 563
 564     /* Update outer/inner flops */
 565
 566     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*122);
 567 }
 568 /*
 569  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse2_double
 570  * Electrostatics interaction: Coulomb
 571  * VdW interaction:            CubicSplineTable
 572  * Geometry:                   Water3-Particle
 573  * Calculate force/pot:        Force
 574  */
 575 void
 576 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse2_double
 577                     (t_nblist                    * gmx_restrict       nlist,
 578                      rvec                        * gmx_restrict          xx,
 579                      rvec                        * gmx_restrict          ff,
 580                      struct t_forcerec           * gmx_restrict          fr,
 581                      t_mdatoms                   * gmx_restrict     mdatoms,
 582                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 583                      t_nrnb                      * gmx_restrict        nrnb)
 584 {
 585     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 586      * just 0 for non-waters.
 587      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 588      * jnr indices corresponding to data put in the four positions in the SIMD register.
 589      */
 590     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 591     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 592     int              jnrA,jnrB;
 593     int              j_coord_offsetA,j_coord_offsetB;
 594     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 595     real             rcutoff_scalar;
 596     real             *shiftvec,*fshift,*x,*f;
 597     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 598     int              vdwioffset0;
 599     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 600     int              vdwioffset1;
 601     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 602     int              vdwioffset2;
 603     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 604     int              vdwjidx0A,vdwjidx0B;
 605     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 606     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 607     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 608     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 609     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 610     real             *charge;
 611     int              nvdwtype;
 612     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 613     int              *vdwtype;
 614     real             *vdwparam;
 615     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 616     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 617     __m128i          vfitab;
 618     __m128i          ifour       = _mm_set1_epi32(4);
 619     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 620     real             *vftab;
 621     __m128d          dummy_mask,cutoff_mask;
 622     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 623     __m128d          one     = _mm_set1_pd(1.0);
 624     __m128d          two     = _mm_set1_pd(2.0);
 625     x                = xx[0];
 626     f                = ff[0];
 627
 628     nri              = nlist->nri;
 629     iinr             = nlist->iinr;
 630     jindex           = nlist->jindex;
 631     jjnr             = nlist->jjnr;
 632     shiftidx         = nlist->shift;
 633     gid              = nlist->gid;
 634     shiftvec         = fr->shift_vec[0];
 635     fshift           = fr->fshift[0];
 636     facel            = _mm_set1_pd(fr->ic->epsfac);
 637     charge           = mdatoms->chargeA;
 638     nvdwtype         = fr->ntype;
 639     vdwparam         = fr->nbfp;
 640     vdwtype          = mdatoms->typeA;
 641
 642     vftab            = kernel_data->table_vdw->data;
 643     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 644
 645     /* Setup water-specific parameters */
 646     inr              = nlist->iinr[0];
 647     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 648     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 649     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 650     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 651
 652     /* Avoid stupid compiler warnings */
 653     jnrA = jnrB = 0;
 654     j_coord_offsetA = 0;
 655     j_coord_offsetB = 0;
 656
 657     outeriter        = 0;
 658     inneriter        = 0;
 659
 660     /* Start outer loop over neighborlists */
 661     for(iidx=0; iidx<nri; iidx++)
 662     {
 663         /* Load shift vector for this list */
 664         i_shift_offset   = DIM*shiftidx[iidx];
 665
 666         /* Load limits for loop over neighbors */
 667         j_index_start    = jindex[iidx];
 668         j_index_end      = jindex[iidx+1];
 669
 670         /* Get outer coordinate index */
 671         inr              = iinr[iidx];
 672         i_coord_offset   = DIM*inr;
 673
 674         /* Load i particle coords and add shift vector */
 675         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 676                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 677
 678         fix0             = _mm_setzero_pd();
 679         fiy0             = _mm_setzero_pd();
 680         fiz0             = _mm_setzero_pd();
 681         fix1             = _mm_setzero_pd();
 682         fiy1             = _mm_setzero_pd();
 683         fiz1             = _mm_setzero_pd();
 684         fix2             = _mm_setzero_pd();
 685         fiy2             = _mm_setzero_pd();
 686         fiz2             = _mm_setzero_pd();
 687
 688         /* Start inner kernel loop */
 689         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 690         {
 691
 692             /* Get j neighbor index, and coordinate index */
 693             jnrA             = jjnr[jidx];
 694             jnrB             = jjnr[jidx+1];
 695             j_coord_offsetA  = DIM*jnrA;
 696             j_coord_offsetB  = DIM*jnrB;
 697
 698             /* load j atom coordinates */
 699             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 700                                               &jx0,&jy0,&jz0);
 701
 702             /* Calculate displacement vector */
 703             dx00             = _mm_sub_pd(ix0,jx0);
 704             dy00             = _mm_sub_pd(iy0,jy0);
 705             dz00             = _mm_sub_pd(iz0,jz0);
 706             dx10             = _mm_sub_pd(ix1,jx0);
 707             dy10             = _mm_sub_pd(iy1,jy0);
 708             dz10             = _mm_sub_pd(iz1,jz0);
 709             dx20             = _mm_sub_pd(ix2,jx0);
 710             dy20             = _mm_sub_pd(iy2,jy0);
 711             dz20             = _mm_sub_pd(iz2,jz0);
 712
 713             /* Calculate squared distance and things based on it */
 714             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 715             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 716             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 717
 718             rinv00           = sse2_invsqrt_d(rsq00);
 719             rinv10           = sse2_invsqrt_d(rsq10);
 720             rinv20           = sse2_invsqrt_d(rsq20);
 721
 722             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 723             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 724             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 725
 726             /* Load parameters for j particles */
 727             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 728             vdwjidx0A        = 2*vdwtype[jnrA+0];
 729             vdwjidx0B        = 2*vdwtype[jnrB+0];
 730
 731             fjx0             = _mm_setzero_pd();
 732             fjy0             = _mm_setzero_pd();
 733             fjz0             = _mm_setzero_pd();
 734
 735             /**************************
 736              * CALCULATE INTERACTIONS *
 737              **************************/
 738
 739             r00              = _mm_mul_pd(rsq00,rinv00);
 740
 741             /* Compute parameters for interactions between i and j atoms */
 742             qq00             = _mm_mul_pd(iq0,jq0);
 743             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 744                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 745
 746             /* Calculate table index by multiplying r with table scale and truncate to integer */
 747             rt               = _mm_mul_pd(r00,vftabscale);
 748             vfitab           = _mm_cvttpd_epi32(rt);
 749             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 750             vfitab           = _mm_slli_epi32(vfitab,3);
 751
 752             /* COULOMB ELECTROSTATICS */
 753             velec            = _mm_mul_pd(qq00,rinv00);
 754             felec            = _mm_mul_pd(velec,rinvsq00);
 755
 756             /* CUBIC SPLINE TABLE DISPERSION */
 757             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 758             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 759             GMX_MM_TRANSPOSE2_PD(Y,F);
 760             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 761             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 762             GMX_MM_TRANSPOSE2_PD(G,H);
 763             Heps             = _mm_mul_pd(vfeps,H);
 764             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 765             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 766             fvdw6            = _mm_mul_pd(c6_00,FF);
 767
 768             /* CUBIC SPLINE TABLE REPULSION */
 769             vfitab           = _mm_add_epi32(vfitab,ifour);
 770             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 771             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 772             GMX_MM_TRANSPOSE2_PD(Y,F);
 773             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 774             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 775             GMX_MM_TRANSPOSE2_PD(G,H);
 776             Heps             = _mm_mul_pd(vfeps,H);
 777             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 778             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 779             fvdw12           = _mm_mul_pd(c12_00,FF);
 780             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 781
 782             fscal            = _mm_add_pd(felec,fvdw);
 783
 784             /* Calculate temporary vectorial force */
 785             tx               = _mm_mul_pd(fscal,dx00);
 786             ty               = _mm_mul_pd(fscal,dy00);
 787             tz               = _mm_mul_pd(fscal,dz00);
 788
 789             /* Update vectorial force */
 790             fix0             = _mm_add_pd(fix0,tx);
 791             fiy0             = _mm_add_pd(fiy0,ty);
 792             fiz0             = _mm_add_pd(fiz0,tz);
 793
 794             fjx0             = _mm_add_pd(fjx0,tx);
 795             fjy0             = _mm_add_pd(fjy0,ty);
 796             fjz0             = _mm_add_pd(fjz0,tz);
 797
 798             /**************************
 799              * CALCULATE INTERACTIONS *
 800              **************************/
 801
 802             /* Compute parameters for interactions between i and j atoms */
 803             qq10             = _mm_mul_pd(iq1,jq0);
 804
 805             /* COULOMB ELECTROSTATICS */
 806             velec            = _mm_mul_pd(qq10,rinv10);
 807             felec            = _mm_mul_pd(velec,rinvsq10);
 808
 809             fscal            = felec;
 810
 811             /* Calculate temporary vectorial force */
 812             tx               = _mm_mul_pd(fscal,dx10);
 813             ty               = _mm_mul_pd(fscal,dy10);
 814             tz               = _mm_mul_pd(fscal,dz10);
 815
 816             /* Update vectorial force */
 817             fix1             = _mm_add_pd(fix1,tx);
 818             fiy1             = _mm_add_pd(fiy1,ty);
 819             fiz1             = _mm_add_pd(fiz1,tz);
 820
 821             fjx0             = _mm_add_pd(fjx0,tx);
 822             fjy0             = _mm_add_pd(fjy0,ty);
 823             fjz0             = _mm_add_pd(fjz0,tz);
 824
 825             /**************************
 826              * CALCULATE INTERACTIONS *
 827              **************************/
 828
 829             /* Compute parameters for interactions between i and j atoms */
 830             qq20             = _mm_mul_pd(iq2,jq0);
 831
 832             /* COULOMB ELECTROSTATICS */
 833             velec            = _mm_mul_pd(qq20,rinv20);
 834             felec            = _mm_mul_pd(velec,rinvsq20);
 835
 836             fscal            = felec;
 837
 838             /* Calculate temporary vectorial force */
 839             tx               = _mm_mul_pd(fscal,dx20);
 840             ty               = _mm_mul_pd(fscal,dy20);
 841             tz               = _mm_mul_pd(fscal,dz20);
 842
 843             /* Update vectorial force */
 844             fix2             = _mm_add_pd(fix2,tx);
 845             fiy2             = _mm_add_pd(fiy2,ty);
 846             fiz2             = _mm_add_pd(fiz2,tz);
 847
 848             fjx0             = _mm_add_pd(fjx0,tx);
 849             fjy0             = _mm_add_pd(fjy0,ty);
 850             fjz0             = _mm_add_pd(fjz0,tz);
 851
 852             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 853
 854             /* Inner loop uses 111 flops */
 855         }
 856
 857         if(jidx<j_index_end)
 858         {
 859
 860             jnrA             = jjnr[jidx];
 861             j_coord_offsetA  = DIM*jnrA;
 862
 863             /* load j atom coordinates */
 864             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 865                                               &jx0,&jy0,&jz0);
 866
 867             /* Calculate displacement vector */
 868             dx00             = _mm_sub_pd(ix0,jx0);
 869             dy00             = _mm_sub_pd(iy0,jy0);
 870             dz00             = _mm_sub_pd(iz0,jz0);
 871             dx10             = _mm_sub_pd(ix1,jx0);
 872             dy10             = _mm_sub_pd(iy1,jy0);
 873             dz10             = _mm_sub_pd(iz1,jz0);
 874             dx20             = _mm_sub_pd(ix2,jx0);
 875             dy20             = _mm_sub_pd(iy2,jy0);
 876             dz20             = _mm_sub_pd(iz2,jz0);
 877
 878             /* Calculate squared distance and things based on it */
 879             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 880             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 881             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 882
 883             rinv00           = sse2_invsqrt_d(rsq00);
 884             rinv10           = sse2_invsqrt_d(rsq10);
 885             rinv20           = sse2_invsqrt_d(rsq20);
 886
 887             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 888             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 889             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 890
 891             /* Load parameters for j particles */
 892             jq0              = _mm_load_sd(charge+jnrA+0);
 893             vdwjidx0A        = 2*vdwtype[jnrA+0];
 894
 895             fjx0             = _mm_setzero_pd();
 896             fjy0             = _mm_setzero_pd();
 897             fjz0             = _mm_setzero_pd();
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             r00              = _mm_mul_pd(rsq00,rinv00);
 904
 905             /* Compute parameters for interactions between i and j atoms */
 906             qq00             = _mm_mul_pd(iq0,jq0);
 907             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 908
 909             /* Calculate table index by multiplying r with table scale and truncate to integer */
 910             rt               = _mm_mul_pd(r00,vftabscale);
 911             vfitab           = _mm_cvttpd_epi32(rt);
 912             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 913             vfitab           = _mm_slli_epi32(vfitab,3);
 914
 915             /* COULOMB ELECTROSTATICS */
 916             velec            = _mm_mul_pd(qq00,rinv00);
 917             felec            = _mm_mul_pd(velec,rinvsq00);
 918
 919             /* CUBIC SPLINE TABLE DISPERSION */
 920             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 921             F                = _mm_setzero_pd();
 922             GMX_MM_TRANSPOSE2_PD(Y,F);
 923             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 924             H                = _mm_setzero_pd();
 925             GMX_MM_TRANSPOSE2_PD(G,H);
 926             Heps             = _mm_mul_pd(vfeps,H);
 927             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 928             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 929             fvdw6            = _mm_mul_pd(c6_00,FF);
 930
 931             /* CUBIC SPLINE TABLE REPULSION */
 932             vfitab           = _mm_add_epi32(vfitab,ifour);
 933             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 934             F                = _mm_setzero_pd();
 935             GMX_MM_TRANSPOSE2_PD(Y,F);
 936             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 937             H                = _mm_setzero_pd();
 938             GMX_MM_TRANSPOSE2_PD(G,H);
 939             Heps             = _mm_mul_pd(vfeps,H);
 940             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 941             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 942             fvdw12           = _mm_mul_pd(c12_00,FF);
 943             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 944
 945             fscal            = _mm_add_pd(felec,fvdw);
 946
 947             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 948
 949             /* Calculate temporary vectorial force */
 950             tx               = _mm_mul_pd(fscal,dx00);
 951             ty               = _mm_mul_pd(fscal,dy00);
 952             tz               = _mm_mul_pd(fscal,dz00);
 953
 954             /* Update vectorial force */
 955             fix0             = _mm_add_pd(fix0,tx);
 956             fiy0             = _mm_add_pd(fiy0,ty);
 957             fiz0             = _mm_add_pd(fiz0,tz);
 958
 959             fjx0             = _mm_add_pd(fjx0,tx);
 960             fjy0             = _mm_add_pd(fjy0,ty);
 961             fjz0             = _mm_add_pd(fjz0,tz);
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             /* Compute parameters for interactions between i and j atoms */
 968             qq10             = _mm_mul_pd(iq1,jq0);
 969
 970             /* COULOMB ELECTROSTATICS */
 971             velec            = _mm_mul_pd(qq10,rinv10);
 972             felec            = _mm_mul_pd(velec,rinvsq10);
 973
 974             fscal            = felec;
 975
 976             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 977
 978             /* Calculate temporary vectorial force */
 979             tx               = _mm_mul_pd(fscal,dx10);
 980             ty               = _mm_mul_pd(fscal,dy10);
 981             tz               = _mm_mul_pd(fscal,dz10);
 982
 983             /* Update vectorial force */
 984             fix1             = _mm_add_pd(fix1,tx);
 985             fiy1             = _mm_add_pd(fiy1,ty);
 986             fiz1             = _mm_add_pd(fiz1,tz);
 987
 988             fjx0             = _mm_add_pd(fjx0,tx);
 989             fjy0             = _mm_add_pd(fjy0,ty);
 990             fjz0             = _mm_add_pd(fjz0,tz);
 991
 992             /**************************
 993              * CALCULATE INTERACTIONS *
 994              **************************/
 995
 996             /* Compute parameters for interactions between i and j atoms */
 997             qq20             = _mm_mul_pd(iq2,jq0);
 998
 999             /* COULOMB ELECTROSTATICS */
1000             velec            = _mm_mul_pd(qq20,rinv20);
1001             felec            = _mm_mul_pd(velec,rinvsq20);
1002
1003             fscal            = felec;
1004
1005             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1006
1007             /* Calculate temporary vectorial force */
1008             tx               = _mm_mul_pd(fscal,dx20);
1009             ty               = _mm_mul_pd(fscal,dy20);
1010             tz               = _mm_mul_pd(fscal,dz20);
1011
1012             /* Update vectorial force */
1013             fix2             = _mm_add_pd(fix2,tx);
1014             fiy2             = _mm_add_pd(fiy2,ty);
1015             fiz2             = _mm_add_pd(fiz2,tz);
1016
1017             fjx0             = _mm_add_pd(fjx0,tx);
1018             fjy0             = _mm_add_pd(fjy0,ty);
1019             fjz0             = _mm_add_pd(fjz0,tz);
1020
1021             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1022
1023             /* Inner loop uses 111 flops */
1024         }
1025
1026         /* End of innermost loop */
1027
1028         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1029                                               f+i_coord_offset,fshift+i_shift_offset);
1030
1031         /* Increment number of inner iterations */
1032         inneriter                  += j_index_end - j_index_start;
1033
1034         /* Outer loop uses 18 flops */
1035     }
1036
1037     /* Increment number of outer iterations */
1038     outeriter        += nri;
1039
1040     /* Update outer/inner flops */
1041
1042     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*111);
1043 }