src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse4_1_double
  54  * Electrostatics interaction: CubicSplineTable
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_elec_vdw->data;
 130     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 164
 165         fix0             = _mm_setzero_pd();
 166         fiy0             = _mm_setzero_pd();
 167         fiz0             = _mm_setzero_pd();
 168         fix1             = _mm_setzero_pd();
 169         fiy1             = _mm_setzero_pd();
 170         fiz1             = _mm_setzero_pd();
 171         fix2             = _mm_setzero_pd();
 172         fiy2             = _mm_setzero_pd();
 173         fiz2             = _mm_setzero_pd();
 174         fix3             = _mm_setzero_pd();
 175         fiy3             = _mm_setzero_pd();
 176         fiz3             = _mm_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_pd();
 180         vvdwsum          = _mm_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_pd(ix0,jx0);
 198             dy00             = _mm_sub_pd(iy0,jy0);
 199             dz00             = _mm_sub_pd(iz0,jz0);
 200             dx10             = _mm_sub_pd(ix1,jx0);
 201             dy10             = _mm_sub_pd(iy1,jy0);
 202             dz10             = _mm_sub_pd(iz1,jz0);
 203             dx20             = _mm_sub_pd(ix2,jx0);
 204             dy20             = _mm_sub_pd(iy2,jy0);
 205             dz20             = _mm_sub_pd(iz2,jz0);
 206             dx30             = _mm_sub_pd(ix3,jx0);
 207             dy30             = _mm_sub_pd(iy3,jy0);
 208             dz30             = _mm_sub_pd(iz3,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 212             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 213             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 214             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 215
 216             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 217             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 218             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 219             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 223             vdwjidx0A        = 2*vdwtype[jnrA+0];
 224             vdwjidx0B        = 2*vdwtype[jnrB+0];
 225
 226             fjx0             = _mm_setzero_pd();
 227             fjy0             = _mm_setzero_pd();
 228             fjz0             = _mm_setzero_pd();
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             r00              = _mm_mul_pd(rsq00,rinv00);
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 238                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 239
 240             /* Calculate table index by multiplying r with table scale and truncate to integer */
 241             rt               = _mm_mul_pd(r00,vftabscale);
 242             vfitab           = _mm_cvttpd_epi32(rt);
 243             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 244             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 245
 246             /* CUBIC SPLINE TABLE DISPERSION */
 247             vfitab           = _mm_add_epi32(vfitab,ifour);
 248             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 249             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 250             GMX_MM_TRANSPOSE2_PD(Y,F);
 251             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 252             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 253             GMX_MM_TRANSPOSE2_PD(G,H);
 254             Heps             = _mm_mul_pd(vfeps,H);
 255             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 256             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 257             vvdw6            = _mm_mul_pd(c6_00,VV);
 258             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 259             fvdw6            = _mm_mul_pd(c6_00,FF);
 260
 261             /* CUBIC SPLINE TABLE REPULSION */
 262             vfitab           = _mm_add_epi32(vfitab,ifour);
 263             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 264             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 265             GMX_MM_TRANSPOSE2_PD(Y,F);
 266             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 267             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 268             GMX_MM_TRANSPOSE2_PD(G,H);
 269             Heps             = _mm_mul_pd(vfeps,H);
 270             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 271             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 272             vvdw12           = _mm_mul_pd(c12_00,VV);
 273             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 274             fvdw12           = _mm_mul_pd(c12_00,FF);
 275             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 276             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 280
 281             fscal            = fvdw;
 282
 283             /* Calculate temporary vectorial force */
 284             tx               = _mm_mul_pd(fscal,dx00);
 285             ty               = _mm_mul_pd(fscal,dy00);
 286             tz               = _mm_mul_pd(fscal,dz00);
 287
 288             /* Update vectorial force */
 289             fix0             = _mm_add_pd(fix0,tx);
 290             fiy0             = _mm_add_pd(fiy0,ty);
 291             fiz0             = _mm_add_pd(fiz0,tz);
 292
 293             fjx0             = _mm_add_pd(fjx0,tx);
 294             fjy0             = _mm_add_pd(fjy0,ty);
 295             fjz0             = _mm_add_pd(fjz0,tz);
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             r10              = _mm_mul_pd(rsq10,rinv10);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_pd(iq1,jq0);
 305
 306             /* Calculate table index by multiplying r with table scale and truncate to integer */
 307             rt               = _mm_mul_pd(r10,vftabscale);
 308             vfitab           = _mm_cvttpd_epi32(rt);
 309             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 310             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 311
 312             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 313             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 314             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 315             GMX_MM_TRANSPOSE2_PD(Y,F);
 316             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 317             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 318             GMX_MM_TRANSPOSE2_PD(G,H);
 319             Heps             = _mm_mul_pd(vfeps,H);
 320             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 321             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 322             velec            = _mm_mul_pd(qq10,VV);
 323             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 324             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm_add_pd(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm_mul_pd(fscal,dx10);
 333             ty               = _mm_mul_pd(fscal,dy10);
 334             tz               = _mm_mul_pd(fscal,dz10);
 335
 336             /* Update vectorial force */
 337             fix1             = _mm_add_pd(fix1,tx);
 338             fiy1             = _mm_add_pd(fiy1,ty);
 339             fiz1             = _mm_add_pd(fiz1,tz);
 340
 341             fjx0             = _mm_add_pd(fjx0,tx);
 342             fjy0             = _mm_add_pd(fjy0,ty);
 343             fjz0             = _mm_add_pd(fjz0,tz);
 344
 345             /**************************
 346              * CALCULATE INTERACTIONS *
 347              **************************/
 348
 349             r20              = _mm_mul_pd(rsq20,rinv20);
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq20             = _mm_mul_pd(iq2,jq0);
 353
 354             /* Calculate table index by multiplying r with table scale and truncate to integer */
 355             rt               = _mm_mul_pd(r20,vftabscale);
 356             vfitab           = _mm_cvttpd_epi32(rt);
 357             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 358             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 359
 360             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 361             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 362             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 363             GMX_MM_TRANSPOSE2_PD(Y,F);
 364             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 365             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 366             GMX_MM_TRANSPOSE2_PD(G,H);
 367             Heps             = _mm_mul_pd(vfeps,H);
 368             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 369             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 370             velec            = _mm_mul_pd(qq20,VV);
 371             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 372             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 373
 374             /* Update potential sum for this i atom from the interaction with this j atom. */
 375             velecsum         = _mm_add_pd(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             /* Calculate temporary vectorial force */
 380             tx               = _mm_mul_pd(fscal,dx20);
 381             ty               = _mm_mul_pd(fscal,dy20);
 382             tz               = _mm_mul_pd(fscal,dz20);
 383
 384             /* Update vectorial force */
 385             fix2             = _mm_add_pd(fix2,tx);
 386             fiy2             = _mm_add_pd(fiy2,ty);
 387             fiz2             = _mm_add_pd(fiz2,tz);
 388
 389             fjx0             = _mm_add_pd(fjx0,tx);
 390             fjy0             = _mm_add_pd(fjy0,ty);
 391             fjz0             = _mm_add_pd(fjz0,tz);
 392
 393             /**************************
 394              * CALCULATE INTERACTIONS *
 395              **************************/
 396
 397             r30              = _mm_mul_pd(rsq30,rinv30);
 398
 399             /* Compute parameters for interactions between i and j atoms */
 400             qq30             = _mm_mul_pd(iq3,jq0);
 401
 402             /* Calculate table index by multiplying r with table scale and truncate to integer */
 403             rt               = _mm_mul_pd(r30,vftabscale);
 404             vfitab           = _mm_cvttpd_epi32(rt);
 405             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 406             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 407
 408             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 409             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 410             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 411             GMX_MM_TRANSPOSE2_PD(Y,F);
 412             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 413             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 414             GMX_MM_TRANSPOSE2_PD(G,H);
 415             Heps             = _mm_mul_pd(vfeps,H);
 416             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 417             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 418             velec            = _mm_mul_pd(qq30,VV);
 419             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 420             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velecsum         = _mm_add_pd(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427             /* Calculate temporary vectorial force */
 428             tx               = _mm_mul_pd(fscal,dx30);
 429             ty               = _mm_mul_pd(fscal,dy30);
 430             tz               = _mm_mul_pd(fscal,dz30);
 431
 432             /* Update vectorial force */
 433             fix3             = _mm_add_pd(fix3,tx);
 434             fiy3             = _mm_add_pd(fiy3,ty);
 435             fiz3             = _mm_add_pd(fiz3,tz);
 436
 437             fjx0             = _mm_add_pd(fjx0,tx);
 438             fjy0             = _mm_add_pd(fjy0,ty);
 439             fjz0             = _mm_add_pd(fjz0,tz);
 440
 441             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 442
 443             /* Inner loop uses 188 flops */
 444         }
 445
 446         if(jidx<j_index_end)
 447         {
 448
 449             jnrA             = jjnr[jidx];
 450             j_coord_offsetA  = DIM*jnrA;
 451
 452             /* load j atom coordinates */
 453             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 454                                               &jx0,&jy0,&jz0);
 455
 456             /* Calculate displacement vector */
 457             dx00             = _mm_sub_pd(ix0,jx0);
 458             dy00             = _mm_sub_pd(iy0,jy0);
 459             dz00             = _mm_sub_pd(iz0,jz0);
 460             dx10             = _mm_sub_pd(ix1,jx0);
 461             dy10             = _mm_sub_pd(iy1,jy0);
 462             dz10             = _mm_sub_pd(iz1,jz0);
 463             dx20             = _mm_sub_pd(ix2,jx0);
 464             dy20             = _mm_sub_pd(iy2,jy0);
 465             dz20             = _mm_sub_pd(iz2,jz0);
 466             dx30             = _mm_sub_pd(ix3,jx0);
 467             dy30             = _mm_sub_pd(iy3,jy0);
 468             dz30             = _mm_sub_pd(iz3,jz0);
 469
 470             /* Calculate squared distance and things based on it */
 471             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 472             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 473             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 474             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 475
 476             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 477             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 478             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 479             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 480
 481             /* Load parameters for j particles */
 482             jq0              = _mm_load_sd(charge+jnrA+0);
 483             vdwjidx0A        = 2*vdwtype[jnrA+0];
 484
 485             fjx0             = _mm_setzero_pd();
 486             fjy0             = _mm_setzero_pd();
 487             fjz0             = _mm_setzero_pd();
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             r00              = _mm_mul_pd(rsq00,rinv00);
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 497
 498             /* Calculate table index by multiplying r with table scale and truncate to integer */
 499             rt               = _mm_mul_pd(r00,vftabscale);
 500             vfitab           = _mm_cvttpd_epi32(rt);
 501             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 502             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 503
 504             /* CUBIC SPLINE TABLE DISPERSION */
 505             vfitab           = _mm_add_epi32(vfitab,ifour);
 506             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 507             F                = _mm_setzero_pd();
 508             GMX_MM_TRANSPOSE2_PD(Y,F);
 509             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 510             H                = _mm_setzero_pd();
 511             GMX_MM_TRANSPOSE2_PD(G,H);
 512             Heps             = _mm_mul_pd(vfeps,H);
 513             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 514             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 515             vvdw6            = _mm_mul_pd(c6_00,VV);
 516             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 517             fvdw6            = _mm_mul_pd(c6_00,FF);
 518
 519             /* CUBIC SPLINE TABLE REPULSION */
 520             vfitab           = _mm_add_epi32(vfitab,ifour);
 521             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 522             F                = _mm_setzero_pd();
 523             GMX_MM_TRANSPOSE2_PD(Y,F);
 524             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 525             H                = _mm_setzero_pd();
 526             GMX_MM_TRANSPOSE2_PD(G,H);
 527             Heps             = _mm_mul_pd(vfeps,H);
 528             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 529             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 530             vvdw12           = _mm_mul_pd(c12_00,VV);
 531             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 532             fvdw12           = _mm_mul_pd(c12_00,FF);
 533             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 534             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 535
 536             /* Update potential sum for this i atom from the interaction with this j atom. */
 537             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 538             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 539
 540             fscal            = fvdw;
 541
 542             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 543
 544             /* Calculate temporary vectorial force */
 545             tx               = _mm_mul_pd(fscal,dx00);
 546             ty               = _mm_mul_pd(fscal,dy00);
 547             tz               = _mm_mul_pd(fscal,dz00);
 548
 549             /* Update vectorial force */
 550             fix0             = _mm_add_pd(fix0,tx);
 551             fiy0             = _mm_add_pd(fiy0,ty);
 552             fiz0             = _mm_add_pd(fiz0,tz);
 553
 554             fjx0             = _mm_add_pd(fjx0,tx);
 555             fjy0             = _mm_add_pd(fjy0,ty);
 556             fjz0             = _mm_add_pd(fjz0,tz);
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             r10              = _mm_mul_pd(rsq10,rinv10);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq10             = _mm_mul_pd(iq1,jq0);
 566
 567             /* Calculate table index by multiplying r with table scale and truncate to integer */
 568             rt               = _mm_mul_pd(r10,vftabscale);
 569             vfitab           = _mm_cvttpd_epi32(rt);
 570             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 571             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 572
 573             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 574             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 575             F                = _mm_setzero_pd();
 576             GMX_MM_TRANSPOSE2_PD(Y,F);
 577             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 578             H                = _mm_setzero_pd();
 579             GMX_MM_TRANSPOSE2_PD(G,H);
 580             Heps             = _mm_mul_pd(vfeps,H);
 581             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 582             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 583             velec            = _mm_mul_pd(qq10,VV);
 584             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 585             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 586
 587             /* Update potential sum for this i atom from the interaction with this j atom. */
 588             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 589             velecsum         = _mm_add_pd(velecsum,velec);
 590
 591             fscal            = felec;
 592
 593             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm_mul_pd(fscal,dx10);
 597             ty               = _mm_mul_pd(fscal,dy10);
 598             tz               = _mm_mul_pd(fscal,dz10);
 599
 600             /* Update vectorial force */
 601             fix1             = _mm_add_pd(fix1,tx);
 602             fiy1             = _mm_add_pd(fiy1,ty);
 603             fiz1             = _mm_add_pd(fiz1,tz);
 604
 605             fjx0             = _mm_add_pd(fjx0,tx);
 606             fjy0             = _mm_add_pd(fjy0,ty);
 607             fjz0             = _mm_add_pd(fjz0,tz);
 608
 609             /**************************
 610              * CALCULATE INTERACTIONS *
 611              **************************/
 612
 613             r20              = _mm_mul_pd(rsq20,rinv20);
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq20             = _mm_mul_pd(iq2,jq0);
 617
 618             /* Calculate table index by multiplying r with table scale and truncate to integer */
 619             rt               = _mm_mul_pd(r20,vftabscale);
 620             vfitab           = _mm_cvttpd_epi32(rt);
 621             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 622             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 623
 624             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 625             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 626             F                = _mm_setzero_pd();
 627             GMX_MM_TRANSPOSE2_PD(Y,F);
 628             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 629             H                = _mm_setzero_pd();
 630             GMX_MM_TRANSPOSE2_PD(G,H);
 631             Heps             = _mm_mul_pd(vfeps,H);
 632             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 633             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 634             velec            = _mm_mul_pd(qq20,VV);
 635             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 636             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 637
 638             /* Update potential sum for this i atom from the interaction with this j atom. */
 639             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 640             velecsum         = _mm_add_pd(velecsum,velec);
 641
 642             fscal            = felec;
 643
 644             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 645
 646             /* Calculate temporary vectorial force */
 647             tx               = _mm_mul_pd(fscal,dx20);
 648             ty               = _mm_mul_pd(fscal,dy20);
 649             tz               = _mm_mul_pd(fscal,dz20);
 650
 651             /* Update vectorial force */
 652             fix2             = _mm_add_pd(fix2,tx);
 653             fiy2             = _mm_add_pd(fiy2,ty);
 654             fiz2             = _mm_add_pd(fiz2,tz);
 655
 656             fjx0             = _mm_add_pd(fjx0,tx);
 657             fjy0             = _mm_add_pd(fjy0,ty);
 658             fjz0             = _mm_add_pd(fjz0,tz);
 659
 660             /**************************
 661              * CALCULATE INTERACTIONS *
 662              **************************/
 663
 664             r30              = _mm_mul_pd(rsq30,rinv30);
 665
 666             /* Compute parameters for interactions between i and j atoms */
 667             qq30             = _mm_mul_pd(iq3,jq0);
 668
 669             /* Calculate table index by multiplying r with table scale and truncate to integer */
 670             rt               = _mm_mul_pd(r30,vftabscale);
 671             vfitab           = _mm_cvttpd_epi32(rt);
 672             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 673             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 674
 675             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 676             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 677             F                = _mm_setzero_pd();
 678             GMX_MM_TRANSPOSE2_PD(Y,F);
 679             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 680             H                = _mm_setzero_pd();
 681             GMX_MM_TRANSPOSE2_PD(G,H);
 682             Heps             = _mm_mul_pd(vfeps,H);
 683             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 684             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 685             velec            = _mm_mul_pd(qq30,VV);
 686             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 687             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
 688
 689             /* Update potential sum for this i atom from the interaction with this j atom. */
 690             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 691             velecsum         = _mm_add_pd(velecsum,velec);
 692
 693             fscal            = felec;
 694
 695             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 696
 697             /* Calculate temporary vectorial force */
 698             tx               = _mm_mul_pd(fscal,dx30);
 699             ty               = _mm_mul_pd(fscal,dy30);
 700             tz               = _mm_mul_pd(fscal,dz30);
 701
 702             /* Update vectorial force */
 703             fix3             = _mm_add_pd(fix3,tx);
 704             fiy3             = _mm_add_pd(fiy3,ty);
 705             fiz3             = _mm_add_pd(fiz3,tz);
 706
 707             fjx0             = _mm_add_pd(fjx0,tx);
 708             fjy0             = _mm_add_pd(fjy0,ty);
 709             fjz0             = _mm_add_pd(fjz0,tz);
 710
 711             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 712
 713             /* Inner loop uses 188 flops */
 714         }
 715
 716         /* End of innermost loop */
 717
 718         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 719                                               f+i_coord_offset,fshift+i_shift_offset);
 720
 721         ggid                        = gid[iidx];
 722         /* Update potential energies */
 723         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 724         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 725
 726         /* Increment number of inner iterations */
 727         inneriter                  += j_index_end - j_index_start;
 728
 729         /* Outer loop uses 26 flops */
 730     }
 731
 732     /* Increment number of outer iterations */
 733     outeriter        += nri;
 734
 735     /* Update outer/inner flops */
 736
 737     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*188);
 738 }
 739 /*
 740  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse4_1_double
 741  * Electrostatics interaction: CubicSplineTable
 742  * VdW interaction:            CubicSplineTable
 743  * Geometry:                   Water4-Particle
 744  * Calculate force/pot:        Force
 745  */
 746 void
 747 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse4_1_double
 748                     (t_nblist                    * gmx_restrict       nlist,
 749                      rvec                        * gmx_restrict          xx,
 750                      rvec                        * gmx_restrict          ff,
 751                      t_forcerec                  * gmx_restrict          fr,
 752                      t_mdatoms                   * gmx_restrict     mdatoms,
 753                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 754                      t_nrnb                      * gmx_restrict        nrnb)
 755 {
 756     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 757      * just 0 for non-waters.
 758      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 759      * jnr indices corresponding to data put in the four positions in the SIMD register.
 760      */
 761     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 762     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 763     int              jnrA,jnrB;
 764     int              j_coord_offsetA,j_coord_offsetB;
 765     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 766     real             rcutoff_scalar;
 767     real             *shiftvec,*fshift,*x,*f;
 768     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 769     int              vdwioffset0;
 770     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 771     int              vdwioffset1;
 772     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 773     int              vdwioffset2;
 774     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 775     int              vdwioffset3;
 776     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 777     int              vdwjidx0A,vdwjidx0B;
 778     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 779     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 780     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 781     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 782     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 783     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 784     real             *charge;
 785     int              nvdwtype;
 786     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 787     int              *vdwtype;
 788     real             *vdwparam;
 789     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 790     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 791     __m128i          vfitab;
 792     __m128i          ifour       = _mm_set1_epi32(4);
 793     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 794     real             *vftab;
 795     __m128d          dummy_mask,cutoff_mask;
 796     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 797     __m128d          one     = _mm_set1_pd(1.0);
 798     __m128d          two     = _mm_set1_pd(2.0);
 799     x                = xx[0];
 800     f                = ff[0];
 801
 802     nri              = nlist->nri;
 803     iinr             = nlist->iinr;
 804     jindex           = nlist->jindex;
 805     jjnr             = nlist->jjnr;
 806     shiftidx         = nlist->shift;
 807     gid              = nlist->gid;
 808     shiftvec         = fr->shift_vec[0];
 809     fshift           = fr->fshift[0];
 810     facel            = _mm_set1_pd(fr->epsfac);
 811     charge           = mdatoms->chargeA;
 812     nvdwtype         = fr->ntype;
 813     vdwparam         = fr->nbfp;
 814     vdwtype          = mdatoms->typeA;
 815
 816     vftab            = kernel_data->table_elec_vdw->data;
 817     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 818
 819     /* Setup water-specific parameters */
 820     inr              = nlist->iinr[0];
 821     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 822     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 823     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 824     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 825
 826     /* Avoid stupid compiler warnings */
 827     jnrA = jnrB = 0;
 828     j_coord_offsetA = 0;
 829     j_coord_offsetB = 0;
 830
 831     outeriter        = 0;
 832     inneriter        = 0;
 833
 834     /* Start outer loop over neighborlists */
 835     for(iidx=0; iidx<nri; iidx++)
 836     {
 837         /* Load shift vector for this list */
 838         i_shift_offset   = DIM*shiftidx[iidx];
 839
 840         /* Load limits for loop over neighbors */
 841         j_index_start    = jindex[iidx];
 842         j_index_end      = jindex[iidx+1];
 843
 844         /* Get outer coordinate index */
 845         inr              = iinr[iidx];
 846         i_coord_offset   = DIM*inr;
 847
 848         /* Load i particle coords and add shift vector */
 849         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 850                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 851
 852         fix0             = _mm_setzero_pd();
 853         fiy0             = _mm_setzero_pd();
 854         fiz0             = _mm_setzero_pd();
 855         fix1             = _mm_setzero_pd();
 856         fiy1             = _mm_setzero_pd();
 857         fiz1             = _mm_setzero_pd();
 858         fix2             = _mm_setzero_pd();
 859         fiy2             = _mm_setzero_pd();
 860         fiz2             = _mm_setzero_pd();
 861         fix3             = _mm_setzero_pd();
 862         fiy3             = _mm_setzero_pd();
 863         fiz3             = _mm_setzero_pd();
 864
 865         /* Start inner kernel loop */
 866         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 867         {
 868
 869             /* Get j neighbor index, and coordinate index */
 870             jnrA             = jjnr[jidx];
 871             jnrB             = jjnr[jidx+1];
 872             j_coord_offsetA  = DIM*jnrA;
 873             j_coord_offsetB  = DIM*jnrB;
 874
 875             /* load j atom coordinates */
 876             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 877                                               &jx0,&jy0,&jz0);
 878
 879             /* Calculate displacement vector */
 880             dx00             = _mm_sub_pd(ix0,jx0);
 881             dy00             = _mm_sub_pd(iy0,jy0);
 882             dz00             = _mm_sub_pd(iz0,jz0);
 883             dx10             = _mm_sub_pd(ix1,jx0);
 884             dy10             = _mm_sub_pd(iy1,jy0);
 885             dz10             = _mm_sub_pd(iz1,jz0);
 886             dx20             = _mm_sub_pd(ix2,jx0);
 887             dy20             = _mm_sub_pd(iy2,jy0);
 888             dz20             = _mm_sub_pd(iz2,jz0);
 889             dx30             = _mm_sub_pd(ix3,jx0);
 890             dy30             = _mm_sub_pd(iy3,jy0);
 891             dz30             = _mm_sub_pd(iz3,jz0);
 892
 893             /* Calculate squared distance and things based on it */
 894             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 895             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 896             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 897             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 898
 899             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 900             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 901             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 902             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 903
 904             /* Load parameters for j particles */
 905             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 906             vdwjidx0A        = 2*vdwtype[jnrA+0];
 907             vdwjidx0B        = 2*vdwtype[jnrB+0];
 908
 909             fjx0             = _mm_setzero_pd();
 910             fjy0             = _mm_setzero_pd();
 911             fjz0             = _mm_setzero_pd();
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             r00              = _mm_mul_pd(rsq00,rinv00);
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 921                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 922
 923             /* Calculate table index by multiplying r with table scale and truncate to integer */
 924             rt               = _mm_mul_pd(r00,vftabscale);
 925             vfitab           = _mm_cvttpd_epi32(rt);
 926             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 927             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 928
 929             /* CUBIC SPLINE TABLE DISPERSION */
 930             vfitab           = _mm_add_epi32(vfitab,ifour);
 931             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 932             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 933             GMX_MM_TRANSPOSE2_PD(Y,F);
 934             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 935             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 936             GMX_MM_TRANSPOSE2_PD(G,H);
 937             Heps             = _mm_mul_pd(vfeps,H);
 938             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 939             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 940             fvdw6            = _mm_mul_pd(c6_00,FF);
 941
 942             /* CUBIC SPLINE TABLE REPULSION */
 943             vfitab           = _mm_add_epi32(vfitab,ifour);
 944             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 945             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 946             GMX_MM_TRANSPOSE2_PD(Y,F);
 947             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 948             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 949             GMX_MM_TRANSPOSE2_PD(G,H);
 950             Heps             = _mm_mul_pd(vfeps,H);
 951             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 952             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 953             fvdw12           = _mm_mul_pd(c12_00,FF);
 954             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 955
 956             fscal            = fvdw;
 957
 958             /* Calculate temporary vectorial force */
 959             tx               = _mm_mul_pd(fscal,dx00);
 960             ty               = _mm_mul_pd(fscal,dy00);
 961             tz               = _mm_mul_pd(fscal,dz00);
 962
 963             /* Update vectorial force */
 964             fix0             = _mm_add_pd(fix0,tx);
 965             fiy0             = _mm_add_pd(fiy0,ty);
 966             fiz0             = _mm_add_pd(fiz0,tz);
 967
 968             fjx0             = _mm_add_pd(fjx0,tx);
 969             fjy0             = _mm_add_pd(fjy0,ty);
 970             fjz0             = _mm_add_pd(fjz0,tz);
 971
 972             /**************************
 973              * CALCULATE INTERACTIONS *
 974              **************************/
 975
 976             r10              = _mm_mul_pd(rsq10,rinv10);
 977
 978             /* Compute parameters for interactions between i and j atoms */
 979             qq10             = _mm_mul_pd(iq1,jq0);
 980
 981             /* Calculate table index by multiplying r with table scale and truncate to integer */
 982             rt               = _mm_mul_pd(r10,vftabscale);
 983             vfitab           = _mm_cvttpd_epi32(rt);
 984             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 985             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 986
 987             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 988             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 989             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 990             GMX_MM_TRANSPOSE2_PD(Y,F);
 991             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 992             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 993             GMX_MM_TRANSPOSE2_PD(G,H);
 994             Heps             = _mm_mul_pd(vfeps,H);
 995             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 996             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 997             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 998
 999             fscal            = felec;
1000
1001             /* Calculate temporary vectorial force */
1002             tx               = _mm_mul_pd(fscal,dx10);
1003             ty               = _mm_mul_pd(fscal,dy10);
1004             tz               = _mm_mul_pd(fscal,dz10);
1005
1006             /* Update vectorial force */
1007             fix1             = _mm_add_pd(fix1,tx);
1008             fiy1             = _mm_add_pd(fiy1,ty);
1009             fiz1             = _mm_add_pd(fiz1,tz);
1010
1011             fjx0             = _mm_add_pd(fjx0,tx);
1012             fjy0             = _mm_add_pd(fjy0,ty);
1013             fjz0             = _mm_add_pd(fjz0,tz);
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             r20              = _mm_mul_pd(rsq20,rinv20);
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq20             = _mm_mul_pd(iq2,jq0);
1023
1024             /* Calculate table index by multiplying r with table scale and truncate to integer */
1025             rt               = _mm_mul_pd(r20,vftabscale);
1026             vfitab           = _mm_cvttpd_epi32(rt);
1027             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1028             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1029
1030             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1031             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1032             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
1033             GMX_MM_TRANSPOSE2_PD(Y,F);
1034             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1035             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
1036             GMX_MM_TRANSPOSE2_PD(G,H);
1037             Heps             = _mm_mul_pd(vfeps,H);
1038             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1039             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1040             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1041
1042             fscal            = felec;
1043
1044             /* Calculate temporary vectorial force */
1045             tx               = _mm_mul_pd(fscal,dx20);
1046             ty               = _mm_mul_pd(fscal,dy20);
1047             tz               = _mm_mul_pd(fscal,dz20);
1048
1049             /* Update vectorial force */
1050             fix2             = _mm_add_pd(fix2,tx);
1051             fiy2             = _mm_add_pd(fiy2,ty);
1052             fiz2             = _mm_add_pd(fiz2,tz);
1053
1054             fjx0             = _mm_add_pd(fjx0,tx);
1055             fjy0             = _mm_add_pd(fjy0,ty);
1056             fjz0             = _mm_add_pd(fjz0,tz);
1057
1058             /**************************
1059              * CALCULATE INTERACTIONS *
1060              **************************/
1061
1062             r30              = _mm_mul_pd(rsq30,rinv30);
1063
1064             /* Compute parameters for interactions between i and j atoms */
1065             qq30             = _mm_mul_pd(iq3,jq0);
1066
1067             /* Calculate table index by multiplying r with table scale and truncate to integer */
1068             rt               = _mm_mul_pd(r30,vftabscale);
1069             vfitab           = _mm_cvttpd_epi32(rt);
1070             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1071             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1072
1073             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1074             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1075             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
1076             GMX_MM_TRANSPOSE2_PD(Y,F);
1077             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1078             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
1079             GMX_MM_TRANSPOSE2_PD(G,H);
1080             Heps             = _mm_mul_pd(vfeps,H);
1081             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1082             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1083             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1084
1085             fscal            = felec;
1086
1087             /* Calculate temporary vectorial force */
1088             tx               = _mm_mul_pd(fscal,dx30);
1089             ty               = _mm_mul_pd(fscal,dy30);
1090             tz               = _mm_mul_pd(fscal,dz30);
1091
1092             /* Update vectorial force */
1093             fix3             = _mm_add_pd(fix3,tx);
1094             fiy3             = _mm_add_pd(fiy3,ty);
1095             fiz3             = _mm_add_pd(fiz3,tz);
1096
1097             fjx0             = _mm_add_pd(fjx0,tx);
1098             fjy0             = _mm_add_pd(fjy0,ty);
1099             fjz0             = _mm_add_pd(fjz0,tz);
1100
1101             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1102
1103             /* Inner loop uses 168 flops */
1104         }
1105
1106         if(jidx<j_index_end)
1107         {
1108
1109             jnrA             = jjnr[jidx];
1110             j_coord_offsetA  = DIM*jnrA;
1111
1112             /* load j atom coordinates */
1113             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1114                                               &jx0,&jy0,&jz0);
1115
1116             /* Calculate displacement vector */
1117             dx00             = _mm_sub_pd(ix0,jx0);
1118             dy00             = _mm_sub_pd(iy0,jy0);
1119             dz00             = _mm_sub_pd(iz0,jz0);
1120             dx10             = _mm_sub_pd(ix1,jx0);
1121             dy10             = _mm_sub_pd(iy1,jy0);
1122             dz10             = _mm_sub_pd(iz1,jz0);
1123             dx20             = _mm_sub_pd(ix2,jx0);
1124             dy20             = _mm_sub_pd(iy2,jy0);
1125             dz20             = _mm_sub_pd(iz2,jz0);
1126             dx30             = _mm_sub_pd(ix3,jx0);
1127             dy30             = _mm_sub_pd(iy3,jy0);
1128             dz30             = _mm_sub_pd(iz3,jz0);
1129
1130             /* Calculate squared distance and things based on it */
1131             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1132             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1133             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1134             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1135
1136             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1137             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1138             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1139             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1140
1141             /* Load parameters for j particles */
1142             jq0              = _mm_load_sd(charge+jnrA+0);
1143             vdwjidx0A        = 2*vdwtype[jnrA+0];
1144
1145             fjx0             = _mm_setzero_pd();
1146             fjy0             = _mm_setzero_pd();
1147             fjz0             = _mm_setzero_pd();
1148
1149             /**************************
1150              * CALCULATE INTERACTIONS *
1151              **************************/
1152
1153             r00              = _mm_mul_pd(rsq00,rinv00);
1154
1155             /* Compute parameters for interactions between i and j atoms */
1156             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1157
1158             /* Calculate table index by multiplying r with table scale and truncate to integer */
1159             rt               = _mm_mul_pd(r00,vftabscale);
1160             vfitab           = _mm_cvttpd_epi32(rt);
1161             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1162             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1163
1164             /* CUBIC SPLINE TABLE DISPERSION */
1165             vfitab           = _mm_add_epi32(vfitab,ifour);
1166             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1167             F                = _mm_setzero_pd();
1168             GMX_MM_TRANSPOSE2_PD(Y,F);
1169             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1170             H                = _mm_setzero_pd();
1171             GMX_MM_TRANSPOSE2_PD(G,H);
1172             Heps             = _mm_mul_pd(vfeps,H);
1173             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1174             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1175             fvdw6            = _mm_mul_pd(c6_00,FF);
1176
1177             /* CUBIC SPLINE TABLE REPULSION */
1178             vfitab           = _mm_add_epi32(vfitab,ifour);
1179             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1180             F                = _mm_setzero_pd();
1181             GMX_MM_TRANSPOSE2_PD(Y,F);
1182             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1183             H                = _mm_setzero_pd();
1184             GMX_MM_TRANSPOSE2_PD(G,H);
1185             Heps             = _mm_mul_pd(vfeps,H);
1186             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1187             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1188             fvdw12           = _mm_mul_pd(c12_00,FF);
1189             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1190
1191             fscal            = fvdw;
1192
1193             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1194
1195             /* Calculate temporary vectorial force */
1196             tx               = _mm_mul_pd(fscal,dx00);
1197             ty               = _mm_mul_pd(fscal,dy00);
1198             tz               = _mm_mul_pd(fscal,dz00);
1199
1200             /* Update vectorial force */
1201             fix0             = _mm_add_pd(fix0,tx);
1202             fiy0             = _mm_add_pd(fiy0,ty);
1203             fiz0             = _mm_add_pd(fiz0,tz);
1204
1205             fjx0             = _mm_add_pd(fjx0,tx);
1206             fjy0             = _mm_add_pd(fjy0,ty);
1207             fjz0             = _mm_add_pd(fjz0,tz);
1208
1209             /**************************
1210              * CALCULATE INTERACTIONS *
1211              **************************/
1212
1213             r10              = _mm_mul_pd(rsq10,rinv10);
1214
1215             /* Compute parameters for interactions between i and j atoms */
1216             qq10             = _mm_mul_pd(iq1,jq0);
1217
1218             /* Calculate table index by multiplying r with table scale and truncate to integer */
1219             rt               = _mm_mul_pd(r10,vftabscale);
1220             vfitab           = _mm_cvttpd_epi32(rt);
1221             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1222             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1223
1224             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1225             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1226             F                = _mm_setzero_pd();
1227             GMX_MM_TRANSPOSE2_PD(Y,F);
1228             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1229             H                = _mm_setzero_pd();
1230             GMX_MM_TRANSPOSE2_PD(G,H);
1231             Heps             = _mm_mul_pd(vfeps,H);
1232             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1233             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1234             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1235
1236             fscal            = felec;
1237
1238             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1239
1240             /* Calculate temporary vectorial force */
1241             tx               = _mm_mul_pd(fscal,dx10);
1242             ty               = _mm_mul_pd(fscal,dy10);
1243             tz               = _mm_mul_pd(fscal,dz10);
1244
1245             /* Update vectorial force */
1246             fix1             = _mm_add_pd(fix1,tx);
1247             fiy1             = _mm_add_pd(fiy1,ty);
1248             fiz1             = _mm_add_pd(fiz1,tz);
1249
1250             fjx0             = _mm_add_pd(fjx0,tx);
1251             fjy0             = _mm_add_pd(fjy0,ty);
1252             fjz0             = _mm_add_pd(fjz0,tz);
1253
1254             /**************************
1255              * CALCULATE INTERACTIONS *
1256              **************************/
1257
1258             r20              = _mm_mul_pd(rsq20,rinv20);
1259
1260             /* Compute parameters for interactions between i and j atoms */
1261             qq20             = _mm_mul_pd(iq2,jq0);
1262
1263             /* Calculate table index by multiplying r with table scale and truncate to integer */
1264             rt               = _mm_mul_pd(r20,vftabscale);
1265             vfitab           = _mm_cvttpd_epi32(rt);
1266             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1267             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1268
1269             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1270             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1271             F                = _mm_setzero_pd();
1272             GMX_MM_TRANSPOSE2_PD(Y,F);
1273             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1274             H                = _mm_setzero_pd();
1275             GMX_MM_TRANSPOSE2_PD(G,H);
1276             Heps             = _mm_mul_pd(vfeps,H);
1277             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1278             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1279             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1280
1281             fscal            = felec;
1282
1283             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1284
1285             /* Calculate temporary vectorial force */
1286             tx               = _mm_mul_pd(fscal,dx20);
1287             ty               = _mm_mul_pd(fscal,dy20);
1288             tz               = _mm_mul_pd(fscal,dz20);
1289
1290             /* Update vectorial force */
1291             fix2             = _mm_add_pd(fix2,tx);
1292             fiy2             = _mm_add_pd(fiy2,ty);
1293             fiz2             = _mm_add_pd(fiz2,tz);
1294
1295             fjx0             = _mm_add_pd(fjx0,tx);
1296             fjy0             = _mm_add_pd(fjy0,ty);
1297             fjz0             = _mm_add_pd(fjz0,tz);
1298
1299             /**************************
1300              * CALCULATE INTERACTIONS *
1301              **************************/
1302
1303             r30              = _mm_mul_pd(rsq30,rinv30);
1304
1305             /* Compute parameters for interactions between i and j atoms */
1306             qq30             = _mm_mul_pd(iq3,jq0);
1307
1308             /* Calculate table index by multiplying r with table scale and truncate to integer */
1309             rt               = _mm_mul_pd(r30,vftabscale);
1310             vfitab           = _mm_cvttpd_epi32(rt);
1311             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1312             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1313
1314             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1315             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1316             F                = _mm_setzero_pd();
1317             GMX_MM_TRANSPOSE2_PD(Y,F);
1318             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1319             H                = _mm_setzero_pd();
1320             GMX_MM_TRANSPOSE2_PD(G,H);
1321             Heps             = _mm_mul_pd(vfeps,H);
1322             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1323             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1324             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq30,FF),_mm_mul_pd(vftabscale,rinv30)));
1325
1326             fscal            = felec;
1327
1328             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1329
1330             /* Calculate temporary vectorial force */
1331             tx               = _mm_mul_pd(fscal,dx30);
1332             ty               = _mm_mul_pd(fscal,dy30);
1333             tz               = _mm_mul_pd(fscal,dz30);
1334
1335             /* Update vectorial force */
1336             fix3             = _mm_add_pd(fix3,tx);
1337             fiy3             = _mm_add_pd(fiy3,ty);
1338             fiz3             = _mm_add_pd(fiz3,tz);
1339
1340             fjx0             = _mm_add_pd(fjx0,tx);
1341             fjy0             = _mm_add_pd(fjy0,ty);
1342             fjz0             = _mm_add_pd(fjz0,tz);
1343
1344             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1345
1346             /* Inner loop uses 168 flops */
1347         }
1348
1349         /* End of innermost loop */
1350
1351         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1352                                               f+i_coord_offset,fshift+i_shift_offset);
1353
1354         /* Increment number of inner iterations */
1355         inneriter                  += j_index_end - j_index_start;
1356
1357         /* Outer loop uses 24 flops */
1358     }
1359
1360     /* Increment number of outer iterations */
1361     outeriter        += nri;
1362
1363     /* Update outer/inner flops */
1364
1365     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*168);
1366 }