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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse2_double
  52  * Electrostatics interaction: Coulomb
  53  * VdW interaction:            CubicSplineTable
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m128d          dummy_mask,cutoff_mask;
 104     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 105     __m128d          one     = _mm_set1_pd(1.0);
 106     __m128d          two     = _mm_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_pd(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     vftab            = kernel_data->table_vdw->data;
 125     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 130     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 131     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 159
 160         fix0             = _mm_setzero_pd();
 161         fiy0             = _mm_setzero_pd();
 162         fiz0             = _mm_setzero_pd();
 163         fix1             = _mm_setzero_pd();
 164         fiy1             = _mm_setzero_pd();
 165         fiz1             = _mm_setzero_pd();
 166         fix2             = _mm_setzero_pd();
 167         fiy2             = _mm_setzero_pd();
 168         fiz2             = _mm_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_pd();
 172         vvdwsum          = _mm_setzero_pd();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_pd(ix0,jx0);
 190             dy00             = _mm_sub_pd(iy0,jy0);
 191             dz00             = _mm_sub_pd(iz0,jz0);
 192             dx10             = _mm_sub_pd(ix1,jx0);
 193             dy10             = _mm_sub_pd(iy1,jy0);
 194             dz10             = _mm_sub_pd(iz1,jz0);
 195             dx20             = _mm_sub_pd(ix2,jx0);
 196             dy20             = _mm_sub_pd(iy2,jy0);
 197             dz20             = _mm_sub_pd(iz2,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 201             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 202             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 203
 204             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 205             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 206             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 207
 208             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 209             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 210             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216
 217             fjx0             = _mm_setzero_pd();
 218             fjy0             = _mm_setzero_pd();
 219             fjz0             = _mm_setzero_pd();
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_pd(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm_mul_pd(iq0,jq0);
 229             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 231
 232             /* Calculate table index by multiplying r with table scale and truncate to integer */
 233             rt               = _mm_mul_pd(r00,vftabscale);
 234             vfitab           = _mm_cvttpd_epi32(rt);
 235             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 236             vfitab           = _mm_slli_epi32(vfitab,3);
 237
 238             /* COULOMB ELECTROSTATICS */
 239             velec            = _mm_mul_pd(qq00,rinv00);
 240             felec            = _mm_mul_pd(velec,rinvsq00);
 241
 242             /* CUBIC SPLINE TABLE DISPERSION */
 243             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 244             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 245             GMX_MM_TRANSPOSE2_PD(Y,F);
 246             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 247             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 248             GMX_MM_TRANSPOSE2_PD(G,H);
 249             Heps             = _mm_mul_pd(vfeps,H);
 250             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 251             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 252             vvdw6            = _mm_mul_pd(c6_00,VV);
 253             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 254             fvdw6            = _mm_mul_pd(c6_00,FF);
 255
 256             /* CUBIC SPLINE TABLE REPULSION */
 257             vfitab           = _mm_add_epi32(vfitab,ifour);
 258             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 259             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 260             GMX_MM_TRANSPOSE2_PD(Y,F);
 261             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 262             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 263             GMX_MM_TRANSPOSE2_PD(G,H);
 264             Heps             = _mm_mul_pd(vfeps,H);
 265             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 266             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 267             vvdw12           = _mm_mul_pd(c12_00,VV);
 268             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 269             fvdw12           = _mm_mul_pd(c12_00,FF);
 270             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 271             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             velecsum         = _mm_add_pd(velecsum,velec);
 275             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 276
 277             fscal            = _mm_add_pd(felec,fvdw);
 278
 279             /* Calculate temporary vectorial force */
 280             tx               = _mm_mul_pd(fscal,dx00);
 281             ty               = _mm_mul_pd(fscal,dy00);
 282             tz               = _mm_mul_pd(fscal,dz00);
 283
 284             /* Update vectorial force */
 285             fix0             = _mm_add_pd(fix0,tx);
 286             fiy0             = _mm_add_pd(fiy0,ty);
 287             fiz0             = _mm_add_pd(fiz0,tz);
 288
 289             fjx0             = _mm_add_pd(fjx0,tx);
 290             fjy0             = _mm_add_pd(fjy0,ty);
 291             fjz0             = _mm_add_pd(fjz0,tz);
 292
 293             /**************************
 294              * CALCULATE INTERACTIONS *
 295              **************************/
 296
 297             /* Compute parameters for interactions between i and j atoms */
 298             qq10             = _mm_mul_pd(iq1,jq0);
 299
 300             /* COULOMB ELECTROSTATICS */
 301             velec            = _mm_mul_pd(qq10,rinv10);
 302             felec            = _mm_mul_pd(velec,rinvsq10);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velecsum         = _mm_add_pd(velecsum,velec);
 306
 307             fscal            = felec;
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm_mul_pd(fscal,dx10);
 311             ty               = _mm_mul_pd(fscal,dy10);
 312             tz               = _mm_mul_pd(fscal,dz10);
 313
 314             /* Update vectorial force */
 315             fix1             = _mm_add_pd(fix1,tx);
 316             fiy1             = _mm_add_pd(fiy1,ty);
 317             fiz1             = _mm_add_pd(fiz1,tz);
 318
 319             fjx0             = _mm_add_pd(fjx0,tx);
 320             fjy0             = _mm_add_pd(fjy0,ty);
 321             fjz0             = _mm_add_pd(fjz0,tz);
 322
 323             /**************************
 324              * CALCULATE INTERACTIONS *
 325              **************************/
 326
 327             /* Compute parameters for interactions between i and j atoms */
 328             qq20             = _mm_mul_pd(iq2,jq0);
 329
 330             /* COULOMB ELECTROSTATICS */
 331             velec            = _mm_mul_pd(qq20,rinv20);
 332             felec            = _mm_mul_pd(velec,rinvsq20);
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velecsum         = _mm_add_pd(velecsum,velec);
 336
 337             fscal            = felec;
 338
 339             /* Calculate temporary vectorial force */
 340             tx               = _mm_mul_pd(fscal,dx20);
 341             ty               = _mm_mul_pd(fscal,dy20);
 342             tz               = _mm_mul_pd(fscal,dz20);
 343
 344             /* Update vectorial force */
 345             fix2             = _mm_add_pd(fix2,tx);
 346             fiy2             = _mm_add_pd(fiy2,ty);
 347             fiz2             = _mm_add_pd(fiz2,tz);
 348
 349             fjx0             = _mm_add_pd(fjx0,tx);
 350             fjy0             = _mm_add_pd(fjy0,ty);
 351             fjz0             = _mm_add_pd(fjz0,tz);
 352
 353             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 354
 355             /* Inner loop uses 122 flops */
 356         }
 357
 358         if(jidx<j_index_end)
 359         {
 360
 361             jnrA             = jjnr[jidx];
 362             j_coord_offsetA  = DIM*jnrA;
 363
 364             /* load j atom coordinates */
 365             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 366                                               &jx0,&jy0,&jz0);
 367
 368             /* Calculate displacement vector */
 369             dx00             = _mm_sub_pd(ix0,jx0);
 370             dy00             = _mm_sub_pd(iy0,jy0);
 371             dz00             = _mm_sub_pd(iz0,jz0);
 372             dx10             = _mm_sub_pd(ix1,jx0);
 373             dy10             = _mm_sub_pd(iy1,jy0);
 374             dz10             = _mm_sub_pd(iz1,jz0);
 375             dx20             = _mm_sub_pd(ix2,jx0);
 376             dy20             = _mm_sub_pd(iy2,jy0);
 377             dz20             = _mm_sub_pd(iz2,jz0);
 378
 379             /* Calculate squared distance and things based on it */
 380             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 381             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 382             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 383
 384             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 385             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 386             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 387
 388             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 389             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 390             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 391
 392             /* Load parameters for j particles */
 393             jq0              = _mm_load_sd(charge+jnrA+0);
 394             vdwjidx0A        = 2*vdwtype[jnrA+0];
 395
 396             fjx0             = _mm_setzero_pd();
 397             fjy0             = _mm_setzero_pd();
 398             fjz0             = _mm_setzero_pd();
 399
 400             /**************************
 401              * CALCULATE INTERACTIONS *
 402              **************************/
 403
 404             r00              = _mm_mul_pd(rsq00,rinv00);
 405
 406             /* Compute parameters for interactions between i and j atoms */
 407             qq00             = _mm_mul_pd(iq0,jq0);
 408             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 409
 410             /* Calculate table index by multiplying r with table scale and truncate to integer */
 411             rt               = _mm_mul_pd(r00,vftabscale);
 412             vfitab           = _mm_cvttpd_epi32(rt);
 413             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 414             vfitab           = _mm_slli_epi32(vfitab,3);
 415
 416             /* COULOMB ELECTROSTATICS */
 417             velec            = _mm_mul_pd(qq00,rinv00);
 418             felec            = _mm_mul_pd(velec,rinvsq00);
 419
 420             /* CUBIC SPLINE TABLE DISPERSION */
 421             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 422             F                = _mm_setzero_pd();
 423             GMX_MM_TRANSPOSE2_PD(Y,F);
 424             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 425             H                = _mm_setzero_pd();
 426             GMX_MM_TRANSPOSE2_PD(G,H);
 427             Heps             = _mm_mul_pd(vfeps,H);
 428             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 429             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 430             vvdw6            = _mm_mul_pd(c6_00,VV);
 431             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 432             fvdw6            = _mm_mul_pd(c6_00,FF);
 433
 434             /* CUBIC SPLINE TABLE REPULSION */
 435             vfitab           = _mm_add_epi32(vfitab,ifour);
 436             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 437             F                = _mm_setzero_pd();
 438             GMX_MM_TRANSPOSE2_PD(Y,F);
 439             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 440             H                = _mm_setzero_pd();
 441             GMX_MM_TRANSPOSE2_PD(G,H);
 442             Heps             = _mm_mul_pd(vfeps,H);
 443             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 444             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 445             vvdw12           = _mm_mul_pd(c12_00,VV);
 446             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 447             fvdw12           = _mm_mul_pd(c12_00,FF);
 448             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 449             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 450
 451             /* Update potential sum for this i atom from the interaction with this j atom. */
 452             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 453             velecsum         = _mm_add_pd(velecsum,velec);
 454             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 455             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 456
 457             fscal            = _mm_add_pd(felec,fvdw);
 458
 459             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 460
 461             /* Calculate temporary vectorial force */
 462             tx               = _mm_mul_pd(fscal,dx00);
 463             ty               = _mm_mul_pd(fscal,dy00);
 464             tz               = _mm_mul_pd(fscal,dz00);
 465
 466             /* Update vectorial force */
 467             fix0             = _mm_add_pd(fix0,tx);
 468             fiy0             = _mm_add_pd(fiy0,ty);
 469             fiz0             = _mm_add_pd(fiz0,tz);
 470
 471             fjx0             = _mm_add_pd(fjx0,tx);
 472             fjy0             = _mm_add_pd(fjy0,ty);
 473             fjz0             = _mm_add_pd(fjz0,tz);
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             /* Compute parameters for interactions between i and j atoms */
 480             qq10             = _mm_mul_pd(iq1,jq0);
 481
 482             /* COULOMB ELECTROSTATICS */
 483             velec            = _mm_mul_pd(qq10,rinv10);
 484             felec            = _mm_mul_pd(velec,rinvsq10);
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 488             velecsum         = _mm_add_pd(velecsum,velec);
 489
 490             fscal            = felec;
 491
 492             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 493
 494             /* Calculate temporary vectorial force */
 495             tx               = _mm_mul_pd(fscal,dx10);
 496             ty               = _mm_mul_pd(fscal,dy10);
 497             tz               = _mm_mul_pd(fscal,dz10);
 498
 499             /* Update vectorial force */
 500             fix1             = _mm_add_pd(fix1,tx);
 501             fiy1             = _mm_add_pd(fiy1,ty);
 502             fiz1             = _mm_add_pd(fiz1,tz);
 503
 504             fjx0             = _mm_add_pd(fjx0,tx);
 505             fjy0             = _mm_add_pd(fjy0,ty);
 506             fjz0             = _mm_add_pd(fjz0,tz);
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             /* Compute parameters for interactions between i and j atoms */
 513             qq20             = _mm_mul_pd(iq2,jq0);
 514
 515             /* COULOMB ELECTROSTATICS */
 516             velec            = _mm_mul_pd(qq20,rinv20);
 517             felec            = _mm_mul_pd(velec,rinvsq20);
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 521             velecsum         = _mm_add_pd(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 526
 527             /* Calculate temporary vectorial force */
 528             tx               = _mm_mul_pd(fscal,dx20);
 529             ty               = _mm_mul_pd(fscal,dy20);
 530             tz               = _mm_mul_pd(fscal,dz20);
 531
 532             /* Update vectorial force */
 533             fix2             = _mm_add_pd(fix2,tx);
 534             fiy2             = _mm_add_pd(fiy2,ty);
 535             fiz2             = _mm_add_pd(fiz2,tz);
 536
 537             fjx0             = _mm_add_pd(fjx0,tx);
 538             fjy0             = _mm_add_pd(fjy0,ty);
 539             fjz0             = _mm_add_pd(fjz0,tz);
 540
 541             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 542
 543             /* Inner loop uses 122 flops */
 544         }
 545
 546         /* End of innermost loop */
 547
 548         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 549                                               f+i_coord_offset,fshift+i_shift_offset);
 550
 551         ggid                        = gid[iidx];
 552         /* Update potential energies */
 553         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 554         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 555
 556         /* Increment number of inner iterations */
 557         inneriter                  += j_index_end - j_index_start;
 558
 559         /* Outer loop uses 20 flops */
 560     }
 561
 562     /* Increment number of outer iterations */
 563     outeriter        += nri;
 564
 565     /* Update outer/inner flops */
 566
 567     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*122);
 568 }
 569 /*
 570  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse2_double
 571  * Electrostatics interaction: Coulomb
 572  * VdW interaction:            CubicSplineTable
 573  * Geometry:                   Water3-Particle
 574  * Calculate force/pot:        Force
 575  */
 576 void
 577 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse2_double
 578                     (t_nblist                    * gmx_restrict       nlist,
 579                      rvec                        * gmx_restrict          xx,
 580                      rvec                        * gmx_restrict          ff,
 581                      t_forcerec                  * gmx_restrict          fr,
 582                      t_mdatoms                   * gmx_restrict     mdatoms,
 583                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 584                      t_nrnb                      * gmx_restrict        nrnb)
 585 {
 586     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 587      * just 0 for non-waters.
 588      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 589      * jnr indices corresponding to data put in the four positions in the SIMD register.
 590      */
 591     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 592     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 593     int              jnrA,jnrB;
 594     int              j_coord_offsetA,j_coord_offsetB;
 595     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 596     real             rcutoff_scalar;
 597     real             *shiftvec,*fshift,*x,*f;
 598     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 599     int              vdwioffset0;
 600     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 601     int              vdwioffset1;
 602     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 603     int              vdwioffset2;
 604     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 605     int              vdwjidx0A,vdwjidx0B;
 606     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 607     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 608     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 609     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 610     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 611     real             *charge;
 612     int              nvdwtype;
 613     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 614     int              *vdwtype;
 615     real             *vdwparam;
 616     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 617     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 618     __m128i          vfitab;
 619     __m128i          ifour       = _mm_set1_epi32(4);
 620     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 621     real             *vftab;
 622     __m128d          dummy_mask,cutoff_mask;
 623     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 624     __m128d          one     = _mm_set1_pd(1.0);
 625     __m128d          two     = _mm_set1_pd(2.0);
 626     x                = xx[0];
 627     f                = ff[0];
 628
 629     nri              = nlist->nri;
 630     iinr             = nlist->iinr;
 631     jindex           = nlist->jindex;
 632     jjnr             = nlist->jjnr;
 633     shiftidx         = nlist->shift;
 634     gid              = nlist->gid;
 635     shiftvec         = fr->shift_vec[0];
 636     fshift           = fr->fshift[0];
 637     facel            = _mm_set1_pd(fr->epsfac);
 638     charge           = mdatoms->chargeA;
 639     nvdwtype         = fr->ntype;
 640     vdwparam         = fr->nbfp;
 641     vdwtype          = mdatoms->typeA;
 642
 643     vftab            = kernel_data->table_vdw->data;
 644     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 645
 646     /* Setup water-specific parameters */
 647     inr              = nlist->iinr[0];
 648     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 649     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 650     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 651     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 652
 653     /* Avoid stupid compiler warnings */
 654     jnrA = jnrB = 0;
 655     j_coord_offsetA = 0;
 656     j_coord_offsetB = 0;
 657
 658     outeriter        = 0;
 659     inneriter        = 0;
 660
 661     /* Start outer loop over neighborlists */
 662     for(iidx=0; iidx<nri; iidx++)
 663     {
 664         /* Load shift vector for this list */
 665         i_shift_offset   = DIM*shiftidx[iidx];
 666
 667         /* Load limits for loop over neighbors */
 668         j_index_start    = jindex[iidx];
 669         j_index_end      = jindex[iidx+1];
 670
 671         /* Get outer coordinate index */
 672         inr              = iinr[iidx];
 673         i_coord_offset   = DIM*inr;
 674
 675         /* Load i particle coords and add shift vector */
 676         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 677                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 678
 679         fix0             = _mm_setzero_pd();
 680         fiy0             = _mm_setzero_pd();
 681         fiz0             = _mm_setzero_pd();
 682         fix1             = _mm_setzero_pd();
 683         fiy1             = _mm_setzero_pd();
 684         fiz1             = _mm_setzero_pd();
 685         fix2             = _mm_setzero_pd();
 686         fiy2             = _mm_setzero_pd();
 687         fiz2             = _mm_setzero_pd();
 688
 689         /* Start inner kernel loop */
 690         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 691         {
 692
 693             /* Get j neighbor index, and coordinate index */
 694             jnrA             = jjnr[jidx];
 695             jnrB             = jjnr[jidx+1];
 696             j_coord_offsetA  = DIM*jnrA;
 697             j_coord_offsetB  = DIM*jnrB;
 698
 699             /* load j atom coordinates */
 700             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 701                                               &jx0,&jy0,&jz0);
 702
 703             /* Calculate displacement vector */
 704             dx00             = _mm_sub_pd(ix0,jx0);
 705             dy00             = _mm_sub_pd(iy0,jy0);
 706             dz00             = _mm_sub_pd(iz0,jz0);
 707             dx10             = _mm_sub_pd(ix1,jx0);
 708             dy10             = _mm_sub_pd(iy1,jy0);
 709             dz10             = _mm_sub_pd(iz1,jz0);
 710             dx20             = _mm_sub_pd(ix2,jx0);
 711             dy20             = _mm_sub_pd(iy2,jy0);
 712             dz20             = _mm_sub_pd(iz2,jz0);
 713
 714             /* Calculate squared distance and things based on it */
 715             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 716             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 717             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 718
 719             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 720             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 721             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 722
 723             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 724             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 725             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 726
 727             /* Load parameters for j particles */
 728             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 729             vdwjidx0A        = 2*vdwtype[jnrA+0];
 730             vdwjidx0B        = 2*vdwtype[jnrB+0];
 731
 732             fjx0             = _mm_setzero_pd();
 733             fjy0             = _mm_setzero_pd();
 734             fjz0             = _mm_setzero_pd();
 735
 736             /**************************
 737              * CALCULATE INTERACTIONS *
 738              **************************/
 739
 740             r00              = _mm_mul_pd(rsq00,rinv00);
 741
 742             /* Compute parameters for interactions between i and j atoms */
 743             qq00             = _mm_mul_pd(iq0,jq0);
 744             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 745                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 746
 747             /* Calculate table index by multiplying r with table scale and truncate to integer */
 748             rt               = _mm_mul_pd(r00,vftabscale);
 749             vfitab           = _mm_cvttpd_epi32(rt);
 750             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 751             vfitab           = _mm_slli_epi32(vfitab,3);
 752
 753             /* COULOMB ELECTROSTATICS */
 754             velec            = _mm_mul_pd(qq00,rinv00);
 755             felec            = _mm_mul_pd(velec,rinvsq00);
 756
 757             /* CUBIC SPLINE TABLE DISPERSION */
 758             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 759             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 760             GMX_MM_TRANSPOSE2_PD(Y,F);
 761             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 762             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 763             GMX_MM_TRANSPOSE2_PD(G,H);
 764             Heps             = _mm_mul_pd(vfeps,H);
 765             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 766             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 767             fvdw6            = _mm_mul_pd(c6_00,FF);
 768
 769             /* CUBIC SPLINE TABLE REPULSION */
 770             vfitab           = _mm_add_epi32(vfitab,ifour);
 771             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 772             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 773             GMX_MM_TRANSPOSE2_PD(Y,F);
 774             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 775             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 776             GMX_MM_TRANSPOSE2_PD(G,H);
 777             Heps             = _mm_mul_pd(vfeps,H);
 778             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 779             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 780             fvdw12           = _mm_mul_pd(c12_00,FF);
 781             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 782
 783             fscal            = _mm_add_pd(felec,fvdw);
 784
 785             /* Calculate temporary vectorial force */
 786             tx               = _mm_mul_pd(fscal,dx00);
 787             ty               = _mm_mul_pd(fscal,dy00);
 788             tz               = _mm_mul_pd(fscal,dz00);
 789
 790             /* Update vectorial force */
 791             fix0             = _mm_add_pd(fix0,tx);
 792             fiy0             = _mm_add_pd(fiy0,ty);
 793             fiz0             = _mm_add_pd(fiz0,tz);
 794
 795             fjx0             = _mm_add_pd(fjx0,tx);
 796             fjy0             = _mm_add_pd(fjy0,ty);
 797             fjz0             = _mm_add_pd(fjz0,tz);
 798
 799             /**************************
 800              * CALCULATE INTERACTIONS *
 801              **************************/
 802
 803             /* Compute parameters for interactions between i and j atoms */
 804             qq10             = _mm_mul_pd(iq1,jq0);
 805
 806             /* COULOMB ELECTROSTATICS */
 807             velec            = _mm_mul_pd(qq10,rinv10);
 808             felec            = _mm_mul_pd(velec,rinvsq10);
 809
 810             fscal            = felec;
 811
 812             /* Calculate temporary vectorial force */
 813             tx               = _mm_mul_pd(fscal,dx10);
 814             ty               = _mm_mul_pd(fscal,dy10);
 815             tz               = _mm_mul_pd(fscal,dz10);
 816
 817             /* Update vectorial force */
 818             fix1             = _mm_add_pd(fix1,tx);
 819             fiy1             = _mm_add_pd(fiy1,ty);
 820             fiz1             = _mm_add_pd(fiz1,tz);
 821
 822             fjx0             = _mm_add_pd(fjx0,tx);
 823             fjy0             = _mm_add_pd(fjy0,ty);
 824             fjz0             = _mm_add_pd(fjz0,tz);
 825
 826             /**************************
 827              * CALCULATE INTERACTIONS *
 828              **************************/
 829
 830             /* Compute parameters for interactions between i and j atoms */
 831             qq20             = _mm_mul_pd(iq2,jq0);
 832
 833             /* COULOMB ELECTROSTATICS */
 834             velec            = _mm_mul_pd(qq20,rinv20);
 835             felec            = _mm_mul_pd(velec,rinvsq20);
 836
 837             fscal            = felec;
 838
 839             /* Calculate temporary vectorial force */
 840             tx               = _mm_mul_pd(fscal,dx20);
 841             ty               = _mm_mul_pd(fscal,dy20);
 842             tz               = _mm_mul_pd(fscal,dz20);
 843
 844             /* Update vectorial force */
 845             fix2             = _mm_add_pd(fix2,tx);
 846             fiy2             = _mm_add_pd(fiy2,ty);
 847             fiz2             = _mm_add_pd(fiz2,tz);
 848
 849             fjx0             = _mm_add_pd(fjx0,tx);
 850             fjy0             = _mm_add_pd(fjy0,ty);
 851             fjz0             = _mm_add_pd(fjz0,tz);
 852
 853             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 854
 855             /* Inner loop uses 111 flops */
 856         }
 857
 858         if(jidx<j_index_end)
 859         {
 860
 861             jnrA             = jjnr[jidx];
 862             j_coord_offsetA  = DIM*jnrA;
 863
 864             /* load j atom coordinates */
 865             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 866                                               &jx0,&jy0,&jz0);
 867
 868             /* Calculate displacement vector */
 869             dx00             = _mm_sub_pd(ix0,jx0);
 870             dy00             = _mm_sub_pd(iy0,jy0);
 871             dz00             = _mm_sub_pd(iz0,jz0);
 872             dx10             = _mm_sub_pd(ix1,jx0);
 873             dy10             = _mm_sub_pd(iy1,jy0);
 874             dz10             = _mm_sub_pd(iz1,jz0);
 875             dx20             = _mm_sub_pd(ix2,jx0);
 876             dy20             = _mm_sub_pd(iy2,jy0);
 877             dz20             = _mm_sub_pd(iz2,jz0);
 878
 879             /* Calculate squared distance and things based on it */
 880             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 881             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 882             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 883
 884             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 885             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 886             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 887
 888             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 889             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 890             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 891
 892             /* Load parameters for j particles */
 893             jq0              = _mm_load_sd(charge+jnrA+0);
 894             vdwjidx0A        = 2*vdwtype[jnrA+0];
 895
 896             fjx0             = _mm_setzero_pd();
 897             fjy0             = _mm_setzero_pd();
 898             fjz0             = _mm_setzero_pd();
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             r00              = _mm_mul_pd(rsq00,rinv00);
 905
 906             /* Compute parameters for interactions between i and j atoms */
 907             qq00             = _mm_mul_pd(iq0,jq0);
 908             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 909
 910             /* Calculate table index by multiplying r with table scale and truncate to integer */
 911             rt               = _mm_mul_pd(r00,vftabscale);
 912             vfitab           = _mm_cvttpd_epi32(rt);
 913             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 914             vfitab           = _mm_slli_epi32(vfitab,3);
 915
 916             /* COULOMB ELECTROSTATICS */
 917             velec            = _mm_mul_pd(qq00,rinv00);
 918             felec            = _mm_mul_pd(velec,rinvsq00);
 919
 920             /* CUBIC SPLINE TABLE DISPERSION */
 921             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 922             F                = _mm_setzero_pd();
 923             GMX_MM_TRANSPOSE2_PD(Y,F);
 924             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 925             H                = _mm_setzero_pd();
 926             GMX_MM_TRANSPOSE2_PD(G,H);
 927             Heps             = _mm_mul_pd(vfeps,H);
 928             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 929             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 930             fvdw6            = _mm_mul_pd(c6_00,FF);
 931
 932             /* CUBIC SPLINE TABLE REPULSION */
 933             vfitab           = _mm_add_epi32(vfitab,ifour);
 934             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 935             F                = _mm_setzero_pd();
 936             GMX_MM_TRANSPOSE2_PD(Y,F);
 937             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 938             H                = _mm_setzero_pd();
 939             GMX_MM_TRANSPOSE2_PD(G,H);
 940             Heps             = _mm_mul_pd(vfeps,H);
 941             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 942             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 943             fvdw12           = _mm_mul_pd(c12_00,FF);
 944             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 945
 946             fscal            = _mm_add_pd(felec,fvdw);
 947
 948             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 949
 950             /* Calculate temporary vectorial force */
 951             tx               = _mm_mul_pd(fscal,dx00);
 952             ty               = _mm_mul_pd(fscal,dy00);
 953             tz               = _mm_mul_pd(fscal,dz00);
 954
 955             /* Update vectorial force */
 956             fix0             = _mm_add_pd(fix0,tx);
 957             fiy0             = _mm_add_pd(fiy0,ty);
 958             fiz0             = _mm_add_pd(fiz0,tz);
 959
 960             fjx0             = _mm_add_pd(fjx0,tx);
 961             fjy0             = _mm_add_pd(fjy0,ty);
 962             fjz0             = _mm_add_pd(fjz0,tz);
 963
 964             /**************************
 965              * CALCULATE INTERACTIONS *
 966              **************************/
 967
 968             /* Compute parameters for interactions between i and j atoms */
 969             qq10             = _mm_mul_pd(iq1,jq0);
 970
 971             /* COULOMB ELECTROSTATICS */
 972             velec            = _mm_mul_pd(qq10,rinv10);
 973             felec            = _mm_mul_pd(velec,rinvsq10);
 974
 975             fscal            = felec;
 976
 977             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 978
 979             /* Calculate temporary vectorial force */
 980             tx               = _mm_mul_pd(fscal,dx10);
 981             ty               = _mm_mul_pd(fscal,dy10);
 982             tz               = _mm_mul_pd(fscal,dz10);
 983
 984             /* Update vectorial force */
 985             fix1             = _mm_add_pd(fix1,tx);
 986             fiy1             = _mm_add_pd(fiy1,ty);
 987             fiz1             = _mm_add_pd(fiz1,tz);
 988
 989             fjx0             = _mm_add_pd(fjx0,tx);
 990             fjy0             = _mm_add_pd(fjy0,ty);
 991             fjz0             = _mm_add_pd(fjz0,tz);
 992
 993             /**************************
 994              * CALCULATE INTERACTIONS *
 995              **************************/
 996
 997             /* Compute parameters for interactions between i and j atoms */
 998             qq20             = _mm_mul_pd(iq2,jq0);
 999
1000             /* COULOMB ELECTROSTATICS */
1001             velec            = _mm_mul_pd(qq20,rinv20);
1002             felec            = _mm_mul_pd(velec,rinvsq20);
1003
1004             fscal            = felec;
1005
1006             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm_mul_pd(fscal,dx20);
1010             ty               = _mm_mul_pd(fscal,dy20);
1011             tz               = _mm_mul_pd(fscal,dz20);
1012
1013             /* Update vectorial force */
1014             fix2             = _mm_add_pd(fix2,tx);
1015             fiy2             = _mm_add_pd(fiy2,ty);
1016             fiz2             = _mm_add_pd(fiz2,tz);
1017
1018             fjx0             = _mm_add_pd(fjx0,tx);
1019             fjy0             = _mm_add_pd(fjy0,ty);
1020             fjz0             = _mm_add_pd(fjz0,tz);
1021
1022             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1023
1024             /* Inner loop uses 111 flops */
1025         }
1026
1027         /* End of innermost loop */
1028
1029         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1030                                               f+i_coord_offset,fshift+i_shift_offset);
1031
1032         /* Increment number of inner iterations */
1033         inneriter                  += j_index_end - j_index_start;
1034
1035         /* Outer loop uses 18 flops */
1036     }
1037
1038     /* Increment number of outer iterations */
1039     outeriter        += nri;
1040
1041     /* Update outer/inner flops */
1042
1043     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*111);
1044 }