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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_double.h"
  50 #include "kernelutil_x86_sse2_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse2_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse2_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              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          vfitab;
 102     __m128i          ifour       = _mm_set1_epi32(4);
 103     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 104     real             *vftab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_vdw->data;
 127     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 132     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 161
 162         fix0             = _mm_setzero_pd();
 163         fiy0             = _mm_setzero_pd();
 164         fiz0             = _mm_setzero_pd();
 165         fix1             = _mm_setzero_pd();
 166         fiy1             = _mm_setzero_pd();
 167         fiz1             = _mm_setzero_pd();
 168         fix2             = _mm_setzero_pd();
 169         fiy2             = _mm_setzero_pd();
 170         fiz2             = _mm_setzero_pd();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_pd();
 174         vvdwsum          = _mm_setzero_pd();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_pd(ix0,jx0);
 192             dy00             = _mm_sub_pd(iy0,jy0);
 193             dz00             = _mm_sub_pd(iz0,jz0);
 194             dx10             = _mm_sub_pd(ix1,jx0);
 195             dy10             = _mm_sub_pd(iy1,jy0);
 196             dz10             = _mm_sub_pd(iz1,jz0);
 197             dx20             = _mm_sub_pd(ix2,jx0);
 198             dy20             = _mm_sub_pd(iy2,jy0);
 199             dz20             = _mm_sub_pd(iz2,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 203             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 204             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 205
 206             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 207             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 208             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 209
 210             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 211             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 212             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218
 219             fjx0             = _mm_setzero_pd();
 220             fjy0             = _mm_setzero_pd();
 221             fjz0             = _mm_setzero_pd();
 222
 223             /**************************
 224              * CALCULATE INTERACTIONS *
 225              **************************/
 226
 227             r00              = _mm_mul_pd(rsq00,rinv00);
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq00             = _mm_mul_pd(iq0,jq0);
 231             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 232                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 233
 234             /* Calculate table index by multiplying r with table scale and truncate to integer */
 235             rt               = _mm_mul_pd(r00,vftabscale);
 236             vfitab           = _mm_cvttpd_epi32(rt);
 237             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 238             vfitab           = _mm_slli_epi32(vfitab,3);
 239
 240             /* COULOMB ELECTROSTATICS */
 241             velec            = _mm_mul_pd(qq00,rinv00);
 242             felec            = _mm_mul_pd(velec,rinvsq00);
 243
 244             /* CUBIC SPLINE TABLE DISPERSION */
 245             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 246             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 247             GMX_MM_TRANSPOSE2_PD(Y,F);
 248             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 249             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 250             GMX_MM_TRANSPOSE2_PD(G,H);
 251             Heps             = _mm_mul_pd(vfeps,H);
 252             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 253             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 254             vvdw6            = _mm_mul_pd(c6_00,VV);
 255             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 256             fvdw6            = _mm_mul_pd(c6_00,FF);
 257
 258             /* CUBIC SPLINE TABLE REPULSION */
 259             vfitab           = _mm_add_epi32(vfitab,ifour);
 260             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 261             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 262             GMX_MM_TRANSPOSE2_PD(Y,F);
 263             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 264             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 265             GMX_MM_TRANSPOSE2_PD(G,H);
 266             Heps             = _mm_mul_pd(vfeps,H);
 267             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 268             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 269             vvdw12           = _mm_mul_pd(c12_00,VV);
 270             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 271             fvdw12           = _mm_mul_pd(c12_00,FF);
 272             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 273             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velecsum         = _mm_add_pd(velecsum,velec);
 277             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 278
 279             fscal            = _mm_add_pd(felec,fvdw);
 280
 281             /* Calculate temporary vectorial force */
 282             tx               = _mm_mul_pd(fscal,dx00);
 283             ty               = _mm_mul_pd(fscal,dy00);
 284             tz               = _mm_mul_pd(fscal,dz00);
 285
 286             /* Update vectorial force */
 287             fix0             = _mm_add_pd(fix0,tx);
 288             fiy0             = _mm_add_pd(fiy0,ty);
 289             fiz0             = _mm_add_pd(fiz0,tz);
 290
 291             fjx0             = _mm_add_pd(fjx0,tx);
 292             fjy0             = _mm_add_pd(fjy0,ty);
 293             fjz0             = _mm_add_pd(fjz0,tz);
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq10             = _mm_mul_pd(iq1,jq0);
 301
 302             /* COULOMB ELECTROSTATICS */
 303             velec            = _mm_mul_pd(qq10,rinv10);
 304             felec            = _mm_mul_pd(velec,rinvsq10);
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velecsum         = _mm_add_pd(velecsum,velec);
 308
 309             fscal            = felec;
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm_mul_pd(fscal,dx10);
 313             ty               = _mm_mul_pd(fscal,dy10);
 314             tz               = _mm_mul_pd(fscal,dz10);
 315
 316             /* Update vectorial force */
 317             fix1             = _mm_add_pd(fix1,tx);
 318             fiy1             = _mm_add_pd(fiy1,ty);
 319             fiz1             = _mm_add_pd(fiz1,tz);
 320
 321             fjx0             = _mm_add_pd(fjx0,tx);
 322             fjy0             = _mm_add_pd(fjy0,ty);
 323             fjz0             = _mm_add_pd(fjz0,tz);
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             /* Compute parameters for interactions between i and j atoms */
 330             qq20             = _mm_mul_pd(iq2,jq0);
 331
 332             /* COULOMB ELECTROSTATICS */
 333             velec            = _mm_mul_pd(qq20,rinv20);
 334             felec            = _mm_mul_pd(velec,rinvsq20);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velecsum         = _mm_add_pd(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             /* Calculate temporary vectorial force */
 342             tx               = _mm_mul_pd(fscal,dx20);
 343             ty               = _mm_mul_pd(fscal,dy20);
 344             tz               = _mm_mul_pd(fscal,dz20);
 345
 346             /* Update vectorial force */
 347             fix2             = _mm_add_pd(fix2,tx);
 348             fiy2             = _mm_add_pd(fiy2,ty);
 349             fiz2             = _mm_add_pd(fiz2,tz);
 350
 351             fjx0             = _mm_add_pd(fjx0,tx);
 352             fjy0             = _mm_add_pd(fjy0,ty);
 353             fjz0             = _mm_add_pd(fjz0,tz);
 354
 355             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 356
 357             /* Inner loop uses 122 flops */
 358         }
 359
 360         if(jidx<j_index_end)
 361         {
 362
 363             jnrA             = jjnr[jidx];
 364             j_coord_offsetA  = DIM*jnrA;
 365
 366             /* load j atom coordinates */
 367             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 368                                               &jx0,&jy0,&jz0);
 369
 370             /* Calculate displacement vector */
 371             dx00             = _mm_sub_pd(ix0,jx0);
 372             dy00             = _mm_sub_pd(iy0,jy0);
 373             dz00             = _mm_sub_pd(iz0,jz0);
 374             dx10             = _mm_sub_pd(ix1,jx0);
 375             dy10             = _mm_sub_pd(iy1,jy0);
 376             dz10             = _mm_sub_pd(iz1,jz0);
 377             dx20             = _mm_sub_pd(ix2,jx0);
 378             dy20             = _mm_sub_pd(iy2,jy0);
 379             dz20             = _mm_sub_pd(iz2,jz0);
 380
 381             /* Calculate squared distance and things based on it */
 382             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 383             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 384             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 385
 386             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 387             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 388             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 389
 390             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 391             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 392             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 393
 394             /* Load parameters for j particles */
 395             jq0              = _mm_load_sd(charge+jnrA+0);
 396             vdwjidx0A        = 2*vdwtype[jnrA+0];
 397
 398             fjx0             = _mm_setzero_pd();
 399             fjy0             = _mm_setzero_pd();
 400             fjz0             = _mm_setzero_pd();
 401
 402             /**************************
 403              * CALCULATE INTERACTIONS *
 404              **************************/
 405
 406             r00              = _mm_mul_pd(rsq00,rinv00);
 407
 408             /* Compute parameters for interactions between i and j atoms */
 409             qq00             = _mm_mul_pd(iq0,jq0);
 410             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 411
 412             /* Calculate table index by multiplying r with table scale and truncate to integer */
 413             rt               = _mm_mul_pd(r00,vftabscale);
 414             vfitab           = _mm_cvttpd_epi32(rt);
 415             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 416             vfitab           = _mm_slli_epi32(vfitab,3);
 417
 418             /* COULOMB ELECTROSTATICS */
 419             velec            = _mm_mul_pd(qq00,rinv00);
 420             felec            = _mm_mul_pd(velec,rinvsq00);
 421
 422             /* CUBIC SPLINE TABLE DISPERSION */
 423             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 424             F                = _mm_setzero_pd();
 425             GMX_MM_TRANSPOSE2_PD(Y,F);
 426             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 427             H                = _mm_setzero_pd();
 428             GMX_MM_TRANSPOSE2_PD(G,H);
 429             Heps             = _mm_mul_pd(vfeps,H);
 430             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 431             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 432             vvdw6            = _mm_mul_pd(c6_00,VV);
 433             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 434             fvdw6            = _mm_mul_pd(c6_00,FF);
 435
 436             /* CUBIC SPLINE TABLE REPULSION */
 437             vfitab           = _mm_add_epi32(vfitab,ifour);
 438             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 439             F                = _mm_setzero_pd();
 440             GMX_MM_TRANSPOSE2_PD(Y,F);
 441             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 442             H                = _mm_setzero_pd();
 443             GMX_MM_TRANSPOSE2_PD(G,H);
 444             Heps             = _mm_mul_pd(vfeps,H);
 445             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 446             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 447             vvdw12           = _mm_mul_pd(c12_00,VV);
 448             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 449             fvdw12           = _mm_mul_pd(c12_00,FF);
 450             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 451             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 452
 453             /* Update potential sum for this i atom from the interaction with this j atom. */
 454             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 455             velecsum         = _mm_add_pd(velecsum,velec);
 456             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 457             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 458
 459             fscal            = _mm_add_pd(felec,fvdw);
 460
 461             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 462
 463             /* Calculate temporary vectorial force */
 464             tx               = _mm_mul_pd(fscal,dx00);
 465             ty               = _mm_mul_pd(fscal,dy00);
 466             tz               = _mm_mul_pd(fscal,dz00);
 467
 468             /* Update vectorial force */
 469             fix0             = _mm_add_pd(fix0,tx);
 470             fiy0             = _mm_add_pd(fiy0,ty);
 471             fiz0             = _mm_add_pd(fiz0,tz);
 472
 473             fjx0             = _mm_add_pd(fjx0,tx);
 474             fjy0             = _mm_add_pd(fjy0,ty);
 475             fjz0             = _mm_add_pd(fjz0,tz);
 476
 477             /**************************
 478              * CALCULATE INTERACTIONS *
 479              **************************/
 480
 481             /* Compute parameters for interactions between i and j atoms */
 482             qq10             = _mm_mul_pd(iq1,jq0);
 483
 484             /* COULOMB ELECTROSTATICS */
 485             velec            = _mm_mul_pd(qq10,rinv10);
 486             felec            = _mm_mul_pd(velec,rinvsq10);
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 490             velecsum         = _mm_add_pd(velecsum,velec);
 491
 492             fscal            = felec;
 493
 494             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 495
 496             /* Calculate temporary vectorial force */
 497             tx               = _mm_mul_pd(fscal,dx10);
 498             ty               = _mm_mul_pd(fscal,dy10);
 499             tz               = _mm_mul_pd(fscal,dz10);
 500
 501             /* Update vectorial force */
 502             fix1             = _mm_add_pd(fix1,tx);
 503             fiy1             = _mm_add_pd(fiy1,ty);
 504             fiz1             = _mm_add_pd(fiz1,tz);
 505
 506             fjx0             = _mm_add_pd(fjx0,tx);
 507             fjy0             = _mm_add_pd(fjy0,ty);
 508             fjz0             = _mm_add_pd(fjz0,tz);
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq20             = _mm_mul_pd(iq2,jq0);
 516
 517             /* COULOMB ELECTROSTATICS */
 518             velec            = _mm_mul_pd(qq20,rinv20);
 519             felec            = _mm_mul_pd(velec,rinvsq20);
 520
 521             /* Update potential sum for this i atom from the interaction with this j atom. */
 522             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 523             velecsum         = _mm_add_pd(velecsum,velec);
 524
 525             fscal            = felec;
 526
 527             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 528
 529             /* Calculate temporary vectorial force */
 530             tx               = _mm_mul_pd(fscal,dx20);
 531             ty               = _mm_mul_pd(fscal,dy20);
 532             tz               = _mm_mul_pd(fscal,dz20);
 533
 534             /* Update vectorial force */
 535             fix2             = _mm_add_pd(fix2,tx);
 536             fiy2             = _mm_add_pd(fiy2,ty);
 537             fiz2             = _mm_add_pd(fiz2,tz);
 538
 539             fjx0             = _mm_add_pd(fjx0,tx);
 540             fjy0             = _mm_add_pd(fjy0,ty);
 541             fjz0             = _mm_add_pd(fjz0,tz);
 542
 543             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 544
 545             /* Inner loop uses 122 flops */
 546         }
 547
 548         /* End of innermost loop */
 549
 550         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 551                                               f+i_coord_offset,fshift+i_shift_offset);
 552
 553         ggid                        = gid[iidx];
 554         /* Update potential energies */
 555         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 556         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 557
 558         /* Increment number of inner iterations */
 559         inneriter                  += j_index_end - j_index_start;
 560
 561         /* Outer loop uses 20 flops */
 562     }
 563
 564     /* Increment number of outer iterations */
 565     outeriter        += nri;
 566
 567     /* Update outer/inner flops */
 568
 569     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*122);
 570 }
 571 /*
 572  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse2_double
 573  * Electrostatics interaction: Coulomb
 574  * VdW interaction:            CubicSplineTable
 575  * Geometry:                   Water3-Particle
 576  * Calculate force/pot:        Force
 577  */
 578 void
 579 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse2_double
 580                     (t_nblist                    * gmx_restrict       nlist,
 581                      rvec                        * gmx_restrict          xx,
 582                      rvec                        * gmx_restrict          ff,
 583                      t_forcerec                  * gmx_restrict          fr,
 584                      t_mdatoms                   * gmx_restrict     mdatoms,
 585                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 586                      t_nrnb                      * gmx_restrict        nrnb)
 587 {
 588     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 589      * just 0 for non-waters.
 590      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 591      * jnr indices corresponding to data put in the four positions in the SIMD register.
 592      */
 593     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 594     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 595     int              jnrA,jnrB;
 596     int              j_coord_offsetA,j_coord_offsetB;
 597     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 598     real             rcutoff_scalar;
 599     real             *shiftvec,*fshift,*x,*f;
 600     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 601     int              vdwioffset0;
 602     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 603     int              vdwioffset1;
 604     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 605     int              vdwioffset2;
 606     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 607     int              vdwjidx0A,vdwjidx0B;
 608     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 609     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 610     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 611     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 612     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 613     real             *charge;
 614     int              nvdwtype;
 615     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 616     int              *vdwtype;
 617     real             *vdwparam;
 618     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 619     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 620     __m128i          vfitab;
 621     __m128i          ifour       = _mm_set1_epi32(4);
 622     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 623     real             *vftab;
 624     __m128d          dummy_mask,cutoff_mask;
 625     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 626     __m128d          one     = _mm_set1_pd(1.0);
 627     __m128d          two     = _mm_set1_pd(2.0);
 628     x                = xx[0];
 629     f                = ff[0];
 630
 631     nri              = nlist->nri;
 632     iinr             = nlist->iinr;
 633     jindex           = nlist->jindex;
 634     jjnr             = nlist->jjnr;
 635     shiftidx         = nlist->shift;
 636     gid              = nlist->gid;
 637     shiftvec         = fr->shift_vec[0];
 638     fshift           = fr->fshift[0];
 639     facel            = _mm_set1_pd(fr->epsfac);
 640     charge           = mdatoms->chargeA;
 641     nvdwtype         = fr->ntype;
 642     vdwparam         = fr->nbfp;
 643     vdwtype          = mdatoms->typeA;
 644
 645     vftab            = kernel_data->table_vdw->data;
 646     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 647
 648     /* Setup water-specific parameters */
 649     inr              = nlist->iinr[0];
 650     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 651     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 652     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 653     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 654
 655     /* Avoid stupid compiler warnings */
 656     jnrA = jnrB = 0;
 657     j_coord_offsetA = 0;
 658     j_coord_offsetB = 0;
 659
 660     outeriter        = 0;
 661     inneriter        = 0;
 662
 663     /* Start outer loop over neighborlists */
 664     for(iidx=0; iidx<nri; iidx++)
 665     {
 666         /* Load shift vector for this list */
 667         i_shift_offset   = DIM*shiftidx[iidx];
 668
 669         /* Load limits for loop over neighbors */
 670         j_index_start    = jindex[iidx];
 671         j_index_end      = jindex[iidx+1];
 672
 673         /* Get outer coordinate index */
 674         inr              = iinr[iidx];
 675         i_coord_offset   = DIM*inr;
 676
 677         /* Load i particle coords and add shift vector */
 678         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 679                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 680
 681         fix0             = _mm_setzero_pd();
 682         fiy0             = _mm_setzero_pd();
 683         fiz0             = _mm_setzero_pd();
 684         fix1             = _mm_setzero_pd();
 685         fiy1             = _mm_setzero_pd();
 686         fiz1             = _mm_setzero_pd();
 687         fix2             = _mm_setzero_pd();
 688         fiy2             = _mm_setzero_pd();
 689         fiz2             = _mm_setzero_pd();
 690
 691         /* Start inner kernel loop */
 692         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 693         {
 694
 695             /* Get j neighbor index, and coordinate index */
 696             jnrA             = jjnr[jidx];
 697             jnrB             = jjnr[jidx+1];
 698             j_coord_offsetA  = DIM*jnrA;
 699             j_coord_offsetB  = DIM*jnrB;
 700
 701             /* load j atom coordinates */
 702             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 703                                               &jx0,&jy0,&jz0);
 704
 705             /* Calculate displacement vector */
 706             dx00             = _mm_sub_pd(ix0,jx0);
 707             dy00             = _mm_sub_pd(iy0,jy0);
 708             dz00             = _mm_sub_pd(iz0,jz0);
 709             dx10             = _mm_sub_pd(ix1,jx0);
 710             dy10             = _mm_sub_pd(iy1,jy0);
 711             dz10             = _mm_sub_pd(iz1,jz0);
 712             dx20             = _mm_sub_pd(ix2,jx0);
 713             dy20             = _mm_sub_pd(iy2,jy0);
 714             dz20             = _mm_sub_pd(iz2,jz0);
 715
 716             /* Calculate squared distance and things based on it */
 717             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 718             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 719             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 720
 721             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 722             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 723             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 724
 725             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 726             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 727             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 728
 729             /* Load parameters for j particles */
 730             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 731             vdwjidx0A        = 2*vdwtype[jnrA+0];
 732             vdwjidx0B        = 2*vdwtype[jnrB+0];
 733
 734             fjx0             = _mm_setzero_pd();
 735             fjy0             = _mm_setzero_pd();
 736             fjz0             = _mm_setzero_pd();
 737
 738             /**************************
 739              * CALCULATE INTERACTIONS *
 740              **************************/
 741
 742             r00              = _mm_mul_pd(rsq00,rinv00);
 743
 744             /* Compute parameters for interactions between i and j atoms */
 745             qq00             = _mm_mul_pd(iq0,jq0);
 746             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 747                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 748
 749             /* Calculate table index by multiplying r with table scale and truncate to integer */
 750             rt               = _mm_mul_pd(r00,vftabscale);
 751             vfitab           = _mm_cvttpd_epi32(rt);
 752             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 753             vfitab           = _mm_slli_epi32(vfitab,3);
 754
 755             /* COULOMB ELECTROSTATICS */
 756             velec            = _mm_mul_pd(qq00,rinv00);
 757             felec            = _mm_mul_pd(velec,rinvsq00);
 758
 759             /* CUBIC SPLINE TABLE DISPERSION */
 760             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 761             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 762             GMX_MM_TRANSPOSE2_PD(Y,F);
 763             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 764             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 765             GMX_MM_TRANSPOSE2_PD(G,H);
 766             Heps             = _mm_mul_pd(vfeps,H);
 767             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 768             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 769             fvdw6            = _mm_mul_pd(c6_00,FF);
 770
 771             /* CUBIC SPLINE TABLE REPULSION */
 772             vfitab           = _mm_add_epi32(vfitab,ifour);
 773             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 774             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 775             GMX_MM_TRANSPOSE2_PD(Y,F);
 776             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 777             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 778             GMX_MM_TRANSPOSE2_PD(G,H);
 779             Heps             = _mm_mul_pd(vfeps,H);
 780             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 781             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 782             fvdw12           = _mm_mul_pd(c12_00,FF);
 783             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 784
 785             fscal            = _mm_add_pd(felec,fvdw);
 786
 787             /* Calculate temporary vectorial force */
 788             tx               = _mm_mul_pd(fscal,dx00);
 789             ty               = _mm_mul_pd(fscal,dy00);
 790             tz               = _mm_mul_pd(fscal,dz00);
 791
 792             /* Update vectorial force */
 793             fix0             = _mm_add_pd(fix0,tx);
 794             fiy0             = _mm_add_pd(fiy0,ty);
 795             fiz0             = _mm_add_pd(fiz0,tz);
 796
 797             fjx0             = _mm_add_pd(fjx0,tx);
 798             fjy0             = _mm_add_pd(fjy0,ty);
 799             fjz0             = _mm_add_pd(fjz0,tz);
 800
 801             /**************************
 802              * CALCULATE INTERACTIONS *
 803              **************************/
 804
 805             /* Compute parameters for interactions between i and j atoms */
 806             qq10             = _mm_mul_pd(iq1,jq0);
 807
 808             /* COULOMB ELECTROSTATICS */
 809             velec            = _mm_mul_pd(qq10,rinv10);
 810             felec            = _mm_mul_pd(velec,rinvsq10);
 811
 812             fscal            = felec;
 813
 814             /* Calculate temporary vectorial force */
 815             tx               = _mm_mul_pd(fscal,dx10);
 816             ty               = _mm_mul_pd(fscal,dy10);
 817             tz               = _mm_mul_pd(fscal,dz10);
 818
 819             /* Update vectorial force */
 820             fix1             = _mm_add_pd(fix1,tx);
 821             fiy1             = _mm_add_pd(fiy1,ty);
 822             fiz1             = _mm_add_pd(fiz1,tz);
 823
 824             fjx0             = _mm_add_pd(fjx0,tx);
 825             fjy0             = _mm_add_pd(fjy0,ty);
 826             fjz0             = _mm_add_pd(fjz0,tz);
 827
 828             /**************************
 829              * CALCULATE INTERACTIONS *
 830              **************************/
 831
 832             /* Compute parameters for interactions between i and j atoms */
 833             qq20             = _mm_mul_pd(iq2,jq0);
 834
 835             /* COULOMB ELECTROSTATICS */
 836             velec            = _mm_mul_pd(qq20,rinv20);
 837             felec            = _mm_mul_pd(velec,rinvsq20);
 838
 839             fscal            = felec;
 840
 841             /* Calculate temporary vectorial force */
 842             tx               = _mm_mul_pd(fscal,dx20);
 843             ty               = _mm_mul_pd(fscal,dy20);
 844             tz               = _mm_mul_pd(fscal,dz20);
 845
 846             /* Update vectorial force */
 847             fix2             = _mm_add_pd(fix2,tx);
 848             fiy2             = _mm_add_pd(fiy2,ty);
 849             fiz2             = _mm_add_pd(fiz2,tz);
 850
 851             fjx0             = _mm_add_pd(fjx0,tx);
 852             fjy0             = _mm_add_pd(fjy0,ty);
 853             fjz0             = _mm_add_pd(fjz0,tz);
 854
 855             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 856
 857             /* Inner loop uses 111 flops */
 858         }
 859
 860         if(jidx<j_index_end)
 861         {
 862
 863             jnrA             = jjnr[jidx];
 864             j_coord_offsetA  = DIM*jnrA;
 865
 866             /* load j atom coordinates */
 867             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 868                                               &jx0,&jy0,&jz0);
 869
 870             /* Calculate displacement vector */
 871             dx00             = _mm_sub_pd(ix0,jx0);
 872             dy00             = _mm_sub_pd(iy0,jy0);
 873             dz00             = _mm_sub_pd(iz0,jz0);
 874             dx10             = _mm_sub_pd(ix1,jx0);
 875             dy10             = _mm_sub_pd(iy1,jy0);
 876             dz10             = _mm_sub_pd(iz1,jz0);
 877             dx20             = _mm_sub_pd(ix2,jx0);
 878             dy20             = _mm_sub_pd(iy2,jy0);
 879             dz20             = _mm_sub_pd(iz2,jz0);
 880
 881             /* Calculate squared distance and things based on it */
 882             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 883             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 884             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 885
 886             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 887             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 888             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 889
 890             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 891             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 892             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 893
 894             /* Load parameters for j particles */
 895             jq0              = _mm_load_sd(charge+jnrA+0);
 896             vdwjidx0A        = 2*vdwtype[jnrA+0];
 897
 898             fjx0             = _mm_setzero_pd();
 899             fjy0             = _mm_setzero_pd();
 900             fjz0             = _mm_setzero_pd();
 901
 902             /**************************
 903              * CALCULATE INTERACTIONS *
 904              **************************/
 905
 906             r00              = _mm_mul_pd(rsq00,rinv00);
 907
 908             /* Compute parameters for interactions between i and j atoms */
 909             qq00             = _mm_mul_pd(iq0,jq0);
 910             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 911
 912             /* Calculate table index by multiplying r with table scale and truncate to integer */
 913             rt               = _mm_mul_pd(r00,vftabscale);
 914             vfitab           = _mm_cvttpd_epi32(rt);
 915             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 916             vfitab           = _mm_slli_epi32(vfitab,3);
 917
 918             /* COULOMB ELECTROSTATICS */
 919             velec            = _mm_mul_pd(qq00,rinv00);
 920             felec            = _mm_mul_pd(velec,rinvsq00);
 921
 922             /* CUBIC SPLINE TABLE DISPERSION */
 923             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 924             F                = _mm_setzero_pd();
 925             GMX_MM_TRANSPOSE2_PD(Y,F);
 926             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 927             H                = _mm_setzero_pd();
 928             GMX_MM_TRANSPOSE2_PD(G,H);
 929             Heps             = _mm_mul_pd(vfeps,H);
 930             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 931             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 932             fvdw6            = _mm_mul_pd(c6_00,FF);
 933
 934             /* CUBIC SPLINE TABLE REPULSION */
 935             vfitab           = _mm_add_epi32(vfitab,ifour);
 936             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 937             F                = _mm_setzero_pd();
 938             GMX_MM_TRANSPOSE2_PD(Y,F);
 939             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 940             H                = _mm_setzero_pd();
 941             GMX_MM_TRANSPOSE2_PD(G,H);
 942             Heps             = _mm_mul_pd(vfeps,H);
 943             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 944             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 945             fvdw12           = _mm_mul_pd(c12_00,FF);
 946             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 947
 948             fscal            = _mm_add_pd(felec,fvdw);
 949
 950             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 951
 952             /* Calculate temporary vectorial force */
 953             tx               = _mm_mul_pd(fscal,dx00);
 954             ty               = _mm_mul_pd(fscal,dy00);
 955             tz               = _mm_mul_pd(fscal,dz00);
 956
 957             /* Update vectorial force */
 958             fix0             = _mm_add_pd(fix0,tx);
 959             fiy0             = _mm_add_pd(fiy0,ty);
 960             fiz0             = _mm_add_pd(fiz0,tz);
 961
 962             fjx0             = _mm_add_pd(fjx0,tx);
 963             fjy0             = _mm_add_pd(fjy0,ty);
 964             fjz0             = _mm_add_pd(fjz0,tz);
 965
 966             /**************************
 967              * CALCULATE INTERACTIONS *
 968              **************************/
 969
 970             /* Compute parameters for interactions between i and j atoms */
 971             qq10             = _mm_mul_pd(iq1,jq0);
 972
 973             /* COULOMB ELECTROSTATICS */
 974             velec            = _mm_mul_pd(qq10,rinv10);
 975             felec            = _mm_mul_pd(velec,rinvsq10);
 976
 977             fscal            = felec;
 978
 979             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 980
 981             /* Calculate temporary vectorial force */
 982             tx               = _mm_mul_pd(fscal,dx10);
 983             ty               = _mm_mul_pd(fscal,dy10);
 984             tz               = _mm_mul_pd(fscal,dz10);
 985
 986             /* Update vectorial force */
 987             fix1             = _mm_add_pd(fix1,tx);
 988             fiy1             = _mm_add_pd(fiy1,ty);
 989             fiz1             = _mm_add_pd(fiz1,tz);
 990
 991             fjx0             = _mm_add_pd(fjx0,tx);
 992             fjy0             = _mm_add_pd(fjy0,ty);
 993             fjz0             = _mm_add_pd(fjz0,tz);
 994
 995             /**************************
 996              * CALCULATE INTERACTIONS *
 997              **************************/
 998
 999             /* Compute parameters for interactions between i and j atoms */
1000             qq20             = _mm_mul_pd(iq2,jq0);
1001
1002             /* COULOMB ELECTROSTATICS */
1003             velec            = _mm_mul_pd(qq20,rinv20);
1004             felec            = _mm_mul_pd(velec,rinvsq20);
1005
1006             fscal            = felec;
1007
1008             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1009
1010             /* Calculate temporary vectorial force */
1011             tx               = _mm_mul_pd(fscal,dx20);
1012             ty               = _mm_mul_pd(fscal,dy20);
1013             tz               = _mm_mul_pd(fscal,dz20);
1014
1015             /* Update vectorial force */
1016             fix2             = _mm_add_pd(fix2,tx);
1017             fiy2             = _mm_add_pd(fiy2,ty);
1018             fiz2             = _mm_add_pd(fiz2,tz);
1019
1020             fjx0             = _mm_add_pd(fjx0,tx);
1021             fjy0             = _mm_add_pd(fjy0,ty);
1022             fjz0             = _mm_add_pd(fjz0,tz);
1023
1024             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1025
1026             /* Inner loop uses 111 flops */
1027         }
1028
1029         /* End of innermost loop */
1030
1031         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1032                                               f+i_coord_offset,fshift+i_shift_offset);
1033
1034         /* Increment number of inner iterations */
1035         inneriter                  += j_index_end - j_index_start;
1036
1037         /* Outer loop uses 18 flops */
1038     }
1039
1040     /* Increment number of outer iterations */
1041     outeriter        += nri;
1042
1043     /* Update outer/inner flops */
1044
1045     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*111);
1046 }