src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse4_1_double
  54  * Electrostatics interaction: ReactionField
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              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     krf              = _mm_set1_pd(fr->ic->k_rf);
 123     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 124     crf              = _mm_set1_pd(fr->ic->c_rf);
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 135     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 136     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 140     rcutoff_scalar   = fr->rcoulomb;
 141     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 142     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 143
 144     /* Avoid stupid compiler warnings */
 145     jnrA = jnrB = 0;
 146     j_coord_offsetA = 0;
 147     j_coord_offsetB = 0;
 148
 149     outeriter        = 0;
 150     inneriter        = 0;
 151
 152     /* Start outer loop over neighborlists */
 153     for(iidx=0; iidx<nri; iidx++)
 154     {
 155         /* Load shift vector for this list */
 156         i_shift_offset   = DIM*shiftidx[iidx];
 157
 158         /* Load limits for loop over neighbors */
 159         j_index_start    = jindex[iidx];
 160         j_index_end      = jindex[iidx+1];
 161
 162         /* Get outer coordinate index */
 163         inr              = iinr[iidx];
 164         i_coord_offset   = DIM*inr;
 165
 166         /* Load i particle coords and add shift vector */
 167         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 168                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 169
 170         fix0             = _mm_setzero_pd();
 171         fiy0             = _mm_setzero_pd();
 172         fiz0             = _mm_setzero_pd();
 173         fix1             = _mm_setzero_pd();
 174         fiy1             = _mm_setzero_pd();
 175         fiz1             = _mm_setzero_pd();
 176         fix2             = _mm_setzero_pd();
 177         fiy2             = _mm_setzero_pd();
 178         fiz2             = _mm_setzero_pd();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_pd();
 182         vvdwsum          = _mm_setzero_pd();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_pd(ix0,jx0);
 200             dy00             = _mm_sub_pd(iy0,jy0);
 201             dz00             = _mm_sub_pd(iz0,jz0);
 202             dx10             = _mm_sub_pd(ix1,jx0);
 203             dy10             = _mm_sub_pd(iy1,jy0);
 204             dz10             = _mm_sub_pd(iz1,jz0);
 205             dx20             = _mm_sub_pd(ix2,jx0);
 206             dy20             = _mm_sub_pd(iy2,jy0);
 207             dz20             = _mm_sub_pd(iz2,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 213
 214             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 215             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 216             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 217
 218             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 219             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226
 227             fjx0             = _mm_setzero_pd();
 228             fjy0             = _mm_setzero_pd();
 229             fjz0             = _mm_setzero_pd();
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             if (gmx_mm_any_lt(rsq00,rcutoff2))
 236             {
 237
 238             r00              = _mm_mul_pd(rsq00,rinv00);
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             qq00             = _mm_mul_pd(iq0,jq0);
 242             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 243                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 244
 245             /* Calculate table index by multiplying r with table scale and truncate to integer */
 246             rt               = _mm_mul_pd(r00,vftabscale);
 247             vfitab           = _mm_cvttpd_epi32(rt);
 248             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 249             vfitab           = _mm_slli_epi32(vfitab,3);
 250
 251             /* REACTION-FIELD ELECTROSTATICS */
 252             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 253             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 254
 255             /* CUBIC SPLINE TABLE DISPERSION */
 256             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 257             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 258             GMX_MM_TRANSPOSE2_PD(Y,F);
 259             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 260             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 261             GMX_MM_TRANSPOSE2_PD(G,H);
 262             Heps             = _mm_mul_pd(vfeps,H);
 263             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 264             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 265             vvdw6            = _mm_mul_pd(c6_00,VV);
 266             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 267             fvdw6            = _mm_mul_pd(c6_00,FF);
 268
 269             /* CUBIC SPLINE TABLE REPULSION */
 270             vfitab           = _mm_add_epi32(vfitab,ifour);
 271             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 272             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 273             GMX_MM_TRANSPOSE2_PD(Y,F);
 274             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 275             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 276             GMX_MM_TRANSPOSE2_PD(G,H);
 277             Heps             = _mm_mul_pd(vfeps,H);
 278             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 279             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 280             vvdw12           = _mm_mul_pd(c12_00,VV);
 281             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 282             fvdw12           = _mm_mul_pd(c12_00,FF);
 283             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 284             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 285
 286             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             velec            = _mm_and_pd(velec,cutoff_mask);
 290             velecsum         = _mm_add_pd(velecsum,velec);
 291             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 292             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 293
 294             fscal            = _mm_add_pd(felec,fvdw);
 295
 296             fscal            = _mm_and_pd(fscal,cutoff_mask);
 297
 298             /* Calculate temporary vectorial force */
 299             tx               = _mm_mul_pd(fscal,dx00);
 300             ty               = _mm_mul_pd(fscal,dy00);
 301             tz               = _mm_mul_pd(fscal,dz00);
 302
 303             /* Update vectorial force */
 304             fix0             = _mm_add_pd(fix0,tx);
 305             fiy0             = _mm_add_pd(fiy0,ty);
 306             fiz0             = _mm_add_pd(fiz0,tz);
 307
 308             fjx0             = _mm_add_pd(fjx0,tx);
 309             fjy0             = _mm_add_pd(fjy0,ty);
 310             fjz0             = _mm_add_pd(fjz0,tz);
 311
 312             }
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             if (gmx_mm_any_lt(rsq10,rcutoff2))
 319             {
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq10             = _mm_mul_pd(iq1,jq0);
 323
 324             /* REACTION-FIELD ELECTROSTATICS */
 325             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 326             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 327
 328             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velec            = _mm_and_pd(velec,cutoff_mask);
 332             velecsum         = _mm_add_pd(velecsum,velec);
 333
 334             fscal            = felec;
 335
 336             fscal            = _mm_and_pd(fscal,cutoff_mask);
 337
 338             /* Calculate temporary vectorial force */
 339             tx               = _mm_mul_pd(fscal,dx10);
 340             ty               = _mm_mul_pd(fscal,dy10);
 341             tz               = _mm_mul_pd(fscal,dz10);
 342
 343             /* Update vectorial force */
 344             fix1             = _mm_add_pd(fix1,tx);
 345             fiy1             = _mm_add_pd(fiy1,ty);
 346             fiz1             = _mm_add_pd(fiz1,tz);
 347
 348             fjx0             = _mm_add_pd(fjx0,tx);
 349             fjy0             = _mm_add_pd(fjy0,ty);
 350             fjz0             = _mm_add_pd(fjz0,tz);
 351
 352             }
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             if (gmx_mm_any_lt(rsq20,rcutoff2))
 359             {
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq20             = _mm_mul_pd(iq2,jq0);
 363
 364             /* REACTION-FIELD ELECTROSTATICS */
 365             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 366             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 367
 368             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm_and_pd(velec,cutoff_mask);
 372             velecsum         = _mm_add_pd(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             fscal            = _mm_and_pd(fscal,cutoff_mask);
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm_mul_pd(fscal,dx20);
 380             ty               = _mm_mul_pd(fscal,dy20);
 381             tz               = _mm_mul_pd(fscal,dz20);
 382
 383             /* Update vectorial force */
 384             fix2             = _mm_add_pd(fix2,tx);
 385             fiy2             = _mm_add_pd(fiy2,ty);
 386             fiz2             = _mm_add_pd(fiz2,tz);
 387
 388             fjx0             = _mm_add_pd(fjx0,tx);
 389             fjy0             = _mm_add_pd(fjy0,ty);
 390             fjz0             = _mm_add_pd(fjz0,tz);
 391
 392             }
 393
 394             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 395
 396             /* Inner loop uses 147 flops */
 397         }
 398
 399         if(jidx<j_index_end)
 400         {
 401
 402             jnrA             = jjnr[jidx];
 403             j_coord_offsetA  = DIM*jnrA;
 404
 405             /* load j atom coordinates */
 406             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 407                                               &jx0,&jy0,&jz0);
 408
 409             /* Calculate displacement vector */
 410             dx00             = _mm_sub_pd(ix0,jx0);
 411             dy00             = _mm_sub_pd(iy0,jy0);
 412             dz00             = _mm_sub_pd(iz0,jz0);
 413             dx10             = _mm_sub_pd(ix1,jx0);
 414             dy10             = _mm_sub_pd(iy1,jy0);
 415             dz10             = _mm_sub_pd(iz1,jz0);
 416             dx20             = _mm_sub_pd(ix2,jx0);
 417             dy20             = _mm_sub_pd(iy2,jy0);
 418             dz20             = _mm_sub_pd(iz2,jz0);
 419
 420             /* Calculate squared distance and things based on it */
 421             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 422             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 423             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 424
 425             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 426             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 427             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 428
 429             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 430             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 431             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 432
 433             /* Load parameters for j particles */
 434             jq0              = _mm_load_sd(charge+jnrA+0);
 435             vdwjidx0A        = 2*vdwtype[jnrA+0];
 436
 437             fjx0             = _mm_setzero_pd();
 438             fjy0             = _mm_setzero_pd();
 439             fjz0             = _mm_setzero_pd();
 440
 441             /**************************
 442              * CALCULATE INTERACTIONS *
 443              **************************/
 444
 445             if (gmx_mm_any_lt(rsq00,rcutoff2))
 446             {
 447
 448             r00              = _mm_mul_pd(rsq00,rinv00);
 449
 450             /* Compute parameters for interactions between i and j atoms */
 451             qq00             = _mm_mul_pd(iq0,jq0);
 452             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 453
 454             /* Calculate table index by multiplying r with table scale and truncate to integer */
 455             rt               = _mm_mul_pd(r00,vftabscale);
 456             vfitab           = _mm_cvttpd_epi32(rt);
 457             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 458             vfitab           = _mm_slli_epi32(vfitab,3);
 459
 460             /* REACTION-FIELD ELECTROSTATICS */
 461             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 462             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 463
 464             /* CUBIC SPLINE TABLE DISPERSION */
 465             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 466             F                = _mm_setzero_pd();
 467             GMX_MM_TRANSPOSE2_PD(Y,F);
 468             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 469             H                = _mm_setzero_pd();
 470             GMX_MM_TRANSPOSE2_PD(G,H);
 471             Heps             = _mm_mul_pd(vfeps,H);
 472             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 473             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 474             vvdw6            = _mm_mul_pd(c6_00,VV);
 475             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 476             fvdw6            = _mm_mul_pd(c6_00,FF);
 477
 478             /* CUBIC SPLINE TABLE REPULSION */
 479             vfitab           = _mm_add_epi32(vfitab,ifour);
 480             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 481             F                = _mm_setzero_pd();
 482             GMX_MM_TRANSPOSE2_PD(Y,F);
 483             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 484             H                = _mm_setzero_pd();
 485             GMX_MM_TRANSPOSE2_PD(G,H);
 486             Heps             = _mm_mul_pd(vfeps,H);
 487             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 488             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 489             vvdw12           = _mm_mul_pd(c12_00,VV);
 490             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 491             fvdw12           = _mm_mul_pd(c12_00,FF);
 492             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 493             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 494
 495             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 496
 497             /* Update potential sum for this i atom from the interaction with this j atom. */
 498             velec            = _mm_and_pd(velec,cutoff_mask);
 499             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 500             velecsum         = _mm_add_pd(velecsum,velec);
 501             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 502             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 503             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 504
 505             fscal            = _mm_add_pd(felec,fvdw);
 506
 507             fscal            = _mm_and_pd(fscal,cutoff_mask);
 508
 509             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 510
 511             /* Calculate temporary vectorial force */
 512             tx               = _mm_mul_pd(fscal,dx00);
 513             ty               = _mm_mul_pd(fscal,dy00);
 514             tz               = _mm_mul_pd(fscal,dz00);
 515
 516             /* Update vectorial force */
 517             fix0             = _mm_add_pd(fix0,tx);
 518             fiy0             = _mm_add_pd(fiy0,ty);
 519             fiz0             = _mm_add_pd(fiz0,tz);
 520
 521             fjx0             = _mm_add_pd(fjx0,tx);
 522             fjy0             = _mm_add_pd(fjy0,ty);
 523             fjz0             = _mm_add_pd(fjz0,tz);
 524
 525             }
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             if (gmx_mm_any_lt(rsq10,rcutoff2))
 532             {
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq10             = _mm_mul_pd(iq1,jq0);
 536
 537             /* REACTION-FIELD ELECTROSTATICS */
 538             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_add_pd(rinv10,_mm_mul_pd(krf,rsq10)),crf));
 539             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 540
 541             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm_and_pd(velec,cutoff_mask);
 545             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 546             velecsum         = _mm_add_pd(velecsum,velec);
 547
 548             fscal            = felec;
 549
 550             fscal            = _mm_and_pd(fscal,cutoff_mask);
 551
 552             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm_mul_pd(fscal,dx10);
 556             ty               = _mm_mul_pd(fscal,dy10);
 557             tz               = _mm_mul_pd(fscal,dz10);
 558
 559             /* Update vectorial force */
 560             fix1             = _mm_add_pd(fix1,tx);
 561             fiy1             = _mm_add_pd(fiy1,ty);
 562             fiz1             = _mm_add_pd(fiz1,tz);
 563
 564             fjx0             = _mm_add_pd(fjx0,tx);
 565             fjy0             = _mm_add_pd(fjy0,ty);
 566             fjz0             = _mm_add_pd(fjz0,tz);
 567
 568             }
 569
 570             /**************************
 571              * CALCULATE INTERACTIONS *
 572              **************************/
 573
 574             if (gmx_mm_any_lt(rsq20,rcutoff2))
 575             {
 576
 577             /* Compute parameters for interactions between i and j atoms */
 578             qq20             = _mm_mul_pd(iq2,jq0);
 579
 580             /* REACTION-FIELD ELECTROSTATICS */
 581             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_add_pd(rinv20,_mm_mul_pd(krf,rsq20)),crf));
 582             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 583
 584             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm_and_pd(velec,cutoff_mask);
 588             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 589             velecsum         = _mm_add_pd(velecsum,velec);
 590
 591             fscal            = felec;
 592
 593             fscal            = _mm_and_pd(fscal,cutoff_mask);
 594
 595             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 596
 597             /* Calculate temporary vectorial force */
 598             tx               = _mm_mul_pd(fscal,dx20);
 599             ty               = _mm_mul_pd(fscal,dy20);
 600             tz               = _mm_mul_pd(fscal,dz20);
 601
 602             /* Update vectorial force */
 603             fix2             = _mm_add_pd(fix2,tx);
 604             fiy2             = _mm_add_pd(fiy2,ty);
 605             fiz2             = _mm_add_pd(fiz2,tz);
 606
 607             fjx0             = _mm_add_pd(fjx0,tx);
 608             fjy0             = _mm_add_pd(fjy0,ty);
 609             fjz0             = _mm_add_pd(fjz0,tz);
 610
 611             }
 612
 613             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 614
 615             /* Inner loop uses 147 flops */
 616         }
 617
 618         /* End of innermost loop */
 619
 620         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 621                                               f+i_coord_offset,fshift+i_shift_offset);
 622
 623         ggid                        = gid[iidx];
 624         /* Update potential energies */
 625         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 626         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 627
 628         /* Increment number of inner iterations */
 629         inneriter                  += j_index_end - j_index_start;
 630
 631         /* Outer loop uses 20 flops */
 632     }
 633
 634     /* Increment number of outer iterations */
 635     outeriter        += nri;
 636
 637     /* Update outer/inner flops */
 638
 639     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*147);
 640 }
 641 /*
 642  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse4_1_double
 643  * Electrostatics interaction: ReactionField
 644  * VdW interaction:            CubicSplineTable
 645  * Geometry:                   Water3-Particle
 646  * Calculate force/pot:        Force
 647  */
 648 void
 649 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse4_1_double
 650                     (t_nblist                    * gmx_restrict       nlist,
 651                      rvec                        * gmx_restrict          xx,
 652                      rvec                        * gmx_restrict          ff,
 653                      t_forcerec                  * gmx_restrict          fr,
 654                      t_mdatoms                   * gmx_restrict     mdatoms,
 655                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 656                      t_nrnb                      * gmx_restrict        nrnb)
 657 {
 658     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 659      * just 0 for non-waters.
 660      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 661      * jnr indices corresponding to data put in the four positions in the SIMD register.
 662      */
 663     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 664     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 665     int              jnrA,jnrB;
 666     int              j_coord_offsetA,j_coord_offsetB;
 667     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 668     real             rcutoff_scalar;
 669     real             *shiftvec,*fshift,*x,*f;
 670     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 671     int              vdwioffset0;
 672     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 673     int              vdwioffset1;
 674     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 675     int              vdwioffset2;
 676     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 677     int              vdwjidx0A,vdwjidx0B;
 678     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 679     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 680     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 681     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 682     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 683     real             *charge;
 684     int              nvdwtype;
 685     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 686     int              *vdwtype;
 687     real             *vdwparam;
 688     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 689     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 690     __m128i          vfitab;
 691     __m128i          ifour       = _mm_set1_epi32(4);
 692     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 693     real             *vftab;
 694     __m128d          dummy_mask,cutoff_mask;
 695     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 696     __m128d          one     = _mm_set1_pd(1.0);
 697     __m128d          two     = _mm_set1_pd(2.0);
 698     x                = xx[0];
 699     f                = ff[0];
 700
 701     nri              = nlist->nri;
 702     iinr             = nlist->iinr;
 703     jindex           = nlist->jindex;
 704     jjnr             = nlist->jjnr;
 705     shiftidx         = nlist->shift;
 706     gid              = nlist->gid;
 707     shiftvec         = fr->shift_vec[0];
 708     fshift           = fr->fshift[0];
 709     facel            = _mm_set1_pd(fr->epsfac);
 710     charge           = mdatoms->chargeA;
 711     krf              = _mm_set1_pd(fr->ic->k_rf);
 712     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 713     crf              = _mm_set1_pd(fr->ic->c_rf);
 714     nvdwtype         = fr->ntype;
 715     vdwparam         = fr->nbfp;
 716     vdwtype          = mdatoms->typeA;
 717
 718     vftab            = kernel_data->table_vdw->data;
 719     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 720
 721     /* Setup water-specific parameters */
 722     inr              = nlist->iinr[0];
 723     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 724     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 725     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 726     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 727
 728     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 729     rcutoff_scalar   = fr->rcoulomb;
 730     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 731     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 732
 733     /* Avoid stupid compiler warnings */
 734     jnrA = jnrB = 0;
 735     j_coord_offsetA = 0;
 736     j_coord_offsetB = 0;
 737
 738     outeriter        = 0;
 739     inneriter        = 0;
 740
 741     /* Start outer loop over neighborlists */
 742     for(iidx=0; iidx<nri; iidx++)
 743     {
 744         /* Load shift vector for this list */
 745         i_shift_offset   = DIM*shiftidx[iidx];
 746
 747         /* Load limits for loop over neighbors */
 748         j_index_start    = jindex[iidx];
 749         j_index_end      = jindex[iidx+1];
 750
 751         /* Get outer coordinate index */
 752         inr              = iinr[iidx];
 753         i_coord_offset   = DIM*inr;
 754
 755         /* Load i particle coords and add shift vector */
 756         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 757                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 758
 759         fix0             = _mm_setzero_pd();
 760         fiy0             = _mm_setzero_pd();
 761         fiz0             = _mm_setzero_pd();
 762         fix1             = _mm_setzero_pd();
 763         fiy1             = _mm_setzero_pd();
 764         fiz1             = _mm_setzero_pd();
 765         fix2             = _mm_setzero_pd();
 766         fiy2             = _mm_setzero_pd();
 767         fiz2             = _mm_setzero_pd();
 768
 769         /* Start inner kernel loop */
 770         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 771         {
 772
 773             /* Get j neighbor index, and coordinate index */
 774             jnrA             = jjnr[jidx];
 775             jnrB             = jjnr[jidx+1];
 776             j_coord_offsetA  = DIM*jnrA;
 777             j_coord_offsetB  = DIM*jnrB;
 778
 779             /* load j atom coordinates */
 780             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 781                                               &jx0,&jy0,&jz0);
 782
 783             /* Calculate displacement vector */
 784             dx00             = _mm_sub_pd(ix0,jx0);
 785             dy00             = _mm_sub_pd(iy0,jy0);
 786             dz00             = _mm_sub_pd(iz0,jz0);
 787             dx10             = _mm_sub_pd(ix1,jx0);
 788             dy10             = _mm_sub_pd(iy1,jy0);
 789             dz10             = _mm_sub_pd(iz1,jz0);
 790             dx20             = _mm_sub_pd(ix2,jx0);
 791             dy20             = _mm_sub_pd(iy2,jy0);
 792             dz20             = _mm_sub_pd(iz2,jz0);
 793
 794             /* Calculate squared distance and things based on it */
 795             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 796             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 797             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 798
 799             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 800             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 801             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 802
 803             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 804             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 805             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 806
 807             /* Load parameters for j particles */
 808             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 809             vdwjidx0A        = 2*vdwtype[jnrA+0];
 810             vdwjidx0B        = 2*vdwtype[jnrB+0];
 811
 812             fjx0             = _mm_setzero_pd();
 813             fjy0             = _mm_setzero_pd();
 814             fjz0             = _mm_setzero_pd();
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             if (gmx_mm_any_lt(rsq00,rcutoff2))
 821             {
 822
 823             r00              = _mm_mul_pd(rsq00,rinv00);
 824
 825             /* Compute parameters for interactions between i and j atoms */
 826             qq00             = _mm_mul_pd(iq0,jq0);
 827             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 828                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 829
 830             /* Calculate table index by multiplying r with table scale and truncate to integer */
 831             rt               = _mm_mul_pd(r00,vftabscale);
 832             vfitab           = _mm_cvttpd_epi32(rt);
 833             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 834             vfitab           = _mm_slli_epi32(vfitab,3);
 835
 836             /* REACTION-FIELD ELECTROSTATICS */
 837             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 838
 839             /* CUBIC SPLINE TABLE DISPERSION */
 840             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 841             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 842             GMX_MM_TRANSPOSE2_PD(Y,F);
 843             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 844             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 845             GMX_MM_TRANSPOSE2_PD(G,H);
 846             Heps             = _mm_mul_pd(vfeps,H);
 847             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 848             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 849             fvdw6            = _mm_mul_pd(c6_00,FF);
 850
 851             /* CUBIC SPLINE TABLE REPULSION */
 852             vfitab           = _mm_add_epi32(vfitab,ifour);
 853             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 854             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 855             GMX_MM_TRANSPOSE2_PD(Y,F);
 856             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 857             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 858             GMX_MM_TRANSPOSE2_PD(G,H);
 859             Heps             = _mm_mul_pd(vfeps,H);
 860             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 861             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 862             fvdw12           = _mm_mul_pd(c12_00,FF);
 863             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 864
 865             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 866
 867             fscal            = _mm_add_pd(felec,fvdw);
 868
 869             fscal            = _mm_and_pd(fscal,cutoff_mask);
 870
 871             /* Calculate temporary vectorial force */
 872             tx               = _mm_mul_pd(fscal,dx00);
 873             ty               = _mm_mul_pd(fscal,dy00);
 874             tz               = _mm_mul_pd(fscal,dz00);
 875
 876             /* Update vectorial force */
 877             fix0             = _mm_add_pd(fix0,tx);
 878             fiy0             = _mm_add_pd(fiy0,ty);
 879             fiz0             = _mm_add_pd(fiz0,tz);
 880
 881             fjx0             = _mm_add_pd(fjx0,tx);
 882             fjy0             = _mm_add_pd(fjy0,ty);
 883             fjz0             = _mm_add_pd(fjz0,tz);
 884
 885             }
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             if (gmx_mm_any_lt(rsq10,rcutoff2))
 892             {
 893
 894             /* Compute parameters for interactions between i and j atoms */
 895             qq10             = _mm_mul_pd(iq1,jq0);
 896
 897             /* REACTION-FIELD ELECTROSTATICS */
 898             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
 899
 900             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 901
 902             fscal            = felec;
 903
 904             fscal            = _mm_and_pd(fscal,cutoff_mask);
 905
 906             /* Calculate temporary vectorial force */
 907             tx               = _mm_mul_pd(fscal,dx10);
 908             ty               = _mm_mul_pd(fscal,dy10);
 909             tz               = _mm_mul_pd(fscal,dz10);
 910
 911             /* Update vectorial force */
 912             fix1             = _mm_add_pd(fix1,tx);
 913             fiy1             = _mm_add_pd(fiy1,ty);
 914             fiz1             = _mm_add_pd(fiz1,tz);
 915
 916             fjx0             = _mm_add_pd(fjx0,tx);
 917             fjy0             = _mm_add_pd(fjy0,ty);
 918             fjz0             = _mm_add_pd(fjz0,tz);
 919
 920             }
 921
 922             /**************************
 923              * CALCULATE INTERACTIONS *
 924              **************************/
 925
 926             if (gmx_mm_any_lt(rsq20,rcutoff2))
 927             {
 928
 929             /* Compute parameters for interactions between i and j atoms */
 930             qq20             = _mm_mul_pd(iq2,jq0);
 931
 932             /* REACTION-FIELD ELECTROSTATICS */
 933             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
 934
 935             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 936
 937             fscal            = felec;
 938
 939             fscal            = _mm_and_pd(fscal,cutoff_mask);
 940
 941             /* Calculate temporary vectorial force */
 942             tx               = _mm_mul_pd(fscal,dx20);
 943             ty               = _mm_mul_pd(fscal,dy20);
 944             tz               = _mm_mul_pd(fscal,dz20);
 945
 946             /* Update vectorial force */
 947             fix2             = _mm_add_pd(fix2,tx);
 948             fiy2             = _mm_add_pd(fiy2,ty);
 949             fiz2             = _mm_add_pd(fiz2,tz);
 950
 951             fjx0             = _mm_add_pd(fjx0,tx);
 952             fjy0             = _mm_add_pd(fjy0,ty);
 953             fjz0             = _mm_add_pd(fjz0,tz);
 954
 955             }
 956
 957             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 958
 959             /* Inner loop uses 120 flops */
 960         }
 961
 962         if(jidx<j_index_end)
 963         {
 964
 965             jnrA             = jjnr[jidx];
 966             j_coord_offsetA  = DIM*jnrA;
 967
 968             /* load j atom coordinates */
 969             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 970                                               &jx0,&jy0,&jz0);
 971
 972             /* Calculate displacement vector */
 973             dx00             = _mm_sub_pd(ix0,jx0);
 974             dy00             = _mm_sub_pd(iy0,jy0);
 975             dz00             = _mm_sub_pd(iz0,jz0);
 976             dx10             = _mm_sub_pd(ix1,jx0);
 977             dy10             = _mm_sub_pd(iy1,jy0);
 978             dz10             = _mm_sub_pd(iz1,jz0);
 979             dx20             = _mm_sub_pd(ix2,jx0);
 980             dy20             = _mm_sub_pd(iy2,jy0);
 981             dz20             = _mm_sub_pd(iz2,jz0);
 982
 983             /* Calculate squared distance and things based on it */
 984             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 985             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 986             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 987
 988             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 989             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 990             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 991
 992             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 993             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 994             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 995
 996             /* Load parameters for j particles */
 997             jq0              = _mm_load_sd(charge+jnrA+0);
 998             vdwjidx0A        = 2*vdwtype[jnrA+0];
 999
1000             fjx0             = _mm_setzero_pd();
1001             fjy0             = _mm_setzero_pd();
1002             fjz0             = _mm_setzero_pd();
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             if (gmx_mm_any_lt(rsq00,rcutoff2))
1009             {
1010
1011             r00              = _mm_mul_pd(rsq00,rinv00);
1012
1013             /* Compute parameters for interactions between i and j atoms */
1014             qq00             = _mm_mul_pd(iq0,jq0);
1015             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1016
1017             /* Calculate table index by multiplying r with table scale and truncate to integer */
1018             rt               = _mm_mul_pd(r00,vftabscale);
1019             vfitab           = _mm_cvttpd_epi32(rt);
1020             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1021             vfitab           = _mm_slli_epi32(vfitab,3);
1022
1023             /* REACTION-FIELD ELECTROSTATICS */
1024             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
1025
1026             /* CUBIC SPLINE TABLE DISPERSION */
1027             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1028             F                = _mm_setzero_pd();
1029             GMX_MM_TRANSPOSE2_PD(Y,F);
1030             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1031             H                = _mm_setzero_pd();
1032             GMX_MM_TRANSPOSE2_PD(G,H);
1033             Heps             = _mm_mul_pd(vfeps,H);
1034             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1035             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1036             fvdw6            = _mm_mul_pd(c6_00,FF);
1037
1038             /* CUBIC SPLINE TABLE REPULSION */
1039             vfitab           = _mm_add_epi32(vfitab,ifour);
1040             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1041             F                = _mm_setzero_pd();
1042             GMX_MM_TRANSPOSE2_PD(Y,F);
1043             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1044             H                = _mm_setzero_pd();
1045             GMX_MM_TRANSPOSE2_PD(G,H);
1046             Heps             = _mm_mul_pd(vfeps,H);
1047             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1048             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1049             fvdw12           = _mm_mul_pd(c12_00,FF);
1050             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1051
1052             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1053
1054             fscal            = _mm_add_pd(felec,fvdw);
1055
1056             fscal            = _mm_and_pd(fscal,cutoff_mask);
1057
1058             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1059
1060             /* Calculate temporary vectorial force */
1061             tx               = _mm_mul_pd(fscal,dx00);
1062             ty               = _mm_mul_pd(fscal,dy00);
1063             tz               = _mm_mul_pd(fscal,dz00);
1064
1065             /* Update vectorial force */
1066             fix0             = _mm_add_pd(fix0,tx);
1067             fiy0             = _mm_add_pd(fiy0,ty);
1068             fiz0             = _mm_add_pd(fiz0,tz);
1069
1070             fjx0             = _mm_add_pd(fjx0,tx);
1071             fjy0             = _mm_add_pd(fjy0,ty);
1072             fjz0             = _mm_add_pd(fjz0,tz);
1073
1074             }
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             if (gmx_mm_any_lt(rsq10,rcutoff2))
1081             {
1082
1083             /* Compute parameters for interactions between i and j atoms */
1084             qq10             = _mm_mul_pd(iq1,jq0);
1085
1086             /* REACTION-FIELD ELECTROSTATICS */
1087             felec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_mul_pd(rinv10,rinvsq10),krf2));
1088
1089             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1090
1091             fscal            = felec;
1092
1093             fscal            = _mm_and_pd(fscal,cutoff_mask);
1094
1095             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1096
1097             /* Calculate temporary vectorial force */
1098             tx               = _mm_mul_pd(fscal,dx10);
1099             ty               = _mm_mul_pd(fscal,dy10);
1100             tz               = _mm_mul_pd(fscal,dz10);
1101
1102             /* Update vectorial force */
1103             fix1             = _mm_add_pd(fix1,tx);
1104             fiy1             = _mm_add_pd(fiy1,ty);
1105             fiz1             = _mm_add_pd(fiz1,tz);
1106
1107             fjx0             = _mm_add_pd(fjx0,tx);
1108             fjy0             = _mm_add_pd(fjy0,ty);
1109             fjz0             = _mm_add_pd(fjz0,tz);
1110
1111             }
1112
1113             /**************************
1114              * CALCULATE INTERACTIONS *
1115              **************************/
1116
1117             if (gmx_mm_any_lt(rsq20,rcutoff2))
1118             {
1119
1120             /* Compute parameters for interactions between i and j atoms */
1121             qq20             = _mm_mul_pd(iq2,jq0);
1122
1123             /* REACTION-FIELD ELECTROSTATICS */
1124             felec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_mul_pd(rinv20,rinvsq20),krf2));
1125
1126             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1127
1128             fscal            = felec;
1129
1130             fscal            = _mm_and_pd(fscal,cutoff_mask);
1131
1132             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1133
1134             /* Calculate temporary vectorial force */
1135             tx               = _mm_mul_pd(fscal,dx20);
1136             ty               = _mm_mul_pd(fscal,dy20);
1137             tz               = _mm_mul_pd(fscal,dz20);
1138
1139             /* Update vectorial force */
1140             fix2             = _mm_add_pd(fix2,tx);
1141             fiy2             = _mm_add_pd(fiy2,ty);
1142             fiz2             = _mm_add_pd(fiz2,tz);
1143
1144             fjx0             = _mm_add_pd(fjx0,tx);
1145             fjy0             = _mm_add_pd(fjy0,ty);
1146             fjz0             = _mm_add_pd(fjz0,tz);
1147
1148             }
1149
1150             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1151
1152             /* Inner loop uses 120 flops */
1153         }
1154
1155         /* End of innermost loop */
1156
1157         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1158                                               f+i_coord_offset,fshift+i_shift_offset);
1159
1160         /* Increment number of inner iterations */
1161         inneriter                  += j_index_end - j_index_start;
1162
1163         /* Outer loop uses 18 flops */
1164     }
1165
1166     /* Increment number of outer iterations */
1167     outeriter        += nri;
1168
1169     /* Update outer/inner flops */
1170
1171     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*120);
1172 }