src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_sse4_1_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 sse4_1_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_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128d          dummy_mask,cutoff_mask;
 109     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 110     __m128d          one     = _mm_set1_pd(1.0);
 111     __m128d          two     = _mm_set1_pd(2.0);
 112     x                = xx[0];
 113     f                = ff[0];
 114
 115     nri              = nlist->nri;
 116     iinr             = nlist->iinr;
 117     jindex           = nlist->jindex;
 118     jjnr             = nlist->jjnr;
 119     shiftidx         = nlist->shift;
 120     gid              = nlist->gid;
 121     shiftvec         = fr->shift_vec[0];
 122     fshift           = fr->fshift[0];
 123     facel            = _mm_set1_pd(fr->epsfac);
 124     charge           = mdatoms->chargeA;
 125     nvdwtype         = fr->ntype;
 126     vdwparam         = fr->nbfp;
 127     vdwtype          = mdatoms->typeA;
 128
 129     vftab            = kernel_data->table_vdw->data;
 130     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 163                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 164
 165         fix0             = _mm_setzero_pd();
 166         fiy0             = _mm_setzero_pd();
 167         fiz0             = _mm_setzero_pd();
 168         fix1             = _mm_setzero_pd();
 169         fiy1             = _mm_setzero_pd();
 170         fiz1             = _mm_setzero_pd();
 171         fix2             = _mm_setzero_pd();
 172         fiy2             = _mm_setzero_pd();
 173         fiz2             = _mm_setzero_pd();
 174         fix3             = _mm_setzero_pd();
 175         fiy3             = _mm_setzero_pd();
 176         fiz3             = _mm_setzero_pd();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_pd();
 180         vvdwsum          = _mm_setzero_pd();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_pd(ix0,jx0);
 198             dy00             = _mm_sub_pd(iy0,jy0);
 199             dz00             = _mm_sub_pd(iz0,jz0);
 200             dx10             = _mm_sub_pd(ix1,jx0);
 201             dy10             = _mm_sub_pd(iy1,jy0);
 202             dz10             = _mm_sub_pd(iz1,jz0);
 203             dx20             = _mm_sub_pd(ix2,jx0);
 204             dy20             = _mm_sub_pd(iy2,jy0);
 205             dz20             = _mm_sub_pd(iz2,jz0);
 206             dx30             = _mm_sub_pd(ix3,jx0);
 207             dy30             = _mm_sub_pd(iy3,jy0);
 208             dz30             = _mm_sub_pd(iz3,jz0);
 209
 210             /* Calculate squared distance and things based on it */
 211             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 212             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 213             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 214             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 215
 216             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 217             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 218             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 219             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 220
 221             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 223             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 227             vdwjidx0A        = 2*vdwtype[jnrA+0];
 228             vdwjidx0B        = 2*vdwtype[jnrB+0];
 229
 230             fjx0             = _mm_setzero_pd();
 231             fjy0             = _mm_setzero_pd();
 232             fjz0             = _mm_setzero_pd();
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             r00              = _mm_mul_pd(rsq00,rinv00);
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 242                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 243
 244             /* Calculate table index by multiplying r with table scale and truncate to integer */
 245             rt               = _mm_mul_pd(r00,vftabscale);
 246             vfitab           = _mm_cvttpd_epi32(rt);
 247             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 248             vfitab           = _mm_slli_epi32(vfitab,3);
 249
 250             /* CUBIC SPLINE TABLE DISPERSION */
 251             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 252             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 253             GMX_MM_TRANSPOSE2_PD(Y,F);
 254             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 255             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 256             GMX_MM_TRANSPOSE2_PD(G,H);
 257             Heps             = _mm_mul_pd(vfeps,H);
 258             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 259             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 260             vvdw6            = _mm_mul_pd(c6_00,VV);
 261             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 262             fvdw6            = _mm_mul_pd(c6_00,FF);
 263
 264             /* CUBIC SPLINE TABLE REPULSION */
 265             vfitab           = _mm_add_epi32(vfitab,ifour);
 266             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 267             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 268             GMX_MM_TRANSPOSE2_PD(Y,F);
 269             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 270             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 271             GMX_MM_TRANSPOSE2_PD(G,H);
 272             Heps             = _mm_mul_pd(vfeps,H);
 273             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 274             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 275             vvdw12           = _mm_mul_pd(c12_00,VV);
 276             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 277             fvdw12           = _mm_mul_pd(c12_00,FF);
 278             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 279             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 283
 284             fscal            = fvdw;
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_pd(fscal,dx00);
 288             ty               = _mm_mul_pd(fscal,dy00);
 289             tz               = _mm_mul_pd(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm_add_pd(fix0,tx);
 293             fiy0             = _mm_add_pd(fiy0,ty);
 294             fiz0             = _mm_add_pd(fiz0,tz);
 295
 296             fjx0             = _mm_add_pd(fjx0,tx);
 297             fjy0             = _mm_add_pd(fjy0,ty);
 298             fjz0             = _mm_add_pd(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm_mul_pd(iq1,jq0);
 306
 307             /* COULOMB ELECTROSTATICS */
 308             velec            = _mm_mul_pd(qq10,rinv10);
 309             felec            = _mm_mul_pd(velec,rinvsq10);
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velecsum         = _mm_add_pd(velecsum,velec);
 313
 314             fscal            = felec;
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm_mul_pd(fscal,dx10);
 318             ty               = _mm_mul_pd(fscal,dy10);
 319             tz               = _mm_mul_pd(fscal,dz10);
 320
 321             /* Update vectorial force */
 322             fix1             = _mm_add_pd(fix1,tx);
 323             fiy1             = _mm_add_pd(fiy1,ty);
 324             fiz1             = _mm_add_pd(fiz1,tz);
 325
 326             fjx0             = _mm_add_pd(fjx0,tx);
 327             fjy0             = _mm_add_pd(fjy0,ty);
 328             fjz0             = _mm_add_pd(fjz0,tz);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq20             = _mm_mul_pd(iq2,jq0);
 336
 337             /* COULOMB ELECTROSTATICS */
 338             velec            = _mm_mul_pd(qq20,rinv20);
 339             felec            = _mm_mul_pd(velec,rinvsq20);
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_pd(fscal,dx20);
 348             ty               = _mm_mul_pd(fscal,dy20);
 349             tz               = _mm_mul_pd(fscal,dz20);
 350
 351             /* Update vectorial force */
 352             fix2             = _mm_add_pd(fix2,tx);
 353             fiy2             = _mm_add_pd(fiy2,ty);
 354             fiz2             = _mm_add_pd(fiz2,tz);
 355
 356             fjx0             = _mm_add_pd(fjx0,tx);
 357             fjy0             = _mm_add_pd(fjy0,ty);
 358             fjz0             = _mm_add_pd(fjz0,tz);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq30             = _mm_mul_pd(iq3,jq0);
 366
 367             /* COULOMB ELECTROSTATICS */
 368             velec            = _mm_mul_pd(qq30,rinv30);
 369             felec            = _mm_mul_pd(velec,rinvsq30);
 370
 371             /* Update potential sum for this i atom from the interaction with this j atom. */
 372             velecsum         = _mm_add_pd(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             /* Calculate temporary vectorial force */
 377             tx               = _mm_mul_pd(fscal,dx30);
 378             ty               = _mm_mul_pd(fscal,dy30);
 379             tz               = _mm_mul_pd(fscal,dz30);
 380
 381             /* Update vectorial force */
 382             fix3             = _mm_add_pd(fix3,tx);
 383             fiy3             = _mm_add_pd(fiy3,ty);
 384             fiz3             = _mm_add_pd(fiz3,tz);
 385
 386             fjx0             = _mm_add_pd(fjx0,tx);
 387             fjy0             = _mm_add_pd(fjy0,ty);
 388             fjz0             = _mm_add_pd(fjz0,tz);
 389
 390             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 391
 392             /* Inner loop uses 143 flops */
 393         }
 394
 395         if(jidx<j_index_end)
 396         {
 397
 398             jnrA             = jjnr[jidx];
 399             j_coord_offsetA  = DIM*jnrA;
 400
 401             /* load j atom coordinates */
 402             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 403                                               &jx0,&jy0,&jz0);
 404
 405             /* Calculate displacement vector */
 406             dx00             = _mm_sub_pd(ix0,jx0);
 407             dy00             = _mm_sub_pd(iy0,jy0);
 408             dz00             = _mm_sub_pd(iz0,jz0);
 409             dx10             = _mm_sub_pd(ix1,jx0);
 410             dy10             = _mm_sub_pd(iy1,jy0);
 411             dz10             = _mm_sub_pd(iz1,jz0);
 412             dx20             = _mm_sub_pd(ix2,jx0);
 413             dy20             = _mm_sub_pd(iy2,jy0);
 414             dz20             = _mm_sub_pd(iz2,jz0);
 415             dx30             = _mm_sub_pd(ix3,jx0);
 416             dy30             = _mm_sub_pd(iy3,jy0);
 417             dz30             = _mm_sub_pd(iz3,jz0);
 418
 419             /* Calculate squared distance and things based on it */
 420             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 421             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 422             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 423             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 424
 425             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 426             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 427             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 428             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 429
 430             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 431             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 432             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 433
 434             /* Load parameters for j particles */
 435             jq0              = _mm_load_sd(charge+jnrA+0);
 436             vdwjidx0A        = 2*vdwtype[jnrA+0];
 437
 438             fjx0             = _mm_setzero_pd();
 439             fjy0             = _mm_setzero_pd();
 440             fjz0             = _mm_setzero_pd();
 441
 442             /**************************
 443              * CALCULATE INTERACTIONS *
 444              **************************/
 445
 446             r00              = _mm_mul_pd(rsq00,rinv00);
 447
 448             /* Compute parameters for interactions between i and j atoms */
 449             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 450
 451             /* Calculate table index by multiplying r with table scale and truncate to integer */
 452             rt               = _mm_mul_pd(r00,vftabscale);
 453             vfitab           = _mm_cvttpd_epi32(rt);
 454             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 455             vfitab           = _mm_slli_epi32(vfitab,3);
 456
 457             /* CUBIC SPLINE TABLE DISPERSION */
 458             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 459             F                = _mm_setzero_pd();
 460             GMX_MM_TRANSPOSE2_PD(Y,F);
 461             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 462             H                = _mm_setzero_pd();
 463             GMX_MM_TRANSPOSE2_PD(G,H);
 464             Heps             = _mm_mul_pd(vfeps,H);
 465             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 466             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 467             vvdw6            = _mm_mul_pd(c6_00,VV);
 468             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 469             fvdw6            = _mm_mul_pd(c6_00,FF);
 470
 471             /* CUBIC SPLINE TABLE REPULSION */
 472             vfitab           = _mm_add_epi32(vfitab,ifour);
 473             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 474             F                = _mm_setzero_pd();
 475             GMX_MM_TRANSPOSE2_PD(Y,F);
 476             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 477             H                = _mm_setzero_pd();
 478             GMX_MM_TRANSPOSE2_PD(G,H);
 479             Heps             = _mm_mul_pd(vfeps,H);
 480             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 481             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 482             vvdw12           = _mm_mul_pd(c12_00,VV);
 483             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 484             fvdw12           = _mm_mul_pd(c12_00,FF);
 485             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 486             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 490             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 491
 492             fscal            = fvdw;
 493
 494             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 495
 496             /* Calculate temporary vectorial force */
 497             tx               = _mm_mul_pd(fscal,dx00);
 498             ty               = _mm_mul_pd(fscal,dy00);
 499             tz               = _mm_mul_pd(fscal,dz00);
 500
 501             /* Update vectorial force */
 502             fix0             = _mm_add_pd(fix0,tx);
 503             fiy0             = _mm_add_pd(fiy0,ty);
 504             fiz0             = _mm_add_pd(fiz0,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             qq10             = _mm_mul_pd(iq1,jq0);
 516
 517             /* COULOMB ELECTROSTATICS */
 518             velec            = _mm_mul_pd(qq10,rinv10);
 519             felec            = _mm_mul_pd(velec,rinvsq10);
 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,dx10);
 531             ty               = _mm_mul_pd(fscal,dy10);
 532             tz               = _mm_mul_pd(fscal,dz10);
 533
 534             /* Update vectorial force */
 535             fix1             = _mm_add_pd(fix1,tx);
 536             fiy1             = _mm_add_pd(fiy1,ty);
 537             fiz1             = _mm_add_pd(fiz1,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             /**************************
 544              * CALCULATE INTERACTIONS *
 545              **************************/
 546
 547             /* Compute parameters for interactions between i and j atoms */
 548             qq20             = _mm_mul_pd(iq2,jq0);
 549
 550             /* COULOMB ELECTROSTATICS */
 551             velec            = _mm_mul_pd(qq20,rinv20);
 552             felec            = _mm_mul_pd(velec,rinvsq20);
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 556             velecsum         = _mm_add_pd(velecsum,velec);
 557
 558             fscal            = felec;
 559
 560             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 561
 562             /* Calculate temporary vectorial force */
 563             tx               = _mm_mul_pd(fscal,dx20);
 564             ty               = _mm_mul_pd(fscal,dy20);
 565             tz               = _mm_mul_pd(fscal,dz20);
 566
 567             /* Update vectorial force */
 568             fix2             = _mm_add_pd(fix2,tx);
 569             fiy2             = _mm_add_pd(fiy2,ty);
 570             fiz2             = _mm_add_pd(fiz2,tz);
 571
 572             fjx0             = _mm_add_pd(fjx0,tx);
 573             fjy0             = _mm_add_pd(fjy0,ty);
 574             fjz0             = _mm_add_pd(fjz0,tz);
 575
 576             /**************************
 577              * CALCULATE INTERACTIONS *
 578              **************************/
 579
 580             /* Compute parameters for interactions between i and j atoms */
 581             qq30             = _mm_mul_pd(iq3,jq0);
 582
 583             /* COULOMB ELECTROSTATICS */
 584             velec            = _mm_mul_pd(qq30,rinv30);
 585             felec            = _mm_mul_pd(velec,rinvsq30);
 586
 587             /* Update potential sum for this i atom from the interaction with this j atom. */
 588             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 589             velecsum         = _mm_add_pd(velecsum,velec);
 590
 591             fscal            = felec;
 592
 593             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 594
 595             /* Calculate temporary vectorial force */
 596             tx               = _mm_mul_pd(fscal,dx30);
 597             ty               = _mm_mul_pd(fscal,dy30);
 598             tz               = _mm_mul_pd(fscal,dz30);
 599
 600             /* Update vectorial force */
 601             fix3             = _mm_add_pd(fix3,tx);
 602             fiy3             = _mm_add_pd(fiy3,ty);
 603             fiz3             = _mm_add_pd(fiz3,tz);
 604
 605             fjx0             = _mm_add_pd(fjx0,tx);
 606             fjy0             = _mm_add_pd(fjy0,ty);
 607             fjz0             = _mm_add_pd(fjz0,tz);
 608
 609             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 610
 611             /* Inner loop uses 143 flops */
 612         }
 613
 614         /* End of innermost loop */
 615
 616         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 617                                               f+i_coord_offset,fshift+i_shift_offset);
 618
 619         ggid                        = gid[iidx];
 620         /* Update potential energies */
 621         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 622         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 623
 624         /* Increment number of inner iterations */
 625         inneriter                  += j_index_end - j_index_start;
 626
 627         /* Outer loop uses 26 flops */
 628     }
 629
 630     /* Increment number of outer iterations */
 631     outeriter        += nri;
 632
 633     /* Update outer/inner flops */
 634
 635     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*143);
 636 }
 637 /*
 638  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_double
 639  * Electrostatics interaction: Coulomb
 640  * VdW interaction:            CubicSplineTable
 641  * Geometry:                   Water4-Particle
 642  * Calculate force/pot:        Force
 643  */
 644 void
 645 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_double
 646                     (t_nblist                    * gmx_restrict       nlist,
 647                      rvec                        * gmx_restrict          xx,
 648                      rvec                        * gmx_restrict          ff,
 649                      t_forcerec                  * gmx_restrict          fr,
 650                      t_mdatoms                   * gmx_restrict     mdatoms,
 651                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 652                      t_nrnb                      * gmx_restrict        nrnb)
 653 {
 654     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 655      * just 0 for non-waters.
 656      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 657      * jnr indices corresponding to data put in the four positions in the SIMD register.
 658      */
 659     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 660     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 661     int              jnrA,jnrB;
 662     int              j_coord_offsetA,j_coord_offsetB;
 663     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 664     real             rcutoff_scalar;
 665     real             *shiftvec,*fshift,*x,*f;
 666     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 667     int              vdwioffset0;
 668     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 669     int              vdwioffset1;
 670     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 671     int              vdwioffset2;
 672     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 673     int              vdwioffset3;
 674     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 675     int              vdwjidx0A,vdwjidx0B;
 676     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 677     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 678     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 679     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 680     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 681     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 682     real             *charge;
 683     int              nvdwtype;
 684     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 685     int              *vdwtype;
 686     real             *vdwparam;
 687     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 688     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 689     __m128i          vfitab;
 690     __m128i          ifour       = _mm_set1_epi32(4);
 691     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 692     real             *vftab;
 693     __m128d          dummy_mask,cutoff_mask;
 694     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 695     __m128d          one     = _mm_set1_pd(1.0);
 696     __m128d          two     = _mm_set1_pd(2.0);
 697     x                = xx[0];
 698     f                = ff[0];
 699
 700     nri              = nlist->nri;
 701     iinr             = nlist->iinr;
 702     jindex           = nlist->jindex;
 703     jjnr             = nlist->jjnr;
 704     shiftidx         = nlist->shift;
 705     gid              = nlist->gid;
 706     shiftvec         = fr->shift_vec[0];
 707     fshift           = fr->fshift[0];
 708     facel            = _mm_set1_pd(fr->epsfac);
 709     charge           = mdatoms->chargeA;
 710     nvdwtype         = fr->ntype;
 711     vdwparam         = fr->nbfp;
 712     vdwtype          = mdatoms->typeA;
 713
 714     vftab            = kernel_data->table_vdw->data;
 715     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 716
 717     /* Setup water-specific parameters */
 718     inr              = nlist->iinr[0];
 719     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 720     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 721     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 722     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 723
 724     /* Avoid stupid compiler warnings */
 725     jnrA = jnrB = 0;
 726     j_coord_offsetA = 0;
 727     j_coord_offsetB = 0;
 728
 729     outeriter        = 0;
 730     inneriter        = 0;
 731
 732     /* Start outer loop over neighborlists */
 733     for(iidx=0; iidx<nri; iidx++)
 734     {
 735         /* Load shift vector for this list */
 736         i_shift_offset   = DIM*shiftidx[iidx];
 737
 738         /* Load limits for loop over neighbors */
 739         j_index_start    = jindex[iidx];
 740         j_index_end      = jindex[iidx+1];
 741
 742         /* Get outer coordinate index */
 743         inr              = iinr[iidx];
 744         i_coord_offset   = DIM*inr;
 745
 746         /* Load i particle coords and add shift vector */
 747         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 748                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 749
 750         fix0             = _mm_setzero_pd();
 751         fiy0             = _mm_setzero_pd();
 752         fiz0             = _mm_setzero_pd();
 753         fix1             = _mm_setzero_pd();
 754         fiy1             = _mm_setzero_pd();
 755         fiz1             = _mm_setzero_pd();
 756         fix2             = _mm_setzero_pd();
 757         fiy2             = _mm_setzero_pd();
 758         fiz2             = _mm_setzero_pd();
 759         fix3             = _mm_setzero_pd();
 760         fiy3             = _mm_setzero_pd();
 761         fiz3             = _mm_setzero_pd();
 762
 763         /* Start inner kernel loop */
 764         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 765         {
 766
 767             /* Get j neighbor index, and coordinate index */
 768             jnrA             = jjnr[jidx];
 769             jnrB             = jjnr[jidx+1];
 770             j_coord_offsetA  = DIM*jnrA;
 771             j_coord_offsetB  = DIM*jnrB;
 772
 773             /* load j atom coordinates */
 774             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 775                                               &jx0,&jy0,&jz0);
 776
 777             /* Calculate displacement vector */
 778             dx00             = _mm_sub_pd(ix0,jx0);
 779             dy00             = _mm_sub_pd(iy0,jy0);
 780             dz00             = _mm_sub_pd(iz0,jz0);
 781             dx10             = _mm_sub_pd(ix1,jx0);
 782             dy10             = _mm_sub_pd(iy1,jy0);
 783             dz10             = _mm_sub_pd(iz1,jz0);
 784             dx20             = _mm_sub_pd(ix2,jx0);
 785             dy20             = _mm_sub_pd(iy2,jy0);
 786             dz20             = _mm_sub_pd(iz2,jz0);
 787             dx30             = _mm_sub_pd(ix3,jx0);
 788             dy30             = _mm_sub_pd(iy3,jy0);
 789             dz30             = _mm_sub_pd(iz3,jz0);
 790
 791             /* Calculate squared distance and things based on it */
 792             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 793             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 794             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 795             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 796
 797             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 798             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 799             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 800             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 801
 802             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 803             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 804             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 805
 806             /* Load parameters for j particles */
 807             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 808             vdwjidx0A        = 2*vdwtype[jnrA+0];
 809             vdwjidx0B        = 2*vdwtype[jnrB+0];
 810
 811             fjx0             = _mm_setzero_pd();
 812             fjy0             = _mm_setzero_pd();
 813             fjz0             = _mm_setzero_pd();
 814
 815             /**************************
 816              * CALCULATE INTERACTIONS *
 817              **************************/
 818
 819             r00              = _mm_mul_pd(rsq00,rinv00);
 820
 821             /* Compute parameters for interactions between i and j atoms */
 822             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 823                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 824
 825             /* Calculate table index by multiplying r with table scale and truncate to integer */
 826             rt               = _mm_mul_pd(r00,vftabscale);
 827             vfitab           = _mm_cvttpd_epi32(rt);
 828             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 829             vfitab           = _mm_slli_epi32(vfitab,3);
 830
 831             /* CUBIC SPLINE TABLE DISPERSION */
 832             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 833             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 834             GMX_MM_TRANSPOSE2_PD(Y,F);
 835             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 836             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 837             GMX_MM_TRANSPOSE2_PD(G,H);
 838             Heps             = _mm_mul_pd(vfeps,H);
 839             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 840             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 841             fvdw6            = _mm_mul_pd(c6_00,FF);
 842
 843             /* CUBIC SPLINE TABLE REPULSION */
 844             vfitab           = _mm_add_epi32(vfitab,ifour);
 845             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 846             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 847             GMX_MM_TRANSPOSE2_PD(Y,F);
 848             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 849             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 850             GMX_MM_TRANSPOSE2_PD(G,H);
 851             Heps             = _mm_mul_pd(vfeps,H);
 852             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 853             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 854             fvdw12           = _mm_mul_pd(c12_00,FF);
 855             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 856
 857             fscal            = fvdw;
 858
 859             /* Calculate temporary vectorial force */
 860             tx               = _mm_mul_pd(fscal,dx00);
 861             ty               = _mm_mul_pd(fscal,dy00);
 862             tz               = _mm_mul_pd(fscal,dz00);
 863
 864             /* Update vectorial force */
 865             fix0             = _mm_add_pd(fix0,tx);
 866             fiy0             = _mm_add_pd(fiy0,ty);
 867             fiz0             = _mm_add_pd(fiz0,tz);
 868
 869             fjx0             = _mm_add_pd(fjx0,tx);
 870             fjy0             = _mm_add_pd(fjy0,ty);
 871             fjz0             = _mm_add_pd(fjz0,tz);
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             /* Compute parameters for interactions between i and j atoms */
 878             qq10             = _mm_mul_pd(iq1,jq0);
 879
 880             /* COULOMB ELECTROSTATICS */
 881             velec            = _mm_mul_pd(qq10,rinv10);
 882             felec            = _mm_mul_pd(velec,rinvsq10);
 883
 884             fscal            = felec;
 885
 886             /* Calculate temporary vectorial force */
 887             tx               = _mm_mul_pd(fscal,dx10);
 888             ty               = _mm_mul_pd(fscal,dy10);
 889             tz               = _mm_mul_pd(fscal,dz10);
 890
 891             /* Update vectorial force */
 892             fix1             = _mm_add_pd(fix1,tx);
 893             fiy1             = _mm_add_pd(fiy1,ty);
 894             fiz1             = _mm_add_pd(fiz1,tz);
 895
 896             fjx0             = _mm_add_pd(fjx0,tx);
 897             fjy0             = _mm_add_pd(fjy0,ty);
 898             fjz0             = _mm_add_pd(fjz0,tz);
 899
 900             /**************************
 901              * CALCULATE INTERACTIONS *
 902              **************************/
 903
 904             /* Compute parameters for interactions between i and j atoms */
 905             qq20             = _mm_mul_pd(iq2,jq0);
 906
 907             /* COULOMB ELECTROSTATICS */
 908             velec            = _mm_mul_pd(qq20,rinv20);
 909             felec            = _mm_mul_pd(velec,rinvsq20);
 910
 911             fscal            = felec;
 912
 913             /* Calculate temporary vectorial force */
 914             tx               = _mm_mul_pd(fscal,dx20);
 915             ty               = _mm_mul_pd(fscal,dy20);
 916             tz               = _mm_mul_pd(fscal,dz20);
 917
 918             /* Update vectorial force */
 919             fix2             = _mm_add_pd(fix2,tx);
 920             fiy2             = _mm_add_pd(fiy2,ty);
 921             fiz2             = _mm_add_pd(fiz2,tz);
 922
 923             fjx0             = _mm_add_pd(fjx0,tx);
 924             fjy0             = _mm_add_pd(fjy0,ty);
 925             fjz0             = _mm_add_pd(fjz0,tz);
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             /* Compute parameters for interactions between i and j atoms */
 932             qq30             = _mm_mul_pd(iq3,jq0);
 933
 934             /* COULOMB ELECTROSTATICS */
 935             velec            = _mm_mul_pd(qq30,rinv30);
 936             felec            = _mm_mul_pd(velec,rinvsq30);
 937
 938             fscal            = felec;
 939
 940             /* Calculate temporary vectorial force */
 941             tx               = _mm_mul_pd(fscal,dx30);
 942             ty               = _mm_mul_pd(fscal,dy30);
 943             tz               = _mm_mul_pd(fscal,dz30);
 944
 945             /* Update vectorial force */
 946             fix3             = _mm_add_pd(fix3,tx);
 947             fiy3             = _mm_add_pd(fiy3,ty);
 948             fiz3             = _mm_add_pd(fiz3,tz);
 949
 950             fjx0             = _mm_add_pd(fjx0,tx);
 951             fjy0             = _mm_add_pd(fjy0,ty);
 952             fjz0             = _mm_add_pd(fjz0,tz);
 953
 954             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 955
 956             /* Inner loop uses 132 flops */
 957         }
 958
 959         if(jidx<j_index_end)
 960         {
 961
 962             jnrA             = jjnr[jidx];
 963             j_coord_offsetA  = DIM*jnrA;
 964
 965             /* load j atom coordinates */
 966             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 967                                               &jx0,&jy0,&jz0);
 968
 969             /* Calculate displacement vector */
 970             dx00             = _mm_sub_pd(ix0,jx0);
 971             dy00             = _mm_sub_pd(iy0,jy0);
 972             dz00             = _mm_sub_pd(iz0,jz0);
 973             dx10             = _mm_sub_pd(ix1,jx0);
 974             dy10             = _mm_sub_pd(iy1,jy0);
 975             dz10             = _mm_sub_pd(iz1,jz0);
 976             dx20             = _mm_sub_pd(ix2,jx0);
 977             dy20             = _mm_sub_pd(iy2,jy0);
 978             dz20             = _mm_sub_pd(iz2,jz0);
 979             dx30             = _mm_sub_pd(ix3,jx0);
 980             dy30             = _mm_sub_pd(iy3,jy0);
 981             dz30             = _mm_sub_pd(iz3,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             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 988
 989             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 990             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 991             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 992             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 993
 994             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 995             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 996             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 997
 998             /* Load parameters for j particles */
 999             jq0              = _mm_load_sd(charge+jnrA+0);
1000             vdwjidx0A        = 2*vdwtype[jnrA+0];
1001
1002             fjx0             = _mm_setzero_pd();
1003             fjy0             = _mm_setzero_pd();
1004             fjz0             = _mm_setzero_pd();
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             r00              = _mm_mul_pd(rsq00,rinv00);
1011
1012             /* Compute parameters for interactions between i and j atoms */
1013             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1014
1015             /* Calculate table index by multiplying r with table scale and truncate to integer */
1016             rt               = _mm_mul_pd(r00,vftabscale);
1017             vfitab           = _mm_cvttpd_epi32(rt);
1018             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1019             vfitab           = _mm_slli_epi32(vfitab,3);
1020
1021             /* CUBIC SPLINE TABLE DISPERSION */
1022             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1023             F                = _mm_setzero_pd();
1024             GMX_MM_TRANSPOSE2_PD(Y,F);
1025             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1026             H                = _mm_setzero_pd();
1027             GMX_MM_TRANSPOSE2_PD(G,H);
1028             Heps             = _mm_mul_pd(vfeps,H);
1029             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1030             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1031             fvdw6            = _mm_mul_pd(c6_00,FF);
1032
1033             /* CUBIC SPLINE TABLE REPULSION */
1034             vfitab           = _mm_add_epi32(vfitab,ifour);
1035             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1036             F                = _mm_setzero_pd();
1037             GMX_MM_TRANSPOSE2_PD(Y,F);
1038             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1039             H                = _mm_setzero_pd();
1040             GMX_MM_TRANSPOSE2_PD(G,H);
1041             Heps             = _mm_mul_pd(vfeps,H);
1042             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1043             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1044             fvdw12           = _mm_mul_pd(c12_00,FF);
1045             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1046
1047             fscal            = fvdw;
1048
1049             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1050
1051             /* Calculate temporary vectorial force */
1052             tx               = _mm_mul_pd(fscal,dx00);
1053             ty               = _mm_mul_pd(fscal,dy00);
1054             tz               = _mm_mul_pd(fscal,dz00);
1055
1056             /* Update vectorial force */
1057             fix0             = _mm_add_pd(fix0,tx);
1058             fiy0             = _mm_add_pd(fiy0,ty);
1059             fiz0             = _mm_add_pd(fiz0,tz);
1060
1061             fjx0             = _mm_add_pd(fjx0,tx);
1062             fjy0             = _mm_add_pd(fjy0,ty);
1063             fjz0             = _mm_add_pd(fjz0,tz);
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             /* Compute parameters for interactions between i and j atoms */
1070             qq10             = _mm_mul_pd(iq1,jq0);
1071
1072             /* COULOMB ELECTROSTATICS */
1073             velec            = _mm_mul_pd(qq10,rinv10);
1074             felec            = _mm_mul_pd(velec,rinvsq10);
1075
1076             fscal            = felec;
1077
1078             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm_mul_pd(fscal,dx10);
1082             ty               = _mm_mul_pd(fscal,dy10);
1083             tz               = _mm_mul_pd(fscal,dz10);
1084
1085             /* Update vectorial force */
1086             fix1             = _mm_add_pd(fix1,tx);
1087             fiy1             = _mm_add_pd(fiy1,ty);
1088             fiz1             = _mm_add_pd(fiz1,tz);
1089
1090             fjx0             = _mm_add_pd(fjx0,tx);
1091             fjy0             = _mm_add_pd(fjy0,ty);
1092             fjz0             = _mm_add_pd(fjz0,tz);
1093
1094             /**************************
1095              * CALCULATE INTERACTIONS *
1096              **************************/
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq20             = _mm_mul_pd(iq2,jq0);
1100
1101             /* COULOMB ELECTROSTATICS */
1102             velec            = _mm_mul_pd(qq20,rinv20);
1103             felec            = _mm_mul_pd(velec,rinvsq20);
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm_mul_pd(fscal,dx20);
1111             ty               = _mm_mul_pd(fscal,dy20);
1112             tz               = _mm_mul_pd(fscal,dz20);
1113
1114             /* Update vectorial force */
1115             fix2             = _mm_add_pd(fix2,tx);
1116             fiy2             = _mm_add_pd(fiy2,ty);
1117             fiz2             = _mm_add_pd(fiz2,tz);
1118
1119             fjx0             = _mm_add_pd(fjx0,tx);
1120             fjy0             = _mm_add_pd(fjy0,ty);
1121             fjz0             = _mm_add_pd(fjz0,tz);
1122
1123             /**************************
1124              * CALCULATE INTERACTIONS *
1125              **************************/
1126
1127             /* Compute parameters for interactions between i and j atoms */
1128             qq30             = _mm_mul_pd(iq3,jq0);
1129
1130             /* COULOMB ELECTROSTATICS */
1131             velec            = _mm_mul_pd(qq30,rinv30);
1132             felec            = _mm_mul_pd(velec,rinvsq30);
1133
1134             fscal            = felec;
1135
1136             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1137
1138             /* Calculate temporary vectorial force */
1139             tx               = _mm_mul_pd(fscal,dx30);
1140             ty               = _mm_mul_pd(fscal,dy30);
1141             tz               = _mm_mul_pd(fscal,dz30);
1142
1143             /* Update vectorial force */
1144             fix3             = _mm_add_pd(fix3,tx);
1145             fiy3             = _mm_add_pd(fiy3,ty);
1146             fiz3             = _mm_add_pd(fiz3,tz);
1147
1148             fjx0             = _mm_add_pd(fjx0,tx);
1149             fjy0             = _mm_add_pd(fjy0,ty);
1150             fjz0             = _mm_add_pd(fjz0,tz);
1151
1152             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1153
1154             /* Inner loop uses 132 flops */
1155         }
1156
1157         /* End of innermost loop */
1158
1159         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1160                                               f+i_coord_offset,fshift+i_shift_offset);
1161
1162         /* Increment number of inner iterations */
1163         inneriter                  += j_index_end - j_index_start;
1164
1165         /* Outer loop uses 24 flops */
1166     }
1167
1168     /* Increment number of outer iterations */
1169     outeriter        += nri;
1170
1171     /* Update outer/inner flops */
1172
1173     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1174 }