src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_sse4_1_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse4_1_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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->ic->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     vftab            = kernel_data->table_vdw->data;
 127     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 132     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 133     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* Avoid stupid compiler warnings */
 137     jnrA = jnrB = 0;
 138     j_coord_offsetA = 0;
 139     j_coord_offsetB = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 161
 162         fix0             = _mm_setzero_pd();
 163         fiy0             = _mm_setzero_pd();
 164         fiz0             = _mm_setzero_pd();
 165         fix1             = _mm_setzero_pd();
 166         fiy1             = _mm_setzero_pd();
 167         fiz1             = _mm_setzero_pd();
 168         fix2             = _mm_setzero_pd();
 169         fiy2             = _mm_setzero_pd();
 170         fiz2             = _mm_setzero_pd();
 171         fix3             = _mm_setzero_pd();
 172         fiy3             = _mm_setzero_pd();
 173         fiz3             = _mm_setzero_pd();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_pd();
 177         vvdwsum          = _mm_setzero_pd();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188
 189             /* load j atom coordinates */
 190             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                               &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm_sub_pd(ix0,jx0);
 195             dy00             = _mm_sub_pd(iy0,jy0);
 196             dz00             = _mm_sub_pd(iz0,jz0);
 197             dx10             = _mm_sub_pd(ix1,jx0);
 198             dy10             = _mm_sub_pd(iy1,jy0);
 199             dz10             = _mm_sub_pd(iz1,jz0);
 200             dx20             = _mm_sub_pd(ix2,jx0);
 201             dy20             = _mm_sub_pd(iy2,jy0);
 202             dz20             = _mm_sub_pd(iz2,jz0);
 203             dx30             = _mm_sub_pd(ix3,jx0);
 204             dy30             = _mm_sub_pd(iy3,jy0);
 205             dz30             = _mm_sub_pd(iz3,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 209             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 210             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 211             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 212
 213             rinv00           = sse41_invsqrt_d(rsq00);
 214             rinv10           = sse41_invsqrt_d(rsq10);
 215             rinv20           = sse41_invsqrt_d(rsq20);
 216             rinv30           = sse41_invsqrt_d(rsq30);
 217
 218             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 219             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 220             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 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             r00              = _mm_mul_pd(rsq00,rinv00);
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 240
 241             /* Calculate table index by multiplying r with table scale and truncate to integer */
 242             rt               = _mm_mul_pd(r00,vftabscale);
 243             vfitab           = _mm_cvttpd_epi32(rt);
 244             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 245             vfitab           = _mm_slli_epi32(vfitab,3);
 246
 247             /* CUBIC SPLINE TABLE DISPERSION */
 248             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 249             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 250             GMX_MM_TRANSPOSE2_PD(Y,F);
 251             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 252             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 253             GMX_MM_TRANSPOSE2_PD(G,H);
 254             Heps             = _mm_mul_pd(vfeps,H);
 255             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 256             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 257             vvdw6            = _mm_mul_pd(c6_00,VV);
 258             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 259             fvdw6            = _mm_mul_pd(c6_00,FF);
 260
 261             /* CUBIC SPLINE TABLE REPULSION */
 262             vfitab           = _mm_add_epi32(vfitab,ifour);
 263             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 264             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 265             GMX_MM_TRANSPOSE2_PD(Y,F);
 266             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 267             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 268             GMX_MM_TRANSPOSE2_PD(G,H);
 269             Heps             = _mm_mul_pd(vfeps,H);
 270             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 271             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 272             vvdw12           = _mm_mul_pd(c12_00,VV);
 273             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 274             fvdw12           = _mm_mul_pd(c12_00,FF);
 275             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 276             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 277
 278             /* Update potential sum for this i atom from the interaction with this j atom. */
 279             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 280
 281             fscal            = fvdw;
 282
 283             /* Calculate temporary vectorial force */
 284             tx               = _mm_mul_pd(fscal,dx00);
 285             ty               = _mm_mul_pd(fscal,dy00);
 286             tz               = _mm_mul_pd(fscal,dz00);
 287
 288             /* Update vectorial force */
 289             fix0             = _mm_add_pd(fix0,tx);
 290             fiy0             = _mm_add_pd(fiy0,ty);
 291             fiz0             = _mm_add_pd(fiz0,tz);
 292
 293             fjx0             = _mm_add_pd(fjx0,tx);
 294             fjy0             = _mm_add_pd(fjy0,ty);
 295             fjz0             = _mm_add_pd(fjz0,tz);
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq10             = _mm_mul_pd(iq1,jq0);
 303
 304             /* COULOMB ELECTROSTATICS */
 305             velec            = _mm_mul_pd(qq10,rinv10);
 306             felec            = _mm_mul_pd(velec,rinvsq10);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             velecsum         = _mm_add_pd(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             /* Calculate temporary vectorial force */
 314             tx               = _mm_mul_pd(fscal,dx10);
 315             ty               = _mm_mul_pd(fscal,dy10);
 316             tz               = _mm_mul_pd(fscal,dz10);
 317
 318             /* Update vectorial force */
 319             fix1             = _mm_add_pd(fix1,tx);
 320             fiy1             = _mm_add_pd(fiy1,ty);
 321             fiz1             = _mm_add_pd(fiz1,tz);
 322
 323             fjx0             = _mm_add_pd(fjx0,tx);
 324             fjy0             = _mm_add_pd(fjy0,ty);
 325             fjz0             = _mm_add_pd(fjz0,tz);
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq20             = _mm_mul_pd(iq2,jq0);
 333
 334             /* COULOMB ELECTROSTATICS */
 335             velec            = _mm_mul_pd(qq20,rinv20);
 336             felec            = _mm_mul_pd(velec,rinvsq20);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velecsum         = _mm_add_pd(velecsum,velec);
 340
 341             fscal            = felec;
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm_mul_pd(fscal,dx20);
 345             ty               = _mm_mul_pd(fscal,dy20);
 346             tz               = _mm_mul_pd(fscal,dz20);
 347
 348             /* Update vectorial force */
 349             fix2             = _mm_add_pd(fix2,tx);
 350             fiy2             = _mm_add_pd(fiy2,ty);
 351             fiz2             = _mm_add_pd(fiz2,tz);
 352
 353             fjx0             = _mm_add_pd(fjx0,tx);
 354             fjy0             = _mm_add_pd(fjy0,ty);
 355             fjz0             = _mm_add_pd(fjz0,tz);
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq30             = _mm_mul_pd(iq3,jq0);
 363
 364             /* COULOMB ELECTROSTATICS */
 365             velec            = _mm_mul_pd(qq30,rinv30);
 366             felec            = _mm_mul_pd(velec,rinvsq30);
 367
 368             /* Update potential sum for this i atom from the interaction with this j atom. */
 369             velecsum         = _mm_add_pd(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             /* Calculate temporary vectorial force */
 374             tx               = _mm_mul_pd(fscal,dx30);
 375             ty               = _mm_mul_pd(fscal,dy30);
 376             tz               = _mm_mul_pd(fscal,dz30);
 377
 378             /* Update vectorial force */
 379             fix3             = _mm_add_pd(fix3,tx);
 380             fiy3             = _mm_add_pd(fiy3,ty);
 381             fiz3             = _mm_add_pd(fiz3,tz);
 382
 383             fjx0             = _mm_add_pd(fjx0,tx);
 384             fjy0             = _mm_add_pd(fjy0,ty);
 385             fjz0             = _mm_add_pd(fjz0,tz);
 386
 387             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 388
 389             /* Inner loop uses 143 flops */
 390         }
 391
 392         if(jidx<j_index_end)
 393         {
 394
 395             jnrA             = jjnr[jidx];
 396             j_coord_offsetA  = DIM*jnrA;
 397
 398             /* load j atom coordinates */
 399             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 400                                               &jx0,&jy0,&jz0);
 401
 402             /* Calculate displacement vector */
 403             dx00             = _mm_sub_pd(ix0,jx0);
 404             dy00             = _mm_sub_pd(iy0,jy0);
 405             dz00             = _mm_sub_pd(iz0,jz0);
 406             dx10             = _mm_sub_pd(ix1,jx0);
 407             dy10             = _mm_sub_pd(iy1,jy0);
 408             dz10             = _mm_sub_pd(iz1,jz0);
 409             dx20             = _mm_sub_pd(ix2,jx0);
 410             dy20             = _mm_sub_pd(iy2,jy0);
 411             dz20             = _mm_sub_pd(iz2,jz0);
 412             dx30             = _mm_sub_pd(ix3,jx0);
 413             dy30             = _mm_sub_pd(iy3,jy0);
 414             dz30             = _mm_sub_pd(iz3,jz0);
 415
 416             /* Calculate squared distance and things based on it */
 417             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 418             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 419             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 420             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 421
 422             rinv00           = sse41_invsqrt_d(rsq00);
 423             rinv10           = sse41_invsqrt_d(rsq10);
 424             rinv20           = sse41_invsqrt_d(rsq20);
 425             rinv30           = sse41_invsqrt_d(rsq30);
 426
 427             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 428             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 429             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 430
 431             /* Load parameters for j particles */
 432             jq0              = _mm_load_sd(charge+jnrA+0);
 433             vdwjidx0A        = 2*vdwtype[jnrA+0];
 434
 435             fjx0             = _mm_setzero_pd();
 436             fjy0             = _mm_setzero_pd();
 437             fjz0             = _mm_setzero_pd();
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             r00              = _mm_mul_pd(rsq00,rinv00);
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 447
 448             /* Calculate table index by multiplying r with table scale and truncate to integer */
 449             rt               = _mm_mul_pd(r00,vftabscale);
 450             vfitab           = _mm_cvttpd_epi32(rt);
 451             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 452             vfitab           = _mm_slli_epi32(vfitab,3);
 453
 454             /* CUBIC SPLINE TABLE DISPERSION */
 455             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 456             F                = _mm_setzero_pd();
 457             GMX_MM_TRANSPOSE2_PD(Y,F);
 458             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 459             H                = _mm_setzero_pd();
 460             GMX_MM_TRANSPOSE2_PD(G,H);
 461             Heps             = _mm_mul_pd(vfeps,H);
 462             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 463             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 464             vvdw6            = _mm_mul_pd(c6_00,VV);
 465             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 466             fvdw6            = _mm_mul_pd(c6_00,FF);
 467
 468             /* CUBIC SPLINE TABLE REPULSION */
 469             vfitab           = _mm_add_epi32(vfitab,ifour);
 470             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 471             F                = _mm_setzero_pd();
 472             GMX_MM_TRANSPOSE2_PD(Y,F);
 473             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 474             H                = _mm_setzero_pd();
 475             GMX_MM_TRANSPOSE2_PD(G,H);
 476             Heps             = _mm_mul_pd(vfeps,H);
 477             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 478             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 479             vvdw12           = _mm_mul_pd(c12_00,VV);
 480             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 481             fvdw12           = _mm_mul_pd(c12_00,FF);
 482             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 483             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 484
 485             /* Update potential sum for this i atom from the interaction with this j atom. */
 486             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 487             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 488
 489             fscal            = fvdw;
 490
 491             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 492
 493             /* Calculate temporary vectorial force */
 494             tx               = _mm_mul_pd(fscal,dx00);
 495             ty               = _mm_mul_pd(fscal,dy00);
 496             tz               = _mm_mul_pd(fscal,dz00);
 497
 498             /* Update vectorial force */
 499             fix0             = _mm_add_pd(fix0,tx);
 500             fiy0             = _mm_add_pd(fiy0,ty);
 501             fiz0             = _mm_add_pd(fiz0,tz);
 502
 503             fjx0             = _mm_add_pd(fjx0,tx);
 504             fjy0             = _mm_add_pd(fjy0,ty);
 505             fjz0             = _mm_add_pd(fjz0,tz);
 506
 507             /**************************
 508              * CALCULATE INTERACTIONS *
 509              **************************/
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             qq10             = _mm_mul_pd(iq1,jq0);
 513
 514             /* COULOMB ELECTROSTATICS */
 515             velec            = _mm_mul_pd(qq10,rinv10);
 516             felec            = _mm_mul_pd(velec,rinvsq10);
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 520             velecsum         = _mm_add_pd(velecsum,velec);
 521
 522             fscal            = felec;
 523
 524             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 525
 526             /* Calculate temporary vectorial force */
 527             tx               = _mm_mul_pd(fscal,dx10);
 528             ty               = _mm_mul_pd(fscal,dy10);
 529             tz               = _mm_mul_pd(fscal,dz10);
 530
 531             /* Update vectorial force */
 532             fix1             = _mm_add_pd(fix1,tx);
 533             fiy1             = _mm_add_pd(fiy1,ty);
 534             fiz1             = _mm_add_pd(fiz1,tz);
 535
 536             fjx0             = _mm_add_pd(fjx0,tx);
 537             fjy0             = _mm_add_pd(fjy0,ty);
 538             fjz0             = _mm_add_pd(fjz0,tz);
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             /* Compute parameters for interactions between i and j atoms */
 545             qq20             = _mm_mul_pd(iq2,jq0);
 546
 547             /* COULOMB ELECTROSTATICS */
 548             velec            = _mm_mul_pd(qq20,rinv20);
 549             felec            = _mm_mul_pd(velec,rinvsq20);
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 553             velecsum         = _mm_add_pd(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm_mul_pd(fscal,dx20);
 561             ty               = _mm_mul_pd(fscal,dy20);
 562             tz               = _mm_mul_pd(fscal,dz20);
 563
 564             /* Update vectorial force */
 565             fix2             = _mm_add_pd(fix2,tx);
 566             fiy2             = _mm_add_pd(fiy2,ty);
 567             fiz2             = _mm_add_pd(fiz2,tz);
 568
 569             fjx0             = _mm_add_pd(fjx0,tx);
 570             fjy0             = _mm_add_pd(fjy0,ty);
 571             fjz0             = _mm_add_pd(fjz0,tz);
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             /* Compute parameters for interactions between i and j atoms */
 578             qq30             = _mm_mul_pd(iq3,jq0);
 579
 580             /* COULOMB ELECTROSTATICS */
 581             velec            = _mm_mul_pd(qq30,rinv30);
 582             felec            = _mm_mul_pd(velec,rinvsq30);
 583
 584             /* Update potential sum for this i atom from the interaction with this j atom. */
 585             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 586             velecsum         = _mm_add_pd(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 591
 592             /* Calculate temporary vectorial force */
 593             tx               = _mm_mul_pd(fscal,dx30);
 594             ty               = _mm_mul_pd(fscal,dy30);
 595             tz               = _mm_mul_pd(fscal,dz30);
 596
 597             /* Update vectorial force */
 598             fix3             = _mm_add_pd(fix3,tx);
 599             fiy3             = _mm_add_pd(fiy3,ty);
 600             fiz3             = _mm_add_pd(fiz3,tz);
 601
 602             fjx0             = _mm_add_pd(fjx0,tx);
 603             fjy0             = _mm_add_pd(fjy0,ty);
 604             fjz0             = _mm_add_pd(fjz0,tz);
 605
 606             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 607
 608             /* Inner loop uses 143 flops */
 609         }
 610
 611         /* End of innermost loop */
 612
 613         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 614                                               f+i_coord_offset,fshift+i_shift_offset);
 615
 616         ggid                        = gid[iidx];
 617         /* Update potential energies */
 618         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 619         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 26 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*143);
 633 }
 634 /*
 635  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_double
 636  * Electrostatics interaction: Coulomb
 637  * VdW interaction:            CubicSplineTable
 638  * Geometry:                   Water4-Particle
 639  * Calculate force/pot:        Force
 640  */
 641 void
 642 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse4_1_double
 643                     (t_nblist                    * gmx_restrict       nlist,
 644                      rvec                        * gmx_restrict          xx,
 645                      rvec                        * gmx_restrict          ff,
 646                      struct t_forcerec           * gmx_restrict          fr,
 647                      t_mdatoms                   * gmx_restrict     mdatoms,
 648                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 649                      t_nrnb                      * gmx_restrict        nrnb)
 650 {
 651     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 652      * just 0 for non-waters.
 653      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 654      * jnr indices corresponding to data put in the four positions in the SIMD register.
 655      */
 656     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 657     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 658     int              jnrA,jnrB;
 659     int              j_coord_offsetA,j_coord_offsetB;
 660     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 661     real             rcutoff_scalar;
 662     real             *shiftvec,*fshift,*x,*f;
 663     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 664     int              vdwioffset0;
 665     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 666     int              vdwioffset1;
 667     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 668     int              vdwioffset2;
 669     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 670     int              vdwioffset3;
 671     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 672     int              vdwjidx0A,vdwjidx0B;
 673     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 674     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 675     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 676     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 677     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 678     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 679     real             *charge;
 680     int              nvdwtype;
 681     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 682     int              *vdwtype;
 683     real             *vdwparam;
 684     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 685     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 686     __m128i          vfitab;
 687     __m128i          ifour       = _mm_set1_epi32(4);
 688     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 689     real             *vftab;
 690     __m128d          dummy_mask,cutoff_mask;
 691     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 692     __m128d          one     = _mm_set1_pd(1.0);
 693     __m128d          two     = _mm_set1_pd(2.0);
 694     x                = xx[0];
 695     f                = ff[0];
 696
 697     nri              = nlist->nri;
 698     iinr             = nlist->iinr;
 699     jindex           = nlist->jindex;
 700     jjnr             = nlist->jjnr;
 701     shiftidx         = nlist->shift;
 702     gid              = nlist->gid;
 703     shiftvec         = fr->shift_vec[0];
 704     fshift           = fr->fshift[0];
 705     facel            = _mm_set1_pd(fr->ic->epsfac);
 706     charge           = mdatoms->chargeA;
 707     nvdwtype         = fr->ntype;
 708     vdwparam         = fr->nbfp;
 709     vdwtype          = mdatoms->typeA;
 710
 711     vftab            = kernel_data->table_vdw->data;
 712     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 713
 714     /* Setup water-specific parameters */
 715     inr              = nlist->iinr[0];
 716     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 717     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 718     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 719     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 720
 721     /* Avoid stupid compiler warnings */
 722     jnrA = jnrB = 0;
 723     j_coord_offsetA = 0;
 724     j_coord_offsetB = 0;
 725
 726     outeriter        = 0;
 727     inneriter        = 0;
 728
 729     /* Start outer loop over neighborlists */
 730     for(iidx=0; iidx<nri; iidx++)
 731     {
 732         /* Load shift vector for this list */
 733         i_shift_offset   = DIM*shiftidx[iidx];
 734
 735         /* Load limits for loop over neighbors */
 736         j_index_start    = jindex[iidx];
 737         j_index_end      = jindex[iidx+1];
 738
 739         /* Get outer coordinate index */
 740         inr              = iinr[iidx];
 741         i_coord_offset   = DIM*inr;
 742
 743         /* Load i particle coords and add shift vector */
 744         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 745                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 746
 747         fix0             = _mm_setzero_pd();
 748         fiy0             = _mm_setzero_pd();
 749         fiz0             = _mm_setzero_pd();
 750         fix1             = _mm_setzero_pd();
 751         fiy1             = _mm_setzero_pd();
 752         fiz1             = _mm_setzero_pd();
 753         fix2             = _mm_setzero_pd();
 754         fiy2             = _mm_setzero_pd();
 755         fiz2             = _mm_setzero_pd();
 756         fix3             = _mm_setzero_pd();
 757         fiy3             = _mm_setzero_pd();
 758         fiz3             = _mm_setzero_pd();
 759
 760         /* Start inner kernel loop */
 761         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 762         {
 763
 764             /* Get j neighbor index, and coordinate index */
 765             jnrA             = jjnr[jidx];
 766             jnrB             = jjnr[jidx+1];
 767             j_coord_offsetA  = DIM*jnrA;
 768             j_coord_offsetB  = DIM*jnrB;
 769
 770             /* load j atom coordinates */
 771             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 772                                               &jx0,&jy0,&jz0);
 773
 774             /* Calculate displacement vector */
 775             dx00             = _mm_sub_pd(ix0,jx0);
 776             dy00             = _mm_sub_pd(iy0,jy0);
 777             dz00             = _mm_sub_pd(iz0,jz0);
 778             dx10             = _mm_sub_pd(ix1,jx0);
 779             dy10             = _mm_sub_pd(iy1,jy0);
 780             dz10             = _mm_sub_pd(iz1,jz0);
 781             dx20             = _mm_sub_pd(ix2,jx0);
 782             dy20             = _mm_sub_pd(iy2,jy0);
 783             dz20             = _mm_sub_pd(iz2,jz0);
 784             dx30             = _mm_sub_pd(ix3,jx0);
 785             dy30             = _mm_sub_pd(iy3,jy0);
 786             dz30             = _mm_sub_pd(iz3,jz0);
 787
 788             /* Calculate squared distance and things based on it */
 789             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 790             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 791             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 792             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 793
 794             rinv00           = sse41_invsqrt_d(rsq00);
 795             rinv10           = sse41_invsqrt_d(rsq10);
 796             rinv20           = sse41_invsqrt_d(rsq20);
 797             rinv30           = sse41_invsqrt_d(rsq30);
 798
 799             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 800             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 801             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 802
 803             /* Load parameters for j particles */
 804             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 805             vdwjidx0A        = 2*vdwtype[jnrA+0];
 806             vdwjidx0B        = 2*vdwtype[jnrB+0];
 807
 808             fjx0             = _mm_setzero_pd();
 809             fjy0             = _mm_setzero_pd();
 810             fjz0             = _mm_setzero_pd();
 811
 812             /**************************
 813              * CALCULATE INTERACTIONS *
 814              **************************/
 815
 816             r00              = _mm_mul_pd(rsq00,rinv00);
 817
 818             /* Compute parameters for interactions between i and j atoms */
 819             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 820                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 821
 822             /* Calculate table index by multiplying r with table scale and truncate to integer */
 823             rt               = _mm_mul_pd(r00,vftabscale);
 824             vfitab           = _mm_cvttpd_epi32(rt);
 825             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 826             vfitab           = _mm_slli_epi32(vfitab,3);
 827
 828             /* CUBIC SPLINE TABLE DISPERSION */
 829             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 830             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 831             GMX_MM_TRANSPOSE2_PD(Y,F);
 832             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 833             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 834             GMX_MM_TRANSPOSE2_PD(G,H);
 835             Heps             = _mm_mul_pd(vfeps,H);
 836             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 837             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 838             fvdw6            = _mm_mul_pd(c6_00,FF);
 839
 840             /* CUBIC SPLINE TABLE REPULSION */
 841             vfitab           = _mm_add_epi32(vfitab,ifour);
 842             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 843             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 844             GMX_MM_TRANSPOSE2_PD(Y,F);
 845             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 846             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 847             GMX_MM_TRANSPOSE2_PD(G,H);
 848             Heps             = _mm_mul_pd(vfeps,H);
 849             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 850             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 851             fvdw12           = _mm_mul_pd(c12_00,FF);
 852             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 853
 854             fscal            = fvdw;
 855
 856             /* Calculate temporary vectorial force */
 857             tx               = _mm_mul_pd(fscal,dx00);
 858             ty               = _mm_mul_pd(fscal,dy00);
 859             tz               = _mm_mul_pd(fscal,dz00);
 860
 861             /* Update vectorial force */
 862             fix0             = _mm_add_pd(fix0,tx);
 863             fiy0             = _mm_add_pd(fiy0,ty);
 864             fiz0             = _mm_add_pd(fiz0,tz);
 865
 866             fjx0             = _mm_add_pd(fjx0,tx);
 867             fjy0             = _mm_add_pd(fjy0,ty);
 868             fjz0             = _mm_add_pd(fjz0,tz);
 869
 870             /**************************
 871              * CALCULATE INTERACTIONS *
 872              **************************/
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq10             = _mm_mul_pd(iq1,jq0);
 876
 877             /* COULOMB ELECTROSTATICS */
 878             velec            = _mm_mul_pd(qq10,rinv10);
 879             felec            = _mm_mul_pd(velec,rinvsq10);
 880
 881             fscal            = felec;
 882
 883             /* Calculate temporary vectorial force */
 884             tx               = _mm_mul_pd(fscal,dx10);
 885             ty               = _mm_mul_pd(fscal,dy10);
 886             tz               = _mm_mul_pd(fscal,dz10);
 887
 888             /* Update vectorial force */
 889             fix1             = _mm_add_pd(fix1,tx);
 890             fiy1             = _mm_add_pd(fiy1,ty);
 891             fiz1             = _mm_add_pd(fiz1,tz);
 892
 893             fjx0             = _mm_add_pd(fjx0,tx);
 894             fjy0             = _mm_add_pd(fjy0,ty);
 895             fjz0             = _mm_add_pd(fjz0,tz);
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq20             = _mm_mul_pd(iq2,jq0);
 903
 904             /* COULOMB ELECTROSTATICS */
 905             velec            = _mm_mul_pd(qq20,rinv20);
 906             felec            = _mm_mul_pd(velec,rinvsq20);
 907
 908             fscal            = felec;
 909
 910             /* Calculate temporary vectorial force */
 911             tx               = _mm_mul_pd(fscal,dx20);
 912             ty               = _mm_mul_pd(fscal,dy20);
 913             tz               = _mm_mul_pd(fscal,dz20);
 914
 915             /* Update vectorial force */
 916             fix2             = _mm_add_pd(fix2,tx);
 917             fiy2             = _mm_add_pd(fiy2,ty);
 918             fiz2             = _mm_add_pd(fiz2,tz);
 919
 920             fjx0             = _mm_add_pd(fjx0,tx);
 921             fjy0             = _mm_add_pd(fjy0,ty);
 922             fjz0             = _mm_add_pd(fjz0,tz);
 923
 924             /**************************
 925              * CALCULATE INTERACTIONS *
 926              **************************/
 927
 928             /* Compute parameters for interactions between i and j atoms */
 929             qq30             = _mm_mul_pd(iq3,jq0);
 930
 931             /* COULOMB ELECTROSTATICS */
 932             velec            = _mm_mul_pd(qq30,rinv30);
 933             felec            = _mm_mul_pd(velec,rinvsq30);
 934
 935             fscal            = felec;
 936
 937             /* Calculate temporary vectorial force */
 938             tx               = _mm_mul_pd(fscal,dx30);
 939             ty               = _mm_mul_pd(fscal,dy30);
 940             tz               = _mm_mul_pd(fscal,dz30);
 941
 942             /* Update vectorial force */
 943             fix3             = _mm_add_pd(fix3,tx);
 944             fiy3             = _mm_add_pd(fiy3,ty);
 945             fiz3             = _mm_add_pd(fiz3,tz);
 946
 947             fjx0             = _mm_add_pd(fjx0,tx);
 948             fjy0             = _mm_add_pd(fjy0,ty);
 949             fjz0             = _mm_add_pd(fjz0,tz);
 950
 951             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 952
 953             /* Inner loop uses 132 flops */
 954         }
 955
 956         if(jidx<j_index_end)
 957         {
 958
 959             jnrA             = jjnr[jidx];
 960             j_coord_offsetA  = DIM*jnrA;
 961
 962             /* load j atom coordinates */
 963             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 964                                               &jx0,&jy0,&jz0);
 965
 966             /* Calculate displacement vector */
 967             dx00             = _mm_sub_pd(ix0,jx0);
 968             dy00             = _mm_sub_pd(iy0,jy0);
 969             dz00             = _mm_sub_pd(iz0,jz0);
 970             dx10             = _mm_sub_pd(ix1,jx0);
 971             dy10             = _mm_sub_pd(iy1,jy0);
 972             dz10             = _mm_sub_pd(iz1,jz0);
 973             dx20             = _mm_sub_pd(ix2,jx0);
 974             dy20             = _mm_sub_pd(iy2,jy0);
 975             dz20             = _mm_sub_pd(iz2,jz0);
 976             dx30             = _mm_sub_pd(ix3,jx0);
 977             dy30             = _mm_sub_pd(iy3,jy0);
 978             dz30             = _mm_sub_pd(iz3,jz0);
 979
 980             /* Calculate squared distance and things based on it */
 981             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 982             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 983             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 984             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 985
 986             rinv00           = sse41_invsqrt_d(rsq00);
 987             rinv10           = sse41_invsqrt_d(rsq10);
 988             rinv20           = sse41_invsqrt_d(rsq20);
 989             rinv30           = sse41_invsqrt_d(rsq30);
 990
 991             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 992             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 993             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 994
 995             /* Load parameters for j particles */
 996             jq0              = _mm_load_sd(charge+jnrA+0);
 997             vdwjidx0A        = 2*vdwtype[jnrA+0];
 998
 999             fjx0             = _mm_setzero_pd();
1000             fjy0             = _mm_setzero_pd();
1001             fjz0             = _mm_setzero_pd();
1002
1003             /**************************
1004              * CALCULATE INTERACTIONS *
1005              **************************/
1006
1007             r00              = _mm_mul_pd(rsq00,rinv00);
1008
1009             /* Compute parameters for interactions between i and j atoms */
1010             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1011
1012             /* Calculate table index by multiplying r with table scale and truncate to integer */
1013             rt               = _mm_mul_pd(r00,vftabscale);
1014             vfitab           = _mm_cvttpd_epi32(rt);
1015             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1016             vfitab           = _mm_slli_epi32(vfitab,3);
1017
1018             /* CUBIC SPLINE TABLE DISPERSION */
1019             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1020             F                = _mm_setzero_pd();
1021             GMX_MM_TRANSPOSE2_PD(Y,F);
1022             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1023             H                = _mm_setzero_pd();
1024             GMX_MM_TRANSPOSE2_PD(G,H);
1025             Heps             = _mm_mul_pd(vfeps,H);
1026             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1027             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1028             fvdw6            = _mm_mul_pd(c6_00,FF);
1029
1030             /* CUBIC SPLINE TABLE REPULSION */
1031             vfitab           = _mm_add_epi32(vfitab,ifour);
1032             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1033             F                = _mm_setzero_pd();
1034             GMX_MM_TRANSPOSE2_PD(Y,F);
1035             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1036             H                = _mm_setzero_pd();
1037             GMX_MM_TRANSPOSE2_PD(G,H);
1038             Heps             = _mm_mul_pd(vfeps,H);
1039             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1040             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1041             fvdw12           = _mm_mul_pd(c12_00,FF);
1042             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1043
1044             fscal            = fvdw;
1045
1046             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1047
1048             /* Calculate temporary vectorial force */
1049             tx               = _mm_mul_pd(fscal,dx00);
1050             ty               = _mm_mul_pd(fscal,dy00);
1051             tz               = _mm_mul_pd(fscal,dz00);
1052
1053             /* Update vectorial force */
1054             fix0             = _mm_add_pd(fix0,tx);
1055             fiy0             = _mm_add_pd(fiy0,ty);
1056             fiz0             = _mm_add_pd(fiz0,tz);
1057
1058             fjx0             = _mm_add_pd(fjx0,tx);
1059             fjy0             = _mm_add_pd(fjy0,ty);
1060             fjz0             = _mm_add_pd(fjz0,tz);
1061
1062             /**************************
1063              * CALCULATE INTERACTIONS *
1064              **************************/
1065
1066             /* Compute parameters for interactions between i and j atoms */
1067             qq10             = _mm_mul_pd(iq1,jq0);
1068
1069             /* COULOMB ELECTROSTATICS */
1070             velec            = _mm_mul_pd(qq10,rinv10);
1071             felec            = _mm_mul_pd(velec,rinvsq10);
1072
1073             fscal            = felec;
1074
1075             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1076
1077             /* Calculate temporary vectorial force */
1078             tx               = _mm_mul_pd(fscal,dx10);
1079             ty               = _mm_mul_pd(fscal,dy10);
1080             tz               = _mm_mul_pd(fscal,dz10);
1081
1082             /* Update vectorial force */
1083             fix1             = _mm_add_pd(fix1,tx);
1084             fiy1             = _mm_add_pd(fiy1,ty);
1085             fiz1             = _mm_add_pd(fiz1,tz);
1086
1087             fjx0             = _mm_add_pd(fjx0,tx);
1088             fjy0             = _mm_add_pd(fjy0,ty);
1089             fjz0             = _mm_add_pd(fjz0,tz);
1090
1091             /**************************
1092              * CALCULATE INTERACTIONS *
1093              **************************/
1094
1095             /* Compute parameters for interactions between i and j atoms */
1096             qq20             = _mm_mul_pd(iq2,jq0);
1097
1098             /* COULOMB ELECTROSTATICS */
1099             velec            = _mm_mul_pd(qq20,rinv20);
1100             felec            = _mm_mul_pd(velec,rinvsq20);
1101
1102             fscal            = felec;
1103
1104             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1105
1106             /* Calculate temporary vectorial force */
1107             tx               = _mm_mul_pd(fscal,dx20);
1108             ty               = _mm_mul_pd(fscal,dy20);
1109             tz               = _mm_mul_pd(fscal,dz20);
1110
1111             /* Update vectorial force */
1112             fix2             = _mm_add_pd(fix2,tx);
1113             fiy2             = _mm_add_pd(fiy2,ty);
1114             fiz2             = _mm_add_pd(fiz2,tz);
1115
1116             fjx0             = _mm_add_pd(fjx0,tx);
1117             fjy0             = _mm_add_pd(fjy0,ty);
1118             fjz0             = _mm_add_pd(fjz0,tz);
1119
1120             /**************************
1121              * CALCULATE INTERACTIONS *
1122              **************************/
1123
1124             /* Compute parameters for interactions between i and j atoms */
1125             qq30             = _mm_mul_pd(iq3,jq0);
1126
1127             /* COULOMB ELECTROSTATICS */
1128             velec            = _mm_mul_pd(qq30,rinv30);
1129             felec            = _mm_mul_pd(velec,rinvsq30);
1130
1131             fscal            = felec;
1132
1133             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1134
1135             /* Calculate temporary vectorial force */
1136             tx               = _mm_mul_pd(fscal,dx30);
1137             ty               = _mm_mul_pd(fscal,dy30);
1138             tz               = _mm_mul_pd(fscal,dz30);
1139
1140             /* Update vectorial force */
1141             fix3             = _mm_add_pd(fix3,tx);
1142             fiy3             = _mm_add_pd(fiy3,ty);
1143             fiz3             = _mm_add_pd(fiz3,tz);
1144
1145             fjx0             = _mm_add_pd(fjx0,tx);
1146             fjy0             = _mm_add_pd(fjy0,ty);
1147             fjz0             = _mm_add_pd(fjz0,tz);
1148
1149             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1150
1151             /* Inner loop uses 132 flops */
1152         }
1153
1154         /* End of innermost loop */
1155
1156         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1157                                               f+i_coord_offset,fshift+i_shift_offset);
1158
1159         /* Increment number of inner iterations */
1160         inneriter                  += j_index_end - j_index_start;
1161
1162         /* Outer loop uses 24 flops */
1163     }
1164
1165     /* Increment number of outer iterations */
1166     outeriter        += nri;
1167
1168     /* Update outer/inner flops */
1169
1170     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1171 }