src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_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              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128i          vfitab;
  99     __m128i          ifour       = _mm_set1_epi32(4);
 100     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 101     real             *vftab;
 102     __m128d          dummy_mask,cutoff_mask;
 103     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 104     __m128d          one     = _mm_set1_pd(1.0);
 105     __m128d          two     = _mm_set1_pd(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm_set1_pd(fr->ic->epsfac);
 118     charge           = mdatoms->chargeA;
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     vftab            = kernel_data->table_elec->data;
 124     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 125
 126     /* Setup water-specific parameters */
 127     inr              = nlist->iinr[0];
 128     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 129     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 130     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 131     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 158
 159         fix0             = _mm_setzero_pd();
 160         fiy0             = _mm_setzero_pd();
 161         fiz0             = _mm_setzero_pd();
 162         fix1             = _mm_setzero_pd();
 163         fiy1             = _mm_setzero_pd();
 164         fiz1             = _mm_setzero_pd();
 165         fix2             = _mm_setzero_pd();
 166         fiy2             = _mm_setzero_pd();
 167         fiz2             = _mm_setzero_pd();
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm_setzero_pd();
 171         vvdwsum          = _mm_setzero_pd();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                               &jx0,&jy0,&jz0);
 186
 187             /* Calculate displacement vector */
 188             dx00             = _mm_sub_pd(ix0,jx0);
 189             dy00             = _mm_sub_pd(iy0,jy0);
 190             dz00             = _mm_sub_pd(iz0,jz0);
 191             dx10             = _mm_sub_pd(ix1,jx0);
 192             dy10             = _mm_sub_pd(iy1,jy0);
 193             dz10             = _mm_sub_pd(iz1,jz0);
 194             dx20             = _mm_sub_pd(ix2,jx0);
 195             dy20             = _mm_sub_pd(iy2,jy0);
 196             dz20             = _mm_sub_pd(iz2,jz0);
 197
 198             /* Calculate squared distance and things based on it */
 199             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 200             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 201             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 202
 203             rinv00           = sse2_invsqrt_d(rsq00);
 204             rinv10           = sse2_invsqrt_d(rsq10);
 205             rinv20           = sse2_invsqrt_d(rsq20);
 206
 207             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 211             vdwjidx0A        = 2*vdwtype[jnrA+0];
 212             vdwjidx0B        = 2*vdwtype[jnrB+0];
 213
 214             fjx0             = _mm_setzero_pd();
 215             fjy0             = _mm_setzero_pd();
 216             fjz0             = _mm_setzero_pd();
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             r00              = _mm_mul_pd(rsq00,rinv00);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq00             = _mm_mul_pd(iq0,jq0);
 226             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 227                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 228
 229             /* Calculate table index by multiplying r with table scale and truncate to integer */
 230             rt               = _mm_mul_pd(r00,vftabscale);
 231             vfitab           = _mm_cvttpd_epi32(rt);
 232             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 233             vfitab           = _mm_slli_epi32(vfitab,2);
 234
 235             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 236             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 237             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 238             GMX_MM_TRANSPOSE2_PD(Y,F);
 239             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 240             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 241             GMX_MM_TRANSPOSE2_PD(G,H);
 242             Heps             = _mm_mul_pd(vfeps,H);
 243             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 244             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 245             velec            = _mm_mul_pd(qq00,VV);
 246             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 247             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 248
 249             /* LENNARD-JONES DISPERSION/REPULSION */
 250
 251             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 252             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 253             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 254             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 255             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 256
 257             /* Update potential sum for this i atom from the interaction with this j atom. */
 258             velecsum         = _mm_add_pd(velecsum,velec);
 259             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 260
 261             fscal            = _mm_add_pd(felec,fvdw);
 262
 263             /* Calculate temporary vectorial force */
 264             tx               = _mm_mul_pd(fscal,dx00);
 265             ty               = _mm_mul_pd(fscal,dy00);
 266             tz               = _mm_mul_pd(fscal,dz00);
 267
 268             /* Update vectorial force */
 269             fix0             = _mm_add_pd(fix0,tx);
 270             fiy0             = _mm_add_pd(fiy0,ty);
 271             fiz0             = _mm_add_pd(fiz0,tz);
 272
 273             fjx0             = _mm_add_pd(fjx0,tx);
 274             fjy0             = _mm_add_pd(fjy0,ty);
 275             fjz0             = _mm_add_pd(fjz0,tz);
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             r10              = _mm_mul_pd(rsq10,rinv10);
 282
 283             /* Compute parameters for interactions between i and j atoms */
 284             qq10             = _mm_mul_pd(iq1,jq0);
 285
 286             /* Calculate table index by multiplying r with table scale and truncate to integer */
 287             rt               = _mm_mul_pd(r10,vftabscale);
 288             vfitab           = _mm_cvttpd_epi32(rt);
 289             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 290             vfitab           = _mm_slli_epi32(vfitab,2);
 291
 292             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 293             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 294             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 295             GMX_MM_TRANSPOSE2_PD(Y,F);
 296             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 297             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 298             GMX_MM_TRANSPOSE2_PD(G,H);
 299             Heps             = _mm_mul_pd(vfeps,H);
 300             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 301             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 302             velec            = _mm_mul_pd(qq10,VV);
 303             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 304             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 305
 306             /* Update potential sum for this i atom from the interaction with this j atom. */
 307             velecsum         = _mm_add_pd(velecsum,velec);
 308
 309             fscal            = felec;
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm_mul_pd(fscal,dx10);
 313             ty               = _mm_mul_pd(fscal,dy10);
 314             tz               = _mm_mul_pd(fscal,dz10);
 315
 316             /* Update vectorial force */
 317             fix1             = _mm_add_pd(fix1,tx);
 318             fiy1             = _mm_add_pd(fiy1,ty);
 319             fiz1             = _mm_add_pd(fiz1,tz);
 320
 321             fjx0             = _mm_add_pd(fjx0,tx);
 322             fjy0             = _mm_add_pd(fjy0,ty);
 323             fjz0             = _mm_add_pd(fjz0,tz);
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             r20              = _mm_mul_pd(rsq20,rinv20);
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq20             = _mm_mul_pd(iq2,jq0);
 333
 334             /* Calculate table index by multiplying r with table scale and truncate to integer */
 335             rt               = _mm_mul_pd(r20,vftabscale);
 336             vfitab           = _mm_cvttpd_epi32(rt);
 337             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 338             vfitab           = _mm_slli_epi32(vfitab,2);
 339
 340             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 341             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 342             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 343             GMX_MM_TRANSPOSE2_PD(Y,F);
 344             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 345             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 346             GMX_MM_TRANSPOSE2_PD(G,H);
 347             Heps             = _mm_mul_pd(vfeps,H);
 348             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 349             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 350             velec            = _mm_mul_pd(qq20,VV);
 351             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 352             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 353
 354             /* Update potential sum for this i atom from the interaction with this j atom. */
 355             velecsum         = _mm_add_pd(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359             /* Calculate temporary vectorial force */
 360             tx               = _mm_mul_pd(fscal,dx20);
 361             ty               = _mm_mul_pd(fscal,dy20);
 362             tz               = _mm_mul_pd(fscal,dz20);
 363
 364             /* Update vectorial force */
 365             fix2             = _mm_add_pd(fix2,tx);
 366             fiy2             = _mm_add_pd(fiy2,ty);
 367             fiz2             = _mm_add_pd(fiz2,tz);
 368
 369             fjx0             = _mm_add_pd(fjx0,tx);
 370             fjy0             = _mm_add_pd(fjy0,ty);
 371             fjz0             = _mm_add_pd(fjz0,tz);
 372
 373             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 374
 375             /* Inner loop uses 145 flops */
 376         }
 377
 378         if(jidx<j_index_end)
 379         {
 380
 381             jnrA             = jjnr[jidx];
 382             j_coord_offsetA  = DIM*jnrA;
 383
 384             /* load j atom coordinates */
 385             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 386                                               &jx0,&jy0,&jz0);
 387
 388             /* Calculate displacement vector */
 389             dx00             = _mm_sub_pd(ix0,jx0);
 390             dy00             = _mm_sub_pd(iy0,jy0);
 391             dz00             = _mm_sub_pd(iz0,jz0);
 392             dx10             = _mm_sub_pd(ix1,jx0);
 393             dy10             = _mm_sub_pd(iy1,jy0);
 394             dz10             = _mm_sub_pd(iz1,jz0);
 395             dx20             = _mm_sub_pd(ix2,jx0);
 396             dy20             = _mm_sub_pd(iy2,jy0);
 397             dz20             = _mm_sub_pd(iz2,jz0);
 398
 399             /* Calculate squared distance and things based on it */
 400             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 401             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 402             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 403
 404             rinv00           = sse2_invsqrt_d(rsq00);
 405             rinv10           = sse2_invsqrt_d(rsq10);
 406             rinv20           = sse2_invsqrt_d(rsq20);
 407
 408             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 409
 410             /* Load parameters for j particles */
 411             jq0              = _mm_load_sd(charge+jnrA+0);
 412             vdwjidx0A        = 2*vdwtype[jnrA+0];
 413
 414             fjx0             = _mm_setzero_pd();
 415             fjy0             = _mm_setzero_pd();
 416             fjz0             = _mm_setzero_pd();
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             r00              = _mm_mul_pd(rsq00,rinv00);
 423
 424             /* Compute parameters for interactions between i and j atoms */
 425             qq00             = _mm_mul_pd(iq0,jq0);
 426             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 427
 428             /* Calculate table index by multiplying r with table scale and truncate to integer */
 429             rt               = _mm_mul_pd(r00,vftabscale);
 430             vfitab           = _mm_cvttpd_epi32(rt);
 431             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 432             vfitab           = _mm_slli_epi32(vfitab,2);
 433
 434             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 435             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 436             F                = _mm_setzero_pd();
 437             GMX_MM_TRANSPOSE2_PD(Y,F);
 438             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 439             H                = _mm_setzero_pd();
 440             GMX_MM_TRANSPOSE2_PD(G,H);
 441             Heps             = _mm_mul_pd(vfeps,H);
 442             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 443             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 444             velec            = _mm_mul_pd(qq00,VV);
 445             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 446             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 447
 448             /* LENNARD-JONES DISPERSION/REPULSION */
 449
 450             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 451             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 452             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 453             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 454             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 458             velecsum         = _mm_add_pd(velecsum,velec);
 459             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 460             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 461
 462             fscal            = _mm_add_pd(felec,fvdw);
 463
 464             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 465
 466             /* Calculate temporary vectorial force */
 467             tx               = _mm_mul_pd(fscal,dx00);
 468             ty               = _mm_mul_pd(fscal,dy00);
 469             tz               = _mm_mul_pd(fscal,dz00);
 470
 471             /* Update vectorial force */
 472             fix0             = _mm_add_pd(fix0,tx);
 473             fiy0             = _mm_add_pd(fiy0,ty);
 474             fiz0             = _mm_add_pd(fiz0,tz);
 475
 476             fjx0             = _mm_add_pd(fjx0,tx);
 477             fjy0             = _mm_add_pd(fjy0,ty);
 478             fjz0             = _mm_add_pd(fjz0,tz);
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r10              = _mm_mul_pd(rsq10,rinv10);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq10             = _mm_mul_pd(iq1,jq0);
 488
 489             /* Calculate table index by multiplying r with table scale and truncate to integer */
 490             rt               = _mm_mul_pd(r10,vftabscale);
 491             vfitab           = _mm_cvttpd_epi32(rt);
 492             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 493             vfitab           = _mm_slli_epi32(vfitab,2);
 494
 495             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 496             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 497             F                = _mm_setzero_pd();
 498             GMX_MM_TRANSPOSE2_PD(Y,F);
 499             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 500             H                = _mm_setzero_pd();
 501             GMX_MM_TRANSPOSE2_PD(G,H);
 502             Heps             = _mm_mul_pd(vfeps,H);
 503             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 504             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 505             velec            = _mm_mul_pd(qq10,VV);
 506             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 507             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 511             velecsum         = _mm_add_pd(velecsum,velec);
 512
 513             fscal            = felec;
 514
 515             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 516
 517             /* Calculate temporary vectorial force */
 518             tx               = _mm_mul_pd(fscal,dx10);
 519             ty               = _mm_mul_pd(fscal,dy10);
 520             tz               = _mm_mul_pd(fscal,dz10);
 521
 522             /* Update vectorial force */
 523             fix1             = _mm_add_pd(fix1,tx);
 524             fiy1             = _mm_add_pd(fiy1,ty);
 525             fiz1             = _mm_add_pd(fiz1,tz);
 526
 527             fjx0             = _mm_add_pd(fjx0,tx);
 528             fjy0             = _mm_add_pd(fjy0,ty);
 529             fjz0             = _mm_add_pd(fjz0,tz);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r20              = _mm_mul_pd(rsq20,rinv20);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq20             = _mm_mul_pd(iq2,jq0);
 539
 540             /* Calculate table index by multiplying r with table scale and truncate to integer */
 541             rt               = _mm_mul_pd(r20,vftabscale);
 542             vfitab           = _mm_cvttpd_epi32(rt);
 543             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 544             vfitab           = _mm_slli_epi32(vfitab,2);
 545
 546             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 547             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 548             F                = _mm_setzero_pd();
 549             GMX_MM_TRANSPOSE2_PD(Y,F);
 550             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 551             H                = _mm_setzero_pd();
 552             GMX_MM_TRANSPOSE2_PD(G,H);
 553             Heps             = _mm_mul_pd(vfeps,H);
 554             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 555             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 556             velec            = _mm_mul_pd(qq20,VV);
 557             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 558             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 559
 560             /* Update potential sum for this i atom from the interaction with this j atom. */
 561             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 562             velecsum         = _mm_add_pd(velecsum,velec);
 563
 564             fscal            = felec;
 565
 566             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm_mul_pd(fscal,dx20);
 570             ty               = _mm_mul_pd(fscal,dy20);
 571             tz               = _mm_mul_pd(fscal,dz20);
 572
 573             /* Update vectorial force */
 574             fix2             = _mm_add_pd(fix2,tx);
 575             fiy2             = _mm_add_pd(fiy2,ty);
 576             fiz2             = _mm_add_pd(fiz2,tz);
 577
 578             fjx0             = _mm_add_pd(fjx0,tx);
 579             fjy0             = _mm_add_pd(fjy0,ty);
 580             fjz0             = _mm_add_pd(fjz0,tz);
 581
 582             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 583
 584             /* Inner loop uses 145 flops */
 585         }
 586
 587         /* End of innermost loop */
 588
 589         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 590                                               f+i_coord_offset,fshift+i_shift_offset);
 591
 592         ggid                        = gid[iidx];
 593         /* Update potential energies */
 594         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 595         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 596
 597         /* Increment number of inner iterations */
 598         inneriter                  += j_index_end - j_index_start;
 599
 600         /* Outer loop uses 20 flops */
 601     }
 602
 603     /* Increment number of outer iterations */
 604     outeriter        += nri;
 605
 606     /* Update outer/inner flops */
 607
 608     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 609 }
 610 /*
 611  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_double
 612  * Electrostatics interaction: CubicSplineTable
 613  * VdW interaction:            LennardJones
 614  * Geometry:                   Water3-Particle
 615  * Calculate force/pot:        Force
 616  */
 617 void
 618 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_double
 619                     (t_nblist                    * gmx_restrict       nlist,
 620                      rvec                        * gmx_restrict          xx,
 621                      rvec                        * gmx_restrict          ff,
 622                      struct t_forcerec           * gmx_restrict          fr,
 623                      t_mdatoms                   * gmx_restrict     mdatoms,
 624                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 625                      t_nrnb                      * gmx_restrict        nrnb)
 626 {
 627     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 628      * just 0 for non-waters.
 629      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 630      * jnr indices corresponding to data put in the four positions in the SIMD register.
 631      */
 632     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 633     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 634     int              jnrA,jnrB;
 635     int              j_coord_offsetA,j_coord_offsetB;
 636     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 637     real             rcutoff_scalar;
 638     real             *shiftvec,*fshift,*x,*f;
 639     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 640     int              vdwioffset0;
 641     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 642     int              vdwioffset1;
 643     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 644     int              vdwioffset2;
 645     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 646     int              vdwjidx0A,vdwjidx0B;
 647     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 648     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 649     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 650     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 651     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 652     real             *charge;
 653     int              nvdwtype;
 654     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 655     int              *vdwtype;
 656     real             *vdwparam;
 657     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 658     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 659     __m128i          vfitab;
 660     __m128i          ifour       = _mm_set1_epi32(4);
 661     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 662     real             *vftab;
 663     __m128d          dummy_mask,cutoff_mask;
 664     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 665     __m128d          one     = _mm_set1_pd(1.0);
 666     __m128d          two     = _mm_set1_pd(2.0);
 667     x                = xx[0];
 668     f                = ff[0];
 669
 670     nri              = nlist->nri;
 671     iinr             = nlist->iinr;
 672     jindex           = nlist->jindex;
 673     jjnr             = nlist->jjnr;
 674     shiftidx         = nlist->shift;
 675     gid              = nlist->gid;
 676     shiftvec         = fr->shift_vec[0];
 677     fshift           = fr->fshift[0];
 678     facel            = _mm_set1_pd(fr->ic->epsfac);
 679     charge           = mdatoms->chargeA;
 680     nvdwtype         = fr->ntype;
 681     vdwparam         = fr->nbfp;
 682     vdwtype          = mdatoms->typeA;
 683
 684     vftab            = kernel_data->table_elec->data;
 685     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 686
 687     /* Setup water-specific parameters */
 688     inr              = nlist->iinr[0];
 689     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 690     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 691     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 692     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 693
 694     /* Avoid stupid compiler warnings */
 695     jnrA = jnrB = 0;
 696     j_coord_offsetA = 0;
 697     j_coord_offsetB = 0;
 698
 699     outeriter        = 0;
 700     inneriter        = 0;
 701
 702     /* Start outer loop over neighborlists */
 703     for(iidx=0; iidx<nri; iidx++)
 704     {
 705         /* Load shift vector for this list */
 706         i_shift_offset   = DIM*shiftidx[iidx];
 707
 708         /* Load limits for loop over neighbors */
 709         j_index_start    = jindex[iidx];
 710         j_index_end      = jindex[iidx+1];
 711
 712         /* Get outer coordinate index */
 713         inr              = iinr[iidx];
 714         i_coord_offset   = DIM*inr;
 715
 716         /* Load i particle coords and add shift vector */
 717         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 718                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 719
 720         fix0             = _mm_setzero_pd();
 721         fiy0             = _mm_setzero_pd();
 722         fiz0             = _mm_setzero_pd();
 723         fix1             = _mm_setzero_pd();
 724         fiy1             = _mm_setzero_pd();
 725         fiz1             = _mm_setzero_pd();
 726         fix2             = _mm_setzero_pd();
 727         fiy2             = _mm_setzero_pd();
 728         fiz2             = _mm_setzero_pd();
 729
 730         /* Start inner kernel loop */
 731         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 732         {
 733
 734             /* Get j neighbor index, and coordinate index */
 735             jnrA             = jjnr[jidx];
 736             jnrB             = jjnr[jidx+1];
 737             j_coord_offsetA  = DIM*jnrA;
 738             j_coord_offsetB  = DIM*jnrB;
 739
 740             /* load j atom coordinates */
 741             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 742                                               &jx0,&jy0,&jz0);
 743
 744             /* Calculate displacement vector */
 745             dx00             = _mm_sub_pd(ix0,jx0);
 746             dy00             = _mm_sub_pd(iy0,jy0);
 747             dz00             = _mm_sub_pd(iz0,jz0);
 748             dx10             = _mm_sub_pd(ix1,jx0);
 749             dy10             = _mm_sub_pd(iy1,jy0);
 750             dz10             = _mm_sub_pd(iz1,jz0);
 751             dx20             = _mm_sub_pd(ix2,jx0);
 752             dy20             = _mm_sub_pd(iy2,jy0);
 753             dz20             = _mm_sub_pd(iz2,jz0);
 754
 755             /* Calculate squared distance and things based on it */
 756             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 757             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 758             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 759
 760             rinv00           = sse2_invsqrt_d(rsq00);
 761             rinv10           = sse2_invsqrt_d(rsq10);
 762             rinv20           = sse2_invsqrt_d(rsq20);
 763
 764             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 765
 766             /* Load parameters for j particles */
 767             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 768             vdwjidx0A        = 2*vdwtype[jnrA+0];
 769             vdwjidx0B        = 2*vdwtype[jnrB+0];
 770
 771             fjx0             = _mm_setzero_pd();
 772             fjy0             = _mm_setzero_pd();
 773             fjz0             = _mm_setzero_pd();
 774
 775             /**************************
 776              * CALCULATE INTERACTIONS *
 777              **************************/
 778
 779             r00              = _mm_mul_pd(rsq00,rinv00);
 780
 781             /* Compute parameters for interactions between i and j atoms */
 782             qq00             = _mm_mul_pd(iq0,jq0);
 783             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 784                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 785
 786             /* Calculate table index by multiplying r with table scale and truncate to integer */
 787             rt               = _mm_mul_pd(r00,vftabscale);
 788             vfitab           = _mm_cvttpd_epi32(rt);
 789             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 790             vfitab           = _mm_slli_epi32(vfitab,2);
 791
 792             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 793             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 794             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 795             GMX_MM_TRANSPOSE2_PD(Y,F);
 796             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 797             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 798             GMX_MM_TRANSPOSE2_PD(G,H);
 799             Heps             = _mm_mul_pd(vfeps,H);
 800             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 801             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 802             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 803
 804             /* LENNARD-JONES DISPERSION/REPULSION */
 805
 806             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 807             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 808
 809             fscal            = _mm_add_pd(felec,fvdw);
 810
 811             /* Calculate temporary vectorial force */
 812             tx               = _mm_mul_pd(fscal,dx00);
 813             ty               = _mm_mul_pd(fscal,dy00);
 814             tz               = _mm_mul_pd(fscal,dz00);
 815
 816             /* Update vectorial force */
 817             fix0             = _mm_add_pd(fix0,tx);
 818             fiy0             = _mm_add_pd(fiy0,ty);
 819             fiz0             = _mm_add_pd(fiz0,tz);
 820
 821             fjx0             = _mm_add_pd(fjx0,tx);
 822             fjy0             = _mm_add_pd(fjy0,ty);
 823             fjz0             = _mm_add_pd(fjz0,tz);
 824
 825             /**************************
 826              * CALCULATE INTERACTIONS *
 827              **************************/
 828
 829             r10              = _mm_mul_pd(rsq10,rinv10);
 830
 831             /* Compute parameters for interactions between i and j atoms */
 832             qq10             = _mm_mul_pd(iq1,jq0);
 833
 834             /* Calculate table index by multiplying r with table scale and truncate to integer */
 835             rt               = _mm_mul_pd(r10,vftabscale);
 836             vfitab           = _mm_cvttpd_epi32(rt);
 837             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 838             vfitab           = _mm_slli_epi32(vfitab,2);
 839
 840             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 841             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 842             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 843             GMX_MM_TRANSPOSE2_PD(Y,F);
 844             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 845             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 846             GMX_MM_TRANSPOSE2_PD(G,H);
 847             Heps             = _mm_mul_pd(vfeps,H);
 848             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 849             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 850             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 851
 852             fscal            = felec;
 853
 854             /* Calculate temporary vectorial force */
 855             tx               = _mm_mul_pd(fscal,dx10);
 856             ty               = _mm_mul_pd(fscal,dy10);
 857             tz               = _mm_mul_pd(fscal,dz10);
 858
 859             /* Update vectorial force */
 860             fix1             = _mm_add_pd(fix1,tx);
 861             fiy1             = _mm_add_pd(fiy1,ty);
 862             fiz1             = _mm_add_pd(fiz1,tz);
 863
 864             fjx0             = _mm_add_pd(fjx0,tx);
 865             fjy0             = _mm_add_pd(fjy0,ty);
 866             fjz0             = _mm_add_pd(fjz0,tz);
 867
 868             /**************************
 869              * CALCULATE INTERACTIONS *
 870              **************************/
 871
 872             r20              = _mm_mul_pd(rsq20,rinv20);
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq20             = _mm_mul_pd(iq2,jq0);
 876
 877             /* Calculate table index by multiplying r with table scale and truncate to integer */
 878             rt               = _mm_mul_pd(r20,vftabscale);
 879             vfitab           = _mm_cvttpd_epi32(rt);
 880             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 881             vfitab           = _mm_slli_epi32(vfitab,2);
 882
 883             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 884             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 885             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 886             GMX_MM_TRANSPOSE2_PD(Y,F);
 887             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 888             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 889             GMX_MM_TRANSPOSE2_PD(G,H);
 890             Heps             = _mm_mul_pd(vfeps,H);
 891             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 892             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 893             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 894
 895             fscal            = felec;
 896
 897             /* Calculate temporary vectorial force */
 898             tx               = _mm_mul_pd(fscal,dx20);
 899             ty               = _mm_mul_pd(fscal,dy20);
 900             tz               = _mm_mul_pd(fscal,dz20);
 901
 902             /* Update vectorial force */
 903             fix2             = _mm_add_pd(fix2,tx);
 904             fiy2             = _mm_add_pd(fiy2,ty);
 905             fiz2             = _mm_add_pd(fiz2,tz);
 906
 907             fjx0             = _mm_add_pd(fjx0,tx);
 908             fjy0             = _mm_add_pd(fjy0,ty);
 909             fjz0             = _mm_add_pd(fjz0,tz);
 910
 911             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 912
 913             /* Inner loop uses 128 flops */
 914         }
 915
 916         if(jidx<j_index_end)
 917         {
 918
 919             jnrA             = jjnr[jidx];
 920             j_coord_offsetA  = DIM*jnrA;
 921
 922             /* load j atom coordinates */
 923             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 924                                               &jx0,&jy0,&jz0);
 925
 926             /* Calculate displacement vector */
 927             dx00             = _mm_sub_pd(ix0,jx0);
 928             dy00             = _mm_sub_pd(iy0,jy0);
 929             dz00             = _mm_sub_pd(iz0,jz0);
 930             dx10             = _mm_sub_pd(ix1,jx0);
 931             dy10             = _mm_sub_pd(iy1,jy0);
 932             dz10             = _mm_sub_pd(iz1,jz0);
 933             dx20             = _mm_sub_pd(ix2,jx0);
 934             dy20             = _mm_sub_pd(iy2,jy0);
 935             dz20             = _mm_sub_pd(iz2,jz0);
 936
 937             /* Calculate squared distance and things based on it */
 938             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 939             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 940             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 941
 942             rinv00           = sse2_invsqrt_d(rsq00);
 943             rinv10           = sse2_invsqrt_d(rsq10);
 944             rinv20           = sse2_invsqrt_d(rsq20);
 945
 946             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 947
 948             /* Load parameters for j particles */
 949             jq0              = _mm_load_sd(charge+jnrA+0);
 950             vdwjidx0A        = 2*vdwtype[jnrA+0];
 951
 952             fjx0             = _mm_setzero_pd();
 953             fjy0             = _mm_setzero_pd();
 954             fjz0             = _mm_setzero_pd();
 955
 956             /**************************
 957              * CALCULATE INTERACTIONS *
 958              **************************/
 959
 960             r00              = _mm_mul_pd(rsq00,rinv00);
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             qq00             = _mm_mul_pd(iq0,jq0);
 964             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 965
 966             /* Calculate table index by multiplying r with table scale and truncate to integer */
 967             rt               = _mm_mul_pd(r00,vftabscale);
 968             vfitab           = _mm_cvttpd_epi32(rt);
 969             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 970             vfitab           = _mm_slli_epi32(vfitab,2);
 971
 972             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 973             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 974             F                = _mm_setzero_pd();
 975             GMX_MM_TRANSPOSE2_PD(Y,F);
 976             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 977             H                = _mm_setzero_pd();
 978             GMX_MM_TRANSPOSE2_PD(G,H);
 979             Heps             = _mm_mul_pd(vfeps,H);
 980             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 981             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 982             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 983
 984             /* LENNARD-JONES DISPERSION/REPULSION */
 985
 986             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 987             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 988
 989             fscal            = _mm_add_pd(felec,fvdw);
 990
 991             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 992
 993             /* Calculate temporary vectorial force */
 994             tx               = _mm_mul_pd(fscal,dx00);
 995             ty               = _mm_mul_pd(fscal,dy00);
 996             tz               = _mm_mul_pd(fscal,dz00);
 997
 998             /* Update vectorial force */
 999             fix0             = _mm_add_pd(fix0,tx);
1000             fiy0             = _mm_add_pd(fiy0,ty);
1001             fiz0             = _mm_add_pd(fiz0,tz);
1002
1003             fjx0             = _mm_add_pd(fjx0,tx);
1004             fjy0             = _mm_add_pd(fjy0,ty);
1005             fjz0             = _mm_add_pd(fjz0,tz);
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             r10              = _mm_mul_pd(rsq10,rinv10);
1012
1013             /* Compute parameters for interactions between i and j atoms */
1014             qq10             = _mm_mul_pd(iq1,jq0);
1015
1016             /* Calculate table index by multiplying r with table scale and truncate to integer */
1017             rt               = _mm_mul_pd(r10,vftabscale);
1018             vfitab           = _mm_cvttpd_epi32(rt);
1019             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1020             vfitab           = _mm_slli_epi32(vfitab,2);
1021
1022             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1023             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1024             F                = _mm_setzero_pd();
1025             GMX_MM_TRANSPOSE2_PD(Y,F);
1026             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1027             H                = _mm_setzero_pd();
1028             GMX_MM_TRANSPOSE2_PD(G,H);
1029             Heps             = _mm_mul_pd(vfeps,H);
1030             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1031             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1032             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1033
1034             fscal            = felec;
1035
1036             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1037
1038             /* Calculate temporary vectorial force */
1039             tx               = _mm_mul_pd(fscal,dx10);
1040             ty               = _mm_mul_pd(fscal,dy10);
1041             tz               = _mm_mul_pd(fscal,dz10);
1042
1043             /* Update vectorial force */
1044             fix1             = _mm_add_pd(fix1,tx);
1045             fiy1             = _mm_add_pd(fiy1,ty);
1046             fiz1             = _mm_add_pd(fiz1,tz);
1047
1048             fjx0             = _mm_add_pd(fjx0,tx);
1049             fjy0             = _mm_add_pd(fjy0,ty);
1050             fjz0             = _mm_add_pd(fjz0,tz);
1051
1052             /**************************
1053              * CALCULATE INTERACTIONS *
1054              **************************/
1055
1056             r20              = _mm_mul_pd(rsq20,rinv20);
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             qq20             = _mm_mul_pd(iq2,jq0);
1060
1061             /* Calculate table index by multiplying r with table scale and truncate to integer */
1062             rt               = _mm_mul_pd(r20,vftabscale);
1063             vfitab           = _mm_cvttpd_epi32(rt);
1064             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1065             vfitab           = _mm_slli_epi32(vfitab,2);
1066
1067             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1068             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1069             F                = _mm_setzero_pd();
1070             GMX_MM_TRANSPOSE2_PD(Y,F);
1071             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1072             H                = _mm_setzero_pd();
1073             GMX_MM_TRANSPOSE2_PD(G,H);
1074             Heps             = _mm_mul_pd(vfeps,H);
1075             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1076             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1077             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1078
1079             fscal            = felec;
1080
1081             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1082
1083             /* Calculate temporary vectorial force */
1084             tx               = _mm_mul_pd(fscal,dx20);
1085             ty               = _mm_mul_pd(fscal,dy20);
1086             tz               = _mm_mul_pd(fscal,dz20);
1087
1088             /* Update vectorial force */
1089             fix2             = _mm_add_pd(fix2,tx);
1090             fiy2             = _mm_add_pd(fiy2,ty);
1091             fiz2             = _mm_add_pd(fiz2,tz);
1092
1093             fjx0             = _mm_add_pd(fjx0,tx);
1094             fjy0             = _mm_add_pd(fjy0,ty);
1095             fjz0             = _mm_add_pd(fjz0,tz);
1096
1097             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1098
1099             /* Inner loop uses 128 flops */
1100         }
1101
1102         /* End of innermost loop */
1103
1104         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1105                                               f+i_coord_offset,fshift+i_shift_offset);
1106
1107         /* Increment number of inner iterations */
1108         inneriter                  += j_index_end - j_index_start;
1109
1110         /* Outer loop uses 18 flops */
1111     }
1112
1113     /* Increment number of outer iterations */
1114     outeriter        += nri;
1115
1116     /* Update outer/inner flops */
1117
1118     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*128);
1119 }