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, 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 "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_double
  52  * Electrostatics interaction: CubicSplineTable
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128i          vfitab;
 100     __m128i          ifour       = _mm_set1_epi32(4);
 101     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 102     real             *vftab;
 103     __m128d          dummy_mask,cutoff_mask;
 104     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 105     __m128d          one     = _mm_set1_pd(1.0);
 106     __m128d          two     = _mm_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_pd(fr->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     vftab            = kernel_data->table_elec->data;
 125     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 130     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 131     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 159
 160         fix0             = _mm_setzero_pd();
 161         fiy0             = _mm_setzero_pd();
 162         fiz0             = _mm_setzero_pd();
 163         fix1             = _mm_setzero_pd();
 164         fiy1             = _mm_setzero_pd();
 165         fiz1             = _mm_setzero_pd();
 166         fix2             = _mm_setzero_pd();
 167         fiy2             = _mm_setzero_pd();
 168         fiz2             = _mm_setzero_pd();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_pd();
 172         vvdwsum          = _mm_setzero_pd();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_pd(ix0,jx0);
 190             dy00             = _mm_sub_pd(iy0,jy0);
 191             dz00             = _mm_sub_pd(iz0,jz0);
 192             dx10             = _mm_sub_pd(ix1,jx0);
 193             dy10             = _mm_sub_pd(iy1,jy0);
 194             dz10             = _mm_sub_pd(iz1,jz0);
 195             dx20             = _mm_sub_pd(ix2,jx0);
 196             dy20             = _mm_sub_pd(iy2,jy0);
 197             dz20             = _mm_sub_pd(iz2,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 201             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 202             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 203
 204             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 205             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 206             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 207
 208             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 209
 210             /* Load parameters for j particles */
 211             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 212             vdwjidx0A        = 2*vdwtype[jnrA+0];
 213             vdwjidx0B        = 2*vdwtype[jnrB+0];
 214
 215             fjx0             = _mm_setzero_pd();
 216             fjy0             = _mm_setzero_pd();
 217             fjz0             = _mm_setzero_pd();
 218
 219             /**************************
 220              * CALCULATE INTERACTIONS *
 221              **************************/
 222
 223             r00              = _mm_mul_pd(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_pd(iq0,jq0);
 227             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 228                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 229
 230             /* Calculate table index by multiplying r with table scale and truncate to integer */
 231             rt               = _mm_mul_pd(r00,vftabscale);
 232             vfitab           = _mm_cvttpd_epi32(rt);
 233             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 234             vfitab           = _mm_slli_epi32(vfitab,2);
 235
 236             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 237             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 238             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 239             GMX_MM_TRANSPOSE2_PD(Y,F);
 240             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 241             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 242             GMX_MM_TRANSPOSE2_PD(G,H);
 243             Heps             = _mm_mul_pd(vfeps,H);
 244             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 245             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 246             velec            = _mm_mul_pd(qq00,VV);
 247             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 248             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 249
 250             /* LENNARD-JONES DISPERSION/REPULSION */
 251
 252             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 253             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 254             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 255             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 256             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 257
 258             /* Update potential sum for this i atom from the interaction with this j atom. */
 259             velecsum         = _mm_add_pd(velecsum,velec);
 260             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 261
 262             fscal            = _mm_add_pd(felec,fvdw);
 263
 264             /* Calculate temporary vectorial force */
 265             tx               = _mm_mul_pd(fscal,dx00);
 266             ty               = _mm_mul_pd(fscal,dy00);
 267             tz               = _mm_mul_pd(fscal,dz00);
 268
 269             /* Update vectorial force */
 270             fix0             = _mm_add_pd(fix0,tx);
 271             fiy0             = _mm_add_pd(fiy0,ty);
 272             fiz0             = _mm_add_pd(fiz0,tz);
 273
 274             fjx0             = _mm_add_pd(fjx0,tx);
 275             fjy0             = _mm_add_pd(fjy0,ty);
 276             fjz0             = _mm_add_pd(fjz0,tz);
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             r10              = _mm_mul_pd(rsq10,rinv10);
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             qq10             = _mm_mul_pd(iq1,jq0);
 286
 287             /* Calculate table index by multiplying r with table scale and truncate to integer */
 288             rt               = _mm_mul_pd(r10,vftabscale);
 289             vfitab           = _mm_cvttpd_epi32(rt);
 290             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 291             vfitab           = _mm_slli_epi32(vfitab,2);
 292
 293             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 294             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 295             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 296             GMX_MM_TRANSPOSE2_PD(Y,F);
 297             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 298             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 299             GMX_MM_TRANSPOSE2_PD(G,H);
 300             Heps             = _mm_mul_pd(vfeps,H);
 301             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 302             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 303             velec            = _mm_mul_pd(qq10,VV);
 304             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 305             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velecsum         = _mm_add_pd(velecsum,velec);
 309
 310             fscal            = felec;
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm_mul_pd(fscal,dx10);
 314             ty               = _mm_mul_pd(fscal,dy10);
 315             tz               = _mm_mul_pd(fscal,dz10);
 316
 317             /* Update vectorial force */
 318             fix1             = _mm_add_pd(fix1,tx);
 319             fiy1             = _mm_add_pd(fiy1,ty);
 320             fiz1             = _mm_add_pd(fiz1,tz);
 321
 322             fjx0             = _mm_add_pd(fjx0,tx);
 323             fjy0             = _mm_add_pd(fjy0,ty);
 324             fjz0             = _mm_add_pd(fjz0,tz);
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             r20              = _mm_mul_pd(rsq20,rinv20);
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq20             = _mm_mul_pd(iq2,jq0);
 334
 335             /* Calculate table index by multiplying r with table scale and truncate to integer */
 336             rt               = _mm_mul_pd(r20,vftabscale);
 337             vfitab           = _mm_cvttpd_epi32(rt);
 338             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 339             vfitab           = _mm_slli_epi32(vfitab,2);
 340
 341             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 342             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 343             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 344             GMX_MM_TRANSPOSE2_PD(Y,F);
 345             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 346             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 347             GMX_MM_TRANSPOSE2_PD(G,H);
 348             Heps             = _mm_mul_pd(vfeps,H);
 349             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 350             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 351             velec            = _mm_mul_pd(qq20,VV);
 352             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 353             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm_add_pd(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm_mul_pd(fscal,dx20);
 362             ty               = _mm_mul_pd(fscal,dy20);
 363             tz               = _mm_mul_pd(fscal,dz20);
 364
 365             /* Update vectorial force */
 366             fix2             = _mm_add_pd(fix2,tx);
 367             fiy2             = _mm_add_pd(fiy2,ty);
 368             fiz2             = _mm_add_pd(fiz2,tz);
 369
 370             fjx0             = _mm_add_pd(fjx0,tx);
 371             fjy0             = _mm_add_pd(fjy0,ty);
 372             fjz0             = _mm_add_pd(fjz0,tz);
 373
 374             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 375
 376             /* Inner loop uses 145 flops */
 377         }
 378
 379         if(jidx<j_index_end)
 380         {
 381
 382             jnrA             = jjnr[jidx];
 383             j_coord_offsetA  = DIM*jnrA;
 384
 385             /* load j atom coordinates */
 386             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 387                                               &jx0,&jy0,&jz0);
 388
 389             /* Calculate displacement vector */
 390             dx00             = _mm_sub_pd(ix0,jx0);
 391             dy00             = _mm_sub_pd(iy0,jy0);
 392             dz00             = _mm_sub_pd(iz0,jz0);
 393             dx10             = _mm_sub_pd(ix1,jx0);
 394             dy10             = _mm_sub_pd(iy1,jy0);
 395             dz10             = _mm_sub_pd(iz1,jz0);
 396             dx20             = _mm_sub_pd(ix2,jx0);
 397             dy20             = _mm_sub_pd(iy2,jy0);
 398             dz20             = _mm_sub_pd(iz2,jz0);
 399
 400             /* Calculate squared distance and things based on it */
 401             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 402             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 403             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 404
 405             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 406             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 407             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 408
 409             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 410
 411             /* Load parameters for j particles */
 412             jq0              = _mm_load_sd(charge+jnrA+0);
 413             vdwjidx0A        = 2*vdwtype[jnrA+0];
 414
 415             fjx0             = _mm_setzero_pd();
 416             fjy0             = _mm_setzero_pd();
 417             fjz0             = _mm_setzero_pd();
 418
 419             /**************************
 420              * CALCULATE INTERACTIONS *
 421              **************************/
 422
 423             r00              = _mm_mul_pd(rsq00,rinv00);
 424
 425             /* Compute parameters for interactions between i and j atoms */
 426             qq00             = _mm_mul_pd(iq0,jq0);
 427             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 428
 429             /* Calculate table index by multiplying r with table scale and truncate to integer */
 430             rt               = _mm_mul_pd(r00,vftabscale);
 431             vfitab           = _mm_cvttpd_epi32(rt);
 432             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 433             vfitab           = _mm_slli_epi32(vfitab,2);
 434
 435             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 436             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 437             F                = _mm_setzero_pd();
 438             GMX_MM_TRANSPOSE2_PD(Y,F);
 439             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 440             H                = _mm_setzero_pd();
 441             GMX_MM_TRANSPOSE2_PD(G,H);
 442             Heps             = _mm_mul_pd(vfeps,H);
 443             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 444             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 445             velec            = _mm_mul_pd(qq00,VV);
 446             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 447             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 448
 449             /* LENNARD-JONES DISPERSION/REPULSION */
 450
 451             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 452             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 453             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 454             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 455             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 459             velecsum         = _mm_add_pd(velecsum,velec);
 460             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 461             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 462
 463             fscal            = _mm_add_pd(felec,fvdw);
 464
 465             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 466
 467             /* Calculate temporary vectorial force */
 468             tx               = _mm_mul_pd(fscal,dx00);
 469             ty               = _mm_mul_pd(fscal,dy00);
 470             tz               = _mm_mul_pd(fscal,dz00);
 471
 472             /* Update vectorial force */
 473             fix0             = _mm_add_pd(fix0,tx);
 474             fiy0             = _mm_add_pd(fiy0,ty);
 475             fiz0             = _mm_add_pd(fiz0,tz);
 476
 477             fjx0             = _mm_add_pd(fjx0,tx);
 478             fjy0             = _mm_add_pd(fjy0,ty);
 479             fjz0             = _mm_add_pd(fjz0,tz);
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             r10              = _mm_mul_pd(rsq10,rinv10);
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq10             = _mm_mul_pd(iq1,jq0);
 489
 490             /* Calculate table index by multiplying r with table scale and truncate to integer */
 491             rt               = _mm_mul_pd(r10,vftabscale);
 492             vfitab           = _mm_cvttpd_epi32(rt);
 493             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 494             vfitab           = _mm_slli_epi32(vfitab,2);
 495
 496             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 497             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 498             F                = _mm_setzero_pd();
 499             GMX_MM_TRANSPOSE2_PD(Y,F);
 500             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 501             H                = _mm_setzero_pd();
 502             GMX_MM_TRANSPOSE2_PD(G,H);
 503             Heps             = _mm_mul_pd(vfeps,H);
 504             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 505             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 506             velec            = _mm_mul_pd(qq10,VV);
 507             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 508             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 512             velecsum         = _mm_add_pd(velecsum,velec);
 513
 514             fscal            = felec;
 515
 516             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm_mul_pd(fscal,dx10);
 520             ty               = _mm_mul_pd(fscal,dy10);
 521             tz               = _mm_mul_pd(fscal,dz10);
 522
 523             /* Update vectorial force */
 524             fix1             = _mm_add_pd(fix1,tx);
 525             fiy1             = _mm_add_pd(fiy1,ty);
 526             fiz1             = _mm_add_pd(fiz1,tz);
 527
 528             fjx0             = _mm_add_pd(fjx0,tx);
 529             fjy0             = _mm_add_pd(fjy0,ty);
 530             fjz0             = _mm_add_pd(fjz0,tz);
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             r20              = _mm_mul_pd(rsq20,rinv20);
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             qq20             = _mm_mul_pd(iq2,jq0);
 540
 541             /* Calculate table index by multiplying r with table scale and truncate to integer */
 542             rt               = _mm_mul_pd(r20,vftabscale);
 543             vfitab           = _mm_cvttpd_epi32(rt);
 544             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 545             vfitab           = _mm_slli_epi32(vfitab,2);
 546
 547             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 548             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 549             F                = _mm_setzero_pd();
 550             GMX_MM_TRANSPOSE2_PD(Y,F);
 551             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 552             H                = _mm_setzero_pd();
 553             GMX_MM_TRANSPOSE2_PD(G,H);
 554             Heps             = _mm_mul_pd(vfeps,H);
 555             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 556             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 557             velec            = _mm_mul_pd(qq20,VV);
 558             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 559             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 563             velecsum         = _mm_add_pd(velecsum,velec);
 564
 565             fscal            = felec;
 566
 567             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 568
 569             /* Calculate temporary vectorial force */
 570             tx               = _mm_mul_pd(fscal,dx20);
 571             ty               = _mm_mul_pd(fscal,dy20);
 572             tz               = _mm_mul_pd(fscal,dz20);
 573
 574             /* Update vectorial force */
 575             fix2             = _mm_add_pd(fix2,tx);
 576             fiy2             = _mm_add_pd(fiy2,ty);
 577             fiz2             = _mm_add_pd(fiz2,tz);
 578
 579             fjx0             = _mm_add_pd(fjx0,tx);
 580             fjy0             = _mm_add_pd(fjy0,ty);
 581             fjz0             = _mm_add_pd(fjz0,tz);
 582
 583             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 584
 585             /* Inner loop uses 145 flops */
 586         }
 587
 588         /* End of innermost loop */
 589
 590         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 591                                               f+i_coord_offset,fshift+i_shift_offset);
 592
 593         ggid                        = gid[iidx];
 594         /* Update potential energies */
 595         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 596         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 597
 598         /* Increment number of inner iterations */
 599         inneriter                  += j_index_end - j_index_start;
 600
 601         /* Outer loop uses 20 flops */
 602     }
 603
 604     /* Increment number of outer iterations */
 605     outeriter        += nri;
 606
 607     /* Update outer/inner flops */
 608
 609     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 610 }
 611 /*
 612  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_double
 613  * Electrostatics interaction: CubicSplineTable
 614  * VdW interaction:            LennardJones
 615  * Geometry:                   Water3-Particle
 616  * Calculate force/pot:        Force
 617  */
 618 void
 619 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_double
 620                     (t_nblist                    * gmx_restrict       nlist,
 621                      rvec                        * gmx_restrict          xx,
 622                      rvec                        * gmx_restrict          ff,
 623                      t_forcerec                  * gmx_restrict          fr,
 624                      t_mdatoms                   * gmx_restrict     mdatoms,
 625                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 626                      t_nrnb                      * gmx_restrict        nrnb)
 627 {
 628     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 629      * just 0 for non-waters.
 630      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 631      * jnr indices corresponding to data put in the four positions in the SIMD register.
 632      */
 633     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 634     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 635     int              jnrA,jnrB;
 636     int              j_coord_offsetA,j_coord_offsetB;
 637     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 638     real             rcutoff_scalar;
 639     real             *shiftvec,*fshift,*x,*f;
 640     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 641     int              vdwioffset0;
 642     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 643     int              vdwioffset1;
 644     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 645     int              vdwioffset2;
 646     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 647     int              vdwjidx0A,vdwjidx0B;
 648     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 649     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 650     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 651     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 652     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 653     real             *charge;
 654     int              nvdwtype;
 655     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 656     int              *vdwtype;
 657     real             *vdwparam;
 658     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 659     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 660     __m128i          vfitab;
 661     __m128i          ifour       = _mm_set1_epi32(4);
 662     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 663     real             *vftab;
 664     __m128d          dummy_mask,cutoff_mask;
 665     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 666     __m128d          one     = _mm_set1_pd(1.0);
 667     __m128d          two     = _mm_set1_pd(2.0);
 668     x                = xx[0];
 669     f                = ff[0];
 670
 671     nri              = nlist->nri;
 672     iinr             = nlist->iinr;
 673     jindex           = nlist->jindex;
 674     jjnr             = nlist->jjnr;
 675     shiftidx         = nlist->shift;
 676     gid              = nlist->gid;
 677     shiftvec         = fr->shift_vec[0];
 678     fshift           = fr->fshift[0];
 679     facel            = _mm_set1_pd(fr->epsfac);
 680     charge           = mdatoms->chargeA;
 681     nvdwtype         = fr->ntype;
 682     vdwparam         = fr->nbfp;
 683     vdwtype          = mdatoms->typeA;
 684
 685     vftab            = kernel_data->table_elec->data;
 686     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 687
 688     /* Setup water-specific parameters */
 689     inr              = nlist->iinr[0];
 690     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 691     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 692     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 693     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 694
 695     /* Avoid stupid compiler warnings */
 696     jnrA = jnrB = 0;
 697     j_coord_offsetA = 0;
 698     j_coord_offsetB = 0;
 699
 700     outeriter        = 0;
 701     inneriter        = 0;
 702
 703     /* Start outer loop over neighborlists */
 704     for(iidx=0; iidx<nri; iidx++)
 705     {
 706         /* Load shift vector for this list */
 707         i_shift_offset   = DIM*shiftidx[iidx];
 708
 709         /* Load limits for loop over neighbors */
 710         j_index_start    = jindex[iidx];
 711         j_index_end      = jindex[iidx+1];
 712
 713         /* Get outer coordinate index */
 714         inr              = iinr[iidx];
 715         i_coord_offset   = DIM*inr;
 716
 717         /* Load i particle coords and add shift vector */
 718         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 719                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 720
 721         fix0             = _mm_setzero_pd();
 722         fiy0             = _mm_setzero_pd();
 723         fiz0             = _mm_setzero_pd();
 724         fix1             = _mm_setzero_pd();
 725         fiy1             = _mm_setzero_pd();
 726         fiz1             = _mm_setzero_pd();
 727         fix2             = _mm_setzero_pd();
 728         fiy2             = _mm_setzero_pd();
 729         fiz2             = _mm_setzero_pd();
 730
 731         /* Start inner kernel loop */
 732         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 733         {
 734
 735             /* Get j neighbor index, and coordinate index */
 736             jnrA             = jjnr[jidx];
 737             jnrB             = jjnr[jidx+1];
 738             j_coord_offsetA  = DIM*jnrA;
 739             j_coord_offsetB  = DIM*jnrB;
 740
 741             /* load j atom coordinates */
 742             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 743                                               &jx0,&jy0,&jz0);
 744
 745             /* Calculate displacement vector */
 746             dx00             = _mm_sub_pd(ix0,jx0);
 747             dy00             = _mm_sub_pd(iy0,jy0);
 748             dz00             = _mm_sub_pd(iz0,jz0);
 749             dx10             = _mm_sub_pd(ix1,jx0);
 750             dy10             = _mm_sub_pd(iy1,jy0);
 751             dz10             = _mm_sub_pd(iz1,jz0);
 752             dx20             = _mm_sub_pd(ix2,jx0);
 753             dy20             = _mm_sub_pd(iy2,jy0);
 754             dz20             = _mm_sub_pd(iz2,jz0);
 755
 756             /* Calculate squared distance and things based on it */
 757             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 758             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 759             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 760
 761             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 762             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 763             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 764
 765             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 766
 767             /* Load parameters for j particles */
 768             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 769             vdwjidx0A        = 2*vdwtype[jnrA+0];
 770             vdwjidx0B        = 2*vdwtype[jnrB+0];
 771
 772             fjx0             = _mm_setzero_pd();
 773             fjy0             = _mm_setzero_pd();
 774             fjz0             = _mm_setzero_pd();
 775
 776             /**************************
 777              * CALCULATE INTERACTIONS *
 778              **************************/
 779
 780             r00              = _mm_mul_pd(rsq00,rinv00);
 781
 782             /* Compute parameters for interactions between i and j atoms */
 783             qq00             = _mm_mul_pd(iq0,jq0);
 784             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 785                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 786
 787             /* Calculate table index by multiplying r with table scale and truncate to integer */
 788             rt               = _mm_mul_pd(r00,vftabscale);
 789             vfitab           = _mm_cvttpd_epi32(rt);
 790             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 791             vfitab           = _mm_slli_epi32(vfitab,2);
 792
 793             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 794             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 795             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 796             GMX_MM_TRANSPOSE2_PD(Y,F);
 797             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 798             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 799             GMX_MM_TRANSPOSE2_PD(G,H);
 800             Heps             = _mm_mul_pd(vfeps,H);
 801             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 802             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 803             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 804
 805             /* LENNARD-JONES DISPERSION/REPULSION */
 806
 807             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 808             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 809
 810             fscal            = _mm_add_pd(felec,fvdw);
 811
 812             /* Calculate temporary vectorial force */
 813             tx               = _mm_mul_pd(fscal,dx00);
 814             ty               = _mm_mul_pd(fscal,dy00);
 815             tz               = _mm_mul_pd(fscal,dz00);
 816
 817             /* Update vectorial force */
 818             fix0             = _mm_add_pd(fix0,tx);
 819             fiy0             = _mm_add_pd(fiy0,ty);
 820             fiz0             = _mm_add_pd(fiz0,tz);
 821
 822             fjx0             = _mm_add_pd(fjx0,tx);
 823             fjy0             = _mm_add_pd(fjy0,ty);
 824             fjz0             = _mm_add_pd(fjz0,tz);
 825
 826             /**************************
 827              * CALCULATE INTERACTIONS *
 828              **************************/
 829
 830             r10              = _mm_mul_pd(rsq10,rinv10);
 831
 832             /* Compute parameters for interactions between i and j atoms */
 833             qq10             = _mm_mul_pd(iq1,jq0);
 834
 835             /* Calculate table index by multiplying r with table scale and truncate to integer */
 836             rt               = _mm_mul_pd(r10,vftabscale);
 837             vfitab           = _mm_cvttpd_epi32(rt);
 838             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 839             vfitab           = _mm_slli_epi32(vfitab,2);
 840
 841             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 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             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 852
 853             fscal            = felec;
 854
 855             /* Calculate temporary vectorial force */
 856             tx               = _mm_mul_pd(fscal,dx10);
 857             ty               = _mm_mul_pd(fscal,dy10);
 858             tz               = _mm_mul_pd(fscal,dz10);
 859
 860             /* Update vectorial force */
 861             fix1             = _mm_add_pd(fix1,tx);
 862             fiy1             = _mm_add_pd(fiy1,ty);
 863             fiz1             = _mm_add_pd(fiz1,tz);
 864
 865             fjx0             = _mm_add_pd(fjx0,tx);
 866             fjy0             = _mm_add_pd(fjy0,ty);
 867             fjz0             = _mm_add_pd(fjz0,tz);
 868
 869             /**************************
 870              * CALCULATE INTERACTIONS *
 871              **************************/
 872
 873             r20              = _mm_mul_pd(rsq20,rinv20);
 874
 875             /* Compute parameters for interactions between i and j atoms */
 876             qq20             = _mm_mul_pd(iq2,jq0);
 877
 878             /* Calculate table index by multiplying r with table scale and truncate to integer */
 879             rt               = _mm_mul_pd(r20,vftabscale);
 880             vfitab           = _mm_cvttpd_epi32(rt);
 881             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 882             vfitab           = _mm_slli_epi32(vfitab,2);
 883
 884             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 885             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 886             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 887             GMX_MM_TRANSPOSE2_PD(Y,F);
 888             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 889             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 890             GMX_MM_TRANSPOSE2_PD(G,H);
 891             Heps             = _mm_mul_pd(vfeps,H);
 892             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 893             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 894             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 895
 896             fscal            = felec;
 897
 898             /* Calculate temporary vectorial force */
 899             tx               = _mm_mul_pd(fscal,dx20);
 900             ty               = _mm_mul_pd(fscal,dy20);
 901             tz               = _mm_mul_pd(fscal,dz20);
 902
 903             /* Update vectorial force */
 904             fix2             = _mm_add_pd(fix2,tx);
 905             fiy2             = _mm_add_pd(fiy2,ty);
 906             fiz2             = _mm_add_pd(fiz2,tz);
 907
 908             fjx0             = _mm_add_pd(fjx0,tx);
 909             fjy0             = _mm_add_pd(fjy0,ty);
 910             fjz0             = _mm_add_pd(fjz0,tz);
 911
 912             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 913
 914             /* Inner loop uses 128 flops */
 915         }
 916
 917         if(jidx<j_index_end)
 918         {
 919
 920             jnrA             = jjnr[jidx];
 921             j_coord_offsetA  = DIM*jnrA;
 922
 923             /* load j atom coordinates */
 924             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 925                                               &jx0,&jy0,&jz0);
 926
 927             /* Calculate displacement vector */
 928             dx00             = _mm_sub_pd(ix0,jx0);
 929             dy00             = _mm_sub_pd(iy0,jy0);
 930             dz00             = _mm_sub_pd(iz0,jz0);
 931             dx10             = _mm_sub_pd(ix1,jx0);
 932             dy10             = _mm_sub_pd(iy1,jy0);
 933             dz10             = _mm_sub_pd(iz1,jz0);
 934             dx20             = _mm_sub_pd(ix2,jx0);
 935             dy20             = _mm_sub_pd(iy2,jy0);
 936             dz20             = _mm_sub_pd(iz2,jz0);
 937
 938             /* Calculate squared distance and things based on it */
 939             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 940             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 941             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 942
 943             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 944             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 945             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 946
 947             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 948
 949             /* Load parameters for j particles */
 950             jq0              = _mm_load_sd(charge+jnrA+0);
 951             vdwjidx0A        = 2*vdwtype[jnrA+0];
 952
 953             fjx0             = _mm_setzero_pd();
 954             fjy0             = _mm_setzero_pd();
 955             fjz0             = _mm_setzero_pd();
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             r00              = _mm_mul_pd(rsq00,rinv00);
 962
 963             /* Compute parameters for interactions between i and j atoms */
 964             qq00             = _mm_mul_pd(iq0,jq0);
 965             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 966
 967             /* Calculate table index by multiplying r with table scale and truncate to integer */
 968             rt               = _mm_mul_pd(r00,vftabscale);
 969             vfitab           = _mm_cvttpd_epi32(rt);
 970             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 971             vfitab           = _mm_slli_epi32(vfitab,2);
 972
 973             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 974             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 975             F                = _mm_setzero_pd();
 976             GMX_MM_TRANSPOSE2_PD(Y,F);
 977             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 978             H                = _mm_setzero_pd();
 979             GMX_MM_TRANSPOSE2_PD(G,H);
 980             Heps             = _mm_mul_pd(vfeps,H);
 981             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 982             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 983             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 984
 985             /* LENNARD-JONES DISPERSION/REPULSION */
 986
 987             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 988             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 989
 990             fscal            = _mm_add_pd(felec,fvdw);
 991
 992             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 993
 994             /* Calculate temporary vectorial force */
 995             tx               = _mm_mul_pd(fscal,dx00);
 996             ty               = _mm_mul_pd(fscal,dy00);
 997             tz               = _mm_mul_pd(fscal,dz00);
 998
 999             /* Update vectorial force */
1000             fix0             = _mm_add_pd(fix0,tx);
1001             fiy0             = _mm_add_pd(fiy0,ty);
1002             fiz0             = _mm_add_pd(fiz0,tz);
1003
1004             fjx0             = _mm_add_pd(fjx0,tx);
1005             fjy0             = _mm_add_pd(fjy0,ty);
1006             fjz0             = _mm_add_pd(fjz0,tz);
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             r10              = _mm_mul_pd(rsq10,rinv10);
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq10             = _mm_mul_pd(iq1,jq0);
1016
1017             /* Calculate table index by multiplying r with table scale and truncate to integer */
1018             rt               = _mm_mul_pd(r10,vftabscale);
1019             vfitab           = _mm_cvttpd_epi32(rt);
1020             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1021             vfitab           = _mm_slli_epi32(vfitab,2);
1022
1023             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1024             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1025             F                = _mm_setzero_pd();
1026             GMX_MM_TRANSPOSE2_PD(Y,F);
1027             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1028             H                = _mm_setzero_pd();
1029             GMX_MM_TRANSPOSE2_PD(G,H);
1030             Heps             = _mm_mul_pd(vfeps,H);
1031             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1032             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1033             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1034
1035             fscal            = felec;
1036
1037             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm_mul_pd(fscal,dx10);
1041             ty               = _mm_mul_pd(fscal,dy10);
1042             tz               = _mm_mul_pd(fscal,dz10);
1043
1044             /* Update vectorial force */
1045             fix1             = _mm_add_pd(fix1,tx);
1046             fiy1             = _mm_add_pd(fiy1,ty);
1047             fiz1             = _mm_add_pd(fiz1,tz);
1048
1049             fjx0             = _mm_add_pd(fjx0,tx);
1050             fjy0             = _mm_add_pd(fjy0,ty);
1051             fjz0             = _mm_add_pd(fjz0,tz);
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             r20              = _mm_mul_pd(rsq20,rinv20);
1058
1059             /* Compute parameters for interactions between i and j atoms */
1060             qq20             = _mm_mul_pd(iq2,jq0);
1061
1062             /* Calculate table index by multiplying r with table scale and truncate to integer */
1063             rt               = _mm_mul_pd(r20,vftabscale);
1064             vfitab           = _mm_cvttpd_epi32(rt);
1065             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1066             vfitab           = _mm_slli_epi32(vfitab,2);
1067
1068             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1069             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1070             F                = _mm_setzero_pd();
1071             GMX_MM_TRANSPOSE2_PD(Y,F);
1072             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1073             H                = _mm_setzero_pd();
1074             GMX_MM_TRANSPOSE2_PD(G,H);
1075             Heps             = _mm_mul_pd(vfeps,H);
1076             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1077             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1078             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1079
1080             fscal            = felec;
1081
1082             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm_mul_pd(fscal,dx20);
1086             ty               = _mm_mul_pd(fscal,dy20);
1087             tz               = _mm_mul_pd(fscal,dz20);
1088
1089             /* Update vectorial force */
1090             fix2             = _mm_add_pd(fix2,tx);
1091             fiy2             = _mm_add_pd(fiy2,ty);
1092             fiz2             = _mm_add_pd(fiz2,tz);
1093
1094             fjx0             = _mm_add_pd(fjx0,tx);
1095             fjy0             = _mm_add_pd(fjy0,ty);
1096             fjz0             = _mm_add_pd(fjz0,tz);
1097
1098             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1099
1100             /* Inner loop uses 128 flops */
1101         }
1102
1103         /* End of innermost loop */
1104
1105         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1106                                               f+i_coord_offset,fshift+i_shift_offset);
1107
1108         /* Increment number of inner iterations */
1109         inneriter                  += j_index_end - j_index_start;
1110
1111         /* Outer loop uses 18 flops */
1112     }
1113
1114     /* Increment number of outer iterations */
1115     outeriter        += nri;
1116
1117     /* Update outer/inner flops */
1118
1119     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*128);
1120 }