src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_sse2_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS 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_VdwCSTab_GeomW3P1_VF_sse2_double
  51  * Electrostatics interaction: CubicSplineTable
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCSTab_VdwCSTab_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_vdw->data;
 124     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->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             /* Load parameters for j particles */
 208             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 209             vdwjidx0A        = 2*vdwtype[jnrA+0];
 210             vdwjidx0B        = 2*vdwtype[jnrB+0];
 211
 212             fjx0             = _mm_setzero_pd();
 213             fjy0             = _mm_setzero_pd();
 214             fjz0             = _mm_setzero_pd();
 215
 216             /**************************
 217              * CALCULATE INTERACTIONS *
 218              **************************/
 219
 220             r00              = _mm_mul_pd(rsq00,rinv00);
 221
 222             /* Compute parameters for interactions between i and j atoms */
 223             qq00             = _mm_mul_pd(iq0,jq0);
 224             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 225                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 226
 227             /* Calculate table index by multiplying r with table scale and truncate to integer */
 228             rt               = _mm_mul_pd(r00,vftabscale);
 229             vfitab           = _mm_cvttpd_epi32(rt);
 230             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 231             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 232
 233             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 234             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 235             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 236             GMX_MM_TRANSPOSE2_PD(Y,F);
 237             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 238             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 239             GMX_MM_TRANSPOSE2_PD(G,H);
 240             Heps             = _mm_mul_pd(vfeps,H);
 241             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 242             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 243             velec            = _mm_mul_pd(qq00,VV);
 244             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 245             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 246
 247             /* CUBIC SPLINE TABLE DISPERSION */
 248             vfitab           = _mm_add_epi32(vfitab,ifour);
 249             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 250             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 251             GMX_MM_TRANSPOSE2_PD(Y,F);
 252             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 253             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 254             GMX_MM_TRANSPOSE2_PD(G,H);
 255             Heps             = _mm_mul_pd(vfeps,H);
 256             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 257             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 258             vvdw6            = _mm_mul_pd(c6_00,VV);
 259             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 260             fvdw6            = _mm_mul_pd(c6_00,FF);
 261
 262             /* CUBIC SPLINE TABLE REPULSION */
 263             vfitab           = _mm_add_epi32(vfitab,ifour);
 264             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 265             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 266             GMX_MM_TRANSPOSE2_PD(Y,F);
 267             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 268             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 269             GMX_MM_TRANSPOSE2_PD(G,H);
 270             Heps             = _mm_mul_pd(vfeps,H);
 271             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 272             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 273             vvdw12           = _mm_mul_pd(c12_00,VV);
 274             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 275             fvdw12           = _mm_mul_pd(c12_00,FF);
 276             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 277             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             velecsum         = _mm_add_pd(velecsum,velec);
 281             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 282
 283             fscal            = _mm_add_pd(felec,fvdw);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm_mul_pd(fscal,dx00);
 287             ty               = _mm_mul_pd(fscal,dy00);
 288             tz               = _mm_mul_pd(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm_add_pd(fix0,tx);
 292             fiy0             = _mm_add_pd(fiy0,ty);
 293             fiz0             = _mm_add_pd(fiz0,tz);
 294
 295             fjx0             = _mm_add_pd(fjx0,tx);
 296             fjy0             = _mm_add_pd(fjy0,ty);
 297             fjz0             = _mm_add_pd(fjz0,tz);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             r10              = _mm_mul_pd(rsq10,rinv10);
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_pd(iq1,jq0);
 307
 308             /* Calculate table index by multiplying r with table scale and truncate to integer */
 309             rt               = _mm_mul_pd(r10,vftabscale);
 310             vfitab           = _mm_cvttpd_epi32(rt);
 311             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 312             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 313
 314             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 315             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 316             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 317             GMX_MM_TRANSPOSE2_PD(Y,F);
 318             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 319             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 320             GMX_MM_TRANSPOSE2_PD(G,H);
 321             Heps             = _mm_mul_pd(vfeps,H);
 322             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 323             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 324             velec            = _mm_mul_pd(qq10,VV);
 325             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 326             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_pd(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_pd(fscal,dx10);
 335             ty               = _mm_mul_pd(fscal,dy10);
 336             tz               = _mm_mul_pd(fscal,dz10);
 337
 338             /* Update vectorial force */
 339             fix1             = _mm_add_pd(fix1,tx);
 340             fiy1             = _mm_add_pd(fiy1,ty);
 341             fiz1             = _mm_add_pd(fiz1,tz);
 342
 343             fjx0             = _mm_add_pd(fjx0,tx);
 344             fjy0             = _mm_add_pd(fjy0,ty);
 345             fjz0             = _mm_add_pd(fjz0,tz);
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             r20              = _mm_mul_pd(rsq20,rinv20);
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq20             = _mm_mul_pd(iq2,jq0);
 355
 356             /* Calculate table index by multiplying r with table scale and truncate to integer */
 357             rt               = _mm_mul_pd(r20,vftabscale);
 358             vfitab           = _mm_cvttpd_epi32(rt);
 359             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 360             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 361
 362             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 363             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 364             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 365             GMX_MM_TRANSPOSE2_PD(Y,F);
 366             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 367             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 368             GMX_MM_TRANSPOSE2_PD(G,H);
 369             Heps             = _mm_mul_pd(vfeps,H);
 370             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 371             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 372             velec            = _mm_mul_pd(qq20,VV);
 373             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 374             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm_add_pd(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm_mul_pd(fscal,dx20);
 383             ty               = _mm_mul_pd(fscal,dy20);
 384             tz               = _mm_mul_pd(fscal,dz20);
 385
 386             /* Update vectorial force */
 387             fix2             = _mm_add_pd(fix2,tx);
 388             fiy2             = _mm_add_pd(fiy2,ty);
 389             fiz2             = _mm_add_pd(fiz2,tz);
 390
 391             fjx0             = _mm_add_pd(fjx0,tx);
 392             fjy0             = _mm_add_pd(fjy0,ty);
 393             fjz0             = _mm_add_pd(fjz0,tz);
 394
 395             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 396
 397             /* Inner loop uses 162 flops */
 398         }
 399
 400         if(jidx<j_index_end)
 401         {
 402
 403             jnrA             = jjnr[jidx];
 404             j_coord_offsetA  = DIM*jnrA;
 405
 406             /* load j atom coordinates */
 407             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 408                                               &jx0,&jy0,&jz0);
 409
 410             /* Calculate displacement vector */
 411             dx00             = _mm_sub_pd(ix0,jx0);
 412             dy00             = _mm_sub_pd(iy0,jy0);
 413             dz00             = _mm_sub_pd(iz0,jz0);
 414             dx10             = _mm_sub_pd(ix1,jx0);
 415             dy10             = _mm_sub_pd(iy1,jy0);
 416             dz10             = _mm_sub_pd(iz1,jz0);
 417             dx20             = _mm_sub_pd(ix2,jx0);
 418             dy20             = _mm_sub_pd(iy2,jy0);
 419             dz20             = _mm_sub_pd(iz2,jz0);
 420
 421             /* Calculate squared distance and things based on it */
 422             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 423             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 424             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 425
 426             rinv00           = sse2_invsqrt_d(rsq00);
 427             rinv10           = sse2_invsqrt_d(rsq10);
 428             rinv20           = sse2_invsqrt_d(rsq20);
 429
 430             /* Load parameters for j particles */
 431             jq0              = _mm_load_sd(charge+jnrA+0);
 432             vdwjidx0A        = 2*vdwtype[jnrA+0];
 433
 434             fjx0             = _mm_setzero_pd();
 435             fjy0             = _mm_setzero_pd();
 436             fjz0             = _mm_setzero_pd();
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             r00              = _mm_mul_pd(rsq00,rinv00);
 443
 444             /* Compute parameters for interactions between i and j atoms */
 445             qq00             = _mm_mul_pd(iq0,jq0);
 446             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 447
 448             /* Calculate table index by multiplying r with table scale and truncate to integer */
 449             rt               = _mm_mul_pd(r00,vftabscale);
 450             vfitab           = _mm_cvttpd_epi32(rt);
 451             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 452             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 453
 454             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 455             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 456             F                = _mm_setzero_pd();
 457             GMX_MM_TRANSPOSE2_PD(Y,F);
 458             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 459             H                = _mm_setzero_pd();
 460             GMX_MM_TRANSPOSE2_PD(G,H);
 461             Heps             = _mm_mul_pd(vfeps,H);
 462             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 463             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 464             velec            = _mm_mul_pd(qq00,VV);
 465             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 466             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 467
 468             /* CUBIC SPLINE TABLE DISPERSION */
 469             vfitab           = _mm_add_epi32(vfitab,ifour);
 470             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 471             F                = _mm_setzero_pd();
 472             GMX_MM_TRANSPOSE2_PD(Y,F);
 473             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 474             H                = _mm_setzero_pd();
 475             GMX_MM_TRANSPOSE2_PD(G,H);
 476             Heps             = _mm_mul_pd(vfeps,H);
 477             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 478             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 479             vvdw6            = _mm_mul_pd(c6_00,VV);
 480             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 481             fvdw6            = _mm_mul_pd(c6_00,FF);
 482
 483             /* CUBIC SPLINE TABLE REPULSION */
 484             vfitab           = _mm_add_epi32(vfitab,ifour);
 485             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 486             F                = _mm_setzero_pd();
 487             GMX_MM_TRANSPOSE2_PD(Y,F);
 488             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 489             H                = _mm_setzero_pd();
 490             GMX_MM_TRANSPOSE2_PD(G,H);
 491             Heps             = _mm_mul_pd(vfeps,H);
 492             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 493             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 494             vvdw12           = _mm_mul_pd(c12_00,VV);
 495             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 496             fvdw12           = _mm_mul_pd(c12_00,FF);
 497             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 498             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 502             velecsum         = _mm_add_pd(velecsum,velec);
 503             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 504             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 505
 506             fscal            = _mm_add_pd(felec,fvdw);
 507
 508             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 509
 510             /* Calculate temporary vectorial force */
 511             tx               = _mm_mul_pd(fscal,dx00);
 512             ty               = _mm_mul_pd(fscal,dy00);
 513             tz               = _mm_mul_pd(fscal,dz00);
 514
 515             /* Update vectorial force */
 516             fix0             = _mm_add_pd(fix0,tx);
 517             fiy0             = _mm_add_pd(fiy0,ty);
 518             fiz0             = _mm_add_pd(fiz0,tz);
 519
 520             fjx0             = _mm_add_pd(fjx0,tx);
 521             fjy0             = _mm_add_pd(fjy0,ty);
 522             fjz0             = _mm_add_pd(fjz0,tz);
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r10              = _mm_mul_pd(rsq10,rinv10);
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq10             = _mm_mul_pd(iq1,jq0);
 532
 533             /* Calculate table index by multiplying r with table scale and truncate to integer */
 534             rt               = _mm_mul_pd(r10,vftabscale);
 535             vfitab           = _mm_cvttpd_epi32(rt);
 536             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 537             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 538
 539             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 540             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 541             F                = _mm_setzero_pd();
 542             GMX_MM_TRANSPOSE2_PD(Y,F);
 543             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 544             H                = _mm_setzero_pd();
 545             GMX_MM_TRANSPOSE2_PD(G,H);
 546             Heps             = _mm_mul_pd(vfeps,H);
 547             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 548             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 549             velec            = _mm_mul_pd(qq10,VV);
 550             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 551             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 552
 553             /* Update potential sum for this i atom from the interaction with this j atom. */
 554             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 555             velecsum         = _mm_add_pd(velecsum,velec);
 556
 557             fscal            = felec;
 558
 559             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm_mul_pd(fscal,dx10);
 563             ty               = _mm_mul_pd(fscal,dy10);
 564             tz               = _mm_mul_pd(fscal,dz10);
 565
 566             /* Update vectorial force */
 567             fix1             = _mm_add_pd(fix1,tx);
 568             fiy1             = _mm_add_pd(fiy1,ty);
 569             fiz1             = _mm_add_pd(fiz1,tz);
 570
 571             fjx0             = _mm_add_pd(fjx0,tx);
 572             fjy0             = _mm_add_pd(fjy0,ty);
 573             fjz0             = _mm_add_pd(fjz0,tz);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             r20              = _mm_mul_pd(rsq20,rinv20);
 580
 581             /* Compute parameters for interactions between i and j atoms */
 582             qq20             = _mm_mul_pd(iq2,jq0);
 583
 584             /* Calculate table index by multiplying r with table scale and truncate to integer */
 585             rt               = _mm_mul_pd(r20,vftabscale);
 586             vfitab           = _mm_cvttpd_epi32(rt);
 587             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 588             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 589
 590             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 591             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 592             F                = _mm_setzero_pd();
 593             GMX_MM_TRANSPOSE2_PD(Y,F);
 594             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 595             H                = _mm_setzero_pd();
 596             GMX_MM_TRANSPOSE2_PD(G,H);
 597             Heps             = _mm_mul_pd(vfeps,H);
 598             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 599             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 600             velec            = _mm_mul_pd(qq20,VV);
 601             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 602             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 603
 604             /* Update potential sum for this i atom from the interaction with this j atom. */
 605             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 606             velecsum         = _mm_add_pd(velecsum,velec);
 607
 608             fscal            = felec;
 609
 610             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 611
 612             /* Calculate temporary vectorial force */
 613             tx               = _mm_mul_pd(fscal,dx20);
 614             ty               = _mm_mul_pd(fscal,dy20);
 615             tz               = _mm_mul_pd(fscal,dz20);
 616
 617             /* Update vectorial force */
 618             fix2             = _mm_add_pd(fix2,tx);
 619             fiy2             = _mm_add_pd(fiy2,ty);
 620             fiz2             = _mm_add_pd(fiz2,tz);
 621
 622             fjx0             = _mm_add_pd(fjx0,tx);
 623             fjy0             = _mm_add_pd(fjy0,ty);
 624             fjz0             = _mm_add_pd(fjz0,tz);
 625
 626             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 627
 628             /* Inner loop uses 162 flops */
 629         }
 630
 631         /* End of innermost loop */
 632
 633         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 634                                               f+i_coord_offset,fshift+i_shift_offset);
 635
 636         ggid                        = gid[iidx];
 637         /* Update potential energies */
 638         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 639         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 640
 641         /* Increment number of inner iterations */
 642         inneriter                  += j_index_end - j_index_start;
 643
 644         /* Outer loop uses 20 flops */
 645     }
 646
 647     /* Increment number of outer iterations */
 648     outeriter        += nri;
 649
 650     /* Update outer/inner flops */
 651
 652     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*162);
 653 }
 654 /*
 655  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse2_double
 656  * Electrostatics interaction: CubicSplineTable
 657  * VdW interaction:            CubicSplineTable
 658  * Geometry:                   Water3-Particle
 659  * Calculate force/pot:        Force
 660  */
 661 void
 662 nb_kernel_ElecCSTab_VdwCSTab_GeomW3P1_F_sse2_double
 663                     (t_nblist                    * gmx_restrict       nlist,
 664                      rvec                        * gmx_restrict          xx,
 665                      rvec                        * gmx_restrict          ff,
 666                      struct t_forcerec           * gmx_restrict          fr,
 667                      t_mdatoms                   * gmx_restrict     mdatoms,
 668                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 669                      t_nrnb                      * gmx_restrict        nrnb)
 670 {
 671     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 672      * just 0 for non-waters.
 673      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 674      * jnr indices corresponding to data put in the four positions in the SIMD register.
 675      */
 676     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 677     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 678     int              jnrA,jnrB;
 679     int              j_coord_offsetA,j_coord_offsetB;
 680     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 681     real             rcutoff_scalar;
 682     real             *shiftvec,*fshift,*x,*f;
 683     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 684     int              vdwioffset0;
 685     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 686     int              vdwioffset1;
 687     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 688     int              vdwioffset2;
 689     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 690     int              vdwjidx0A,vdwjidx0B;
 691     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 692     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 693     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 694     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 695     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 696     real             *charge;
 697     int              nvdwtype;
 698     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 699     int              *vdwtype;
 700     real             *vdwparam;
 701     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 702     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 703     __m128i          vfitab;
 704     __m128i          ifour       = _mm_set1_epi32(4);
 705     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 706     real             *vftab;
 707     __m128d          dummy_mask,cutoff_mask;
 708     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 709     __m128d          one     = _mm_set1_pd(1.0);
 710     __m128d          two     = _mm_set1_pd(2.0);
 711     x                = xx[0];
 712     f                = ff[0];
 713
 714     nri              = nlist->nri;
 715     iinr             = nlist->iinr;
 716     jindex           = nlist->jindex;
 717     jjnr             = nlist->jjnr;
 718     shiftidx         = nlist->shift;
 719     gid              = nlist->gid;
 720     shiftvec         = fr->shift_vec[0];
 721     fshift           = fr->fshift[0];
 722     facel            = _mm_set1_pd(fr->ic->epsfac);
 723     charge           = mdatoms->chargeA;
 724     nvdwtype         = fr->ntype;
 725     vdwparam         = fr->nbfp;
 726     vdwtype          = mdatoms->typeA;
 727
 728     vftab            = kernel_data->table_elec_vdw->data;
 729     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 730
 731     /* Setup water-specific parameters */
 732     inr              = nlist->iinr[0];
 733     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 734     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 735     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 736     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 737
 738     /* Avoid stupid compiler warnings */
 739     jnrA = jnrB = 0;
 740     j_coord_offsetA = 0;
 741     j_coord_offsetB = 0;
 742
 743     outeriter        = 0;
 744     inneriter        = 0;
 745
 746     /* Start outer loop over neighborlists */
 747     for(iidx=0; iidx<nri; iidx++)
 748     {
 749         /* Load shift vector for this list */
 750         i_shift_offset   = DIM*shiftidx[iidx];
 751
 752         /* Load limits for loop over neighbors */
 753         j_index_start    = jindex[iidx];
 754         j_index_end      = jindex[iidx+1];
 755
 756         /* Get outer coordinate index */
 757         inr              = iinr[iidx];
 758         i_coord_offset   = DIM*inr;
 759
 760         /* Load i particle coords and add shift vector */
 761         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 762                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 763
 764         fix0             = _mm_setzero_pd();
 765         fiy0             = _mm_setzero_pd();
 766         fiz0             = _mm_setzero_pd();
 767         fix1             = _mm_setzero_pd();
 768         fiy1             = _mm_setzero_pd();
 769         fiz1             = _mm_setzero_pd();
 770         fix2             = _mm_setzero_pd();
 771         fiy2             = _mm_setzero_pd();
 772         fiz2             = _mm_setzero_pd();
 773
 774         /* Start inner kernel loop */
 775         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 776         {
 777
 778             /* Get j neighbor index, and coordinate index */
 779             jnrA             = jjnr[jidx];
 780             jnrB             = jjnr[jidx+1];
 781             j_coord_offsetA  = DIM*jnrA;
 782             j_coord_offsetB  = DIM*jnrB;
 783
 784             /* load j atom coordinates */
 785             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 786                                               &jx0,&jy0,&jz0);
 787
 788             /* Calculate displacement vector */
 789             dx00             = _mm_sub_pd(ix0,jx0);
 790             dy00             = _mm_sub_pd(iy0,jy0);
 791             dz00             = _mm_sub_pd(iz0,jz0);
 792             dx10             = _mm_sub_pd(ix1,jx0);
 793             dy10             = _mm_sub_pd(iy1,jy0);
 794             dz10             = _mm_sub_pd(iz1,jz0);
 795             dx20             = _mm_sub_pd(ix2,jx0);
 796             dy20             = _mm_sub_pd(iy2,jy0);
 797             dz20             = _mm_sub_pd(iz2,jz0);
 798
 799             /* Calculate squared distance and things based on it */
 800             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 801             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 802             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 803
 804             rinv00           = sse2_invsqrt_d(rsq00);
 805             rinv10           = sse2_invsqrt_d(rsq10);
 806             rinv20           = sse2_invsqrt_d(rsq20);
 807
 808             /* Load parameters for j particles */
 809             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 810             vdwjidx0A        = 2*vdwtype[jnrA+0];
 811             vdwjidx0B        = 2*vdwtype[jnrB+0];
 812
 813             fjx0             = _mm_setzero_pd();
 814             fjy0             = _mm_setzero_pd();
 815             fjz0             = _mm_setzero_pd();
 816
 817             /**************************
 818              * CALCULATE INTERACTIONS *
 819              **************************/
 820
 821             r00              = _mm_mul_pd(rsq00,rinv00);
 822
 823             /* Compute parameters for interactions between i and j atoms */
 824             qq00             = _mm_mul_pd(iq0,jq0);
 825             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 826                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 827
 828             /* Calculate table index by multiplying r with table scale and truncate to integer */
 829             rt               = _mm_mul_pd(r00,vftabscale);
 830             vfitab           = _mm_cvttpd_epi32(rt);
 831             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 832             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 833
 834             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 835             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 836             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 837             GMX_MM_TRANSPOSE2_PD(Y,F);
 838             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 839             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 840             GMX_MM_TRANSPOSE2_PD(G,H);
 841             Heps             = _mm_mul_pd(vfeps,H);
 842             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 843             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 844             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 845
 846             /* CUBIC SPLINE TABLE DISPERSION */
 847             vfitab           = _mm_add_epi32(vfitab,ifour);
 848             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 849             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 850             GMX_MM_TRANSPOSE2_PD(Y,F);
 851             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 852             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 853             GMX_MM_TRANSPOSE2_PD(G,H);
 854             Heps             = _mm_mul_pd(vfeps,H);
 855             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 856             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 857             fvdw6            = _mm_mul_pd(c6_00,FF);
 858
 859             /* CUBIC SPLINE TABLE REPULSION */
 860             vfitab           = _mm_add_epi32(vfitab,ifour);
 861             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 862             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 863             GMX_MM_TRANSPOSE2_PD(Y,F);
 864             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 865             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 866             GMX_MM_TRANSPOSE2_PD(G,H);
 867             Heps             = _mm_mul_pd(vfeps,H);
 868             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 869             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 870             fvdw12           = _mm_mul_pd(c12_00,FF);
 871             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 872
 873             fscal            = _mm_add_pd(felec,fvdw);
 874
 875             /* Calculate temporary vectorial force */
 876             tx               = _mm_mul_pd(fscal,dx00);
 877             ty               = _mm_mul_pd(fscal,dy00);
 878             tz               = _mm_mul_pd(fscal,dz00);
 879
 880             /* Update vectorial force */
 881             fix0             = _mm_add_pd(fix0,tx);
 882             fiy0             = _mm_add_pd(fiy0,ty);
 883             fiz0             = _mm_add_pd(fiz0,tz);
 884
 885             fjx0             = _mm_add_pd(fjx0,tx);
 886             fjy0             = _mm_add_pd(fjy0,ty);
 887             fjz0             = _mm_add_pd(fjz0,tz);
 888
 889             /**************************
 890              * CALCULATE INTERACTIONS *
 891              **************************/
 892
 893             r10              = _mm_mul_pd(rsq10,rinv10);
 894
 895             /* Compute parameters for interactions between i and j atoms */
 896             qq10             = _mm_mul_pd(iq1,jq0);
 897
 898             /* Calculate table index by multiplying r with table scale and truncate to integer */
 899             rt               = _mm_mul_pd(r10,vftabscale);
 900             vfitab           = _mm_cvttpd_epi32(rt);
 901             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 902             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 903
 904             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 905             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 906             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 907             GMX_MM_TRANSPOSE2_PD(Y,F);
 908             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 909             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 910             GMX_MM_TRANSPOSE2_PD(G,H);
 911             Heps             = _mm_mul_pd(vfeps,H);
 912             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 913             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 914             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
 915
 916             fscal            = felec;
 917
 918             /* Calculate temporary vectorial force */
 919             tx               = _mm_mul_pd(fscal,dx10);
 920             ty               = _mm_mul_pd(fscal,dy10);
 921             tz               = _mm_mul_pd(fscal,dz10);
 922
 923             /* Update vectorial force */
 924             fix1             = _mm_add_pd(fix1,tx);
 925             fiy1             = _mm_add_pd(fiy1,ty);
 926             fiz1             = _mm_add_pd(fiz1,tz);
 927
 928             fjx0             = _mm_add_pd(fjx0,tx);
 929             fjy0             = _mm_add_pd(fjy0,ty);
 930             fjz0             = _mm_add_pd(fjz0,tz);
 931
 932             /**************************
 933              * CALCULATE INTERACTIONS *
 934              **************************/
 935
 936             r20              = _mm_mul_pd(rsq20,rinv20);
 937
 938             /* Compute parameters for interactions between i and j atoms */
 939             qq20             = _mm_mul_pd(iq2,jq0);
 940
 941             /* Calculate table index by multiplying r with table scale and truncate to integer */
 942             rt               = _mm_mul_pd(r20,vftabscale);
 943             vfitab           = _mm_cvttpd_epi32(rt);
 944             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 945             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 946
 947             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 948             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 949             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 950             GMX_MM_TRANSPOSE2_PD(Y,F);
 951             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 952             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 953             GMX_MM_TRANSPOSE2_PD(G,H);
 954             Heps             = _mm_mul_pd(vfeps,H);
 955             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 956             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 957             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
 958
 959             fscal            = felec;
 960
 961             /* Calculate temporary vectorial force */
 962             tx               = _mm_mul_pd(fscal,dx20);
 963             ty               = _mm_mul_pd(fscal,dy20);
 964             tz               = _mm_mul_pd(fscal,dz20);
 965
 966             /* Update vectorial force */
 967             fix2             = _mm_add_pd(fix2,tx);
 968             fiy2             = _mm_add_pd(fiy2,ty);
 969             fiz2             = _mm_add_pd(fiz2,tz);
 970
 971             fjx0             = _mm_add_pd(fjx0,tx);
 972             fjy0             = _mm_add_pd(fjy0,ty);
 973             fjz0             = _mm_add_pd(fjz0,tz);
 974
 975             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 976
 977             /* Inner loop uses 142 flops */
 978         }
 979
 980         if(jidx<j_index_end)
 981         {
 982
 983             jnrA             = jjnr[jidx];
 984             j_coord_offsetA  = DIM*jnrA;
 985
 986             /* load j atom coordinates */
 987             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 988                                               &jx0,&jy0,&jz0);
 989
 990             /* Calculate displacement vector */
 991             dx00             = _mm_sub_pd(ix0,jx0);
 992             dy00             = _mm_sub_pd(iy0,jy0);
 993             dz00             = _mm_sub_pd(iz0,jz0);
 994             dx10             = _mm_sub_pd(ix1,jx0);
 995             dy10             = _mm_sub_pd(iy1,jy0);
 996             dz10             = _mm_sub_pd(iz1,jz0);
 997             dx20             = _mm_sub_pd(ix2,jx0);
 998             dy20             = _mm_sub_pd(iy2,jy0);
 999             dz20             = _mm_sub_pd(iz2,jz0);
1000
1001             /* Calculate squared distance and things based on it */
1002             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1003             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1004             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1005
1006             rinv00           = sse2_invsqrt_d(rsq00);
1007             rinv10           = sse2_invsqrt_d(rsq10);
1008             rinv20           = sse2_invsqrt_d(rsq20);
1009
1010             /* Load parameters for j particles */
1011             jq0              = _mm_load_sd(charge+jnrA+0);
1012             vdwjidx0A        = 2*vdwtype[jnrA+0];
1013
1014             fjx0             = _mm_setzero_pd();
1015             fjy0             = _mm_setzero_pd();
1016             fjz0             = _mm_setzero_pd();
1017
1018             /**************************
1019              * CALCULATE INTERACTIONS *
1020              **************************/
1021
1022             r00              = _mm_mul_pd(rsq00,rinv00);
1023
1024             /* Compute parameters for interactions between i and j atoms */
1025             qq00             = _mm_mul_pd(iq0,jq0);
1026             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1027
1028             /* Calculate table index by multiplying r with table scale and truncate to integer */
1029             rt               = _mm_mul_pd(r00,vftabscale);
1030             vfitab           = _mm_cvttpd_epi32(rt);
1031             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1032             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1033
1034             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1035             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1036             F                = _mm_setzero_pd();
1037             GMX_MM_TRANSPOSE2_PD(Y,F);
1038             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1039             H                = _mm_setzero_pd();
1040             GMX_MM_TRANSPOSE2_PD(G,H);
1041             Heps             = _mm_mul_pd(vfeps,H);
1042             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1043             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1044             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
1045
1046             /* CUBIC SPLINE TABLE DISPERSION */
1047             vfitab           = _mm_add_epi32(vfitab,ifour);
1048             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1049             F                = _mm_setzero_pd();
1050             GMX_MM_TRANSPOSE2_PD(Y,F);
1051             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1052             H                = _mm_setzero_pd();
1053             GMX_MM_TRANSPOSE2_PD(G,H);
1054             Heps             = _mm_mul_pd(vfeps,H);
1055             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1056             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1057             fvdw6            = _mm_mul_pd(c6_00,FF);
1058
1059             /* CUBIC SPLINE TABLE REPULSION */
1060             vfitab           = _mm_add_epi32(vfitab,ifour);
1061             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1062             F                = _mm_setzero_pd();
1063             GMX_MM_TRANSPOSE2_PD(Y,F);
1064             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1065             H                = _mm_setzero_pd();
1066             GMX_MM_TRANSPOSE2_PD(G,H);
1067             Heps             = _mm_mul_pd(vfeps,H);
1068             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1069             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1070             fvdw12           = _mm_mul_pd(c12_00,FF);
1071             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1072
1073             fscal            = _mm_add_pd(felec,fvdw);
1074
1075             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1076
1077             /* Calculate temporary vectorial force */
1078             tx               = _mm_mul_pd(fscal,dx00);
1079             ty               = _mm_mul_pd(fscal,dy00);
1080             tz               = _mm_mul_pd(fscal,dz00);
1081
1082             /* Update vectorial force */
1083             fix0             = _mm_add_pd(fix0,tx);
1084             fiy0             = _mm_add_pd(fiy0,ty);
1085             fiz0             = _mm_add_pd(fiz0,tz);
1086
1087             fjx0             = _mm_add_pd(fjx0,tx);
1088             fjy0             = _mm_add_pd(fjy0,ty);
1089             fjz0             = _mm_add_pd(fjz0,tz);
1090
1091             /**************************
1092              * CALCULATE INTERACTIONS *
1093              **************************/
1094
1095             r10              = _mm_mul_pd(rsq10,rinv10);
1096
1097             /* Compute parameters for interactions between i and j atoms */
1098             qq10             = _mm_mul_pd(iq1,jq0);
1099
1100             /* Calculate table index by multiplying r with table scale and truncate to integer */
1101             rt               = _mm_mul_pd(r10,vftabscale);
1102             vfitab           = _mm_cvttpd_epi32(rt);
1103             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1104             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1105
1106             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1107             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1108             F                = _mm_setzero_pd();
1109             GMX_MM_TRANSPOSE2_PD(Y,F);
1110             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1111             H                = _mm_setzero_pd();
1112             GMX_MM_TRANSPOSE2_PD(G,H);
1113             Heps             = _mm_mul_pd(vfeps,H);
1114             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1115             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1116             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq10,FF),_mm_mul_pd(vftabscale,rinv10)));
1117
1118             fscal            = felec;
1119
1120             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1121
1122             /* Calculate temporary vectorial force */
1123             tx               = _mm_mul_pd(fscal,dx10);
1124             ty               = _mm_mul_pd(fscal,dy10);
1125             tz               = _mm_mul_pd(fscal,dz10);
1126
1127             /* Update vectorial force */
1128             fix1             = _mm_add_pd(fix1,tx);
1129             fiy1             = _mm_add_pd(fiy1,ty);
1130             fiz1             = _mm_add_pd(fiz1,tz);
1131
1132             fjx0             = _mm_add_pd(fjx0,tx);
1133             fjy0             = _mm_add_pd(fjy0,ty);
1134             fjz0             = _mm_add_pd(fjz0,tz);
1135
1136             /**************************
1137              * CALCULATE INTERACTIONS *
1138              **************************/
1139
1140             r20              = _mm_mul_pd(rsq20,rinv20);
1141
1142             /* Compute parameters for interactions between i and j atoms */
1143             qq20             = _mm_mul_pd(iq2,jq0);
1144
1145             /* Calculate table index by multiplying r with table scale and truncate to integer */
1146             rt               = _mm_mul_pd(r20,vftabscale);
1147             vfitab           = _mm_cvttpd_epi32(rt);
1148             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
1149             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1150
1151             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1152             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
1153             F                = _mm_setzero_pd();
1154             GMX_MM_TRANSPOSE2_PD(Y,F);
1155             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
1156             H                = _mm_setzero_pd();
1157             GMX_MM_TRANSPOSE2_PD(G,H);
1158             Heps             = _mm_mul_pd(vfeps,H);
1159             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
1160             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
1161             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq20,FF),_mm_mul_pd(vftabscale,rinv20)));
1162
1163             fscal            = felec;
1164
1165             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1166
1167             /* Calculate temporary vectorial force */
1168             tx               = _mm_mul_pd(fscal,dx20);
1169             ty               = _mm_mul_pd(fscal,dy20);
1170             tz               = _mm_mul_pd(fscal,dz20);
1171
1172             /* Update vectorial force */
1173             fix2             = _mm_add_pd(fix2,tx);
1174             fiy2             = _mm_add_pd(fiy2,ty);
1175             fiz2             = _mm_add_pd(fiz2,tz);
1176
1177             fjx0             = _mm_add_pd(fjx0,tx);
1178             fjy0             = _mm_add_pd(fjy0,ty);
1179             fjz0             = _mm_add_pd(fjz0,tz);
1180
1181             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1182
1183             /* Inner loop uses 142 flops */
1184         }
1185
1186         /* End of innermost loop */
1187
1188         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1189                                               f+i_coord_offset,fshift+i_shift_offset);
1190
1191         /* Increment number of inner iterations */
1192         inneriter                  += j_index_end - j_index_start;
1193
1194         /* Outer loop uses 18 flops */
1195     }
1196
1197     /* Increment number of outer iterations */
1198     outeriter        += nri;
1199
1200     /* Update outer/inner flops */
1201
1202     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*142);
1203 }