src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse4_1_double
  54  * Electrostatics interaction: Coulomb
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwjidx0A,vdwjidx0B;
  85     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  86     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  87     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  88     real             *charge;
  89     int              nvdwtype;
  90     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  91     int              *vdwtype;
  92     real             *vdwparam;
  93     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  94     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  95     __m128i          vfitab;
  96     __m128i          ifour       = _mm_set1_epi32(4);
  97     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  98     real             *vftab;
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     nvdwtype         = fr->ntype;
 117     vdwparam         = fr->nbfp;
 118     vdwtype          = mdatoms->typeA;
 119
 120     vftab            = kernel_data->table_vdw->data;
 121     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 147
 148         fix0             = _mm_setzero_pd();
 149         fiy0             = _mm_setzero_pd();
 150         fiz0             = _mm_setzero_pd();
 151
 152         /* Load parameters for i particles */
 153         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 154         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_pd();
 158         vvdwsum          = _mm_setzero_pd();
 159
 160         /* Start inner kernel loop */
 161         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 162         {
 163
 164             /* Get j neighbor index, and coordinate index */
 165             jnrA             = jjnr[jidx];
 166             jnrB             = jjnr[jidx+1];
 167             j_coord_offsetA  = DIM*jnrA;
 168             j_coord_offsetB  = DIM*jnrB;
 169
 170             /* load j atom coordinates */
 171             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_pd(ix0,jx0);
 176             dy00             = _mm_sub_pd(iy0,jy0);
 177             dz00             = _mm_sub_pd(iz0,jz0);
 178
 179             /* Calculate squared distance and things based on it */
 180             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 181
 182             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 183
 184             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 185
 186             /* Load parameters for j particles */
 187             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 188             vdwjidx0A        = 2*vdwtype[jnrA+0];
 189             vdwjidx0B        = 2*vdwtype[jnrB+0];
 190
 191             /**************************
 192              * CALCULATE INTERACTIONS *
 193              **************************/
 194
 195             r00              = _mm_mul_pd(rsq00,rinv00);
 196
 197             /* Compute parameters for interactions between i and j atoms */
 198             qq00             = _mm_mul_pd(iq0,jq0);
 199             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 200                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 201
 202             /* Calculate table index by multiplying r with table scale and truncate to integer */
 203             rt               = _mm_mul_pd(r00,vftabscale);
 204             vfitab           = _mm_cvttpd_epi32(rt);
 205             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 206             vfitab           = _mm_slli_epi32(vfitab,3);
 207
 208             /* COULOMB ELECTROSTATICS */
 209             velec            = _mm_mul_pd(qq00,rinv00);
 210             felec            = _mm_mul_pd(velec,rinvsq00);
 211
 212             /* CUBIC SPLINE TABLE DISPERSION */
 213             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 214             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 215             GMX_MM_TRANSPOSE2_PD(Y,F);
 216             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 217             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 218             GMX_MM_TRANSPOSE2_PD(G,H);
 219             Heps             = _mm_mul_pd(vfeps,H);
 220             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 221             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 222             vvdw6            = _mm_mul_pd(c6_00,VV);
 223             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 224             fvdw6            = _mm_mul_pd(c6_00,FF);
 225
 226             /* CUBIC SPLINE TABLE REPULSION */
 227             vfitab           = _mm_add_epi32(vfitab,ifour);
 228             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 229             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 230             GMX_MM_TRANSPOSE2_PD(Y,F);
 231             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 232             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 233             GMX_MM_TRANSPOSE2_PD(G,H);
 234             Heps             = _mm_mul_pd(vfeps,H);
 235             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 236             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 237             vvdw12           = _mm_mul_pd(c12_00,VV);
 238             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 239             fvdw12           = _mm_mul_pd(c12_00,FF);
 240             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 241             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velecsum         = _mm_add_pd(velecsum,velec);
 245             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 246
 247             fscal            = _mm_add_pd(felec,fvdw);
 248
 249             /* Calculate temporary vectorial force */
 250             tx               = _mm_mul_pd(fscal,dx00);
 251             ty               = _mm_mul_pd(fscal,dy00);
 252             tz               = _mm_mul_pd(fscal,dz00);
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_add_pd(fix0,tx);
 256             fiy0             = _mm_add_pd(fiy0,ty);
 257             fiz0             = _mm_add_pd(fiz0,tz);
 258
 259             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 260
 261             /* Inner loop uses 63 flops */
 262         }
 263
 264         if(jidx<j_index_end)
 265         {
 266
 267             jnrA             = jjnr[jidx];
 268             j_coord_offsetA  = DIM*jnrA;
 269
 270             /* load j atom coordinates */
 271             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 272                                               &jx0,&jy0,&jz0);
 273
 274             /* Calculate displacement vector */
 275             dx00             = _mm_sub_pd(ix0,jx0);
 276             dy00             = _mm_sub_pd(iy0,jy0);
 277             dz00             = _mm_sub_pd(iz0,jz0);
 278
 279             /* Calculate squared distance and things based on it */
 280             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 281
 282             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 283
 284             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 285
 286             /* Load parameters for j particles */
 287             jq0              = _mm_load_sd(charge+jnrA+0);
 288             vdwjidx0A        = 2*vdwtype[jnrA+0];
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             r00              = _mm_mul_pd(rsq00,rinv00);
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq00             = _mm_mul_pd(iq0,jq0);
 298             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 299
 300             /* Calculate table index by multiplying r with table scale and truncate to integer */
 301             rt               = _mm_mul_pd(r00,vftabscale);
 302             vfitab           = _mm_cvttpd_epi32(rt);
 303             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 304             vfitab           = _mm_slli_epi32(vfitab,3);
 305
 306             /* COULOMB ELECTROSTATICS */
 307             velec            = _mm_mul_pd(qq00,rinv00);
 308             felec            = _mm_mul_pd(velec,rinvsq00);
 309
 310             /* CUBIC SPLINE TABLE DISPERSION */
 311             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 312             F                = _mm_setzero_pd();
 313             GMX_MM_TRANSPOSE2_PD(Y,F);
 314             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 315             H                = _mm_setzero_pd();
 316             GMX_MM_TRANSPOSE2_PD(G,H);
 317             Heps             = _mm_mul_pd(vfeps,H);
 318             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 319             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 320             vvdw6            = _mm_mul_pd(c6_00,VV);
 321             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 322             fvdw6            = _mm_mul_pd(c6_00,FF);
 323
 324             /* CUBIC SPLINE TABLE REPULSION */
 325             vfitab           = _mm_add_epi32(vfitab,ifour);
 326             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 327             F                = _mm_setzero_pd();
 328             GMX_MM_TRANSPOSE2_PD(Y,F);
 329             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 330             H                = _mm_setzero_pd();
 331             GMX_MM_TRANSPOSE2_PD(G,H);
 332             Heps             = _mm_mul_pd(vfeps,H);
 333             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 334             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 335             vvdw12           = _mm_mul_pd(c12_00,VV);
 336             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 337             fvdw12           = _mm_mul_pd(c12_00,FF);
 338             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 339             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 343             velecsum         = _mm_add_pd(velecsum,velec);
 344             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 345             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 346
 347             fscal            = _mm_add_pd(felec,fvdw);
 348
 349             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm_mul_pd(fscal,dx00);
 353             ty               = _mm_mul_pd(fscal,dy00);
 354             tz               = _mm_mul_pd(fscal,dz00);
 355
 356             /* Update vectorial force */
 357             fix0             = _mm_add_pd(fix0,tx);
 358             fiy0             = _mm_add_pd(fiy0,ty);
 359             fiz0             = _mm_add_pd(fiz0,tz);
 360
 361             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 362
 363             /* Inner loop uses 63 flops */
 364         }
 365
 366         /* End of innermost loop */
 367
 368         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 369                                               f+i_coord_offset,fshift+i_shift_offset);
 370
 371         ggid                        = gid[iidx];
 372         /* Update potential energies */
 373         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 374         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 375
 376         /* Increment number of inner iterations */
 377         inneriter                  += j_index_end - j_index_start;
 378
 379         /* Outer loop uses 9 flops */
 380     }
 381
 382     /* Increment number of outer iterations */
 383     outeriter        += nri;
 384
 385     /* Update outer/inner flops */
 386
 387     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*63);
 388 }
 389 /*
 390  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse4_1_double
 391  * Electrostatics interaction: Coulomb
 392  * VdW interaction:            CubicSplineTable
 393  * Geometry:                   Particle-Particle
 394  * Calculate force/pot:        Force
 395  */
 396 void
 397 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse4_1_double
 398                     (t_nblist                    * gmx_restrict       nlist,
 399                      rvec                        * gmx_restrict          xx,
 400                      rvec                        * gmx_restrict          ff,
 401                      t_forcerec                  * gmx_restrict          fr,
 402                      t_mdatoms                   * gmx_restrict     mdatoms,
 403                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 404                      t_nrnb                      * gmx_restrict        nrnb)
 405 {
 406     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 407      * just 0 for non-waters.
 408      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 409      * jnr indices corresponding to data put in the four positions in the SIMD register.
 410      */
 411     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 412     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 413     int              jnrA,jnrB;
 414     int              j_coord_offsetA,j_coord_offsetB;
 415     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 416     real             rcutoff_scalar;
 417     real             *shiftvec,*fshift,*x,*f;
 418     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 419     int              vdwioffset0;
 420     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 421     int              vdwjidx0A,vdwjidx0B;
 422     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 423     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 424     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 425     real             *charge;
 426     int              nvdwtype;
 427     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 428     int              *vdwtype;
 429     real             *vdwparam;
 430     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 431     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 432     __m128i          vfitab;
 433     __m128i          ifour       = _mm_set1_epi32(4);
 434     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 435     real             *vftab;
 436     __m128d          dummy_mask,cutoff_mask;
 437     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 438     __m128d          one     = _mm_set1_pd(1.0);
 439     __m128d          two     = _mm_set1_pd(2.0);
 440     x                = xx[0];
 441     f                = ff[0];
 442
 443     nri              = nlist->nri;
 444     iinr             = nlist->iinr;
 445     jindex           = nlist->jindex;
 446     jjnr             = nlist->jjnr;
 447     shiftidx         = nlist->shift;
 448     gid              = nlist->gid;
 449     shiftvec         = fr->shift_vec[0];
 450     fshift           = fr->fshift[0];
 451     facel            = _mm_set1_pd(fr->epsfac);
 452     charge           = mdatoms->chargeA;
 453     nvdwtype         = fr->ntype;
 454     vdwparam         = fr->nbfp;
 455     vdwtype          = mdatoms->typeA;
 456
 457     vftab            = kernel_data->table_vdw->data;
 458     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 459
 460     /* Avoid stupid compiler warnings */
 461     jnrA = jnrB = 0;
 462     j_coord_offsetA = 0;
 463     j_coord_offsetB = 0;
 464
 465     outeriter        = 0;
 466     inneriter        = 0;
 467
 468     /* Start outer loop over neighborlists */
 469     for(iidx=0; iidx<nri; iidx++)
 470     {
 471         /* Load shift vector for this list */
 472         i_shift_offset   = DIM*shiftidx[iidx];
 473
 474         /* Load limits for loop over neighbors */
 475         j_index_start    = jindex[iidx];
 476         j_index_end      = jindex[iidx+1];
 477
 478         /* Get outer coordinate index */
 479         inr              = iinr[iidx];
 480         i_coord_offset   = DIM*inr;
 481
 482         /* Load i particle coords and add shift vector */
 483         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 484
 485         fix0             = _mm_setzero_pd();
 486         fiy0             = _mm_setzero_pd();
 487         fiz0             = _mm_setzero_pd();
 488
 489         /* Load parameters for i particles */
 490         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 491         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 492
 493         /* Start inner kernel loop */
 494         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 495         {
 496
 497             /* Get j neighbor index, and coordinate index */
 498             jnrA             = jjnr[jidx];
 499             jnrB             = jjnr[jidx+1];
 500             j_coord_offsetA  = DIM*jnrA;
 501             j_coord_offsetB  = DIM*jnrB;
 502
 503             /* load j atom coordinates */
 504             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 505                                               &jx0,&jy0,&jz0);
 506
 507             /* Calculate displacement vector */
 508             dx00             = _mm_sub_pd(ix0,jx0);
 509             dy00             = _mm_sub_pd(iy0,jy0);
 510             dz00             = _mm_sub_pd(iz0,jz0);
 511
 512             /* Calculate squared distance and things based on it */
 513             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 514
 515             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 516
 517             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 521             vdwjidx0A        = 2*vdwtype[jnrA+0];
 522             vdwjidx0B        = 2*vdwtype[jnrB+0];
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r00              = _mm_mul_pd(rsq00,rinv00);
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq00             = _mm_mul_pd(iq0,jq0);
 532             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 533                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 534
 535             /* Calculate table index by multiplying r with table scale and truncate to integer */
 536             rt               = _mm_mul_pd(r00,vftabscale);
 537             vfitab           = _mm_cvttpd_epi32(rt);
 538             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 539             vfitab           = _mm_slli_epi32(vfitab,3);
 540
 541             /* COULOMB ELECTROSTATICS */
 542             velec            = _mm_mul_pd(qq00,rinv00);
 543             felec            = _mm_mul_pd(velec,rinvsq00);
 544
 545             /* CUBIC SPLINE TABLE DISPERSION */
 546             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 547             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 548             GMX_MM_TRANSPOSE2_PD(Y,F);
 549             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 550             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 551             GMX_MM_TRANSPOSE2_PD(G,H);
 552             Heps             = _mm_mul_pd(vfeps,H);
 553             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 554             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 555             fvdw6            = _mm_mul_pd(c6_00,FF);
 556
 557             /* CUBIC SPLINE TABLE REPULSION */
 558             vfitab           = _mm_add_epi32(vfitab,ifour);
 559             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 560             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 561             GMX_MM_TRANSPOSE2_PD(Y,F);
 562             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 563             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 564             GMX_MM_TRANSPOSE2_PD(G,H);
 565             Heps             = _mm_mul_pd(vfeps,H);
 566             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 567             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 568             fvdw12           = _mm_mul_pd(c12_00,FF);
 569             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 570
 571             fscal            = _mm_add_pd(felec,fvdw);
 572
 573             /* Calculate temporary vectorial force */
 574             tx               = _mm_mul_pd(fscal,dx00);
 575             ty               = _mm_mul_pd(fscal,dy00);
 576             tz               = _mm_mul_pd(fscal,dz00);
 577
 578             /* Update vectorial force */
 579             fix0             = _mm_add_pd(fix0,tx);
 580             fiy0             = _mm_add_pd(fiy0,ty);
 581             fiz0             = _mm_add_pd(fiz0,tz);
 582
 583             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 584
 585             /* Inner loop uses 54 flops */
 586         }
 587
 588         if(jidx<j_index_end)
 589         {
 590
 591             jnrA             = jjnr[jidx];
 592             j_coord_offsetA  = DIM*jnrA;
 593
 594             /* load j atom coordinates */
 595             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 596                                               &jx0,&jy0,&jz0);
 597
 598             /* Calculate displacement vector */
 599             dx00             = _mm_sub_pd(ix0,jx0);
 600             dy00             = _mm_sub_pd(iy0,jy0);
 601             dz00             = _mm_sub_pd(iz0,jz0);
 602
 603             /* Calculate squared distance and things based on it */
 604             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 605
 606             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 607
 608             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 609
 610             /* Load parameters for j particles */
 611             jq0              = _mm_load_sd(charge+jnrA+0);
 612             vdwjidx0A        = 2*vdwtype[jnrA+0];
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             r00              = _mm_mul_pd(rsq00,rinv00);
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq00             = _mm_mul_pd(iq0,jq0);
 622             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 623
 624             /* Calculate table index by multiplying r with table scale and truncate to integer */
 625             rt               = _mm_mul_pd(r00,vftabscale);
 626             vfitab           = _mm_cvttpd_epi32(rt);
 627             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 628             vfitab           = _mm_slli_epi32(vfitab,3);
 629
 630             /* COULOMB ELECTROSTATICS */
 631             velec            = _mm_mul_pd(qq00,rinv00);
 632             felec            = _mm_mul_pd(velec,rinvsq00);
 633
 634             /* CUBIC SPLINE TABLE DISPERSION */
 635             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 636             F                = _mm_setzero_pd();
 637             GMX_MM_TRANSPOSE2_PD(Y,F);
 638             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 639             H                = _mm_setzero_pd();
 640             GMX_MM_TRANSPOSE2_PD(G,H);
 641             Heps             = _mm_mul_pd(vfeps,H);
 642             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 643             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 644             fvdw6            = _mm_mul_pd(c6_00,FF);
 645
 646             /* CUBIC SPLINE TABLE REPULSION */
 647             vfitab           = _mm_add_epi32(vfitab,ifour);
 648             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 649             F                = _mm_setzero_pd();
 650             GMX_MM_TRANSPOSE2_PD(Y,F);
 651             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 652             H                = _mm_setzero_pd();
 653             GMX_MM_TRANSPOSE2_PD(G,H);
 654             Heps             = _mm_mul_pd(vfeps,H);
 655             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 656             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 657             fvdw12           = _mm_mul_pd(c12_00,FF);
 658             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 659
 660             fscal            = _mm_add_pd(felec,fvdw);
 661
 662             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 663
 664             /* Calculate temporary vectorial force */
 665             tx               = _mm_mul_pd(fscal,dx00);
 666             ty               = _mm_mul_pd(fscal,dy00);
 667             tz               = _mm_mul_pd(fscal,dz00);
 668
 669             /* Update vectorial force */
 670             fix0             = _mm_add_pd(fix0,tx);
 671             fiy0             = _mm_add_pd(fiy0,ty);
 672             fiz0             = _mm_add_pd(fiz0,tz);
 673
 674             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 675
 676             /* Inner loop uses 54 flops */
 677         }
 678
 679         /* End of innermost loop */
 680
 681         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 682                                               f+i_coord_offset,fshift+i_shift_offset);
 683
 684         /* Increment number of inner iterations */
 685         inneriter                  += j_index_end - j_index_start;
 686
 687         /* Outer loop uses 7 flops */
 688     }
 689
 690     /* Increment number of outer iterations */
 691     outeriter        += nri;
 692
 693     /* Update outer/inner flops */
 694
 695     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*54);
 696 }