src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_VF_sse2_double
  51  * Electrostatics interaction: Coulomb
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_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              vdwjidx0A,vdwjidx0B;
  82     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  91     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  92     __m128i          vfitab;
  93     __m128i          ifour       = _mm_set1_epi32(4);
  94     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  95     real             *vftab;
  96     __m128d          dummy_mask,cutoff_mask;
  97     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  98     __m128d          one     = _mm_set1_pd(1.0);
  99     __m128d          two     = _mm_set1_pd(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_pd(fr->ic->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     vftab            = kernel_data->table_vdw->data;
 118     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _mm_setzero_pd();
 146         fiy0             = _mm_setzero_pd();
 147         fiz0             = _mm_setzero_pd();
 148
 149         /* Load parameters for i particles */
 150         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 151         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 152
 153         /* Reset potential sums */
 154         velecsum         = _mm_setzero_pd();
 155         vvdwsum          = _mm_setzero_pd();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm_sub_pd(ix0,jx0);
 173             dy00             = _mm_sub_pd(iy0,jy0);
 174             dz00             = _mm_sub_pd(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 178
 179             rinv00           = sse2_invsqrt_d(rsq00);
 180
 181             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 182
 183             /* Load parameters for j particles */
 184             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 185             vdwjidx0A        = 2*vdwtype[jnrA+0];
 186             vdwjidx0B        = 2*vdwtype[jnrB+0];
 187
 188             /**************************
 189              * CALCULATE INTERACTIONS *
 190              **************************/
 191
 192             r00              = _mm_mul_pd(rsq00,rinv00);
 193
 194             /* Compute parameters for interactions between i and j atoms */
 195             qq00             = _mm_mul_pd(iq0,jq0);
 196             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 197                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 198
 199             /* Calculate table index by multiplying r with table scale and truncate to integer */
 200             rt               = _mm_mul_pd(r00,vftabscale);
 201             vfitab           = _mm_cvttpd_epi32(rt);
 202             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 203             vfitab           = _mm_slli_epi32(vfitab,3);
 204
 205             /* COULOMB ELECTROSTATICS */
 206             velec            = _mm_mul_pd(qq00,rinv00);
 207             felec            = _mm_mul_pd(velec,rinvsq00);
 208
 209             /* CUBIC SPLINE TABLE DISPERSION */
 210             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 211             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 212             GMX_MM_TRANSPOSE2_PD(Y,F);
 213             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 214             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 215             GMX_MM_TRANSPOSE2_PD(G,H);
 216             Heps             = _mm_mul_pd(vfeps,H);
 217             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 218             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 219             vvdw6            = _mm_mul_pd(c6_00,VV);
 220             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 221             fvdw6            = _mm_mul_pd(c6_00,FF);
 222
 223             /* CUBIC SPLINE TABLE REPULSION */
 224             vfitab           = _mm_add_epi32(vfitab,ifour);
 225             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 226             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 227             GMX_MM_TRANSPOSE2_PD(Y,F);
 228             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 229             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 230             GMX_MM_TRANSPOSE2_PD(G,H);
 231             Heps             = _mm_mul_pd(vfeps,H);
 232             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 233             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 234             vvdw12           = _mm_mul_pd(c12_00,VV);
 235             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 236             fvdw12           = _mm_mul_pd(c12_00,FF);
 237             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 238             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 239
 240             /* Update potential sum for this i atom from the interaction with this j atom. */
 241             velecsum         = _mm_add_pd(velecsum,velec);
 242             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 243
 244             fscal            = _mm_add_pd(felec,fvdw);
 245
 246             /* Calculate temporary vectorial force */
 247             tx               = _mm_mul_pd(fscal,dx00);
 248             ty               = _mm_mul_pd(fscal,dy00);
 249             tz               = _mm_mul_pd(fscal,dz00);
 250
 251             /* Update vectorial force */
 252             fix0             = _mm_add_pd(fix0,tx);
 253             fiy0             = _mm_add_pd(fiy0,ty);
 254             fiz0             = _mm_add_pd(fiz0,tz);
 255
 256             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 257
 258             /* Inner loop uses 63 flops */
 259         }
 260
 261         if(jidx<j_index_end)
 262         {
 263
 264             jnrA             = jjnr[jidx];
 265             j_coord_offsetA  = DIM*jnrA;
 266
 267             /* load j atom coordinates */
 268             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 269                                               &jx0,&jy0,&jz0);
 270
 271             /* Calculate displacement vector */
 272             dx00             = _mm_sub_pd(ix0,jx0);
 273             dy00             = _mm_sub_pd(iy0,jy0);
 274             dz00             = _mm_sub_pd(iz0,jz0);
 275
 276             /* Calculate squared distance and things based on it */
 277             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 278
 279             rinv00           = sse2_invsqrt_d(rsq00);
 280
 281             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 282
 283             /* Load parameters for j particles */
 284             jq0              = _mm_load_sd(charge+jnrA+0);
 285             vdwjidx0A        = 2*vdwtype[jnrA+0];
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             r00              = _mm_mul_pd(rsq00,rinv00);
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             qq00             = _mm_mul_pd(iq0,jq0);
 295             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 296
 297             /* Calculate table index by multiplying r with table scale and truncate to integer */
 298             rt               = _mm_mul_pd(r00,vftabscale);
 299             vfitab           = _mm_cvttpd_epi32(rt);
 300             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 301             vfitab           = _mm_slli_epi32(vfitab,3);
 302
 303             /* COULOMB ELECTROSTATICS */
 304             velec            = _mm_mul_pd(qq00,rinv00);
 305             felec            = _mm_mul_pd(velec,rinvsq00);
 306
 307             /* CUBIC SPLINE TABLE DISPERSION */
 308             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 309             F                = _mm_setzero_pd();
 310             GMX_MM_TRANSPOSE2_PD(Y,F);
 311             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 312             H                = _mm_setzero_pd();
 313             GMX_MM_TRANSPOSE2_PD(G,H);
 314             Heps             = _mm_mul_pd(vfeps,H);
 315             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 316             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 317             vvdw6            = _mm_mul_pd(c6_00,VV);
 318             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 319             fvdw6            = _mm_mul_pd(c6_00,FF);
 320
 321             /* CUBIC SPLINE TABLE REPULSION */
 322             vfitab           = _mm_add_epi32(vfitab,ifour);
 323             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 324             F                = _mm_setzero_pd();
 325             GMX_MM_TRANSPOSE2_PD(Y,F);
 326             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 327             H                = _mm_setzero_pd();
 328             GMX_MM_TRANSPOSE2_PD(G,H);
 329             Heps             = _mm_mul_pd(vfeps,H);
 330             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 331             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 332             vvdw12           = _mm_mul_pd(c12_00,VV);
 333             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 334             fvdw12           = _mm_mul_pd(c12_00,FF);
 335             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 336             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 340             velecsum         = _mm_add_pd(velecsum,velec);
 341             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 342             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 343
 344             fscal            = _mm_add_pd(felec,fvdw);
 345
 346             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_pd(fscal,dx00);
 350             ty               = _mm_mul_pd(fscal,dy00);
 351             tz               = _mm_mul_pd(fscal,dz00);
 352
 353             /* Update vectorial force */
 354             fix0             = _mm_add_pd(fix0,tx);
 355             fiy0             = _mm_add_pd(fiy0,ty);
 356             fiz0             = _mm_add_pd(fiz0,tz);
 357
 358             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 359
 360             /* Inner loop uses 63 flops */
 361         }
 362
 363         /* End of innermost loop */
 364
 365         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 366                                               f+i_coord_offset,fshift+i_shift_offset);
 367
 368         ggid                        = gid[iidx];
 369         /* Update potential energies */
 370         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 371         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 372
 373         /* Increment number of inner iterations */
 374         inneriter                  += j_index_end - j_index_start;
 375
 376         /* Outer loop uses 9 flops */
 377     }
 378
 379     /* Increment number of outer iterations */
 380     outeriter        += nri;
 381
 382     /* Update outer/inner flops */
 383
 384     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*63);
 385 }
 386 /*
 387  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse2_double
 388  * Electrostatics interaction: Coulomb
 389  * VdW interaction:            CubicSplineTable
 390  * Geometry:                   Particle-Particle
 391  * Calculate force/pot:        Force
 392  */
 393 void
 394 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_sse2_double
 395                     (t_nblist                    * gmx_restrict       nlist,
 396                      rvec                        * gmx_restrict          xx,
 397                      rvec                        * gmx_restrict          ff,
 398                      struct t_forcerec           * gmx_restrict          fr,
 399                      t_mdatoms                   * gmx_restrict     mdatoms,
 400                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 401                      t_nrnb                      * gmx_restrict        nrnb)
 402 {
 403     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 404      * just 0 for non-waters.
 405      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 406      * jnr indices corresponding to data put in the four positions in the SIMD register.
 407      */
 408     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 409     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 410     int              jnrA,jnrB;
 411     int              j_coord_offsetA,j_coord_offsetB;
 412     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 413     real             rcutoff_scalar;
 414     real             *shiftvec,*fshift,*x,*f;
 415     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 416     int              vdwioffset0;
 417     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 418     int              vdwjidx0A,vdwjidx0B;
 419     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 420     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 421     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 422     real             *charge;
 423     int              nvdwtype;
 424     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 425     int              *vdwtype;
 426     real             *vdwparam;
 427     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 428     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 429     __m128i          vfitab;
 430     __m128i          ifour       = _mm_set1_epi32(4);
 431     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 432     real             *vftab;
 433     __m128d          dummy_mask,cutoff_mask;
 434     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 435     __m128d          one     = _mm_set1_pd(1.0);
 436     __m128d          two     = _mm_set1_pd(2.0);
 437     x                = xx[0];
 438     f                = ff[0];
 439
 440     nri              = nlist->nri;
 441     iinr             = nlist->iinr;
 442     jindex           = nlist->jindex;
 443     jjnr             = nlist->jjnr;
 444     shiftidx         = nlist->shift;
 445     gid              = nlist->gid;
 446     shiftvec         = fr->shift_vec[0];
 447     fshift           = fr->fshift[0];
 448     facel            = _mm_set1_pd(fr->ic->epsfac);
 449     charge           = mdatoms->chargeA;
 450     nvdwtype         = fr->ntype;
 451     vdwparam         = fr->nbfp;
 452     vdwtype          = mdatoms->typeA;
 453
 454     vftab            = kernel_data->table_vdw->data;
 455     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 456
 457     /* Avoid stupid compiler warnings */
 458     jnrA = jnrB = 0;
 459     j_coord_offsetA = 0;
 460     j_coord_offsetB = 0;
 461
 462     outeriter        = 0;
 463     inneriter        = 0;
 464
 465     /* Start outer loop over neighborlists */
 466     for(iidx=0; iidx<nri; iidx++)
 467     {
 468         /* Load shift vector for this list */
 469         i_shift_offset   = DIM*shiftidx[iidx];
 470
 471         /* Load limits for loop over neighbors */
 472         j_index_start    = jindex[iidx];
 473         j_index_end      = jindex[iidx+1];
 474
 475         /* Get outer coordinate index */
 476         inr              = iinr[iidx];
 477         i_coord_offset   = DIM*inr;
 478
 479         /* Load i particle coords and add shift vector */
 480         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 481
 482         fix0             = _mm_setzero_pd();
 483         fiy0             = _mm_setzero_pd();
 484         fiz0             = _mm_setzero_pd();
 485
 486         /* Load parameters for i particles */
 487         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 488         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 489
 490         /* Start inner kernel loop */
 491         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 492         {
 493
 494             /* Get j neighbor index, and coordinate index */
 495             jnrA             = jjnr[jidx];
 496             jnrB             = jjnr[jidx+1];
 497             j_coord_offsetA  = DIM*jnrA;
 498             j_coord_offsetB  = DIM*jnrB;
 499
 500             /* load j atom coordinates */
 501             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 502                                               &jx0,&jy0,&jz0);
 503
 504             /* Calculate displacement vector */
 505             dx00             = _mm_sub_pd(ix0,jx0);
 506             dy00             = _mm_sub_pd(iy0,jy0);
 507             dz00             = _mm_sub_pd(iz0,jz0);
 508
 509             /* Calculate squared distance and things based on it */
 510             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 511
 512             rinv00           = sse2_invsqrt_d(rsq00);
 513
 514             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 515
 516             /* Load parameters for j particles */
 517             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 518             vdwjidx0A        = 2*vdwtype[jnrA+0];
 519             vdwjidx0B        = 2*vdwtype[jnrB+0];
 520
 521             /**************************
 522              * CALCULATE INTERACTIONS *
 523              **************************/
 524
 525             r00              = _mm_mul_pd(rsq00,rinv00);
 526
 527             /* Compute parameters for interactions between i and j atoms */
 528             qq00             = _mm_mul_pd(iq0,jq0);
 529             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 530                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 531
 532             /* Calculate table index by multiplying r with table scale and truncate to integer */
 533             rt               = _mm_mul_pd(r00,vftabscale);
 534             vfitab           = _mm_cvttpd_epi32(rt);
 535             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 536             vfitab           = _mm_slli_epi32(vfitab,3);
 537
 538             /* COULOMB ELECTROSTATICS */
 539             velec            = _mm_mul_pd(qq00,rinv00);
 540             felec            = _mm_mul_pd(velec,rinvsq00);
 541
 542             /* CUBIC SPLINE TABLE DISPERSION */
 543             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 544             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 545             GMX_MM_TRANSPOSE2_PD(Y,F);
 546             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 547             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 548             GMX_MM_TRANSPOSE2_PD(G,H);
 549             Heps             = _mm_mul_pd(vfeps,H);
 550             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 551             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 552             fvdw6            = _mm_mul_pd(c6_00,FF);
 553
 554             /* CUBIC SPLINE TABLE REPULSION */
 555             vfitab           = _mm_add_epi32(vfitab,ifour);
 556             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 557             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 558             GMX_MM_TRANSPOSE2_PD(Y,F);
 559             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 560             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 561             GMX_MM_TRANSPOSE2_PD(G,H);
 562             Heps             = _mm_mul_pd(vfeps,H);
 563             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 564             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 565             fvdw12           = _mm_mul_pd(c12_00,FF);
 566             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 567
 568             fscal            = _mm_add_pd(felec,fvdw);
 569
 570             /* Calculate temporary vectorial force */
 571             tx               = _mm_mul_pd(fscal,dx00);
 572             ty               = _mm_mul_pd(fscal,dy00);
 573             tz               = _mm_mul_pd(fscal,dz00);
 574
 575             /* Update vectorial force */
 576             fix0             = _mm_add_pd(fix0,tx);
 577             fiy0             = _mm_add_pd(fiy0,ty);
 578             fiz0             = _mm_add_pd(fiz0,tz);
 579
 580             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 581
 582             /* Inner loop uses 54 flops */
 583         }
 584
 585         if(jidx<j_index_end)
 586         {
 587
 588             jnrA             = jjnr[jidx];
 589             j_coord_offsetA  = DIM*jnrA;
 590
 591             /* load j atom coordinates */
 592             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 593                                               &jx0,&jy0,&jz0);
 594
 595             /* Calculate displacement vector */
 596             dx00             = _mm_sub_pd(ix0,jx0);
 597             dy00             = _mm_sub_pd(iy0,jy0);
 598             dz00             = _mm_sub_pd(iz0,jz0);
 599
 600             /* Calculate squared distance and things based on it */
 601             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 602
 603             rinv00           = sse2_invsqrt_d(rsq00);
 604
 605             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 606
 607             /* Load parameters for j particles */
 608             jq0              = _mm_load_sd(charge+jnrA+0);
 609             vdwjidx0A        = 2*vdwtype[jnrA+0];
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             r00              = _mm_mul_pd(rsq00,rinv00);
 616
 617             /* Compute parameters for interactions between i and j atoms */
 618             qq00             = _mm_mul_pd(iq0,jq0);
 619             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 620
 621             /* Calculate table index by multiplying r with table scale and truncate to integer */
 622             rt               = _mm_mul_pd(r00,vftabscale);
 623             vfitab           = _mm_cvttpd_epi32(rt);
 624             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 625             vfitab           = _mm_slli_epi32(vfitab,3);
 626
 627             /* COULOMB ELECTROSTATICS */
 628             velec            = _mm_mul_pd(qq00,rinv00);
 629             felec            = _mm_mul_pd(velec,rinvsq00);
 630
 631             /* CUBIC SPLINE TABLE DISPERSION */
 632             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 633             F                = _mm_setzero_pd();
 634             GMX_MM_TRANSPOSE2_PD(Y,F);
 635             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 636             H                = _mm_setzero_pd();
 637             GMX_MM_TRANSPOSE2_PD(G,H);
 638             Heps             = _mm_mul_pd(vfeps,H);
 639             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 640             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 641             fvdw6            = _mm_mul_pd(c6_00,FF);
 642
 643             /* CUBIC SPLINE TABLE REPULSION */
 644             vfitab           = _mm_add_epi32(vfitab,ifour);
 645             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 646             F                = _mm_setzero_pd();
 647             GMX_MM_TRANSPOSE2_PD(Y,F);
 648             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 649             H                = _mm_setzero_pd();
 650             GMX_MM_TRANSPOSE2_PD(G,H);
 651             Heps             = _mm_mul_pd(vfeps,H);
 652             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 653             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 654             fvdw12           = _mm_mul_pd(c12_00,FF);
 655             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 656
 657             fscal            = _mm_add_pd(felec,fvdw);
 658
 659             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm_mul_pd(fscal,dx00);
 663             ty               = _mm_mul_pd(fscal,dy00);
 664             tz               = _mm_mul_pd(fscal,dz00);
 665
 666             /* Update vectorial force */
 667             fix0             = _mm_add_pd(fix0,tx);
 668             fiy0             = _mm_add_pd(fiy0,ty);
 669             fiz0             = _mm_add_pd(fiz0,tz);
 670
 671             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 672
 673             /* Inner loop uses 54 flops */
 674         }
 675
 676         /* End of innermost loop */
 677
 678         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 679                                               f+i_coord_offset,fshift+i_shift_offset);
 680
 681         /* Increment number of inner iterations */
 682         inneriter                  += j_index_end - j_index_start;
 683
 684         /* Outer loop uses 7 flops */
 685     }
 686
 687     /* Increment number of outer iterations */
 688     outeriter        += nri;
 689
 690     /* Update outer/inner flops */
 691
 692     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*54);
 693 }