src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecRF_VdwCSTab_GeomP1P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 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/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse4_1_double
  51  * Electrostatics interaction: ReactionField
  52  * VdW interaction:            CubicSplineTable
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_VF_sse4_1_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     krf              = _mm_set1_pd(fr->ic->k_rf);
 114     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 115     crf              = _mm_set1_pd(fr->ic->c_rf);
 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           = sse41_invsqrt_d(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             /* REACTION-FIELD ELECTROSTATICS */
 209             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 210             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 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 67 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           = sse41_invsqrt_d(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             /* REACTION-FIELD ELECTROSTATICS */
 307             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_add_pd(rinv00,_mm_mul_pd(krf,rsq00)),crf));
 308             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 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 67 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*67);
 388 }
 389 /*
 390  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse4_1_double
 391  * Electrostatics interaction: ReactionField
 392  * VdW interaction:            CubicSplineTable
 393  * Geometry:                   Particle-Particle
 394  * Calculate force/pot:        Force
 395  */
 396 void
 397 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_sse4_1_double
 398                     (t_nblist                    * gmx_restrict       nlist,
 399                      rvec                        * gmx_restrict          xx,
 400                      rvec                        * gmx_restrict          ff,
 401                      struct 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->ic->epsfac);
 452     charge           = mdatoms->chargeA;
 453     krf              = _mm_set1_pd(fr->ic->k_rf);
 454     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 455     crf              = _mm_set1_pd(fr->ic->c_rf);
 456     nvdwtype         = fr->ntype;
 457     vdwparam         = fr->nbfp;
 458     vdwtype          = mdatoms->typeA;
 459
 460     vftab            = kernel_data->table_vdw->data;
 461     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 462
 463     /* Avoid stupid compiler warnings */
 464     jnrA = jnrB = 0;
 465     j_coord_offsetA = 0;
 466     j_coord_offsetB = 0;
 467
 468     outeriter        = 0;
 469     inneriter        = 0;
 470
 471     /* Start outer loop over neighborlists */
 472     for(iidx=0; iidx<nri; iidx++)
 473     {
 474         /* Load shift vector for this list */
 475         i_shift_offset   = DIM*shiftidx[iidx];
 476
 477         /* Load limits for loop over neighbors */
 478         j_index_start    = jindex[iidx];
 479         j_index_end      = jindex[iidx+1];
 480
 481         /* Get outer coordinate index */
 482         inr              = iinr[iidx];
 483         i_coord_offset   = DIM*inr;
 484
 485         /* Load i particle coords and add shift vector */
 486         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 487
 488         fix0             = _mm_setzero_pd();
 489         fiy0             = _mm_setzero_pd();
 490         fiz0             = _mm_setzero_pd();
 491
 492         /* Load parameters for i particles */
 493         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 494         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 495
 496         /* Start inner kernel loop */
 497         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 498         {
 499
 500             /* Get j neighbor index, and coordinate index */
 501             jnrA             = jjnr[jidx];
 502             jnrB             = jjnr[jidx+1];
 503             j_coord_offsetA  = DIM*jnrA;
 504             j_coord_offsetB  = DIM*jnrB;
 505
 506             /* load j atom coordinates */
 507             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 508                                               &jx0,&jy0,&jz0);
 509
 510             /* Calculate displacement vector */
 511             dx00             = _mm_sub_pd(ix0,jx0);
 512             dy00             = _mm_sub_pd(iy0,jy0);
 513             dz00             = _mm_sub_pd(iz0,jz0);
 514
 515             /* Calculate squared distance and things based on it */
 516             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 517
 518             rinv00           = sse41_invsqrt_d(rsq00);
 519
 520             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 521
 522             /* Load parameters for j particles */
 523             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 524             vdwjidx0A        = 2*vdwtype[jnrA+0];
 525             vdwjidx0B        = 2*vdwtype[jnrB+0];
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             r00              = _mm_mul_pd(rsq00,rinv00);
 532
 533             /* Compute parameters for interactions between i and j atoms */
 534             qq00             = _mm_mul_pd(iq0,jq0);
 535             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 536                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 537
 538             /* Calculate table index by multiplying r with table scale and truncate to integer */
 539             rt               = _mm_mul_pd(r00,vftabscale);
 540             vfitab           = _mm_cvttpd_epi32(rt);
 541             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 542             vfitab           = _mm_slli_epi32(vfitab,3);
 543
 544             /* REACTION-FIELD ELECTROSTATICS */
 545             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 546
 547             /* CUBIC SPLINE TABLE DISPERSION */
 548             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 549             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 550             GMX_MM_TRANSPOSE2_PD(Y,F);
 551             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 552             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 553             GMX_MM_TRANSPOSE2_PD(G,H);
 554             Heps             = _mm_mul_pd(vfeps,H);
 555             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 556             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 557             fvdw6            = _mm_mul_pd(c6_00,FF);
 558
 559             /* CUBIC SPLINE TABLE REPULSION */
 560             vfitab           = _mm_add_epi32(vfitab,ifour);
 561             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 562             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 563             GMX_MM_TRANSPOSE2_PD(Y,F);
 564             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 565             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 566             GMX_MM_TRANSPOSE2_PD(G,H);
 567             Heps             = _mm_mul_pd(vfeps,H);
 568             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 569             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 570             fvdw12           = _mm_mul_pd(c12_00,FF);
 571             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 572
 573             fscal            = _mm_add_pd(felec,fvdw);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm_mul_pd(fscal,dx00);
 577             ty               = _mm_mul_pd(fscal,dy00);
 578             tz               = _mm_mul_pd(fscal,dz00);
 579
 580             /* Update vectorial force */
 581             fix0             = _mm_add_pd(fix0,tx);
 582             fiy0             = _mm_add_pd(fiy0,ty);
 583             fiz0             = _mm_add_pd(fiz0,tz);
 584
 585             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 586
 587             /* Inner loop uses 54 flops */
 588         }
 589
 590         if(jidx<j_index_end)
 591         {
 592
 593             jnrA             = jjnr[jidx];
 594             j_coord_offsetA  = DIM*jnrA;
 595
 596             /* load j atom coordinates */
 597             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 598                                               &jx0,&jy0,&jz0);
 599
 600             /* Calculate displacement vector */
 601             dx00             = _mm_sub_pd(ix0,jx0);
 602             dy00             = _mm_sub_pd(iy0,jy0);
 603             dz00             = _mm_sub_pd(iz0,jz0);
 604
 605             /* Calculate squared distance and things based on it */
 606             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 607
 608             rinv00           = sse41_invsqrt_d(rsq00);
 609
 610             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 611
 612             /* Load parameters for j particles */
 613             jq0              = _mm_load_sd(charge+jnrA+0);
 614             vdwjidx0A        = 2*vdwtype[jnrA+0];
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             r00              = _mm_mul_pd(rsq00,rinv00);
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq00             = _mm_mul_pd(iq0,jq0);
 624             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 625
 626             /* Calculate table index by multiplying r with table scale and truncate to integer */
 627             rt               = _mm_mul_pd(r00,vftabscale);
 628             vfitab           = _mm_cvttpd_epi32(rt);
 629             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 630             vfitab           = _mm_slli_epi32(vfitab,3);
 631
 632             /* REACTION-FIELD ELECTROSTATICS */
 633             felec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_mul_pd(rinv00,rinvsq00),krf2));
 634
 635             /* CUBIC SPLINE TABLE DISPERSION */
 636             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 637             F                = _mm_setzero_pd();
 638             GMX_MM_TRANSPOSE2_PD(Y,F);
 639             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 640             H                = _mm_setzero_pd();
 641             GMX_MM_TRANSPOSE2_PD(G,H);
 642             Heps             = _mm_mul_pd(vfeps,H);
 643             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 644             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 645             fvdw6            = _mm_mul_pd(c6_00,FF);
 646
 647             /* CUBIC SPLINE TABLE REPULSION */
 648             vfitab           = _mm_add_epi32(vfitab,ifour);
 649             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 650             F                = _mm_setzero_pd();
 651             GMX_MM_TRANSPOSE2_PD(Y,F);
 652             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 653             H                = _mm_setzero_pd();
 654             GMX_MM_TRANSPOSE2_PD(G,H);
 655             Heps             = _mm_mul_pd(vfeps,H);
 656             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 657             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 658             fvdw12           = _mm_mul_pd(c12_00,FF);
 659             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 660
 661             fscal            = _mm_add_pd(felec,fvdw);
 662
 663             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 664
 665             /* Calculate temporary vectorial force */
 666             tx               = _mm_mul_pd(fscal,dx00);
 667             ty               = _mm_mul_pd(fscal,dy00);
 668             tz               = _mm_mul_pd(fscal,dz00);
 669
 670             /* Update vectorial force */
 671             fix0             = _mm_add_pd(fix0,tx);
 672             fiy0             = _mm_add_pd(fiy0,ty);
 673             fiz0             = _mm_add_pd(fiz0,tz);
 674
 675             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 676
 677             /* Inner loop uses 54 flops */
 678         }
 679
 680         /* End of innermost loop */
 681
 682         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 683                                               f+i_coord_offset,fshift+i_shift_offset);
 684
 685         /* Increment number of inner iterations */
 686         inneriter                  += j_index_end - j_index_start;
 687
 688         /* Outer loop uses 7 flops */
 689     }
 690
 691     /* Increment number of outer iterations */
 692     outeriter        += nri;
 693
 694     /* Update outer/inner flops */
 695
 696     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*54);
 697 }