src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecNone_VdwLJSw_GeomP1P1_sparc64_hpc_ace_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_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_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: None
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_VF_sparc64_hpc_ace_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 double precision SIMD, 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     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     int              nvdwtype;
  85     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  86     int              *vdwtype;
  87     real             *vdwparam;
  88     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  89     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  90     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  91     real             rswitch_scalar,d_scalar;
  92     _fjsp_v2r8       itab_tmp;
  93     _fjsp_v2r8       dummy_mask,cutoff_mask;
  94     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  95     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  96     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  97
  98     x                = xx[0];
  99     f                = ff[0];
 100
 101     nri              = nlist->nri;
 102     iinr             = nlist->iinr;
 103     jindex           = nlist->jindex;
 104     jjnr             = nlist->jjnr;
 105     shiftidx         = nlist->shift;
 106     gid              = nlist->gid;
 107     shiftvec         = fr->shift_vec[0];
 108     fshift           = fr->fshift[0];
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     rcutoff_scalar   = fr->ic->rvdw;
 114     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 115     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 116
 117     rswitch_scalar   = fr->ic->rvdw_switch;
 118     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 119     /* Setup switch parameters */
 120     d_scalar         = rcutoff_scalar-rswitch_scalar;
 121     d                = gmx_fjsp_set1_v2r8(d_scalar);
 122     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 123     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 124     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 125     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 126     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 127     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 153
 154         fix0             = _fjsp_setzero_v2r8();
 155         fiy0             = _fjsp_setzero_v2r8();
 156         fiz0             = _fjsp_setzero_v2r8();
 157
 158         /* Load parameters for i particles */
 159         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 160
 161         /* Reset potential sums */
 162         vvdwsum          = _fjsp_setzero_v2r8();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173
 174             /* load j atom coordinates */
 175             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 180             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 181             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 185
 186             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 187
 188             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 189
 190             /* Load parameters for j particles */
 191             vdwjidx0A        = 2*vdwtype[jnrA+0];
 192             vdwjidx0B        = 2*vdwtype[jnrB+0];
 193
 194             /**************************
 195              * CALCULATE INTERACTIONS *
 196              **************************/
 197
 198             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 199             {
 200
 201             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 202
 203             /* Compute parameters for interactions between i and j atoms */
 204             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 205                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 206
 207             /* LENNARD-JONES DISPERSION/REPULSION */
 208
 209             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 210             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 211             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 212             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 213             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 214
 215             d                = _fjsp_sub_v2r8(r00,rswitch);
 216             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 217             d2               = _fjsp_mul_v2r8(d,d);
 218             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 219
 220             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 221
 222             /* Evaluate switch function */
 223             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 224             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 225             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 226             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 227
 228             /* Update potential sum for this i atom from the interaction with this j atom. */
 229             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 230             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 231
 232             fscal            = fvdw;
 233
 234             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 235
 236             /* Update vectorial force */
 237             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 238             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 239             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 240
 241             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 242
 243             }
 244
 245             /* Inner loop uses 62 flops */
 246         }
 247
 248         if(jidx<j_index_end)
 249         {
 250
 251             jnrA             = jjnr[jidx];
 252             j_coord_offsetA  = DIM*jnrA;
 253
 254             /* load j atom coordinates */
 255             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 256                                               &jx0,&jy0,&jz0);
 257
 258             /* Calculate displacement vector */
 259             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 260             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 261             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 262
 263             /* Calculate squared distance and things based on it */
 264             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 265
 266             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 267
 268             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 269
 270             /* Load parameters for j particles */
 271             vdwjidx0A        = 2*vdwtype[jnrA+0];
 272
 273             /**************************
 274              * CALCULATE INTERACTIONS *
 275              **************************/
 276
 277             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 278             {
 279
 280             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 284                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 285
 286             /* LENNARD-JONES DISPERSION/REPULSION */
 287
 288             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 289             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 290             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 291             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 292             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 293
 294             d                = _fjsp_sub_v2r8(r00,rswitch);
 295             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 296             d2               = _fjsp_mul_v2r8(d,d);
 297             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 298
 299             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 300
 301             /* Evaluate switch function */
 302             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 303             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 304             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 305             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 309             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 310             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 311
 312             fscal            = fvdw;
 313
 314             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 315
 316             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 317
 318             /* Update vectorial force */
 319             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 320             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 321             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 322
 323             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 324
 325             }
 326
 327             /* Inner loop uses 62 flops */
 328         }
 329
 330         /* End of innermost loop */
 331
 332         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 333                                               f+i_coord_offset,fshift+i_shift_offset);
 334
 335         ggid                        = gid[iidx];
 336         /* Update potential energies */
 337         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 338
 339         /* Increment number of inner iterations */
 340         inneriter                  += j_index_end - j_index_start;
 341
 342         /* Outer loop uses 7 flops */
 343     }
 344
 345     /* Increment number of outer iterations */
 346     outeriter        += nri;
 347
 348     /* Update outer/inner flops */
 349
 350     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*62);
 351 }
 352 /*
 353  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 354  * Electrostatics interaction: None
 355  * VdW interaction:            LennardJones
 356  * Geometry:                   Particle-Particle
 357  * Calculate force/pot:        Force
 358  */
 359 void
 360 nb_kernel_ElecNone_VdwLJSw_GeomP1P1_F_sparc64_hpc_ace_double
 361                     (t_nblist                    * gmx_restrict       nlist,
 362                      rvec                        * gmx_restrict          xx,
 363                      rvec                        * gmx_restrict          ff,
 364                      struct t_forcerec           * gmx_restrict          fr,
 365                      t_mdatoms                   * gmx_restrict     mdatoms,
 366                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 367                      t_nrnb                      * gmx_restrict        nrnb)
 368 {
 369     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 370      * just 0 for non-waters.
 371      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 372      * jnr indices corresponding to data put in the four positions in the SIMD register.
 373      */
 374     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 375     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 376     int              jnrA,jnrB;
 377     int              j_coord_offsetA,j_coord_offsetB;
 378     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 379     real             rcutoff_scalar;
 380     real             *shiftvec,*fshift,*x,*f;
 381     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 382     int              vdwioffset0;
 383     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 384     int              vdwjidx0A,vdwjidx0B;
 385     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 386     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 387     int              nvdwtype;
 388     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 389     int              *vdwtype;
 390     real             *vdwparam;
 391     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 392     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 393     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 394     real             rswitch_scalar,d_scalar;
 395     _fjsp_v2r8       itab_tmp;
 396     _fjsp_v2r8       dummy_mask,cutoff_mask;
 397     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 398     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 399     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 400
 401     x                = xx[0];
 402     f                = ff[0];
 403
 404     nri              = nlist->nri;
 405     iinr             = nlist->iinr;
 406     jindex           = nlist->jindex;
 407     jjnr             = nlist->jjnr;
 408     shiftidx         = nlist->shift;
 409     gid              = nlist->gid;
 410     shiftvec         = fr->shift_vec[0];
 411     fshift           = fr->fshift[0];
 412     nvdwtype         = fr->ntype;
 413     vdwparam         = fr->nbfp;
 414     vdwtype          = mdatoms->typeA;
 415
 416     rcutoff_scalar   = fr->ic->rvdw;
 417     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 418     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 419
 420     rswitch_scalar   = fr->ic->rvdw_switch;
 421     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 422     /* Setup switch parameters */
 423     d_scalar         = rcutoff_scalar-rswitch_scalar;
 424     d                = gmx_fjsp_set1_v2r8(d_scalar);
 425     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 426     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 427     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 428     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 429     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 430     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 431
 432     /* Avoid stupid compiler warnings */
 433     jnrA = jnrB = 0;
 434     j_coord_offsetA = 0;
 435     j_coord_offsetB = 0;
 436
 437     outeriter        = 0;
 438     inneriter        = 0;
 439
 440     /* Start outer loop over neighborlists */
 441     for(iidx=0; iidx<nri; iidx++)
 442     {
 443         /* Load shift vector for this list */
 444         i_shift_offset   = DIM*shiftidx[iidx];
 445
 446         /* Load limits for loop over neighbors */
 447         j_index_start    = jindex[iidx];
 448         j_index_end      = jindex[iidx+1];
 449
 450         /* Get outer coordinate index */
 451         inr              = iinr[iidx];
 452         i_coord_offset   = DIM*inr;
 453
 454         /* Load i particle coords and add shift vector */
 455         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 456
 457         fix0             = _fjsp_setzero_v2r8();
 458         fiy0             = _fjsp_setzero_v2r8();
 459         fiz0             = _fjsp_setzero_v2r8();
 460
 461         /* Load parameters for i particles */
 462         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 463
 464         /* Start inner kernel loop */
 465         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 466         {
 467
 468             /* Get j neighbor index, and coordinate index */
 469             jnrA             = jjnr[jidx];
 470             jnrB             = jjnr[jidx+1];
 471             j_coord_offsetA  = DIM*jnrA;
 472             j_coord_offsetB  = DIM*jnrB;
 473
 474             /* load j atom coordinates */
 475             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 476                                               &jx0,&jy0,&jz0);
 477
 478             /* Calculate displacement vector */
 479             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 480             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 481             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 482
 483             /* Calculate squared distance and things based on it */
 484             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 485
 486             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 487
 488             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 489
 490             /* Load parameters for j particles */
 491             vdwjidx0A        = 2*vdwtype[jnrA+0];
 492             vdwjidx0B        = 2*vdwtype[jnrB+0];
 493
 494             /**************************
 495              * CALCULATE INTERACTIONS *
 496              **************************/
 497
 498             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 499             {
 500
 501             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 505                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 506
 507             /* LENNARD-JONES DISPERSION/REPULSION */
 508
 509             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 510             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 511             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 512             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 513             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 514
 515             d                = _fjsp_sub_v2r8(r00,rswitch);
 516             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 517             d2               = _fjsp_mul_v2r8(d,d);
 518             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 519
 520             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 521
 522             /* Evaluate switch function */
 523             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 524             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 525             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 526
 527             fscal            = fvdw;
 528
 529             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 530
 531             /* Update vectorial force */
 532             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 533             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 534             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 535
 536             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 537
 538             }
 539
 540             /* Inner loop uses 59 flops */
 541         }
 542
 543         if(jidx<j_index_end)
 544         {
 545
 546             jnrA             = jjnr[jidx];
 547             j_coord_offsetA  = DIM*jnrA;
 548
 549             /* load j atom coordinates */
 550             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 551                                               &jx0,&jy0,&jz0);
 552
 553             /* Calculate displacement vector */
 554             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 555             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 556             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 557
 558             /* Calculate squared distance and things based on it */
 559             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 560
 561             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 562
 563             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 564
 565             /* Load parameters for j particles */
 566             vdwjidx0A        = 2*vdwtype[jnrA+0];
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 573             {
 574
 575             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 576
 577             /* Compute parameters for interactions between i and j atoms */
 578             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 579                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 580
 581             /* LENNARD-JONES DISPERSION/REPULSION */
 582
 583             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 584             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 585             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 586             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 587             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 588
 589             d                = _fjsp_sub_v2r8(r00,rswitch);
 590             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 591             d2               = _fjsp_mul_v2r8(d,d);
 592             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 593
 594             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 595
 596             /* Evaluate switch function */
 597             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 598             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv00,_fjsp_mul_v2r8(vvdw,dsw)) );
 599             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 600
 601             fscal            = fvdw;
 602
 603             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 604
 605             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 606
 607             /* Update vectorial force */
 608             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 609             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 610             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 611
 612             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 613
 614             }
 615
 616             /* Inner loop uses 59 flops */
 617         }
 618
 619         /* End of innermost loop */
 620
 621         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 622                                               f+i_coord_offset,fshift+i_shift_offset);
 623
 624         /* Increment number of inner iterations */
 625         inneriter                  += j_index_end - j_index_start;
 626
 627         /* Outer loop uses 6 flops */
 628     }
 629
 630     /* Increment number of outer iterations */
 631     outeriter        += nri;
 632
 633     /* Update outer/inner flops */
 634
 635     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*59);
 636 }