src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecNone_VdwLJEwSh_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_VdwLJEwSh_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: None
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecNone_VdwLJEwSh_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           c6grid_00;
  91     real                 *vdwgridparam;
  92     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  93     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
  94     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
  95     _fjsp_v2r8       itab_tmp;
  96     _fjsp_v2r8       dummy_mask,cutoff_mask;
  97     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  98     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  99     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 100
 101     x                = xx[0];
 102     f                = ff[0];
 103
 104     nri              = nlist->nri;
 105     iinr             = nlist->iinr;
 106     jindex           = nlist->jindex;
 107     jjnr             = nlist->jjnr;
 108     shiftidx         = nlist->shift;
 109     gid              = nlist->gid;
 110     shiftvec         = fr->shift_vec[0];
 111     fshift           = fr->fshift[0];
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115     vdwgridparam     = fr->ljpme_c6grid;
 116     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 117     ewclj            = gmx_fjsp_set1_v2r8(fr->ic->ewaldcoeff_lj);
 118     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 119
 120     rcutoff_scalar   = fr->ic->rvdw;
 121     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 122     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 123
 124     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 125     rvdw             = gmx_fjsp_set1_v2r8(fr->ic->rvdw);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _fjsp_setzero_v2r8();
 153         fiy0             = _fjsp_setzero_v2r8();
 154         fiz0             = _fjsp_setzero_v2r8();
 155
 156         /* Load parameters for i particles */
 157         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 158
 159         /* Reset potential sums */
 160         vvdwsum          = _fjsp_setzero_v2r8();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             j_coord_offsetA  = DIM*jnrA;
 170             j_coord_offsetB  = DIM*jnrB;
 171
 172             /* load j atom coordinates */
 173             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 174                                               &jx0,&jy0,&jz0);
 175
 176             /* Calculate displacement vector */
 177             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 178             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 179             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 180
 181             /* Calculate squared distance and things based on it */
 182             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 183
 184             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 185
 186             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 187
 188             /* Load parameters for j particles */
 189             vdwjidx0A        = 2*vdwtype[jnrA+0];
 190             vdwjidx0B        = 2*vdwtype[jnrB+0];
 191
 192             /**************************
 193              * CALCULATE INTERACTIONS *
 194              **************************/
 195
 196             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 197             {
 198
 199             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 200
 201             /* Compute parameters for interactions between i and j atoms */
 202             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 203                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 204
 205             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 206                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 207
 208             /* Analytical LJ-PME */
 209             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 210             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 211             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 212             exponent         = gmx_simd_exp_d(ewcljrsq);
 213             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 214             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 215             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 216             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 217             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 218             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 219                                _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw6,_fjsp_madd_v2r8(c6grid_00,sh_lj_ewald,_fjsp_mul_v2r8(c6_00,sh_vdw_invrcut6))),one_sixth));
 220             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 221             fvdw             = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq00);
 222
 223             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 224
 225             /* Update potential sum for this i atom from the interaction with this j atom. */
 226             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 227             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 228
 229             fscal            = fvdw;
 230
 231             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 232
 233             /* Update vectorial force */
 234             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 235             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 236             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 237
 238             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 239
 240             }
 241
 242             /* Inner loop uses 59 flops */
 243         }
 244
 245         if(jidx<j_index_end)
 246         {
 247
 248             jnrA             = jjnr[jidx];
 249             j_coord_offsetA  = DIM*jnrA;
 250
 251             /* load j atom coordinates */
 252             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 253                                               &jx0,&jy0,&jz0);
 254
 255             /* Calculate displacement vector */
 256             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 257             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 258             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 259
 260             /* Calculate squared distance and things based on it */
 261             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 262
 263             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 264
 265             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 266
 267             /* Load parameters for j particles */
 268             vdwjidx0A        = 2*vdwtype[jnrA+0];
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 275             {
 276
 277             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 278
 279             /* Compute parameters for interactions between i and j atoms */
 280             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 281                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 282
 283             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 284                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 285
 286             /* Analytical LJ-PME */
 287             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 288             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 289             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 290             exponent         = gmx_simd_exp_d(ewcljrsq);
 291             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 292             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 293             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 294             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 295             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 296             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 297                                _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw6,_fjsp_madd_v2r8(c6grid_00,sh_lj_ewald,_fjsp_mul_v2r8(c6_00,sh_vdw_invrcut6))),one_sixth));
 298             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 299             fvdw             = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq00);
 300
 301             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 305             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 306             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 307
 308             fscal            = fvdw;
 309
 310             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 311
 312             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 313
 314             /* Update vectorial force */
 315             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 316             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 317             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 318
 319             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 320
 321             }
 322
 323             /* Inner loop uses 59 flops */
 324         }
 325
 326         /* End of innermost loop */
 327
 328         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 329                                               f+i_coord_offset,fshift+i_shift_offset);
 330
 331         ggid                        = gid[iidx];
 332         /* Update potential energies */
 333         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 334
 335         /* Increment number of inner iterations */
 336         inneriter                  += j_index_end - j_index_start;
 337
 338         /* Outer loop uses 7 flops */
 339     }
 340
 341     /* Increment number of outer iterations */
 342     outeriter        += nri;
 343
 344     /* Update outer/inner flops */
 345
 346     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 347 }
 348 /*
 349  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sparc64_hpc_ace_double
 350  * Electrostatics interaction: None
 351  * VdW interaction:            LJEwald
 352  * Geometry:                   Particle-Particle
 353  * Calculate force/pot:        Force
 354  */
 355 void
 356 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sparc64_hpc_ace_double
 357                     (t_nblist                    * gmx_restrict       nlist,
 358                      rvec                        * gmx_restrict          xx,
 359                      rvec                        * gmx_restrict          ff,
 360                      struct t_forcerec           * gmx_restrict          fr,
 361                      t_mdatoms                   * gmx_restrict     mdatoms,
 362                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 363                      t_nrnb                      * gmx_restrict        nrnb)
 364 {
 365     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 366      * just 0 for non-waters.
 367      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 368      * jnr indices corresponding to data put in the four positions in the SIMD register.
 369      */
 370     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 371     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 372     int              jnrA,jnrB;
 373     int              j_coord_offsetA,j_coord_offsetB;
 374     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 375     real             rcutoff_scalar;
 376     real             *shiftvec,*fshift,*x,*f;
 377     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 378     int              vdwioffset0;
 379     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 380     int              vdwjidx0A,vdwjidx0B;
 381     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 382     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 383     int              nvdwtype;
 384     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 385     int              *vdwtype;
 386     real             *vdwparam;
 387     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 388     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 389     _fjsp_v2r8           c6grid_00;
 390     real                 *vdwgridparam;
 391     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 392     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 393     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 394     _fjsp_v2r8       itab_tmp;
 395     _fjsp_v2r8       dummy_mask,cutoff_mask;
 396     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 397     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 398     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 399
 400     x                = xx[0];
 401     f                = ff[0];
 402
 403     nri              = nlist->nri;
 404     iinr             = nlist->iinr;
 405     jindex           = nlist->jindex;
 406     jjnr             = nlist->jjnr;
 407     shiftidx         = nlist->shift;
 408     gid              = nlist->gid;
 409     shiftvec         = fr->shift_vec[0];
 410     fshift           = fr->fshift[0];
 411     nvdwtype         = fr->ntype;
 412     vdwparam         = fr->nbfp;
 413     vdwtype          = mdatoms->typeA;
 414     vdwgridparam     = fr->ljpme_c6grid;
 415     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 416     ewclj            = gmx_fjsp_set1_v2r8(fr->ic->ewaldcoeff_lj);
 417     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 418
 419     rcutoff_scalar   = fr->ic->rvdw;
 420     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 421     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 422
 423     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 424     rvdw             = gmx_fjsp_set1_v2r8(fr->ic->rvdw);
 425
 426     /* Avoid stupid compiler warnings */
 427     jnrA = jnrB = 0;
 428     j_coord_offsetA = 0;
 429     j_coord_offsetB = 0;
 430
 431     outeriter        = 0;
 432     inneriter        = 0;
 433
 434     /* Start outer loop over neighborlists */
 435     for(iidx=0; iidx<nri; iidx++)
 436     {
 437         /* Load shift vector for this list */
 438         i_shift_offset   = DIM*shiftidx[iidx];
 439
 440         /* Load limits for loop over neighbors */
 441         j_index_start    = jindex[iidx];
 442         j_index_end      = jindex[iidx+1];
 443
 444         /* Get outer coordinate index */
 445         inr              = iinr[iidx];
 446         i_coord_offset   = DIM*inr;
 447
 448         /* Load i particle coords and add shift vector */
 449         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 450
 451         fix0             = _fjsp_setzero_v2r8();
 452         fiy0             = _fjsp_setzero_v2r8();
 453         fiz0             = _fjsp_setzero_v2r8();
 454
 455         /* Load parameters for i particles */
 456         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 457
 458         /* Start inner kernel loop */
 459         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 460         {
 461
 462             /* Get j neighbor index, and coordinate index */
 463             jnrA             = jjnr[jidx];
 464             jnrB             = jjnr[jidx+1];
 465             j_coord_offsetA  = DIM*jnrA;
 466             j_coord_offsetB  = DIM*jnrB;
 467
 468             /* load j atom coordinates */
 469             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 470                                               &jx0,&jy0,&jz0);
 471
 472             /* Calculate displacement vector */
 473             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 474             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 475             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 476
 477             /* Calculate squared distance and things based on it */
 478             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 479
 480             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 481
 482             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 483
 484             /* Load parameters for j particles */
 485             vdwjidx0A        = 2*vdwtype[jnrA+0];
 486             vdwjidx0B        = 2*vdwtype[jnrB+0];
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 493             {
 494
 495             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 499                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 500
 501             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 502                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 503
 504             /* Analytical LJ-PME */
 505             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 506             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 507             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 508             exponent         = gmx_simd_exp_d(ewcljrsq);
 509             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 510             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 511             /* f6A = 6 * C6grid * (1 - poly) */
 512             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 513             /* f6B = C6grid * exponent * beta^6 */
 514             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 515             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 516             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 517
 518             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 519
 520             fscal            = fvdw;
 521
 522             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 523
 524             /* Update vectorial force */
 525             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 526             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 527             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 528
 529             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 530
 531             }
 532
 533             /* Inner loop uses 51 flops */
 534         }
 535
 536         if(jidx<j_index_end)
 537         {
 538
 539             jnrA             = jjnr[jidx];
 540             j_coord_offsetA  = DIM*jnrA;
 541
 542             /* load j atom coordinates */
 543             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 544                                               &jx0,&jy0,&jz0);
 545
 546             /* Calculate displacement vector */
 547             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 548             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 549             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 550
 551             /* Calculate squared distance and things based on it */
 552             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 553
 554             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 555
 556             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 557
 558             /* Load parameters for j particles */
 559             vdwjidx0A        = 2*vdwtype[jnrA+0];
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 566             {
 567
 568             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 572                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 573
 574             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 575                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 576
 577             /* Analytical LJ-PME */
 578             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 579             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 580             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 581             exponent         = gmx_simd_exp_d(ewcljrsq);
 582             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 583             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 584             /* f6A = 6 * C6grid * (1 - poly) */
 585             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 586             /* f6B = C6grid * exponent * beta^6 */
 587             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 588             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 589             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 590
 591             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 592
 593             fscal            = fvdw;
 594
 595             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 596
 597             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 598
 599             /* Update vectorial force */
 600             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 601             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 602             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 603
 604             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 605
 606             }
 607
 608             /* Inner loop uses 51 flops */
 609         }
 610
 611         /* End of innermost loop */
 612
 613         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 614                                               f+i_coord_offset,fshift+i_shift_offset);
 615
 616         /* Increment number of inner iterations */
 617         inneriter                  += j_index_end - j_index_start;
 618
 619         /* Outer loop uses 6 flops */
 620     }
 621
 622     /* Increment number of outer iterations */
 623     outeriter        += nri;
 624
 625     /* Update outer/inner flops */
 626
 627     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*51);
 628 }