src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecNone_VdwLJEw_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_VdwLJEw_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_VdwLJEw_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     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124
 125     outeriter        = 0;
 126     inneriter        = 0;
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 144
 145         fix0             = _fjsp_setzero_v2r8();
 146         fiy0             = _fjsp_setzero_v2r8();
 147         fiz0             = _fjsp_setzero_v2r8();
 148
 149         /* Load parameters for i particles */
 150         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152         /* Reset potential sums */
 153         vvdwsum          = _fjsp_setzero_v2r8();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             j_coord_offsetA  = DIM*jnrA;
 163             j_coord_offsetB  = DIM*jnrB;
 164
 165             /* load j atom coordinates */
 166             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 167                                               &jx0,&jy0,&jz0);
 168
 169             /* Calculate displacement vector */
 170             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 171             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 172             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 173
 174             /* Calculate squared distance and things based on it */
 175             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 176
 177             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 178
 179             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 180
 181             /* Load parameters for j particles */
 182             vdwjidx0A        = 2*vdwtype[jnrA+0];
 183             vdwjidx0B        = 2*vdwtype[jnrB+0];
 184
 185             /**************************
 186              * CALCULATE INTERACTIONS *
 187              **************************/
 188
 189             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 193                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 194
 195             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 196                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 197
 198             /* Analytical LJ-PME */
 199             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 200             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 201             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 202             exponent         = gmx_simd_exp_d(ewcljrsq);
 203             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 204             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 205             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 206             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 207             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 208             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 209             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 210             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);
 211
 212             /* Update potential sum for this i atom from the interaction with this j atom. */
 213             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 214
 215             fscal            = fvdw;
 216
 217             /* Update vectorial force */
 218             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 219             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 220             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 221
 222             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 223
 224             /* Inner loop uses 50 flops */
 225         }
 226
 227         if(jidx<j_index_end)
 228         {
 229
 230             jnrA             = jjnr[jidx];
 231             j_coord_offsetA  = DIM*jnrA;
 232
 233             /* load j atom coordinates */
 234             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 235                                               &jx0,&jy0,&jz0);
 236
 237             /* Calculate displacement vector */
 238             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 239             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 240             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 241
 242             /* Calculate squared distance and things based on it */
 243             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 244
 245             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 246
 247             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 248
 249             /* Load parameters for j particles */
 250             vdwjidx0A        = 2*vdwtype[jnrA+0];
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 257
 258             /* Compute parameters for interactions between i and j atoms */
 259             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 260                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 261
 262             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 263                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 264
 265             /* Analytical LJ-PME */
 266             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 267             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 268             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 269             exponent         = gmx_simd_exp_d(ewcljrsq);
 270             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 271             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 272             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 273             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 274             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 275             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 276             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 277             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);
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 281             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 282
 283             fscal            = fvdw;
 284
 285             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 286
 287             /* Update vectorial force */
 288             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 289             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 290             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 291
 292             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 293
 294             /* Inner loop uses 50 flops */
 295         }
 296
 297         /* End of innermost loop */
 298
 299         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 300                                               f+i_coord_offset,fshift+i_shift_offset);
 301
 302         ggid                        = gid[iidx];
 303         /* Update potential energies */
 304         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 305
 306         /* Increment number of inner iterations */
 307         inneriter                  += j_index_end - j_index_start;
 308
 309         /* Outer loop uses 7 flops */
 310     }
 311
 312     /* Increment number of outer iterations */
 313     outeriter        += nri;
 314
 315     /* Update outer/inner flops */
 316
 317     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*50);
 318 }
 319 /*
 320  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sparc64_hpc_ace_double
 321  * Electrostatics interaction: None
 322  * VdW interaction:            LJEwald
 323  * Geometry:                   Particle-Particle
 324  * Calculate force/pot:        Force
 325  */
 326 void
 327 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sparc64_hpc_ace_double
 328                     (t_nblist                    * gmx_restrict       nlist,
 329                      rvec                        * gmx_restrict          xx,
 330                      rvec                        * gmx_restrict          ff,
 331                      struct t_forcerec           * gmx_restrict          fr,
 332                      t_mdatoms                   * gmx_restrict     mdatoms,
 333                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 334                      t_nrnb                      * gmx_restrict        nrnb)
 335 {
 336     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 337      * just 0 for non-waters.
 338      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 339      * jnr indices corresponding to data put in the four positions in the SIMD register.
 340      */
 341     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 342     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 343     int              jnrA,jnrB;
 344     int              j_coord_offsetA,j_coord_offsetB;
 345     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 346     real             rcutoff_scalar;
 347     real             *shiftvec,*fshift,*x,*f;
 348     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 349     int              vdwioffset0;
 350     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 351     int              vdwjidx0A,vdwjidx0B;
 352     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 353     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 354     int              nvdwtype;
 355     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 356     int              *vdwtype;
 357     real             *vdwparam;
 358     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 359     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 360     _fjsp_v2r8           c6grid_00;
 361     real                 *vdwgridparam;
 362     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 363     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 364     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 365     _fjsp_v2r8       itab_tmp;
 366     _fjsp_v2r8       dummy_mask,cutoff_mask;
 367     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 368     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 369     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 370
 371     x                = xx[0];
 372     f                = ff[0];
 373
 374     nri              = nlist->nri;
 375     iinr             = nlist->iinr;
 376     jindex           = nlist->jindex;
 377     jjnr             = nlist->jjnr;
 378     shiftidx         = nlist->shift;
 379     gid              = nlist->gid;
 380     shiftvec         = fr->shift_vec[0];
 381     fshift           = fr->fshift[0];
 382     nvdwtype         = fr->ntype;
 383     vdwparam         = fr->nbfp;
 384     vdwtype          = mdatoms->typeA;
 385     vdwgridparam     = fr->ljpme_c6grid;
 386     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 387     ewclj            = gmx_fjsp_set1_v2r8(fr->ic->ewaldcoeff_lj);
 388     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 389
 390     /* Avoid stupid compiler warnings */
 391     jnrA = jnrB = 0;
 392     j_coord_offsetA = 0;
 393     j_coord_offsetB = 0;
 394
 395     outeriter        = 0;
 396     inneriter        = 0;
 397
 398     /* Start outer loop over neighborlists */
 399     for(iidx=0; iidx<nri; iidx++)
 400     {
 401         /* Load shift vector for this list */
 402         i_shift_offset   = DIM*shiftidx[iidx];
 403
 404         /* Load limits for loop over neighbors */
 405         j_index_start    = jindex[iidx];
 406         j_index_end      = jindex[iidx+1];
 407
 408         /* Get outer coordinate index */
 409         inr              = iinr[iidx];
 410         i_coord_offset   = DIM*inr;
 411
 412         /* Load i particle coords and add shift vector */
 413         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 414
 415         fix0             = _fjsp_setzero_v2r8();
 416         fiy0             = _fjsp_setzero_v2r8();
 417         fiz0             = _fjsp_setzero_v2r8();
 418
 419         /* Load parameters for i particles */
 420         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 421
 422         /* Start inner kernel loop */
 423         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 424         {
 425
 426             /* Get j neighbor index, and coordinate index */
 427             jnrA             = jjnr[jidx];
 428             jnrB             = jjnr[jidx+1];
 429             j_coord_offsetA  = DIM*jnrA;
 430             j_coord_offsetB  = DIM*jnrB;
 431
 432             /* load j atom coordinates */
 433             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 434                                               &jx0,&jy0,&jz0);
 435
 436             /* Calculate displacement vector */
 437             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 438             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 439             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 443
 444             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 445
 446             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 447
 448             /* Load parameters for j particles */
 449             vdwjidx0A        = 2*vdwtype[jnrA+0];
 450             vdwjidx0B        = 2*vdwtype[jnrB+0];
 451
 452             /**************************
 453              * CALCULATE INTERACTIONS *
 454              **************************/
 455
 456             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 457
 458             /* Compute parameters for interactions between i and j atoms */
 459             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 460                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 461
 462             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 463                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 464
 465             /* Analytical LJ-PME */
 466             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 467             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 468             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 469             exponent         = gmx_simd_exp_d(ewcljrsq);
 470             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 471             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 472             /* f6A = 6 * C6grid * (1 - poly) */
 473             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 474             /* f6B = C6grid * exponent * beta^6 */
 475             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 476             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 477             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 478
 479             fscal            = fvdw;
 480
 481             /* Update vectorial force */
 482             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 483             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 484             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 485
 486             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 487
 488             /* Inner loop uses 48 flops */
 489         }
 490
 491         if(jidx<j_index_end)
 492         {
 493
 494             jnrA             = jjnr[jidx];
 495             j_coord_offsetA  = DIM*jnrA;
 496
 497             /* load j atom coordinates */
 498             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 499                                               &jx0,&jy0,&jz0);
 500
 501             /* Calculate displacement vector */
 502             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 503             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 504             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 505
 506             /* Calculate squared distance and things based on it */
 507             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 508
 509             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 510
 511             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 512
 513             /* Load parameters for j particles */
 514             vdwjidx0A        = 2*vdwtype[jnrA+0];
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 524                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 525
 526             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 527                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 528
 529             /* Analytical LJ-PME */
 530             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 531             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 532             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 533             exponent         = gmx_simd_exp_d(ewcljrsq);
 534             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 535             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 536             /* f6A = 6 * C6grid * (1 - poly) */
 537             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 538             /* f6B = C6grid * exponent * beta^6 */
 539             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 540             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 541             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 542
 543             fscal            = fvdw;
 544
 545             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 546
 547             /* Update vectorial force */
 548             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 549             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 550             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 551
 552             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 553
 554             /* Inner loop uses 48 flops */
 555         }
 556
 557         /* End of innermost loop */
 558
 559         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 560                                               f+i_coord_offset,fshift+i_shift_offset);
 561
 562         /* Increment number of inner iterations */
 563         inneriter                  += j_index_end - j_index_start;
 564
 565         /* Outer loop uses 6 flops */
 566     }
 567
 568     /* Increment number of outer iterations */
 569     outeriter        += nri;
 570
 571     /* Update outer/inner flops */
 572
 573     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*48);
 574 }