src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecNone_VdwLJEw_GeomP1P1_sparc64_hpc_ace_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/legacyheaders/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: None
  54  * VdW interaction:            LJEwald
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_VF_sparc64_hpc_ace_double
  60                     (t_nblist                    * gmx_restrict       nlist,
  61                      rvec                        * gmx_restrict          xx,
  62                      rvec                        * gmx_restrict          ff,
  63                      t_forcerec                  * gmx_restrict          fr,
  64                      t_mdatoms                   * gmx_restrict     mdatoms,
  65                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  66                      t_nrnb                      * gmx_restrict        nrnb)
  67 {
  68     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  69      * just 0 for non-waters.
  70      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  71      * jnr indices corresponding to data put in the four positions in the SIMD register.
  72      */
  73     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  74     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  75     int              jnrA,jnrB;
  76     int              j_coord_offsetA,j_coord_offsetB;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  81     int              vdwioffset0;
  82     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  83     int              vdwjidx0A,vdwjidx0B;
  84     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  85     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  86     int              nvdwtype;
  87     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  91     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  92     _fjsp_v2r8           c6grid_00;
  93     real                 *vdwgridparam;
  94     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  95     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
  96     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
  97     _fjsp_v2r8       itab_tmp;
  98     _fjsp_v2r8       dummy_mask,cutoff_mask;
  99     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 100     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 101     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 102
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     nvdwtype         = fr->ntype;
 115     vdwparam         = fr->nbfp;
 116     vdwtype          = mdatoms->typeA;
 117     vdwgridparam     = fr->ljpme_c6grid;
 118     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 119     ewclj            = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
 120     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _fjsp_setzero_v2r8();
 148         fiy0             = _fjsp_setzero_v2r8();
 149         fiz0             = _fjsp_setzero_v2r8();
 150
 151         /* Load parameters for i particles */
 152         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 153
 154         /* Reset potential sums */
 155         vvdwsum          = _fjsp_setzero_v2r8();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 173             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 174             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 180
 181             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 182
 183             /* Load parameters for j particles */
 184             vdwjidx0A        = 2*vdwtype[jnrA+0];
 185             vdwjidx0B        = 2*vdwtype[jnrB+0];
 186
 187             /**************************
 188              * CALCULATE INTERACTIONS *
 189              **************************/
 190
 191             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 192
 193             /* Compute parameters for interactions between i and j atoms */
 194             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 195                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 196
 197             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 198                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 199
 200             /* Analytical LJ-PME */
 201             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 202             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 203             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 204             exponent         = gmx_simd_exp_d(ewcljrsq);
 205             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 206             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 207             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 208             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 209             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 210             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 211             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 212             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);
 213
 214             /* Update potential sum for this i atom from the interaction with this j atom. */
 215             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 216
 217             fscal            = fvdw;
 218
 219             /* Update vectorial force */
 220             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 221             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 222             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 223
 224             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 225
 226             /* Inner loop uses 50 flops */
 227         }
 228
 229         if(jidx<j_index_end)
 230         {
 231
 232             jnrA             = jjnr[jidx];
 233             j_coord_offsetA  = DIM*jnrA;
 234
 235             /* load j atom coordinates */
 236             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 237                                               &jx0,&jy0,&jz0);
 238
 239             /* Calculate displacement vector */
 240             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 241             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 242             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 243
 244             /* Calculate squared distance and things based on it */
 245             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 246
 247             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 248
 249             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 250
 251             /* Load parameters for j particles */
 252             vdwjidx0A        = 2*vdwtype[jnrA+0];
 253
 254             /**************************
 255              * CALCULATE INTERACTIONS *
 256              **************************/
 257
 258             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 262                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 263
 264             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 265                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 266
 267             /* Analytical LJ-PME */
 268             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 269             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 270             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 271             exponent         = gmx_simd_exp_d(ewcljrsq);
 272             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 273             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 274             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 275             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 276             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 277             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 278             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 279             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);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 283             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 284
 285             fscal            = fvdw;
 286
 287             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 288
 289             /* Update vectorial force */
 290             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 291             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 292             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 293
 294             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 295
 296             /* Inner loop uses 50 flops */
 297         }
 298
 299         /* End of innermost loop */
 300
 301         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 302                                               f+i_coord_offset,fshift+i_shift_offset);
 303
 304         ggid                        = gid[iidx];
 305         /* Update potential energies */
 306         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 307
 308         /* Increment number of inner iterations */
 309         inneriter                  += j_index_end - j_index_start;
 310
 311         /* Outer loop uses 7 flops */
 312     }
 313
 314     /* Increment number of outer iterations */
 315     outeriter        += nri;
 316
 317     /* Update outer/inner flops */
 318
 319     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*50);
 320 }
 321 /*
 322  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sparc64_hpc_ace_double
 323  * Electrostatics interaction: None
 324  * VdW interaction:            LJEwald
 325  * Geometry:                   Particle-Particle
 326  * Calculate force/pot:        Force
 327  */
 328 void
 329 nb_kernel_ElecNone_VdwLJEw_GeomP1P1_F_sparc64_hpc_ace_double
 330                     (t_nblist                    * gmx_restrict       nlist,
 331                      rvec                        * gmx_restrict          xx,
 332                      rvec                        * gmx_restrict          ff,
 333                      t_forcerec                  * gmx_restrict          fr,
 334                      t_mdatoms                   * gmx_restrict     mdatoms,
 335                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 336                      t_nrnb                      * gmx_restrict        nrnb)
 337 {
 338     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 339      * just 0 for non-waters.
 340      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 341      * jnr indices corresponding to data put in the four positions in the SIMD register.
 342      */
 343     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 344     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 345     int              jnrA,jnrB;
 346     int              j_coord_offsetA,j_coord_offsetB;
 347     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 348     real             rcutoff_scalar;
 349     real             *shiftvec,*fshift,*x,*f;
 350     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 351     int              vdwioffset0;
 352     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 353     int              vdwjidx0A,vdwjidx0B;
 354     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 355     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 356     int              nvdwtype;
 357     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 358     int              *vdwtype;
 359     real             *vdwparam;
 360     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 361     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 362     _fjsp_v2r8           c6grid_00;
 363     real                 *vdwgridparam;
 364     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 365     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 366     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 367     _fjsp_v2r8       itab_tmp;
 368     _fjsp_v2r8       dummy_mask,cutoff_mask;
 369     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 370     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 371     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 372
 373     x                = xx[0];
 374     f                = ff[0];
 375
 376     nri              = nlist->nri;
 377     iinr             = nlist->iinr;
 378     jindex           = nlist->jindex;
 379     jjnr             = nlist->jjnr;
 380     shiftidx         = nlist->shift;
 381     gid              = nlist->gid;
 382     shiftvec         = fr->shift_vec[0];
 383     fshift           = fr->fshift[0];
 384     nvdwtype         = fr->ntype;
 385     vdwparam         = fr->nbfp;
 386     vdwtype          = mdatoms->typeA;
 387     vdwgridparam     = fr->ljpme_c6grid;
 388     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 389     ewclj            = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
 390     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 391
 392     /* Avoid stupid compiler warnings */
 393     jnrA = jnrB = 0;
 394     j_coord_offsetA = 0;
 395     j_coord_offsetB = 0;
 396
 397     outeriter        = 0;
 398     inneriter        = 0;
 399
 400     /* Start outer loop over neighborlists */
 401     for(iidx=0; iidx<nri; iidx++)
 402     {
 403         /* Load shift vector for this list */
 404         i_shift_offset   = DIM*shiftidx[iidx];
 405
 406         /* Load limits for loop over neighbors */
 407         j_index_start    = jindex[iidx];
 408         j_index_end      = jindex[iidx+1];
 409
 410         /* Get outer coordinate index */
 411         inr              = iinr[iidx];
 412         i_coord_offset   = DIM*inr;
 413
 414         /* Load i particle coords and add shift vector */
 415         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 416
 417         fix0             = _fjsp_setzero_v2r8();
 418         fiy0             = _fjsp_setzero_v2r8();
 419         fiz0             = _fjsp_setzero_v2r8();
 420
 421         /* Load parameters for i particles */
 422         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 423
 424         /* Start inner kernel loop */
 425         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 426         {
 427
 428             /* Get j neighbor index, and coordinate index */
 429             jnrA             = jjnr[jidx];
 430             jnrB             = jjnr[jidx+1];
 431             j_coord_offsetA  = DIM*jnrA;
 432             j_coord_offsetB  = DIM*jnrB;
 433
 434             /* load j atom coordinates */
 435             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 436                                               &jx0,&jy0,&jz0);
 437
 438             /* Calculate displacement vector */
 439             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 440             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 441             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 442
 443             /* Calculate squared distance and things based on it */
 444             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 445
 446             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 447
 448             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 449
 450             /* Load parameters for j particles */
 451             vdwjidx0A        = 2*vdwtype[jnrA+0];
 452             vdwjidx0B        = 2*vdwtype[jnrB+0];
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 459
 460             /* Compute parameters for interactions between i and j atoms */
 461             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 462                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 463
 464             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 465                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 466
 467             /* Analytical LJ-PME */
 468             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 469             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 470             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 471             exponent         = gmx_simd_exp_d(ewcljrsq);
 472             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 473             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 474             /* f6A = 6 * C6grid * (1 - poly) */
 475             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 476             /* f6B = C6grid * exponent * beta^6 */
 477             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 478             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 479             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 480
 481             fscal            = fvdw;
 482
 483             /* Update vectorial force */
 484             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 485             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 486             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 487
 488             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 489
 490             /* Inner loop uses 48 flops */
 491         }
 492
 493         if(jidx<j_index_end)
 494         {
 495
 496             jnrA             = jjnr[jidx];
 497             j_coord_offsetA  = DIM*jnrA;
 498
 499             /* load j atom coordinates */
 500             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 501                                               &jx0,&jy0,&jz0);
 502
 503             /* Calculate displacement vector */
 504             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 505             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 506             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 507
 508             /* Calculate squared distance and things based on it */
 509             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 510
 511             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 512
 513             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 514
 515             /* Load parameters for j particles */
 516             vdwjidx0A        = 2*vdwtype[jnrA+0];
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 526                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 527
 528             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 529                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 530
 531             /* Analytical LJ-PME */
 532             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 533             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 534             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 535             exponent         = gmx_simd_exp_d(ewcljrsq);
 536             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 537             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 538             /* f6A = 6 * C6grid * (1 - poly) */
 539             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 540             /* f6B = C6grid * exponent * beta^6 */
 541             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 542             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 543             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 544
 545             fscal            = fvdw;
 546
 547             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 548
 549             /* Update vectorial force */
 550             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 551             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 552             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 553
 554             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 555
 556             /* Inner loop uses 48 flops */
 557         }
 558
 559         /* End of innermost loop */
 560
 561         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 562                                               f+i_coord_offset,fshift+i_shift_offset);
 563
 564         /* Increment number of inner iterations */
 565         inneriter                  += j_index_end - j_index_start;
 566
 567         /* Outer loop uses 6 flops */
 568     }
 569
 570     /* Increment number of outer iterations */
 571     outeriter        += nri;
 572
 573     /* Update outer/inner flops */
 574
 575     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*48);
 576 }