src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecNone_VdwLJEwSh_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_VdwLJEwSh_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_VdwLJEwSh_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     rcutoff_scalar   = fr->rvdw;
 123     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 124     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 125
 126     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 127     rvdw             = gmx_fjsp_set1_v2r8(fr->rvdw);
 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             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 208                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 209
 210             /* Analytical LJ-PME */
 211             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 212             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 213             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 214             exponent         = gmx_simd_exp_d(ewcljrsq);
 215             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 216             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 217             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 218             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 219             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 220             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 221                                _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));
 222             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 223             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);
 224
 225             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 226
 227             /* Update potential sum for this i atom from the interaction with this j atom. */
 228             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 229             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 230
 231             fscal            = fvdw;
 232
 233             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 234
 235             /* Update vectorial force */
 236             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 237             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 238             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 239
 240             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 241
 242             }
 243
 244             /* Inner loop uses 59 flops */
 245         }
 246
 247         if(jidx<j_index_end)
 248         {
 249
 250             jnrA             = jjnr[jidx];
 251             j_coord_offsetA  = DIM*jnrA;
 252
 253             /* load j atom coordinates */
 254             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 255                                               &jx0,&jy0,&jz0);
 256
 257             /* Calculate displacement vector */
 258             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 259             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 260             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 261
 262             /* Calculate squared distance and things based on it */
 263             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 264
 265             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 266
 267             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 268
 269             /* Load parameters for j particles */
 270             vdwjidx0A        = 2*vdwtype[jnrA+0];
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 277             {
 278
 279             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 280
 281             /* Compute parameters for interactions between i and j atoms */
 282             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 283                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 284
 285             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 286                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 287
 288             /* Analytical LJ-PME */
 289             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 290             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 291             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 292             exponent         = gmx_simd_exp_d(ewcljrsq);
 293             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 294             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 295             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 296             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 297             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 298             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 299                                _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));
 300             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 301             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);
 302
 303             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 307             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 308             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 309
 310             fscal            = fvdw;
 311
 312             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 313
 314             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 315
 316             /* Update vectorial force */
 317             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 318             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 319             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 320
 321             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 322
 323             }
 324
 325             /* Inner loop uses 59 flops */
 326         }
 327
 328         /* End of innermost loop */
 329
 330         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 331                                               f+i_coord_offset,fshift+i_shift_offset);
 332
 333         ggid                        = gid[iidx];
 334         /* Update potential energies */
 335         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 336
 337         /* Increment number of inner iterations */
 338         inneriter                  += j_index_end - j_index_start;
 339
 340         /* Outer loop uses 7 flops */
 341     }
 342
 343     /* Increment number of outer iterations */
 344     outeriter        += nri;
 345
 346     /* Update outer/inner flops */
 347
 348     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_VF,outeriter*7 + inneriter*59);
 349 }
 350 /*
 351  * Gromacs nonbonded kernel:   nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sparc64_hpc_ace_double
 352  * Electrostatics interaction: None
 353  * VdW interaction:            LJEwald
 354  * Geometry:                   Particle-Particle
 355  * Calculate force/pot:        Force
 356  */
 357 void
 358 nb_kernel_ElecNone_VdwLJEwSh_GeomP1P1_F_sparc64_hpc_ace_double
 359                     (t_nblist                    * gmx_restrict       nlist,
 360                      rvec                        * gmx_restrict          xx,
 361                      rvec                        * gmx_restrict          ff,
 362                      t_forcerec                  * gmx_restrict          fr,
 363                      t_mdatoms                   * gmx_restrict     mdatoms,
 364                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 365                      t_nrnb                      * gmx_restrict        nrnb)
 366 {
 367     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 368      * just 0 for non-waters.
 369      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 370      * jnr indices corresponding to data put in the four positions in the SIMD register.
 371      */
 372     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 373     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 374     int              jnrA,jnrB;
 375     int              j_coord_offsetA,j_coord_offsetB;
 376     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 377     real             rcutoff_scalar;
 378     real             *shiftvec,*fshift,*x,*f;
 379     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 380     int              vdwioffset0;
 381     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 382     int              vdwjidx0A,vdwjidx0B;
 383     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 384     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 385     int              nvdwtype;
 386     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 387     int              *vdwtype;
 388     real             *vdwparam;
 389     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 390     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 391     _fjsp_v2r8           c6grid_00;
 392     real                 *vdwgridparam;
 393     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 394     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 395     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 396     _fjsp_v2r8       itab_tmp;
 397     _fjsp_v2r8       dummy_mask,cutoff_mask;
 398     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 399     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 400     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 401
 402     x                = xx[0];
 403     f                = ff[0];
 404
 405     nri              = nlist->nri;
 406     iinr             = nlist->iinr;
 407     jindex           = nlist->jindex;
 408     jjnr             = nlist->jjnr;
 409     shiftidx         = nlist->shift;
 410     gid              = nlist->gid;
 411     shiftvec         = fr->shift_vec[0];
 412     fshift           = fr->fshift[0];
 413     nvdwtype         = fr->ntype;
 414     vdwparam         = fr->nbfp;
 415     vdwtype          = mdatoms->typeA;
 416     vdwgridparam     = fr->ljpme_c6grid;
 417     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 418     ewclj            = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
 419     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 420
 421     rcutoff_scalar   = fr->rvdw;
 422     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 423     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 424
 425     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 426     rvdw             = gmx_fjsp_set1_v2r8(fr->rvdw);
 427
 428     /* Avoid stupid compiler warnings */
 429     jnrA = jnrB = 0;
 430     j_coord_offsetA = 0;
 431     j_coord_offsetB = 0;
 432
 433     outeriter        = 0;
 434     inneriter        = 0;
 435
 436     /* Start outer loop over neighborlists */
 437     for(iidx=0; iidx<nri; iidx++)
 438     {
 439         /* Load shift vector for this list */
 440         i_shift_offset   = DIM*shiftidx[iidx];
 441
 442         /* Load limits for loop over neighbors */
 443         j_index_start    = jindex[iidx];
 444         j_index_end      = jindex[iidx+1];
 445
 446         /* Get outer coordinate index */
 447         inr              = iinr[iidx];
 448         i_coord_offset   = DIM*inr;
 449
 450         /* Load i particle coords and add shift vector */
 451         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 452
 453         fix0             = _fjsp_setzero_v2r8();
 454         fiy0             = _fjsp_setzero_v2r8();
 455         fiz0             = _fjsp_setzero_v2r8();
 456
 457         /* Load parameters for i particles */
 458         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 459
 460         /* Start inner kernel loop */
 461         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 462         {
 463
 464             /* Get j neighbor index, and coordinate index */
 465             jnrA             = jjnr[jidx];
 466             jnrB             = jjnr[jidx+1];
 467             j_coord_offsetA  = DIM*jnrA;
 468             j_coord_offsetB  = DIM*jnrB;
 469
 470             /* load j atom coordinates */
 471             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 472                                               &jx0,&jy0,&jz0);
 473
 474             /* Calculate displacement vector */
 475             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 476             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 477             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 481
 482             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 483
 484             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 485
 486             /* Load parameters for j particles */
 487             vdwjidx0A        = 2*vdwtype[jnrA+0];
 488             vdwjidx0B        = 2*vdwtype[jnrB+0];
 489
 490             /**************************
 491              * CALCULATE INTERACTIONS *
 492              **************************/
 493
 494             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 495             {
 496
 497             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 498
 499             /* Compute parameters for interactions between i and j atoms */
 500             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 501                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 502
 503             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 504                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 505
 506             /* Analytical LJ-PME */
 507             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 508             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 509             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 510             exponent         = gmx_simd_exp_d(ewcljrsq);
 511             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 512             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 513             /* f6A = 6 * C6grid * (1 - poly) */
 514             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 515             /* f6B = C6grid * exponent * beta^6 */
 516             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 517             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 518             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 519
 520             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 521
 522             fscal            = fvdw;
 523
 524             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 525
 526             /* Update vectorial force */
 527             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 528             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 529             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 530
 531             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 532
 533             }
 534
 535             /* Inner loop uses 51 flops */
 536         }
 537
 538         if(jidx<j_index_end)
 539         {
 540
 541             jnrA             = jjnr[jidx];
 542             j_coord_offsetA  = DIM*jnrA;
 543
 544             /* load j atom coordinates */
 545             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 546                                               &jx0,&jy0,&jz0);
 547
 548             /* Calculate displacement vector */
 549             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 550             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 551             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 552
 553             /* Calculate squared distance and things based on it */
 554             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 555
 556             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 557
 558             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 559
 560             /* Load parameters for j particles */
 561             vdwjidx0A        = 2*vdwtype[jnrA+0];
 562
 563             /**************************
 564              * CALCULATE INTERACTIONS *
 565              **************************/
 566
 567             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 568             {
 569
 570             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 574                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 575
 576             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 577                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 578
 579             /* Analytical LJ-PME */
 580             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 581             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 582             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 583             exponent         = gmx_simd_exp_d(ewcljrsq);
 584             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 585             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 586             /* f6A = 6 * C6grid * (1 - poly) */
 587             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 588             /* f6B = C6grid * exponent * beta^6 */
 589             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 590             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 591             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 592
 593             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 594
 595             fscal            = fvdw;
 596
 597             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 598
 599             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 600
 601             /* Update vectorial force */
 602             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 603             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 604             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 605
 606             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 607
 608             }
 609
 610             /* Inner loop uses 51 flops */
 611         }
 612
 613         /* End of innermost loop */
 614
 615         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 616                                               f+i_coord_offset,fshift+i_shift_offset);
 617
 618         /* Increment number of inner iterations */
 619         inneriter                  += j_index_end - j_index_start;
 620
 621         /* Outer loop uses 6 flops */
 622     }
 623
 624     /* Increment number of outer iterations */
 625     outeriter        += nri;
 626
 627     /* Update outer/inner flops */
 628
 629     inc_nrnb(nrnb,eNR_NBKERNEL_VDW_F,outeriter*6 + inneriter*51);
 630 }