src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_sparc64_hpc_ace_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: Ewald
  54  * VdW interaction:            LennardJones
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecEwSh_VdwLJSh_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     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     int              nvdwtype;
  89     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  90     int              *vdwtype;
  91     real             *vdwparam;
  92     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  93     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  94     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  95     real             *ewtab;
  96     _fjsp_v2r8       itab_tmp;
  97     _fjsp_v2r8       dummy_mask,cutoff_mask;
  98     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  99     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 100     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 101
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 122     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 127     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 128
 129     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 130     rvdw             = gmx_fjsp_set1_v2r8(fr->rvdw);
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 156
 157         fix0             = _fjsp_setzero_v2r8();
 158         fiy0             = _fjsp_setzero_v2r8();
 159         fiz0             = _fjsp_setzero_v2r8();
 160
 161         /* Load parameters for i particles */
 162         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 163         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 164
 165         /* Reset potential sums */
 166         velecsum         = _fjsp_setzero_v2r8();
 167         vvdwsum          = _fjsp_setzero_v2r8();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178
 179             /* load j atom coordinates */
 180             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 185             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 186             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 187
 188             /* Calculate squared distance and things based on it */
 189             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 190
 191             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 192
 193             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 197             vdwjidx0A        = 2*vdwtype[jnrA+0];
 198             vdwjidx0B        = 2*vdwtype[jnrB+0];
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 203
 204             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 205             {
 206
 207             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 208
 209             /* Compute parameters for interactions between i and j atoms */
 210             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 211             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 212                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 213
 214             /* EWALD ELECTROSTATICS */
 215
 216             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 217             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 218             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 219             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 220             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 221
 222             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 223             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 224             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 225             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 226             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 227             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 228             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 229             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 230             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
 231             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 232
 233             /* LENNARD-JONES DISPERSION/REPULSION */
 234
 235             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 236             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 237             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 238             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 239                                            _fjsp_mul_v2r8(_fjsp_nmsub_v2r8( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 240             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 241
 242             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 246             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 247             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 248             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 249
 250             fscal            = _fjsp_add_v2r8(felec,fvdw);
 251
 252             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 253
 254             /* Update vectorial force */
 255             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 256             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 257             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 258
 259             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 260
 261             }
 262
 263             /* Inner loop uses 67 flops */
 264         }
 265
 266         if(jidx<j_index_end)
 267         {
 268
 269             jnrA             = jjnr[jidx];
 270             j_coord_offsetA  = DIM*jnrA;
 271
 272             /* load j atom coordinates */
 273             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 274                                               &jx0,&jy0,&jz0);
 275
 276             /* Calculate displacement vector */
 277             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 278             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 279             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 280
 281             /* Calculate squared distance and things based on it */
 282             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 283
 284             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 285
 286             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 287
 288             /* Load parameters for j particles */
 289             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 290             vdwjidx0A        = 2*vdwtype[jnrA+0];
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 297             {
 298
 299             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 303             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 304
 305             /* EWALD ELECTROSTATICS */
 306
 307             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 308             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 309             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 310             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 311             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 312
 313             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 314             ewtabD           = _fjsp_setzero_v2r8();
 315             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 316             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 317             ewtabFn          = _fjsp_setzero_v2r8();
 318             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 319             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 320             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 321             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
 322             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 323
 324             /* LENNARD-JONES DISPERSION/REPULSION */
 325
 326             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 327             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 328             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 329             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 330                                            _fjsp_mul_v2r8(_fjsp_nmsub_v2r8( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 331             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 332
 333             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 337             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 338             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 339             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 340             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 341             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 342
 343             fscal            = _fjsp_add_v2r8(felec,fvdw);
 344
 345             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 346
 347             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 348
 349             /* Update vectorial force */
 350             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 351             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 352             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 353
 354             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 355
 356             }
 357
 358             /* Inner loop uses 67 flops */
 359         }
 360
 361         /* End of innermost loop */
 362
 363         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 364                                               f+i_coord_offset,fshift+i_shift_offset);
 365
 366         ggid                        = gid[iidx];
 367         /* Update potential energies */
 368         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 369         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 370
 371         /* Increment number of inner iterations */
 372         inneriter                  += j_index_end - j_index_start;
 373
 374         /* Outer loop uses 9 flops */
 375     }
 376
 377     /* Increment number of outer iterations */
 378     outeriter        += nri;
 379
 380     /* Update outer/inner flops */
 381
 382     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*67);
 383 }
 384 /*
 385  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sparc64_hpc_ace_double
 386  * Electrostatics interaction: Ewald
 387  * VdW interaction:            LennardJones
 388  * Geometry:                   Particle-Particle
 389  * Calculate force/pot:        Force
 390  */
 391 void
 392 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sparc64_hpc_ace_double
 393                     (t_nblist                    * gmx_restrict       nlist,
 394                      rvec                        * gmx_restrict          xx,
 395                      rvec                        * gmx_restrict          ff,
 396                      t_forcerec                  * gmx_restrict          fr,
 397                      t_mdatoms                   * gmx_restrict     mdatoms,
 398                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 399                      t_nrnb                      * gmx_restrict        nrnb)
 400 {
 401     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 402      * just 0 for non-waters.
 403      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 404      * jnr indices corresponding to data put in the four positions in the SIMD register.
 405      */
 406     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 407     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 408     int              jnrA,jnrB;
 409     int              j_coord_offsetA,j_coord_offsetB;
 410     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 411     real             rcutoff_scalar;
 412     real             *shiftvec,*fshift,*x,*f;
 413     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 414     int              vdwioffset0;
 415     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 416     int              vdwjidx0A,vdwjidx0B;
 417     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 418     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 419     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 420     real             *charge;
 421     int              nvdwtype;
 422     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 423     int              *vdwtype;
 424     real             *vdwparam;
 425     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 426     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 427     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 428     real             *ewtab;
 429     _fjsp_v2r8       itab_tmp;
 430     _fjsp_v2r8       dummy_mask,cutoff_mask;
 431     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 432     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 433     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 434
 435     x                = xx[0];
 436     f                = ff[0];
 437
 438     nri              = nlist->nri;
 439     iinr             = nlist->iinr;
 440     jindex           = nlist->jindex;
 441     jjnr             = nlist->jjnr;
 442     shiftidx         = nlist->shift;
 443     gid              = nlist->gid;
 444     shiftvec         = fr->shift_vec[0];
 445     fshift           = fr->fshift[0];
 446     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 447     charge           = mdatoms->chargeA;
 448     nvdwtype         = fr->ntype;
 449     vdwparam         = fr->nbfp;
 450     vdwtype          = mdatoms->typeA;
 451
 452     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 453     ewtab            = fr->ic->tabq_coul_F;
 454     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 455     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 456
 457     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 458     rcutoff_scalar   = fr->rcoulomb;
 459     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 460     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 461
 462     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 463     rvdw             = gmx_fjsp_set1_v2r8(fr->rvdw);
 464
 465     /* Avoid stupid compiler warnings */
 466     jnrA = jnrB = 0;
 467     j_coord_offsetA = 0;
 468     j_coord_offsetB = 0;
 469
 470     outeriter        = 0;
 471     inneriter        = 0;
 472
 473     /* Start outer loop over neighborlists */
 474     for(iidx=0; iidx<nri; iidx++)
 475     {
 476         /* Load shift vector for this list */
 477         i_shift_offset   = DIM*shiftidx[iidx];
 478
 479         /* Load limits for loop over neighbors */
 480         j_index_start    = jindex[iidx];
 481         j_index_end      = jindex[iidx+1];
 482
 483         /* Get outer coordinate index */
 484         inr              = iinr[iidx];
 485         i_coord_offset   = DIM*inr;
 486
 487         /* Load i particle coords and add shift vector */
 488         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 489
 490         fix0             = _fjsp_setzero_v2r8();
 491         fiy0             = _fjsp_setzero_v2r8();
 492         fiz0             = _fjsp_setzero_v2r8();
 493
 494         /* Load parameters for i particles */
 495         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 496         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 497
 498         /* Start inner kernel loop */
 499         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 500         {
 501
 502             /* Get j neighbor index, and coordinate index */
 503             jnrA             = jjnr[jidx];
 504             jnrB             = jjnr[jidx+1];
 505             j_coord_offsetA  = DIM*jnrA;
 506             j_coord_offsetB  = DIM*jnrB;
 507
 508             /* load j atom coordinates */
 509             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 510                                               &jx0,&jy0,&jz0);
 511
 512             /* Calculate displacement vector */
 513             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 514             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 515             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 516
 517             /* Calculate squared distance and things based on it */
 518             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 519
 520             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 521
 522             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 523
 524             /* Load parameters for j particles */
 525             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 526             vdwjidx0A        = 2*vdwtype[jnrA+0];
 527             vdwjidx0B        = 2*vdwtype[jnrB+0];
 528
 529             /**************************
 530              * CALCULATE INTERACTIONS *
 531              **************************/
 532
 533             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 534             {
 535
 536             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 540             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 541                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 542
 543             /* EWALD ELECTROSTATICS */
 544
 545             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 546             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 547             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 548             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 549             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 550
 551             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 552                                          &ewtabF,&ewtabFn);
 553             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 554             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 555
 556             /* LENNARD-JONES DISPERSION/REPULSION */
 557
 558             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 559             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 560
 561             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 562
 563             fscal            = _fjsp_add_v2r8(felec,fvdw);
 564
 565             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 566
 567             /* Update vectorial force */
 568             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 569             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 570             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 571
 572             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 573
 574             }
 575
 576             /* Inner loop uses 49 flops */
 577         }
 578
 579         if(jidx<j_index_end)
 580         {
 581
 582             jnrA             = jjnr[jidx];
 583             j_coord_offsetA  = DIM*jnrA;
 584
 585             /* load j atom coordinates */
 586             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 587                                               &jx0,&jy0,&jz0);
 588
 589             /* Calculate displacement vector */
 590             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 591             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 592             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 596
 597             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 598
 599             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 600
 601             /* Load parameters for j particles */
 602             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 603             vdwjidx0A        = 2*vdwtype[jnrA+0];
 604
 605             /**************************
 606              * CALCULATE INTERACTIONS *
 607              **************************/
 608
 609             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 610             {
 611
 612             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 616             gmx_fjsp_load_1pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 617
 618             /* EWALD ELECTROSTATICS */
 619
 620             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 621             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 622             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 623             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 624             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 625
 626             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 627             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 628             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 629
 630             /* LENNARD-JONES DISPERSION/REPULSION */
 631
 632             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 633             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 634
 635             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 636
 637             fscal            = _fjsp_add_v2r8(felec,fvdw);
 638
 639             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 640
 641             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 642
 643             /* Update vectorial force */
 644             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 645             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 646             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 647
 648             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 649
 650             }
 651
 652             /* Inner loop uses 49 flops */
 653         }
 654
 655         /* End of innermost loop */
 656
 657         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 658                                               f+i_coord_offset,fshift+i_shift_offset);
 659
 660         /* Increment number of inner iterations */
 661         inneriter                  += j_index_end - j_index_start;
 662
 663         /* Outer loop uses 7 flops */
 664     }
 665
 666     /* Increment number of outer iterations */
 667     outeriter        += nri;
 668
 669     /* Update outer/inner flops */
 670
 671     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*49);
 672 }