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) 2012,2013,2014, 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_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_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 304                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 305
 306             /* EWALD ELECTROSTATICS */
 307
 308             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 309             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 310             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 311             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 312             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 313
 314             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 315             ewtabD           = _fjsp_setzero_v2r8();
 316             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 317             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 318             ewtabFn          = _fjsp_setzero_v2r8();
 319             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 320             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 321             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 322             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
 323             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 324
 325             /* LENNARD-JONES DISPERSION/REPULSION */
 326
 327             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 328             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 329             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 330             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 331                                            _fjsp_mul_v2r8(_fjsp_nmsub_v2r8( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 332             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 333
 334             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 338             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 339             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 340             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 341             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 342             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 343
 344             fscal            = _fjsp_add_v2r8(felec,fvdw);
 345
 346             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 347
 348             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 349
 350             /* Update vectorial force */
 351             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 352             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 353             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 354
 355             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 356
 357             }
 358
 359             /* Inner loop uses 67 flops */
 360         }
 361
 362         /* End of innermost loop */
 363
 364         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 365                                               f+i_coord_offset,fshift+i_shift_offset);
 366
 367         ggid                        = gid[iidx];
 368         /* Update potential energies */
 369         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 370         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 371
 372         /* Increment number of inner iterations */
 373         inneriter                  += j_index_end - j_index_start;
 374
 375         /* Outer loop uses 9 flops */
 376     }
 377
 378     /* Increment number of outer iterations */
 379     outeriter        += nri;
 380
 381     /* Update outer/inner flops */
 382
 383     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*67);
 384 }
 385 /*
 386  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sparc64_hpc_ace_double
 387  * Electrostatics interaction: Ewald
 388  * VdW interaction:            LennardJones
 389  * Geometry:                   Particle-Particle
 390  * Calculate force/pot:        Force
 391  */
 392 void
 393 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sparc64_hpc_ace_double
 394                     (t_nblist                    * gmx_restrict       nlist,
 395                      rvec                        * gmx_restrict          xx,
 396                      rvec                        * gmx_restrict          ff,
 397                      t_forcerec                  * gmx_restrict          fr,
 398                      t_mdatoms                   * gmx_restrict     mdatoms,
 399                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 400                      t_nrnb                      * gmx_restrict        nrnb)
 401 {
 402     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 403      * just 0 for non-waters.
 404      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 405      * jnr indices corresponding to data put in the four positions in the SIMD register.
 406      */
 407     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 408     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 409     int              jnrA,jnrB;
 410     int              j_coord_offsetA,j_coord_offsetB;
 411     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 412     real             rcutoff_scalar;
 413     real             *shiftvec,*fshift,*x,*f;
 414     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 415     int              vdwioffset0;
 416     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 417     int              vdwjidx0A,vdwjidx0B;
 418     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 419     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 420     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 421     real             *charge;
 422     int              nvdwtype;
 423     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 424     int              *vdwtype;
 425     real             *vdwparam;
 426     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 427     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 428     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 429     real             *ewtab;
 430     _fjsp_v2r8       itab_tmp;
 431     _fjsp_v2r8       dummy_mask,cutoff_mask;
 432     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 433     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 434     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 435
 436     x                = xx[0];
 437     f                = ff[0];
 438
 439     nri              = nlist->nri;
 440     iinr             = nlist->iinr;
 441     jindex           = nlist->jindex;
 442     jjnr             = nlist->jjnr;
 443     shiftidx         = nlist->shift;
 444     gid              = nlist->gid;
 445     shiftvec         = fr->shift_vec[0];
 446     fshift           = fr->fshift[0];
 447     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 448     charge           = mdatoms->chargeA;
 449     nvdwtype         = fr->ntype;
 450     vdwparam         = fr->nbfp;
 451     vdwtype          = mdatoms->typeA;
 452
 453     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 454     ewtab            = fr->ic->tabq_coul_F;
 455     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 456     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 457
 458     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 459     rcutoff_scalar   = fr->rcoulomb;
 460     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 461     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 462
 463     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 464     rvdw             = gmx_fjsp_set1_v2r8(fr->rvdw);
 465
 466     /* Avoid stupid compiler warnings */
 467     jnrA = jnrB = 0;
 468     j_coord_offsetA = 0;
 469     j_coord_offsetB = 0;
 470
 471     outeriter        = 0;
 472     inneriter        = 0;
 473
 474     /* Start outer loop over neighborlists */
 475     for(iidx=0; iidx<nri; iidx++)
 476     {
 477         /* Load shift vector for this list */
 478         i_shift_offset   = DIM*shiftidx[iidx];
 479
 480         /* Load limits for loop over neighbors */
 481         j_index_start    = jindex[iidx];
 482         j_index_end      = jindex[iidx+1];
 483
 484         /* Get outer coordinate index */
 485         inr              = iinr[iidx];
 486         i_coord_offset   = DIM*inr;
 487
 488         /* Load i particle coords and add shift vector */
 489         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 490
 491         fix0             = _fjsp_setzero_v2r8();
 492         fiy0             = _fjsp_setzero_v2r8();
 493         fiz0             = _fjsp_setzero_v2r8();
 494
 495         /* Load parameters for i particles */
 496         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 497         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 498
 499         /* Start inner kernel loop */
 500         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 501         {
 502
 503             /* Get j neighbor index, and coordinate index */
 504             jnrA             = jjnr[jidx];
 505             jnrB             = jjnr[jidx+1];
 506             j_coord_offsetA  = DIM*jnrA;
 507             j_coord_offsetB  = DIM*jnrB;
 508
 509             /* load j atom coordinates */
 510             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 511                                               &jx0,&jy0,&jz0);
 512
 513             /* Calculate displacement vector */
 514             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 515             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 516             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 517
 518             /* Calculate squared distance and things based on it */
 519             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 520
 521             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 522
 523             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 524
 525             /* Load parameters for j particles */
 526             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 527             vdwjidx0A        = 2*vdwtype[jnrA+0];
 528             vdwjidx0B        = 2*vdwtype[jnrB+0];
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 535             {
 536
 537             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 541             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 542                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 543
 544             /* EWALD ELECTROSTATICS */
 545
 546             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 547             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 548             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 549             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 550             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 551
 552             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 553                                          &ewtabF,&ewtabFn);
 554             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 555             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 556
 557             /* LENNARD-JONES DISPERSION/REPULSION */
 558
 559             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 560             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 561
 562             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 563
 564             fscal            = _fjsp_add_v2r8(felec,fvdw);
 565
 566             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 567
 568             /* Update vectorial force */
 569             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 570             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 571             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 572
 573             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 574
 575             }
 576
 577             /* Inner loop uses 49 flops */
 578         }
 579
 580         if(jidx<j_index_end)
 581         {
 582
 583             jnrA             = jjnr[jidx];
 584             j_coord_offsetA  = DIM*jnrA;
 585
 586             /* load j atom coordinates */
 587             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 588                                               &jx0,&jy0,&jz0);
 589
 590             /* Calculate displacement vector */
 591             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 592             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 593             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 594
 595             /* Calculate squared distance and things based on it */
 596             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 597
 598             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 599
 600             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 601
 602             /* Load parameters for j particles */
 603             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 604             vdwjidx0A        = 2*vdwtype[jnrA+0];
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 611             {
 612
 613             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 617             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 618                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 619
 620             /* EWALD ELECTROSTATICS */
 621
 622             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 623             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 624             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 625             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 626             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 627
 628             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 629             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 630             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 631
 632             /* LENNARD-JONES DISPERSION/REPULSION */
 633
 634             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 635             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 636
 637             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 638
 639             fscal            = _fjsp_add_v2r8(felec,fvdw);
 640
 641             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 642
 643             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 644
 645             /* Update vectorial force */
 646             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 647             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 648             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 649
 650             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 651
 652             }
 653
 654             /* Inner loop uses 49 flops */
 655         }
 656
 657         /* End of innermost loop */
 658
 659         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 660                                               f+i_coord_offset,fshift+i_shift_offset);
 661
 662         /* Increment number of inner iterations */
 663         inneriter                  += j_index_end - j_index_start;
 664
 665         /* Outer loop uses 7 flops */
 666     }
 667
 668     /* Increment number of outer iterations */
 669     outeriter        += nri;
 670
 671     /* Update outer/inner flops */
 672
 673     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*49);
 674 }