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

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_VF_sparc64_hpc_ace_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  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       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  93     real             *ewtab;
  94     _fjsp_v2r8       itab_tmp;
  95     _fjsp_v2r8       dummy_mask,cutoff_mask;
  96     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  97     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  98     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  99
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 118     ewtab            = fr->ic->tabq_coul_FDV0;
 119     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 120     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->ic->rcoulomb;
 124     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 125     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 126
 127     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 128     rvdw             = gmx_fjsp_set1_v2r8(fr->ic->rvdw);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 154
 155         fix0             = _fjsp_setzero_v2r8();
 156         fiy0             = _fjsp_setzero_v2r8();
 157         fiz0             = _fjsp_setzero_v2r8();
 158
 159         /* Load parameters for i particles */
 160         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 161         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 162
 163         /* Reset potential sums */
 164         velecsum         = _fjsp_setzero_v2r8();
 165         vvdwsum          = _fjsp_setzero_v2r8();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 183             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 184             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 185
 186             /* Calculate squared distance and things based on it */
 187             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 188
 189             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 190
 191             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 192
 193             /* Load parameters for j particles */
 194             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 195             vdwjidx0A        = 2*vdwtype[jnrA+0];
 196             vdwjidx0B        = 2*vdwtype[jnrB+0];
 197
 198             /**************************
 199              * CALCULATE INTERACTIONS *
 200              **************************/
 201
 202             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 203             {
 204
 205             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 209             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 210                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 211
 212             /* EWALD ELECTROSTATICS */
 213
 214             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 215             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 216             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 217             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 218             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 219
 220             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 221             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 222             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 223             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 224             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 225             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 226             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 227             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 228             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
 229             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 230
 231             /* LENNARD-JONES DISPERSION/REPULSION */
 232
 233             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 234             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 235             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 236             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 237                                            _fjsp_mul_v2r8(_fjsp_nmsub_v2r8( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 238             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 239
 240             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 241
 242             /* Update potential sum for this i atom from the interaction with this j atom. */
 243             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 244             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 245             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 246             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 247
 248             fscal            = _fjsp_add_v2r8(felec,fvdw);
 249
 250             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 251
 252             /* Update vectorial force */
 253             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 254             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 255             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 256
 257             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 258
 259             }
 260
 261             /* Inner loop uses 67 flops */
 262         }
 263
 264         if(jidx<j_index_end)
 265         {
 266
 267             jnrA             = jjnr[jidx];
 268             j_coord_offsetA  = DIM*jnrA;
 269
 270             /* load j atom coordinates */
 271             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 272                                               &jx0,&jy0,&jz0);
 273
 274             /* Calculate displacement vector */
 275             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 276             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 277             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 278
 279             /* Calculate squared distance and things based on it */
 280             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 281
 282             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 283
 284             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 285
 286             /* Load parameters for j particles */
 287             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 288             vdwjidx0A        = 2*vdwtype[jnrA+0];
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 295             {
 296
 297             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 298
 299             /* Compute parameters for interactions between i and j atoms */
 300             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 301             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 302                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 303
 304             /* EWALD ELECTROSTATICS */
 305
 306             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 307             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 308             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 309             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 310             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 311
 312             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 313             ewtabD           = _fjsp_setzero_v2r8();
 314             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 315             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 316             ewtabFn          = _fjsp_setzero_v2r8();
 317             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 318             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 319             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 320             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
 321             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 322
 323             /* LENNARD-JONES DISPERSION/REPULSION */
 324
 325             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 326             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 327             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 328             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 329                                            _fjsp_mul_v2r8(_fjsp_nmsub_v2r8( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 330             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 331
 332             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 333
 334             /* Update potential sum for this i atom from the interaction with this j atom. */
 335             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 336             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 337             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 338             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 339             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 340             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 341
 342             fscal            = _fjsp_add_v2r8(felec,fvdw);
 343
 344             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 345
 346             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 347
 348             /* Update vectorial force */
 349             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 350             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 351             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 352
 353             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 354
 355             }
 356
 357             /* Inner loop uses 67 flops */
 358         }
 359
 360         /* End of innermost loop */
 361
 362         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 363                                               f+i_coord_offset,fshift+i_shift_offset);
 364
 365         ggid                        = gid[iidx];
 366         /* Update potential energies */
 367         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 368         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 369
 370         /* Increment number of inner iterations */
 371         inneriter                  += j_index_end - j_index_start;
 372
 373         /* Outer loop uses 9 flops */
 374     }
 375
 376     /* Increment number of outer iterations */
 377     outeriter        += nri;
 378
 379     /* Update outer/inner flops */
 380
 381     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*67);
 382 }
 383 /*
 384  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sparc64_hpc_ace_double
 385  * Electrostatics interaction: Ewald
 386  * VdW interaction:            LennardJones
 387  * Geometry:                   Particle-Particle
 388  * Calculate force/pot:        Force
 389  */
 390 void
 391 nb_kernel_ElecEwSh_VdwLJSh_GeomP1P1_F_sparc64_hpc_ace_double
 392                     (t_nblist                    * gmx_restrict       nlist,
 393                      rvec                        * gmx_restrict          xx,
 394                      rvec                        * gmx_restrict          ff,
 395                      struct t_forcerec           * gmx_restrict          fr,
 396                      t_mdatoms                   * gmx_restrict     mdatoms,
 397                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 398                      t_nrnb                      * gmx_restrict        nrnb)
 399 {
 400     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 401      * just 0 for non-waters.
 402      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 403      * jnr indices corresponding to data put in the four positions in the SIMD register.
 404      */
 405     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 406     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 407     int              jnrA,jnrB;
 408     int              j_coord_offsetA,j_coord_offsetB;
 409     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 410     real             rcutoff_scalar;
 411     real             *shiftvec,*fshift,*x,*f;
 412     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 413     int              vdwioffset0;
 414     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 415     int              vdwjidx0A,vdwjidx0B;
 416     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 417     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 418     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 419     real             *charge;
 420     int              nvdwtype;
 421     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 422     int              *vdwtype;
 423     real             *vdwparam;
 424     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 425     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 426     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 427     real             *ewtab;
 428     _fjsp_v2r8       itab_tmp;
 429     _fjsp_v2r8       dummy_mask,cutoff_mask;
 430     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 431     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 432     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 433
 434     x                = xx[0];
 435     f                = ff[0];
 436
 437     nri              = nlist->nri;
 438     iinr             = nlist->iinr;
 439     jindex           = nlist->jindex;
 440     jjnr             = nlist->jjnr;
 441     shiftidx         = nlist->shift;
 442     gid              = nlist->gid;
 443     shiftvec         = fr->shift_vec[0];
 444     fshift           = fr->fshift[0];
 445     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 446     charge           = mdatoms->chargeA;
 447     nvdwtype         = fr->ntype;
 448     vdwparam         = fr->nbfp;
 449     vdwtype          = mdatoms->typeA;
 450
 451     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 452     ewtab            = fr->ic->tabq_coul_F;
 453     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 454     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 455
 456     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 457     rcutoff_scalar   = fr->ic->rcoulomb;
 458     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 459     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 460
 461     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 462     rvdw             = gmx_fjsp_set1_v2r8(fr->ic->rvdw);
 463
 464     /* Avoid stupid compiler warnings */
 465     jnrA = jnrB = 0;
 466     j_coord_offsetA = 0;
 467     j_coord_offsetB = 0;
 468
 469     outeriter        = 0;
 470     inneriter        = 0;
 471
 472     /* Start outer loop over neighborlists */
 473     for(iidx=0; iidx<nri; iidx++)
 474     {
 475         /* Load shift vector for this list */
 476         i_shift_offset   = DIM*shiftidx[iidx];
 477
 478         /* Load limits for loop over neighbors */
 479         j_index_start    = jindex[iidx];
 480         j_index_end      = jindex[iidx+1];
 481
 482         /* Get outer coordinate index */
 483         inr              = iinr[iidx];
 484         i_coord_offset   = DIM*inr;
 485
 486         /* Load i particle coords and add shift vector */
 487         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 488
 489         fix0             = _fjsp_setzero_v2r8();
 490         fiy0             = _fjsp_setzero_v2r8();
 491         fiz0             = _fjsp_setzero_v2r8();
 492
 493         /* Load parameters for i particles */
 494         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 495         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 496
 497         /* Start inner kernel loop */
 498         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 499         {
 500
 501             /* Get j neighbor index, and coordinate index */
 502             jnrA             = jjnr[jidx];
 503             jnrB             = jjnr[jidx+1];
 504             j_coord_offsetA  = DIM*jnrA;
 505             j_coord_offsetB  = DIM*jnrB;
 506
 507             /* load j atom coordinates */
 508             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 509                                               &jx0,&jy0,&jz0);
 510
 511             /* Calculate displacement vector */
 512             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 513             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 514             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 515
 516             /* Calculate squared distance and things based on it */
 517             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 518
 519             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 520
 521             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 522
 523             /* Load parameters for j particles */
 524             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 525             vdwjidx0A        = 2*vdwtype[jnrA+0];
 526             vdwjidx0B        = 2*vdwtype[jnrB+0];
 527
 528             /**************************
 529              * CALCULATE INTERACTIONS *
 530              **************************/
 531
 532             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 533             {
 534
 535             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 539             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 540                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 541
 542             /* EWALD ELECTROSTATICS */
 543
 544             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 545             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 546             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 547             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 548             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 549
 550             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 551                                          &ewtabF,&ewtabFn);
 552             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 553             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 554
 555             /* LENNARD-JONES DISPERSION/REPULSION */
 556
 557             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 558             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 559
 560             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 561
 562             fscal            = _fjsp_add_v2r8(felec,fvdw);
 563
 564             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 565
 566             /* Update vectorial force */
 567             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 568             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 569             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 570
 571             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 572
 573             }
 574
 575             /* Inner loop uses 49 flops */
 576         }
 577
 578         if(jidx<j_index_end)
 579         {
 580
 581             jnrA             = jjnr[jidx];
 582             j_coord_offsetA  = DIM*jnrA;
 583
 584             /* load j atom coordinates */
 585             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 586                                               &jx0,&jy0,&jz0);
 587
 588             /* Calculate displacement vector */
 589             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 590             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 591             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 592
 593             /* Calculate squared distance and things based on it */
 594             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 595
 596             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 597
 598             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 599
 600             /* Load parameters for j particles */
 601             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 602             vdwjidx0A        = 2*vdwtype[jnrA+0];
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 609             {
 610
 611             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 615             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 616                                          vdwparam+vdwioffset0+vdwjidx0B,&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 }