src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEw_VdwLJ_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_ElecEw_VdwLJ_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_ElecEw_VdwLJ_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     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135
 136         /* Load limits for loop over neighbors */
 137         j_index_start    = jindex[iidx];
 138         j_index_end      = jindex[iidx+1];
 139
 140         /* Get outer coordinate index */
 141         inr              = iinr[iidx];
 142         i_coord_offset   = DIM*inr;
 143
 144         /* Load i particle coords and add shift vector */
 145         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 146
 147         fix0             = _fjsp_setzero_v2r8();
 148         fiy0             = _fjsp_setzero_v2r8();
 149         fiz0             = _fjsp_setzero_v2r8();
 150
 151         /* Load parameters for i particles */
 152         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 153         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155         /* Reset potential sums */
 156         velecsum         = _fjsp_setzero_v2r8();
 157         vvdwsum          = _fjsp_setzero_v2r8();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             j_coord_offsetA  = DIM*jnrA;
 167             j_coord_offsetB  = DIM*jnrB;
 168
 169             /* load j atom coordinates */
 170             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               &jx0,&jy0,&jz0);
 172
 173             /* Calculate displacement vector */
 174             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 175             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 176             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 180
 181             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 182
 183             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 184
 185             /* Load parameters for j particles */
 186             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 187             vdwjidx0A        = 2*vdwtype[jnrA+0];
 188             vdwjidx0B        = 2*vdwtype[jnrB+0];
 189
 190             /**************************
 191              * CALCULATE INTERACTIONS *
 192              **************************/
 193
 194             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 195
 196             /* Compute parameters for interactions between i and j atoms */
 197             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 198             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 199                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 200
 201             /* EWALD ELECTROSTATICS */
 202
 203             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 204             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 205             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 206             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 207             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 208
 209             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 210             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 211             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 212             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 213             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 214             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 215             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 216             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 217             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 218             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 219
 220             /* LENNARD-JONES DISPERSION/REPULSION */
 221
 222             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 223             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 224             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 225             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 226             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 227
 228             /* Update potential sum for this i atom from the interaction with this j atom. */
 229             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 230             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 231
 232             fscal            = _fjsp_add_v2r8(felec,fvdw);
 233
 234             /* Update vectorial force */
 235             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 236             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 237             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 238
 239             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 240
 241             /* Inner loop uses 56 flops */
 242         }
 243
 244         if(jidx<j_index_end)
 245         {
 246
 247             jnrA             = jjnr[jidx];
 248             j_coord_offsetA  = DIM*jnrA;
 249
 250             /* load j atom coordinates */
 251             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 252                                               &jx0,&jy0,&jz0);
 253
 254             /* Calculate displacement vector */
 255             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 256             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 257             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 258
 259             /* Calculate squared distance and things based on it */
 260             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 261
 262             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 263
 264             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 265
 266             /* Load parameters for j particles */
 267             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 268             vdwjidx0A        = 2*vdwtype[jnrA+0];
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 278             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 279                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 280
 281             /* EWALD ELECTROSTATICS */
 282
 283             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 284             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 285             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 286             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 287             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 288
 289             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 290             ewtabD           = _fjsp_setzero_v2r8();
 291             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 292             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 293             ewtabFn          = _fjsp_setzero_v2r8();
 294             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 295             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 296             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 297             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(rinv00,velec));
 298             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 299
 300             /* LENNARD-JONES DISPERSION/REPULSION */
 301
 302             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 303             vvdw6            = _fjsp_mul_v2r8(c6_00,rinvsix);
 304             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 305             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 306             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq00);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 310             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 311             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 312             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 313
 314             fscal            = _fjsp_add_v2r8(felec,fvdw);
 315
 316             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 317
 318             /* Update vectorial force */
 319             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 320             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 321             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 322
 323             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 324
 325             /* Inner loop uses 56 flops */
 326         }
 327
 328         /* End of innermost loop */
 329
 330         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 331                                               f+i_coord_offset,fshift+i_shift_offset);
 332
 333         ggid                        = gid[iidx];
 334         /* Update potential energies */
 335         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 336         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 337
 338         /* Increment number of inner iterations */
 339         inneriter                  += j_index_end - j_index_start;
 340
 341         /* Outer loop uses 9 flops */
 342     }
 343
 344     /* Increment number of outer iterations */
 345     outeriter        += nri;
 346
 347     /* Update outer/inner flops */
 348
 349     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*56);
 350 }
 351 /*
 352  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 353  * Electrostatics interaction: Ewald
 354  * VdW interaction:            LennardJones
 355  * Geometry:                   Particle-Particle
 356  * Calculate force/pot:        Force
 357  */
 358 void
 359 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sparc64_hpc_ace_double
 360                     (t_nblist                    * gmx_restrict       nlist,
 361                      rvec                        * gmx_restrict          xx,
 362                      rvec                        * gmx_restrict          ff,
 363                      struct t_forcerec           * gmx_restrict          fr,
 364                      t_mdatoms                   * gmx_restrict     mdatoms,
 365                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 366                      t_nrnb                      * gmx_restrict        nrnb)
 367 {
 368     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 369      * just 0 for non-waters.
 370      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 371      * jnr indices corresponding to data put in the four positions in the SIMD register.
 372      */
 373     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 374     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 375     int              jnrA,jnrB;
 376     int              j_coord_offsetA,j_coord_offsetB;
 377     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 378     real             rcutoff_scalar;
 379     real             *shiftvec,*fshift,*x,*f;
 380     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 381     int              vdwioffset0;
 382     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 383     int              vdwjidx0A,vdwjidx0B;
 384     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 385     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 386     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 387     real             *charge;
 388     int              nvdwtype;
 389     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 390     int              *vdwtype;
 391     real             *vdwparam;
 392     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 393     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 394     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 395     real             *ewtab;
 396     _fjsp_v2r8       itab_tmp;
 397     _fjsp_v2r8       dummy_mask,cutoff_mask;
 398     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 399     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 400     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 401
 402     x                = xx[0];
 403     f                = ff[0];
 404
 405     nri              = nlist->nri;
 406     iinr             = nlist->iinr;
 407     jindex           = nlist->jindex;
 408     jjnr             = nlist->jjnr;
 409     shiftidx         = nlist->shift;
 410     gid              = nlist->gid;
 411     shiftvec         = fr->shift_vec[0];
 412     fshift           = fr->fshift[0];
 413     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 414     charge           = mdatoms->chargeA;
 415     nvdwtype         = fr->ntype;
 416     vdwparam         = fr->nbfp;
 417     vdwtype          = mdatoms->typeA;
 418
 419     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 420     ewtab            = fr->ic->tabq_coul_F;
 421     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 422     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 423
 424     /* Avoid stupid compiler warnings */
 425     jnrA = jnrB = 0;
 426     j_coord_offsetA = 0;
 427     j_coord_offsetB = 0;
 428
 429     outeriter        = 0;
 430     inneriter        = 0;
 431
 432     /* Start outer loop over neighborlists */
 433     for(iidx=0; iidx<nri; iidx++)
 434     {
 435         /* Load shift vector for this list */
 436         i_shift_offset   = DIM*shiftidx[iidx];
 437
 438         /* Load limits for loop over neighbors */
 439         j_index_start    = jindex[iidx];
 440         j_index_end      = jindex[iidx+1];
 441
 442         /* Get outer coordinate index */
 443         inr              = iinr[iidx];
 444         i_coord_offset   = DIM*inr;
 445
 446         /* Load i particle coords and add shift vector */
 447         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 448
 449         fix0             = _fjsp_setzero_v2r8();
 450         fiy0             = _fjsp_setzero_v2r8();
 451         fiz0             = _fjsp_setzero_v2r8();
 452
 453         /* Load parameters for i particles */
 454         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 455         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 456
 457         /* Start inner kernel loop */
 458         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 459         {
 460
 461             /* Get j neighbor index, and coordinate index */
 462             jnrA             = jjnr[jidx];
 463             jnrB             = jjnr[jidx+1];
 464             j_coord_offsetA  = DIM*jnrA;
 465             j_coord_offsetB  = DIM*jnrB;
 466
 467             /* load j atom coordinates */
 468             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 469                                               &jx0,&jy0,&jz0);
 470
 471             /* Calculate displacement vector */
 472             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 473             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 474             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 475
 476             /* Calculate squared distance and things based on it */
 477             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 478
 479             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 480
 481             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 482
 483             /* Load parameters for j particles */
 484             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 485             vdwjidx0A        = 2*vdwtype[jnrA+0];
 486             vdwjidx0B        = 2*vdwtype[jnrB+0];
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 493
 494             /* Compute parameters for interactions between i and j atoms */
 495             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 496             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 497                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 498
 499             /* EWALD ELECTROSTATICS */
 500
 501             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 502             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 503             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 504             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 505             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 506
 507             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 508                                          &ewtabF,&ewtabFn);
 509             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 510             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 511
 512             /* LENNARD-JONES DISPERSION/REPULSION */
 513
 514             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 515             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 516
 517             fscal            = _fjsp_add_v2r8(felec,fvdw);
 518
 519             /* Update vectorial force */
 520             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 521             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 522             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 523
 524             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 525
 526             /* Inner loop uses 46 flops */
 527         }
 528
 529         if(jidx<j_index_end)
 530         {
 531
 532             jnrA             = jjnr[jidx];
 533             j_coord_offsetA  = DIM*jnrA;
 534
 535             /* load j atom coordinates */
 536             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 537                                               &jx0,&jy0,&jz0);
 538
 539             /* Calculate displacement vector */
 540             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 541             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 542             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 543
 544             /* Calculate squared distance and things based on it */
 545             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 546
 547             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 548
 549             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 550
 551             /* Load parameters for j particles */
 552             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 553             vdwjidx0A        = 2*vdwtype[jnrA+0];
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 560
 561             /* Compute parameters for interactions between i and j atoms */
 562             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 563             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 564                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 565
 566             /* EWALD ELECTROSTATICS */
 567
 568             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 569             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 570             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 571             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 572             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 573
 574             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 575             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 576             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 577
 578             /* LENNARD-JONES DISPERSION/REPULSION */
 579
 580             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 581             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,c6_00),_fjsp_mul_v2r8(rinvsix,rinvsq00));
 582
 583             fscal            = _fjsp_add_v2r8(felec,fvdw);
 584
 585             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 586
 587             /* Update vectorial force */
 588             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 589             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 590             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 591
 592             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 593
 594             /* Inner loop uses 46 flops */
 595         }
 596
 597         /* End of innermost loop */
 598
 599         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 600                                               f+i_coord_offset,fshift+i_shift_offset);
 601
 602         /* Increment number of inner iterations */
 603         inneriter                  += j_index_end - j_index_start;
 604
 605         /* Outer loop uses 7 flops */
 606     }
 607
 608     /* Increment number of outer iterations */
 609     outeriter        += nri;
 610
 611     /* Update outer/inner flops */
 612
 613     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*46);
 614 }