src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecGB_VdwNone_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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: GeneralizedBorn
  52  * VdW interaction:            None
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sparc64_hpc_ace_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      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     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
  87     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
  88     real             *invsqrta,*dvda,*gbtab;
  89     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  90     real             *vftab;
  91     _fjsp_v2r8       itab_tmp;
  92     _fjsp_v2r8       dummy_mask,cutoff_mask;
  93     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  94     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  95     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  96
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110
 111     invsqrta         = fr->invsqrta;
 112     dvda             = fr->dvda;
 113     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 114     gbtab            = fr->gbtab.data;
 115     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 116
 117     /* Avoid stupid compiler warnings */
 118     jnrA = jnrB = 0;
 119     j_coord_offsetA = 0;
 120     j_coord_offsetB = 0;
 121
 122     outeriter        = 0;
 123     inneriter        = 0;
 124
 125     /* Start outer loop over neighborlists */
 126     for(iidx=0; iidx<nri; iidx++)
 127     {
 128         /* Load shift vector for this list */
 129         i_shift_offset   = DIM*shiftidx[iidx];
 130
 131         /* Load limits for loop over neighbors */
 132         j_index_start    = jindex[iidx];
 133         j_index_end      = jindex[iidx+1];
 134
 135         /* Get outer coordinate index */
 136         inr              = iinr[iidx];
 137         i_coord_offset   = DIM*inr;
 138
 139         /* Load i particle coords and add shift vector */
 140         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 141
 142         fix0             = _fjsp_setzero_v2r8();
 143         fiy0             = _fjsp_setzero_v2r8();
 144         fiz0             = _fjsp_setzero_v2r8();
 145
 146         /* Load parameters for i particles */
 147         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 148         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 149
 150         /* Reset potential sums */
 151         velecsum         = _fjsp_setzero_v2r8();
 152         vgbsum           = _fjsp_setzero_v2r8();
 153         dvdasum          = _fjsp_setzero_v2r8();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             j_coord_offsetA  = DIM*jnrA;
 163             j_coord_offsetB  = DIM*jnrB;
 164
 165             /* load j atom coordinates */
 166             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 167                                               &jx0,&jy0,&jz0);
 168
 169             /* Calculate displacement vector */
 170             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 171             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 172             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 173
 174             /* Calculate squared distance and things based on it */
 175             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 176
 177             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 178
 179             /* Load parameters for j particles */
 180             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 181             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 182
 183             /**************************
 184              * CALCULATE INTERACTIONS *
 185              **************************/
 186
 187             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 188
 189             /* Compute parameters for interactions between i and j atoms */
 190             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 191
 192             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 193             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 194             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 195             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 196
 197             /* Calculate generalized born table index - this is a separate table from the normal one,
 198              * but we use the same procedure by multiplying r with scale and truncating to integer.
 199              */
 200             rt               = _fjsp_mul_v2r8(r00,gbscale);
 201             itab_tmp         = _fjsp_dtox_v2r8(rt);
 202             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 203             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 204
 205             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 206             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 207             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 208             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 209             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 210             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 211             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 212             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 213             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 214
 215             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 216             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 217             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 218             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 219             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 220             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 221             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 222             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 223
 224             /* Update potential sum for this i atom from the interaction with this j atom. */
 225             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 226             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 227
 228             fscal            = felec;
 229
 230             /* Update vectorial force */
 231             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 232             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 233             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 234
 235             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 236
 237             /* Inner loop uses 61 flops */
 238         }
 239
 240         if(jidx<j_index_end)
 241         {
 242
 243             jnrA             = jjnr[jidx];
 244             j_coord_offsetA  = DIM*jnrA;
 245
 246             /* load j atom coordinates */
 247             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 248                                               &jx0,&jy0,&jz0);
 249
 250             /* Calculate displacement vector */
 251             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 252             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 253             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 254
 255             /* Calculate squared distance and things based on it */
 256             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 257
 258             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 259
 260             /* Load parameters for j particles */
 261             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 262             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 263
 264             /**************************
 265              * CALCULATE INTERACTIONS *
 266              **************************/
 267
 268             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 272
 273             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 274             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 275             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 276             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 277
 278             /* Calculate generalized born table index - this is a separate table from the normal one,
 279              * but we use the same procedure by multiplying r with scale and truncating to integer.
 280              */
 281             rt               = _fjsp_mul_v2r8(r00,gbscale);
 282             itab_tmp         = _fjsp_dtox_v2r8(rt);
 283             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 284             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 285
 286             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 287             F                = _fjsp_setzero_v2r8();
 288             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 289             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 290             H                = _fjsp_setzero_v2r8();
 291             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 292             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 293             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 294             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 295
 296             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 297             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 298             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 299             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 300             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 301             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 302             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 303             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 307             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 308             vgb              = _fjsp_unpacklo_v2r8(vgb,_fjsp_setzero_v2r8());
 309             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 310
 311             fscal            = felec;
 312
 313             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 314
 315             /* Update vectorial force */
 316             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 317             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 318             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 319
 320             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 321
 322             /* Inner loop uses 61 flops */
 323         }
 324
 325         /* End of innermost loop */
 326
 327         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 328                                               f+i_coord_offset,fshift+i_shift_offset);
 329
 330         ggid                        = gid[iidx];
 331         /* Update potential energies */
 332         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 333         gmx_fjsp_update_1pot_v2r8(vgbsum,kernel_data->energygrp_polarization+ggid);
 334         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 335         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 336
 337         /* Increment number of inner iterations */
 338         inneriter                  += j_index_end - j_index_start;
 339
 340         /* Outer loop uses 9 flops */
 341     }
 342
 343     /* Increment number of outer iterations */
 344     outeriter        += nri;
 345
 346     /* Update outer/inner flops */
 347
 348     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*61);
 349 }
 350 /*
 351  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sparc64_hpc_ace_double
 352  * Electrostatics interaction: GeneralizedBorn
 353  * VdW interaction:            None
 354  * Geometry:                   Particle-Particle
 355  * Calculate force/pot:        Force
 356  */
 357 void
 358 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sparc64_hpc_ace_double
 359                     (t_nblist                    * gmx_restrict       nlist,
 360                      rvec                        * gmx_restrict          xx,
 361                      rvec                        * gmx_restrict          ff,
 362                      t_forcerec                  * gmx_restrict          fr,
 363                      t_mdatoms                   * gmx_restrict     mdatoms,
 364                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 365                      t_nrnb                      * gmx_restrict        nrnb)
 366 {
 367     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 368      * just 0 for non-waters.
 369      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 370      * jnr indices corresponding to data put in the four positions in the SIMD register.
 371      */
 372     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 373     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 374     int              jnrA,jnrB;
 375     int              j_coord_offsetA,j_coord_offsetB;
 376     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 377     real             rcutoff_scalar;
 378     real             *shiftvec,*fshift,*x,*f;
 379     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 380     int              vdwioffset0;
 381     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 382     int              vdwjidx0A,vdwjidx0B;
 383     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 384     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 385     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 386     real             *charge;
 387     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
 388     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
 389     real             *invsqrta,*dvda,*gbtab;
 390     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 391     real             *vftab;
 392     _fjsp_v2r8       itab_tmp;
 393     _fjsp_v2r8       dummy_mask,cutoff_mask;
 394     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 395     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 396     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 397
 398     x                = xx[0];
 399     f                = ff[0];
 400
 401     nri              = nlist->nri;
 402     iinr             = nlist->iinr;
 403     jindex           = nlist->jindex;
 404     jjnr             = nlist->jjnr;
 405     shiftidx         = nlist->shift;
 406     gid              = nlist->gid;
 407     shiftvec         = fr->shift_vec[0];
 408     fshift           = fr->fshift[0];
 409     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 410     charge           = mdatoms->chargeA;
 411
 412     invsqrta         = fr->invsqrta;
 413     dvda             = fr->dvda;
 414     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 415     gbtab            = fr->gbtab.data;
 416     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 417
 418     /* Avoid stupid compiler warnings */
 419     jnrA = jnrB = 0;
 420     j_coord_offsetA = 0;
 421     j_coord_offsetB = 0;
 422
 423     outeriter        = 0;
 424     inneriter        = 0;
 425
 426     /* Start outer loop over neighborlists */
 427     for(iidx=0; iidx<nri; iidx++)
 428     {
 429         /* Load shift vector for this list */
 430         i_shift_offset   = DIM*shiftidx[iidx];
 431
 432         /* Load limits for loop over neighbors */
 433         j_index_start    = jindex[iidx];
 434         j_index_end      = jindex[iidx+1];
 435
 436         /* Get outer coordinate index */
 437         inr              = iinr[iidx];
 438         i_coord_offset   = DIM*inr;
 439
 440         /* Load i particle coords and add shift vector */
 441         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 442
 443         fix0             = _fjsp_setzero_v2r8();
 444         fiy0             = _fjsp_setzero_v2r8();
 445         fiz0             = _fjsp_setzero_v2r8();
 446
 447         /* Load parameters for i particles */
 448         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 449         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 450
 451         dvdasum          = _fjsp_setzero_v2r8();
 452
 453         /* Start inner kernel loop */
 454         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 455         {
 456
 457             /* Get j neighbor index, and coordinate index */
 458             jnrA             = jjnr[jidx];
 459             jnrB             = jjnr[jidx+1];
 460             j_coord_offsetA  = DIM*jnrA;
 461             j_coord_offsetB  = DIM*jnrB;
 462
 463             /* load j atom coordinates */
 464             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 465                                               &jx0,&jy0,&jz0);
 466
 467             /* Calculate displacement vector */
 468             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 469             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 470             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 471
 472             /* Calculate squared distance and things based on it */
 473             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 474
 475             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 476
 477             /* Load parameters for j particles */
 478             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 479             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 489
 490             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 491             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 492             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 493             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 494
 495             /* Calculate generalized born table index - this is a separate table from the normal one,
 496              * but we use the same procedure by multiplying r with scale and truncating to integer.
 497              */
 498             rt               = _fjsp_mul_v2r8(r00,gbscale);
 499             itab_tmp         = _fjsp_dtox_v2r8(rt);
 500             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 501             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 502
 503             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 504             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 505             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 506             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 507             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 508             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 509             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 510             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 511             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 512
 513             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 514             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 515             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 516             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 517             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 518             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 519             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 520             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 521
 522             fscal            = felec;
 523
 524             /* Update vectorial force */
 525             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 526             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 527             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 528
 529             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 530
 531             /* Inner loop uses 59 flops */
 532         }
 533
 534         if(jidx<j_index_end)
 535         {
 536
 537             jnrA             = jjnr[jidx];
 538             j_coord_offsetA  = DIM*jnrA;
 539
 540             /* load j atom coordinates */
 541             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 542                                               &jx0,&jy0,&jz0);
 543
 544             /* Calculate displacement vector */
 545             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 546             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 547             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 548
 549             /* Calculate squared distance and things based on it */
 550             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 551
 552             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 553
 554             /* Load parameters for j particles */
 555             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 556             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 557
 558             /**************************
 559              * CALCULATE INTERACTIONS *
 560              **************************/
 561
 562             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 563
 564             /* Compute parameters for interactions between i and j atoms */
 565             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 566
 567             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 568             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 569             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 570             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 571
 572             /* Calculate generalized born table index - this is a separate table from the normal one,
 573              * but we use the same procedure by multiplying r with scale and truncating to integer.
 574              */
 575             rt               = _fjsp_mul_v2r8(r00,gbscale);
 576             itab_tmp         = _fjsp_dtox_v2r8(rt);
 577             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 578             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 579
 580             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 581             F                = _fjsp_setzero_v2r8();
 582             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 583             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 584             H                = _fjsp_setzero_v2r8();
 585             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 586             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 587             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 588             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 589
 590             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 591             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 592             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 593             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 594             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 595             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 596             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 597             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 598
 599             fscal            = felec;
 600
 601             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 602
 603             /* Update vectorial force */
 604             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 605             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 606             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 607
 608             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 609
 610             /* Inner loop uses 59 flops */
 611         }
 612
 613         /* End of innermost loop */
 614
 615         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 616                                               f+i_coord_offset,fshift+i_shift_offset);
 617
 618         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 619         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 7 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*59);
 633 }