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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "kernelutil_sparc64_hpc_ace_double.h"
  50
  51 /*
  52  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sparc64_hpc_ace_double
  53  * Electrostatics interaction: GeneralizedBorn
  54  * VdW interaction:            None
  55  * Geometry:                   Particle-Particle
  56  * Calculate force/pot:        PotentialAndForce
  57  */
  58 void
  59 nb_kernel_ElecGB_VdwNone_GeomP1P1_VF_sparc64_hpc_ace_double
  60                     (t_nblist                    * gmx_restrict       nlist,
  61                      rvec                        * gmx_restrict          xx,
  62                      rvec                        * gmx_restrict          ff,
  63                      t_forcerec                  * gmx_restrict          fr,
  64                      t_mdatoms                   * gmx_restrict     mdatoms,
  65                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  66                      t_nrnb                      * gmx_restrict        nrnb)
  67 {
  68     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  69      * just 0 for non-waters.
  70      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  71      * jnr indices corresponding to data put in the four positions in the SIMD register.
  72      */
  73     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  74     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  75     int              jnrA,jnrB;
  76     int              j_coord_offsetA,j_coord_offsetB;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  81     int              vdwioffset0;
  82     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  83     int              vdwjidx0A,vdwjidx0B;
  84     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  85     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  86     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  87     real             *charge;
  88     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
  89     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
  90     real             *invsqrta,*dvda,*gbtab;
  91     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  92     real             *vftab;
  93     _fjsp_v2r8       itab_tmp;
  94     _fjsp_v2r8       dummy_mask,cutoff_mask;
  95     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
  96     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
  97     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
  98
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112
 113     invsqrta         = fr->invsqrta;
 114     dvda             = fr->dvda;
 115     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 116     gbtab            = fr->gbtab.data;
 117     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123
 124     outeriter        = 0;
 125     inneriter        = 0;
 126
 127     /* Start outer loop over neighborlists */
 128     for(iidx=0; iidx<nri; iidx++)
 129     {
 130         /* Load shift vector for this list */
 131         i_shift_offset   = DIM*shiftidx[iidx];
 132
 133         /* Load limits for loop over neighbors */
 134         j_index_start    = jindex[iidx];
 135         j_index_end      = jindex[iidx+1];
 136
 137         /* Get outer coordinate index */
 138         inr              = iinr[iidx];
 139         i_coord_offset   = DIM*inr;
 140
 141         /* Load i particle coords and add shift vector */
 142         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 143
 144         fix0             = _fjsp_setzero_v2r8();
 145         fiy0             = _fjsp_setzero_v2r8();
 146         fiz0             = _fjsp_setzero_v2r8();
 147
 148         /* Load parameters for i particles */
 149         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 150         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 151
 152         /* Reset potential sums */
 153         velecsum         = _fjsp_setzero_v2r8();
 154         vgbsum           = _fjsp_setzero_v2r8();
 155         dvdasum          = _fjsp_setzero_v2r8();
 156
 157         /* Start inner kernel loop */
 158         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 159         {
 160
 161             /* Get j neighbor index, and coordinate index */
 162             jnrA             = jjnr[jidx];
 163             jnrB             = jjnr[jidx+1];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166
 167             /* load j atom coordinates */
 168             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 173             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 174             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 175
 176             /* Calculate squared distance and things based on it */
 177             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 178
 179             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 180
 181             /* Load parameters for j particles */
 182             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 183             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 184
 185             /**************************
 186              * CALCULATE INTERACTIONS *
 187              **************************/
 188
 189             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 193
 194             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 195             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 196             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 197             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 198
 199             /* Calculate generalized born table index - this is a separate table from the normal one,
 200              * but we use the same procedure by multiplying r with scale and truncating to integer.
 201              */
 202             rt               = _fjsp_mul_v2r8(r00,gbscale);
 203             itab_tmp         = _fjsp_dtox_v2r8(rt);
 204             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 205             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 206
 207             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 208             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 209             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 210             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 211             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 212             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 213             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 214             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 215             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 216
 217             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 218             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 219             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 220             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 221             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 222             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 223             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 224             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 225
 226             /* Update potential sum for this i atom from the interaction with this j atom. */
 227             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 228             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 229
 230             fscal            = felec;
 231
 232             /* Update vectorial force */
 233             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 234             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 235             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 236
 237             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 238
 239             /* Inner loop uses 61 flops */
 240         }
 241
 242         if(jidx<j_index_end)
 243         {
 244
 245             jnrA             = jjnr[jidx];
 246             j_coord_offsetA  = DIM*jnrA;
 247
 248             /* load j atom coordinates */
 249             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 250                                               &jx0,&jy0,&jz0);
 251
 252             /* Calculate displacement vector */
 253             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 254             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 255             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 256
 257             /* Calculate squared distance and things based on it */
 258             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 259
 260             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 261
 262             /* Load parameters for j particles */
 263             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 264             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 274
 275             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 276             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 277             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 278             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 279
 280             /* Calculate generalized born table index - this is a separate table from the normal one,
 281              * but we use the same procedure by multiplying r with scale and truncating to integer.
 282              */
 283             rt               = _fjsp_mul_v2r8(r00,gbscale);
 284             itab_tmp         = _fjsp_dtox_v2r8(rt);
 285             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 286             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 287
 288             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 289             F                = _fjsp_setzero_v2r8();
 290             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 291             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 292             H                = _fjsp_setzero_v2r8();
 293             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 294             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 295             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 296             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 297
 298             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 299             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 300             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 301             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 302             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 303             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 304             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 305             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 309             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 310             vgb              = _fjsp_unpacklo_v2r8(vgb,_fjsp_setzero_v2r8());
 311             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 312
 313             fscal            = felec;
 314
 315             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 316
 317             /* Update vectorial force */
 318             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 319             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 320             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 321
 322             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 323
 324             /* Inner loop uses 61 flops */
 325         }
 326
 327         /* End of innermost loop */
 328
 329         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 330                                               f+i_coord_offset,fshift+i_shift_offset);
 331
 332         ggid                        = gid[iidx];
 333         /* Update potential energies */
 334         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 335         gmx_fjsp_update_1pot_v2r8(vgbsum,kernel_data->energygrp_polarization+ggid);
 336         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 337         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 338
 339         /* Increment number of inner iterations */
 340         inneriter                  += j_index_end - j_index_start;
 341
 342         /* Outer loop uses 9 flops */
 343     }
 344
 345     /* Increment number of outer iterations */
 346     outeriter        += nri;
 347
 348     /* Update outer/inner flops */
 349
 350     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*9 + inneriter*61);
 351 }
 352 /*
 353  * Gromacs nonbonded kernel:   nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sparc64_hpc_ace_double
 354  * Electrostatics interaction: GeneralizedBorn
 355  * VdW interaction:            None
 356  * Geometry:                   Particle-Particle
 357  * Calculate force/pot:        Force
 358  */
 359 void
 360 nb_kernel_ElecGB_VdwNone_GeomP1P1_F_sparc64_hpc_ace_double
 361                     (t_nblist                    * gmx_restrict       nlist,
 362                      rvec                        * gmx_restrict          xx,
 363                      rvec                        * gmx_restrict          ff,
 364                      t_forcerec                  * gmx_restrict          fr,
 365                      t_mdatoms                   * gmx_restrict     mdatoms,
 366                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 367                      t_nrnb                      * gmx_restrict        nrnb)
 368 {
 369     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 370      * just 0 for non-waters.
 371      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 372      * jnr indices corresponding to data put in the four positions in the SIMD register.
 373      */
 374     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 375     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 376     int              jnrA,jnrB;
 377     int              j_coord_offsetA,j_coord_offsetB;
 378     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 379     real             rcutoff_scalar;
 380     real             *shiftvec,*fshift,*x,*f;
 381     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 382     int              vdwioffset0;
 383     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 384     int              vdwjidx0A,vdwjidx0B;
 385     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 386     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 387     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 388     real             *charge;
 389     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
 390     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
 391     real             *invsqrta,*dvda,*gbtab;
 392     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 393     real             *vftab;
 394     _fjsp_v2r8       itab_tmp;
 395     _fjsp_v2r8       dummy_mask,cutoff_mask;
 396     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 397     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 398     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 399
 400     x                = xx[0];
 401     f                = ff[0];
 402
 403     nri              = nlist->nri;
 404     iinr             = nlist->iinr;
 405     jindex           = nlist->jindex;
 406     jjnr             = nlist->jjnr;
 407     shiftidx         = nlist->shift;
 408     gid              = nlist->gid;
 409     shiftvec         = fr->shift_vec[0];
 410     fshift           = fr->fshift[0];
 411     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 412     charge           = mdatoms->chargeA;
 413
 414     invsqrta         = fr->invsqrta;
 415     dvda             = fr->dvda;
 416     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 417     gbtab            = fr->gbtab.data;
 418     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 419
 420     /* Avoid stupid compiler warnings */
 421     jnrA = jnrB = 0;
 422     j_coord_offsetA = 0;
 423     j_coord_offsetB = 0;
 424
 425     outeriter        = 0;
 426     inneriter        = 0;
 427
 428     /* Start outer loop over neighborlists */
 429     for(iidx=0; iidx<nri; iidx++)
 430     {
 431         /* Load shift vector for this list */
 432         i_shift_offset   = DIM*shiftidx[iidx];
 433
 434         /* Load limits for loop over neighbors */
 435         j_index_start    = jindex[iidx];
 436         j_index_end      = jindex[iidx+1];
 437
 438         /* Get outer coordinate index */
 439         inr              = iinr[iidx];
 440         i_coord_offset   = DIM*inr;
 441
 442         /* Load i particle coords and add shift vector */
 443         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 444
 445         fix0             = _fjsp_setzero_v2r8();
 446         fiy0             = _fjsp_setzero_v2r8();
 447         fiz0             = _fjsp_setzero_v2r8();
 448
 449         /* Load parameters for i particles */
 450         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 451         isai0            = gmx_fjsp_load1_v2r8(invsqrta+inr+0);
 452
 453         dvdasum          = _fjsp_setzero_v2r8();
 454
 455         /* Start inner kernel loop */
 456         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 457         {
 458
 459             /* Get j neighbor index, and coordinate index */
 460             jnrA             = jjnr[jidx];
 461             jnrB             = jjnr[jidx+1];
 462             j_coord_offsetA  = DIM*jnrA;
 463             j_coord_offsetB  = DIM*jnrB;
 464
 465             /* load j atom coordinates */
 466             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 467                                               &jx0,&jy0,&jz0);
 468
 469             /* Calculate displacement vector */
 470             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 471             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 472             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 473
 474             /* Calculate squared distance and things based on it */
 475             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 476
 477             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 478
 479             /* Load parameters for j particles */
 480             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 481             isaj0            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+0,invsqrta+jnrB+0);
 482
 483             /**************************
 484              * CALCULATE INTERACTIONS *
 485              **************************/
 486
 487             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 488
 489             /* Compute parameters for interactions between i and j atoms */
 490             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 491
 492             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 493             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 494             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 495             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 496
 497             /* Calculate generalized born table index - this is a separate table from the normal one,
 498              * but we use the same procedure by multiplying r with scale and truncating to integer.
 499              */
 500             rt               = _fjsp_mul_v2r8(r00,gbscale);
 501             itab_tmp         = _fjsp_dtox_v2r8(rt);
 502             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 503             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 504
 505             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 506             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 507             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 508             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 509             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 510             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 511             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 512             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 513             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 514
 515             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 516             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 517             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 518             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 519             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 520             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 521             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 522             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 523
 524             fscal            = felec;
 525
 526             /* Update vectorial force */
 527             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 528             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 529             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 530
 531             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 532
 533             /* Inner loop uses 59 flops */
 534         }
 535
 536         if(jidx<j_index_end)
 537         {
 538
 539             jnrA             = jjnr[jidx];
 540             j_coord_offsetA  = DIM*jnrA;
 541
 542             /* load j atom coordinates */
 543             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 544                                               &jx0,&jy0,&jz0);
 545
 546             /* Calculate displacement vector */
 547             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 548             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 549             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 550
 551             /* Calculate squared distance and things based on it */
 552             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 553
 554             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 555
 556             /* Load parameters for j particles */
 557             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 558             isaj0            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+0);
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 565
 566             /* Compute parameters for interactions between i and j atoms */
 567             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 568
 569             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 570             isaprod          = _fjsp_mul_v2r8(isai0,isaj0);
 571             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq00,_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 572             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 573
 574             /* Calculate generalized born table index - this is a separate table from the normal one,
 575              * but we use the same procedure by multiplying r with scale and truncating to integer.
 576              */
 577             rt               = _fjsp_mul_v2r8(r00,gbscale);
 578             itab_tmp         = _fjsp_dtox_v2r8(rt);
 579             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 580             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 581
 582             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 583             F                = _fjsp_setzero_v2r8();
 584             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 585             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 586             H                = _fjsp_setzero_v2r8();
 587             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 588             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 589             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 590             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 591
 592             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 593             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 594             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 595             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r00,vgb));
 596             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 597             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj0,isaj0)));
 598             velec            = _fjsp_mul_v2r8(qq00,rinv00);
 599             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv00,fgb),rinv00);
 600
 601             fscal            = felec;
 602
 603             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 604
 605             /* Update vectorial force */
 606             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 607             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 608             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 609
 610             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 611
 612             /* Inner loop uses 59 flops */
 613         }
 614
 615         /* End of innermost loop */
 616
 617         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 618                                               f+i_coord_offset,fshift+i_shift_offset);
 619
 620         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai0,isai0));
 621         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
 622
 623         /* Increment number of inner iterations */
 624         inneriter                  += j_index_end - j_index_start;
 625
 626         /* Outer loop uses 7 flops */
 627     }
 628
 629     /* Increment number of outer iterations */
 630     outeriter        += nri;
 631
 632     /* Update outer/inner flops */
 633
 634     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*59);
 635 }