src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_template_sparc64_hpc_ace_double.pre

   1 /* #if 0 */
   2 /*
   3  * This file is part of the GROMACS molecular simulation package.
   4  *
   5  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   6  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   7  * and including many others, as listed in the AUTHORS file in the
   8  * top-level source directory and at http://www.gromacs.org.
   9  *
  10  * GROMACS is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public License
  12  * as published by the Free Software Foundation; either version 2.1
  13  * of the License, or (at your option) any later version.
  14  *
  15  * GROMACS is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with GROMACS; if not, see
  22  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  23  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  24  *
  25  * If you want to redistribute modifications to GROMACS, please
  26  * consider that scientific software is very special. Version
  27  * control is crucial - bugs must be traceable. We will be happy to
  28  * consider code for inclusion in the official distribution, but
  29  * derived work must not be called official GROMACS. Details are found
  30  * in the README & COPYING files - if they are missing, get the
  31  * official version at http://www.gromacs.org.
  32  *
  33  * To help us fund GROMACS development, we humbly ask that you cite
  34  * the research papers on the package. Check out http://www.gromacs.org.
  35  */
  36 #error This file must be processed with the Gromacs pre-preprocessor
  37 /* #endif */
  38 /* #if INCLUDE_HEADER */
  39 #include "gmxpre.h"
  40
  41 #include "config.h"
  42
  43 #include <math.h>
  44
  45 #include "../nb_kernel.h"
  46 #include "gromacs/gmxlib/nrnb.h"
  47
  48 #include "kernelutil_sparc64_hpc_ace_double.h"
  49 /* #endif */
  50
  51 /* ## List of variables set by the generating script:                                    */
  52 /* ##                                                                                    */
  53 /* ## Setttings that apply to the entire kernel:                                         */
  54 /* ## KERNEL_ELEC:           String, choice for electrostatic interactions               */
  55 /* ## KERNEL_VDW:            String, choice for van der Waals interactions               */
  56 /* ## KERNEL_NAME:           String, name of this kernel                                 */
  57 /* ## KERNEL_VF:             String telling if we calculate potential, force, or both    */
  58 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
  59 /* ##                                                                                    */
  60 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops:             */
  61 /* ## PARTICLES_I[]/         Arrays with lists of i/j particles to use in kernel. It is  */
  62 /* ## PARTICLES_J[]:         just [0] for particle geometry, but can be longer for water */
  63 /* ## PARTICLES_ELEC_I[]/    Arrays with lists of i/j particle that have electrostatics  */
  64 /* ## PARTICLES_ELEC_J[]:    interactions that should be calculated in this kernel.      */
  65 /* ## PARTICLES_VDW_I[]/     Arrays with the list of i/j particle that have VdW          */
  66 /* ## PARTICLES_VDW_J[]:     interactions that should be calculated in this kernel.      */
  67 /* ##                                                                                    */
  68 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle)   */
  69 /* ## PAIRS_IJ[]:            Array with (i,j) tuples of pairs for which interactions     */
  70 /* ##                        should be calculated in this kernel. Zero-charge particles  */
  71 /* ##                        do not have interactions with particles without vdw, and    */
  72 /* ##                        Vdw-only interactions are not evaluated in a no-vdw-kernel. */
  73 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
  74 /* ##                        For each i-j pair, the element [I][J] is a list of strings  */
  75 /* ##                        defining properties/flags of this interaction. Examples     */
  76 /* ##                        include 'electrostatics'/'vdw' if that type of interaction  */
  77 /* ##                        should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values  */
  78 /* ##                        are needed, and 'exactcutoff' or 'shift','switch' to        */
  79 /* ##                        decide if the force/potential should be modified. This way  */
  80 /* ##                        we only calculate values absolutely needed for each case.   */
  81
  82 /* ## Calculate the size and offset for (merged/interleaved) table data */
  83
  84 /*
  85  * Gromacs nonbonded kernel:   {KERNEL_NAME}
  86  * Electrostatics interaction: {KERNEL_ELEC}
  87  * VdW interaction:            {KERNEL_VDW}
  88  * Geometry:                   {GEOMETRY_I}-{GEOMETRY_J}
  89  * Calculate force/pot:        {KERNEL_VF}
  90  */
  91 void
  92 {KERNEL_NAME}
  93                     (t_nblist                    * gmx_restrict       nlist,
  94                      rvec                        * gmx_restrict          xx,
  95                      rvec                        * gmx_restrict          ff,
  96                      struct t_forcerec           * gmx_restrict          fr,
  97                      t_mdatoms                   * gmx_restrict     mdatoms,
  98                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  99                      t_nrnb                      * gmx_restrict        nrnb)
 100 {
 101     /* ## Not all variables are used for all kernels, but any optimizing compiler fixes that, */
 102     /* ## so there is no point in going to extremes to exclude variables that are not needed. */
 103     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 104      * just 0 for non-waters.
 105      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 106      * jnr indices corresponding to data put in the four positions in the SIMD register.
 107      */
 108     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 109     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 110     int              jnrA,jnrB;
 111     int              j_coord_offsetA,j_coord_offsetB;
 112     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 113     real             rcutoff_scalar;
 114     real             *shiftvec,*fshift,*x,*f;
 115     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 116     /* #for I in PARTICLES_I */
 117     int              vdwioffset{I};
 118     _fjsp_v2r8       ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 119     /* #endfor */
 120     /* #for J in PARTICLES_J */
 121     int              vdwjidx{J}A,vdwjidx{J}B;
 122     _fjsp_v2r8       jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 123     /* #endfor */
 124     /* #for I,J in PAIRS_IJ */
 125     _fjsp_v2r8       dx{I}{J},dy{I}{J},dz{I}{J},rsq{I}{J},rinv{I}{J},rinvsq{I}{J},r{I}{J},qq{I}{J},c6_{I}{J},c12_{I}{J};
 126     /* #endfor */
 127     /* #if KERNEL_ELEC != 'None' */
 128     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 129     real             *charge;
 130     /* #endif */
 131     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 132     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
 133     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
 134     real             *invsqrta,*dvda,*gbtab;
 135     /* #endif */
 136     /* #if KERNEL_VDW != 'None' */
 137     int              nvdwtype;
 138     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 139     int              *vdwtype;
 140     real             *vdwparam;
 141     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 142     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 143     /* #endif */
 144     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 145     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 146     real             *vftab;
 147     /* #endif */
 148     /* #if 'LJEwald' in KERNEL_VDW */
 149     /* #for I,J in PAIRS_IJ */
 150     _fjsp_v2r8           c6grid_{I}{J};
 151     /* #endfor */
 152     real                 *vdwgridparam;
 153     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 154     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 155     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 156     /* #endif */
 157     /* #if 'Ewald' in KERNEL_ELEC */
 158     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 159     real             *ewtab;
 160     /* #endif */
 161     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 162     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 163     real             rswitch_scalar,d_scalar;
 164     /* #endif */
 165     _fjsp_v2r8       itab_tmp;
 166     _fjsp_v2r8       dummy_mask,cutoff_mask;
 167     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 168     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 169     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 170
 171     x                = xx[0];
 172     f                = ff[0];
 173
 174     nri              = nlist->nri;
 175     iinr             = nlist->iinr;
 176     jindex           = nlist->jindex;
 177     jjnr             = nlist->jjnr;
 178     shiftidx         = nlist->shift;
 179     gid              = nlist->gid;
 180     shiftvec         = fr->shift_vec[0];
 181     fshift           = fr->fshift[0];
 182     /* #if KERNEL_ELEC != 'None' */
 183     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 184     charge           = mdatoms->chargeA;
 185     /*     #if 'ReactionField' in KERNEL_ELEC */
 186     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 187     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 188     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 189     /*     #endif */
 190     /* #endif */
 191     /* #if KERNEL_VDW != 'None' */
 192     nvdwtype         = fr->ntype;
 193     vdwparam         = fr->nbfp;
 194     vdwtype          = mdatoms->typeA;
 195     /* #endif */
 196     /* #if 'LJEwald' in KERNEL_VDW */
 197     vdwgridparam     = fr->ljpme_c6grid;
 198     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 199     ewclj            = gmx_fjsp_set1_v2r8(fr->ic->ewaldcoeff_lj);
 200     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 201     /* #endif */
 202
 203     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 204     vftab            = kernel_data->table_elec_vdw->data;
 205     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec_vdw->scale);
 206     /* #elif 'Table' in KERNEL_ELEC */
 207     vftab            = kernel_data->table_elec->data;
 208     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec->scale);
 209     /* #elif 'Table' in KERNEL_VDW */
 210     vftab            = kernel_data->table_vdw->data;
 211     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 212     /* #endif */
 213
 214     /* #if 'Ewald' in KERNEL_ELEC */
 215     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 216     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 217     ewtab            = fr->ic->tabq_coul_F;
 218     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 219     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 220     /*     #else */
 221     ewtab            = fr->ic->tabq_coul_FDV0;
 222     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 223     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 224      /*     #endif */
 225     /* #endif */
 226
 227     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 228     invsqrta         = fr->invsqrta;
 229     dvda             = fr->dvda;
 230     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab->scale);
 231     gbtab            = fr->gbtab->data;
 232     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 233     /* #endif */
 234
 235     /* #if 'Water' in GEOMETRY_I */
 236     /* Setup water-specific parameters */
 237     inr              = nlist->iinr[0];
 238     /*     #for I in PARTICLES_ELEC_I */
 239     iq{I}              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+{I}]));
 240     /*     #endfor */
 241     /*     #for I in PARTICLES_VDW_I */
 242     vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 243     /*     #endfor */
 244     /* #endif */
 245
 246     /* #if 'Water' in GEOMETRY_J */
 247     /*     #for J in PARTICLES_ELEC_J */
 248     jq{J}              = gmx_fjsp_set1_v2r8(charge[inr+{J}]);
 249     /*     #endfor */
 250     /*     #for J in PARTICLES_VDW_J */
 251     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 252     /*     #endfor */
 253     /*     #for I,J in PAIRS_IJ */
 254     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 255     qq{I}{J}             = _fjsp_mul_v2r8(iq{I},jq{J});
 256     /*         #endif */
 257     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 258     /*             #if 'LJEwald' in KERNEL_VDW */
 259     c6_{I}{J}            = gmx_fjsp_set1_v2r8(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 260     c12_{I}{J}           = gmx_fjsp_set1_v2r8(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 261     c6grid_{I}{J}        = gmx_fjsp_set1_v2r8(vdwgridparam[vdwioffset{I}+vdwjidx{J}A]);
 262     /*             #else */
 263     c6_{I}{J}            = gmx_fjsp_set1_v2r8(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 264     c12_{I}{J}           = gmx_fjsp_set1_v2r8(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 265     /*             #endif */
 266     /*         #endif */
 267     /*     #endfor */
 268     /* #endif */
 269
 270     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 271     /*     #if KERNEL_ELEC!='None' */
 272     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 273     rcutoff_scalar   = fr->ic->rcoulomb;
 274     /*     #else */
 275     rcutoff_scalar   = fr->ic->rvdw;
 276     /*     #endif */
 277     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 278     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 279     /* #endif */
 280
 281     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 282     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 283     rvdw             = gmx_fjsp_set1_v2r8(fr->ic->rvdw);
 284     /* #endif */
 285
 286     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 287     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 288     rswitch_scalar   = fr->ic->rcoulomb_switch;
 289     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 290     /*     #else */
 291     rswitch_scalar   = fr->ic->rvdw_switch;
 292     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 293     /*     #endif */
 294     /* Setup switch parameters */
 295     d_scalar         = rcutoff_scalar-rswitch_scalar;
 296     d                = gmx_fjsp_set1_v2r8(d_scalar);
 297     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 298     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 299     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 300     /*     #if 'Force' in KERNEL_VF */
 301     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 302     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 303     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 304     /*     #endif */
 305     /* #endif */
 306
 307     /* Avoid stupid compiler warnings */
 308     jnrA = jnrB = 0;
 309     j_coord_offsetA = 0;
 310     j_coord_offsetB = 0;
 311
 312     /* ## Keep track of the floating point operations we issue for reporting! */
 313     /* #define OUTERFLOPS 0 */
 314     outeriter        = 0;
 315     inneriter        = 0;
 316
 317     /* Start outer loop over neighborlists */
 318     for(iidx=0; iidx<nri; iidx++)
 319     {
 320         /* Load shift vector for this list */
 321         i_shift_offset   = DIM*shiftidx[iidx];
 322
 323         /* Load limits for loop over neighbors */
 324         j_index_start    = jindex[iidx];
 325         j_index_end      = jindex[iidx+1];
 326
 327         /* Get outer coordinate index */
 328         inr              = iinr[iidx];
 329         i_coord_offset   = DIM*inr;
 330
 331         /* Load i particle coords and add shift vector */
 332         /* #if GEOMETRY_I == 'Particle' */
 333         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 334         /* #elif GEOMETRY_I == 'Water3' */
 335         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 336                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 337         /* #elif GEOMETRY_I == 'Water4' */
 338         /*     #if 0 in PARTICLES_I                 */
 339         gmx_fjsp_load_shift_and_4rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 340                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 341         /*     #else                                */
 342         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 343                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 344         /*     #endif                               */
 345         /* #endif                                   */
 346
 347         /* #if 'Force' in KERNEL_VF */
 348         /*     #for I in PARTICLES_I */
 349         fix{I}             = _fjsp_setzero_v2r8();
 350         fiy{I}             = _fjsp_setzero_v2r8();
 351         fiz{I}             = _fjsp_setzero_v2r8();
 352         /*     #endfor */
 353         /* #endif */
 354
 355         /* ## For water we already preloaded parameters at the start of the kernel */
 356         /* #if not 'Water' in GEOMETRY_I */
 357         /* Load parameters for i particles */
 358         /*     #for I in PARTICLES_ELEC_I */
 359         iq{I}              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+{I}));
 360         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 361         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 362         isai{I}            = gmx_fjsp_load1_v2r8(invsqrta+inr+{I});
 363         /*         #endif */
 364         /*     #endfor */
 365         /*     #for I in PARTICLES_VDW_I */
 366         vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 367         /*     #endfor */
 368         /* #endif */
 369
 370         /* #if 'Potential' in KERNEL_VF */
 371         /* Reset potential sums */
 372         /*     #if KERNEL_ELEC != 'None' */
 373         velecsum         = _fjsp_setzero_v2r8();
 374         /*     #endif */
 375         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 376         vgbsum           = _fjsp_setzero_v2r8();
 377         /*     #endif */
 378         /*     #if KERNEL_VDW != 'None' */
 379         vvdwsum          = _fjsp_setzero_v2r8();
 380         /*     #endif */
 381         /* #endif */
 382         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 383         dvdasum          = _fjsp_setzero_v2r8();
 384         /*     #endif */
 385
 386         /* #for ROUND in ['Loop','Epilogue'] */
 387
 388         /* #if ROUND =='Loop' */
 389         /* Start inner kernel loop */
 390         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 391         {
 392         /* ## First round is normal loop (next statement resets indentation) */
 393         /*     #if 0 */
 394         }
 395         /*     #endif */
 396         /* #else */
 397         if(jidx<j_index_end)
 398         {
 399         /* ## Second round is epilogue */
 400         /* #endif */
 401         /* #define INNERFLOPS 0 */
 402
 403             /* #if ROUND =='Loop' */
 404             /* Get j neighbor index, and coordinate index */
 405             jnrA             = jjnr[jidx];
 406             jnrB             = jjnr[jidx+1];
 407             j_coord_offsetA  = DIM*jnrA;
 408             j_coord_offsetB  = DIM*jnrB;
 409
 410             /* load j atom coordinates */
 411             /*     #if GEOMETRY_J == 'Particle'             */
 412             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 413                                               &jx0,&jy0,&jz0);
 414             /*     #elif GEOMETRY_J == 'Water3'             */
 415             gmx_fjsp_load_3rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 416                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 417             /*     #elif GEOMETRY_J == 'Water4'             */
 418             /*         #if 0 in PARTICLES_J                 */
 419             gmx_fjsp_load_4rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 420                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 421                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 422             /*         #else                                */
 423             gmx_fjsp_load_3rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 424                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 425             /*         #endif                               */
 426             /*     #endif                                   */
 427             /* #else */
 428             jnrA             = jjnr[jidx];
 429             j_coord_offsetA  = DIM*jnrA;
 430
 431             /* load j atom coordinates */
 432             /*     #if GEOMETRY_J == 'Particle'             */
 433             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 434                                               &jx0,&jy0,&jz0);
 435             /*     #elif GEOMETRY_J == 'Water3'             */
 436             gmx_fjsp_load_3rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 437                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 438             /*     #elif GEOMETRY_J == 'Water4'             */
 439             /*         #if 0 in PARTICLES_J                 */
 440             gmx_fjsp_load_4rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 441                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 442                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 443             /*         #else                                */
 444             gmx_fjsp_load_3rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA+DIM,
 445                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 446             /*         #endif                               */
 447             /*     #endif                                   */
 448             /* #endif */
 449
 450             /* Calculate displacement vector */
 451             /* #for I,J in PAIRS_IJ */
 452             dx{I}{J}             = _fjsp_sub_v2r8(ix{I},jx{J});
 453             dy{I}{J}             = _fjsp_sub_v2r8(iy{I},jy{J});
 454             dz{I}{J}             = _fjsp_sub_v2r8(iz{I},jz{J});
 455             /*     #define INNERFLOPS INNERFLOPS+3 */
 456             /* #endfor */
 457
 458             /* Calculate squared distance and things based on it */
 459             /* #for I,J in PAIRS_IJ */
 460             rsq{I}{J}            = gmx_fjsp_calc_rsq_v2r8(dx{I}{J},dy{I}{J},dz{I}{J});
 461             /*     #define INNERFLOPS INNERFLOPS+5 */
 462             /* #endfor */
 463
 464             /* #for I,J in PAIRS_IJ */
 465             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 466             rinv{I}{J}           = gmx_fjsp_invsqrt_v2r8(rsq{I}{J});
 467             /*         #define INNERFLOPS INNERFLOPS+5 */
 468             /*     #endif */
 469             /* #endfor */
 470
 471             /* #for I,J in PAIRS_IJ */
 472             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 473             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 474             rinvsq{I}{J}         = gmx_fjsp_inv_v2r8(rsq{I}{J});
 475             /*             #define INNERFLOPS INNERFLOPS+4 */
 476             /*         #else */
 477             rinvsq{I}{J}         = _fjsp_mul_v2r8(rinv{I}{J},rinv{I}{J});
 478             /*             #define INNERFLOPS INNERFLOPS+1 */
 479             /*         #endif */
 480             /*     #endif */
 481             /* #endfor */
 482
 483             /* #if not 'Water' in GEOMETRY_J */
 484             /* Load parameters for j particles */
 485             /*     #for J in PARTICLES_ELEC_J */
 486             /*         #if ROUND =='Loop' */
 487             jq{J}              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+{J},charge+jnrB+{J});
 488             /*         #else */
 489             jq{J}              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+{J});
 490             /*         #endif */
 491             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 492             /*             #if ROUND =='Loop' */
 493             isaj{J}            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+{J},invsqrta+jnrB+{J});
 494             /*             #else */
 495             isaj{J}            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+{J});
 496             /*             #endif */
 497             /*         #endif */
 498             /*     #endfor */
 499             /*     #for J in PARTICLES_VDW_J */
 500             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 501             /*         #if ROUND =='Loop' */
 502             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 503             /*         #endif */
 504             /*     #endfor */
 505             /* #endif */
 506
 507             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 508             /*     #for J in PARTICLES_J */
 509             fjx{J}             = _fjsp_setzero_v2r8();
 510             fjy{J}             = _fjsp_setzero_v2r8();
 511             fjz{J}             = _fjsp_setzero_v2r8();
 512             /*     #endfor */
 513             /* #endif */
 514
 515             /* #for I,J in PAIRS_IJ */
 516
 517             /**************************
 518              * CALCULATE INTERACTIONS *
 519              **************************/
 520
 521             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 522             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 523             if (gmx_fjsp_any_lt_v2r8(rsq{I}{J},rcutoff2))
 524             {
 525                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 526             }
 527             /*         #endif */
 528             /*         #define INNERFLOPS INNERFLOPS+1 */
 529             /*     #endif */
 530
 531             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 532             r{I}{J}              = _fjsp_mul_v2r8(rsq{I}{J},rinv{I}{J});
 533              /*         #define INNERFLOPS INNERFLOPS+1 */
 534             /*     #endif */
 535
 536             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 537             /*     #if not 'Water' in GEOMETRY_J */
 538             /* Compute parameters for interactions between i and j atoms */
 539             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 540             qq{I}{J}             = _fjsp_mul_v2r8(iq{I},jq{J});
 541             /*             #define INNERFLOPS INNERFLOPS+1 */
 542             /*         #endif */
 543             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 544             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset{I}+vdwjidx{J}A,
 545                                          vdwparam+vdwioffset{I}+vdwjidx{J}B,&c6_{I}{J},&c12_{I}{J});
 546
 547             /*             #if 'LJEwald' in KERNEL_VDW */
 548             c6grid_{I}{J}       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset{I}+vdwjidx{J}A,
 549                                                                    vdwgridparam+vdwioffset{I}+vdwjidx{J}B);
 550             /*             #endif */
 551             /*         #endif */
 552             /*     #endif */
 553
 554             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 555             /* Calculate table index by multiplying r with table scale and truncate to integer */
 556             rt               = _fjsp_mul_v2r8(r{I}{J},vftabscale);
 557             itab_tmp         = _fjsp_dtox_v2r8(rt);
 558             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 559             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 560             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 561
 562             /*         #define INNERFLOPS INNERFLOPS+4                          */
 563             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW     */
 564             /*             ## 3 tables, 4 data per point: multiply index by 12 */
 565             vfconv.i[0]     *= 12;
 566             vfconv.i[1]     *= 12;
 567             /*         #elif 'Table' in KERNEL_ELEC                             */
 568             /*             ## 1 table, 4 data per point: multiply index by 4   */
 569             vfconv.i[0]     *= 4;
 570             vfconv.i[1]     *= 4;
 571             /*         #elif 'Table' in KERNEL_VDW                              */
 572             /*             ## 2 tables, 4 data per point: multiply index by 8  */
 573             vfconv.i[0]     *= 8;
 574             vfconv.i[1]     *= 8;
 575             /*         #endif                                                   */
 576             /*     #endif */
 577
 578             /*     ## ELECTROSTATIC INTERACTIONS */
 579             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 580
 581             /*         #if KERNEL_ELEC=='Coulomb' */
 582
 583             /* COULOMB ELECTROSTATICS */
 584             velec            = _fjsp_mul_v2r8(qq{I}{J},rinv{I}{J});
 585             /*             #define INNERFLOPS INNERFLOPS+1 */
 586             /*             #if 'Force' in KERNEL_VF */
 587             felec            = _fjsp_mul_v2r8(velec,rinvsq{I}{J});
 588             /*                 #define INNERFLOPS INNERFLOPS+2 */
 589             /*             #endif */
 590
 591             /*         #elif KERNEL_ELEC=='ReactionField' */
 592
 593             /* REACTION-FIELD ELECTROSTATICS */
 594             /*             #if 'Potential' in KERNEL_VF */
 595             velec            = _fjsp_mul_v2r8(qq{I}{J},_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq{I}{J},rinv{I}{J}),crf));
 596             /*                 #define INNERFLOPS INNERFLOPS+4 */
 597             /*             #endif */
 598             /*             #if 'Force' in KERNEL_VF */
 599             felec            = _fjsp_mul_v2r8(qq{I}{J},_fjsp_msub_v2r8(rinv{I}{J},rinvsq{I}{J},krf2));
 600             /*                 #define INNERFLOPS INNERFLOPS+3 */
 601             /*             #endif */
 602
 603             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 604
 605             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 606             isaprod          = _fjsp_mul_v2r8(isai{I},isaj{J});
 607             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq{I}{J},_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 608             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 609             /*             #define INNERFLOPS INNERFLOPS+5 */
 610
 611             /* Calculate generalized born table index - this is a separate table from the normal one,
 612              * but we use the same procedure by multiplying r with scale and truncating to integer.
 613              */
 614             rt               = _fjsp_mul_v2r8(r{I}{J},gbscale);
 615             itab_tmp         = _fjsp_dtox_v2r8(rt);
 616             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 617             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 618
 619             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 620             /*             #if ROUND == 'Loop' */
 621             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 622             /*             #else */
 623             F                = _fjsp_setzero_v2r8();
 624             /*             #endif */
 625             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 626             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 627             /*             #if ROUND == 'Loop' */
 628             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 629             /*             #else */
 630             H                = _fjsp_setzero_v2r8();
 631             /*             #endif */
 632             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 633             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 634             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 635             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 636             /*             #define INNERFLOPS INNERFLOPS+10 */
 637
 638             /*             #if 'Force' in KERNEL_VF */
 639             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 640             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 641             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 642             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r{I}{J},vgb));
 643             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 644             /*             #if ROUND == 'Loop' */
 645             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj{J},isaj{J})));
 646             /*             #else */
 647             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj{J},isaj{J})));
 648             /*             #endif */
 649             /*                 #define INNERFLOPS INNERFLOPS+13 */
 650             /*             #endif */
 651             velec            = _fjsp_mul_v2r8(qq{I}{J},rinv{I}{J});
 652             /*                 #define INNERFLOPS INNERFLOPS+1 */
 653             /*             #if 'Force' in KERNEL_VF */
 654             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv{I}{J},fgb),rinv{I}{J});
 655             /*                 #define INNERFLOPS INNERFLOPS+3 */
 656             /*             #endif */
 657
 658             /*         #elif KERNEL_ELEC=='Ewald' */
 659             /* EWALD ELECTROSTATICS */
 660
 661             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 662             ewrt             = _fjsp_mul_v2r8(r{I}{J},ewtabscale);
 663             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 664             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 665             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 666
 667             /*             #define INNERFLOPS INNERFLOPS+4 */
 668             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 669             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 670             /*                 #if ROUND == 'Loop' */
 671             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 672             /*                 #else */
 673             ewtabD           = _fjsp_setzero_v2r8();
 674             /*                 #endif */
 675             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 676             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 677             /*                 #if ROUND == 'Loop' */
 678             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 679             /*                 #else */
 680             ewtabFn          = _fjsp_setzero_v2r8();
 681             /*                 #endif */
 682             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 683             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 684             /*                 #define INNERFLOPS INNERFLOPS+2 */
 685             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 686             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 687             velec            = _fjsp_mul_v2r8(qq{I}{J},_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv{I}{J},sh_ewald),velec));
 688             /*                     #define INNERFLOPS INNERFLOPS+7 */
 689             /*                 #else */
 690             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 691             velec            = _fjsp_mul_v2r8(qq{I}{J},_fjsp_sub_v2r8(rinv{I}{J},velec));
 692             /*                     #define INNERFLOPS INNERFLOPS+6 */
 693             /*                 #endif */
 694             /*                 #if 'Force' in KERNEL_VF */
 695             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq{I}{J},rinv{I}{J}),_fjsp_sub_v2r8(rinvsq{I}{J},felec));
 696             /*                      #define INNERFLOPS INNERFLOPS+3 */
 697             /*                 #endif */
 698             /*             #elif KERNEL_VF=='Force' */
 699             /*                 #if ROUND == 'Loop' */
 700             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 701                                          &ewtabF,&ewtabFn);
 702             /*                 #else */
 703             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 704             /*                 #endif */
 705             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 706             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq{I}{J},rinv{I}{J}),_fjsp_sub_v2r8(rinvsq{I}{J},felec));
 707             /*                 #define INNERFLOPS INNERFLOPS+7 */
 708             /*             #endif */
 709
 710             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 711
 712             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 713             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 714             /*             #if ROUND == 'Loop' */
 715             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 716             /*             #else */
 717             F                = _fjsp_setzero_v2r8();
 718             /*             #endif */
 719             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 720             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 721             /*             #if ROUND == 'Loop' */
 722             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 723             /*             #else */
 724             H                = _fjsp_setzero_v2r8();
 725             /*             #endif */
 726             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 727             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 728             /*             #define INNERFLOPS INNERFLOPS+4 */
 729             /*             #if 'Potential' in KERNEL_VF */
 730             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 731             velec            = _fjsp_mul_v2r8(qq{I}{J},VV);
 732             /*                 #define INNERFLOPS INNERFLOPS+3 */
 733             /*             #endif */
 734             /*             #if 'Force' in KERNEL_VF */
 735             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 736             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq{I}{J},FF),_fjsp_mul_v2r8(vftabscale,rinv{I}{J})));
 737             /*                 #define INNERFLOPS INNERFLOPS+7 */
 738             /*             #endif */
 739             /*         #endif */
 740             /*         ## End of check for electrostatics interaction forms */
 741             /*     #endif */
 742             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 743
 744             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 745
 746             /*         #if KERNEL_VDW=='LennardJones' */
 747
 748             /* LENNARD-JONES DISPERSION/REPULSION */
 749
 750             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 751             /*             #define INNERFLOPS INNERFLOPS+2 */
 752             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 753             vvdw6            = _fjsp_mul_v2r8(c6_{I}{J},rinvsix);
 754             vvdw12           = _fjsp_mul_v2r8(c12_{I}{J},_fjsp_mul_v2r8(rinvsix,rinvsix));
 755             /*                 #define INNERFLOPS INNERFLOPS+3 */
 756             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 757             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_{I}{J},_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 758                                            _fjsp_mul_v2r8(_fjsp_nmsub_v2r8( c6_{I}{J},sh_vdw_invrcut6,vvdw6),one_sixth));
 759             /*                     #define INNERFLOPS INNERFLOPS+8 */
 760             /*                 #else */
 761             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 762             /*                     #define INNERFLOPS INNERFLOPS+3 */
 763             /*                 #endif */
 764             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 765             /*                 #if 'Force' in KERNEL_VF */
 766             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq{I}{J});
 767             /*                     #define INNERFLOPS INNERFLOPS+2 */
 768             /*                 #endif */
 769             /*             #elif KERNEL_VF=='Force' */
 770             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 771             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_{I}{J},rinvsix,c6_{I}{J}),_fjsp_mul_v2r8(rinvsix,rinvsq{I}{J}));
 772             /*                 #define INNERFLOPS INNERFLOPS+4 */
 773             /*             #endif */
 774
 775             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 776
 777             /* CUBIC SPLINE TABLE DISPERSION */
 778             /*             #if 'Table' in KERNEL_ELEC */
 779             vfconv.i[0]       += 4;
 780             vfconv.i[1]       += 4;
 781             /*             #endif                     */
 782             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 783             /*             #if ROUND == 'Loop' */
 784             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 785             /*             #else */
 786             F                = _fjsp_setzero_v2r8();
 787             /*             #endif */
 788             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 789             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 790             /*             #if ROUND == 'Loop' */
 791             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 792             /*             #else */
 793             H                = _fjsp_setzero_v2r8();
 794             /*             #endif */
 795             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 796             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 797             /*             #define INNERFLOPS INNERFLOPS+4 */
 798             /*             #if 'Potential' in KERNEL_VF */
 799             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 800             vvdw6            = _fjsp_mul_v2r8(c6_{I}{J},VV);
 801             /*                 #define INNERFLOPS INNERFLOPS+3 */
 802             /*             #endif */
 803             /*             #if 'Force' in KERNEL_VF */
 804             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 805             fvdw6            = _fjsp_mul_v2r8(c6_{I}{J},FF);
 806             /*                 #define INNERFLOPS INNERFLOPS+4 */
 807             /*             #endif */
 808
 809             /* CUBIC SPLINE TABLE REPULSION */
 810             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 811             /*             #if ROUND == 'Loop' */
 812             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 813             /*             #else */
 814             F                = _fjsp_setzero_v2r8();
 815             /*             #endif */
 816             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 817             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 818             /*             #if ROUND == 'Loop' */
 819             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 820             /*             #else */
 821             H                = _fjsp_setzero_v2r8();
 822             /*             #endif */
 823             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 824             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 825             /*             #define INNERFLOPS INNERFLOPS+4 */
 826             /*             #if 'Potential' in KERNEL_VF */
 827             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 828             vvdw12           = _fjsp_mul_v2r8(c12_{I}{J},VV);
 829             /*                 #define INNERFLOPS INNERFLOPS+3 */
 830             /*             #endif */
 831             /*             #if 'Force' in KERNEL_VF */
 832             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 833             fvdw12           = _fjsp_mul_v2r8(c12_{I}{J},FF);
 834             /*                 #define INNERFLOPS INNERFLOPS+5 */
 835             /*             #endif */
 836             /*             #if 'Potential' in KERNEL_VF */
 837             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 838             /*                 #define INNERFLOPS INNERFLOPS+1 */
 839             /*             #endif */
 840             /*             #if 'Force' in KERNEL_VF */
 841             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv{I}{J})));
 842             /*                 #define INNERFLOPS INNERFLOPS+4 */
 843             /*             #endif */
 844
 845             /*         #elif KERNEL_VDW=='LJEwald' */
 846
 847             /* Analytical LJ-PME */
 848             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 849             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq{I}{J});
 850             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 851             exponent         = gmx_simd_exp_d(ewcljrsq);
 852             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 853             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 854             /*                 #define INNERFLOPS INNERFLOPS+9 */
 855             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 856             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 857             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_{I}{J},_fjsp_sub_v2r8(poly,one),c6_{I}{J}),rinvsix);
 858             vvdw12           = _fjsp_mul_v2r8(c12_{I}{J},_fjsp_mul_v2r8(rinvsix,rinvsix));
 859             /*                 #define INNERFLOPS INNERFLOPS+5 */
 860             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 861             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_{I}{J},_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 862                                _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw6,_fjsp_madd_v2r8(c6grid_{I}{J},sh_lj_ewald,_fjsp_mul_v2r8(c6_{I}{J},sh_vdw_invrcut6))),one_sixth));
 863             /*                     #define INNERFLOPS INNERFLOPS+7 */
 864             /*                 #else */
 865             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 866             /*                 #define INNERFLOPS INNERFLOPS+2 */
 867             /*                 #endif */
 868             /*                  ## Check for force inside potential check, i.e. this means we already did the potential part */
 869             /*                  #if 'Force' in KERNEL_VF */
 870             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 871             fvdw             = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_{I}{J},one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq{I}{J});
 872             /*                 #define INNERFLOPS INNERFLOPS+6 */
 873             /*                  #endif */
 874             /*              #elif KERNEL_VF=='Force' */
 875             /* f6A = 6 * C6grid * (1 - poly) */
 876             f6A              = _fjsp_mul_v2r8(c6grid_{I}{J},_fjsp_sub_v2r8(one,poly));
 877             /* f6B = C6grid * exponent * beta^6 */
 878             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_{I}{J},one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 879             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 880             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_{I}{J},rinvsix,_fjsp_sub_v2r8(c6_{I}{J},f6A)),rinvsix,f6B),rinvsq{I}{J});
 881             /*                 #define INNERFLOPS INNERFLOPS+12 */
 882             /*              #endif */
 883             /*         #endif */
 884             /*         ## End of check for vdw interaction forms */
 885             /*     #endif */
 886             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 887
 888             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 889             d                = _fjsp_sub_v2r8(r{I}{J},rswitch);
 890             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 891             d2               = _fjsp_mul_v2r8(d,d);
 892             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 893             /*         #define INNERFLOPS INNERFLOPS+10 */
 894
 895             /*         #if 'Force' in KERNEL_VF */
 896             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 897             /*             #define INNERFLOPS INNERFLOPS+5 */
 898             /*         #endif */
 899
 900             /* Evaluate switch function */
 901             /*         #if 'Force' in KERNEL_VF */
 902             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 903             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 904             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv{I}{J},_fjsp_mul_v2r8(velec,dsw)) );
 905             /*                 #define INNERFLOPS INNERFLOPS+4 */
 906             /*             #endif */
 907             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 908             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv{I}{J},_fjsp_mul_v2r8(vvdw,dsw)) );
 909             /*                 #define INNERFLOPS INNERFLOPS+4 */
 910             /*             #endif */
 911             /*         #endif */
 912             /*         #if 'Potential' in KERNEL_VF */
 913             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 914             velec            = _fjsp_mul_v2r8(velec,sw);
 915             /*                 #define INNERFLOPS INNERFLOPS+1 */
 916             /*             #endif */
 917             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 918             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 919             /*                 #define INNERFLOPS INNERFLOPS+1 */
 920             /*             #endif */
 921             /*         #endif */
 922             /*     #endif */
 923             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 924             cutoff_mask      = _fjsp_cmplt_v2r8(rsq{I}{J},rcutoff2);
 925             /*         #define INNERFLOPS INNERFLOPS+1 */
 926             /*     #endif */
 927
 928             /*     #if 'Potential' in KERNEL_VF */
 929             /* Update potential sum for this i atom from the interaction with this j atom. */
 930             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 931             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 932             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 933             /*                 #define INNERFLOPS INNERFLOPS+1 */
 934             /*             #endif                                       */
 935             /*             #if ROUND == 'Epilogue' */
 936             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 937             /*             #endif */
 938             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 939             /*             #define INNERFLOPS INNERFLOPS+1 */
 940             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 941             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 942             vgb              = _fjsp_and_v2r8(vgb,cutoff_mask);
 943             /*                 #define INNERFLOPS INNERFLOPS+1 */
 944             /*             #endif                                       */
 945             /*             #if ROUND == 'Epilogue' */
 946             vgb              = _fjsp_unpacklo_v2r8(vgb,_fjsp_setzero_v2r8());
 947             /*             #endif */
 948             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 949             /*                 #define INNERFLOPS INNERFLOPS+1 */
 950             /*             #endif */
 951             /*         #endif */
 952             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 953             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 954             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 955             /*                 #define INNERFLOPS INNERFLOPS+1 */
 956             /*             #endif                                       */
 957             /*             #if ROUND == 'Epilogue' */
 958             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 959             /*             #endif */
 960             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 961             /*             #define INNERFLOPS INNERFLOPS+1 */
 962             /*         #endif */
 963             /*     #endif */
 964
 965             /*     #if 'Force' in KERNEL_VF */
 966
 967             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 968             fscal            = _fjsp_add_v2r8(felec,fvdw);
 969             /*             #define INNERFLOPS INNERFLOPS+1 */
 970             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 971             fscal            = felec;
 972             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 973             fscal            = fvdw;
 974             /*        #endif */
 975
 976             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 977             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 978             /*                 #define INNERFLOPS INNERFLOPS+1 */
 979             /*             #endif                                       */
 980
 981             /*             #if ROUND == 'Epilogue' */
 982             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 983             /*             #endif */
 984
 985             /* ## Construction of vectorial force built into FMA instructions now */
 986             /* #define INNERFLOPS INNERFLOPS+3      */
 987
 988             /* Update vectorial force */
 989             fix{I}             = _fjsp_madd_v2r8(dx{I}{J},fscal,fix{I});
 990             fiy{I}             = _fjsp_madd_v2r8(dy{I}{J},fscal,fiy{I});
 991             fiz{I}             = _fjsp_madd_v2r8(dz{I}{J},fscal,fiz{I});
 992             /*             #define INNERFLOPS INNERFLOPS+6 */
 993
 994             /* #if GEOMETRY_I == 'Particle'             */
 995             /*     #if ROUND == 'Loop' */
 996             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx{I}{J},dy{I}{J},dz{I}{J});
 997             /*     #else */
 998             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx{I}{J},dy{I}{J},dz{I}{J});
 999             /*     #endif */
1000             /*     #define INNERFLOPS INNERFLOPS+3      */
1001             /* #else                                    */
1002             fjx{J}             = _fjsp_madd_v2r8(dx{I}{J},fscal,fjx{J});
1003             fjy{J}             = _fjsp_madd_v2r8(dy{I}{J},fscal,fjy{J});
1004             fjz{J}             = _fjsp_madd_v2r8(dz{I}{J},fscal,fjz{J});
1005             /*     #define INNERFLOPS INNERFLOPS+3      */
1006             /* #endif                                   */
1007
1008             /*     #endif */
1009
1010             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
1011             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
1012             {
1013                 /*     #endif */
1014             }
1015             /*     #endif */
1016             /*    ## End of check for the interaction being outside the cutoff */
1017
1018             /* #endfor */
1019             /* ## End of loop over i-j interaction pairs */
1020
1021             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
1022             /*     #if ROUND == 'Loop' */
1023             gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1024             /*     #else */
1025             gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1026             /*     #endif */
1027             /*     #define INNERFLOPS INNERFLOPS+3      */
1028             /* #elif GEOMETRY_J == 'Water3'             */
1029             /*     #if ROUND == 'Loop' */
1030             gmx_fjsp_decrement_3rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1031             /*     #else */
1032             gmx_fjsp_decrement_3rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1033             /*     #endif */
1034             /*     #define INNERFLOPS INNERFLOPS+9      */
1035             /* #elif GEOMETRY_J == 'Water4'             */
1036             /*     #if 0 in PARTICLES_J                 */
1037             /*         #if ROUND == 'Loop' */
1038             gmx_fjsp_decrement_4rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1039             /*         #else */
1040             gmx_fjsp_decrement_4rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1041             /*         #endif */
1042             /*         #define INNERFLOPS INNERFLOPS+12 */
1043             /*     #else                                */
1044             /*         #if ROUND == 'Loop' */
1045             gmx_fjsp_decrement_3rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1046             /*         #else */
1047             gmx_fjsp_decrement_3rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1048             /*         #endif */
1049             /*         #define INNERFLOPS INNERFLOPS+9  */
1050             /*     #endif                               */
1051             /* #endif                                   */
1052
1053             /* Inner loop uses {INNERFLOPS} flops */
1054         }
1055
1056         /* #endfor */
1057
1058         /* End of innermost loop */
1059
1060         /* #if 'Force' in KERNEL_VF */
1061         /*     #if GEOMETRY_I == 'Particle'            */
1062         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
1063                                               f+i_coord_offset,fshift+i_shift_offset);
1064         /*         #define OUTERFLOPS OUTERFLOPS+6     */
1065         /*     #elif GEOMETRY_I == 'Water3'            */
1066         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1067                                               f+i_coord_offset,fshift+i_shift_offset);
1068         /*         #define OUTERFLOPS OUTERFLOPS+18    */
1069         /*     #elif GEOMETRY_I == 'Water4'            */
1070         /*         #if 0 in PARTICLES_I                */
1071         gmx_fjsp_update_iforce_4atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1072                                               f+i_coord_offset,fshift+i_shift_offset);
1073         /*             #define OUTERFLOPS OUTERFLOPS+24    */
1074         /*         #else                               */
1075         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1076                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1077         /*             #define OUTERFLOPS OUTERFLOPS+18    */
1078         /*         #endif                              */
1079         /*     #endif                                  */
1080         /* #endif                                      */
1081
1082         /* #if 'Potential' in KERNEL_VF */
1083         ggid                        = gid[iidx];
1084         /* Update potential energies */
1085         /*     #if KERNEL_ELEC != 'None' */
1086         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
1087         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1088         /*     #endif */
1089         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
1090         gmx_fjsp_update_1pot_v2r8(vgbsum,kernel_data->energygrp_polarization+ggid);
1091         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1092         /*     #endif */
1093         /*     #if KERNEL_VDW != 'None' */
1094         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
1095         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1096         /*     #endif */
1097         /* #endif */
1098         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1099         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai{I},isai{I}));
1100         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
1101         /*     #endif */
1102
1103         /* Increment number of inner iterations */
1104         inneriter                  += j_index_end - j_index_start;
1105
1106         /* Outer loop uses {OUTERFLOPS} flops */
1107     }
1108
1109     /* Increment number of outer iterations */
1110     outeriter        += nri;
1111
1112     /* Update outer/inner flops */
1113     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1114     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1115     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1116     /* #if GEOMETRY_I == 'Water3'            */
1117     /*     #define ISUFFIX '_W3'             */
1118     /* #elif GEOMETRY_I == 'Water4'          */
1119     /*     #define ISUFFIX '_W4'             */
1120     /* #else                                 */
1121     /*     #define ISUFFIX ''                */
1122     /* #endif                                */
1123     /* #if GEOMETRY_J == 'Water3'            */
1124     /*     #define JSUFFIX 'W3'              */
1125     /* #elif GEOMETRY_J == 'Water4'          */
1126     /*     #define JSUFFIX 'W4'              */
1127     /* #else                                 */
1128     /*     #define JSUFFIX ''                */
1129     /* #endif                                */
1130     /* #if 'PotentialAndForce' in KERNEL_VF  */
1131     /*     #define VFSUFFIX  '_VF'           */
1132     /* #elif 'Potential' in KERNEL_VF        */
1133     /*     #define VFSUFFIX '_V'             */
1134     /* #else                                 */
1135     /*     #define VFSUFFIX '_F'             */
1136     /* #endif                                */
1137
1138     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1139     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1140     /* #elif KERNEL_ELEC != 'None' */
1141     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1142     /* #else */
1143     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1144     /* #endif  */
1145 }