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, 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 #ifdef HAVE_CONFIG_H
  40 #include <config.h>
  41 #endif
  42
  43 #include <math.h>
  44
  45 #include "../nb_kernel.h"
  46 #include "types/simple.h"
  47 #include "gromacs/legacyheaders/vec.h"
  48 #include "nrnb.h"
  49
  50 #include "kernelutil_sparc64_hpc_ace_double.h"
  51 /* #endif */
  52
  53 /* ## List of variables set by the generating script:                                    */
  54 /* ##                                                                                    */
  55 /* ## Setttings that apply to the entire kernel:                                         */
  56 /* ## KERNEL_ELEC:           String, choice for electrostatic interactions               */
  57 /* ## KERNEL_VDW:            String, choice for van der Waals interactions               */
  58 /* ## KERNEL_NAME:           String, name of this kernel                                 */
  59 /* ## KERNEL_VF:             String telling if we calculate potential, force, or both    */
  60 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
  61 /* ##                                                                                    */
  62 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops:             */
  63 /* ## PARTICLES_I[]/         Arrays with lists of i/j particles to use in kernel. It is  */
  64 /* ## PARTICLES_J[]:         just [0] for particle geometry, but can be longer for water */
  65 /* ## PARTICLES_ELEC_I[]/    Arrays with lists of i/j particle that have electrostatics  */
  66 /* ## PARTICLES_ELEC_J[]:    interactions that should be calculated in this kernel.      */
  67 /* ## PARTICLES_VDW_I[]/     Arrays with the list of i/j particle that have VdW          */
  68 /* ## PARTICLES_VDW_J[]:     interactions that should be calculated in this kernel.      */
  69 /* ##                                                                                    */
  70 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle)   */
  71 /* ## PAIRS_IJ[]:            Array with (i,j) tuples of pairs for which interactions     */
  72 /* ##                        should be calculated in this kernel. Zero-charge particles  */
  73 /* ##                        do not have interactions with particles without vdw, and    */
  74 /* ##                        Vdw-only interactions are not evaluated in a no-vdw-kernel. */
  75 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
  76 /* ##                        For each i-j pair, the element [I][J] is a list of strings  */
  77 /* ##                        defining properties/flags of this interaction. Examples     */
  78 /* ##                        include 'electrostatics'/'vdw' if that type of interaction  */
  79 /* ##                        should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values  */
  80 /* ##                        are needed, and 'exactcutoff' or 'shift','switch' to        */
  81 /* ##                        decide if the force/potential should be modified. This way  */
  82 /* ##                        we only calculate values absolutely needed for each case.   */
  83
  84 /* ## Calculate the size and offset for (merged/interleaved) table data */
  85
  86 /*
  87  * Gromacs nonbonded kernel:   {KERNEL_NAME}
  88  * Electrostatics interaction: {KERNEL_ELEC}
  89  * VdW interaction:            {KERNEL_VDW}
  90  * Geometry:                   {GEOMETRY_I}-{GEOMETRY_J}
  91  * Calculate force/pot:        {KERNEL_VF}
  92  */
  93 void
  94 {KERNEL_NAME}
  95                     (t_nblist                    * gmx_restrict       nlist,
  96                      rvec                        * gmx_restrict          xx,
  97                      rvec                        * gmx_restrict          ff,
  98                      t_forcerec                  * gmx_restrict          fr,
  99                      t_mdatoms                   * gmx_restrict     mdatoms,
 100                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 101                      t_nrnb                      * gmx_restrict        nrnb)
 102 {
 103     /* ## Not all variables are used for all kernels, but any optimizing compiler fixes that, */
 104     /* ## so there is no point in going to extremes to exclude variables that are not needed. */
 105     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 106      * just 0 for non-waters.
 107      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 108      * jnr indices corresponding to data put in the four positions in the SIMD register.
 109      */
 110     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 111     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 112     int              jnrA,jnrB;
 113     int              j_coord_offsetA,j_coord_offsetB;
 114     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 115     real             rcutoff_scalar;
 116     real             *shiftvec,*fshift,*x,*f;
 117     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 118     /* #for I in PARTICLES_I */
 119     int              vdwioffset{I};
 120     _fjsp_v2r8       ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 121     /* #endfor */
 122     /* #for J in PARTICLES_J */
 123     int              vdwjidx{J}A,vdwjidx{J}B;
 124     _fjsp_v2r8       jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 125     /* #endfor */
 126     /* #for I,J in PAIRS_IJ */
 127     _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};
 128     /* #endfor */
 129     /* #if KERNEL_ELEC != 'None' */
 130     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 131     real             *charge;
 132     /* #endif */
 133     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 134     _fjsp_v2r8       vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,twogbeps,dvdatmp;
 135     _fjsp_v2r8       minushalf = gmx_fjsp_set1_v2r8(-0.5);
 136     real             *invsqrta,*dvda,*gbtab;
 137     /* #endif */
 138     /* #if KERNEL_VDW != 'None' */
 139     int              nvdwtype;
 140     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 141     int              *vdwtype;
 142     real             *vdwparam;
 143     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 144     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 145     /* #endif */
 146     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 147     _fjsp_v2r8       rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 148     real             *vftab;
 149     /* #endif */
 150     /* #if 'LJEwald' in KERNEL_VDW */
 151     /* #for I,J in PAIRS_IJ */
 152     _fjsp_v2r8           c6grid_{I}{J};
 153     /* #endfor */
 154     real                 *vdwgridparam;
 155     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 156     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 157     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 158     /* #endif */
 159     /* #if 'Ewald' in KERNEL_ELEC */
 160     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 161     real             *ewtab;
 162     /* #endif */
 163     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 164     _fjsp_v2r8       rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 165     real             rswitch_scalar,d_scalar;
 166     /* #endif */
 167     _fjsp_v2r8       itab_tmp;
 168     _fjsp_v2r8       dummy_mask,cutoff_mask;
 169     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 170     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 171     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 172
 173     x                = xx[0];
 174     f                = ff[0];
 175
 176     nri              = nlist->nri;
 177     iinr             = nlist->iinr;
 178     jindex           = nlist->jindex;
 179     jjnr             = nlist->jjnr;
 180     shiftidx         = nlist->shift;
 181     gid              = nlist->gid;
 182     shiftvec         = fr->shift_vec[0];
 183     fshift           = fr->fshift[0];
 184     /* #if KERNEL_ELEC != 'None' */
 185     facel            = gmx_fjsp_set1_v2r8(fr->epsfac);
 186     charge           = mdatoms->chargeA;
 187     /*     #if 'ReactionField' in KERNEL_ELEC */
 188     krf              = gmx_fjsp_set1_v2r8(fr->ic->k_rf);
 189     krf2             = gmx_fjsp_set1_v2r8(fr->ic->k_rf*2.0);
 190     crf              = gmx_fjsp_set1_v2r8(fr->ic->c_rf);
 191     /*     #endif */
 192     /* #endif */
 193     /* #if KERNEL_VDW != 'None' */
 194     nvdwtype         = fr->ntype;
 195     vdwparam         = fr->nbfp;
 196     vdwtype          = mdatoms->typeA;
 197     /* #endif */
 198     /* #if 'LJEwald' in KERNEL_VDW */
 199     vdwgridparam     = fr->ljpme_c6grid;
 200     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 201     ewclj            = gmx_fjsp_set1_v2r8(fr->ewaldcoeff_lj);
 202     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 203     /* #endif */
 204
 205     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 206     vftab            = kernel_data->table_elec_vdw->data;
 207     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec_vdw->scale);
 208     /* #elif 'Table' in KERNEL_ELEC */
 209     vftab            = kernel_data->table_elec->data;
 210     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_elec->scale);
 211     /* #elif 'Table' in KERNEL_VDW */
 212     vftab            = kernel_data->table_vdw->data;
 213     vftabscale       = gmx_fjsp_set1_v2r8(kernel_data->table_vdw->scale);
 214     /* #endif */
 215
 216     /* #if 'Ewald' in KERNEL_ELEC */
 217     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 218     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 219     ewtab            = fr->ic->tabq_coul_F;
 220     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 221     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 222     /*     #else */
 223     ewtab            = fr->ic->tabq_coul_FDV0;
 224     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 225     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 226      /*     #endif */
 227     /* #endif */
 228
 229     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 230     invsqrta         = fr->invsqrta;
 231     dvda             = fr->dvda;
 232     gbtabscale       = gmx_fjsp_set1_v2r8(fr->gbtab.scale);
 233     gbtab            = fr->gbtab.data;
 234     gbinvepsdiff     = gmx_fjsp_set1_v2r8((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 235     /* #endif */
 236
 237     /* #if 'Water' in GEOMETRY_I */
 238     /* Setup water-specific parameters */
 239     inr              = nlist->iinr[0];
 240     /*     #for I in PARTICLES_ELEC_I */
 241     iq{I}              = _fjsp_mul_v2r8(facel,gmx_fjsp_set1_v2r8(charge[inr+{I}]));
 242     /*     #endfor */
 243     /*     #for I in PARTICLES_VDW_I */
 244     vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 245     /*     #endfor */
 246     /* #endif */
 247
 248     /* #if 'Water' in GEOMETRY_J */
 249     /*     #for J in PARTICLES_ELEC_J */
 250     jq{J}              = gmx_fjsp_set1_v2r8(charge[inr+{J}]);
 251     /*     #endfor */
 252     /*     #for J in PARTICLES_VDW_J */
 253     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 254     /*     #endfor */
 255     /*     #for I,J in PAIRS_IJ */
 256     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 257     qq{I}{J}             = _fjsp_mul_v2r8(iq{I},jq{J});
 258     /*         #endif */
 259     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 260     /*             #if 'LJEwald' in KERNEL_VDW */
 261     c6_{I}{J}            = gmx_fjsp_set1_v2r8(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 262     c12_{I}{J}           = gmx_fjsp_set1_v2r8(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 263     c6grid_{I}{J}        = gmx_fjsp_set1_v2r8(vdwgridparam[vdwioffset{I}+vdwjidx{J}A]);
 264     /*             #else */
 265     c6_{I}{J}            = gmx_fjsp_set1_v2r8(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 266     c12_{I}{J}           = gmx_fjsp_set1_v2r8(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 267     /*             #endif */
 268     /*         #endif */
 269     /*     #endfor */
 270     /* #endif */
 271
 272     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 273     /*     #if KERNEL_ELEC!='None' */
 274     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 275     rcutoff_scalar   = fr->rcoulomb;
 276     /*     #else */
 277     rcutoff_scalar   = fr->rvdw;
 278     /*     #endif */
 279     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 280     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 281     /* #endif */
 282
 283     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 284     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 285     rvdw             = gmx_fjsp_set1_v2r8(fr->rvdw);
 286     /* #endif */
 287
 288     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 289     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 290     rswitch_scalar   = fr->rcoulomb_switch;
 291     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 292     /*     #else */
 293     rswitch_scalar   = fr->rvdw_switch;
 294     rswitch          = gmx_fjsp_set1_v2r8(rswitch_scalar);
 295     /*     #endif */
 296     /* Setup switch parameters */
 297     d_scalar         = rcutoff_scalar-rswitch_scalar;
 298     d                = gmx_fjsp_set1_v2r8(d_scalar);
 299     swV3             = gmx_fjsp_set1_v2r8(-10.0/(d_scalar*d_scalar*d_scalar));
 300     swV4             = gmx_fjsp_set1_v2r8( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 301     swV5             = gmx_fjsp_set1_v2r8( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 302     /*     #if 'Force' in KERNEL_VF */
 303     swF2             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar));
 304     swF3             = gmx_fjsp_set1_v2r8( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 305     swF4             = gmx_fjsp_set1_v2r8(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 306     /*     #endif */
 307     /* #endif */
 308
 309     /* Avoid stupid compiler warnings */
 310     jnrA = jnrB = 0;
 311     j_coord_offsetA = 0;
 312     j_coord_offsetB = 0;
 313
 314     /* ## Keep track of the floating point operations we issue for reporting! */
 315     /* #define OUTERFLOPS 0 */
 316     outeriter        = 0;
 317     inneriter        = 0;
 318
 319     /* Start outer loop over neighborlists */
 320     for(iidx=0; iidx<nri; iidx++)
 321     {
 322         /* Load shift vector for this list */
 323         i_shift_offset   = DIM*shiftidx[iidx];
 324
 325         /* Load limits for loop over neighbors */
 326         j_index_start    = jindex[iidx];
 327         j_index_end      = jindex[iidx+1];
 328
 329         /* Get outer coordinate index */
 330         inr              = iinr[iidx];
 331         i_coord_offset   = DIM*inr;
 332
 333         /* Load i particle coords and add shift vector */
 334         /* #if GEOMETRY_I == 'Particle' */
 335         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 336         /* #elif GEOMETRY_I == 'Water3' */
 337         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 338                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 339         /* #elif GEOMETRY_I == 'Water4' */
 340         /*     #if 0 in PARTICLES_I                 */
 341         gmx_fjsp_load_shift_and_4rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,
 342                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 343         /*     #else                                */
 344         gmx_fjsp_load_shift_and_3rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 345                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 346         /*     #endif                               */
 347         /* #endif                                   */
 348
 349         /* #if 'Force' in KERNEL_VF */
 350         /*     #for I in PARTICLES_I */
 351         fix{I}             = _fjsp_setzero_v2r8();
 352         fiy{I}             = _fjsp_setzero_v2r8();
 353         fiz{I}             = _fjsp_setzero_v2r8();
 354         /*     #endfor */
 355         /* #endif */
 356
 357         /* ## For water we already preloaded parameters at the start of the kernel */
 358         /* #if not 'Water' in GEOMETRY_I */
 359         /* Load parameters for i particles */
 360         /*     #for I in PARTICLES_ELEC_I */
 361         iq{I}              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+{I}));
 362         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 363         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 364         isai{I}            = gmx_fjsp_load1_v2r8(invsqrta+inr+{I});
 365         /*         #endif */
 366         /*     #endfor */
 367         /*     #for I in PARTICLES_VDW_I */
 368         vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 369         /*     #endfor */
 370         /* #endif */
 371
 372         /* #if 'Potential' in KERNEL_VF */
 373         /* Reset potential sums */
 374         /*     #if KERNEL_ELEC != 'None' */
 375         velecsum         = _fjsp_setzero_v2r8();
 376         /*     #endif */
 377         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 378         vgbsum           = _fjsp_setzero_v2r8();
 379         /*     #endif */
 380         /*     #if KERNEL_VDW != 'None' */
 381         vvdwsum          = _fjsp_setzero_v2r8();
 382         /*     #endif */
 383         /* #endif */
 384         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 385         dvdasum          = _fjsp_setzero_v2r8();
 386         /*     #endif */
 387
 388         /* #for ROUND in ['Loop','Epilogue'] */
 389
 390         /* #if ROUND =='Loop' */
 391         /* Start inner kernel loop */
 392         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 393         {
 394         /* ## First round is normal loop (next statement resets indentation) */
 395         /*     #if 0 */
 396         }
 397         /*     #endif */
 398         /* #else */
 399         if(jidx<j_index_end)
 400         {
 401         /* ## Second round is epilogue */
 402         /* #endif */
 403         /* #define INNERFLOPS 0 */
 404
 405             /* #if ROUND =='Loop' */
 406             /* Get j neighbor index, and coordinate index */
 407             jnrA             = jjnr[jidx];
 408             jnrB             = jjnr[jidx+1];
 409             j_coord_offsetA  = DIM*jnrA;
 410             j_coord_offsetB  = DIM*jnrB;
 411
 412             /* load j atom coordinates */
 413             /*     #if GEOMETRY_J == 'Particle'             */
 414             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 415                                               &jx0,&jy0,&jz0);
 416             /*     #elif GEOMETRY_J == 'Water3'             */
 417             gmx_fjsp_load_3rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 418                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 419             /*     #elif GEOMETRY_J == 'Water4'             */
 420             /*         #if 0 in PARTICLES_J                 */
 421             gmx_fjsp_load_4rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 422                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 423                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 424             /*         #else                                */
 425             gmx_fjsp_load_3rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 426                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 427             /*         #endif                               */
 428             /*     #endif                                   */
 429             /* #else */
 430             jnrA             = jjnr[jidx];
 431             j_coord_offsetA  = DIM*jnrA;
 432
 433             /* load j atom coordinates */
 434             /*     #if GEOMETRY_J == 'Particle'             */
 435             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 436                                               &jx0,&jy0,&jz0);
 437             /*     #elif GEOMETRY_J == 'Water3'             */
 438             gmx_fjsp_load_3rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 439                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 440             /*     #elif GEOMETRY_J == 'Water4'             */
 441             /*         #if 0 in PARTICLES_J                 */
 442             gmx_fjsp_load_4rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 443                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 444                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 445             /*         #else                                */
 446             gmx_fjsp_load_3rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA+DIM,
 447                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 448             /*         #endif                               */
 449             /*     #endif                                   */
 450             /* #endif */
 451
 452             /* Calculate displacement vector */
 453             /* #for I,J in PAIRS_IJ */
 454             dx{I}{J}             = _fjsp_sub_v2r8(ix{I},jx{J});
 455             dy{I}{J}             = _fjsp_sub_v2r8(iy{I},jy{J});
 456             dz{I}{J}             = _fjsp_sub_v2r8(iz{I},jz{J});
 457             /*     #define INNERFLOPS INNERFLOPS+3 */
 458             /* #endfor */
 459
 460             /* Calculate squared distance and things based on it */
 461             /* #for I,J in PAIRS_IJ */
 462             rsq{I}{J}            = gmx_fjsp_calc_rsq_v2r8(dx{I}{J},dy{I}{J},dz{I}{J});
 463             /*     #define INNERFLOPS INNERFLOPS+5 */
 464             /* #endfor */
 465
 466             /* #for I,J in PAIRS_IJ */
 467             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 468             rinv{I}{J}           = gmx_fjsp_invsqrt_v2r8(rsq{I}{J});
 469             /*         #define INNERFLOPS INNERFLOPS+5 */
 470             /*     #endif */
 471             /* #endfor */
 472
 473             /* #for I,J in PAIRS_IJ */
 474             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 475             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 476             rinvsq{I}{J}         = gmx_fjsp_inv_v2r8(rsq{I}{J});
 477             /*             #define INNERFLOPS INNERFLOPS+4 */
 478             /*         #else */
 479             rinvsq{I}{J}         = _fjsp_mul_v2r8(rinv{I}{J},rinv{I}{J});
 480             /*             #define INNERFLOPS INNERFLOPS+1 */
 481             /*         #endif */
 482             /*     #endif */
 483             /* #endfor */
 484
 485             /* #if not 'Water' in GEOMETRY_J */
 486             /* Load parameters for j particles */
 487             /*     #for J in PARTICLES_ELEC_J */
 488             /*         #if ROUND =='Loop' */
 489             jq{J}              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+{J},charge+jnrB+{J});
 490             /*         #else */
 491             jq{J}              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+{J});
 492             /*         #endif */
 493             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 494             /*             #if ROUND =='Loop' */
 495             isaj{J}            = gmx_fjsp_load_2real_swizzle_v2r8(invsqrta+jnrA+{J},invsqrta+jnrB+{J});
 496             /*             #else */
 497             isaj{J}            = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),invsqrta+jnrA+{J});
 498             /*             #endif */
 499             /*         #endif */
 500             /*     #endfor */
 501             /*     #for J in PARTICLES_VDW_J */
 502             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 503             /*         #if ROUND =='Loop' */
 504             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 505             /*         #endif */
 506             /*     #endfor */
 507             /* #endif */
 508
 509             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 510             /*     #for J in PARTICLES_J */
 511             fjx{J}             = _fjsp_setzero_v2r8();
 512             fjy{J}             = _fjsp_setzero_v2r8();
 513             fjz{J}             = _fjsp_setzero_v2r8();
 514             /*     #endfor */
 515             /* #endif */
 516
 517             /* #for I,J in PAIRS_IJ */
 518
 519             /**************************
 520              * CALCULATE INTERACTIONS *
 521              **************************/
 522
 523             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 524             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 525             if (gmx_fjsp_any_lt_v2r8(rsq{I}{J},rcutoff2))
 526             {
 527                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 528             }
 529             /*         #endif */
 530             /*         #define INNERFLOPS INNERFLOPS+1 */
 531             /*     #endif */
 532
 533             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 534             r{I}{J}              = _fjsp_mul_v2r8(rsq{I}{J},rinv{I}{J});
 535              /*         #define INNERFLOPS INNERFLOPS+1 */
 536             /*     #endif */
 537
 538             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 539             /*     #if not 'Water' in GEOMETRY_J */
 540             /* Compute parameters for interactions between i and j atoms */
 541             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 542             qq{I}{J}             = _fjsp_mul_v2r8(iq{I},jq{J});
 543             /*             #define INNERFLOPS INNERFLOPS+1 */
 544             /*         #endif */
 545             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 546             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset{I}+vdwjidx{J}A,
 547                                          vdwparam+vdwioffset{I}+vdwjidx{J}B,&c6_{I}{J},&c12_{I}{J});
 548
 549             /*             #if 'LJEwald' in KERNEL_VDW */
 550             c6grid_{I}{J}       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset{I}+vdwjidx{J}A,
 551                                                                    vdwgridparam+vdwioffset{I}+vdwjidx{J}B);
 552             /*             #endif */
 553             /*         #endif */
 554             /*     #endif */
 555
 556             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 557             /* Calculate table index by multiplying r with table scale and truncate to integer */
 558             rt               = _fjsp_mul_v2r8(r{I}{J},vftabscale);
 559             itab_tmp         = _fjsp_dtox_v2r8(rt);
 560             vfeps            = _fjsp_sub_v2r8(rt, _fjsp_xtod_v2r8(itab_tmp));
 561             twovfeps         = _fjsp_add_v2r8(vfeps,vfeps);
 562             _fjsp_store_v2r8(&vfconv.simd,itab_tmp);
 563
 564             /*         #define INNERFLOPS INNERFLOPS+4                          */
 565             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW     */
 566             /*             ## 3 tables, 4 data per point: multiply index by 12 */
 567             vfconv.i[0]     *= 12;
 568             vfconv.i[1]     *= 12;
 569             /*         #elif 'Table' in KERNEL_ELEC                             */
 570             /*             ## 1 table, 4 data per point: multiply index by 4   */
 571             vfconv.i[0]     *= 4;
 572             vfconv.i[1]     *= 4;
 573             /*         #elif 'Table' in KERNEL_VDW                              */
 574             /*             ## 2 tables, 4 data per point: multiply index by 8  */
 575             vfconv.i[0]     *= 8;
 576             vfconv.i[1]     *= 8;
 577             /*         #endif                                                   */
 578             /*     #endif */
 579
 580             /*     ## ELECTROSTATIC INTERACTIONS */
 581             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 582
 583             /*         #if KERNEL_ELEC=='Coulomb' */
 584
 585             /* COULOMB ELECTROSTATICS */
 586             velec            = _fjsp_mul_v2r8(qq{I}{J},rinv{I}{J});
 587             /*             #define INNERFLOPS INNERFLOPS+1 */
 588             /*             #if 'Force' in KERNEL_VF */
 589             felec            = _fjsp_mul_v2r8(velec,rinvsq{I}{J});
 590             /*                 #define INNERFLOPS INNERFLOPS+2 */
 591             /*             #endif */
 592
 593             /*         #elif KERNEL_ELEC=='ReactionField' */
 594
 595             /* REACTION-FIELD ELECTROSTATICS */
 596             /*             #if 'Potential' in KERNEL_VF */
 597             velec            = _fjsp_mul_v2r8(qq{I}{J},_fjsp_sub_v2r8(_fjsp_madd_v2r8(krf,rsq{I}{J},rinv{I}{J}),crf));
 598             /*                 #define INNERFLOPS INNERFLOPS+4 */
 599             /*             #endif */
 600             /*             #if 'Force' in KERNEL_VF */
 601             felec            = _fjsp_mul_v2r8(qq{I}{J},_fjsp_msub_v2r8(rinv{I}{J},rinvsq{I}{J},krf2));
 602             /*                 #define INNERFLOPS INNERFLOPS+3 */
 603             /*             #endif */
 604
 605             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 606
 607             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 608             isaprod          = _fjsp_mul_v2r8(isai{I},isaj{J});
 609             gbqqfactor       = _fjsp_neg_v2r8(_fjsp_mul_v2r8(qq{I}{J},_fjsp_mul_v2r8(isaprod,gbinvepsdiff)));
 610             gbscale          = _fjsp_mul_v2r8(isaprod,gbtabscale);
 611             /*             #define INNERFLOPS INNERFLOPS+5 */
 612
 613             /* Calculate generalized born table index - this is a separate table from the normal one,
 614              * but we use the same procedure by multiplying r with scale and truncating to integer.
 615              */
 616             rt               = _fjsp_mul_v2r8(r{I}{J},gbscale);
 617             itab_tmp         = _fjsp_dtox_v2r8(rt);
 618             gbeps            = _fjsp_sub_v2r8(rt,_fjsp_xtod_v2r8(itab_tmp));
 619             _fjsp_store_v2r8(&gbconv.simd,itab_tmp);
 620
 621             Y                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] );
 622             /*             #if ROUND == 'Loop' */
 623             F                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] );
 624             /*             #else */
 625             F                = _fjsp_setzero_v2r8();
 626             /*             #endif */
 627             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 628             G                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[0] +2);
 629             /*             #if ROUND == 'Loop' */
 630             H                = _fjsp_load_v2r8( gbtab + 4*gbconv.i[1] +2);
 631             /*             #else */
 632             H                = _fjsp_setzero_v2r8();
 633             /*             #endif */
 634             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 635             Fp               = _fjsp_madd_v2r8(gbeps,_fjsp_madd_v2r8(gbeps,H,G),F);
 636             VV               = _fjsp_madd_v2r8(gbeps,Fp,Y);
 637             vgb              = _fjsp_mul_v2r8(gbqqfactor,VV);
 638             /*             #define INNERFLOPS INNERFLOPS+10 */
 639
 640             /*             #if 'Force' in KERNEL_VF */
 641             twogbeps         = _fjsp_add_v2r8(gbeps,gbeps);
 642             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twogbeps,H,G),gbeps,Fp);
 643             fgb              = _fjsp_mul_v2r8(gbqqfactor,_fjsp_mul_v2r8(FF,gbscale));
 644             dvdatmp          = _fjsp_mul_v2r8(minushalf,_fjsp_madd_v2r8(fgb,r{I}{J},vgb));
 645             dvdasum          = _fjsp_add_v2r8(dvdasum,dvdatmp);
 646             /*             #if ROUND == 'Loop' */
 647             gmx_fjsp_increment_2real_swizzle_v2r8(dvda+jnrA,dvda+jnrB,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj{J},isaj{J})));
 648             /*             #else */
 649             gmx_fjsp_increment_1real_v2r8(dvda+jnrA,_fjsp_mul_v2r8(dvdatmp,_fjsp_mul_v2r8(isaj{J},isaj{J})));
 650             /*             #endif */
 651             /*                 #define INNERFLOPS INNERFLOPS+13 */
 652             /*             #endif */
 653             velec            = _fjsp_mul_v2r8(qq{I}{J},rinv{I}{J});
 654             /*                 #define INNERFLOPS INNERFLOPS+1 */
 655             /*             #if 'Force' in KERNEL_VF */
 656             felec            = _fjsp_mul_v2r8(_fjsp_msub_v2r8(velec,rinv{I}{J},fgb),rinv{I}{J});
 657             /*                 #define INNERFLOPS INNERFLOPS+3 */
 658             /*             #endif */
 659
 660             /*         #elif KERNEL_ELEC=='Ewald' */
 661             /* EWALD ELECTROSTATICS */
 662
 663             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 664             ewrt             = _fjsp_mul_v2r8(r{I}{J},ewtabscale);
 665             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 666             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 667             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 668
 669             /*             #define INNERFLOPS INNERFLOPS+4 */
 670             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 671             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 672             /*                 #if ROUND == 'Loop' */
 673             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 674             /*                 #else */
 675             ewtabD           = _fjsp_setzero_v2r8();
 676             /*                 #endif */
 677             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 678             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 679             /*                 #if ROUND == 'Loop' */
 680             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 681             /*                 #else */
 682             ewtabFn          = _fjsp_setzero_v2r8();
 683             /*                 #endif */
 684             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 685             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 686             /*                 #define INNERFLOPS INNERFLOPS+2 */
 687             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 688             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 689             velec            = _fjsp_mul_v2r8(qq{I}{J},_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv{I}{J},sh_ewald),velec));
 690             /*                     #define INNERFLOPS INNERFLOPS+7 */
 691             /*                 #else */
 692             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 693             velec            = _fjsp_mul_v2r8(qq{I}{J},_fjsp_sub_v2r8(rinv{I}{J},velec));
 694             /*                     #define INNERFLOPS INNERFLOPS+6 */
 695             /*                 #endif */
 696             /*                 #if 'Force' in KERNEL_VF */
 697             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq{I}{J},rinv{I}{J}),_fjsp_sub_v2r8(rinvsq{I}{J},felec));
 698             /*                      #define INNERFLOPS INNERFLOPS+3 */
 699             /*                 #endif */
 700             /*             #elif KERNEL_VF=='Force' */
 701             /*                 #if ROUND == 'Loop' */
 702             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 703                                          &ewtabF,&ewtabFn);
 704             /*                 #else */
 705             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 706             /*                 #endif */
 707             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 708             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq{I}{J},rinv{I}{J}),_fjsp_sub_v2r8(rinvsq{I}{J},felec));
 709             /*                 #define INNERFLOPS INNERFLOPS+7 */
 710             /*             #endif */
 711
 712             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 713
 714             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 715             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 716             /*             #if ROUND == 'Loop' */
 717             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 718             /*             #else */
 719             F                = _fjsp_setzero_v2r8();
 720             /*             #endif */
 721             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 722             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] +2);
 723             /*             #if ROUND == 'Loop' */
 724             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] +2);
 725             /*             #else */
 726             H                = _fjsp_setzero_v2r8();
 727             /*             #endif */
 728             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 729             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(vfeps,H,G),F);
 730             /*             #define INNERFLOPS INNERFLOPS+4 */
 731             /*             #if 'Potential' in KERNEL_VF */
 732             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 733             velec            = _fjsp_mul_v2r8(qq{I}{J},VV);
 734             /*                 #define INNERFLOPS INNERFLOPS+3 */
 735             /*             #endif */
 736             /*             #if 'Force' in KERNEL_VF */
 737             FF               = _fjsp_madd_v2r8(_fjsp_madd_v2r8(twovfeps,H,G),vfeps,Fp);
 738             felec            = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_mul_v2r8(qq{I}{J},FF),_fjsp_mul_v2r8(vftabscale,rinv{I}{J})));
 739             /*                 #define INNERFLOPS INNERFLOPS+7 */
 740             /*             #endif */
 741             /*         #endif */
 742             /*         ## End of check for electrostatics interaction forms */
 743             /*     #endif */
 744             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 745
 746             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 747
 748             /*         #if KERNEL_VDW=='LennardJones' */
 749
 750             /* LENNARD-JONES DISPERSION/REPULSION */
 751
 752             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 753             /*             #define INNERFLOPS INNERFLOPS+2 */
 754             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 755             vvdw6            = _fjsp_mul_v2r8(c6_{I}{J},rinvsix);
 756             vvdw12           = _fjsp_mul_v2r8(c12_{I}{J},_fjsp_mul_v2r8(rinvsix,rinvsix));
 757             /*                 #define INNERFLOPS INNERFLOPS+3 */
 758             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 759             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_{I}{J},_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 760                                            _fjsp_mul_v2r8(_fjsp_nmsub_v2r8( c6_{I}{J},sh_vdw_invrcut6,vvdw6),one_sixth));
 761             /*                     #define INNERFLOPS INNERFLOPS+8 */
 762             /*                 #else */
 763             vvdw             = _fjsp_msub_v2r8( vvdw12,one_twelfth, _fjsp_mul_v2r8(vvdw6,one_sixth) );
 764             /*                     #define INNERFLOPS INNERFLOPS+3 */
 765             /*                 #endif */
 766             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 767             /*                 #if 'Force' in KERNEL_VF */
 768             fvdw             = _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw12,vvdw6),rinvsq{I}{J});
 769             /*                     #define INNERFLOPS INNERFLOPS+2 */
 770             /*                 #endif */
 771             /*             #elif KERNEL_VF=='Force' */
 772             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 773             fvdw             = _fjsp_mul_v2r8(_fjsp_msub_v2r8(c12_{I}{J},rinvsix,c6_{I}{J}),_fjsp_mul_v2r8(rinvsix,rinvsq{I}{J}));
 774             /*                 #define INNERFLOPS INNERFLOPS+4 */
 775             /*             #endif */
 776
 777             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 778
 779             /* CUBIC SPLINE TABLE DISPERSION */
 780             /*             #if 'Table' in KERNEL_ELEC */
 781             vfconv.i[0]       += 4;
 782             vfconv.i[1]       += 4;
 783             /*             #endif                     */
 784             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] );
 785             /*             #if ROUND == 'Loop' */
 786             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] );
 787             /*             #else */
 788             F                = _fjsp_setzero_v2r8();
 789             /*             #endif */
 790             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 791             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 2 );
 792             /*             #if ROUND == 'Loop' */
 793             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 2 );
 794             /*             #else */
 795             H                = _fjsp_setzero_v2r8();
 796             /*             #endif */
 797             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 798             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 799             /*             #define INNERFLOPS INNERFLOPS+4 */
 800             /*             #if 'Potential' in KERNEL_VF */
 801             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 802             vvdw6            = _fjsp_mul_v2r8(c6_{I}{J},VV);
 803             /*                 #define INNERFLOPS INNERFLOPS+3 */
 804             /*             #endif */
 805             /*             #if 'Force' in KERNEL_VF */
 806             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 807             fvdw6            = _fjsp_mul_v2r8(c6_{I}{J},FF);
 808             /*                 #define INNERFLOPS INNERFLOPS+4 */
 809             /*             #endif */
 810
 811             /* CUBIC SPLINE TABLE REPULSION */
 812             Y                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 4 );
 813             /*             #if ROUND == 'Loop' */
 814             F                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 4 );
 815             /*             #else */
 816             F                = _fjsp_setzero_v2r8();
 817             /*             #endif */
 818             GMX_FJSP_TRANSPOSE2_V2R8(Y,F);
 819             G                = _fjsp_load_v2r8( vftab + vfconv.i[0] + 6 );
 820             /*             #if ROUND == 'Loop' */
 821             H                = _fjsp_load_v2r8( vftab + vfconv.i[1] + 6 );
 822             /*             #else */
 823             H                = _fjsp_setzero_v2r8();
 824             /*             #endif */
 825             GMX_FJSP_TRANSPOSE2_V2R8(G,H);
 826             Fp               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(H,vfeps,G),F);
 827             /*             #define INNERFLOPS INNERFLOPS+4 */
 828             /*             #if 'Potential' in KERNEL_VF */
 829             VV               = _fjsp_madd_v2r8(vfeps,Fp,Y);
 830             vvdw12           = _fjsp_mul_v2r8(c12_{I}{J},VV);
 831             /*                 #define INNERFLOPS INNERFLOPS+3 */
 832             /*             #endif */
 833             /*             #if 'Force' in KERNEL_VF */
 834             FF               = _fjsp_madd_v2r8(vfeps,_fjsp_madd_v2r8(twovfeps,H,G),Fp);
 835             fvdw12           = _fjsp_mul_v2r8(c12_{I}{J},FF);
 836             /*                 #define INNERFLOPS INNERFLOPS+5 */
 837             /*             #endif */
 838             /*             #if 'Potential' in KERNEL_VF */
 839             vvdw             = _fjsp_add_v2r8(vvdw12,vvdw6);
 840             /*                 #define INNERFLOPS INNERFLOPS+1 */
 841             /*             #endif */
 842             /*             #if 'Force' in KERNEL_VF */
 843             fvdw             = _fjsp_neg_v2r8(_fjsp_mul_v2r8(_fjsp_add_v2r8(fvdw6,fvdw12),_fjsp_mul_v2r8(vftabscale,rinv{I}{J})));
 844             /*                 #define INNERFLOPS INNERFLOPS+4 */
 845             /*             #endif */
 846
 847             /*         #elif KERNEL_VDW=='LJEwald' */
 848
 849             /* Analytical LJ-PME */
 850             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 851             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq{I}{J});
 852             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 853             exponent         = gmx_simd_exp_d(ewcljrsq);
 854             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 855             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 856             /*                 #define INNERFLOPS INNERFLOPS+9 */
 857             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 858             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 859             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_{I}{J},_fjsp_sub_v2r8(poly,one),c6_{I}{J}),rinvsix);
 860             vvdw12           = _fjsp_mul_v2r8(c12_{I}{J},_fjsp_mul_v2r8(rinvsix,rinvsix));
 861             /*                 #define INNERFLOPS INNERFLOPS+5 */
 862             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 863             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_{I}{J},_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 864                                _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));
 865             /*                     #define INNERFLOPS INNERFLOPS+7 */
 866             /*                 #else */
 867             vvdw             = _fjsp_msub_v2r8(vvdw12,one_twelfth,_fjsp_mul_v2r8(vvdw6,one_sixth));
 868             /*                 #define INNERFLOPS INNERFLOPS+2 */
 869             /*                 #endif */
 870             /*                  ## Check for force inside potential check, i.e. this means we already did the potential part */
 871             /*                  #if 'Force' in KERNEL_VF */
 872             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 873             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});
 874             /*                 #define INNERFLOPS INNERFLOPS+6 */
 875             /*                  #endif */
 876             /*              #elif KERNEL_VF=='Force' */
 877             /* f6A = 6 * C6grid * (1 - poly) */
 878             f6A              = _fjsp_mul_v2r8(c6grid_{I}{J},_fjsp_sub_v2r8(one,poly));
 879             /* f6B = C6grid * exponent * beta^6 */
 880             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_{I}{J},one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 881             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 882             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});
 883             /*                 #define INNERFLOPS INNERFLOPS+12 */
 884             /*              #endif */
 885             /*         #endif */
 886             /*         ## End of check for vdw interaction forms */
 887             /*     #endif */
 888             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 889
 890             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 891             d                = _fjsp_sub_v2r8(r{I}{J},rswitch);
 892             d                = _fjsp_max_v2r8(d,_fjsp_setzero_v2r8());
 893             d2               = _fjsp_mul_v2r8(d,d);
 894             sw               = _fjsp_add_v2r8(one,_fjsp_mul_v2r8(d2,_fjsp_mul_v2r8(d,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swV5,swV4),swV3))));
 895             /*         #define INNERFLOPS INNERFLOPS+10 */
 896
 897             /*         #if 'Force' in KERNEL_VF */
 898             dsw              = _fjsp_mul_v2r8(d2,_fjsp_madd_v2r8(d,_fjsp_madd_v2r8(d,swF4,swF3),swF2));
 899             /*             #define INNERFLOPS INNERFLOPS+5 */
 900             /*         #endif */
 901
 902             /* Evaluate switch function */
 903             /*         #if 'Force' in KERNEL_VF */
 904             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 905             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 906             felec            = _fjsp_msub_v2r8( felec,sw , _fjsp_mul_v2r8(rinv{I}{J},_fjsp_mul_v2r8(velec,dsw)) );
 907             /*                 #define INNERFLOPS INNERFLOPS+4 */
 908             /*             #endif */
 909             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 910             fvdw             = _fjsp_msub_v2r8( fvdw,sw , _fjsp_mul_v2r8(rinv{I}{J},_fjsp_mul_v2r8(vvdw,dsw)) );
 911             /*                 #define INNERFLOPS INNERFLOPS+4 */
 912             /*             #endif */
 913             /*         #endif */
 914             /*         #if 'Potential' in KERNEL_VF */
 915             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 916             velec            = _fjsp_mul_v2r8(velec,sw);
 917             /*                 #define INNERFLOPS INNERFLOPS+1 */
 918             /*             #endif */
 919             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 920             vvdw             = _fjsp_mul_v2r8(vvdw,sw);
 921             /*                 #define INNERFLOPS INNERFLOPS+1 */
 922             /*             #endif */
 923             /*         #endif */
 924             /*     #endif */
 925             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 926             cutoff_mask      = _fjsp_cmplt_v2r8(rsq{I}{J},rcutoff2);
 927             /*         #define INNERFLOPS INNERFLOPS+1 */
 928             /*     #endif */
 929
 930             /*     #if 'Potential' in KERNEL_VF */
 931             /* Update potential sum for this i atom from the interaction with this j atom. */
 932             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 933             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 934             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 935             /*                 #define INNERFLOPS INNERFLOPS+1 */
 936             /*             #endif                                       */
 937             /*             #if ROUND == 'Epilogue' */
 938             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 939             /*             #endif */
 940             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 941             /*             #define INNERFLOPS INNERFLOPS+1 */
 942             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 943             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 944             vgb              = _fjsp_and_v2r8(vgb,cutoff_mask);
 945             /*                 #define INNERFLOPS INNERFLOPS+1 */
 946             /*             #endif                                       */
 947             /*             #if ROUND == 'Epilogue' */
 948             vgb              = _fjsp_unpacklo_v2r8(vgb,_fjsp_setzero_v2r8());
 949             /*             #endif */
 950             vgbsum           = _fjsp_add_v2r8(vgbsum,vgb);
 951             /*                 #define INNERFLOPS INNERFLOPS+1 */
 952             /*             #endif */
 953             /*         #endif */
 954             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 955             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 956             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 957             /*                 #define INNERFLOPS INNERFLOPS+1 */
 958             /*             #endif                                       */
 959             /*             #if ROUND == 'Epilogue' */
 960             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 961             /*             #endif */
 962             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 963             /*             #define INNERFLOPS INNERFLOPS+1 */
 964             /*         #endif */
 965             /*     #endif */
 966
 967             /*     #if 'Force' in KERNEL_VF */
 968
 969             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 970             fscal            = _fjsp_add_v2r8(felec,fvdw);
 971             /*             #define INNERFLOPS INNERFLOPS+1 */
 972             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 973             fscal            = felec;
 974             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 975             fscal            = fvdw;
 976             /*        #endif */
 977
 978             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 979             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 980             /*                 #define INNERFLOPS INNERFLOPS+1 */
 981             /*             #endif                                       */
 982
 983             /*             #if ROUND == 'Epilogue' */
 984             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 985             /*             #endif */
 986
 987             /* ## Construction of vectorial force built into FMA instructions now */
 988             /* #define INNERFLOPS INNERFLOPS+3      */
 989
 990             /* Update vectorial force */
 991             fix{I}             = _fjsp_madd_v2r8(dx{I}{J},fscal,fix{I});
 992             fiy{I}             = _fjsp_madd_v2r8(dy{I}{J},fscal,fiy{I});
 993             fiz{I}             = _fjsp_madd_v2r8(dz{I}{J},fscal,fiz{I});
 994             /*             #define INNERFLOPS INNERFLOPS+6 */
 995
 996             /* #if GEOMETRY_I == 'Particle'             */
 997             /*     #if ROUND == 'Loop' */
 998             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});
 999             /*     #else */
1000             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx{I}{J},dy{I}{J},dz{I}{J});
1001             /*     #endif */
1002             /*     #define INNERFLOPS INNERFLOPS+3      */
1003             /* #else                                    */
1004             fjx{J}             = _fjsp_madd_v2r8(dx{I}{J},fscal,fjx{J});
1005             fjy{J}             = _fjsp_madd_v2r8(dy{I}{J},fscal,fjy{J});
1006             fjz{J}             = _fjsp_madd_v2r8(dz{I}{J},fscal,fjz{J});
1007             /*     #define INNERFLOPS INNERFLOPS+3      */
1008             /* #endif                                   */
1009
1010             /*     #endif */
1011
1012             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
1013             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
1014             {
1015                 /*     #endif */
1016             }
1017             /*     #endif */
1018             /*    ## End of check for the interaction being outside the cutoff */
1019
1020             /* #endfor */
1021             /* ## End of loop over i-j interaction pairs */
1022
1023             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
1024             /*     #if ROUND == 'Loop' */
1025             gmx_fjsp_decrement_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1026             /*     #else */
1027             gmx_fjsp_decrement_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1028             /*     #endif */
1029             /*     #define INNERFLOPS INNERFLOPS+3      */
1030             /* #elif GEOMETRY_J == 'Water3'             */
1031             /*     #if ROUND == 'Loop' */
1032             gmx_fjsp_decrement_3rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1033             /*     #else */
1034             gmx_fjsp_decrement_3rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1035             /*     #endif */
1036             /*     #define INNERFLOPS INNERFLOPS+9      */
1037             /* #elif GEOMETRY_J == 'Water4'             */
1038             /*     #if 0 in PARTICLES_J                 */
1039             /*         #if ROUND == 'Loop' */
1040             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);
1041             /*         #else */
1042             gmx_fjsp_decrement_4rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1043             /*         #endif */
1044             /*         #define INNERFLOPS INNERFLOPS+12 */
1045             /*     #else                                */
1046             /*         #if ROUND == 'Loop' */
1047             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);
1048             /*         #else */
1049             gmx_fjsp_decrement_3rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1050             /*         #endif */
1051             /*         #define INNERFLOPS INNERFLOPS+9  */
1052             /*     #endif                               */
1053             /* #endif                                   */
1054
1055             /* Inner loop uses {INNERFLOPS} flops */
1056         }
1057
1058         /* #endfor */
1059
1060         /* End of innermost loop */
1061
1062         /* #if 'Force' in KERNEL_VF */
1063         /*     #if GEOMETRY_I == 'Particle'            */
1064         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
1065                                               f+i_coord_offset,fshift+i_shift_offset);
1066         /*         #define OUTERFLOPS OUTERFLOPS+6     */
1067         /*     #elif GEOMETRY_I == 'Water3'            */
1068         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1069                                               f+i_coord_offset,fshift+i_shift_offset);
1070         /*         #define OUTERFLOPS OUTERFLOPS+18    */
1071         /*     #elif GEOMETRY_I == 'Water4'            */
1072         /*         #if 0 in PARTICLES_I                */
1073         gmx_fjsp_update_iforce_4atom_swizzle_v2r8(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1074                                               f+i_coord_offset,fshift+i_shift_offset);
1075         /*             #define OUTERFLOPS OUTERFLOPS+24    */
1076         /*         #else                               */
1077         gmx_fjsp_update_iforce_3atom_swizzle_v2r8(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1078                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1079         /*             #define OUTERFLOPS OUTERFLOPS+18    */
1080         /*         #endif                              */
1081         /*     #endif                                  */
1082         /* #endif                                      */
1083
1084         /* #if 'Potential' in KERNEL_VF */
1085         ggid                        = gid[iidx];
1086         /* Update potential energies */
1087         /*     #if KERNEL_ELEC != 'None' */
1088         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
1089         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1090         /*     #endif */
1091         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
1092         gmx_fjsp_update_1pot_v2r8(vgbsum,kernel_data->energygrp_polarization+ggid);
1093         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1094         /*     #endif */
1095         /*     #if KERNEL_VDW != 'None' */
1096         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
1097         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1098         /*     #endif */
1099         /* #endif */
1100         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1101         dvdasum = _fjsp_mul_v2r8(dvdasum, _fjsp_mul_v2r8(isai{I},isai{I}));
1102         gmx_fjsp_update_1pot_v2r8(dvdasum,dvda+inr);
1103         /*     #endif */
1104
1105         /* Increment number of inner iterations */
1106         inneriter                  += j_index_end - j_index_start;
1107
1108         /* Outer loop uses {OUTERFLOPS} flops */
1109     }
1110
1111     /* Increment number of outer iterations */
1112     outeriter        += nri;
1113
1114     /* Update outer/inner flops */
1115     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1116     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1117     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1118     /* #if GEOMETRY_I == 'Water3'            */
1119     /*     #define ISUFFIX '_W3'             */
1120     /* #elif GEOMETRY_I == 'Water4'          */
1121     /*     #define ISUFFIX '_W4'             */
1122     /* #else                                 */
1123     /*     #define ISUFFIX ''                */
1124     /* #endif                                */
1125     /* #if GEOMETRY_J == 'Water3'            */
1126     /*     #define JSUFFIX 'W3'              */
1127     /* #elif GEOMETRY_J == 'Water4'          */
1128     /*     #define JSUFFIX 'W4'              */
1129     /* #else                                 */
1130     /*     #define JSUFFIX ''                */
1131     /* #endif                                */
1132     /* #if 'PotentialAndForce' in KERNEL_VF  */
1133     /*     #define VFSUFFIX  '_VF'           */
1134     /* #elif 'Potential' in KERNEL_VF        */
1135     /*     #define VFSUFFIX '_V'             */
1136     /* #else                                 */
1137     /*     #define VFSUFFIX '_F'             */
1138     /* #endif                                */
1139
1140     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1141     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1142     /* #elif KERNEL_ELEC != 'None' */
1143     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1144     /* #else */
1145     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1146     /* #endif  */
1147 }