src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_template_sse4_1_single.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_x86_sse4_1_single.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,C,D refer to j loop unrolling done with SSE, e.g. for the four 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,jnrC,jnrD;
 111     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 112     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 113     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 114     real             rcutoff_scalar;
 115     real             *shiftvec,*fshift,*x,*f;
 116     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 117     real             scratch[4*DIM];
 118     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 119     /* #for I in PARTICLES_I */
 120     int              vdwioffset{I};
 121     __m128           ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 122     /* #endfor */
 123     /* #for J in PARTICLES_J */
 124     int              vdwjidx{J}A,vdwjidx{J}B,vdwjidx{J}C,vdwjidx{J}D;
 125     __m128           jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 126     /* #endfor */
 127     /* #for I,J in PAIRS_IJ */
 128     __m128           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};
 129     /* #endfor */
 130     /* #if KERNEL_ELEC != 'None' */
 131     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 132     real             *charge;
 133     /* #endif */
 134     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 135     __m128i          gbitab;
 136     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 137     __m128           minushalf = _mm_set1_ps(-0.5);
 138     real             *invsqrta,*dvda,*gbtab;
 139     /* #endif */
 140     /* #if KERNEL_VDW != 'None' */
 141     int              nvdwtype;
 142     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 143     int              *vdwtype;
 144     real             *vdwparam;
 145     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 146     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 147     /* #endif */
 148     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 149     __m128i          vfitab;
 150     __m128i          ifour       = _mm_set1_epi32(4);
 151     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 152     real             *vftab;
 153     /* #endif */
 154     /* #if 'LJEwald' in KERNEL_VDW */
 155     /* #for I,J in PAIRS_IJ */
 156     __m128           c6grid_{I}{J};
 157     /* #endfor */
 158     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 159     real             *vdwgridparam;
 160     __m128           one_half  = _mm_set1_ps(0.5);
 161     __m128           minus_one = _mm_set1_ps(-1.0);
 162     /* #endif */
 163     /* #if 'Ewald' in KERNEL_ELEC */
 164     __m128i          ewitab;
 165     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 166     real             *ewtab;
 167     /* #endif */
 168     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 169     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 170     real             rswitch_scalar,d_scalar;
 171     /* #endif */
 172     __m128           dummy_mask,cutoff_mask;
 173     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 174     __m128           one     = _mm_set1_ps(1.0);
 175     __m128           two     = _mm_set1_ps(2.0);
 176     x                = xx[0];
 177     f                = ff[0];
 178
 179     nri              = nlist->nri;
 180     iinr             = nlist->iinr;
 181     jindex           = nlist->jindex;
 182     jjnr             = nlist->jjnr;
 183     shiftidx         = nlist->shift;
 184     gid              = nlist->gid;
 185     shiftvec         = fr->shift_vec[0];
 186     fshift           = fr->fshift[0];
 187     /* #if KERNEL_ELEC != 'None' */
 188     facel            = _mm_set1_ps(fr->ic->epsfac);
 189     charge           = mdatoms->chargeA;
 190     /*     #if 'ReactionField' in KERNEL_ELEC */
 191     krf              = _mm_set1_ps(fr->ic->k_rf);
 192     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 193     crf              = _mm_set1_ps(fr->ic->c_rf);
 194     /*     #endif */
 195     /* #endif */
 196     /* #if KERNEL_VDW != 'None' */
 197     nvdwtype         = fr->ntype;
 198     vdwparam         = fr->nbfp;
 199     vdwtype          = mdatoms->typeA;
 200     /* #endif */
 201     /* #if 'LJEwald' in KERNEL_VDW */
 202     vdwgridparam     = fr->ljpme_c6grid;
 203     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 204     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 205     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 206     /* #endif */
 207
 208     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 209     vftab            = kernel_data->table_elec_vdw->data;
 210     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 211     /* #elif 'Table' in KERNEL_ELEC */
 212     vftab            = kernel_data->table_elec->data;
 213     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 214     /* #elif 'Table' in KERNEL_VDW */
 215     vftab            = kernel_data->table_vdw->data;
 216     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 217     /* #endif */
 218
 219     /* #if 'Ewald' in KERNEL_ELEC */
 220     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 221     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 222     ewtab            = fr->ic->tabq_coul_F;
 223     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 224     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 225     /*     #else */
 226     ewtab            = fr->ic->tabq_coul_FDV0;
 227     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 228     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 229      /*     #endif */
 230     /* #endif */
 231
 232     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 233     invsqrta         = fr->invsqrta;
 234     dvda             = fr->dvda;
 235     gbtabscale       = _mm_set1_ps(fr->gbtab->scale);
 236     gbtab            = fr->gbtab->data;
 237     gbinvepsdiff     = _mm_set1_ps((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 238     /* #endif */
 239
 240     /* #if 'Water' in GEOMETRY_I */
 241     /* Setup water-specific parameters */
 242     inr              = nlist->iinr[0];
 243     /*     #for I in PARTICLES_ELEC_I */
 244     iq{I}              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+{I}]));
 245     /*     #endfor */
 246     /*     #for I in PARTICLES_VDW_I */
 247     vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 248     /*     #endfor */
 249     /* #endif */
 250
 251     /* #if 'Water' in GEOMETRY_J */
 252     /*     #for J in PARTICLES_ELEC_J */
 253     jq{J}              = _mm_set1_ps(charge[inr+{J}]);
 254     /*     #endfor */
 255     /*     #for J in PARTICLES_VDW_J */
 256     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 257     /*     #endfor */
 258     /*     #for I,J in PAIRS_IJ */
 259     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 260     qq{I}{J}             = _mm_mul_ps(iq{I},jq{J});
 261     /*         #endif */
 262     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 263     /*             #if 'LJEwald' in KERNEL_VDW */
 264     c6_{I}{J}            = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 265     c12_{I}{J}           = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 266     c6grid_{I}{J}        = _mm_set1_ps(vdwgridparam[vdwioffset{I}+vdwjidx{J}A]);
 267     /*             #else */
 268     c6_{I}{J}            = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 269     c12_{I}{J}           = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 270     /*             #endif */
 271     /*         #endif */
 272     /*     #endfor */
 273     /* #endif */
 274
 275     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 276     /*     #if KERNEL_ELEC!='None' */
 277     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 278     rcutoff_scalar   = fr->ic->rcoulomb;
 279     /*     #else */
 280     rcutoff_scalar   = fr->ic->rvdw;
 281     /*     #endif */
 282     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 283     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 284     /* #endif */
 285
 286     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 287     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 288     rvdw             = _mm_set1_ps(fr->ic->rvdw);
 289     /* #endif */
 290
 291     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 292     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 293     rswitch_scalar   = fr->ic->rcoulomb_switch;
 294     rswitch          = _mm_set1_ps(rswitch_scalar);
 295     /*     #else */
 296     rswitch_scalar   = fr->ic->rvdw_switch;
 297     rswitch          = _mm_set1_ps(rswitch_scalar);
 298     /*     #endif */
 299     /* Setup switch parameters */
 300     d_scalar         = rcutoff_scalar-rswitch_scalar;
 301     d                = _mm_set1_ps(d_scalar);
 302     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 303     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 304     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 305     /*     #if 'Force' in KERNEL_VF */
 306     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 307     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 308     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 309     /*     #endif */
 310     /* #endif */
 311
 312     /* Avoid stupid compiler warnings */
 313     jnrA = jnrB = jnrC = jnrD = 0;
 314     j_coord_offsetA = 0;
 315     j_coord_offsetB = 0;
 316     j_coord_offsetC = 0;
 317     j_coord_offsetD = 0;
 318
 319     /* ## Keep track of the floating point operations we issue for reporting! */
 320     /* #define OUTERFLOPS 0 */
 321     outeriter        = 0;
 322     inneriter        = 0;
 323
 324     for(iidx=0;iidx<4*DIM;iidx++)
 325     {
 326         scratch[iidx] = 0.0;
 327     }
 328
 329     /* Start outer loop over neighborlists */
 330     for(iidx=0; iidx<nri; iidx++)
 331     {
 332         /* Load shift vector for this list */
 333         i_shift_offset   = DIM*shiftidx[iidx];
 334
 335         /* Load limits for loop over neighbors */
 336         j_index_start    = jindex[iidx];
 337         j_index_end      = jindex[iidx+1];
 338
 339         /* Get outer coordinate index */
 340         inr              = iinr[iidx];
 341         i_coord_offset   = DIM*inr;
 342
 343         /* Load i particle coords and add shift vector */
 344         /* #if GEOMETRY_I == 'Particle' */
 345         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 346         /* #elif GEOMETRY_I == 'Water3' */
 347         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 348                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 349         /* #elif GEOMETRY_I == 'Water4' */
 350         /*     #if 0 in PARTICLES_I                 */
 351         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 352                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 353         /*     #else                                */
 354         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 355                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 356         /*     #endif                               */
 357         /* #endif                                   */
 358
 359         /* #if 'Force' in KERNEL_VF */
 360         /*     #for I in PARTICLES_I */
 361         fix{I}             = _mm_setzero_ps();
 362         fiy{I}             = _mm_setzero_ps();
 363         fiz{I}             = _mm_setzero_ps();
 364         /*     #endfor */
 365         /* #endif */
 366
 367         /* ## For water we already preloaded parameters at the start of the kernel */
 368         /* #if not 'Water' in GEOMETRY_I */
 369         /* Load parameters for i particles */
 370         /*     #for I in PARTICLES_ELEC_I */
 371         iq{I}              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+{I}));
 372         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 373         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 374         isai{I}            = _mm_load1_ps(invsqrta+inr+{I});
 375         /*         #endif */
 376         /*     #endfor */
 377         /*     #for I in PARTICLES_VDW_I */
 378         vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 379         /*     #endfor */
 380         /* #endif */
 381
 382         /* #if 'Potential' in KERNEL_VF */
 383         /* Reset potential sums */
 384         /*     #if KERNEL_ELEC != 'None' */
 385         velecsum         = _mm_setzero_ps();
 386         /*     #endif */
 387         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 388         vgbsum           = _mm_setzero_ps();
 389         /*     #endif */
 390         /*     #if KERNEL_VDW != 'None' */
 391         vvdwsum          = _mm_setzero_ps();
 392         /*     #endif */
 393         /* #endif */
 394         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 395         dvdasum          = _mm_setzero_ps();
 396         /*     #endif */
 397
 398         /* #for ROUND in ['Loop','Epilogue'] */
 399
 400         /* #if ROUND =='Loop' */
 401         /* Start inner kernel loop */
 402         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 403         {
 404         /* ## First round is normal loop (next statement resets indentation) */
 405         /*     #if 0 */
 406         }
 407         /*     #endif */
 408         /* #else */
 409         if(jidx<j_index_end)
 410         {
 411         /* ## Second round is epilogue */
 412         /* #endif */
 413         /* #define INNERFLOPS 0 */
 414
 415             /* Get j neighbor index, and coordinate index */
 416             /* #if ROUND =='Loop' */
 417             jnrA             = jjnr[jidx];
 418             jnrB             = jjnr[jidx+1];
 419             jnrC             = jjnr[jidx+2];
 420             jnrD             = jjnr[jidx+3];
 421             /* #else */
 422             jnrlistA         = jjnr[jidx];
 423             jnrlistB         = jjnr[jidx+1];
 424             jnrlistC         = jjnr[jidx+2];
 425             jnrlistD         = jjnr[jidx+3];
 426             /* Sign of each element will be negative for non-real atoms.
 427              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 428              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 429              */
 430             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 431             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 432             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 433             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 434             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 435             /* #endif */
 436             j_coord_offsetA  = DIM*jnrA;
 437             j_coord_offsetB  = DIM*jnrB;
 438             j_coord_offsetC  = DIM*jnrC;
 439             j_coord_offsetD  = DIM*jnrD;
 440
 441             /* load j atom coordinates */
 442             /* #if GEOMETRY_J == 'Particle'             */
 443             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 444                                               x+j_coord_offsetC,x+j_coord_offsetD,
 445                                               &jx0,&jy0,&jz0);
 446             /* #elif GEOMETRY_J == 'Water3'             */
 447             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 448                                               x+j_coord_offsetC,x+j_coord_offsetD,
 449                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 450             /* #elif GEOMETRY_J == 'Water4'             */
 451             /*     #if 0 in PARTICLES_J                 */
 452             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 453                                               x+j_coord_offsetC,x+j_coord_offsetD,
 454                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 455                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 456             /*     #else                                */
 457             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 458                                               x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 459                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 460             /*     #endif                               */
 461             /* #endif                                   */
 462
 463             /* Calculate displacement vector */
 464             /* #for I,J in PAIRS_IJ */
 465             dx{I}{J}             = _mm_sub_ps(ix{I},jx{J});
 466             dy{I}{J}             = _mm_sub_ps(iy{I},jy{J});
 467             dz{I}{J}             = _mm_sub_ps(iz{I},jz{J});
 468             /*     #define INNERFLOPS INNERFLOPS+3 */
 469             /* #endfor */
 470
 471             /* Calculate squared distance and things based on it */
 472             /* #for I,J in PAIRS_IJ */
 473             rsq{I}{J}            = gmx_mm_calc_rsq_ps(dx{I}{J},dy{I}{J},dz{I}{J});
 474             /*     #define INNERFLOPS INNERFLOPS+5 */
 475             /* #endfor */
 476
 477             /* #for I,J in PAIRS_IJ */
 478             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 479             rinv{I}{J}           = sse41_invsqrt_f(rsq{I}{J});
 480             /*         #define INNERFLOPS INNERFLOPS+5 */
 481             /*     #endif */
 482             /* #endfor */
 483
 484             /* #for I,J in PAIRS_IJ */
 485             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 486             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 487             rinvsq{I}{J}         = sse41_inv_f(rsq{I}{J});
 488             /*             #define INNERFLOPS INNERFLOPS+4 */
 489             /*         #else */
 490             rinvsq{I}{J}         = _mm_mul_ps(rinv{I}{J},rinv{I}{J});
 491             /*             #define INNERFLOPS INNERFLOPS+1 */
 492             /*         #endif */
 493             /*     #endif */
 494             /* #endfor */
 495
 496             /* #if not 'Water' in GEOMETRY_J */
 497             /* Load parameters for j particles */
 498             /*     #for J in PARTICLES_ELEC_J */
 499             jq{J}              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+{J},charge+jnrB+{J},
 500                                                               charge+jnrC+{J},charge+jnrD+{J});
 501             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 502             isaj{J}            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+{J},invsqrta+jnrB+{J},
 503                                                               invsqrta+jnrC+{J},invsqrta+jnrD+{J});
 504             /*         #endif */
 505             /*     #endfor */
 506             /*     #for J in PARTICLES_VDW_J */
 507             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 508             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 509             vdwjidx{J}C        = 2*vdwtype[jnrC+{J}];
 510             vdwjidx{J}D        = 2*vdwtype[jnrD+{J}];
 511             /*     #endfor */
 512             /* #endif */
 513
 514             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 515             /*     #for J in PARTICLES_J */
 516             fjx{J}             = _mm_setzero_ps();
 517             fjy{J}             = _mm_setzero_ps();
 518             fjz{J}             = _mm_setzero_ps();
 519             /*     #endfor */
 520             /* #endif */
 521
 522             /* #for I,J in PAIRS_IJ */
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 529             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 530             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 531             if (gmx_mm_any_lt(rsq{I}{J},rcutoff2))
 532             {
 533                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 534             }
 535             /*         #endif */
 536             /*         #define INNERFLOPS INNERFLOPS+1 */
 537             /*     #endif */
 538
 539             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 540             r{I}{J}              = _mm_mul_ps(rsq{I}{J},rinv{I}{J});
 541             /*         #if ROUND == 'Epilogue' */
 542             r{I}{J}              = _mm_andnot_ps(dummy_mask,r{I}{J});
 543             /*             #define INNERFLOPS INNERFLOPS+1 */
 544             /*         #endif */
 545             /*         #define INNERFLOPS INNERFLOPS+1 */
 546             /*     #endif */
 547
 548             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 549             /*     #if not 'Water' in GEOMETRY_J */
 550             /* Compute parameters for interactions between i and j atoms */
 551             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 552             qq{I}{J}             = _mm_mul_ps(iq{I},jq{J});
 553             /*             #define INNERFLOPS INNERFLOPS+1 */
 554             /*         #endif */
 555             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 556             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset{I}+vdwjidx{J}A,
 557                                          vdwparam+vdwioffset{I}+vdwjidx{J}B,
 558                                          vdwparam+vdwioffset{I}+vdwjidx{J}C,
 559                                          vdwparam+vdwioffset{I}+vdwjidx{J}D,
 560                                          &c6_{I}{J},&c12_{I}{J});
 561
 562             /*         #if 'LJEwald' in KERNEL_VDW */
 563             c6grid_{I}{J}       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset{I}+vdwjidx{J}A,
 564                                                                vdwgridparam+vdwioffset{I}+vdwjidx{J}B,
 565                                                                vdwgridparam+vdwioffset{I}+vdwjidx{J}C,
 566                                                                vdwgridparam+vdwioffset{I}+vdwjidx{J}D);
 567             /*          #endif */
 568             /*         #endif */
 569             /*     #endif */
 570
 571             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 572             /* Calculate table index by multiplying r with table scale and truncate to integer */
 573             rt               = _mm_mul_ps(r{I}{J},vftabscale);
 574             vfitab           = _mm_cvttps_epi32(rt);
 575             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 576             /*         #define INNERFLOPS INNERFLOPS+4                          */
 577             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW     */
 578             /*             ## 3 tables, 4 bytes per point: multiply index by 12 */
 579             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 580             /*         #elif 'Table' in KERNEL_ELEC                             */
 581             /*             ## 1 table, 4 bytes per point: multiply index by 4   */
 582             vfitab           = _mm_slli_epi32(vfitab,2);
 583             /*         #elif 'Table' in KERNEL_VDW                              */
 584             /*             ## 2 tables, 4 bytes per point: multiply index by 8  */
 585             vfitab           = _mm_slli_epi32(vfitab,3);
 586             /*         #endif                                                   */
 587             /*     #endif */
 588
 589             /*     ## ELECTROSTATIC INTERACTIONS */
 590             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 591
 592             /*         #if KERNEL_ELEC=='Coulomb' */
 593
 594             /* COULOMB ELECTROSTATICS */
 595             velec            = _mm_mul_ps(qq{I}{J},rinv{I}{J});
 596             /*             #define INNERFLOPS INNERFLOPS+1 */
 597             /*             #if 'Force' in KERNEL_VF */
 598             felec            = _mm_mul_ps(velec,rinvsq{I}{J});
 599             /*                 #define INNERFLOPS INNERFLOPS+2 */
 600             /*             #endif */
 601
 602             /*         #elif KERNEL_ELEC=='ReactionField' */
 603
 604             /* REACTION-FIELD ELECTROSTATICS */
 605             /*             #if 'Potential' in KERNEL_VF */
 606             velec            = _mm_mul_ps(qq{I}{J},_mm_sub_ps(_mm_add_ps(rinv{I}{J},_mm_mul_ps(krf,rsq{I}{J})),crf));
 607             /*                 #define INNERFLOPS INNERFLOPS+4 */
 608             /*             #endif */
 609             /*             #if 'Force' in KERNEL_VF */
 610             felec            = _mm_mul_ps(qq{I}{J},_mm_sub_ps(_mm_mul_ps(rinv{I}{J},rinvsq{I}{J}),krf2));
 611             /*                 #define INNERFLOPS INNERFLOPS+3 */
 612             /*             #endif */
 613
 614             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 615
 616             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 617             isaprod          = _mm_mul_ps(isai{I},isaj{J});
 618             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq{I}{J},_mm_mul_ps(isaprod,gbinvepsdiff)));
 619             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 620             /*             #define INNERFLOPS INNERFLOPS+5 */
 621
 622             /* Calculate generalized born table index - this is a separate table from the normal one,
 623              * but we use the same procedure by multiplying r with scale and truncating to integer.
 624              */
 625             rt               = _mm_mul_ps(r{I}{J},gbscale);
 626             gbitab           = _mm_cvttps_epi32(rt);
 627             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 628             gbitab           = _mm_slli_epi32(gbitab,2);
 629             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 630             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 631             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 632             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 633             _MM_TRANSPOSE4_PS(Y,F,G,H);
 634             Heps             = _mm_mul_ps(gbeps,H);
 635             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 636             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 637             vgb              = _mm_mul_ps(gbqqfactor,VV);
 638             /*             #define INNERFLOPS INNERFLOPS+10 */
 639
 640             /*             #if 'Force' in KERNEL_VF */
 641             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 642             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 643             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r{I}{J})));
 644             /*                 #if ROUND == 'Epilogue' */
 645             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 646             /*                 #endif */
 647             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 648             /*                 #if ROUND == 'Loop' */
 649             fjptrA           = dvda+jnrA;
 650             fjptrB           = dvda+jnrB;
 651             fjptrC           = dvda+jnrC;
 652             fjptrD           = dvda+jnrD;
 653             /*                 #else */
 654             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 655             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 656             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 657             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 658             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 659             /*                 #endif */
 660             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj{J},isaj{J})));
 661             /*                 #define INNERFLOPS INNERFLOPS+13 */
 662             /*             #endif */
 663             velec            = _mm_mul_ps(qq{I}{J},rinv{I}{J});
 664             /*                 #define INNERFLOPS INNERFLOPS+1 */
 665             /*             #if 'Force' in KERNEL_VF */
 666             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv{I}{J}),fgb),rinv{I}{J});
 667             /*                 #define INNERFLOPS INNERFLOPS+3 */
 668             /*             #endif */
 669
 670             /*         #elif KERNEL_ELEC=='Ewald' */
 671             /* EWALD ELECTROSTATICS */
 672
 673             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 674             ewrt             = _mm_mul_ps(r{I}{J},ewtabscale);
 675             ewitab           = _mm_cvttps_epi32(ewrt);
 676             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 677             /*             #define INNERFLOPS INNERFLOPS+4 */
 678             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 679             ewitab           = _mm_slli_epi32(ewitab,2);
 680             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 681             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 682             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 683             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 684             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 685             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 686             /*                 #define INNERFLOPS INNERFLOPS+2 */
 687             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 688             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 689             velec            = _mm_mul_ps(qq{I}{J},_mm_sub_ps(_mm_sub_ps(rinv{I}{J},sh_ewald),velec));
 690             /*                     #define INNERFLOPS INNERFLOPS+7 */
 691             /*                 #else */
 692             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 693             velec            = _mm_mul_ps(qq{I}{J},_mm_sub_ps(rinv{I}{J},velec));
 694             /*                     #define INNERFLOPS INNERFLOPS+6 */
 695             /*                 #endif */
 696             /*                 #if 'Force' in KERNEL_VF */
 697             felec            = _mm_mul_ps(_mm_mul_ps(qq{I}{J},rinv{I}{J}),_mm_sub_ps(rinvsq{I}{J},felec));
 698             /*                      #define INNERFLOPS INNERFLOPS+3 */
 699             /*                 #endif */
 700             /*             #elif KERNEL_VF=='Force' */
 701             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 702                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 703                                          &ewtabF,&ewtabFn);
 704             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 705             felec            = _mm_mul_ps(_mm_mul_ps(qq{I}{J},rinv{I}{J}),_mm_sub_ps(rinvsq{I}{J},felec));
 706             /*                 #define INNERFLOPS INNERFLOPS+7 */
 707             /*             #endif */
 708
 709             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 710
 711             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 712             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 713             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 714             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 715             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 716             _MM_TRANSPOSE4_PS(Y,F,G,H);
 717             Heps             = _mm_mul_ps(vfeps,H);
 718             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 719             /*             #define INNERFLOPS INNERFLOPS+4 */
 720             /*             #if 'Potential' in KERNEL_VF */
 721             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 722             velec            = _mm_mul_ps(qq{I}{J},VV);
 723             /*                 #define INNERFLOPS INNERFLOPS+3 */
 724             /*             #endif */
 725             /*             #if 'Force' in KERNEL_VF */
 726             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 727             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq{I}{J},FF),_mm_mul_ps(vftabscale,rinv{I}{J})));
 728             /*                 #define INNERFLOPS INNERFLOPS+7 */
 729             /*             #endif */
 730             /*         #endif */
 731             /*         ## End of check for electrostatics interaction forms */
 732             /*     #endif */
 733             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 734
 735             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 736
 737             /*         #if KERNEL_VDW=='LennardJones' */
 738
 739             /* LENNARD-JONES DISPERSION/REPULSION */
 740
 741             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 742             /*             #define INNERFLOPS INNERFLOPS+2 */
 743             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 744             vvdw6            = _mm_mul_ps(c6_{I}{J},rinvsix);
 745             vvdw12           = _mm_mul_ps(c12_{I}{J},_mm_mul_ps(rinvsix,rinvsix));
 746             /*                 #define INNERFLOPS INNERFLOPS+3 */
 747             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 748             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_{I}{J},_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 749                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
 750             /*                     #define INNERFLOPS INNERFLOPS+8 */
 751             /*                 #else */
 752             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 753             /*                     #define INNERFLOPS INNERFLOPS+3 */
 754             /*                 #endif */
 755             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 756             /*                 #if 'Force' in KERNEL_VF */
 757             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq{I}{J});
 758             /*                     #define INNERFLOPS INNERFLOPS+2 */
 759             /*                 #endif */
 760             /*             #elif KERNEL_VF=='Force' */
 761             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 762             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_{I}{J},rinvsix),c6_{I}{J}),_mm_mul_ps(rinvsix,rinvsq{I}{J}));
 763             /*                 #define INNERFLOPS INNERFLOPS+4 */
 764             /*             #endif */
 765
 766             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 767
 768             /* CUBIC SPLINE TABLE DISPERSION */
 769             /*             #if 'Table' in KERNEL_ELEC */
 770             vfitab           = _mm_add_epi32(vfitab,ifour);
 771             /*             #endif                     */
 772             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 773             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 774             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 775             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 776             _MM_TRANSPOSE4_PS(Y,F,G,H);
 777             Heps             = _mm_mul_ps(vfeps,H);
 778             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 779             /*             #define INNERFLOPS INNERFLOPS+4 */
 780             /*             #if 'Potential' in KERNEL_VF */
 781             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 782             vvdw6            = _mm_mul_ps(c6_{I}{J},VV);
 783             /*                 #define INNERFLOPS INNERFLOPS+3 */
 784             /*             #endif */
 785             /*             #if 'Force' in KERNEL_VF */
 786             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 787             fvdw6            = _mm_mul_ps(c6_{I}{J},FF);
 788             /*                 #define INNERFLOPS INNERFLOPS+4 */
 789             /*             #endif */
 790
 791             /* CUBIC SPLINE TABLE REPULSION */
 792             vfitab           = _mm_add_epi32(vfitab,ifour);
 793             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 794             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 795             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 796             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 797             _MM_TRANSPOSE4_PS(Y,F,G,H);
 798             Heps             = _mm_mul_ps(vfeps,H);
 799             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 800             /*             #define INNERFLOPS INNERFLOPS+4 */
 801             /*             #if 'Potential' in KERNEL_VF */
 802             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 803             vvdw12           = _mm_mul_ps(c12_{I}{J},VV);
 804             /*                 #define INNERFLOPS INNERFLOPS+3 */
 805             /*             #endif */
 806             /*             #if 'Force' in KERNEL_VF */
 807             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 808             fvdw12           = _mm_mul_ps(c12_{I}{J},FF);
 809             /*                 #define INNERFLOPS INNERFLOPS+5 */
 810             /*             #endif */
 811             /*             #if 'Potential' in KERNEL_VF */
 812             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 813             /*                 #define INNERFLOPS INNERFLOPS+1 */
 814             /*             #endif */
 815             /*             #if 'Force' in KERNEL_VF */
 816             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv{I}{J})));
 817             /*                 #define INNERFLOPS INNERFLOPS+4 */
 818             /*             #endif */
 819
 820             /*         #elif KERNEL_VDW=='LJEwald' */
 821
 822             /* Analytical LJ-PME */
 823             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 824             ewcljrsq         = _mm_mul_ps(ewclj2,rsq{I}{J});
 825             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 826             exponent         = sse41_exp_f(ewcljrsq);
 827             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 828             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 829             /*             #define INNERFLOPS INNERFLOPS+11 */
 830             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 831             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 832             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_{I}{J},_mm_mul_ps(c6grid_{I}{J},_mm_sub_ps(one,poly))),rinvsix);
 833             vvdw12           = _mm_mul_ps(c12_{I}{J},_mm_mul_ps(rinvsix,rinvsix));
 834             /*                 #define INNERFLOPS INNERFLOPS+6 */
 835             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 836             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_{I}{J},_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))),one_twelfth),
 837                                _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_add_ps(_mm_mul_ps(c6_{I}{J},sh_vdw_invrcut6),_mm_mul_ps(c6grid_{I}{J},sh_lj_ewald))),one_sixth));
 838             /*                 #define INNERFLOPS INNERFLOPS+10 */
 839             /*                 #else */
 840             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 841             /*                 #define INNERFLOPS INNERFLOPS+3 */
 842             /*                 #endif */
 843             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 844             /*                 #if 'Force' in KERNEL_VF */
 845             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 846             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_{I}{J},one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq{I}{J});
 847             /*                 #define INNERFLOPS INNERFLOPS+6 */
 848             /*                 #endif */
 849             /*             #elif KERNEL_VF=='Force' */
 850             /* f6A = 6 * C6grid * (1 - poly) */
 851             f6A              = _mm_mul_ps(c6grid_{I}{J},_mm_sub_ps(one,poly));
 852             /* f6B = C6grid * exponent * beta^6 */
 853             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_{I}{J},one_sixth),_mm_mul_ps(exponent,ewclj6));
 854             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 855             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_{I}{J},rinvsix),_mm_sub_ps(c6_{I}{J},f6A)),rinvsix),f6B),rinvsq{I}{J});
 856             /*                 #define INNERFLOPS INNERFLOPS+11 */
 857             /*             #endif */
 858             /*         #endif */
 859             /*         ## End of check for vdw interaction forms */
 860             /*     #endif */
 861             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 862
 863             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 864             d                = _mm_sub_ps(r{I}{J},rswitch);
 865             d                = _mm_max_ps(d,_mm_setzero_ps());
 866             d2               = _mm_mul_ps(d,d);
 867             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 868             /*         #define INNERFLOPS INNERFLOPS+10 */
 869
 870             /*         #if 'Force' in KERNEL_VF */
 871             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 872             /*             #define INNERFLOPS INNERFLOPS+5 */
 873             /*         #endif */
 874
 875             /* Evaluate switch function */
 876             /*         #if 'Force' in KERNEL_VF */
 877             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 878             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 879             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv{I}{J},_mm_mul_ps(velec,dsw)) );
 880             /*                 #define INNERFLOPS INNERFLOPS+4 */
 881             /*             #endif */
 882             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 883             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv{I}{J},_mm_mul_ps(vvdw,dsw)) );
 884             /*                 #define INNERFLOPS INNERFLOPS+4 */
 885             /*             #endif */
 886             /*         #endif */
 887             /*         #if 'Potential' in KERNEL_VF */
 888             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 889             velec            = _mm_mul_ps(velec,sw);
 890             /*                 #define INNERFLOPS INNERFLOPS+1 */
 891             /*             #endif */
 892             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 893             vvdw             = _mm_mul_ps(vvdw,sw);
 894             /*                 #define INNERFLOPS INNERFLOPS+1 */
 895             /*             #endif */
 896             /*         #endif */
 897             /*     #endif */
 898             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 899             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 900             cutoff_mask      = _mm_cmplt_ps(rsq{I}{J},rcutoff2);
 901             /*         #define INNERFLOPS INNERFLOPS+1 */
 902             /*     #endif */
 903
 904             /*     #if 'Potential' in KERNEL_VF */
 905             /* Update potential sum for this i atom from the interaction with this j atom. */
 906             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 907             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 908             velec            = _mm_and_ps(velec,cutoff_mask);
 909             /*                 #define INNERFLOPS INNERFLOPS+1 */
 910             /*             #endif                                       */
 911             /*             #if ROUND == 'Epilogue' */
 912             velec            = _mm_andnot_ps(dummy_mask,velec);
 913             /*             #endif */
 914             velecsum         = _mm_add_ps(velecsum,velec);
 915             /*             #define INNERFLOPS INNERFLOPS+1 */
 916             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 917             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 918             vgb              = _mm_and_ps(vgb,cutoff_mask);
 919             /*                 #define INNERFLOPS INNERFLOPS+1 */
 920             /*             #endif                                       */
 921             /*             #if ROUND == 'Epilogue' */
 922             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 923             /*             #endif */
 924             vgbsum           = _mm_add_ps(vgbsum,vgb);
 925             /*                 #define INNERFLOPS INNERFLOPS+1 */
 926             /*             #endif */
 927             /*         #endif */
 928             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 929             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 930             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 931             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 932             /*                 #define INNERFLOPS INNERFLOPS+1 */
 933             /*             #endif                                       */
 934             /*             #if ROUND == 'Epilogue' */
 935             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 936             /*             #endif */
 937             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 938             /*             #define INNERFLOPS INNERFLOPS+1 */
 939             /*         #endif */
 940             /*     #endif */
 941
 942             /*     #if 'Force' in KERNEL_VF */
 943
 944             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 945             fscal            = _mm_add_ps(felec,fvdw);
 946             /*             #define INNERFLOPS INNERFLOPS+1 */
 947             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 948             fscal            = felec;
 949             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 950             fscal            = fvdw;
 951             /*        #endif */
 952
 953             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 954             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 955             fscal            = _mm_and_ps(fscal,cutoff_mask);
 956             /*                 #define INNERFLOPS INNERFLOPS+1 */
 957             /*             #endif                                       */
 958
 959             /*             #if ROUND == 'Epilogue' */
 960             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 961             /*             #endif */
 962
 963             /* Calculate temporary vectorial force */
 964             tx               = _mm_mul_ps(fscal,dx{I}{J});
 965             ty               = _mm_mul_ps(fscal,dy{I}{J});
 966             tz               = _mm_mul_ps(fscal,dz{I}{J});
 967
 968             /* Update vectorial force */
 969             fix{I}             = _mm_add_ps(fix{I},tx);
 970             fiy{I}             = _mm_add_ps(fiy{I},ty);
 971             fiz{I}             = _mm_add_ps(fiz{I},tz);
 972             /*             #define INNERFLOPS INNERFLOPS+6 */
 973
 974             /* #if GEOMETRY_I == 'Particle'             */
 975             /*     #if ROUND == 'Loop' */
 976             fjptrA             = f+j_coord_offsetA;
 977             fjptrB             = f+j_coord_offsetB;
 978             fjptrC             = f+j_coord_offsetC;
 979             fjptrD             = f+j_coord_offsetD;
 980             /*     #else */
 981             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 982             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 983             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 984             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 985             /*     #endif */
 986             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 987             /*     #define INNERFLOPS INNERFLOPS+3      */
 988             /* #else                                    */
 989             fjx{J}             = _mm_add_ps(fjx{J},tx);
 990             fjy{J}             = _mm_add_ps(fjy{J},ty);
 991             fjz{J}             = _mm_add_ps(fjz{J},tz);
 992             /*     #define INNERFLOPS INNERFLOPS+3      */
 993             /* #endif                                   */
 994
 995             /*     #endif */
 996
 997             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 998             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 999             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
1000             {
1001                 /*     #endif */
1002             }
1003             /*     #endif */
1004             /*    ## End of check for the interaction being outside the cutoff */
1005
1006             /* #endfor */
1007             /* ## End of loop over i-j interaction pairs */
1008
1009             /* #if GEOMETRY_I != 'Particle' */
1010             /*     #if ROUND == 'Loop' */
1011             fjptrA             = f+j_coord_offsetA;
1012             fjptrB             = f+j_coord_offsetB;
1013             fjptrC             = f+j_coord_offsetC;
1014             fjptrD             = f+j_coord_offsetD;
1015             /*     #else */
1016             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1017             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1018             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1019             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1020             /*     #endif */
1021             /* #endif */
1022
1023             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
1024             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1025             /* #elif GEOMETRY_J == 'Water3'               */
1026             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1027                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1028             /*     #define INNERFLOPS INNERFLOPS+9      */
1029             /* #elif GEOMETRY_J == 'Water4'             */
1030             /*     #if 0 in PARTICLES_J                 */
1031             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1032                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1033                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1034             /*     #define INNERFLOPS INNERFLOPS+12     */
1035             /*     #else                                */
1036             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1037                                                    fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1038             /*     #define INNERFLOPS INNERFLOPS+9      */
1039             /*     #endif                               */
1040             /* #endif                                   */
1041
1042             /* Inner loop uses {INNERFLOPS} flops */
1043         }
1044
1045         /* #endfor */
1046
1047         /* End of innermost loop */
1048
1049         /* #if 'Force' in KERNEL_VF */
1050         /*     #if GEOMETRY_I == 'Particle'            */
1051         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1052                                               f+i_coord_offset,fshift+i_shift_offset);
1053         /*         #define OUTERFLOPS OUTERFLOPS+6     */
1054         /*     #elif GEOMETRY_I == 'Water3'            */
1055         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1056                                               f+i_coord_offset,fshift+i_shift_offset);
1057         /*         #define OUTERFLOPS OUTERFLOPS+18    */
1058         /*     #elif GEOMETRY_I == 'Water4'            */
1059         /*         #if 0 in PARTICLES_I                */
1060         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1061                                               f+i_coord_offset,fshift+i_shift_offset);
1062         /*             #define OUTERFLOPS OUTERFLOPS+24    */
1063         /*         #else                               */
1064         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1065                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1066         /*             #define OUTERFLOPS OUTERFLOPS+18    */
1067         /*         #endif                              */
1068         /*     #endif                                  */
1069         /* #endif                                      */
1070
1071         /* #if 'Potential' in KERNEL_VF */
1072         ggid                        = gid[iidx];
1073         /* Update potential energies */
1074         /*     #if KERNEL_ELEC != 'None' */
1075         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1076         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1077         /*     #endif */
1078         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
1079         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
1080         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1081         /*     #endif */
1082         /*     #if KERNEL_VDW != 'None' */
1083         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1084         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1085         /*     #endif */
1086         /* #endif */
1087         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1088         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai{I},isai{I}));
1089         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
1090         /*     #endif */
1091
1092         /* Increment number of inner iterations */
1093         inneriter                  += j_index_end - j_index_start;
1094
1095         /* Outer loop uses {OUTERFLOPS} flops */
1096     }
1097
1098     /* Increment number of outer iterations */
1099     outeriter        += nri;
1100
1101     /* Update outer/inner flops */
1102     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1103     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1104     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1105     /* #if GEOMETRY_I == 'Water3'            */
1106     /*     #define ISUFFIX '_W3'             */
1107     /* #elif GEOMETRY_I == 'Water4'          */
1108     /*     #define ISUFFIX '_W4'             */
1109     /* #else                                 */
1110     /*     #define ISUFFIX ''                */
1111     /* #endif                                */
1112     /* #if GEOMETRY_J == 'Water3'            */
1113     /*     #define JSUFFIX 'W3'              */
1114     /* #elif GEOMETRY_J == 'Water4'          */
1115     /*     #define JSUFFIX 'W4'              */
1116     /* #else                                 */
1117     /*     #define JSUFFIX ''                */
1118     /* #endif                                */
1119     /* #if 'PotentialAndForce' in KERNEL_VF  */
1120     /*     #define VFSUFFIX  '_VF'           */
1121     /* #elif 'Potential' in KERNEL_VF        */
1122     /*     #define VFSUFFIX '_V'             */
1123     /* #else                                 */
1124     /*     #define VFSUFFIX '_F'             */
1125     /* #endif                                */
1126
1127     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1128     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1129     /* #elif KERNEL_ELEC != 'None' */
1130     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1131     /* #else */
1132     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1133     /* #endif  */
1134 }