src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_template_sse2_single.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 "vec.h"
  48 #include "nrnb.h"
  49
  50 #include "gromacs/simd/math_x86_sse2_single.h"
  51 #include "kernelutil_x86_sse2_single.h"
  52 /* #endif */
  53
  54 /* ## List of variables set by the generating script:                                    */
  55 /* ##                                                                                    */
  56 /* ## Setttings that apply to the entire kernel:                                         */
  57 /* ## KERNEL_ELEC:           String, choice for electrostatic interactions               */
  58 /* ## KERNEL_VDW:            String, choice for van der Waals interactions               */
  59 /* ## KERNEL_NAME:           String, name of this kernel                                 */
  60 /* ## KERNEL_VF:             String telling if we calculate potential, force, or both    */
  61 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
  62 /* ##                                                                                    */
  63 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops:             */
  64 /* ## PARTICLES_I[]/         Arrays with lists of i/j particles to use in kernel. It is  */
  65 /* ## PARTICLES_J[]:         just [0] for particle geometry, but can be longer for water */
  66 /* ## PARTICLES_ELEC_I[]/    Arrays with lists of i/j particle that have electrostatics  */
  67 /* ## PARTICLES_ELEC_J[]:    interactions that should be calculated in this kernel.      */
  68 /* ## PARTICLES_VDW_I[]/     Arrays with the list of i/j particle that have VdW          */
  69 /* ## PARTICLES_VDW_J[]:     interactions that should be calculated in this kernel.      */
  70 /* ##                                                                                    */
  71 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle)   */
  72 /* ## PAIRS_IJ[]:            Array with (i,j) tuples of pairs for which interactions     */
  73 /* ##                        should be calculated in this kernel. Zero-charge particles  */
  74 /* ##                        do not have interactions with particles without vdw, and    */
  75 /* ##                        Vdw-only interactions are not evaluated in a no-vdw-kernel. */
  76 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
  77 /* ##                        For each i-j pair, the element [I][J] is a list of strings  */
  78 /* ##                        defining properties/flags of this interaction. Examples     */
  79 /* ##                        include 'electrostatics'/'vdw' if that type of interaction  */
  80 /* ##                        should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values  */
  81 /* ##                        are needed, and 'exactcutoff' or 'shift','switch' to        */
  82 /* ##                        decide if the force/potential should be modified. This way  */
  83 /* ##                        we only calculate values absolutely needed for each case.   */
  84
  85 /* ## Calculate the size and offset for (merged/interleaved) table data */
  86
  87 /*
  88  * Gromacs nonbonded kernel:   {KERNEL_NAME}
  89  * Electrostatics interaction: {KERNEL_ELEC}
  90  * VdW interaction:            {KERNEL_VDW}
  91  * Geometry:                   {GEOMETRY_I}-{GEOMETRY_J}
  92  * Calculate force/pot:        {KERNEL_VF}
  93  */
  94 void
  95 {KERNEL_NAME}
  96                     (t_nblist                    * gmx_restrict       nlist,
  97                      rvec                        * gmx_restrict          xx,
  98                      rvec                        * gmx_restrict          ff,
  99                      t_forcerec                  * gmx_restrict          fr,
 100                      t_mdatoms                   * gmx_restrict     mdatoms,
 101                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 102                      t_nrnb                      * gmx_restrict        nrnb)
 103 {
 104     /* ## Not all variables are used for all kernels, but any optimizing compiler fixes that, */
 105     /* ## so there is no point in going to extremes to exclude variables that are not needed. */
 106     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 107      * just 0 for non-waters.
 108      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 109      * jnr indices corresponding to data put in the four positions in the SIMD register.
 110      */
 111     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 112     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 113     int              jnrA,jnrB,jnrC,jnrD;
 114     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 115     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 116     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 117     real             rcutoff_scalar;
 118     real             *shiftvec,*fshift,*x,*f;
 119     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 120     real             scratch[4*DIM];
 121     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 122     /* #for I in PARTICLES_I */
 123     int              vdwioffset{I};
 124     __m128           ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 125     /* #endfor */
 126     /* #for J in PARTICLES_J */
 127     int              vdwjidx{J}A,vdwjidx{J}B,vdwjidx{J}C,vdwjidx{J}D;
 128     __m128           jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 129     /* #endfor */
 130     /* #for I,J in PAIRS_IJ */
 131     __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};
 132     /* #endfor */
 133     /* #if KERNEL_ELEC != 'None' */
 134     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 135     real             *charge;
 136     /* #endif */
 137     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 138     __m128i          gbitab;
 139     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 140     __m128           minushalf = _mm_set1_ps(-0.5);
 141     real             *invsqrta,*dvda,*gbtab;
 142     /* #endif */
 143     /* #if KERNEL_VDW != 'None' */
 144     int              nvdwtype;
 145     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 146     int              *vdwtype;
 147     real             *vdwparam;
 148     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 149     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 150     /* #endif */
 151     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 152     __m128i          vfitab;
 153     __m128i          ifour       = _mm_set1_epi32(4);
 154     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 155     real             *vftab;
 156     /* #endif */
 157     /* #if 'LJEwald' in KERNEL_VDW */
 158     /* #for I,J in PAIRS_IJ */
 159     __m128           c6grid_{I}{J};
 160     /* #endfor */
 161     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 162     real             *vdwgridparam;
 163     __m128           one_half = _mm_set1_ps(0.5);
 164     __m128           minus_one = _mm_set1_ps(-1.0);
 165     /* #endif */
 166     /* #if 'Ewald' in KERNEL_ELEC */
 167     __m128i          ewitab;
 168     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 169     real             *ewtab;
 170     /* #endif */
 171     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 172     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 173     real             rswitch_scalar,d_scalar;
 174     /* #endif */
 175     __m128           dummy_mask,cutoff_mask;
 176     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 177     __m128           one     = _mm_set1_ps(1.0);
 178     __m128           two     = _mm_set1_ps(2.0);
 179     x                = xx[0];
 180     f                = ff[0];
 181
 182     nri              = nlist->nri;
 183     iinr             = nlist->iinr;
 184     jindex           = nlist->jindex;
 185     jjnr             = nlist->jjnr;
 186     shiftidx         = nlist->shift;
 187     gid              = nlist->gid;
 188     shiftvec         = fr->shift_vec[0];
 189     fshift           = fr->fshift[0];
 190     /* #if KERNEL_ELEC != 'None' */
 191     facel            = _mm_set1_ps(fr->epsfac);
 192     charge           = mdatoms->chargeA;
 193     /*     #if 'ReactionField' in KERNEL_ELEC */
 194     krf              = _mm_set1_ps(fr->ic->k_rf);
 195     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 196     crf              = _mm_set1_ps(fr->ic->c_rf);
 197     /*     #endif */
 198     /* #endif */
 199     /* #if KERNEL_VDW != 'None' */
 200     nvdwtype         = fr->ntype;
 201     vdwparam         = fr->nbfp;
 202     vdwtype          = mdatoms->typeA;
 203     /* #endif */
 204     /* #if 'LJEwald' in KERNEL_VDW */
 205     vdwgridparam     = fr->ljpme_c6grid;
 206     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 207     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 208     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 209     /* #endif */
 210
 211     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 212     vftab            = kernel_data->table_elec_vdw->data;
 213     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 214     /* #elif 'Table' in KERNEL_ELEC */
 215     vftab            = kernel_data->table_elec->data;
 216     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 217     /* #elif 'Table' in KERNEL_VDW */
 218     vftab            = kernel_data->table_vdw->data;
 219     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 220     /* #endif */
 221
 222     /* #if 'Ewald' in KERNEL_ELEC */
 223     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 224     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 225     ewtab            = fr->ic->tabq_coul_F;
 226     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 227     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 228     /*     #else */
 229     ewtab            = fr->ic->tabq_coul_FDV0;
 230     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 231     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 232      /*     #endif */
 233     /* #endif */
 234
 235     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 236     invsqrta         = fr->invsqrta;
 237     dvda             = fr->dvda;
 238     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 239     gbtab            = fr->gbtab.data;
 240     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 241     /* #endif */
 242
 243     /* #if 'Water' in GEOMETRY_I */
 244     /* Setup water-specific parameters */
 245     inr              = nlist->iinr[0];
 246     /*     #for I in PARTICLES_ELEC_I */
 247     iq{I}              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+{I}]));
 248     /*     #endfor */
 249     /*     #for I in PARTICLES_VDW_I */
 250     vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 251     /*     #endfor */
 252     /* #endif */
 253
 254     /* #if 'Water' in GEOMETRY_J */
 255     /*     #for J in PARTICLES_ELEC_J */
 256     jq{J}              = _mm_set1_ps(charge[inr+{J}]);
 257     /*     #endfor */
 258     /*     #for J in PARTICLES_VDW_J */
 259     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 260     /*     #endfor */
 261     /*     #for I,J in PAIRS_IJ */
 262     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 263     qq{I}{J}             = _mm_mul_ps(iq{I},jq{J});
 264     /*         #endif */
 265     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 266     /*             #if 'LJEwald' in KERNEL_VDW */
 267     c6_{I}{J}            = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 268     c12_{I}{J}           = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 269     c6grid_{I}{J}        = _mm_set1_ps(vdwgridparam[vdwioffset{I}+vdwjidx{J}A]);
 270     /*             #else */
 271     c6_{I}{J}            = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 272     c12_{I}{J}           = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 273     /*             #endif */
 274     /*         #endif */
 275     /*     #endfor */
 276     /* #endif */
 277
 278     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 279     /*     #if KERNEL_ELEC!='None' */
 280     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 281     rcutoff_scalar   = fr->rcoulomb;
 282     /*     #else */
 283     rcutoff_scalar   = fr->rvdw;
 284     /*     #endif */
 285     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 286     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 287     /* #endif */
 288
 289     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 290     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 291     rvdw             = _mm_set1_ps(fr->rvdw);
 292     /* #endif */
 293
 294     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 295     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 296     rswitch_scalar   = fr->rcoulomb_switch;
 297     rswitch          = _mm_set1_ps(rswitch_scalar);
 298     /*     #else */
 299     rswitch_scalar   = fr->rvdw_switch;
 300     rswitch          = _mm_set1_ps(rswitch_scalar);
 301     /*     #endif */
 302     /* Setup switch parameters */
 303     d_scalar         = rcutoff_scalar-rswitch_scalar;
 304     d                = _mm_set1_ps(d_scalar);
 305     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 306     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 307     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 308     /*     #if 'Force' in KERNEL_VF */
 309     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 310     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 311     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 312     /*     #endif */
 313     /* #endif */
 314
 315     /* Avoid stupid compiler warnings */
 316     jnrA = jnrB = jnrC = jnrD = 0;
 317     j_coord_offsetA = 0;
 318     j_coord_offsetB = 0;
 319     j_coord_offsetC = 0;
 320     j_coord_offsetD = 0;
 321
 322     /* ## Keep track of the floating point operations we issue for reporting! */
 323     /* #define OUTERFLOPS 0 */
 324     outeriter        = 0;
 325     inneriter        = 0;
 326
 327     for(iidx=0;iidx<4*DIM;iidx++)
 328     {
 329         scratch[iidx] = 0.0;
 330     }
 331
 332     /* Start outer loop over neighborlists */
 333     for(iidx=0; iidx<nri; iidx++)
 334     {
 335         /* Load shift vector for this list */
 336         i_shift_offset   = DIM*shiftidx[iidx];
 337
 338         /* Load limits for loop over neighbors */
 339         j_index_start    = jindex[iidx];
 340         j_index_end      = jindex[iidx+1];
 341
 342         /* Get outer coordinate index */
 343         inr              = iinr[iidx];
 344         i_coord_offset   = DIM*inr;
 345
 346         /* Load i particle coords and add shift vector */
 347         /* #if GEOMETRY_I == 'Particle' */
 348         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 349         /* #elif GEOMETRY_I == 'Water3' */
 350         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 351                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 352         /* #elif GEOMETRY_I == 'Water4' */
 353         /*     #if 0 in PARTICLES_I                 */
 354         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 355                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 356         /*     #else                                */
 357         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 358                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 359         /*     #endif                               */
 360         /* #endif                                   */
 361
 362         /* #if 'Force' in KERNEL_VF */
 363         /*     #for I in PARTICLES_I */
 364         fix{I}             = _mm_setzero_ps();
 365         fiy{I}             = _mm_setzero_ps();
 366         fiz{I}             = _mm_setzero_ps();
 367         /*     #endfor */
 368         /* #endif */
 369
 370         /* ## For water we already preloaded parameters at the start of the kernel */
 371         /* #if not 'Water' in GEOMETRY_I */
 372         /* Load parameters for i particles */
 373         /*     #for I in PARTICLES_ELEC_I */
 374         iq{I}              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+{I}));
 375         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 376         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 377         isai{I}            = _mm_load1_ps(invsqrta+inr+{I});
 378         /*         #endif */
 379         /*     #endfor */
 380         /*     #for I in PARTICLES_VDW_I */
 381         vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 382         /*     #endfor */
 383         /* #endif */
 384
 385         /* #if 'Potential' in KERNEL_VF */
 386         /* Reset potential sums */
 387         /*     #if KERNEL_ELEC != 'None' */
 388         velecsum         = _mm_setzero_ps();
 389         /*     #endif */
 390         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 391         vgbsum           = _mm_setzero_ps();
 392         /*     #endif */
 393         /*     #if KERNEL_VDW != 'None' */
 394         vvdwsum          = _mm_setzero_ps();
 395         /*     #endif */
 396         /* #endif */
 397         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 398         dvdasum          = _mm_setzero_ps();
 399         /*     #endif */
 400
 401         /* #for ROUND in ['Loop','Epilogue'] */
 402
 403         /* #if ROUND =='Loop' */
 404         /* Start inner kernel loop */
 405         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 406         {
 407         /* ## First round is normal loop (next statement resets indentation) */
 408         /*     #if 0 */
 409         }
 410         /*     #endif */
 411         /* #else */
 412         if(jidx<j_index_end)
 413         {
 414         /* ## Second round is epilogue */
 415         /* #endif */
 416         /* #define INNERFLOPS 0 */
 417
 418             /* Get j neighbor index, and coordinate index */
 419             /* #if ROUND =='Loop' */
 420             jnrA             = jjnr[jidx];
 421             jnrB             = jjnr[jidx+1];
 422             jnrC             = jjnr[jidx+2];
 423             jnrD             = jjnr[jidx+3];
 424             /* #else */
 425             jnrlistA         = jjnr[jidx];
 426             jnrlistB         = jjnr[jidx+1];
 427             jnrlistC         = jjnr[jidx+2];
 428             jnrlistD         = jjnr[jidx+3];
 429             /* Sign of each element will be negative for non-real atoms.
 430              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 431              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 432              */
 433             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 434             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 435             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 436             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 437             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 438             /* #endif */
 439             j_coord_offsetA  = DIM*jnrA;
 440             j_coord_offsetB  = DIM*jnrB;
 441             j_coord_offsetC  = DIM*jnrC;
 442             j_coord_offsetD  = DIM*jnrD;
 443
 444             /* load j atom coordinates */
 445             /* #if GEOMETRY_J == 'Particle'             */
 446             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 447                                               x+j_coord_offsetC,x+j_coord_offsetD,
 448                                               &jx0,&jy0,&jz0);
 449             /* #elif GEOMETRY_J == 'Water3'             */
 450             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 451                                               x+j_coord_offsetC,x+j_coord_offsetD,
 452                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 453             /* #elif GEOMETRY_J == 'Water4'             */
 454             /*     #if 0 in PARTICLES_J                 */
 455             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 456                                               x+j_coord_offsetC,x+j_coord_offsetD,
 457                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 458                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 459             /*     #else                                */
 460             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 461                                               x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 462                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 463             /*     #endif                               */
 464             /* #endif                                   */
 465
 466             /* Calculate displacement vector */
 467             /* #for I,J in PAIRS_IJ */
 468             dx{I}{J}             = _mm_sub_ps(ix{I},jx{J});
 469             dy{I}{J}             = _mm_sub_ps(iy{I},jy{J});
 470             dz{I}{J}             = _mm_sub_ps(iz{I},jz{J});
 471             /*     #define INNERFLOPS INNERFLOPS+3 */
 472             /* #endfor */
 473
 474             /* Calculate squared distance and things based on it */
 475             /* #for I,J in PAIRS_IJ */
 476             rsq{I}{J}            = gmx_mm_calc_rsq_ps(dx{I}{J},dy{I}{J},dz{I}{J});
 477             /*     #define INNERFLOPS INNERFLOPS+5 */
 478             /* #endfor */
 479
 480             /* #for I,J in PAIRS_IJ */
 481             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 482             rinv{I}{J}           = gmx_mm_invsqrt_ps(rsq{I}{J});
 483             /*         #define INNERFLOPS INNERFLOPS+5 */
 484             /*     #endif */
 485             /* #endfor */
 486
 487             /* #for I,J in PAIRS_IJ */
 488             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 489             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 490             rinvsq{I}{J}         = gmx_mm_inv_ps(rsq{I}{J});
 491             /*             #define INNERFLOPS INNERFLOPS+4 */
 492             /*         #else */
 493             rinvsq{I}{J}         = _mm_mul_ps(rinv{I}{J},rinv{I}{J});
 494             /*             #define INNERFLOPS INNERFLOPS+1 */
 495             /*         #endif */
 496             /*     #endif */
 497             /* #endfor */
 498
 499             /* #if not 'Water' in GEOMETRY_J */
 500             /* Load parameters for j particles */
 501             /*     #for J in PARTICLES_ELEC_J */
 502             jq{J}              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+{J},charge+jnrB+{J},
 503                                                               charge+jnrC+{J},charge+jnrD+{J});
 504             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 505             isaj{J}            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+{J},invsqrta+jnrB+{J},
 506                                                               invsqrta+jnrC+{J},invsqrta+jnrD+{J});
 507             /*         #endif */
 508             /*     #endfor */
 509             /*     #for J in PARTICLES_VDW_J */
 510             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 511             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 512             vdwjidx{J}C        = 2*vdwtype[jnrC+{J}];
 513             vdwjidx{J}D        = 2*vdwtype[jnrD+{J}];
 514             /*     #endfor */
 515             /* #endif */
 516
 517             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 518             /*     #for J in PARTICLES_J */
 519             fjx{J}             = _mm_setzero_ps();
 520             fjy{J}             = _mm_setzero_ps();
 521             fjz{J}             = _mm_setzero_ps();
 522             /*     #endfor */
 523             /* #endif */
 524
 525             /* #for I,J in PAIRS_IJ */
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 532             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 533             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 534             if (gmx_mm_any_lt(rsq{I}{J},rcutoff2))
 535             {
 536                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 537             }
 538             /*         #endif */
 539             /*         #define INNERFLOPS INNERFLOPS+1 */
 540             /*     #endif */
 541
 542             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 543             r{I}{J}              = _mm_mul_ps(rsq{I}{J},rinv{I}{J});
 544             /*         #if ROUND == 'Epilogue' */
 545             r{I}{J}              = _mm_andnot_ps(dummy_mask,r{I}{J});
 546             /*             #define INNERFLOPS INNERFLOPS+1 */
 547             /*         #endif */
 548             /*         #define INNERFLOPS INNERFLOPS+1 */
 549             /*     #endif */
 550
 551             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 552             /*     #if not 'Water' in GEOMETRY_J */
 553             /* Compute parameters for interactions between i and j atoms */
 554             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 555             qq{I}{J}             = _mm_mul_ps(iq{I},jq{J});
 556             /*             #define INNERFLOPS INNERFLOPS+1 */
 557             /*         #endif */
 558             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 559             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset{I}+vdwjidx{J}A,
 560                                          vdwparam+vdwioffset{I}+vdwjidx{J}B,
 561                                          vdwparam+vdwioffset{I}+vdwjidx{J}C,
 562                                          vdwparam+vdwioffset{I}+vdwjidx{J}D,
 563                                          &c6_{I}{J},&c12_{I}{J});
 564             /*         #if 'LJEwald' in KERNEL_VDW */
 565             c6grid_{I}{J}       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset{I}+vdwjidx{J}A,
 566                                                                vdwgridparam+vdwioffset{I}+vdwjidx{J}B,
 567                                                                vdwgridparam+vdwioffset{I}+vdwjidx{J}C,
 568                                                                vdwgridparam+vdwioffset{I}+vdwjidx{J}D);
 569             /*          #endif */
 570             /*         #endif */
 571             /*     #endif */
 572
 573             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 574             /* Calculate table index by multiplying r with table scale and truncate to integer */
 575             rt               = _mm_mul_ps(r{I}{J},vftabscale);
 576             vfitab           = _mm_cvttps_epi32(rt);
 577             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 578             /*         #define INNERFLOPS INNERFLOPS+4                          */
 579             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW     */
 580             /*             ## 3 tables, 4 bytes per point: multiply index by 12 */
 581             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 582             /*         #elif 'Table' in KERNEL_ELEC                             */
 583             /*             ## 1 table, 4 bytes per point: multiply index by 4   */
 584             vfitab           = _mm_slli_epi32(vfitab,2);
 585             /*         #elif 'Table' in KERNEL_VDW                              */
 586             /*             ## 2 tables, 4 bytes per point: multiply index by 8  */
 587             vfitab           = _mm_slli_epi32(vfitab,3);
 588             /*         #endif                                                   */
 589             /*     #endif */
 590
 591             /*     ## ELECTROSTATIC INTERACTIONS */
 592             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 593
 594             /*         #if KERNEL_ELEC=='Coulomb' */
 595
 596             /* COULOMB ELECTROSTATICS */
 597             velec            = _mm_mul_ps(qq{I}{J},rinv{I}{J});
 598             /*             #define INNERFLOPS INNERFLOPS+1 */
 599             /*             #if 'Force' in KERNEL_VF */
 600             felec            = _mm_mul_ps(velec,rinvsq{I}{J});
 601             /*                 #define INNERFLOPS INNERFLOPS+2 */
 602             /*             #endif */
 603
 604             /*         #elif KERNEL_ELEC=='ReactionField' */
 605
 606             /* REACTION-FIELD ELECTROSTATICS */
 607             /*             #if 'Potential' in KERNEL_VF */
 608             velec            = _mm_mul_ps(qq{I}{J},_mm_sub_ps(_mm_add_ps(rinv{I}{J},_mm_mul_ps(krf,rsq{I}{J})),crf));
 609             /*                 #define INNERFLOPS INNERFLOPS+4 */
 610             /*             #endif */
 611             /*             #if 'Force' in KERNEL_VF */
 612             felec            = _mm_mul_ps(qq{I}{J},_mm_sub_ps(_mm_mul_ps(rinv{I}{J},rinvsq{I}{J}),krf2));
 613             /*                 #define INNERFLOPS INNERFLOPS+3 */
 614             /*             #endif */
 615
 616             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 617
 618             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 619             isaprod          = _mm_mul_ps(isai{I},isaj{J});
 620             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq{I}{J},_mm_mul_ps(isaprod,gbinvepsdiff)));
 621             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 622             /*             #define INNERFLOPS INNERFLOPS+5 */
 623
 624             /* Calculate generalized born table index - this is a separate table from the normal one,
 625              * but we use the same procedure by multiplying r with scale and truncating to integer.
 626              */
 627             rt               = _mm_mul_ps(r{I}{J},gbscale);
 628             gbitab           = _mm_cvttps_epi32(rt);
 629             gbeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(gbitab));
 630             gbitab           = _mm_slli_epi32(gbitab,2);
 631
 632             Y                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 633             F                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 634             G                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,2) );
 635             H                = _mm_load_ps( gbtab + gmx_mm_extract_epi32(gbitab,3) );
 636             _MM_TRANSPOSE4_PS(Y,F,G,H);
 637             Heps             = _mm_mul_ps(gbeps,H);
 638             Fp               = _mm_add_ps(F,_mm_mul_ps(gbeps,_mm_add_ps(G,Heps)));
 639             VV               = _mm_add_ps(Y,_mm_mul_ps(gbeps,Fp));
 640             vgb              = _mm_mul_ps(gbqqfactor,VV);
 641             /*             #define INNERFLOPS INNERFLOPS+10 */
 642
 643             /*             #if 'Force' in KERNEL_VF */
 644             FF               = _mm_add_ps(Fp,_mm_mul_ps(gbeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 645             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 646             dvdatmp          = _mm_mul_ps(minushalf,_mm_add_ps(vgb,_mm_mul_ps(fgb,r{I}{J})));
 647             /*                 #if ROUND == 'Epilogue' */
 648             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 649             /*                 #endif */
 650             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 651             /*                 #if ROUND == 'Loop' */
 652             fjptrA           = dvda+jnrA;
 653             fjptrB           = dvda+jnrB;
 654             fjptrC           = dvda+jnrC;
 655             fjptrD           = dvda+jnrD;
 656             /*                 #else */
 657             /* 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. */
 658             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 659             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 660             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 661             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 662             /*                 #endif */
 663             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj{J},isaj{J})));
 664             /*                 #define INNERFLOPS INNERFLOPS+13 */
 665             /*             #endif */
 666             velec            = _mm_mul_ps(qq{I}{J},rinv{I}{J});
 667             /*                 #define INNERFLOPS INNERFLOPS+1 */
 668             /*             #if 'Force' in KERNEL_VF */
 669             felec            = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(velec,rinv{I}{J}),fgb),rinv{I}{J});
 670             /*                 #define INNERFLOPS INNERFLOPS+3 */
 671             /*             #endif */
 672
 673             /*         #elif KERNEL_ELEC=='Ewald' */
 674             /* EWALD ELECTROSTATICS */
 675
 676             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 677             ewrt             = _mm_mul_ps(r{I}{J},ewtabscale);
 678             ewitab           = _mm_cvttps_epi32(ewrt);
 679             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 680             /*             #define INNERFLOPS INNERFLOPS+4 */
 681             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 682             ewitab           = _mm_slli_epi32(ewitab,2);
 683             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 684             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 685             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 686             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 687             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 688             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 689             /*                 #define INNERFLOPS INNERFLOPS+2 */
 690             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 691             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 692             velec            = _mm_mul_ps(qq{I}{J},_mm_sub_ps(_mm_sub_ps(rinv{I}{J},sh_ewald),velec));
 693             /*                     #define INNERFLOPS INNERFLOPS+7 */
 694             /*                 #else */
 695             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 696             velec            = _mm_mul_ps(qq{I}{J},_mm_sub_ps(rinv{I}{J},velec));
 697             /*                     #define INNERFLOPS INNERFLOPS+6 */
 698             /*                 #endif */
 699             /*                 #if 'Force' in KERNEL_VF */
 700             felec            = _mm_mul_ps(_mm_mul_ps(qq{I}{J},rinv{I}{J}),_mm_sub_ps(rinvsq{I}{J},felec));
 701             /*                      #define INNERFLOPS INNERFLOPS+3 */
 702             /*                 #endif */
 703             /*             #elif KERNEL_VF=='Force' */
 704             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 705                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 706                                          &ewtabF,&ewtabFn);
 707             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 708             felec            = _mm_mul_ps(_mm_mul_ps(qq{I}{J},rinv{I}{J}),_mm_sub_ps(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                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 716             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 717             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 718             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 719             _MM_TRANSPOSE4_PS(Y,F,G,H);
 720             Heps             = _mm_mul_ps(vfeps,H);
 721             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 722             /*             #define INNERFLOPS INNERFLOPS+4 */
 723             /*             #if 'Potential' in KERNEL_VF */
 724             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 725             velec            = _mm_mul_ps(qq{I}{J},VV);
 726             /*                 #define INNERFLOPS INNERFLOPS+3 */
 727             /*             #endif */
 728             /*             #if 'Force' in KERNEL_VF */
 729             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 730             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq{I}{J},FF),_mm_mul_ps(vftabscale,rinv{I}{J})));
 731             /*                 #define INNERFLOPS INNERFLOPS+7 */
 732             /*             #endif */
 733             /*         #endif */
 734             /*         ## End of check for electrostatics interaction forms */
 735             /*     #endif */
 736             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 737
 738             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 739
 740             /*         #if KERNEL_VDW=='LennardJones' */
 741
 742             /* LENNARD-JONES DISPERSION/REPULSION */
 743
 744             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 745             /*             #define INNERFLOPS INNERFLOPS+2 */
 746             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 747             vvdw6            = _mm_mul_ps(c6_{I}{J},rinvsix);
 748             vvdw12           = _mm_mul_ps(c12_{I}{J},_mm_mul_ps(rinvsix,rinvsix));
 749             /*                 #define INNERFLOPS INNERFLOPS+3 */
 750             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 751             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) ,
 752                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
 753             /*                     #define INNERFLOPS INNERFLOPS+8 */
 754             /*                 #else */
 755             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 756             /*                     #define INNERFLOPS INNERFLOPS+3 */
 757             /*                 #endif */
 758             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 759             /*                 #if 'Force' in KERNEL_VF */
 760             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq{I}{J});
 761             /*                     #define INNERFLOPS INNERFLOPS+2 */
 762             /*                 #endif */
 763             /*             #elif KERNEL_VF=='Force' */
 764             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 765             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_{I}{J},rinvsix),c6_{I}{J}),_mm_mul_ps(rinvsix,rinvsq{I}{J}));
 766             /*                 #define INNERFLOPS INNERFLOPS+4 */
 767             /*             #endif */
 768
 769             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 770
 771             /* CUBIC SPLINE TABLE DISPERSION */
 772             /*             #if 'Table' in KERNEL_ELEC */
 773             vfitab           = _mm_add_epi32(vfitab,ifour);
 774             /*             #endif                     */
 775             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 776             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 777             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 778             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 779             _MM_TRANSPOSE4_PS(Y,F,G,H);
 780             Heps             = _mm_mul_ps(vfeps,H);
 781             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 782             /*             #define INNERFLOPS INNERFLOPS+4 */
 783             /*             #if 'Potential' in KERNEL_VF */
 784             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 785             vvdw6            = _mm_mul_ps(c6_{I}{J},VV);
 786             /*                 #define INNERFLOPS INNERFLOPS+3 */
 787             /*             #endif */
 788             /*             #if 'Force' in KERNEL_VF */
 789             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 790             fvdw6            = _mm_mul_ps(c6_{I}{J},FF);
 791             /*                 #define INNERFLOPS INNERFLOPS+4 */
 792             /*             #endif */
 793
 794             /* CUBIC SPLINE TABLE REPULSION */
 795             vfitab           = _mm_add_epi32(vfitab,ifour);
 796             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 797             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 798             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 799             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 800             _MM_TRANSPOSE4_PS(Y,F,G,H);
 801             Heps             = _mm_mul_ps(vfeps,H);
 802             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 803             /*             #define INNERFLOPS INNERFLOPS+4 */
 804             /*             #if 'Potential' in KERNEL_VF */
 805             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 806             vvdw12           = _mm_mul_ps(c12_{I}{J},VV);
 807             /*                 #define INNERFLOPS INNERFLOPS+3 */
 808             /*             #endif */
 809             /*             #if 'Force' in KERNEL_VF */
 810             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 811             fvdw12           = _mm_mul_ps(c12_{I}{J},FF);
 812             /*                 #define INNERFLOPS INNERFLOPS+5 */
 813             /*             #endif */
 814             /*             #if 'Potential' in KERNEL_VF */
 815             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 816             /*                 #define INNERFLOPS INNERFLOPS+1 */
 817             /*             #endif */
 818             /*             #if 'Force' in KERNEL_VF */
 819             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv{I}{J})));
 820             /*                 #define INNERFLOPS INNERFLOPS+4 */
 821             /*             #endif */
 822
 823             /*         #elif KERNEL_VDW=='LJEwald' */
 824
 825             /* Analytical LJ-PME */
 826             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 827             ewcljrsq         = _mm_mul_ps(ewclj2,rsq{I}{J});
 828             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 829             exponent         = gmx_simd_exp_r(ewcljrsq);
 830             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 831             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 832             /*                 #define INNERFLOPS INNERFLOPS+11 */
 833             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 834             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 835             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_{I}{J},_mm_mul_ps(c6grid_{I}{J},_mm_sub_ps(one,poly))),rinvsix);
 836             vvdw12           = _mm_mul_ps(c12_{I}{J},_mm_mul_ps(rinvsix,rinvsix));
 837             /*                 #define INNERFLOPS INNERFLOPS+6 */
 838             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 839             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) ,
 840                                           _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));
 841             /*                     #define INNERFLOPS INNERFLOPS+10 */
 842             /*                 #else */
 843             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 844             /*                 #define INNERFLOPS INNERFLOPS+3 */
 845             /*                 #endif */
 846             /*                  ## Check for force inside potential check, i.e. this means we already did the potential part */
 847             /*                  #if 'Force' in KERNEL_VF */
 848             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 849             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});
 850             /*                 #define INNERFLOPS INNERFLOPS+6 */
 851             /*                  #endif */
 852             /*              #elif KERNEL_VF=='Force' */
 853             /* f6A = 6 * C6grid * (1 - poly) */
 854             f6A              = _mm_mul_ps(c6grid_{I}{J},_mm_sub_ps(one,poly));
 855             /* f6B = C6grid * exponent * beta^6 */
 856             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_{I}{J},one_sixth),_mm_mul_ps(exponent,ewclj6));
 857             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 858             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});
 859             /*                 #define INNERFLOPS INNERFLOPS+11 */
 860             /*              #endif */
 861             /*         #endif */
 862             /*         ## End of check for vdw interaction forms */
 863             /*     #endif */
 864             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 865
 866             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 867             d                = _mm_sub_ps(r{I}{J},rswitch);
 868             d                = _mm_max_ps(d,_mm_setzero_ps());
 869             d2               = _mm_mul_ps(d,d);
 870             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)))))));
 871             /*         #define INNERFLOPS INNERFLOPS+10 */
 872
 873             /*         #if 'Force' in KERNEL_VF */
 874             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 875             /*             #define INNERFLOPS INNERFLOPS+5 */
 876             /*         #endif */
 877
 878             /* Evaluate switch function */
 879             /*         #if 'Force' in KERNEL_VF */
 880             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 881             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 882             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv{I}{J},_mm_mul_ps(velec,dsw)) );
 883             /*                 #define INNERFLOPS INNERFLOPS+4 */
 884             /*             #endif */
 885             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 886             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv{I}{J},_mm_mul_ps(vvdw,dsw)) );
 887             /*                 #define INNERFLOPS INNERFLOPS+4 */
 888             /*             #endif */
 889             /*         #endif */
 890             /*         #if 'Potential' in KERNEL_VF */
 891             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 892             velec            = _mm_mul_ps(velec,sw);
 893             /*                 #define INNERFLOPS INNERFLOPS+1 */
 894             /*             #endif */
 895             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 896             vvdw             = _mm_mul_ps(vvdw,sw);
 897             /*                 #define INNERFLOPS INNERFLOPS+1 */
 898             /*             #endif */
 899             /*         #endif */
 900             /*     #endif */
 901             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 902             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 903             cutoff_mask      = _mm_cmplt_ps(rsq{I}{J},rcutoff2);
 904             /*         #define INNERFLOPS INNERFLOPS+1 */
 905             /*     #endif */
 906
 907             /*     #if 'Potential' in KERNEL_VF */
 908             /* Update potential sum for this i atom from the interaction with this j atom. */
 909             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 910             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 911             velec            = _mm_and_ps(velec,cutoff_mask);
 912             /*                 #define INNERFLOPS INNERFLOPS+1 */
 913             /*             #endif                                       */
 914             /*             #if ROUND == 'Epilogue' */
 915             velec            = _mm_andnot_ps(dummy_mask,velec);
 916             /*             #endif */
 917             velecsum         = _mm_add_ps(velecsum,velec);
 918             /*             #define INNERFLOPS INNERFLOPS+1 */
 919             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 920             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 921             vgb              = _mm_and_ps(vgb,cutoff_mask);
 922             /*                 #define INNERFLOPS INNERFLOPS+1 */
 923             /*             #endif                                       */
 924             /*             #if ROUND == 'Epilogue' */
 925             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 926             /*             #endif */
 927             vgbsum           = _mm_add_ps(vgbsum,vgb);
 928             /*                 #define INNERFLOPS INNERFLOPS+1 */
 929             /*             #endif */
 930             /*         #endif */
 931             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 932             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 933             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 934             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 935             /*                 #define INNERFLOPS INNERFLOPS+1 */
 936             /*             #endif                                       */
 937             /*             #if ROUND == 'Epilogue' */
 938             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 939             /*             #endif */
 940             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 941             /*             #define INNERFLOPS INNERFLOPS+1 */
 942             /*         #endif */
 943             /*     #endif */
 944
 945             /*     #if 'Force' in KERNEL_VF */
 946
 947             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 948             fscal            = _mm_add_ps(felec,fvdw);
 949             /*             #define INNERFLOPS INNERFLOPS+1 */
 950             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 951             fscal            = felec;
 952             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 953             fscal            = fvdw;
 954             /*        #endif */
 955
 956             /*        ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 957             /*        #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 958             fscal            = _mm_and_ps(fscal,cutoff_mask);
 959             /*             #define INNERFLOPS INNERFLOPS+1 */
 960             /*         #endif                                       */
 961
 962             /*             #if ROUND == 'Epilogue' */
 963             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 964             /*             #endif */
 965
 966             /* Calculate temporary vectorial force */
 967             tx               = _mm_mul_ps(fscal,dx{I}{J});
 968             ty               = _mm_mul_ps(fscal,dy{I}{J});
 969             tz               = _mm_mul_ps(fscal,dz{I}{J});
 970
 971             /* Update vectorial force */
 972             fix{I}             = _mm_add_ps(fix{I},tx);
 973             fiy{I}             = _mm_add_ps(fiy{I},ty);
 974             fiz{I}             = _mm_add_ps(fiz{I},tz);
 975             /*             #define INNERFLOPS INNERFLOPS+6 */
 976
 977             /* #if GEOMETRY_I == 'Particle'             */
 978             /*     #if ROUND == 'Loop' */
 979             fjptrA             = f+j_coord_offsetA;
 980             fjptrB             = f+j_coord_offsetB;
 981             fjptrC             = f+j_coord_offsetC;
 982             fjptrD             = f+j_coord_offsetD;
 983             /*     #else */
 984             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 985             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 986             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 987             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 988             /*     #endif */
 989             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 990             /*     #define INNERFLOPS INNERFLOPS+3      */
 991             /* #else                                    */
 992             fjx{J}             = _mm_add_ps(fjx{J},tx);
 993             fjy{J}             = _mm_add_ps(fjy{J},ty);
 994             fjz{J}             = _mm_add_ps(fjz{J},tz);
 995             /*     #define INNERFLOPS INNERFLOPS+3      */
 996             /* #endif                                   */
 997
 998             /*     #endif */
 999
1000             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
1001             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
1002             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
1003             {
1004                 /*     #endif */
1005             }
1006             /*     #endif */
1007             /*    ## End of check for the interaction being outside the cutoff */
1008
1009             /* #endfor */
1010             /* ## End of loop over i-j interaction pairs */
1011
1012             /* #if GEOMETRY_I != 'Particle' */
1013             /*     #if ROUND == 'Loop' */
1014             fjptrA             = f+j_coord_offsetA;
1015             fjptrB             = f+j_coord_offsetB;
1016             fjptrC             = f+j_coord_offsetC;
1017             fjptrD             = f+j_coord_offsetD;
1018             /*     #else */
1019             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1020             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1021             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1022             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1023             /*     #endif */
1024             /* #endif */
1025
1026             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
1027             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1028             /* #elif GEOMETRY_J == 'Water3'             */
1029             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1030                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1031             /*     #define INNERFLOPS INNERFLOPS+9      */
1032             /* #elif GEOMETRY_J == 'Water4'             */
1033             /*     #if 0 in PARTICLES_J                 */
1034             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1035                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1036                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1037             /*     #define INNERFLOPS INNERFLOPS+12     */
1038             /*     #else                                */
1039             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1040                                                    fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1041             /*     #define INNERFLOPS INNERFLOPS+9      */
1042             /*     #endif                               */
1043             /* #endif                                   */
1044
1045             /* Inner loop uses {INNERFLOPS} flops */
1046         }
1047
1048         /* #endfor */
1049
1050         /* End of innermost loop */
1051
1052         /* #if 'Force' in KERNEL_VF */
1053         /*     #if GEOMETRY_I == 'Particle'            */
1054         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1055                                               f+i_coord_offset,fshift+i_shift_offset);
1056         /*         #define OUTERFLOPS OUTERFLOPS+6     */
1057         /*     #elif GEOMETRY_I == 'Water3'            */
1058         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1059                                               f+i_coord_offset,fshift+i_shift_offset);
1060         /*         #define OUTERFLOPS OUTERFLOPS+18    */
1061         /*     #elif GEOMETRY_I == 'Water4'            */
1062         /*         #if 0 in PARTICLES_I                */
1063         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1064                                               f+i_coord_offset,fshift+i_shift_offset);
1065         /*             #define OUTERFLOPS OUTERFLOPS+24    */
1066         /*         #else                               */
1067         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1068                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1069         /*             #define OUTERFLOPS OUTERFLOPS+18    */
1070         /*         #endif                              */
1071         /*     #endif                                  */
1072         /* #endif                                      */
1073
1074         /* #if 'Potential' in KERNEL_VF */
1075         ggid                        = gid[iidx];
1076         /* Update potential energies */
1077         /*     #if KERNEL_ELEC != 'None' */
1078         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1079         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1080         /*     #endif */
1081         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
1082         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
1083         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1084         /*     #endif */
1085         /*     #if KERNEL_VDW != 'None' */
1086         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1087         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1088         /*     #endif */
1089         /* #endif */
1090         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1091         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai{I},isai{I}));
1092         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
1093         /*     #endif */
1094
1095         /* Increment number of inner iterations */
1096         inneriter                  += j_index_end - j_index_start;
1097
1098         /* Outer loop uses {OUTERFLOPS} flops */
1099     }
1100
1101     /* Increment number of outer iterations */
1102     outeriter        += nri;
1103
1104     /* Update outer/inner flops */
1105     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1106     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1107     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1108     /* #if GEOMETRY_I == 'Water3'            */
1109     /*     #define ISUFFIX '_W3'             */
1110     /* #elif GEOMETRY_I == 'Water4'          */
1111     /*     #define ISUFFIX '_W4'             */
1112     /* #else                                 */
1113     /*     #define ISUFFIX ''                */
1114     /* #endif                                */
1115     /* #if GEOMETRY_J == 'Water3'            */
1116     /*     #define JSUFFIX 'W3'              */
1117     /* #elif GEOMETRY_J == 'Water4'          */
1118     /*     #define JSUFFIX 'W4'              */
1119     /* #else                                 */
1120     /*     #define JSUFFIX ''                */
1121     /* #endif                                */
1122     /* #if 'PotentialAndForce' in KERNEL_VF  */
1123     /*     #define VFSUFFIX  '_VF'           */
1124     /* #elif 'Potential' in KERNEL_VF        */
1125     /*     #define VFSUFFIX '_V'             */
1126     /* #else                                 */
1127     /*     #define VFSUFFIX '_F'             */
1128     /* #endif                                */
1129
1130     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1131     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1132     /* #elif KERNEL_ELEC != 'None' */
1133     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1134     /* #else */
1135     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1136     /* #endif  */
1137 }