src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_template_avx_256_single.pre

   1 /* #if 0 */
   2 /*
   3  * This file is part of the GROMACS molecular simulation package.
   4  *
   5  * Copyright (c) 2012,2013,2014,2015,2017,2018, 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_avx_256_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
  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                      struct 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,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight 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              jnrE,jnrF,jnrG,jnrH;
 115     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 116     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 117     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 118     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 119     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 120     real             rcutoff_scalar;
 121     real             *shiftvec,*fshift,*x,*f;
 122     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 123     real             scratch[4*DIM];
 124     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 125     /* #for I in PARTICLES_I */
 126     real *           vdwioffsetptr{I};
 127     /* #if 'LJEwald' in KERNEL_VDW */
 128     real *           vdwgridioffsetptr{I};
 129     /* #endif                      */
 130     __m256           ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 131     /* #endfor */
 132     /* #for J in PARTICLES_J */
 133     int              vdwjidx{J}A,vdwjidx{J}B,vdwjidx{J}C,vdwjidx{J}D,vdwjidx{J}E,vdwjidx{J}F,vdwjidx{J}G,vdwjidx{J}H;
 134     __m256           jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 135     /* #endfor */
 136     /* #for I,J in PAIRS_IJ */
 137     __m256           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};
 138     /* #endfor */
 139     /* #if KERNEL_ELEC != 'None' */
 140     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 141     real             *charge;
 142     /* #endif */
 143     /* #if KERNEL_VDW != 'None' */
 144     int              nvdwtype;
 145     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 146     int              *vdwtype;
 147     real             *vdwparam;
 148     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 149     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 150     /* #endif */
 151     /* #if 'Table' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 152     __m256i          vfitab;
 153     __m128i          vfitab_lo,vfitab_hi;
 154     __m128i          ifour       = _mm_set1_epi32(4);
 155     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 156     real             *vftab;
 157     /* #endif */
 158     /* #if 'LJEwald' in KERNEL_VDW */
 159     /* #for I,J in PAIRS_IJ */
 160     __m256           c6grid_{I}{J};
 161     /* #endfor */
 162     real             *vdwgridparam;
 163     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 164     __m256           one_half  = _mm256_set1_ps(0.5);
 165     __m256           minus_one = _mm256_set1_ps(-1.0);
 166     /* #endif */
 167     /* #if 'Ewald' in KERNEL_ELEC */
 168     __m256i          ewitab;
 169     __m128i          ewitab_lo,ewitab_hi;
 170     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 171     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 172     real             *ewtab;
 173     /* #endif */
 174     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 175     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 176     real             rswitch_scalar,d_scalar;
 177     /* #endif */
 178     __m256           dummy_mask,cutoff_mask;
 179     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 180     __m256           one     = _mm256_set1_ps(1.0);
 181     __m256           two     = _mm256_set1_ps(2.0);
 182     x                = xx[0];
 183     f                = ff[0];
 184
 185     nri              = nlist->nri;
 186     iinr             = nlist->iinr;
 187     jindex           = nlist->jindex;
 188     jjnr             = nlist->jjnr;
 189     shiftidx         = nlist->shift;
 190     gid              = nlist->gid;
 191     shiftvec         = fr->shift_vec[0];
 192     fshift           = fr->fshift[0];
 193     /* #if KERNEL_ELEC != 'None' */
 194     facel            = _mm256_set1_ps(fr->ic->epsfac);
 195     charge           = mdatoms->chargeA;
 196     /*     #if 'ReactionField' in KERNEL_ELEC */
 197     krf              = _mm256_set1_ps(fr->ic->k_rf);
 198     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 199     crf              = _mm256_set1_ps(fr->ic->c_rf);
 200     /*     #endif */
 201     /* #endif */
 202     /* #if KERNEL_VDW != 'None' */
 203     nvdwtype         = fr->ntype;
 204     vdwparam         = fr->nbfp;
 205     vdwtype          = mdatoms->typeA;
 206     /* #endif */
 207     /* #if 'LJEwald' in KERNEL_VDW */
 208     vdwgridparam     = fr->ljpme_c6grid;
 209     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 210     ewclj            = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
 211     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 212     /* #endif */
 213
 214     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 215     vftab            = kernel_data->table_elec_vdw->data;
 216     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 217     /* #elif 'Table' in KERNEL_ELEC */
 218     vftab            = kernel_data->table_elec->data;
 219     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 220     /* #elif 'Table' in KERNEL_VDW */
 221     vftab            = kernel_data->table_vdw->data;
 222     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 223     /* #endif */
 224
 225     /* #if 'Ewald' in KERNEL_ELEC */
 226     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 227     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 228     beta2            = _mm256_mul_ps(beta,beta);
 229     beta3            = _mm256_mul_ps(beta,beta2);
 230
 231     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 232     ewtab            = fr->ic->tabq_coul_F;
 233     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 234     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 235     /*     #else */
 236     ewtab            = fr->ic->tabq_coul_FDV0;
 237     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 238     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 239      /*     #endif */
 240     /* #endif */
 241
 242     /* #if 'Water' in GEOMETRY_I */
 243     /* Setup water-specific parameters */
 244     inr              = nlist->iinr[0];
 245     /*     #for I in PARTICLES_ELEC_I */
 246     iq{I}              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+{I}]));
 247     /*     #endfor */
 248     /*     #for I in PARTICLES_VDW_I */
 249     vdwioffsetptr{I}   = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
 250     /*         #if 'LJEwald' in KERNEL_VDW */
 251     vdwgridioffsetptr{I} = vdwgridparam+2*nvdwtype*vdwtype[inr+{I}];
 252     /*         #endif */
 253     /*     #endfor */
 254     /* #endif */
 255
 256     /* #if 'Water' in GEOMETRY_J */
 257     /*     #for J in PARTICLES_ELEC_J */
 258     jq{J}              = _mm256_set1_ps(charge[inr+{J}]);
 259     /*     #endfor */
 260     /*     #for J in PARTICLES_VDW_J */
 261     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 262     /*     #endfor */
 263     /*     #for I,J in PAIRS_IJ */
 264     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 265     qq{I}{J}             = _mm256_mul_ps(iq{I},jq{J});
 266     /*         #endif */
 267     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 268         /*         #if 'LJEwald' in KERNEL_VDW */
 269     c6_{I}{J}            = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A]);
 270     c12_{I}{J}           = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A+1]);
 271     c6grid_{I}{J}        = _mm256_set1_ps(vdwgridioffsetptr{I}[vdwjidx{J}A]);
 272         /*         #else */
 273     c6_{I}{J}            = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A]);
 274     c12_{I}{J}           = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A+1]);
 275         /*         #endif */
 276     /*         #endif */
 277     /*     #endfor */
 278     /* #endif */
 279
 280     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 281     /*     #if KERNEL_ELEC!='None' */
 282     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 283     rcutoff_scalar   = fr->ic->rcoulomb;
 284     /*     #else */
 285     rcutoff_scalar   = fr->ic->rvdw;
 286     /*     #endif */
 287     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 288     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 289     /* #endif */
 290
 291     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 292     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 293     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 294     /* #endif */
 295
 296     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 297     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 298     rswitch_scalar   = fr->ic->rcoulomb_switch;
 299     rswitch          = _mm256_set1_ps(rswitch_scalar);
 300     /*     #else */
 301     rswitch_scalar   = fr->ic->rvdw_switch;
 302     rswitch          = _mm256_set1_ps(rswitch_scalar);
 303     /*     #endif */
 304     /* Setup switch parameters */
 305     d_scalar         = rcutoff_scalar-rswitch_scalar;
 306     d                = _mm256_set1_ps(d_scalar);
 307     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 308     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 309     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 310     /*     #if 'Force' in KERNEL_VF */
 311     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 312     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 313     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 314     /*     #endif */
 315     /* #endif */
 316
 317     /* Avoid stupid compiler warnings */
 318     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 319     j_coord_offsetA = 0;
 320     j_coord_offsetB = 0;
 321     j_coord_offsetC = 0;
 322     j_coord_offsetD = 0;
 323     j_coord_offsetE = 0;
 324     j_coord_offsetF = 0;
 325     j_coord_offsetG = 0;
 326     j_coord_offsetH = 0;
 327
 328     /* ## Keep track of the floating point operations we issue for reporting! */
 329     /* #define OUTERFLOPS 0 */
 330     outeriter        = 0;
 331     inneriter        = 0;
 332
 333     for(iidx=0;iidx<4*DIM;iidx++)
 334     {
 335         scratch[iidx] = 0.0;
 336     }
 337
 338     /* Start outer loop over neighborlists */
 339     for(iidx=0; iidx<nri; iidx++)
 340     {
 341         /* Load shift vector for this list */
 342         i_shift_offset   = DIM*shiftidx[iidx];
 343
 344         /* Load limits for loop over neighbors */
 345         j_index_start    = jindex[iidx];
 346         j_index_end      = jindex[iidx+1];
 347
 348         /* Get outer coordinate index */
 349         inr              = iinr[iidx];
 350         i_coord_offset   = DIM*inr;
 351
 352         /* Load i particle coords and add shift vector */
 353         /* #if GEOMETRY_I == 'Particle' */
 354         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 355         /* #elif GEOMETRY_I == 'Water3' */
 356         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 357                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 358         /* #elif GEOMETRY_I == 'Water4' */
 359         /*     #if 0 in PARTICLES_I                 */
 360         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 361                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 362         /*     #else                                */
 363         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 364                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 365         /*     #endif                               */
 366         /* #endif                                   */
 367
 368         /* #if 'Force' in KERNEL_VF */
 369         /*     #for I in PARTICLES_I */
 370         fix{I}             = _mm256_setzero_ps();
 371         fiy{I}             = _mm256_setzero_ps();
 372         fiz{I}             = _mm256_setzero_ps();
 373         /*     #endfor */
 374         /* #endif */
 375
 376         /* ## For water we already preloaded parameters at the start of the kernel */
 377         /* #if not 'Water' in GEOMETRY_I */
 378         /* Load parameters for i particles */
 379         /*     #for I in PARTICLES_ELEC_I */
 380         iq{I}              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+{I}]));
 381         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 382         /*     #endfor */
 383         /*     #for I in PARTICLES_VDW_I */
 384         vdwioffsetptr{I}   = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
 385         /*         #if 'LJEwald' in KERNEL_VDW */
 386         vdwgridioffsetptr{I} = vdwgridparam+2*nvdwtype*vdwtype[inr+{I}];
 387         /*         #endif */
 388         /*     #endfor */
 389         /* #endif */
 390
 391         /* #if 'Potential' in KERNEL_VF */
 392         /* Reset potential sums */
 393         /*     #if KERNEL_ELEC != 'None' */
 394         velecsum         = _mm256_setzero_ps();
 395         /*     #endif */
 396         /*     #if KERNEL_VDW != 'None' */
 397         vvdwsum          = _mm256_setzero_ps();
 398         /*     #endif */
 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+7]>=0; jidx+=8)
 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             jnrE             = jjnr[jidx+4];
 425             jnrF             = jjnr[jidx+5];
 426             jnrG             = jjnr[jidx+6];
 427             jnrH             = jjnr[jidx+7];
 428             /* #else */
 429             jnrlistA         = jjnr[jidx];
 430             jnrlistB         = jjnr[jidx+1];
 431             jnrlistC         = jjnr[jidx+2];
 432             jnrlistD         = jjnr[jidx+3];
 433             jnrlistE         = jjnr[jidx+4];
 434             jnrlistF         = jjnr[jidx+5];
 435             jnrlistG         = jjnr[jidx+6];
 436             jnrlistH         = jjnr[jidx+7];
 437             /* Sign of each element will be negative for non-real atoms.
 438              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 439              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 440              */
 441             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 442                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 443
 444             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 445             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 446             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 447             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 448             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 449             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 450             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 451             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 452             /* #endif */
 453             j_coord_offsetA  = DIM*jnrA;
 454             j_coord_offsetB  = DIM*jnrB;
 455             j_coord_offsetC  = DIM*jnrC;
 456             j_coord_offsetD  = DIM*jnrD;
 457             j_coord_offsetE  = DIM*jnrE;
 458             j_coord_offsetF  = DIM*jnrF;
 459             j_coord_offsetG  = DIM*jnrG;
 460             j_coord_offsetH  = DIM*jnrH;
 461
 462             /* load j atom coordinates */
 463             /* #if GEOMETRY_J == 'Particle'             */
 464             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 465                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 466                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 467                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 468                                                  &jx0,&jy0,&jz0);
 469             /* #elif GEOMETRY_J == 'Water3'             */
 470             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 471                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 472                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 473                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 474                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 475             /* #elif GEOMETRY_J == 'Water4'             */
 476             /*     #if 0 in PARTICLES_J                 */
 477             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 478                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 479                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 480                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 481                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 482                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 483             /*     #else                                */
 484             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 485                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 486                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 487                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 488                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 489             /*     #endif                               */
 490             /* #endif                                   */
 491
 492             /* Calculate displacement vector */
 493             /* #for I,J in PAIRS_IJ */
 494             dx{I}{J}             = _mm256_sub_ps(ix{I},jx{J});
 495             dy{I}{J}             = _mm256_sub_ps(iy{I},jy{J});
 496             dz{I}{J}             = _mm256_sub_ps(iz{I},jz{J});
 497             /*     #define INNERFLOPS INNERFLOPS+3 */
 498             /* #endfor */
 499
 500             /* Calculate squared distance and things based on it */
 501             /* #for I,J in PAIRS_IJ */
 502             rsq{I}{J}            = gmx_mm256_calc_rsq_ps(dx{I}{J},dy{I}{J},dz{I}{J});
 503             /*     #define INNERFLOPS INNERFLOPS+5 */
 504             /* #endfor */
 505
 506             /* #for I,J in PAIRS_IJ */
 507             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 508             rinv{I}{J}           = avx256_invsqrt_f(rsq{I}{J});
 509             /*         #define INNERFLOPS INNERFLOPS+5 */
 510             /*     #endif */
 511             /* #endfor */
 512
 513             /* #for I,J in PAIRS_IJ */
 514             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 515             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 516             rinvsq{I}{J}         = avx256_inv_f(rsq{I}{J});
 517             /*             #define INNERFLOPS INNERFLOPS+4 */
 518             /*         #else */
 519             rinvsq{I}{J}         = _mm256_mul_ps(rinv{I}{J},rinv{I}{J});
 520             /*             #define INNERFLOPS INNERFLOPS+1 */
 521             /*         #endif */
 522             /*     #endif */
 523             /* #endfor */
 524
 525             /* #if not 'Water' in GEOMETRY_J */
 526             /* Load parameters for j particles */
 527             /*     #for J in PARTICLES_ELEC_J */
 528             jq{J}              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+{J},charge+jnrB+{J},
 529                                                                  charge+jnrC+{J},charge+jnrD+{J},
 530                                                                  charge+jnrE+{J},charge+jnrF+{J},
 531                                                                  charge+jnrG+{J},charge+jnrH+{J});
 532             /*     #endfor */
 533             /*     #for J in PARTICLES_VDW_J */
 534             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 535             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 536             vdwjidx{J}C        = 2*vdwtype[jnrC+{J}];
 537             vdwjidx{J}D        = 2*vdwtype[jnrD+{J}];
 538             vdwjidx{J}E        = 2*vdwtype[jnrE+{J}];
 539             vdwjidx{J}F        = 2*vdwtype[jnrF+{J}];
 540             vdwjidx{J}G        = 2*vdwtype[jnrG+{J}];
 541             vdwjidx{J}H        = 2*vdwtype[jnrH+{J}];
 542             /*     #endfor */
 543             /* #endif */
 544
 545             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 546             /*     #for J in PARTICLES_J */
 547             fjx{J}             = _mm256_setzero_ps();
 548             fjy{J}             = _mm256_setzero_ps();
 549             fjz{J}             = _mm256_setzero_ps();
 550             /*     #endfor */
 551             /* #endif */
 552
 553             /* #for I,J in PAIRS_IJ */
 554
 555             /**************************
 556              * CALCULATE INTERACTIONS *
 557              **************************/
 558
 559             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 560             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 561             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 562             if (gmx_mm256_any_lt(rsq{I}{J},rcutoff2))
 563             {
 564                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 565             }
 566             /*         #endif */
 567             /*         #define INNERFLOPS INNERFLOPS+1 */
 568             /*     #endif */
 569
 570             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 571             r{I}{J}              = _mm256_mul_ps(rsq{I}{J},rinv{I}{J});
 572             /*         #if ROUND == 'Epilogue' */
 573             r{I}{J}              = _mm256_andnot_ps(dummy_mask,r{I}{J});
 574             /*             #define INNERFLOPS INNERFLOPS+1 */
 575             /*         #endif */
 576             /*         #define INNERFLOPS INNERFLOPS+1 */
 577             /*     #endif */
 578
 579             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 580             /*     #if not 'Water' in GEOMETRY_J */
 581             /* Compute parameters for interactions between i and j atoms */
 582             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 583             qq{I}{J}             = _mm256_mul_ps(iq{I},jq{J});
 584             /*             #define INNERFLOPS INNERFLOPS+1 */
 585             /*         #endif */
 586             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 587             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr{I}+vdwjidx{J}A,
 588                                             vdwioffsetptr{I}+vdwjidx{J}B,
 589                                             vdwioffsetptr{I}+vdwjidx{J}C,
 590                                             vdwioffsetptr{I}+vdwjidx{J}D,
 591                                             vdwioffsetptr{I}+vdwjidx{J}E,
 592                                             vdwioffsetptr{I}+vdwjidx{J}F,
 593                                             vdwioffsetptr{I}+vdwjidx{J}G,
 594                                             vdwioffsetptr{I}+vdwjidx{J}H,
 595                                             &c6_{I}{J},&c12_{I}{J});
 596
 597             /*         #if 'LJEwald' in KERNEL_VDW */
 598             c6grid_{I}{J}       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr{I}+vdwjidx{J}A,
 599                                                                   vdwgridioffsetptr{I}+vdwjidx{J}B,
 600                                                                   vdwgridioffsetptr{I}+vdwjidx{J}C,
 601                                                                   vdwgridioffsetptr{I}+vdwjidx{J}D,
 602                                                                   vdwgridioffsetptr{I}+vdwjidx{J}E,
 603                                                                   vdwgridioffsetptr{I}+vdwjidx{J}F,
 604                                                                   vdwgridioffsetptr{I}+vdwjidx{J}G,
 605                                                                   vdwgridioffsetptr{I}+vdwjidx{J}H);
 606             /*          #endif */
 607             /*         #endif */
 608             /*     #endif */
 609
 610             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 611             /* Calculate table index by multiplying r with table scale and truncate to integer */
 612             rt               = _mm256_mul_ps(r{I}{J},vftabscale);
 613             vfitab           = _mm256_cvttps_epi32(rt);
 614             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 615             /*         #define INNERFLOPS INNERFLOPS+4                          */
 616             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 617             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 618             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 619             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 620             /*             ## 3 tables, 4 bytes per point: multiply index by 12 */
 621             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 622             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 623             /*         #elif 'Table' in KERNEL_ELEC                             */
 624             /*             ## 1 table, 4 bytes per point: multiply index by 4   */
 625             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 626             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 627             /*         #elif 'Table' in KERNEL_VDW */
 628             /*             ## 2 tables, 4 bytes per point: multiply index by 8  */
 629             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 630             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 631             /*         #endif                                                   */
 632             /*     #endif */
 633
 634             /*     ## ELECTROSTATIC INTERACTIONS */
 635             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 636
 637             /*         #if KERNEL_ELEC=='Coulomb' */
 638
 639             /* COULOMB ELECTROSTATICS */
 640             velec            = _mm256_mul_ps(qq{I}{J},rinv{I}{J});
 641             /*             #define INNERFLOPS INNERFLOPS+1 */
 642             /*             #if 'Force' in KERNEL_VF */
 643             felec            = _mm256_mul_ps(velec,rinvsq{I}{J});
 644             /*                 #define INNERFLOPS INNERFLOPS+1 */
 645             /*             #endif */
 646
 647             /*         #elif KERNEL_ELEC=='ReactionField' */
 648
 649             /* REACTION-FIELD ELECTROSTATICS */
 650             /*             #if 'Potential' in KERNEL_VF */
 651             velec            = _mm256_mul_ps(qq{I}{J},_mm256_sub_ps(_mm256_add_ps(rinv{I}{J},_mm256_mul_ps(krf,rsq{I}{J})),crf));
 652             /*                 #define INNERFLOPS INNERFLOPS+4 */
 653             /*             #endif */
 654             /*             #if 'Force' in KERNEL_VF */
 655             felec            = _mm256_mul_ps(qq{I}{J},_mm256_sub_ps(_mm256_mul_ps(rinv{I}{J},rinvsq{I}{J}),krf2));
 656             /*                 #define INNERFLOPS INNERFLOPS+3 */
 657             /*             #endif */
 658
 659             /*         #elif KERNEL_ELEC=='Ewald' */
 660             /* EWALD ELECTROSTATICS */
 661
 662             /* Analytical PME correction */
 663             zeta2            = _mm256_mul_ps(beta2,rsq{I}{J});
 664             /*             #if 'Force' in KERNEL_VF */
 665             rinv3            = _mm256_mul_ps(rinvsq{I}{J},rinv{I}{J});
 666             pmecorrF         = avx256_pmecorrF_f(zeta2);
 667             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 668             felec            = _mm256_mul_ps(qq{I}{J},felec);
 669             /*                 #define INNERFLOPS INNERFLOPS+31 */
 670             /*             #endif */
 671             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 672             pmecorrV         = avx256_pmecorrV_f(zeta2);
 673             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 674             /*                 #define INNERFLOPS INNERFLOPS+27 */
 675             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 676             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv{I}{J},sh_ewald),pmecorrV);
 677             /*                     #define INNERFLOPS INNERFLOPS+21 */
 678             /*                 #else */
 679             velec            = _mm256_sub_ps(rinv{I}{J},pmecorrV);
 680             /*                 #endif */
 681             velec            = _mm256_mul_ps(qq{I}{J},velec);
 682             /*             #endif */
 683
 684             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 685
 686             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 687             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 688                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 689             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 690                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 691             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 692                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 693             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 694                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 695             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 696             Heps             = _mm256_mul_ps(vfeps,H);
 697             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 698             /*             #define INNERFLOPS INNERFLOPS+4 */
 699             /*             #if 'Potential' in KERNEL_VF */
 700             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 701             velec            = _mm256_mul_ps(qq{I}{J},VV);
 702             /*                 #define INNERFLOPS INNERFLOPS+3 */
 703             /*             #endif */
 704             /*             #if 'Force' in KERNEL_VF */
 705             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 706             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq{I}{J},FF),_mm256_mul_ps(vftabscale,rinv{I}{J})));
 707             /*                 #define INNERFLOPS INNERFLOPS+7 */
 708             /*             #endif */
 709             /*         #endif */
 710             /*         ## End of check for electrostatics interaction forms */
 711             /*     #endif */
 712             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 713
 714             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 715
 716             /*         #if KERNEL_VDW=='LennardJones' */
 717
 718             /* LENNARD-JONES DISPERSION/REPULSION */
 719
 720             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 721             /*             #define INNERFLOPS INNERFLOPS+2 */
 722             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 723             vvdw6            = _mm256_mul_ps(c6_{I}{J},rinvsix);
 724             vvdw12           = _mm256_mul_ps(c12_{I}{J},_mm256_mul_ps(rinvsix,rinvsix));
 725             /*                 #define INNERFLOPS INNERFLOPS+3 */
 726             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 727             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_{I}{J},_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 728                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
 729             /*                     #define INNERFLOPS INNERFLOPS+8 */
 730             /*                 #else */
 731             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 732             /*                     #define INNERFLOPS INNERFLOPS+3 */
 733             /*                 #endif */
 734             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 735             /*                 #if 'Force' in KERNEL_VF */
 736             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq{I}{J});
 737             /*                     #define INNERFLOPS INNERFLOPS+2 */
 738             /*                 #endif */
 739             /*             #elif KERNEL_VF=='Force' */
 740             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 741             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_{I}{J},rinvsix),c6_{I}{J}),_mm256_mul_ps(rinvsix,rinvsq{I}{J}));
 742             /*                 #define INNERFLOPS INNERFLOPS+4 */
 743             /*             #endif */
 744
 745             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 746
 747             /* CUBIC SPLINE TABLE DISPERSION */
 748             /*             #if 'Table' in KERNEL_ELEC */
 749             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 750             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 751             /*             #endif                     */
 752             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 753                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 754             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 755                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 756             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 757                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 758             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 759                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 760             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 761             Heps             = _mm256_mul_ps(vfeps,H);
 762             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 763             /*             #define INNERFLOPS INNERFLOPS+4 */
 764             /*             #if 'Potential' in KERNEL_VF */
 765             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 766             vvdw6            = _mm256_mul_ps(c6_{I}{J},VV);
 767             /*                 #define INNERFLOPS INNERFLOPS+3 */
 768             /*             #endif */
 769             /*             #if 'Force' in KERNEL_VF */
 770             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 771             fvdw6            = _mm256_mul_ps(c6_{I}{J},FF);
 772             /*                 #define INNERFLOPS INNERFLOPS+4 */
 773             /*             #endif */
 774
 775             /* CUBIC SPLINE TABLE REPULSION */
 776             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 777             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 778             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 779                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 780             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 781                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 782             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 783                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 784             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 785                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 786             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 787             Heps             = _mm256_mul_ps(vfeps,H);
 788             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 789             /*             #define INNERFLOPS INNERFLOPS+4 */
 790             /*             #if 'Potential' in KERNEL_VF */
 791             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 792             vvdw12           = _mm256_mul_ps(c12_{I}{J},VV);
 793             /*                 #define INNERFLOPS INNERFLOPS+3 */
 794             /*             #endif */
 795             /*             #if 'Force' in KERNEL_VF */
 796             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 797             fvdw12           = _mm256_mul_ps(c12_{I}{J},FF);
 798             /*                 #define INNERFLOPS INNERFLOPS+5 */
 799             /*             #endif */
 800             /*             #if 'Potential' in KERNEL_VF */
 801             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 802             /*                 #define INNERFLOPS INNERFLOPS+1 */
 803             /*             #endif */
 804             /*             #if 'Force' in KERNEL_VF */
 805             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv{I}{J})));
 806             /*                 #define INNERFLOPS INNERFLOPS+4 */
 807             /*             #endif */
 808
 809             /*         #elif KERNEL_VDW=='LJEwald' */
 810
 811             /* Analytical LJ-PME */
 812             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 813             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq{I}{J});
 814             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 815             exponent         = avx256_exp_f(ewcljrsq);
 816             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 817             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 818             /*                 #define INNERFLOPS INNERFLOPS+11 */
 819             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 820             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 821             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_{I}{J},_mm256_mul_ps(c6grid_{I}{J},_mm256_sub_ps(one,poly))),rinvsix);
 822             vvdw12           = _mm256_mul_ps(c12_{I}{J},_mm256_mul_ps(rinvsix,rinvsix));
 823             /*                 #define INNERFLOPS INNERFLOPS+6 */
 824             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 825             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_{I}{J},_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 826                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_{I}{J},sh_vdw_invrcut6),_mm256_mul_ps(c6grid_{I}{J},sh_lj_ewald))),one_sixth));
 827             /*                     #define INNERFLOPS INNERFLOPS+10 */
 828             /*                 #else */
 829             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 830             /*                 #define INNERFLOPS INNERFLOPS+3 */
 831             /*                 #endif */
 832             /*                  ## Check for force inside potential check, i.e. this means we already did the potential part */
 833             /*                  #if 'Force' in KERNEL_VF */
 834             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 835             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_{I}{J},one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq{I}{J});
 836             /*                 #define INNERFLOPS INNERFLOPS+6 */
 837             /*                  #endif */
 838             /*              #elif KERNEL_VF=='Force' */
 839             /* f6A = 6 * C6grid * (1 - poly) */
 840             f6A              = _mm256_mul_ps(c6grid_{I}{J},_mm256_sub_ps(one,poly));
 841             /* f6B = C6grid * exponent * beta^6 */
 842             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_{I}{J},one_sixth),_mm256_mul_ps(exponent,ewclj6));
 843             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 844             fvdw              = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_{I}{J},rinvsix),_mm256_sub_ps(c6_{I}{J},f6A)),rinvsix),f6B),rinvsq{I}{J});
 845             /*                 #define INNERFLOPS INNERFLOPS+11 */
 846             /*              #endif */
 847             /*         #endif */
 848             /*         ## End of check for vdw interaction forms */
 849             /*     #endif */
 850             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 851
 852             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 853             d                = _mm256_sub_ps(r{I}{J},rswitch);
 854             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 855             d2               = _mm256_mul_ps(d,d);
 856             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 857             /*         #define INNERFLOPS INNERFLOPS+10 */
 858
 859             /*         #if 'Force' in KERNEL_VF */
 860             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 861             /*             #define INNERFLOPS INNERFLOPS+5 */
 862             /*         #endif */
 863
 864             /* Evaluate switch function */
 865             /*         #if 'Force' in KERNEL_VF */
 866             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 867             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 868             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv{I}{J},_mm256_mul_ps(velec,dsw)) );
 869             /*                 #define INNERFLOPS INNERFLOPS+4 */
 870             /*             #endif */
 871             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 872             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv{I}{J},_mm256_mul_ps(vvdw,dsw)) );
 873             /*                 #define INNERFLOPS INNERFLOPS+4 */
 874             /*             #endif */
 875             /*         #endif */
 876             /*         #if 'Potential' in KERNEL_VF */
 877             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 878             velec            = _mm256_mul_ps(velec,sw);
 879             /*                 #define INNERFLOPS INNERFLOPS+1 */
 880             /*             #endif */
 881             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 882             vvdw             = _mm256_mul_ps(vvdw,sw);
 883             /*                 #define INNERFLOPS INNERFLOPS+1 */
 884             /*             #endif */
 885             /*         #endif */
 886             /*     #endif */
 887             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 888             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 889             cutoff_mask      = _mm256_cmp_ps(rsq{I}{J},rcutoff2,_CMP_LT_OQ);
 890             /*         #define INNERFLOPS INNERFLOPS+1 */
 891             /*     #endif */
 892
 893             /*     #if 'Potential' in KERNEL_VF */
 894             /* Update potential sum for this i atom from the interaction with this j atom. */
 895             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 896             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 897             velec            = _mm256_and_ps(velec,cutoff_mask);
 898             /*                 #define INNERFLOPS INNERFLOPS+1 */
 899             /*             #endif                                       */
 900             /*             #if ROUND == 'Epilogue' */
 901             velec            = _mm256_andnot_ps(dummy_mask,velec);
 902             /*             #endif */
 903             velecsum         = _mm256_add_ps(velecsum,velec);
 904             /*             #define INNERFLOPS INNERFLOPS+1 */
 905             /*         #endif */
 906             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 907             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 908             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 909             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 910             /*                 #define INNERFLOPS INNERFLOPS+1 */
 911             /*             #endif                                       */
 912             /*             #if ROUND == 'Epilogue' */
 913             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 914             /*             #endif */
 915             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 916             /*             #define INNERFLOPS INNERFLOPS+1 */
 917             /*         #endif */
 918             /*     #endif */
 919
 920             /*     #if 'Force' in KERNEL_VF */
 921
 922             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 923             fscal            = _mm256_add_ps(felec,fvdw);
 924             /*             #define INNERFLOPS INNERFLOPS+1 */
 925             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 926             fscal            = felec;
 927             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 928             fscal            = fvdw;
 929             /*        #endif */
 930
 931             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 932             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 933             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 934             /*                 #define INNERFLOPS INNERFLOPS+1 */
 935             /*             #endif                                       */
 936
 937             /*             #if ROUND == 'Epilogue' */
 938             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 939             /*             #endif */
 940
 941             /* Calculate temporary vectorial force */
 942             tx               = _mm256_mul_ps(fscal,dx{I}{J});
 943             ty               = _mm256_mul_ps(fscal,dy{I}{J});
 944             tz               = _mm256_mul_ps(fscal,dz{I}{J});
 945
 946             /* Update vectorial force */
 947             fix{I}             = _mm256_add_ps(fix{I},tx);
 948             fiy{I}             = _mm256_add_ps(fiy{I},ty);
 949             fiz{I}             = _mm256_add_ps(fiz{I},tz);
 950             /*             #define INNERFLOPS INNERFLOPS+6 */
 951
 952             /* #if GEOMETRY_I == 'Particle'             */
 953             /*     #if ROUND == 'Loop' */
 954             fjptrA             = f+j_coord_offsetA;
 955             fjptrB             = f+j_coord_offsetB;
 956             fjptrC             = f+j_coord_offsetC;
 957             fjptrD             = f+j_coord_offsetD;
 958             fjptrE             = f+j_coord_offsetE;
 959             fjptrF             = f+j_coord_offsetF;
 960             fjptrG             = f+j_coord_offsetG;
 961             fjptrH             = f+j_coord_offsetH;
 962             /*     #else */
 963             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 964             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 965             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 966             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 967             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 968             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 969             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 970             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 971             /*     #endif */
 972             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 973             /*     #define INNERFLOPS INNERFLOPS+3      */
 974             /* #else                                    */
 975             fjx{J}             = _mm256_add_ps(fjx{J},tx);
 976             fjy{J}             = _mm256_add_ps(fjy{J},ty);
 977             fjz{J}             = _mm256_add_ps(fjz{J},tz);
 978             /*     #define INNERFLOPS INNERFLOPS+3      */
 979             /* #endif                                   */
 980
 981             /*     #endif */
 982
 983             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 984             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 985             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
 986             {
 987                 /*     #endif */
 988             }
 989             /*     #endif */
 990             /*    ## End of check for the interaction being outside the cutoff */
 991
 992             /* #endfor */
 993             /* ## End of loop over i-j interaction pairs */
 994
 995             /* #if GEOMETRY_I != 'Particle' */
 996             /*     #if ROUND == 'Loop' */
 997             fjptrA             = f+j_coord_offsetA;
 998             fjptrB             = f+j_coord_offsetB;
 999             fjptrC             = f+j_coord_offsetC;
1000             fjptrD             = f+j_coord_offsetD;
1001             fjptrE             = f+j_coord_offsetE;
1002             fjptrF             = f+j_coord_offsetF;
1003             fjptrG             = f+j_coord_offsetG;
1004             fjptrH             = f+j_coord_offsetH;
1005             /*     #else */
1006             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1007             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1008             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1009             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1010             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1011             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1012             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1013             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1014             /*     #endif */
1015             /* #endif */
1016
1017             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
1018             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1019             /*     #define INNERFLOPS INNERFLOPS+3      */
1020             /* #elif GEOMETRY_J == 'Water3'             */
1021             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1022                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1023             /*     #define INNERFLOPS INNERFLOPS+9      */
1024             /* #elif GEOMETRY_J == 'Water4'             */
1025             /*     #if 0 in PARTICLES_J                 */
1026             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1027                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1028                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1029             /*     #define INNERFLOPS INNERFLOPS+12     */
1030             /*     #else                                */
1031             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1032                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1033                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1034             /*     #define INNERFLOPS INNERFLOPS+9      */
1035             /*     #endif                               */
1036             /* #endif                                   */
1037
1038             /* Inner loop uses {INNERFLOPS} flops */
1039         }
1040
1041         /* #endfor */
1042
1043         /* End of innermost loop */
1044
1045         /* #if 'Force' in KERNEL_VF */
1046         /*     #if GEOMETRY_I == 'Particle'            */
1047         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1048                                                  f+i_coord_offset,fshift+i_shift_offset);
1049         /*         #define OUTERFLOPS OUTERFLOPS+6     */
1050         /*     #elif GEOMETRY_I == 'Water3'            */
1051         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1052                                                  f+i_coord_offset,fshift+i_shift_offset);
1053         /*         #define OUTERFLOPS OUTERFLOPS+18    */
1054         /*     #elif GEOMETRY_I == 'Water4'            */
1055         /*         #if 0 in PARTICLES_I                */
1056         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1057                                                  f+i_coord_offset,fshift+i_shift_offset);
1058         /*             #define OUTERFLOPS OUTERFLOPS+24    */
1059         /*         #else                               */
1060         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1061                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1062         /*             #define OUTERFLOPS OUTERFLOPS+18    */
1063         /*         #endif                              */
1064         /*     #endif                                  */
1065         /* #endif                                      */
1066
1067         /* #if 'Potential' in KERNEL_VF */
1068         ggid                        = gid[iidx];
1069         /* Update potential energies */
1070         /*     #if KERNEL_ELEC != 'None' */
1071         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1072         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1073         /*     #endif */
1074         /*     #if KERNEL_VDW != 'None' */
1075         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1076         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1077         /*     #endif */
1078         /* #endif */
1079
1080         /* Increment number of inner iterations */
1081         inneriter                  += j_index_end - j_index_start;
1082
1083         /* Outer loop uses {OUTERFLOPS} flops */
1084     }
1085
1086     /* Increment number of outer iterations */
1087     outeriter        += nri;
1088
1089     /* Update outer/inner flops */
1090     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1091     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1092     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1093     /* #if GEOMETRY_I == 'Water3'            */
1094     /*     #define ISUFFIX '_W3'             */
1095     /* #elif GEOMETRY_I == 'Water4'          */
1096     /*     #define ISUFFIX '_W4'             */
1097     /* #else                                 */
1098     /*     #define ISUFFIX ''                */
1099     /* #endif                                */
1100     /* #if GEOMETRY_J == 'Water3'            */
1101     /*     #define JSUFFIX 'W3'              */
1102     /* #elif GEOMETRY_J == 'Water4'          */
1103     /*     #define JSUFFIX 'W4'              */
1104     /* #else                                 */
1105     /*     #define JSUFFIX ''                */
1106     /* #endif                                */
1107     /* #if 'PotentialAndForce' in KERNEL_VF  */
1108     /*     #define VFSUFFIX  '_VF'           */
1109     /* #elif 'Potential' in KERNEL_VF        */
1110     /*     #define VFSUFFIX '_V'             */
1111     /* #else                                 */
1112     /*     #define VFSUFFIX '_F'             */
1113     /* #endif                                */
1114
1115     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1116     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1117     /* #elif KERNEL_ELEC != 'None' */
1118     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1119     /* #else */
1120     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1121     /* #endif  */
1122 }