src/gmxlib/nonbonded/nb_kernel_c/nb_kernel_template_c.pre

   1 /* #if 0 */
   2 #error This file must be processed with the Gromacs pre-preprocessor
   3 /* #endif */
   4 /* #if INCLUDE_HEADER */
   5 #ifdef HAVE_CONFIG_H
   6 #include <config.h>
   7 #endif
   8
   9 #include <math.h>
  10
  11 #include "../nb_kernel.h"
  12 #include "types/simple.h"
  13 #include "vec.h"
  14 #include "nrnb.h"
  15 /* #endif */
  16
  17 /* ## List of variables set by the generating script:                                    */
  18 /* ##                                                                                    */
  19 /* ## Setttings that apply to the entire kernel:                                         */
  20 /* ## KERNEL_ELEC:           String, choice for electrostatic interactions               */
  21 /* ## KERNEL_VDW:            String, choice for van der Waals interactions               */
  22 /* ## KERNEL_NAME:           String, name of this kernel                                 */
  23 /* ## KERNEL_VF:             String telling if we calculate potential, force, or both    */
  24 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
  25 /* ##                                                                                    */
  26 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops:             */
  27 /* ## PARTICLES_I[]/         Arrays with lists of i/j particles to use in kernel. It is  */
  28 /* ## PARTICLES_J[]:         just [0] for particle geometry, but can be longer for water */
  29 /* ## PARTICLES_ELEC_I[]/    Arrays with lists of i/j particle that have electrostatics  */
  30 /* ## PARTICLES_ELEC_J[]:    interactions that should be calculated in this kernel.      */
  31 /* ## PARTICLES_VDW_I[]/     Arrays with the list of i/j particle that have VdW          */
  32 /* ## PARTICLES_VDW_J[]:     interactions that should be calculated in this kernel.      */
  33 /* ##                                                                                    */
  34 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle)   */
  35 /* ## PAIRS_IJ[]:            Array with (i,j) tuples of pairs for which interactions     */
  36 /* ##                        should be calculated in this kernel. Zero-charge particles  */
  37 /* ##                        do not have interactions with particles without vdw, and    */
  38 /* ##                        Vdw-only interactions are not evaluated in a no-vdw-kernel. */
  39 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
  40 /* ##                        For each i-j pair, the element [I][J] is a list of strings  */
  41 /* ##                        defining properties/flags of this interaction. Examples     */
  42 /* ##                        include 'electrostatics'/'vdw' if that type of interaction  */
  43 /* ##                        should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values  */
  44 /* ##                        are needed, and 'exactcutoff' or 'shift','switch' to        */
  45 /* ##                        decide if the force/potential should be modified. This way  */
  46 /* ##                        we only calculate values absolutely needed for each case.   */
  47
  48 /* ## Calculate the size and offset for (merged/interleaved) table data */
  49
  50 /* #if ('CubicSplineTable' in [KERNEL_ELEC,KERNEL_VDW]) or KERNEL_VF=='PotentialAndForce' */
  51 /*     #define TABLE_POINT_SIZE 4 */
  52 /* #else */
  53 /*     #define TABLE_POINT_SIZE 2 */
  54 /* #endif */
  55
  56 /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
  57 /*     #define TABLE_NINTERACTIONS 3 */
  58 /*     #define TABLE_VDW_OFFSET TABLE_POINT_SIZE */
  59 /* #elif 'Table' in KERNEL_ELEC */
  60 /*     #define TABLE_NINTERACTIONS 1 */
  61 /* #elif 'Table' in KERNEL_VDW */
  62 /*     #define TABLE_NINTERACTIONS 2 */
  63 /*     #define TABLE_VDW_OFFSET 0 */
  64 /* #else */
  65 /*     #define TABLE_NINTERACTIONS 0 */
  66 /* #endif */
  67
  68 /* #if 'Buckingham' in KERNEL_VDW */
  69 /*   #define NVDWPARAM 3 */
  70 /* #else */
  71 /*   #define NVDWPARAM 2 */
  72 /* #endif */
  73
  74 /*
  75  * Gromacs nonbonded kernel:   {KERNEL_NAME}
  76  * Electrostatics interaction: {KERNEL_ELEC}
  77  * VdW interaction:            {KERNEL_VDW}
  78  * Geometry:                   {GEOMETRY_I}-{GEOMETRY_J}
  79  * Calculate force/pot:        {KERNEL_VF}
  80  */
  81 void
  82 {KERNEL_NAME}
  83                     (t_nblist * gmx_restrict                nlist,
  84                      rvec * gmx_restrict                    xx,
  85                      rvec * gmx_restrict                    ff,
  86                      t_forcerec * gmx_restrict              fr,
  87                      t_mdatoms * gmx_restrict               mdatoms,
  88                      nb_kernel_data_t * gmx_restrict        kernel_data,
  89                      t_nrnb * gmx_restrict                  nrnb)
  90 {
  91     /* ## Not all variables are used for all kernels, but any optimizing compiler fixes that, */
  92     /* ## so there is no point in going to extremes to exclude variables that are not needed. */
  93     int              i_shift_offset,i_coord_offset,j_coord_offset;
  94     int              j_index_start,j_index_end;
  95     int              nri,inr,ggid,iidx,jidx,jnr,outeriter,inneriter;
  96     real             shX,shY,shZ,tx,ty,tz,fscal,rcutoff,rcutoff2;
  97     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  98     real             *shiftvec,*fshift,*x,*f;
  99     /* #for I in PARTICLES_I */
 100     int              vdwioffset{I};
 101     real             ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 102     /* #endfor */
 103     /* #for J in PARTICLES_J */
 104     int              vdwjidx{J};
 105     real             jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 106     /* #endfor */
 107     /* #for I,J in PAIRS_IJ */
 108     real             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},cexp1_{I}{J},cexp2_{I}{J};
 109     /* #endfor */
 110     /* #if KERNEL_ELEC != 'None' */
 111     real             velec,felec,velecsum,facel,crf,krf,krf2;
 112     real             *charge;
 113     /* #endif */
 114     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 115     int              gbitab;
 116     real             vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 117     real             *invsqrta,*dvda,*gbtab;
 118     /* #endif */
 119     /* #if KERNEL_VDW != 'None' */
 120     int              nvdwtype;
 121     real             rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,br,vvdwexp,sh_vdw_invrcut6;
 122     int              *vdwtype;
 123     real             *vdwparam;
 124     /* #endif */
 125     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 126     int              vfitab;
 127     real             rt,vfeps,vftabscale,Y,F,Geps,Heps2,Fp,VV,FF;
 128     real             *vftab;
 129     /* #endif */
 130     /* #if 'Ewald' in KERNEL_ELEC */
 131     int              ewitab;
 132     real             ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace;
 133     real             *ewtab;
 134     /* #endif */
 135     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 136     real             rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 137     /* #endif */
 138
 139     x                = xx[0];
 140     f                = ff[0];
 141
 142     nri              = nlist->nri;
 143     iinr             = nlist->iinr;
 144     jindex           = nlist->jindex;
 145     jjnr             = nlist->jjnr;
 146     shiftidx         = nlist->shift;
 147     gid              = nlist->gid;
 148     shiftvec         = fr->shift_vec[0];
 149     fshift           = fr->fshift[0];
 150     /* #if KERNEL_ELEC != 'None' */
 151     facel            = fr->epsfac;
 152     charge           = mdatoms->chargeA;
 153     /*     #if 'ReactionField' in KERNEL_ELEC */
 154     krf              = fr->ic->k_rf;
 155     krf2             = krf*2.0;
 156     crf              = fr->ic->c_rf;
 157     /*     #endif */
 158     /* #endif */
 159     /* #if KERNEL_VDW != 'None' */
 160     nvdwtype         = fr->ntype;
 161     vdwparam         = fr->nbfp;
 162     vdwtype          = mdatoms->typeA;
 163     /* #endif */
 164
 165     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 166     vftab            = kernel_data->table_elec_vdw->data;
 167     vftabscale       = kernel_data->table_elec_vdw->scale;
 168     /* #elif 'Table' in KERNEL_ELEC */
 169     vftab            = kernel_data->table_elec->data;
 170     vftabscale       = kernel_data->table_elec->scale;
 171     /* #elif 'Table' in KERNEL_VDW */
 172     vftab            = kernel_data->table_vdw->data;
 173     vftabscale       = kernel_data->table_vdw->scale;
 174     /* #endif */
 175
 176     /* #if 'Ewald' in KERNEL_ELEC */
 177     sh_ewald         = fr->ic->sh_ewald;
 178     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 179     ewtab            = fr->ic->tabq_coul_F;
 180     ewtabscale       = fr->ic->tabq_scale;
 181     ewtabhalfspace   = 0.5/ewtabscale;
 182     /*     #else */
 183     ewtab            = fr->ic->tabq_coul_FDV0;
 184     ewtabscale       = fr->ic->tabq_scale;
 185     ewtabhalfspace   = 0.5/ewtabscale;
 186      /*     #endif */
 187     /* #endif */
 188
 189     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 190     invsqrta         = fr->invsqrta;
 191     dvda             = fr->dvda;
 192     gbtabscale       = fr->gbtab.scale;
 193     gbtab            = fr->gbtab.data;
 194     gbinvepsdiff     = (1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent);
 195     /* #endif */
 196
 197     /* #if 'Water' in GEOMETRY_I */
 198     /* Setup water-specific parameters */
 199     inr              = nlist->iinr[0];
 200     /*     #for I in PARTICLES_ELEC_I */
 201     iq{I}              = facel*charge[inr+{I}];
 202     /*     #endfor */
 203     /*     #for I in PARTICLES_VDW_I */
 204     vdwioffset{I}      = {NVDWPARAM}*nvdwtype*vdwtype[inr+{I}];
 205     /*     #endfor */
 206     /* #endif */
 207
 208     /* #if 'Water' in GEOMETRY_J */
 209     /*     #for J in PARTICLES_ELEC_J */
 210     jq{J}              = charge[inr+{J}];
 211     /*     #endfor */
 212     /*     #for J in PARTICLES_VDW_J */
 213     vdwjidx{J}         = {NVDWPARAM}*vdwtype[inr+{J}];
 214     /*     #endfor */
 215     /*     #for I,J in PAIRS_IJ */
 216     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 217     qq{I}{J}             = iq{I}*jq{J};
 218     /*         #endif */
 219     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 220     /*             #if 'Buckingham' in KERNEL_VDW */
 221     c6_{I}{J}            = vdwparam[vdwioffset{I}+vdwjidx{J}];
 222     cexp1_{I}{J}         = vdwparam[vdwioffset{I}+vdwjidx{J}+1];
 223     cexp2_{I}{J}         = vdwparam[vdwioffset{I}+vdwjidx{J}+2];
 224     /*             #else */
 225     c6_{I}{J}            = vdwparam[vdwioffset{I}+vdwjidx{J}];
 226     c12_{I}{J}           = vdwparam[vdwioffset{I}+vdwjidx{J}+1];
 227     /*             #endif */
 228     /*         #endif */
 229     /*     #endfor */
 230     /* #endif */
 231
 232     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 233     /*     #if KERNEL_ELEC!='None' */
 234     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 235     rcutoff          = fr->rcoulomb;
 236     /*     #else */
 237     rcutoff          = fr->rvdw;
 238     /*     #endif */
 239     rcutoff2         = rcutoff*rcutoff;
 240     /* #endif */
 241
 242     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 243     sh_vdw_invrcut6  = fr->ic->sh_invrc6;
 244     rvdw             = fr->rvdw;
 245     /* #endif */
 246
 247     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 248     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 249     rswitch          = fr->rcoulomb_switch;
 250     /*     #else */
 251     rswitch          = fr->rvdw_switch;
 252     /*     #endif */
 253     /* Setup switch parameters */
 254     d                = rcutoff-rswitch;
 255     swV3             = -10.0/(d*d*d);
 256     swV4             =  15.0/(d*d*d*d);
 257     swV5             =  -6.0/(d*d*d*d*d);
 258     /*     #if 'Force' in KERNEL_VF */
 259     swF2             = -30.0/(d*d*d);
 260     swF3             =  60.0/(d*d*d*d);
 261     swF4             = -30.0/(d*d*d*d*d);
 262     /*     #endif */
 263     /* #endif */
 264
 265     /* ## Keep track of the floating point operations we issue for reporting! */
 266     /* #define OUTERFLOPS 0 */
 267     /* #define INNERFLOPS 0 */
 268     outeriter        = 0;
 269     inneriter        = 0;
 270
 271     /* Start outer loop over neighborlists */
 272     for(iidx=0; iidx<nri; iidx++)
 273     {
 274         /* Load shift vector for this list */
 275         i_shift_offset   = DIM*shiftidx[iidx];
 276         shX              = shiftvec[i_shift_offset+XX];
 277         shY              = shiftvec[i_shift_offset+YY];
 278         shZ              = shiftvec[i_shift_offset+ZZ];
 279
 280         /* Load limits for loop over neighbors */
 281         j_index_start    = jindex[iidx];
 282         j_index_end      = jindex[iidx+1];
 283
 284         /* Get outer coordinate index */
 285         inr              = iinr[iidx];
 286         i_coord_offset   = DIM*inr;
 287
 288         /* Load i particle coords and add shift vector */
 289         /* ## Loop over i particles, but only include ones that we use - skip e.g. vdw-only sites for elec-only kernel */
 290         /*     #for I in PARTICLES_I */
 291         ix{I}              = shX + x[i_coord_offset+DIM*{I}+XX];
 292         iy{I}              = shY + x[i_coord_offset+DIM*{I}+YY];
 293         iz{I}              = shZ + x[i_coord_offset+DIM*{I}+ZZ];
 294         /*     #define OUTERFLOPS OUTERFLOPS+3 */
 295         /* #endfor */
 296
 297         /* #if 'Force' in KERNEL_VF */
 298         /*     #for I in PARTICLES_I */
 299         fix{I}             = 0.0;
 300         fiy{I}             = 0.0;
 301         fiz{I}             = 0.0;
 302         /*     #endfor */
 303         /* #endif */
 304
 305         /* ## For water we already preloaded parameters at the start of the kernel */
 306         /* #if not 'Water' in GEOMETRY_I */
 307         /* Load parameters for i particles */
 308         /*     #for I in PARTICLES_ELEC_I */
 309         iq{I}              = facel*charge[inr+{I}];
 310         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 311         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 312         isai{I}            = invsqrta[inr+{I}];
 313         /*         #endif */
 314         /*     #endfor */
 315         /*     #for I in PARTICLES_VDW_I */
 316         vdwioffset{I}      = {NVDWPARAM}*nvdwtype*vdwtype[inr+{I}];
 317         /*     #endfor */
 318         /* #endif */
 319
 320         /* #if 'Potential' in KERNEL_VF */
 321         /* Reset potential sums */
 322         /*     #if KERNEL_ELEC != 'None' */
 323         velecsum         = 0.0;
 324         /*     #endif */
 325         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 326         vgbsum           = 0.0;
 327         /*     #endif */
 328         /*     #if KERNEL_VDW != 'None' */
 329         vvdwsum          = 0.0;
 330         /*     #endif */
 331         /* #endif */
 332         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 333         dvdasum          = 0.0;
 334         /*     #endif */
 335
 336         /* Start inner kernel loop */
 337         for(jidx=j_index_start; jidx<j_index_end; jidx++)
 338         {
 339             /* Get j neighbor index, and coordinate index */
 340             jnr              = jjnr[jidx];
 341             j_coord_offset   = DIM*jnr;
 342
 343             /* load j atom coordinates */
 344             /* #for J in PARTICLES_J */
 345             jx{J}              = x[j_coord_offset+DIM*{J}+XX];
 346             jy{J}              = x[j_coord_offset+DIM*{J}+YY];
 347             jz{J}              = x[j_coord_offset+DIM*{J}+ZZ];
 348             /* #endfor */
 349
 350             /* Calculate displacement vector */
 351             /* #for I,J in PAIRS_IJ */
 352             dx{I}{J}             = ix{I} - jx{J};
 353             dy{I}{J}             = iy{I} - jy{J};
 354             dz{I}{J}             = iz{I} - jz{J};
 355             /*     #define INNERFLOPS INNERFLOPS+3 */
 356             /* #endfor */
 357
 358             /* Calculate squared distance and things based on it */
 359             /* #for I,J in PAIRS_IJ */
 360             rsq{I}{J}            = dx{I}{J}*dx{I}{J}+dy{I}{J}*dy{I}{J}+dz{I}{J}*dz{I}{J};
 361             /*     #define INNERFLOPS INNERFLOPS+5 */
 362             /* #endfor */
 363
 364             /* #for I,J in PAIRS_IJ */
 365             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 366             rinv{I}{J}           = gmx_invsqrt(rsq{I}{J});
 367             /*         #define INNERFLOPS INNERFLOPS+5 */
 368             /*     #endif */
 369             /* #endfor */
 370
 371             /* #for I,J in PAIRS_IJ */
 372             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 373             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 374             rinvsq{I}{J}         = 1.0/rsq{I}{J};
 375             /*             #define INNERFLOPS INNERFLOPS+4 */
 376             /*         #else */
 377             rinvsq{I}{J}         = rinv{I}{J}*rinv{I}{J};
 378             /*             #define INNERFLOPS INNERFLOPS+1 */
 379             /*         #endif */
 380             /*     #endif */
 381             /* #endfor */
 382
 383             /* #if not 'Water' in GEOMETRY_J */
 384             /* Load parameters for j particles */
 385             /*     #for J in PARTICLES_ELEC_J */
 386             jq{J}              = charge[jnr+{J}];
 387             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 388             isaj{J}           = invsqrta[jnr+{J}];
 389             /*         #endif */
 390             /*     #endfor */
 391             /*     #for J in PARTICLES_VDW_J */
 392             vdwjidx{J}         = {NVDWPARAM}*vdwtype[jnr+{J}];
 393             /*     #endfor */
 394             /* #endif */
 395
 396             /* #for I,J in PAIRS_IJ */
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 403             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 404             if (rsq{I}{J}<rcutoff2)
 405             {
 406                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 407             }
 408             /*         #endif */
 409             /*     #endif */
 410
 411             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 412             r{I}{J}              = rsq{I}{J}*rinv{I}{J};
 413             /*         #define INNERFLOPS INNERFLOPS+1 */
 414             /*     #endif */
 415
 416             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 417             /*     #if not 'Water' in GEOMETRY_J */
 418             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 419             qq{I}{J}             = iq{I}*jq{J};
 420             /*             #define INNERFLOPS INNERFLOPS+1 */
 421             /*         #endif */
 422             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 423             /*             #if KERNEL_VDW=='Buckingham' */
 424             c6_{I}{J}            = vdwparam[vdwioffset{I}+vdwjidx{J}];
 425             cexp1_{I}{J}         = vdwparam[vdwioffset{I}+vdwjidx{J}+1];
 426             cexp2_{I}{J}         = vdwparam[vdwioffset{I}+vdwjidx{J}+2];
 427             /*             #else */
 428             c6_{I}{J}            = vdwparam[vdwioffset{I}+vdwjidx{J}];
 429             c12_{I}{J}           = vdwparam[vdwioffset{I}+vdwjidx{J}+1];
 430             /*             #endif */
 431             /*         #endif */
 432             /*     #endif */
 433
 434             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 435             /* Calculate table index by multiplying r with table scale and truncate to integer */
 436             rt               = r{I}{J}*vftabscale;
 437             vfitab           = rt;
 438             vfeps            = rt-vfitab;
 439             vfitab           = {TABLE_NINTERACTIONS}*{TABLE_POINT_SIZE}*vfitab;
 440             /*         #define INNERFLOPS INNERFLOPS+2 */
 441             /*     #endif */
 442
 443             /*     ## ELECTROSTATIC INTERACTIONS */
 444             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 445
 446             /*         #if KERNEL_ELEC=='Coulomb' */
 447
 448             /* COULOMB ELECTROSTATICS */
 449             velec            = qq{I}{J}*rinv{I}{J};
 450             /*             #define INNERFLOPS INNERFLOPS+1 */
 451             /*             #if 'Force' in KERNEL_VF */
 452             felec            = velec*rinvsq{I}{J};
 453             /*                 #define INNERFLOPS INNERFLOPS+2 */
 454             /*             #endif */
 455
 456             /*         #elif KERNEL_ELEC=='ReactionField' */
 457
 458             /* REACTION-FIELD ELECTROSTATICS */
 459             /*             #if 'Potential' in KERNEL_VF */
 460             velec            = qq{I}{J}*(rinv{I}{J}+krf*rsq{I}{J}-crf);
 461             /*                 #define INNERFLOPS INNERFLOPS+4 */
 462             /*             #endif */
 463             /*             #if 'Force' in KERNEL_VF */
 464             felec            = qq{I}{J}*(rinv{I}{J}*rinvsq{I}{J}-krf2);
 465             /*                 #define INNERFLOPS INNERFLOPS+3 */
 466             /*             #endif */
 467
 468             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 469
 470             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 471             isaprod          = isai{I}*isaj{J};
 472             gbqqfactor       = isaprod*(-qq{I}{J})*gbinvepsdiff;
 473             gbscale          = isaprod*gbtabscale;
 474             dvdaj            = dvda[jnr+{J}];
 475             /*             #define INNERFLOPS INNERFLOPS+5 */
 476
 477             /* Calculate generalized born table index - this is a separate table from the normal one,
 478              * but we use the same procedure by multiplying r with scale and truncating to integer.
 479              */
 480             rt               = r{I}{J}*gbscale;
 481             gbitab           = rt;
 482             gbeps            = rt-gbitab;
 483             gbitab           = 4*gbitab;
 484
 485             Y                = gbtab[gbitab];
 486             F                = gbtab[gbitab+1];
 487             Geps             = gbeps*gbtab[gbitab+2];
 488             Heps2            = gbeps*gbeps*gbtab[gbitab+3];
 489             Fp               = F+Geps+Heps2;
 490             VV               = Y+gbeps*Fp;
 491             vgb              = gbqqfactor*VV;
 492             /*             #define INNERFLOPS INNERFLOPS+10 */
 493
 494             /*             #if 'Force' in KERNEL_VF */
 495             FF               = Fp+Geps+2.0*Heps2;
 496             fgb              = gbqqfactor*FF*gbscale;
 497             dvdatmp          = -0.5*(vgb+fgb*r{I}{J});
 498             dvdasum          = dvdasum + dvdatmp;
 499             dvda[jnr]        = dvdaj+dvdatmp*isaj{J}*isaj{J};
 500             /*                 #define INNERFLOPS INNERFLOPS+13 */
 501             /*             #endif */
 502             velec            = qq{I}{J}*rinv{I}{J};
 503             /*                 #define INNERFLOPS INNERFLOPS+1 */
 504             /*             #if 'Force' in KERNEL_VF */
 505             felec            = (velec*rinv{I}{J}-fgb)*rinv{I}{J};
 506             /*                 #define INNERFLOPS INNERFLOPS+3 */
 507             /*             #endif */
 508
 509             /*         #elif KERNEL_ELEC=='Ewald' */
 510             /* EWALD ELECTROSTATICS */
 511
 512             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 513             ewrt             = r{I}{J}*ewtabscale;
 514             ewitab           = ewrt;
 515             eweps            = ewrt-ewitab;
 516             /*             #define INNERFLOPS INNERFLOPS+2 */
 517             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 518             ewitab           = 4*ewitab;
 519             felec            = ewtab[ewitab]+eweps*ewtab[ewitab+1];
 520             /*                 #define INNERFLOPS INNERFLOPS+4 */
 521             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 522             velec            = qq{I}{J}*((rinv{I}{J}-sh_ewald)-(ewtab[ewitab+2]-ewtabhalfspace*eweps*(ewtab[ewitab]+felec)));
 523             /*                     #define INNERFLOPS INNERFLOPS+7 */
 524             /*                 #else */
 525             velec            = qq{I}{J}*(rinv{I}{J}-(ewtab[ewitab+2]-ewtabhalfspace*eweps*(ewtab[ewitab]+felec)));
 526             /*                     #define INNERFLOPS INNERFLOPS+6 */
 527             /*                 #endif */
 528             /*                 #if 'Force' in KERNEL_VF */
 529             felec            = qq{I}{J}*rinv{I}{J}*(rinvsq{I}{J}-felec);
 530             /*                      #define INNERFLOPS INNERFLOPS+3 */
 531             /*                 #endif */
 532             /*             #elif KERNEL_VF=='Force' */
 533             felec            = (1.0-eweps)*ewtab[ewitab]+eweps*ewtab[ewitab+1];
 534             felec            = qq{I}{J}*rinv{I}{J}*(rinvsq{I}{J}-felec);
 535             /*                 #define INNERFLOPS INNERFLOPS+7 */
 536             /*             #endif */
 537
 538             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 539
 540             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 541             /*             #if 'Force' in KERNEL_VF */
 542             Y                = vftab[vfitab];
 543             /*             #endif */
 544             F                = vftab[vfitab+1];
 545             Geps             = vfeps*vftab[vfitab+2];
 546             Heps2            = vfeps*vfeps*vftab[vfitab+3];
 547             Fp               = F+Geps+Heps2;
 548             /*             #define INNERFLOPS INNERFLOPS+5 */
 549             /*             #if 'Potential' in KERNEL_VF */
 550             VV               = Y+vfeps*Fp;
 551             velec            = qq{I}{J}*VV;
 552             /*                 #define INNERFLOPS INNERFLOPS+3 */
 553             /*             #endif */
 554             /*             #if 'Force' in KERNEL_VF */
 555             FF               = Fp+Geps+2.0*Heps2;
 556             felec            = -qq{I}{J}*FF*vftabscale*rinv{I}{J};
 557             /*                 #define INNERFLOPS INNERFLOPS+7 */
 558             /*             #endif */
 559             /*         #endif */
 560             /*         ## End of check for electrostatics interaction forms */
 561             /*     #endif */
 562             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 563
 564             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 565
 566             /*         #if KERNEL_VDW=='LennardJones' */
 567
 568             /* LENNARD-JONES DISPERSION/REPULSION */
 569
 570             rinvsix          = rinvsq{I}{J}*rinvsq{I}{J}*rinvsq{I}{J};
 571             /*             #define INNERFLOPS INNERFLOPS+2 */
 572             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 573             vvdw6            = c6_{I}{J}*rinvsix;
 574             vvdw12           = c12_{I}{J}*rinvsix*rinvsix;
 575             /*                 #define INNERFLOPS INNERFLOPS+3 */
 576             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 577             vvdw             = (vvdw12 - c12_{I}{J}*sh_vdw_invrcut6*sh_vdw_invrcut6)*(1.0/12.0) - (vvdw6 - c6_{I}{J}*sh_vdw_invrcut6)*(1.0/6.0);
 578             /*                     #define INNERFLOPS INNERFLOPS+8 */
 579             /*                 #else */
 580             vvdw             = vvdw12*(1.0/12.0) - vvdw6*(1.0/6.0);
 581             /*                     #define INNERFLOPS INNERFLOPS+3 */
 582             /*                 #endif */
 583             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 584             /*                 #if 'Force' in KERNEL_VF */
 585             fvdw             = (vvdw12-vvdw6)*rinvsq{I}{J};
 586             /*                     #define INNERFLOPS INNERFLOPS+2 */
 587             /*                 #endif */
 588             /*             #elif KERNEL_VF=='Force' */
 589             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 590             fvdw             = (c12_{I}{J}*rinvsix-c6_{I}{J})*rinvsix*rinvsq{I}{J};
 591             /*                 #define INNERFLOPS INNERFLOPS+4 */
 592             /*             #endif */
 593
 594             /*         #elif KERNEL_VDW=='Buckingham' */
 595
 596             /* BUCKINGHAM DISPERSION/REPULSION */
 597             rinvsix          = rinvsq{I}{J}*rinvsq{I}{J}*rinvsq{I}{J};
 598             vvdw6            = c6_{I}{J}*rinvsix;
 599             br               = cexp2_{I}{J}*r{I}{J};
 600             vvdwexp          = cexp1_{I}{J}*exp(-br);
 601             /*             ## Estimate exp() to 25 flops */
 602             /*             #define INNERFLOPS INNERFLOPS+31 */
 603             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch'  */
 604             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 605             vvdw             = (vvdwexp-cexp1_{I}{J}*exp(-cexp2_{I}{J}*rvdw)) - (vvdw6 - c6_{I}{J}*sh_vdw_invrcut6)*(1.0/6.0);
 606             /*                     #define INNERFLOPS INNERFLOPS+33 */
 607             /*                  #else */
 608             vvdw             = vvdwexp - vvdw6*(1.0/6.0);
 609             /*                     #define INNERFLOPS INNERFLOPS+2 */
 610             /*                 #endif */
 611             /*             #endif */
 612             /*             #if 'Force' in KERNEL_VF */
 613             fvdw             = (br*vvdwexp-vvdw6)*rinvsq{I}{J};
 614             /*                 #define INNERFLOPS INNERFLOPS+3 */
 615             /*             #endif */
 616
 617             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 618
 619             /* CUBIC SPLINE TABLE DISPERSION */
 620             vfitab          += {TABLE_VDW_OFFSET};
 621             /*             #if 'Force' in KERNEL_VF */
 622             Y                = vftab[vfitab];
 623             /*             #endif */
 624             F                = vftab[vfitab+1];
 625             Geps             = vfeps*vftab[vfitab+2];
 626             Heps2            = vfeps*vfeps*vftab[vfitab+3];
 627             Fp               = F+Geps+Heps2;
 628             /*             #define INNERFLOPS INNERFLOPS+5 */
 629             /*             #if 'Potential' in KERNEL_VF */
 630             VV               = Y+vfeps*Fp;
 631             vvdw6            = c6_{I}{J}*VV;
 632             /*                 #define INNERFLOPS INNERFLOPS+3 */
 633             /*             #endif */
 634             /*             #if 'Force' in KERNEL_VF */
 635             FF               = Fp+Geps+2.0*Heps2;
 636             fvdw6            = c6_{I}{J}*FF;
 637             /*                 #define INNERFLOPS INNERFLOPS+4 */
 638             /*             #endif */
 639
 640             /* CUBIC SPLINE TABLE REPULSION */
 641             /*             #if 'Force' in KERNEL_VF */
 642             Y                = vftab[vfitab+4];
 643             /*             #endif */
 644             F                = vftab[vfitab+5];
 645             Geps             = vfeps*vftab[vfitab+6];
 646             Heps2            = vfeps*vfeps*vftab[vfitab+7];
 647             Fp               = F+Geps+Heps2;
 648             /*             #define INNERFLOPS INNERFLOPS+5 */
 649             /*             #if 'Potential' in KERNEL_VF */
 650             VV               = Y+vfeps*Fp;
 651             vvdw12           = c12_{I}{J}*VV;
 652             /*                 #define INNERFLOPS INNERFLOPS+3 */
 653             /*             #endif */
 654             /*             #if 'Force' in KERNEL_VF */
 655             FF               = Fp+Geps+2.0*Heps2;
 656             fvdw12           = c12_{I}{J}*FF;
 657             /*                 #define INNERFLOPS INNERFLOPS+4 */
 658             /*             #endif */
 659             /*             #if 'Potential' in KERNEL_VF */
 660             vvdw             = vvdw12+vvdw6;
 661             /*                 #define INNERFLOPS INNERFLOPS+1 */
 662             /*             #endif */
 663             /*             #if 'Force' in KERNEL_VF */
 664             fvdw             = -(fvdw6+fvdw12)*vftabscale*rinv{I}{J};
 665             /*                 #define INNERFLOPS INNERFLOPS+4 */
 666             /*             #endif */
 667             /*         #endif */
 668             /*         ## End of check for vdw interaction forms */
 669             /*     #endif */
 670             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 671
 672             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 673             d                = r{I}{J}-rswitch;
 674             d                = (d>0.0) ? d : 0.0;
 675             d2               = d*d;
 676             sw               = 1.0+d2*d*(swV3+d*(swV4+d*swV5));
 677             /*         #define INNERFLOPS INNERFLOPS+9 */
 678
 679             /*         #if 'Force' in KERNEL_VF */
 680             dsw              = d2*(swF2+d*(swF3+d*swF4));
 681             /*             #define INNERFLOPS INNERFLOPS+5 */
 682             /*         #endif */
 683
 684             /* Evaluate switch function */
 685             /*         #if 'Force' in KERNEL_VF */
 686             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 687             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 688             felec            = felec*sw - rinv{I}{J}*velec*dsw;
 689             /*                 #define INNERFLOPS INNERFLOPS+3 */
 690             /*             #endif */
 691             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 692             fvdw             = fvdw*sw - rinv{I}{J}*vvdw*dsw;
 693             /*                 #define INNERFLOPS INNERFLOPS+3 */
 694             /*             #endif */
 695             /*         #endif */
 696             /*         #if 'Potential' in KERNEL_VF */
 697             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 698             velec           *= sw;
 699             /*                 #define INNERFLOPS INNERFLOPS+1 */
 700             /*             #endif */
 701             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 702             vvdw            *= sw;
 703             /*                 #define INNERFLOPS INNERFLOPS+1 */
 704             /*             #endif */
 705             /*         #endif */
 706             /*     #endif */
 707
 708             /*     #if 'Potential' in KERNEL_VF */
 709             /* Update potential sums from outer loop */
 710             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 711             velecsum        += velec;
 712             /*             #define INNERFLOPS INNERFLOPS+1 */
 713             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 714             vgbsum          += vgb;
 715             /*                 #define INNERFLOPS INNERFLOPS+1 */
 716             /*             #endif */
 717             /*         #endif */
 718             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 719             vvdwsum         += vvdw;
 720             /*             #define INNERFLOPS INNERFLOPS+1 */
 721             /*         #endif */
 722             /*     #endif */
 723
 724             /*     #if 'Force' in KERNEL_VF */
 725
 726             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 727             fscal            = felec+fvdw;
 728             /*             #define INNERFLOPS INNERFLOPS+1 */
 729             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 730             fscal            = felec;
 731             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 732             fscal            = fvdw;
 733             /*        #endif */
 734
 735             /* Calculate temporary vectorial force */
 736             tx               = fscal*dx{I}{J};
 737             ty               = fscal*dy{I}{J};
 738             tz               = fscal*dz{I}{J};
 739
 740             /* Update vectorial force */
 741             fix{I}            += tx;
 742             fiy{I}            += ty;
 743             fiz{I}            += tz;
 744             f[j_coord_offset+DIM*{J}+XX] -= tx;
 745             f[j_coord_offset+DIM*{J}+YY] -= ty;
 746             f[j_coord_offset+DIM*{J}+ZZ] -= tz;
 747
 748             /*         #define INNERFLOPS INNERFLOPS+9 */
 749             /*     #endif */
 750
 751             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 752             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
 753             {
 754                 /*     #endif */
 755             }
 756             /*     #endif */
 757             /*    ## End of check for the interaction being outside the cutoff */
 758
 759             /* #endfor */
 760             /* ## End of loop over i-j interaction pairs */
 761
 762             /* Inner loop uses {INNERFLOPS} flops */
 763         }
 764         /* End of innermost loop */
 765
 766         /* #if 'Force' in KERNEL_VF */
 767         tx = ty = tz = 0;
 768         /*     #for I in PARTICLES_I */
 769         f[i_coord_offset+DIM*{I}+XX] += fix{I};
 770         f[i_coord_offset+DIM*{I}+YY] += fiy{I};
 771         f[i_coord_offset+DIM*{I}+ZZ] += fiz{I};
 772         tx                         += fix{I};
 773         ty                         += fiy{I};
 774         tz                         += fiz{I};
 775         /*         #define OUTERFLOPS OUTERFLOPS+6 */
 776         /*     #endfor */
 777         fshift[i_shift_offset+XX]  += tx;
 778         fshift[i_shift_offset+YY]  += ty;
 779         fshift[i_shift_offset+ZZ]  += tz;
 780         /*     #define OUTERFLOPS OUTERFLOPS+3 */
 781         /* #endif */
 782
 783         /* #if 'Potential' in KERNEL_VF */
 784         ggid                        = gid[iidx];
 785         /* Update potential energies */
 786         /*     #if KERNEL_ELEC != 'None' */
 787         kernel_data->energygrp_elec[ggid] += velecsum;
 788         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 789         /*     #endif */
 790         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 791         kernel_data->energygrp_polarization[ggid] += vgbsum;
 792         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 793         /*     #endif */
 794         /*     #if KERNEL_VDW != 'None' */
 795         kernel_data->energygrp_vdw[ggid] += vvdwsum;
 796         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 797         /*     #endif */
 798         /* #endif */
 799         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 800         dvda[nri]                   = dvda[nri] + dvdasum*isai{I}*isai{I};
 801         /*     #endif */
 802
 803         /* Increment number of inner iterations */
 804         inneriter                  += j_index_end - j_index_start;
 805
 806         /* Outer loop uses {OUTERFLOPS} flops */
 807     }
 808
 809     /* Increment number of outer iterations */
 810     outeriter        += nri;
 811
 812     /* Update outer/inner flops */
 813     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
 814     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
 815     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
 816     /* #if GEOMETRY_I == 'Water3'            */
 817     /*     #define ISUFFIX '_W3'             */
 818     /* #elif GEOMETRY_I == 'Water4'          */
 819     /*     #define ISUFFIX '_W4'             */
 820     /* #else                                 */
 821     /*     #define ISUFFIX ''                */
 822     /* #endif                                */
 823     /* #if GEOMETRY_J == 'Water3'            */
 824     /*     #define JSUFFIX 'W3'              */
 825     /* #elif GEOMETRY_J == 'Water4'          */
 826     /*     #define JSUFFIX 'W4'              */
 827     /* #else                                 */
 828     /*     #define JSUFFIX ''                */
 829     /* #endif                                */
 830     /* #if 'PotentialAndForce' in KERNEL_VF  */
 831     /*     #define VFSUFFIX  '_VF'           */
 832     /* #elif 'Potential' in KERNEL_VF        */
 833     /*     #define VFSUFFIX '_V'             */
 834     /* #else                                 */
 835     /*     #define VFSUFFIX '_F'             */
 836     /* #endif                                */
 837
 838     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
 839     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
 840     /* #elif KERNEL_ELEC != 'None' */
 841     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
 842     /* #else */
 843     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
 844     /* #endif  */
 845 }