src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_template_avx_256_double.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
  16 #include "gmx_math_x86_avx_256_double.h"
  17 #include "kernelutil_x86_avx_256_double.h"
  18 /* #endif */
  19
  20 /* ## List of variables set by the generating script:                                    */
  21 /* ##                                                                                    */
  22 /* ## Setttings that apply to the entire kernel:                                         */
  23 /* ## KERNEL_ELEC:           String, choice for electrostatic interactions               */
  24 /* ## KERNEL_VDW:            String, choice for van der Waals interactions               */
  25 /* ## KERNEL_NAME:           String, name of this kernel                                 */
  26 /* ## KERNEL_VF:             String telling if we calculate potential, force, or both    */
  27 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
  28 /* ##                                                                                    */
  29 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops:             */
  30 /* ## PARTICLES_I[]/         Arrays with lists of i/j particles to use in kernel. It is  */
  31 /* ## PARTICLES_J[]:         just [0] for particle geometry, but can be longer for water */
  32 /* ## PARTICLES_ELEC_I[]/    Arrays with lists of i/j particle that have electrostatics  */
  33 /* ## PARTICLES_ELEC_J[]:    interactions that should be calculated in this kernel.      */
  34 /* ## PARTICLES_VDW_I[]/     Arrays with the list of i/j particle that have VdW          */
  35 /* ## PARTICLES_VDW_J[]:     interactions that should be calculated in this kernel.      */
  36 /* ##                                                                                    */
  37 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle)   */
  38 /* ## PAIRS_IJ[]:            Array with (i,j) tuples of pairs for which interactions     */
  39 /* ##                        should be calculated in this kernel. Zero-charge particles  */
  40 /* ##                        do not have interactions with particles without vdw, and    */
  41 /* ##                        Vdw-only interactions are not evaluated in a no-vdw-kernel. */
  42 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
  43 /* ##                        For each i-j pair, the element [I][J] is a list of strings  */
  44 /* ##                        defining properties/flags of this interaction. Examples     */
  45 /* ##                        include 'electrostatics'/'vdw' if that type of interaction  */
  46 /* ##                        should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values  */
  47 /* ##                        are needed, and 'exactcutoff' or 'shift','switch' to        */
  48 /* ##                        decide if the force/potential should be modified. This way  */
  49 /* ##                        we only calculate values absolutely needed for each case.   */
  50
  51 /* ## Calculate the size and offset for (merged/interleaved) table data */
  52
  53 /*
  54  * Gromacs nonbonded kernel:   {KERNEL_NAME}
  55  * Electrostatics interaction: {KERNEL_ELEC}
  56  * VdW interaction:            {KERNEL_VDW}
  57  * Geometry:                   {GEOMETRY_I}-{GEOMETRY_J}
  58  * Calculate force/pot:        {KERNEL_VF}
  59  */
  60 void
  61 {KERNEL_NAME}
  62                     (t_nblist * gmx_restrict                nlist,
  63                      rvec * gmx_restrict                    xx,
  64                      rvec * gmx_restrict                    ff,
  65                      t_forcerec * gmx_restrict              fr,
  66                      t_mdatoms * gmx_restrict               mdatoms,
  67                      nb_kernel_data_t * gmx_restrict        kernel_data,
  68                      t_nrnb * gmx_restrict                  nrnb)
  69 {
  70     /* ## Not all variables are used for all kernels, but any optimizing compiler fixes that, */
  71     /* ## so there is no point in going to extremes to exclude variables that are not needed. */
  72     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  73      * just 0 for non-waters.
  74      * Suffixes A,B,C,D refer to j loop unrolling done with AVX, e.g. for the four different
  75      * jnr indices corresponding to data put in the four positions in the SIMD register.
  76      */
  77     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  78     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  79     int              jnrA,jnrB,jnrC,jnrD;
  80     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  81     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  82     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  83     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  84     real             rcutoff_scalar;
  85     real             *shiftvec,*fshift,*x,*f;
  86     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  87     real             scratch[4*DIM];
  88     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  89     /* #for I in PARTICLES_I */
  90     real *           vdwioffsetptr{I};
  91     __m256d          ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
  92     /* #endfor */
  93     /* #for J in PARTICLES_J */
  94     int              vdwjidx{J}A,vdwjidx{J}B,vdwjidx{J}C,vdwjidx{J}D;
  95     __m256d          jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
  96     /* #endfor */
  97     /* #for I,J in PAIRS_IJ */
  98     __m256d          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};
  99     /* #endfor */
 100     /* #if KERNEL_ELEC != 'None' */
 101     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 102     real             *charge;
 103     /* #endif */
 104     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 105     __m128i          gbitab;
 106     __m256d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 107     __m256d          minushalf = _mm256_set1_pd(-0.5);
 108     real             *invsqrta,*dvda,*gbtab;
 109     /* #endif */
 110     /* #if KERNEL_VDW != 'None' */
 111     int              nvdwtype;
 112     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 113     int              *vdwtype;
 114     real             *vdwparam;
 115     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 116     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 117     /* #endif */
 118     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 119     __m128i          vfitab;
 120     __m128i          ifour       = _mm_set1_epi32(4);
 121     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 122     real             *vftab;
 123     /* #endif */
 124     /* #if 'Ewald' in KERNEL_ELEC */
 125     __m128i          ewitab;
 126     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 127     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 128     real             *ewtab;
 129     /* #endif */
 130     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 131     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 132     real             rswitch_scalar,d_scalar;
 133     /* #endif */
 134     __m256d          dummy_mask,cutoff_mask;
 135     __m128           tmpmask0,tmpmask1;
 136     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 137     __m256d          one     = _mm256_set1_pd(1.0);
 138     __m256d          two     = _mm256_set1_pd(2.0);
 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            = _mm256_set1_pd(fr->epsfac);
 152     charge           = mdatoms->chargeA;
 153     /*     #if 'ReactionField' in KERNEL_ELEC */
 154     krf              = _mm256_set1_pd(fr->ic->k_rf);
 155     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 156     crf              = _mm256_set1_pd(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       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 168     /* #elif 'Table' in KERNEL_ELEC */
 169     vftab            = kernel_data->table_elec->data;
 170     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 171     /* #elif 'Table' in KERNEL_VDW */
 172     vftab            = kernel_data->table_vdw->data;
 173     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 174     /* #endif */
 175
 176     /* #if 'Ewald' in KERNEL_ELEC */
 177     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 178     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 179     beta2            = _mm256_mul_pd(beta,beta);
 180     beta3            = _mm256_mul_pd(beta,beta2);
 181
 182     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 183     ewtab            = fr->ic->tabq_coul_F;
 184     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 185     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 186     /*     #else */
 187     ewtab            = fr->ic->tabq_coul_FDV0;
 188     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 189     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 190      /*     #endif */
 191     /* #endif */
 192
 193     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 194     invsqrta         = fr->invsqrta;
 195     dvda             = fr->dvda;
 196     gbtabscale       = _mm256_set1_pd(fr->gbtab.scale);
 197     gbtab            = fr->gbtab.data;
 198     gbinvepsdiff     = _mm256_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 199     /* #endif */
 200
 201     /* #if 'Water' in GEOMETRY_I */
 202     /* Setup water-specific parameters */
 203     inr              = nlist->iinr[0];
 204     /*     #for I in PARTICLES_ELEC_I */
 205     iq{I}              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+{I}]));
 206     /*     #endfor */
 207     /*     #for I in PARTICLES_VDW_I */
 208     vdwioffsetptr{I}   = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
 209     /*     #endfor */
 210     /* #endif */
 211
 212     /* #if 'Water' in GEOMETRY_J */
 213     /*     #for J in PARTICLES_ELEC_J */
 214     jq{J}              = _mm256_set1_pd(charge[inr+{J}]);
 215     /*     #endfor */
 216     /*     #for J in PARTICLES_VDW_J */
 217     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 218     /*     #endfor */
 219     /*     #for I,J in PAIRS_IJ */
 220     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 221     qq{I}{J}             = _mm256_mul_pd(iq{I},jq{J});
 222     /*         #endif */
 223     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 224     c6_{I}{J}            = _mm256_set1_pd(vdwioffsetptr{I}[vdwjidx{J}A]);
 225     c12_{I}{J}           = _mm256_set1_pd(vdwioffsetptr{I}[vdwjidx{J}A+1]);
 226     /*         #endif */
 227     /*     #endfor */
 228     /* #endif */
 229
 230     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 231     /*     #if KERNEL_ELEC!='None' */
 232     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 233     rcutoff_scalar   = fr->rcoulomb;
 234     /*     #else */
 235     rcutoff_scalar   = fr->rvdw;
 236     /*     #endif */
 237     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 238     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 239     /* #endif */
 240
 241     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 242     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 243     rvdw             = _mm256_set1_pd(fr->rvdw);
 244     /* #endif */
 245
 246     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 247     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 248     rswitch_scalar   = fr->rcoulomb_switch;
 249     rswitch          = _mm256_set1_pd(rswitch_scalar);
 250     /*     #else */
 251     rswitch_scalar   = fr->rvdw_switch;
 252     rswitch          = _mm256_set1_pd(rswitch_scalar);
 253     /*     #endif */
 254     /* Setup switch parameters */
 255     d_scalar         = rcutoff_scalar-rswitch_scalar;
 256     d                = _mm256_set1_pd(d_scalar);
 257     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 258     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 259     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 260     /*     #if 'Force' in KERNEL_VF */
 261     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 262     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 263     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 264     /*     #endif */
 265     /* #endif */
 266
 267     /* Avoid stupid compiler warnings */
 268     jnrA = jnrB = jnrC = jnrD = 0;
 269     j_coord_offsetA = 0;
 270     j_coord_offsetB = 0;
 271     j_coord_offsetC = 0;
 272     j_coord_offsetD = 0;
 273
 274     /* ## Keep track of the floating point operations we issue for reporting! */
 275     /* #define OUTERFLOPS 0 */
 276     outeriter        = 0;
 277     inneriter        = 0;
 278
 279     for(iidx=0;iidx<4*DIM;iidx++)
 280     {
 281         scratch[iidx] = 0.0;
 282     }
 283
 284     /* Start outer loop over neighborlists */
 285     for(iidx=0; iidx<nri; iidx++)
 286     {
 287         /* Load shift vector for this list */
 288         i_shift_offset   = DIM*shiftidx[iidx];
 289
 290         /* Load limits for loop over neighbors */
 291         j_index_start    = jindex[iidx];
 292         j_index_end      = jindex[iidx+1];
 293
 294         /* Get outer coordinate index */
 295         inr              = iinr[iidx];
 296         i_coord_offset   = DIM*inr;
 297
 298         /* Load i particle coords and add shift vector */
 299         /* #if GEOMETRY_I == 'Particle' */
 300         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 301         /* #elif GEOMETRY_I == 'Water3' */
 302         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 303                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 304         /* #elif GEOMETRY_I == 'Water4' */
 305         /*     #if 0 in PARTICLES_I                 */
 306         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 307                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 308         /*     #else                                */
 309         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 310                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 311         /*     #endif                               */
 312         /* #endif                                   */
 313
 314         /* #if 'Force' in KERNEL_VF */
 315         /*     #for I in PARTICLES_I */
 316         fix{I}             = _mm256_setzero_pd();
 317         fiy{I}             = _mm256_setzero_pd();
 318         fiz{I}             = _mm256_setzero_pd();
 319         /*     #endfor */
 320         /* #endif */
 321
 322         /* ## For water we already preloaded parameters at the start of the kernel */
 323         /* #if not 'Water' in GEOMETRY_I */
 324         /* Load parameters for i particles */
 325         /*     #for I in PARTICLES_ELEC_I */
 326         iq{I}              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+{I}]));
 327         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 328         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 329         isai{I}            = _mm256_set1_pd(invsqrta[inr+{I}]);
 330         /*         #endif */
 331         /*     #endfor */
 332         /*     #for I in PARTICLES_VDW_I */
 333         vdwioffsetptr{I}   = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
 334         /*     #endfor */
 335         /* #endif */
 336
 337         /* #if 'Potential' in KERNEL_VF */
 338         /* Reset potential sums */
 339         /*     #if KERNEL_ELEC != 'None' */
 340         velecsum         = _mm256_setzero_pd();
 341         /*     #endif */
 342         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 343         vgbsum           = _mm256_setzero_pd();
 344         /*     #endif */
 345         /*     #if KERNEL_VDW != 'None' */
 346         vvdwsum          = _mm256_setzero_pd();
 347         /*     #endif */
 348         /* #endif */
 349         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 350         dvdasum          = _mm256_setzero_pd();
 351         /*     #endif */
 352
 353         /* #for ROUND in ['Loop','Epilogue'] */
 354
 355         /* #if ROUND =='Loop' */
 356         /* Start inner kernel loop */
 357         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 358         {
 359         /* ## First round is normal loop (next statement resets indentation) */
 360         /*     #if 0 */
 361         }
 362         /*     #endif */
 363         /* #else */
 364         if(jidx<j_index_end)
 365         {
 366         /* ## Second round is epilogue */
 367         /* #endif */
 368         /* #define INNERFLOPS 0 */
 369
 370             /* Get j neighbor index, and coordinate index */
 371             /* #if ROUND =='Loop' */
 372             jnrA             = jjnr[jidx];
 373             jnrB             = jjnr[jidx+1];
 374             jnrC             = jjnr[jidx+2];
 375             jnrD             = jjnr[jidx+3];
 376             /* #else */
 377             jnrlistA         = jjnr[jidx];
 378             jnrlistB         = jjnr[jidx+1];
 379             jnrlistC         = jjnr[jidx+2];
 380             jnrlistD         = jjnr[jidx+3];
 381             /* Sign of each element will be negative for non-real atoms.
 382              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 383              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 384              */
 385             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 386
 387             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 388             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 389             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 390
 391             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 392             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 393             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 394             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 395             /* #endif */
 396             j_coord_offsetA  = DIM*jnrA;
 397             j_coord_offsetB  = DIM*jnrB;
 398             j_coord_offsetC  = DIM*jnrC;
 399             j_coord_offsetD  = DIM*jnrD;
 400
 401             /* load j atom coordinates */
 402             /* #if GEOMETRY_J == 'Particle'             */
 403             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 404                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 405                                                  &jx0,&jy0,&jz0);
 406             /* #elif GEOMETRY_J == 'Water3'             */
 407             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 408                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 409                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 410             /* #elif GEOMETRY_J == 'Water4'             */
 411             /*     #if 0 in PARTICLES_J                 */
 412             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 413                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 414                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 415                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 416             /*     #else                                */
 417             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 418                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 419                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 420             /*     #endif                               */
 421             /* #endif                                   */
 422
 423             /* Calculate displacement vector */
 424             /* #for I,J in PAIRS_IJ */
 425             dx{I}{J}             = _mm256_sub_pd(ix{I},jx{J});
 426             dy{I}{J}             = _mm256_sub_pd(iy{I},jy{J});
 427             dz{I}{J}             = _mm256_sub_pd(iz{I},jz{J});
 428             /*     #define INNERFLOPS INNERFLOPS+3 */
 429             /* #endfor */
 430
 431             /* Calculate squared distance and things based on it */
 432             /* #for I,J in PAIRS_IJ */
 433             rsq{I}{J}            = gmx_mm256_calc_rsq_pd(dx{I}{J},dy{I}{J},dz{I}{J});
 434             /*     #define INNERFLOPS INNERFLOPS+5 */
 435             /* #endfor */
 436
 437             /* #for I,J in PAIRS_IJ */
 438             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 439             rinv{I}{J}           = gmx_mm256_invsqrt_pd(rsq{I}{J});
 440             /*         #define INNERFLOPS INNERFLOPS+5 */
 441             /*     #endif */
 442             /* #endfor */
 443
 444             /* #for I,J in PAIRS_IJ */
 445             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 446             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 447             rinvsq{I}{J}         = gmx_mm256_inv_pd(rsq{I}{J});
 448             /*             #define INNERFLOPS INNERFLOPS+4 */
 449             /*         #else */
 450             rinvsq{I}{J}         = _mm256_mul_pd(rinv{I}{J},rinv{I}{J});
 451             /*             #define INNERFLOPS INNERFLOPS+1 */
 452             /*         #endif */
 453             /*     #endif */
 454             /* #endfor */
 455
 456             /* #if not 'Water' in GEOMETRY_J */
 457             /* Load parameters for j particles */
 458             /*     #for J in PARTICLES_ELEC_J */
 459             jq{J}              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+{J},charge+jnrB+{J},
 460                                                                  charge+jnrC+{J},charge+jnrD+{J});
 461             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 462             isaj{J}            = gmx_mm256_load_4real_swizzle_pd(invsqrta+jnrA+{J},invsqrta+jnrB+{J},
 463                                                                  invsqrta+jnrC+{J},invsqrta+jnrD+{J});
 464             /*         #endif */
 465             /*     #endfor */
 466             /*     #for J in PARTICLES_VDW_J */
 467             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 468             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 469             vdwjidx{J}C        = 2*vdwtype[jnrC+{J}];
 470             vdwjidx{J}D        = 2*vdwtype[jnrD+{J}];
 471             /*     #endfor */
 472             /* #endif */
 473
 474             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 475             /*     #for J in PARTICLES_J */
 476             fjx{J}             = _mm256_setzero_pd();
 477             fjy{J}             = _mm256_setzero_pd();
 478             fjz{J}             = _mm256_setzero_pd();
 479             /*     #endfor */
 480             /* #endif */
 481
 482             /* #for I,J in PAIRS_IJ */
 483
 484             /**************************
 485              * CALCULATE INTERACTIONS *
 486              **************************/
 487
 488             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 489             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 490             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 491             if (gmx_mm256_any_lt(rsq{I}{J},rcutoff2))
 492             {
 493                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 494             }
 495             /*         #endif */
 496             /*         #define INNERFLOPS INNERFLOPS+1 */
 497             /*     #endif */
 498
 499             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 500             r{I}{J}              = _mm256_mul_pd(rsq{I}{J},rinv{I}{J});
 501             /*         #if ROUND == 'Epilogue' */
 502             r{I}{J}              = _mm256_andnot_pd(dummy_mask,r{I}{J});
 503             /*             #define INNERFLOPS INNERFLOPS+1 */
 504             /*         #endif */
 505             /*         #define INNERFLOPS INNERFLOPS+1 */
 506             /*     #endif */
 507
 508             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 509             /*     #if not 'Water' in GEOMETRY_J */
 510             /* Compute parameters for interactions between i and j atoms */
 511             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 512             qq{I}{J}             = _mm256_mul_pd(iq{I},jq{J});
 513             /*             #define INNERFLOPS INNERFLOPS+1 */
 514             /*         #endif */
 515             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 516             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr{I}+vdwjidx{J}A,
 517                                             vdwioffsetptr{I}+vdwjidx{J}B,
 518                                             vdwioffsetptr{I}+vdwjidx{J}C,
 519                                             vdwioffsetptr{I}+vdwjidx{J}D,
 520                                             &c6_{I}{J},&c12_{I}{J});
 521             /*         #endif */
 522             /*     #endif */
 523
 524             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 525             /* Calculate table index by multiplying r with table scale and truncate to integer */
 526             rt               = _mm256_mul_pd(r{I}{J},vftabscale);
 527             vfitab           = _mm256_cvttpd_epi32(rt);
 528             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 529             /*         #define INNERFLOPS INNERFLOPS+4                          */
 530             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW     */
 531             /*             ## 3 tables, 4 bytes per point: multiply index by 12 */
 532             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 533             /*         #elif 'Table' in KERNEL_ELEC                             */
 534             /*             ## 1 table, 4 bytes per point: multiply index by 4   */
 535             vfitab           = _mm_slli_epi32(vfitab,2);
 536             /*         #elif 'Table' in KERNEL_VDW                              */
 537             /*             ## 2 tables, 4 bytes per point: multiply index by 8  */
 538             vfitab           = _mm_slli_epi32(vfitab,3);
 539             /*         #endif                                                   */
 540             /*     #endif */
 541
 542             /*     ## ELECTROSTATIC INTERACTIONS */
 543             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 544
 545             /*         #if KERNEL_ELEC=='Coulomb' */
 546
 547             /* COULOMB ELECTROSTATICS */
 548             velec            = _mm256_mul_pd(qq{I}{J},rinv{I}{J});
 549             /*             #define INNERFLOPS INNERFLOPS+1 */
 550             /*             #if 'Force' in KERNEL_VF */
 551             felec            = _mm256_mul_pd(velec,rinvsq{I}{J});
 552             /*                 #define INNERFLOPS INNERFLOPS+1 */
 553             /*             #endif */
 554
 555             /*         #elif KERNEL_ELEC=='ReactionField' */
 556
 557             /* REACTION-FIELD ELECTROSTATICS */
 558             /*             #if 'Potential' in KERNEL_VF */
 559             velec            = _mm256_mul_pd(qq{I}{J},_mm256_sub_pd(_mm256_add_pd(rinv{I}{J},_mm256_mul_pd(krf,rsq{I}{J})),crf));
 560             /*                 #define INNERFLOPS INNERFLOPS+4 */
 561             /*             #endif */
 562             /*             #if 'Force' in KERNEL_VF */
 563             felec            = _mm256_mul_pd(qq{I}{J},_mm256_sub_pd(_mm256_mul_pd(rinv{I}{J},rinvsq{I}{J}),krf2));
 564             /*                 #define INNERFLOPS INNERFLOPS+3 */
 565             /*             #endif */
 566
 567             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 568
 569             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 570             isaprod          = _mm256_mul_pd(isai{I},isaj{J});
 571             gbqqfactor       = _mm256_xor_pd(signbit,_mm256_mul_pd(qq{I}{J},_mm256_mul_pd(isaprod,gbinvepsdiff)));
 572             gbscale          = _mm256_mul_pd(isaprod,gbtabscale);
 573             /*             #define INNERFLOPS INNERFLOPS+5 */
 574
 575             /* Calculate generalized born table index - this is a separate table from the normal one,
 576              * but we use the same procedure by multiplying r with scale and truncating to integer.
 577              */
 578             rt               = _mm256_mul_pd(r{I}{J},gbscale);
 579             gbitab           = _mm256_cvttpd_epi32(rt);
 580             gbeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 581             gbitab           = _mm_slli_epi32(gbitab,2);
 582             Y                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
 583             F                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,1) );
 584             G                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,2) );
 585             H                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,3) );
 586             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 587             Heps             = _mm256_mul_pd(gbeps,H);
 588             Fp               = _mm256_add_pd(F,_mm256_mul_pd(gbeps,_mm256_add_pd(G,Heps)));
 589             VV               = _mm256_add_pd(Y,_mm256_mul_pd(gbeps,Fp));
 590             vgb              = _mm256_mul_pd(gbqqfactor,VV);
 591             /*             #define INNERFLOPS INNERFLOPS+10 */
 592
 593             /*             #if 'Force' in KERNEL_VF */
 594             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(gbeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 595             fgb              = _mm256_mul_pd(gbqqfactor,_mm256_mul_pd(FF,gbscale));
 596             dvdatmp          = _mm256_mul_pd(minushalf,_mm256_add_pd(vgb,_mm256_mul_pd(fgb,r{I}{J})));
 597             /*                 #if ROUND == 'Epilogue' */
 598             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 599             /*                 #endif */
 600             dvdasum          = _mm256_add_pd(dvdasum,dvdatmp);
 601             /*                 #if ROUND == 'Loop' */
 602             fjptrA           = dvda+jnrA;
 603             fjptrB           = dvda+jnrB;
 604             fjptrC           = dvda+jnrC;
 605             fjptrD           = dvda+jnrD;
 606             /*                 #else */
 607             /* The pointers to scratch make sure that this code with compilers that take gmx_restrict seriously (e.g. icc 13) really can't screw things up. */
 608             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 609             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 610             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 611             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 612             /*                 #endif */
 613             gmx_mm256_increment_4real_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 614                                                  _mm256_mul_pd(dvdatmp,_mm256_mul_pd(isaj{J},isaj{J})));
 615             /*                 #define INNERFLOPS INNERFLOPS+12 */
 616             /*             #endif */
 617             velec            = _mm256_mul_pd(qq{I}{J},rinv{I}{J});
 618             /*                 #define INNERFLOPS INNERFLOPS+1 */
 619             /*             #if 'Force' in KERNEL_VF */
 620             felec            = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(velec,rinv{I}{J}),fgb),rinv{I}{J});
 621             /*                 #define INNERFLOPS INNERFLOPS+3 */
 622             /*             #endif */
 623
 624             /*         #elif KERNEL_ELEC=='Ewald' */
 625             /* EWALD ELECTROSTATICS */
 626
 627             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 628             ewrt             = _mm256_mul_pd(r{I}{J},ewtabscale);
 629             ewitab           = _mm256_cvttpd_epi32(ewrt);
 630             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 631             /*             #define INNERFLOPS INNERFLOPS+4 */
 632             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 633             ewitab           = _mm_slli_epi32(ewitab,2);
 634             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 635             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 636             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 637             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 638             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 639             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 640             /*                 #define INNERFLOPS INNERFLOPS+2 */
 641             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 642             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 643             velec            = _mm256_mul_pd(qq{I}{J},_mm256_sub_pd(_mm256_sub_pd(rinv{I}{J},sh_ewald),velec));
 644             /*                     #define INNERFLOPS INNERFLOPS+7 */
 645             /*                 #else */
 646             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 647             velec            = _mm256_mul_pd(qq{I}{J},_mm256_sub_pd(rinv{I}{J},velec));
 648             /*                     #define INNERFLOPS INNERFLOPS+6 */
 649             /*                 #endif */
 650             /*                 #if 'Force' in KERNEL_VF */
 651             felec            = _mm256_mul_pd(_mm256_mul_pd(qq{I}{J},rinv{I}{J}),_mm256_sub_pd(rinvsq{I}{J},felec));
 652             /*                      #define INNERFLOPS INNERFLOPS+3 */
 653             /*                 #endif */
 654             /*             #elif KERNEL_VF=='Force' */
 655             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 656                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 657                                             &ewtabF,&ewtabFn);
 658             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 659             felec            = _mm256_mul_pd(_mm256_mul_pd(qq{I}{J},rinv{I}{J}),_mm256_sub_pd(rinvsq{I}{J},felec));
 660             /*                 #define INNERFLOPS INNERFLOPS+7 */
 661             /*             #endif */
 662
 663             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 664
 665             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 666             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 667             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 668             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 669             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 670             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 671             Heps             = _mm256_mul_pd(vfeps,H);
 672             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 673             /*             #define INNERFLOPS INNERFLOPS+4 */
 674             /*             #if 'Potential' in KERNEL_VF */
 675             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 676             velec            = _mm256_mul_pd(qq{I}{J},VV);
 677             /*                 #define INNERFLOPS INNERFLOPS+3 */
 678             /*             #endif */
 679             /*             #if 'Force' in KERNEL_VF */
 680             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 681             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq{I}{J},FF),_mm256_mul_pd(vftabscale,rinv{I}{J})));
 682             /*                 #define INNERFLOPS INNERFLOPS+7 */
 683             /*             #endif */
 684             /*         #endif */
 685             /*         ## End of check for electrostatics interaction forms */
 686             /*     #endif */
 687             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 688
 689             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 690
 691             /*         #if KERNEL_VDW=='LennardJones' */
 692
 693             /* LENNARD-JONES DISPERSION/REPULSION */
 694
 695             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 696             /*             #define INNERFLOPS INNERFLOPS+2 */
 697             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 698             vvdw6            = _mm256_mul_pd(c6_{I}{J},rinvsix);
 699             vvdw12           = _mm256_mul_pd(c12_{I}{J},_mm256_mul_pd(rinvsix,rinvsix));
 700             /*                 #define INNERFLOPS INNERFLOPS+3 */
 701             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 702             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_{I}{J},_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 703                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
 704             /*                     #define INNERFLOPS INNERFLOPS+8 */
 705             /*                 #else */
 706             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 707             /*                     #define INNERFLOPS INNERFLOPS+3 */
 708             /*                 #endif */
 709             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 710             /*                 #if 'Force' in KERNEL_VF */
 711             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq{I}{J});
 712             /*                     #define INNERFLOPS INNERFLOPS+2 */
 713             /*                 #endif */
 714             /*             #elif KERNEL_VF=='Force' */
 715             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 716             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_{I}{J},rinvsix),c6_{I}{J}),_mm256_mul_pd(rinvsix,rinvsq{I}{J}));
 717             /*                 #define INNERFLOPS INNERFLOPS+4 */
 718             /*             #endif */
 719
 720             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 721
 722             /* CUBIC SPLINE TABLE DISPERSION */
 723             /*             #if 'Table' in KERNEL_ELEC */
 724             vfitab           = _mm_add_epi32(vfitab,ifour);
 725             /*             #endif                     */
 726             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 727             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 728             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 729             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 730             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 731             Heps             = _mm256_mul_pd(vfeps,H);
 732             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 733             /*             #define INNERFLOPS INNERFLOPS+4 */
 734             /*             #if 'Potential' in KERNEL_VF */
 735             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 736             vvdw6            = _mm256_mul_pd(c6_{I}{J},VV);
 737             /*                 #define INNERFLOPS INNERFLOPS+3 */
 738             /*             #endif */
 739             /*             #if 'Force' in KERNEL_VF */
 740             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 741             fvdw6            = _mm256_mul_pd(c6_{I}{J},FF);
 742             /*                 #define INNERFLOPS INNERFLOPS+4 */
 743             /*             #endif */
 744
 745             /* CUBIC SPLINE TABLE REPULSION */
 746             vfitab           = _mm_add_epi32(vfitab,ifour);
 747             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 748             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 749             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 750             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 751             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 752             Heps             = _mm256_mul_pd(vfeps,H);
 753             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 754             /*             #define INNERFLOPS INNERFLOPS+4 */
 755             /*             #if 'Potential' in KERNEL_VF */
 756             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 757             vvdw12           = _mm256_mul_pd(c12_{I}{J},VV);
 758             /*                 #define INNERFLOPS INNERFLOPS+3 */
 759             /*             #endif */
 760             /*             #if 'Force' in KERNEL_VF */
 761             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 762             fvdw12           = _mm256_mul_pd(c12_{I}{J},FF);
 763             /*                 #define INNERFLOPS INNERFLOPS+5 */
 764             /*             #endif */
 765             /*             #if 'Potential' in KERNEL_VF */
 766             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 767             /*                 #define INNERFLOPS INNERFLOPS+1 */
 768             /*             #endif */
 769             /*             #if 'Force' in KERNEL_VF */
 770             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv{I}{J})));
 771             /*                 #define INNERFLOPS INNERFLOPS+4 */
 772             /*             #endif */
 773             /*         #endif */
 774             /*         ## End of check for vdw interaction forms */
 775             /*     #endif */
 776             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 777
 778             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 779             d                = _mm256_sub_pd(r{I}{J},rswitch);
 780             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 781             d2               = _mm256_mul_pd(d,d);
 782             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 783             /*         #define INNERFLOPS INNERFLOPS+10 */
 784
 785             /*         #if 'Force' in KERNEL_VF */
 786             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 787             /*             #define INNERFLOPS INNERFLOPS+5 */
 788             /*         #endif */
 789
 790             /* Evaluate switch function */
 791             /*         #if 'Force' in KERNEL_VF */
 792             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 793             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 794             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv{I}{J},_mm256_mul_pd(velec,dsw)) );
 795             /*                 #define INNERFLOPS INNERFLOPS+4 */
 796             /*             #endif */
 797             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 798             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv{I}{J},_mm256_mul_pd(vvdw,dsw)) );
 799             /*                 #define INNERFLOPS INNERFLOPS+4 */
 800             /*             #endif */
 801             /*         #endif */
 802             /*         #if 'Potential' in KERNEL_VF */
 803             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 804             velec            = _mm256_mul_pd(velec,sw);
 805             /*                 #define INNERFLOPS INNERFLOPS+1 */
 806             /*             #endif */
 807             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 808             vvdw             = _mm256_mul_pd(vvdw,sw);
 809             /*                 #define INNERFLOPS INNERFLOPS+1 */
 810             /*             #endif */
 811             /*         #endif */
 812             /*     #endif */
 813             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 814             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 815             cutoff_mask      = _mm256_cmp_pd(rsq{I}{J},rcutoff2,_CMP_LT_OQ);
 816             /*         #define INNERFLOPS INNERFLOPS+1 */
 817             /*     #endif */
 818
 819             /*     #if 'Potential' in KERNEL_VF */
 820             /* Update potential sum for this i atom from the interaction with this j atom. */
 821             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 822             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 823             velec            = _mm256_and_pd(velec,cutoff_mask);
 824             /*                 #define INNERFLOPS INNERFLOPS+1 */
 825             /*             #endif                                       */
 826             /*             #if ROUND == 'Epilogue' */
 827             velec            = _mm256_andnot_pd(dummy_mask,velec);
 828             /*             #endif */
 829             velecsum         = _mm256_add_pd(velecsum,velec);
 830             /*             #define INNERFLOPS INNERFLOPS+1 */
 831             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 832             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 833             vgb              = _mm256_and_pd(vgb,cutoff_mask);
 834             /*                 #define INNERFLOPS INNERFLOPS+1 */
 835             /*             #endif                                       */
 836             /*             #if ROUND == 'Epilogue' */
 837             vgb              = _mm256_andnot_pd(dummy_mask,vgb);
 838             /*             #endif */
 839             vgbsum           = _mm256_add_pd(vgbsum,vgb);
 840             /*                 #define INNERFLOPS INNERFLOPS+1 */
 841             /*             #endif */
 842             /*         #endif */
 843             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 844             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 845             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 846             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 847             /*                 #define INNERFLOPS INNERFLOPS+1 */
 848             /*             #endif                                       */
 849             /*             #if ROUND == 'Epilogue' */
 850             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 851             /*             #endif */
 852             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 853             /*             #define INNERFLOPS INNERFLOPS+1 */
 854             /*         #endif */
 855             /*     #endif */
 856
 857             /*     #if 'Force' in KERNEL_VF */
 858
 859             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 860             fscal            = _mm256_add_pd(felec,fvdw);
 861             /*             #define INNERFLOPS INNERFLOPS+1 */
 862             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 863             fscal            = felec;
 864             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 865             fscal            = fvdw;
 866             /*        #endif */
 867
 868             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 869             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 870             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 871             /*                 #define INNERFLOPS INNERFLOPS+1 */
 872             /*             #endif                                       */
 873
 874             /*             #if ROUND == 'Epilogue' */
 875             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 876             /*             #endif */
 877
 878             /* Calculate temporary vectorial force */
 879             tx               = _mm256_mul_pd(fscal,dx{I}{J});
 880             ty               = _mm256_mul_pd(fscal,dy{I}{J});
 881             tz               = _mm256_mul_pd(fscal,dz{I}{J});
 882
 883             /* Update vectorial force */
 884             fix{I}             = _mm256_add_pd(fix{I},tx);
 885             fiy{I}             = _mm256_add_pd(fiy{I},ty);
 886             fiz{I}             = _mm256_add_pd(fiz{I},tz);
 887             /*             #define INNERFLOPS INNERFLOPS+6 */
 888
 889             /* #if GEOMETRY_I == 'Particle'             */
 890             /*     #if ROUND == 'Loop' */
 891             fjptrA             = f+j_coord_offsetA;
 892             fjptrB             = f+j_coord_offsetB;
 893             fjptrC             = f+j_coord_offsetC;
 894             fjptrD             = f+j_coord_offsetD;
 895             /*     #else */
 896             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 897             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 898             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 899             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 900             /*     #endif */
 901             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 902             /*     #define INNERFLOPS INNERFLOPS+3      */
 903             /* #else                                    */
 904             fjx{J}             = _mm256_add_pd(fjx{J},tx);
 905             fjy{J}             = _mm256_add_pd(fjy{J},ty);
 906             fjz{J}             = _mm256_add_pd(fjz{J},tz);
 907             /*     #define INNERFLOPS INNERFLOPS+3      */
 908             /* #endif                                   */
 909
 910             /*     #endif */
 911
 912             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 913             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 914             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
 915             {
 916                 /*     #endif */
 917             }
 918             /*     #endif */
 919             /*    ## End of check for the interaction being outside the cutoff */
 920
 921             /* #endfor */
 922             /* ## End of loop over i-j interaction pairs */
 923
 924             /* #if GEOMETRY_I != 'Particle' */
 925             /*     #if ROUND == 'Loop' */
 926             fjptrA             = f+j_coord_offsetA;
 927             fjptrB             = f+j_coord_offsetB;
 928             fjptrC             = f+j_coord_offsetC;
 929             fjptrD             = f+j_coord_offsetD;
 930             /*     #else */
 931             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 932             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 933             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 934             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 935             /*     #endif */
 936             /* #endif */
 937
 938             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
 939             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 940             /*     #define INNERFLOPS INNERFLOPS+3      */
 941             /* #elif GEOMETRY_J == 'Water3'             */
 942             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 943                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 944             /*     #define INNERFLOPS INNERFLOPS+9      */
 945             /* #elif GEOMETRY_J == 'Water4'             */
 946             /*     #if 0 in PARTICLES_J                 */
 947             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 948                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 949                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 950             /*     #define INNERFLOPS INNERFLOPS+12     */
 951             /*     #else                                */
 952             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 953                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 954             /*     #define INNERFLOPS INNERFLOPS+9      */
 955             /*     #endif                               */
 956             /* #endif                                   */
 957
 958             /* Inner loop uses {INNERFLOPS} flops */
 959         }
 960
 961         /* #endfor */
 962
 963         /* End of innermost loop */
 964
 965         /* #if 'Force' in KERNEL_VF */
 966         /*     #if GEOMETRY_I == 'Particle'            */
 967         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 968                                                  f+i_coord_offset,fshift+i_shift_offset);
 969         /*         #define OUTERFLOPS OUTERFLOPS+6     */
 970         /*     #elif GEOMETRY_I == 'Water3'            */
 971         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 972                                                  f+i_coord_offset,fshift+i_shift_offset);
 973         /*         #define OUTERFLOPS OUTERFLOPS+18    */
 974         /*     #elif GEOMETRY_I == 'Water4'            */
 975         /*         #if 0 in PARTICLES_I                */
 976         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 977                                                  f+i_coord_offset,fshift+i_shift_offset);
 978         /*             #define OUTERFLOPS OUTERFLOPS+24    */
 979         /*         #else                               */
 980         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 981                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
 982         /*             #define OUTERFLOPS OUTERFLOPS+18    */
 983         /*         #endif                              */
 984         /*     #endif                                  */
 985         /* #endif                                      */
 986
 987         /* #if 'Potential' in KERNEL_VF */
 988         ggid                        = gid[iidx];
 989         /* Update potential energies */
 990         /*     #if KERNEL_ELEC != 'None' */
 991         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 992         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 993         /*     #endif */
 994         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 995         gmx_mm256_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 996         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 997         /*     #endif */
 998         /*     #if KERNEL_VDW != 'None' */
 999         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1000         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1001         /*     #endif */
1002         /* #endif */
1003         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1004         dvdasum = _mm256_mul_pd(dvdasum, _mm256_mul_pd(isai{I},isai{I}));
1005         gmx_mm256_update_1pot_pd(dvdasum,dvda+inr);
1006         /*     #endif */
1007
1008         /* Increment number of inner iterations */
1009         inneriter                  += j_index_end - j_index_start;
1010
1011         /* Outer loop uses {OUTERFLOPS} flops */
1012     }
1013
1014     /* Increment number of outer iterations */
1015     outeriter        += nri;
1016
1017     /* Update outer/inner flops */
1018     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1019     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1020     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1021     /* #if GEOMETRY_I == 'Water3'            */
1022     /*     #define ISUFFIX '_W3'             */
1023     /* #elif GEOMETRY_I == 'Water4'          */
1024     /*     #define ISUFFIX '_W4'             */
1025     /* #else                                 */
1026     /*     #define ISUFFIX ''                */
1027     /* #endif                                */
1028     /* #if GEOMETRY_J == 'Water3'            */
1029     /*     #define JSUFFIX 'W3'              */
1030     /* #elif GEOMETRY_J == 'Water4'          */
1031     /*     #define JSUFFIX 'W4'              */
1032     /* #else                                 */
1033     /*     #define JSUFFIX ''                */
1034     /* #endif                                */
1035     /* #if 'PotentialAndForce' in KERNEL_VF  */
1036     /*     #define VFSUFFIX  '_VF'           */
1037     /* #elif 'Potential' in KERNEL_VF        */
1038     /*     #define VFSUFFIX '_V'             */
1039     /* #else                                 */
1040     /*     #define VFSUFFIX '_F'             */
1041     /* #endif                                */
1042
1043     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1044     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1045     /* #elif KERNEL_ELEC != 'None' */
1046     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1047     /* #else */
1048     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1049     /* #endif  */
1050 }