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

   1 /* #if 0 */
   2 /*
   3  * This file is part of the GROMACS molecular simulation package.
   4  *
   5  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   6  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   7  * and including many others, as listed in the AUTHORS file in the
   8  * top-level source directory and at http://www.gromacs.org.
   9  *
  10  * GROMACS is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public License
  12  * as published by the Free Software Foundation; either version 2.1
  13  * of the License, or (at your option) any later version.
  14  *
  15  * GROMACS is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with GROMACS; if not, see
  22  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  23  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  24  *
  25  * If you want to redistribute modifications to GROMACS, please
  26  * consider that scientific software is very special. Version
  27  * control is crucial - bugs must be traceable. We will be happy to
  28  * consider code for inclusion in the official distribution, but
  29  * derived work must not be called official GROMACS. Details are found
  30  * in the README & COPYING files - if they are missing, get the
  31  * official version at http://www.gromacs.org.
  32  *
  33  * To help us fund GROMACS development, we humbly ask that you cite
  34  * the research papers on the package. Check out http://www.gromacs.org.
  35  */
  36 #error This file must be processed with the Gromacs pre-preprocessor
  37 /* #endif */
  38 /* #if INCLUDE_HEADER */
  39 #include "gmxpre.h"
  40
  41 #include "config.h"
  42
  43 #include <math.h>
  44
  45 #include "../nb_kernel.h"
  46 #include "gromacs/gmxlib/nrnb.h"
  47
  48 #include "kernelutil_x86_avx_256_single.h"
  49 /* #endif */
  50
  51 /* ## List of variables set by the generating script:                                    */
  52 /* ##                                                                                    */
  53 /* ## Setttings that apply to the entire kernel:                                         */
  54 /* ## KERNEL_ELEC:           String, choice for electrostatic interactions               */
  55 /* ## KERNEL_VDW:            String, choice for van der Waals interactions               */
  56 /* ## KERNEL_NAME:           String, name of this kernel                                 */
  57 /* ## KERNEL_VF:             String telling if we calculate potential, force, or both    */
  58 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
  59 /* ##                                                                                    */
  60 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops:             */
  61 /* ## PARTICLES_I[]/         Arrays with lists of i/j particles to use in kernel. It is  */
  62 /* ## PARTICLES_J[]:         just [0] for particle geometry, but can be longer for water */
  63 /* ## PARTICLES_ELEC_I[]/    Arrays with lists of i/j particle that have electrostatics  */
  64 /* ## PARTICLES_ELEC_J[]:    interactions that should be calculated in this kernel.      */
  65 /* ## PARTICLES_VDW_I[]/     Arrays with the list of i/j particle that have VdW          */
  66 /* ## PARTICLES_VDW_J[]:     interactions that should be calculated in this kernel.      */
  67 /* ##                                                                                    */
  68 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle)   */
  69 /* ## PAIRS_IJ[]:            Array with (i,j) tuples of pairs for which interactions     */
  70 /* ##                        should be calculated in this kernel. Zero-charge particles  */
  71 /* ##                        do not have interactions with particles without vdw, and    */
  72 /* ##                        Vdw-only interactions are not evaluated in a no-vdw-kernel. */
  73 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
  74 /* ##                        For each i-j pair, the element [I][J] is a list of strings  */
  75 /* ##                        defining properties/flags of this interaction. Examples     */
  76 /* ##                        include 'electrostatics'/'vdw' if that type of interaction  */
  77 /* ##                        should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values  */
  78 /* ##                        are needed, and 'exactcutoff' or 'shift','switch' to        */
  79 /* ##                        decide if the force/potential should be modified. This way  */
  80 /* ##                        we only calculate values absolutely needed for each case.   */
  81
  82 /* ## Calculate the size and offset for (merged/interleaved) table data */
  83
  84
  85
  86
  87 /*
  88  * Gromacs nonbonded kernel:   {KERNEL_NAME}
  89  * Electrostatics interaction: {KERNEL_ELEC}
  90  * VdW interaction:            {KERNEL_VDW}
  91  * Geometry:                   {GEOMETRY_I}-{GEOMETRY_J}
  92  * Calculate force/pot:        {KERNEL_VF}
  93  */
  94 void
  95 {KERNEL_NAME}
  96                     (t_nblist                    * gmx_restrict       nlist,
  97                      rvec                        * gmx_restrict          xx,
  98                      rvec                        * gmx_restrict          ff,
  99                      struct t_forcerec           * gmx_restrict          fr,
 100                      t_mdatoms                   * gmx_restrict     mdatoms,
 101                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 102                      t_nrnb                      * gmx_restrict        nrnb)
 103 {
 104     /* ## Not all variables are used for all kernels, but any optimizing compiler fixes that, */
 105     /* ## so there is no point in going to extremes to exclude variables that are not needed. */
 106     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 107      * just 0 for non-waters.
 108      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 109      * jnr indices corresponding to data put in the four positions in the SIMD register.
 110      */
 111     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 112     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 113     int              jnrA,jnrB,jnrC,jnrD;
 114     int              jnrE,jnrF,jnrG,jnrH;
 115     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 116     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 117     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 118     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 119     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 120     real             rcutoff_scalar;
 121     real             *shiftvec,*fshift,*x,*f;
 122     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 123     real             scratch[4*DIM];
 124     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 125     /* #for I in PARTICLES_I */
 126     real *           vdwioffsetptr{I};
 127     /* #if 'LJEwald' in KERNEL_VDW */
 128     real *           vdwgridioffsetptr{I};
 129     /* #endif                      */
 130     __m256           ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 131     /* #endfor */
 132     /* #for J in PARTICLES_J */
 133     int              vdwjidx{J}A,vdwjidx{J}B,vdwjidx{J}C,vdwjidx{J}D,vdwjidx{J}E,vdwjidx{J}F,vdwjidx{J}G,vdwjidx{J}H;
 134     __m256           jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 135     /* #endfor */
 136     /* #for I,J in PAIRS_IJ */
 137     __m256           dx{I}{J},dy{I}{J},dz{I}{J},rsq{I}{J},rinv{I}{J},rinvsq{I}{J},r{I}{J},qq{I}{J},c6_{I}{J},c12_{I}{J};
 138     /* #endfor */
 139     /* #if KERNEL_ELEC != 'None' */
 140     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 141     real             *charge;
 142     /* #endif */
 143     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 144     __m256i          gbitab;
 145     __m128i          gbitab_lo,gbitab_hi;
 146     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 147     __m256           minushalf = _mm256_set1_ps(-0.5);
 148     real             *invsqrta,*dvda,*gbtab;
 149     /* #endif */
 150     /* #if KERNEL_VDW != 'None' */
 151     int              nvdwtype;
 152     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 153     int              *vdwtype;
 154     real             *vdwparam;
 155     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 156     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 157     /* #endif */
 158     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 159     __m256i          vfitab;
 160     __m128i          vfitab_lo,vfitab_hi;
 161     __m128i          ifour       = _mm_set1_epi32(4);
 162     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 163     real             *vftab;
 164     /* #endif */
 165     /* #if 'LJEwald' in KERNEL_VDW */
 166     /* #for I,J in PAIRS_IJ */
 167     __m256           c6grid_{I}{J};
 168     /* #endfor */
 169     real             *vdwgridparam;
 170     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 171     __m256           one_half  = _mm256_set1_ps(0.5);
 172     __m256           minus_one = _mm256_set1_ps(-1.0);
 173     /* #endif */
 174     /* #if 'Ewald' in KERNEL_ELEC */
 175     __m256i          ewitab;
 176     __m128i          ewitab_lo,ewitab_hi;
 177     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 178     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 179     real             *ewtab;
 180     /* #endif */
 181     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 182     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 183     real             rswitch_scalar,d_scalar;
 184     /* #endif */
 185     __m256           dummy_mask,cutoff_mask;
 186     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 187     __m256           one     = _mm256_set1_ps(1.0);
 188     __m256           two     = _mm256_set1_ps(2.0);
 189     x                = xx[0];
 190     f                = ff[0];
 191
 192     nri              = nlist->nri;
 193     iinr             = nlist->iinr;
 194     jindex           = nlist->jindex;
 195     jjnr             = nlist->jjnr;
 196     shiftidx         = nlist->shift;
 197     gid              = nlist->gid;
 198     shiftvec         = fr->shift_vec[0];
 199     fshift           = fr->fshift[0];
 200     /* #if KERNEL_ELEC != 'None' */
 201     facel            = _mm256_set1_ps(fr->ic->epsfac);
 202     charge           = mdatoms->chargeA;
 203     /*     #if 'ReactionField' in KERNEL_ELEC */
 204     krf              = _mm256_set1_ps(fr->ic->k_rf);
 205     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 206     crf              = _mm256_set1_ps(fr->ic->c_rf);
 207     /*     #endif */
 208     /* #endif */
 209     /* #if KERNEL_VDW != 'None' */
 210     nvdwtype         = fr->ntype;
 211     vdwparam         = fr->nbfp;
 212     vdwtype          = mdatoms->typeA;
 213     /* #endif */
 214     /* #if 'LJEwald' in KERNEL_VDW */
 215     vdwgridparam     = fr->ljpme_c6grid;
 216     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 217     ewclj            = _mm256_set1_ps(fr->ic->ewaldcoeff_lj);
 218     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 219     /* #endif */
 220
 221     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 222     vftab            = kernel_data->table_elec_vdw->data;
 223     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 224     /* #elif 'Table' in KERNEL_ELEC */
 225     vftab            = kernel_data->table_elec->data;
 226     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 227     /* #elif 'Table' in KERNEL_VDW */
 228     vftab            = kernel_data->table_vdw->data;
 229     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 230     /* #endif */
 231
 232     /* #if 'Ewald' in KERNEL_ELEC */
 233     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 234     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 235     beta2            = _mm256_mul_ps(beta,beta);
 236     beta3            = _mm256_mul_ps(beta,beta2);
 237
 238     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 239     ewtab            = fr->ic->tabq_coul_F;
 240     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 241     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 242     /*     #else */
 243     ewtab            = fr->ic->tabq_coul_FDV0;
 244     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 245     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 246      /*     #endif */
 247     /* #endif */
 248
 249     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 250     invsqrta         = fr->invsqrta;
 251     dvda             = fr->dvda;
 252     gbtabscale       = _mm256_set1_ps(fr->gbtab->scale);
 253     gbtab            = fr->gbtab->data;
 254     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->ic->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 255     /* #endif */
 256
 257     /* #if 'Water' in GEOMETRY_I */
 258     /* Setup water-specific parameters */
 259     inr              = nlist->iinr[0];
 260     /*     #for I in PARTICLES_ELEC_I */
 261     iq{I}              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+{I}]));
 262     /*     #endfor */
 263     /*     #for I in PARTICLES_VDW_I */
 264     vdwioffsetptr{I}   = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
 265     /*         #if 'LJEwald' in KERNEL_VDW */
 266     vdwgridioffsetptr{I} = vdwgridparam+2*nvdwtype*vdwtype[inr+{I}];
 267     /*         #endif */
 268     /*     #endfor */
 269     /* #endif */
 270
 271     /* #if 'Water' in GEOMETRY_J */
 272     /*     #for J in PARTICLES_ELEC_J */
 273     jq{J}              = _mm256_set1_ps(charge[inr+{J}]);
 274     /*     #endfor */
 275     /*     #for J in PARTICLES_VDW_J */
 276     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 277     /*     #endfor */
 278     /*     #for I,J in PAIRS_IJ */
 279     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 280     qq{I}{J}             = _mm256_mul_ps(iq{I},jq{J});
 281     /*         #endif */
 282     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 283         /*         #if 'LJEwald' in KERNEL_VDW */
 284     c6_{I}{J}            = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A]);
 285     c12_{I}{J}           = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A+1]);
 286     c6grid_{I}{J}        = _mm256_set1_ps(vdwgridioffsetptr{I}[vdwjidx{J}A]);
 287         /*         #else */
 288     c6_{I}{J}            = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A]);
 289     c12_{I}{J}           = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A+1]);
 290         /*         #endif */
 291     /*         #endif */
 292     /*     #endfor */
 293     /* #endif */
 294
 295     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 296     /*     #if KERNEL_ELEC!='None' */
 297     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 298     rcutoff_scalar   = fr->ic->rcoulomb;
 299     /*     #else */
 300     rcutoff_scalar   = fr->ic->rvdw;
 301     /*     #endif */
 302     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 303     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 304     /* #endif */
 305
 306     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 307     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 308     rvdw             = _mm256_set1_ps(fr->ic->rvdw);
 309     /* #endif */
 310
 311     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 312     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 313     rswitch_scalar   = fr->ic->rcoulomb_switch;
 314     rswitch          = _mm256_set1_ps(rswitch_scalar);
 315     /*     #else */
 316     rswitch_scalar   = fr->ic->rvdw_switch;
 317     rswitch          = _mm256_set1_ps(rswitch_scalar);
 318     /*     #endif */
 319     /* Setup switch parameters */
 320     d_scalar         = rcutoff_scalar-rswitch_scalar;
 321     d                = _mm256_set1_ps(d_scalar);
 322     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 323     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 324     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 325     /*     #if 'Force' in KERNEL_VF */
 326     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 327     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 328     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 329     /*     #endif */
 330     /* #endif */
 331
 332     /* Avoid stupid compiler warnings */
 333     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 334     j_coord_offsetA = 0;
 335     j_coord_offsetB = 0;
 336     j_coord_offsetC = 0;
 337     j_coord_offsetD = 0;
 338     j_coord_offsetE = 0;
 339     j_coord_offsetF = 0;
 340     j_coord_offsetG = 0;
 341     j_coord_offsetH = 0;
 342
 343     /* ## Keep track of the floating point operations we issue for reporting! */
 344     /* #define OUTERFLOPS 0 */
 345     outeriter        = 0;
 346     inneriter        = 0;
 347
 348     for(iidx=0;iidx<4*DIM;iidx++)
 349     {
 350         scratch[iidx] = 0.0;
 351     }
 352
 353     /* Start outer loop over neighborlists */
 354     for(iidx=0; iidx<nri; iidx++)
 355     {
 356         /* Load shift vector for this list */
 357         i_shift_offset   = DIM*shiftidx[iidx];
 358
 359         /* Load limits for loop over neighbors */
 360         j_index_start    = jindex[iidx];
 361         j_index_end      = jindex[iidx+1];
 362
 363         /* Get outer coordinate index */
 364         inr              = iinr[iidx];
 365         i_coord_offset   = DIM*inr;
 366
 367         /* Load i particle coords and add shift vector */
 368         /* #if GEOMETRY_I == 'Particle' */
 369         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 370         /* #elif GEOMETRY_I == 'Water3' */
 371         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 372                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 373         /* #elif GEOMETRY_I == 'Water4' */
 374         /*     #if 0 in PARTICLES_I                 */
 375         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 376                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 377         /*     #else                                */
 378         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 379                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 380         /*     #endif                               */
 381         /* #endif                                   */
 382
 383         /* #if 'Force' in KERNEL_VF */
 384         /*     #for I in PARTICLES_I */
 385         fix{I}             = _mm256_setzero_ps();
 386         fiy{I}             = _mm256_setzero_ps();
 387         fiz{I}             = _mm256_setzero_ps();
 388         /*     #endfor */
 389         /* #endif */
 390
 391         /* ## For water we already preloaded parameters at the start of the kernel */
 392         /* #if not 'Water' in GEOMETRY_I */
 393         /* Load parameters for i particles */
 394         /*     #for I in PARTICLES_ELEC_I */
 395         iq{I}              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+{I}]));
 396         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 397         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 398         isai{I}            = _mm256_set1_ps(invsqrta[inr+{I}]);
 399         /*         #endif */
 400         /*     #endfor */
 401         /*     #for I in PARTICLES_VDW_I */
 402         vdwioffsetptr{I}   = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
 403         /*         #if 'LJEwald' in KERNEL_VDW */
 404         vdwgridioffsetptr{I} = vdwgridparam+2*nvdwtype*vdwtype[inr+{I}];
 405         /*         #endif */
 406         /*     #endfor */
 407         /* #endif */
 408
 409         /* #if 'Potential' in KERNEL_VF */
 410         /* Reset potential sums */
 411         /*     #if KERNEL_ELEC != 'None' */
 412         velecsum         = _mm256_setzero_ps();
 413         /*     #endif */
 414         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 415         vgbsum           = _mm256_setzero_ps();
 416         /*     #endif */
 417         /*     #if KERNEL_VDW != 'None' */
 418         vvdwsum          = _mm256_setzero_ps();
 419         /*     #endif */
 420         /* #endif */
 421         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 422         dvdasum          = _mm256_setzero_ps();
 423         /*     #endif */
 424
 425         /* #for ROUND in ['Loop','Epilogue'] */
 426
 427         /* #if ROUND =='Loop' */
 428         /* Start inner kernel loop */
 429         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 430         {
 431         /* ## First round is normal loop (next statement resets indentation) */
 432         /*     #if 0 */
 433         }
 434         /*     #endif */
 435         /* #else */
 436         if(jidx<j_index_end)
 437         {
 438         /* ## Second round is epilogue */
 439         /* #endif */
 440         /* #define INNERFLOPS 0 */
 441
 442             /* Get j neighbor index, and coordinate index */
 443             /* #if ROUND =='Loop' */
 444             jnrA             = jjnr[jidx];
 445             jnrB             = jjnr[jidx+1];
 446             jnrC             = jjnr[jidx+2];
 447             jnrD             = jjnr[jidx+3];
 448             jnrE             = jjnr[jidx+4];
 449             jnrF             = jjnr[jidx+5];
 450             jnrG             = jjnr[jidx+6];
 451             jnrH             = jjnr[jidx+7];
 452             /* #else */
 453             jnrlistA         = jjnr[jidx];
 454             jnrlistB         = jjnr[jidx+1];
 455             jnrlistC         = jjnr[jidx+2];
 456             jnrlistD         = jjnr[jidx+3];
 457             jnrlistE         = jjnr[jidx+4];
 458             jnrlistF         = jjnr[jidx+5];
 459             jnrlistG         = jjnr[jidx+6];
 460             jnrlistH         = jjnr[jidx+7];
 461             /* Sign of each element will be negative for non-real atoms.
 462              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 463              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 464              */
 465             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 466                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 467
 468             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 469             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 470             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 471             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 472             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 473             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 474             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 475             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 476             /* #endif */
 477             j_coord_offsetA  = DIM*jnrA;
 478             j_coord_offsetB  = DIM*jnrB;
 479             j_coord_offsetC  = DIM*jnrC;
 480             j_coord_offsetD  = DIM*jnrD;
 481             j_coord_offsetE  = DIM*jnrE;
 482             j_coord_offsetF  = DIM*jnrF;
 483             j_coord_offsetG  = DIM*jnrG;
 484             j_coord_offsetH  = DIM*jnrH;
 485
 486             /* load j atom coordinates */
 487             /* #if GEOMETRY_J == 'Particle'             */
 488             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 489                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 490                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 491                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 492                                                  &jx0,&jy0,&jz0);
 493             /* #elif GEOMETRY_J == 'Water3'             */
 494             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 495                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 496                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 497                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 498                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 499             /* #elif GEOMETRY_J == 'Water4'             */
 500             /*     #if 0 in PARTICLES_J                 */
 501             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 502                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 503                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 504                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 505                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 506                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 507             /*     #else                                */
 508             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 509                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 510                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 511                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 512                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 513             /*     #endif                               */
 514             /* #endif                                   */
 515
 516             /* Calculate displacement vector */
 517             /* #for I,J in PAIRS_IJ */
 518             dx{I}{J}             = _mm256_sub_ps(ix{I},jx{J});
 519             dy{I}{J}             = _mm256_sub_ps(iy{I},jy{J});
 520             dz{I}{J}             = _mm256_sub_ps(iz{I},jz{J});
 521             /*     #define INNERFLOPS INNERFLOPS+3 */
 522             /* #endfor */
 523
 524             /* Calculate squared distance and things based on it */
 525             /* #for I,J in PAIRS_IJ */
 526             rsq{I}{J}            = gmx_mm256_calc_rsq_ps(dx{I}{J},dy{I}{J},dz{I}{J});
 527             /*     #define INNERFLOPS INNERFLOPS+5 */
 528             /* #endfor */
 529
 530             /* #for I,J in PAIRS_IJ */
 531             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 532             rinv{I}{J}           = avx256_invsqrt_f(rsq{I}{J});
 533             /*         #define INNERFLOPS INNERFLOPS+5 */
 534             /*     #endif */
 535             /* #endfor */
 536
 537             /* #for I,J in PAIRS_IJ */
 538             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 539             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 540             rinvsq{I}{J}         = avx256_inv_f(rsq{I}{J});
 541             /*             #define INNERFLOPS INNERFLOPS+4 */
 542             /*         #else */
 543             rinvsq{I}{J}         = _mm256_mul_ps(rinv{I}{J},rinv{I}{J});
 544             /*             #define INNERFLOPS INNERFLOPS+1 */
 545             /*         #endif */
 546             /*     #endif */
 547             /* #endfor */
 548
 549             /* #if not 'Water' in GEOMETRY_J */
 550             /* Load parameters for j particles */
 551             /*     #for J in PARTICLES_ELEC_J */
 552             jq{J}              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+{J},charge+jnrB+{J},
 553                                                                  charge+jnrC+{J},charge+jnrD+{J},
 554                                                                  charge+jnrE+{J},charge+jnrF+{J},
 555                                                                  charge+jnrG+{J},charge+jnrH+{J});
 556             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 557             isaj{J}            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+{J},invsqrta+jnrB+{J},
 558                                                                  invsqrta+jnrC+{J},invsqrta+jnrD+{J},
 559                                                                  invsqrta+jnrE+{J},invsqrta+jnrF+{J},
 560                                                                  invsqrta+jnrG+{J},invsqrta+jnrH+{J});
 561             /*         #endif */
 562             /*     #endfor */
 563             /*     #for J in PARTICLES_VDW_J */
 564             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 565             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 566             vdwjidx{J}C        = 2*vdwtype[jnrC+{J}];
 567             vdwjidx{J}D        = 2*vdwtype[jnrD+{J}];
 568             vdwjidx{J}E        = 2*vdwtype[jnrE+{J}];
 569             vdwjidx{J}F        = 2*vdwtype[jnrF+{J}];
 570             vdwjidx{J}G        = 2*vdwtype[jnrG+{J}];
 571             vdwjidx{J}H        = 2*vdwtype[jnrH+{J}];
 572             /*     #endfor */
 573             /* #endif */
 574
 575             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 576             /*     #for J in PARTICLES_J */
 577             fjx{J}             = _mm256_setzero_ps();
 578             fjy{J}             = _mm256_setzero_ps();
 579             fjz{J}             = _mm256_setzero_ps();
 580             /*     #endfor */
 581             /* #endif */
 582
 583             /* #for I,J in PAIRS_IJ */
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 590             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 591             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 592             if (gmx_mm256_any_lt(rsq{I}{J},rcutoff2))
 593             {
 594                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 595             }
 596             /*         #endif */
 597             /*         #define INNERFLOPS INNERFLOPS+1 */
 598             /*     #endif */
 599
 600             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 601             r{I}{J}              = _mm256_mul_ps(rsq{I}{J},rinv{I}{J});
 602             /*         #if ROUND == 'Epilogue' */
 603             r{I}{J}              = _mm256_andnot_ps(dummy_mask,r{I}{J});
 604             /*             #define INNERFLOPS INNERFLOPS+1 */
 605             /*         #endif */
 606             /*         #define INNERFLOPS INNERFLOPS+1 */
 607             /*     #endif */
 608
 609             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 610             /*     #if not 'Water' in GEOMETRY_J */
 611             /* Compute parameters for interactions between i and j atoms */
 612             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 613             qq{I}{J}             = _mm256_mul_ps(iq{I},jq{J});
 614             /*             #define INNERFLOPS INNERFLOPS+1 */
 615             /*         #endif */
 616             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 617             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr{I}+vdwjidx{J}A,
 618                                             vdwioffsetptr{I}+vdwjidx{J}B,
 619                                             vdwioffsetptr{I}+vdwjidx{J}C,
 620                                             vdwioffsetptr{I}+vdwjidx{J}D,
 621                                             vdwioffsetptr{I}+vdwjidx{J}E,
 622                                             vdwioffsetptr{I}+vdwjidx{J}F,
 623                                             vdwioffsetptr{I}+vdwjidx{J}G,
 624                                             vdwioffsetptr{I}+vdwjidx{J}H,
 625                                             &c6_{I}{J},&c12_{I}{J});
 626
 627             /*         #if 'LJEwald' in KERNEL_VDW */
 628             c6grid_{I}{J}       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr{I}+vdwjidx{J}A,
 629                                                                   vdwgridioffsetptr{I}+vdwjidx{J}B,
 630                                                                   vdwgridioffsetptr{I}+vdwjidx{J}C,
 631                                                                   vdwgridioffsetptr{I}+vdwjidx{J}D,
 632                                                                   vdwgridioffsetptr{I}+vdwjidx{J}E,
 633                                                                   vdwgridioffsetptr{I}+vdwjidx{J}F,
 634                                                                   vdwgridioffsetptr{I}+vdwjidx{J}G,
 635                                                                   vdwgridioffsetptr{I}+vdwjidx{J}H);
 636             /*          #endif */
 637             /*         #endif */
 638             /*     #endif */
 639
 640             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 641             /* Calculate table index by multiplying r with table scale and truncate to integer */
 642             rt               = _mm256_mul_ps(r{I}{J},vftabscale);
 643             vfitab           = _mm256_cvttps_epi32(rt);
 644             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 645             /*         #define INNERFLOPS INNERFLOPS+4                          */
 646             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 647             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 648             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 649             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 650             /*             ## 3 tables, 4 bytes per point: multiply index by 12 */
 651             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 652             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 653             /*         #elif 'Table' in KERNEL_ELEC                             */
 654             /*             ## 1 table, 4 bytes per point: multiply index by 4   */
 655             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 656             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 657             /*         #elif 'Table' in KERNEL_VDW */
 658             /*             ## 2 tables, 4 bytes per point: multiply index by 8  */
 659             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 660             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 661             /*         #endif                                                   */
 662             /*     #endif */
 663
 664             /*     ## ELECTROSTATIC INTERACTIONS */
 665             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 666
 667             /*         #if KERNEL_ELEC=='Coulomb' */
 668
 669             /* COULOMB ELECTROSTATICS */
 670             velec            = _mm256_mul_ps(qq{I}{J},rinv{I}{J});
 671             /*             #define INNERFLOPS INNERFLOPS+1 */
 672             /*             #if 'Force' in KERNEL_VF */
 673             felec            = _mm256_mul_ps(velec,rinvsq{I}{J});
 674             /*                 #define INNERFLOPS INNERFLOPS+1 */
 675             /*             #endif */
 676
 677             /*         #elif KERNEL_ELEC=='ReactionField' */
 678
 679             /* REACTION-FIELD ELECTROSTATICS */
 680             /*             #if 'Potential' in KERNEL_VF */
 681             velec            = _mm256_mul_ps(qq{I}{J},_mm256_sub_ps(_mm256_add_ps(rinv{I}{J},_mm256_mul_ps(krf,rsq{I}{J})),crf));
 682             /*                 #define INNERFLOPS INNERFLOPS+4 */
 683             /*             #endif */
 684             /*             #if 'Force' in KERNEL_VF */
 685             felec            = _mm256_mul_ps(qq{I}{J},_mm256_sub_ps(_mm256_mul_ps(rinv{I}{J},rinvsq{I}{J}),krf2));
 686             /*                 #define INNERFLOPS INNERFLOPS+3 */
 687             /*             #endif */
 688
 689             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 690
 691             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 692             isaprod          = _mm256_mul_ps(isai{I},isaj{J});
 693             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq{I}{J},_mm256_mul_ps(isaprod,gbinvepsdiff)));
 694             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 695             /*             #define INNERFLOPS INNERFLOPS+5 */
 696
 697             /* Calculate generalized born table index - this is a separate table from the normal one,
 698              * but we use the same procedure by multiplying r with scale and truncating to integer.
 699              */
 700             rt               = _mm256_mul_ps(r{I}{J},gbscale);
 701             gbitab           = _mm256_cvttps_epi32(rt);
 702             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 703             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 704             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 705             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 706             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 707             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 708             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 709                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 710             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 711                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 712             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 713                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 714             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 715                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 716             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 717             Heps             = _mm256_mul_ps(gbeps,H);
 718             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 719             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 720             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 721             /*             #define INNERFLOPS INNERFLOPS+10 */
 722
 723             /*             #if 'Force' in KERNEL_VF */
 724             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 725             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 726             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r{I}{J})));
 727             /*                 #if ROUND == 'Epilogue' */
 728             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 729             /*                 #endif */
 730             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 731             /*                 #if ROUND == 'Loop' */
 732             fjptrA           = dvda+jnrA;
 733             fjptrB           = dvda+jnrB;
 734             fjptrC           = dvda+jnrC;
 735             fjptrD           = dvda+jnrD;
 736             fjptrE           = dvda+jnrE;
 737             fjptrF           = dvda+jnrF;
 738             fjptrG           = dvda+jnrG;
 739             fjptrH           = dvda+jnrH;
 740             /*                 #else */
 741             /* 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. */
 742             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 743             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 744             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 745             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 746             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 747             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 748             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 749             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 750             /*                 #endif */
 751             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 752                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj{J},isaj{J})));
 753             /*                 #define INNERFLOPS INNERFLOPS+12 */
 754             /*             #endif */
 755             velec            = _mm256_mul_ps(qq{I}{J},rinv{I}{J});
 756             /*                 #define INNERFLOPS INNERFLOPS+1 */
 757             /*             #if 'Force' in KERNEL_VF */
 758             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv{I}{J}),fgb),rinv{I}{J});
 759             /*                 #define INNERFLOPS INNERFLOPS+3 */
 760             /*             #endif */
 761
 762             /*         #elif KERNEL_ELEC=='Ewald' */
 763             /* EWALD ELECTROSTATICS */
 764
 765             /* Analytical PME correction */
 766             zeta2            = _mm256_mul_ps(beta2,rsq{I}{J});
 767             /*             #if 'Force' in KERNEL_VF */
 768             rinv3            = _mm256_mul_ps(rinvsq{I}{J},rinv{I}{J});
 769             pmecorrF         = avx256_pmecorrF_f(zeta2);
 770             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 771             felec            = _mm256_mul_ps(qq{I}{J},felec);
 772             /*                 #define INNERFLOPS INNERFLOPS+31 */
 773             /*             #endif */
 774             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 775             pmecorrV         = avx256_pmecorrV_f(zeta2);
 776             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 777             /*                 #define INNERFLOPS INNERFLOPS+27 */
 778             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 779             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv{I}{J},sh_ewald),pmecorrV);
 780             /*                     #define INNERFLOPS INNERFLOPS+21 */
 781             /*                 #else */
 782             velec            = _mm256_sub_ps(rinv{I}{J},pmecorrV);
 783             /*                 #endif */
 784             velec            = _mm256_mul_ps(qq{I}{J},velec);
 785             /*             #endif */
 786
 787             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 788
 789             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 790             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 791                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 792             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 793                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 794             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 795                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 796             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 797                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 798             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 799             Heps             = _mm256_mul_ps(vfeps,H);
 800             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 801             /*             #define INNERFLOPS INNERFLOPS+4 */
 802             /*             #if 'Potential' in KERNEL_VF */
 803             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 804             velec            = _mm256_mul_ps(qq{I}{J},VV);
 805             /*                 #define INNERFLOPS INNERFLOPS+3 */
 806             /*             #endif */
 807             /*             #if 'Force' in KERNEL_VF */
 808             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 809             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq{I}{J},FF),_mm256_mul_ps(vftabscale,rinv{I}{J})));
 810             /*                 #define INNERFLOPS INNERFLOPS+7 */
 811             /*             #endif */
 812             /*         #endif */
 813             /*         ## End of check for electrostatics interaction forms */
 814             /*     #endif */
 815             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 816
 817             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 818
 819             /*         #if KERNEL_VDW=='LennardJones' */
 820
 821             /* LENNARD-JONES DISPERSION/REPULSION */
 822
 823             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 824             /*             #define INNERFLOPS INNERFLOPS+2 */
 825             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 826             vvdw6            = _mm256_mul_ps(c6_{I}{J},rinvsix);
 827             vvdw12           = _mm256_mul_ps(c12_{I}{J},_mm256_mul_ps(rinvsix,rinvsix));
 828             /*                 #define INNERFLOPS INNERFLOPS+3 */
 829             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 830             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_{I}{J},_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 831                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
 832             /*                     #define INNERFLOPS INNERFLOPS+8 */
 833             /*                 #else */
 834             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 835             /*                     #define INNERFLOPS INNERFLOPS+3 */
 836             /*                 #endif */
 837             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 838             /*                 #if 'Force' in KERNEL_VF */
 839             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq{I}{J});
 840             /*                     #define INNERFLOPS INNERFLOPS+2 */
 841             /*                 #endif */
 842             /*             #elif KERNEL_VF=='Force' */
 843             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 844             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_{I}{J},rinvsix),c6_{I}{J}),_mm256_mul_ps(rinvsix,rinvsq{I}{J}));
 845             /*                 #define INNERFLOPS INNERFLOPS+4 */
 846             /*             #endif */
 847
 848             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 849
 850             /* CUBIC SPLINE TABLE DISPERSION */
 851             /*             #if 'Table' in KERNEL_ELEC */
 852             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 853             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 854             /*             #endif                     */
 855             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 856                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 857             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 858                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 859             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 860                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 861             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 862                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 863             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 864             Heps             = _mm256_mul_ps(vfeps,H);
 865             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 866             /*             #define INNERFLOPS INNERFLOPS+4 */
 867             /*             #if 'Potential' in KERNEL_VF */
 868             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 869             vvdw6            = _mm256_mul_ps(c6_{I}{J},VV);
 870             /*                 #define INNERFLOPS INNERFLOPS+3 */
 871             /*             #endif */
 872             /*             #if 'Force' in KERNEL_VF */
 873             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 874             fvdw6            = _mm256_mul_ps(c6_{I}{J},FF);
 875             /*                 #define INNERFLOPS INNERFLOPS+4 */
 876             /*             #endif */
 877
 878             /* CUBIC SPLINE TABLE REPULSION */
 879             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 880             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 881             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 882                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 883             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 884                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 885             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 886                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 887             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 888                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 889             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 890             Heps             = _mm256_mul_ps(vfeps,H);
 891             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 892             /*             #define INNERFLOPS INNERFLOPS+4 */
 893             /*             #if 'Potential' in KERNEL_VF */
 894             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 895             vvdw12           = _mm256_mul_ps(c12_{I}{J},VV);
 896             /*                 #define INNERFLOPS INNERFLOPS+3 */
 897             /*             #endif */
 898             /*             #if 'Force' in KERNEL_VF */
 899             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 900             fvdw12           = _mm256_mul_ps(c12_{I}{J},FF);
 901             /*                 #define INNERFLOPS INNERFLOPS+5 */
 902             /*             #endif */
 903             /*             #if 'Potential' in KERNEL_VF */
 904             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 905             /*                 #define INNERFLOPS INNERFLOPS+1 */
 906             /*             #endif */
 907             /*             #if 'Force' in KERNEL_VF */
 908             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv{I}{J})));
 909             /*                 #define INNERFLOPS INNERFLOPS+4 */
 910             /*             #endif */
 911
 912             /*         #elif KERNEL_VDW=='LJEwald' */
 913
 914             /* Analytical LJ-PME */
 915             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 916             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq{I}{J});
 917             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 918             exponent         = avx256_exp_f(ewcljrsq);
 919             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 920             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 921             /*                 #define INNERFLOPS INNERFLOPS+11 */
 922             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 923             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 924             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_{I}{J},_mm256_mul_ps(c6grid_{I}{J},_mm256_sub_ps(one,poly))),rinvsix);
 925             vvdw12           = _mm256_mul_ps(c12_{I}{J},_mm256_mul_ps(rinvsix,rinvsix));
 926             /*                 #define INNERFLOPS INNERFLOPS+6 */
 927             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 928             vvdw             = _mm256_sub_ps(_mm256_mul_ps( _mm256_sub_ps(vvdw12 , _mm256_mul_ps(c12_{I}{J},_mm256_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 929                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_add_ps(_mm256_mul_ps(c6_{I}{J},sh_vdw_invrcut6),_mm256_mul_ps(c6grid_{I}{J},sh_lj_ewald))),one_sixth));
 930             /*                     #define INNERFLOPS INNERFLOPS+10 */
 931             /*                 #else */
 932             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 933             /*                 #define INNERFLOPS INNERFLOPS+3 */
 934             /*                 #endif */
 935             /*                  ## Check for force inside potential check, i.e. this means we already did the potential part */
 936             /*                  #if 'Force' in KERNEL_VF */
 937             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 938             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,_mm256_sub_ps(vvdw6,_mm256_mul_ps(_mm256_mul_ps(c6grid_{I}{J},one_sixth),_mm256_mul_ps(exponent,ewclj6)))),rinvsq{I}{J});
 939             /*                 #define INNERFLOPS INNERFLOPS+6 */
 940             /*                  #endif */
 941             /*              #elif KERNEL_VF=='Force' */
 942             /* f6A = 6 * C6grid * (1 - poly) */
 943             f6A              = _mm256_mul_ps(c6grid_{I}{J},_mm256_sub_ps(one,poly));
 944             /* f6B = C6grid * exponent * beta^6 */
 945             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_{I}{J},one_sixth),_mm256_mul_ps(exponent,ewclj6));
 946             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 947             fvdw              = _mm256_mul_ps(_mm256_add_ps(_mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_{I}{J},rinvsix),_mm256_sub_ps(c6_{I}{J},f6A)),rinvsix),f6B),rinvsq{I}{J});
 948             /*                 #define INNERFLOPS INNERFLOPS+11 */
 949             /*              #endif */
 950             /*         #endif */
 951             /*         ## End of check for vdw interaction forms */
 952             /*     #endif */
 953             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 954
 955             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 956             d                = _mm256_sub_ps(r{I}{J},rswitch);
 957             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 958             d2               = _mm256_mul_ps(d,d);
 959             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 960             /*         #define INNERFLOPS INNERFLOPS+10 */
 961
 962             /*         #if 'Force' in KERNEL_VF */
 963             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 964             /*             #define INNERFLOPS INNERFLOPS+5 */
 965             /*         #endif */
 966
 967             /* Evaluate switch function */
 968             /*         #if 'Force' in KERNEL_VF */
 969             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 970             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 971             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv{I}{J},_mm256_mul_ps(velec,dsw)) );
 972             /*                 #define INNERFLOPS INNERFLOPS+4 */
 973             /*             #endif */
 974             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 975             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv{I}{J},_mm256_mul_ps(vvdw,dsw)) );
 976             /*                 #define INNERFLOPS INNERFLOPS+4 */
 977             /*             #endif */
 978             /*         #endif */
 979             /*         #if 'Potential' in KERNEL_VF */
 980             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 981             velec            = _mm256_mul_ps(velec,sw);
 982             /*                 #define INNERFLOPS INNERFLOPS+1 */
 983             /*             #endif */
 984             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 985             vvdw             = _mm256_mul_ps(vvdw,sw);
 986             /*                 #define INNERFLOPS INNERFLOPS+1 */
 987             /*             #endif */
 988             /*         #endif */
 989             /*     #endif */
 990             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 991             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 992             cutoff_mask      = _mm256_cmp_ps(rsq{I}{J},rcutoff2,_CMP_LT_OQ);
 993             /*         #define INNERFLOPS INNERFLOPS+1 */
 994             /*     #endif */
 995
 996             /*     #if 'Potential' in KERNEL_VF */
 997             /* Update potential sum for this i atom from the interaction with this j atom. */
 998             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 999             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
1000             velec            = _mm256_and_ps(velec,cutoff_mask);
1001             /*                 #define INNERFLOPS INNERFLOPS+1 */
1002             /*             #endif                                       */
1003             /*             #if ROUND == 'Epilogue' */
1004             velec            = _mm256_andnot_ps(dummy_mask,velec);
1005             /*             #endif */
1006             velecsum         = _mm256_add_ps(velecsum,velec);
1007             /*             #define INNERFLOPS INNERFLOPS+1 */
1008             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
1009             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
1010             vgb              = _mm256_and_ps(vgb,cutoff_mask);
1011             /*                 #define INNERFLOPS INNERFLOPS+1 */
1012             /*             #endif                                       */
1013             /*             #if ROUND == 'Epilogue' */
1014             vgb              = _mm256_andnot_ps(dummy_mask,vgb);
1015             /*             #endif */
1016             vgbsum           = _mm256_add_ps(vgbsum,vgb);
1017             /*                 #define INNERFLOPS INNERFLOPS+1 */
1018             /*             #endif */
1019             /*         #endif */
1020             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
1021             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
1022             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
1023             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
1024             /*                 #define INNERFLOPS INNERFLOPS+1 */
1025             /*             #endif                                       */
1026             /*             #if ROUND == 'Epilogue' */
1027             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
1028             /*             #endif */
1029             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
1030             /*             #define INNERFLOPS INNERFLOPS+1 */
1031             /*         #endif */
1032             /*     #endif */
1033
1034             /*     #if 'Force' in KERNEL_VF */
1035
1036             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
1037             fscal            = _mm256_add_ps(felec,fvdw);
1038             /*             #define INNERFLOPS INNERFLOPS+1 */
1039             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
1040             fscal            = felec;
1041             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
1042             fscal            = fvdw;
1043             /*        #endif */
1044
1045             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
1046             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
1047             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1048             /*                 #define INNERFLOPS INNERFLOPS+1 */
1049             /*             #endif                                       */
1050
1051             /*             #if ROUND == 'Epilogue' */
1052             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1053             /*             #endif */
1054
1055             /* Calculate temporary vectorial force */
1056             tx               = _mm256_mul_ps(fscal,dx{I}{J});
1057             ty               = _mm256_mul_ps(fscal,dy{I}{J});
1058             tz               = _mm256_mul_ps(fscal,dz{I}{J});
1059
1060             /* Update vectorial force */
1061             fix{I}             = _mm256_add_ps(fix{I},tx);
1062             fiy{I}             = _mm256_add_ps(fiy{I},ty);
1063             fiz{I}             = _mm256_add_ps(fiz{I},tz);
1064             /*             #define INNERFLOPS INNERFLOPS+6 */
1065
1066             /* #if GEOMETRY_I == 'Particle'             */
1067             /*     #if ROUND == 'Loop' */
1068             fjptrA             = f+j_coord_offsetA;
1069             fjptrB             = f+j_coord_offsetB;
1070             fjptrC             = f+j_coord_offsetC;
1071             fjptrD             = f+j_coord_offsetD;
1072             fjptrE             = f+j_coord_offsetE;
1073             fjptrF             = f+j_coord_offsetF;
1074             fjptrG             = f+j_coord_offsetG;
1075             fjptrH             = f+j_coord_offsetH;
1076             /*     #else */
1077             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1078             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1079             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1080             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1081             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1082             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1083             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1084             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1085             /*     #endif */
1086             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1087             /*     #define INNERFLOPS INNERFLOPS+3      */
1088             /* #else                                    */
1089             fjx{J}             = _mm256_add_ps(fjx{J},tx);
1090             fjy{J}             = _mm256_add_ps(fjy{J},ty);
1091             fjz{J}             = _mm256_add_ps(fjz{J},tz);
1092             /*     #define INNERFLOPS INNERFLOPS+3      */
1093             /* #endif                                   */
1094
1095             /*     #endif */
1096
1097             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
1098             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
1099             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
1100             {
1101                 /*     #endif */
1102             }
1103             /*     #endif */
1104             /*    ## End of check for the interaction being outside the cutoff */
1105
1106             /* #endfor */
1107             /* ## End of loop over i-j interaction pairs */
1108
1109             /* #if GEOMETRY_I != 'Particle' */
1110             /*     #if ROUND == 'Loop' */
1111             fjptrA             = f+j_coord_offsetA;
1112             fjptrB             = f+j_coord_offsetB;
1113             fjptrC             = f+j_coord_offsetC;
1114             fjptrD             = f+j_coord_offsetD;
1115             fjptrE             = f+j_coord_offsetE;
1116             fjptrF             = f+j_coord_offsetF;
1117             fjptrG             = f+j_coord_offsetG;
1118             fjptrH             = f+j_coord_offsetH;
1119             /*     #else */
1120             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1121             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1122             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1123             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1124             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1125             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1126             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1127             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1128             /*     #endif */
1129             /* #endif */
1130
1131             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
1132             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1133             /*     #define INNERFLOPS INNERFLOPS+3      */
1134             /* #elif GEOMETRY_J == 'Water3'             */
1135             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1136                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1137             /*     #define INNERFLOPS INNERFLOPS+9      */
1138             /* #elif GEOMETRY_J == 'Water4'             */
1139             /*     #if 0 in PARTICLES_J                 */
1140             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1141                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1142                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1143             /*     #define INNERFLOPS INNERFLOPS+12     */
1144             /*     #else                                */
1145             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1146                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1147                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1148             /*     #define INNERFLOPS INNERFLOPS+9      */
1149             /*     #endif                               */
1150             /* #endif                                   */
1151
1152             /* Inner loop uses {INNERFLOPS} flops */
1153         }
1154
1155         /* #endfor */
1156
1157         /* End of innermost loop */
1158
1159         /* #if 'Force' in KERNEL_VF */
1160         /*     #if GEOMETRY_I == 'Particle'            */
1161         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1162                                                  f+i_coord_offset,fshift+i_shift_offset);
1163         /*         #define OUTERFLOPS OUTERFLOPS+6     */
1164         /*     #elif GEOMETRY_I == 'Water3'            */
1165         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1166                                                  f+i_coord_offset,fshift+i_shift_offset);
1167         /*         #define OUTERFLOPS OUTERFLOPS+18    */
1168         /*     #elif GEOMETRY_I == 'Water4'            */
1169         /*         #if 0 in PARTICLES_I                */
1170         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1171                                                  f+i_coord_offset,fshift+i_shift_offset);
1172         /*             #define OUTERFLOPS OUTERFLOPS+24    */
1173         /*         #else                               */
1174         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1175                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1176         /*             #define OUTERFLOPS OUTERFLOPS+18    */
1177         /*         #endif                              */
1178         /*     #endif                                  */
1179         /* #endif                                      */
1180
1181         /* #if 'Potential' in KERNEL_VF */
1182         ggid                        = gid[iidx];
1183         /* Update potential energies */
1184         /*     #if KERNEL_ELEC != 'None' */
1185         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1186         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1187         /*     #endif */
1188         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
1189         gmx_mm256_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
1190         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1191         /*     #endif */
1192         /*     #if KERNEL_VDW != 'None' */
1193         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1194         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1195         /*     #endif */
1196         /* #endif */
1197         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1198         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai{I},isai{I}));
1199         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
1200         /*     #endif */
1201
1202         /* Increment number of inner iterations */
1203         inneriter                  += j_index_end - j_index_start;
1204
1205         /* Outer loop uses {OUTERFLOPS} flops */
1206     }
1207
1208     /* Increment number of outer iterations */
1209     outeriter        += nri;
1210
1211     /* Update outer/inner flops */
1212     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1213     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1214     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1215     /* #if GEOMETRY_I == 'Water3'            */
1216     /*     #define ISUFFIX '_W3'             */
1217     /* #elif GEOMETRY_I == 'Water4'          */
1218     /*     #define ISUFFIX '_W4'             */
1219     /* #else                                 */
1220     /*     #define ISUFFIX ''                */
1221     /* #endif                                */
1222     /* #if GEOMETRY_J == 'Water3'            */
1223     /*     #define JSUFFIX 'W3'              */
1224     /* #elif GEOMETRY_J == 'Water4'          */
1225     /*     #define JSUFFIX 'W4'              */
1226     /* #else                                 */
1227     /*     #define JSUFFIX ''                */
1228     /* #endif                                */
1229     /* #if 'PotentialAndForce' in KERNEL_VF  */
1230     /*     #define VFSUFFIX  '_VF'           */
1231     /* #elif 'Potential' in KERNEL_VF        */
1232     /*     #define VFSUFFIX '_V'             */
1233     /* #else                                 */
1234     /*     #define VFSUFFIX '_F'             */
1235     /* #endif                                */
1236
1237     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1238     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1239     /* #elif KERNEL_ELEC != 'None' */
1240     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1241     /* #else */
1242     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1243     /* #endif  */
1244 }