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, 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 "config.h"
  40
  41 #include <math.h>
  42
  43 #include "../nb_kernel.h"
  44 #include "types/simple.h"
  45 #include "gromacs/math/vec.h"
  46 #include "nrnb.h"
  47
  48 #include "gromacs/simd/math_x86_avx_256_single.h"
  49 #include "kernelutil_x86_avx_256_single.h"
  50 /* #endif */
  51
  52 /* ## List of variables set by the generating script:                                    */
  53 /* ##                                                                                    */
  54 /* ## Setttings that apply to the entire kernel:                                         */
  55 /* ## KERNEL_ELEC:           String, choice for electrostatic interactions               */
  56 /* ## KERNEL_VDW:            String, choice for van der Waals interactions               */
  57 /* ## KERNEL_NAME:           String, name of this kernel                                 */
  58 /* ## KERNEL_VF:             String telling if we calculate potential, force, or both    */
  59 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
  60 /* ##                                                                                    */
  61 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops:             */
  62 /* ## PARTICLES_I[]/         Arrays with lists of i/j particles to use in kernel. It is  */
  63 /* ## PARTICLES_J[]:         just [0] for particle geometry, but can be longer for water */
  64 /* ## PARTICLES_ELEC_I[]/    Arrays with lists of i/j particle that have electrostatics  */
  65 /* ## PARTICLES_ELEC_J[]:    interactions that should be calculated in this kernel.      */
  66 /* ## PARTICLES_VDW_I[]/     Arrays with the list of i/j particle that have VdW          */
  67 /* ## PARTICLES_VDW_J[]:     interactions that should be calculated in this kernel.      */
  68 /* ##                                                                                    */
  69 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle)   */
  70 /* ## PAIRS_IJ[]:            Array with (i,j) tuples of pairs for which interactions     */
  71 /* ##                        should be calculated in this kernel. Zero-charge particles  */
  72 /* ##                        do not have interactions with particles without vdw, and    */
  73 /* ##                        Vdw-only interactions are not evaluated in a no-vdw-kernel. */
  74 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
  75 /* ##                        For each i-j pair, the element [I][J] is a list of strings  */
  76 /* ##                        defining properties/flags of this interaction. Examples     */
  77 /* ##                        include 'electrostatics'/'vdw' if that type of interaction  */
  78 /* ##                        should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values  */
  79 /* ##                        are needed, and 'exactcutoff' or 'shift','switch' to        */
  80 /* ##                        decide if the force/potential should be modified. This way  */
  81 /* ##                        we only calculate values absolutely needed for each case.   */
  82
  83 /* ## Calculate the size and offset for (merged/interleaved) table data */
  84
  85
  86
  87
  88 /*
  89  * Gromacs nonbonded kernel:   {KERNEL_NAME}
  90  * Electrostatics interaction: {KERNEL_ELEC}
  91  * VdW interaction:            {KERNEL_VDW}
  92  * Geometry:                   {GEOMETRY_I}-{GEOMETRY_J}
  93  * Calculate force/pot:        {KERNEL_VF}
  94  */
  95 void
  96 {KERNEL_NAME}
  97                     (t_nblist                    * gmx_restrict       nlist,
  98                      rvec                        * gmx_restrict          xx,
  99                      rvec                        * gmx_restrict          ff,
 100                      t_forcerec                  * gmx_restrict          fr,
 101                      t_mdatoms                   * gmx_restrict     mdatoms,
 102                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 103                      t_nrnb                      * gmx_restrict        nrnb)
 104 {
 105     /* ## Not all variables are used for all kernels, but any optimizing compiler fixes that, */
 106     /* ## so there is no point in going to extremes to exclude variables that are not needed. */
 107     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 108      * just 0 for non-waters.
 109      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 110      * jnr indices corresponding to data put in the four positions in the SIMD register.
 111      */
 112     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 113     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 114     int              jnrA,jnrB,jnrC,jnrD;
 115     int              jnrE,jnrF,jnrG,jnrH;
 116     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 117     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 118     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 119     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 120     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 121     real             rcutoff_scalar;
 122     real             *shiftvec,*fshift,*x,*f;
 123     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 124     real             scratch[4*DIM];
 125     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 126     /* #for I in PARTICLES_I */
 127     real *           vdwioffsetptr{I};
 128     /* #if 'LJEwald' in KERNEL_VDW */
 129     real *           vdwgridioffsetptr{I};
 130     /* #endif                      */
 131     __m256           ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 132     /* #endfor */
 133     /* #for J in PARTICLES_J */
 134     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;
 135     __m256           jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 136     /* #endfor */
 137     /* #for I,J in PAIRS_IJ */
 138     __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};
 139     /* #endfor */
 140     /* #if KERNEL_ELEC != 'None' */
 141     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 142     real             *charge;
 143     /* #endif */
 144     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 145     __m256i          gbitab;
 146     __m128i          gbitab_lo,gbitab_hi;
 147     __m256           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 148     __m256           minushalf = _mm256_set1_ps(-0.5);
 149     real             *invsqrta,*dvda,*gbtab;
 150     /* #endif */
 151     /* #if KERNEL_VDW != 'None' */
 152     int              nvdwtype;
 153     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 154     int              *vdwtype;
 155     real             *vdwparam;
 156     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 157     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 158     /* #endif */
 159     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 160     __m256i          vfitab;
 161     __m128i          vfitab_lo,vfitab_hi;
 162     __m128i          ifour       = _mm_set1_epi32(4);
 163     __m256           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 164     real             *vftab;
 165     /* #endif */
 166     /* #if 'LJEwald' in KERNEL_VDW */
 167     /* #for I,J in PAIRS_IJ */
 168     __m256           c6grid_{I}{J};
 169     /* #endfor */
 170     real             *vdwgridparam;
 171     __m256           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 172     __m256           one_half  = _mm256_set1_ps(0.5);
 173     __m256           minus_one = _mm256_set1_ps(-1.0);
 174     /* #endif */
 175     /* #if 'Ewald' in KERNEL_ELEC */
 176     __m256i          ewitab;
 177     __m128i          ewitab_lo,ewitab_hi;
 178     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 179     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 180     real             *ewtab;
 181     /* #endif */
 182     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 183     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 184     real             rswitch_scalar,d_scalar;
 185     /* #endif */
 186     __m256           dummy_mask,cutoff_mask;
 187     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 188     __m256           one     = _mm256_set1_ps(1.0);
 189     __m256           two     = _mm256_set1_ps(2.0);
 190     x                = xx[0];
 191     f                = ff[0];
 192
 193     nri              = nlist->nri;
 194     iinr             = nlist->iinr;
 195     jindex           = nlist->jindex;
 196     jjnr             = nlist->jjnr;
 197     shiftidx         = nlist->shift;
 198     gid              = nlist->gid;
 199     shiftvec         = fr->shift_vec[0];
 200     fshift           = fr->fshift[0];
 201     /* #if KERNEL_ELEC != 'None' */
 202     facel            = _mm256_set1_ps(fr->epsfac);
 203     charge           = mdatoms->chargeA;
 204     /*     #if 'ReactionField' in KERNEL_ELEC */
 205     krf              = _mm256_set1_ps(fr->ic->k_rf);
 206     krf2             = _mm256_set1_ps(fr->ic->k_rf*2.0);
 207     crf              = _mm256_set1_ps(fr->ic->c_rf);
 208     /*     #endif */
 209     /* #endif */
 210     /* #if KERNEL_VDW != 'None' */
 211     nvdwtype         = fr->ntype;
 212     vdwparam         = fr->nbfp;
 213     vdwtype          = mdatoms->typeA;
 214     /* #endif */
 215     /* #if 'LJEwald' in KERNEL_VDW */
 216     vdwgridparam     = fr->ljpme_c6grid;
 217     sh_lj_ewald      = _mm256_set1_ps(fr->ic->sh_lj_ewald);
 218     ewclj            = _mm256_set1_ps(fr->ewaldcoeff_lj);
 219     ewclj2           = _mm256_mul_ps(minus_one,_mm256_mul_ps(ewclj,ewclj));
 220     /* #endif */
 221
 222     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 223     vftab            = kernel_data->table_elec_vdw->data;
 224     vftabscale       = _mm256_set1_ps(kernel_data->table_elec_vdw->scale);
 225     /* #elif 'Table' in KERNEL_ELEC */
 226     vftab            = kernel_data->table_elec->data;
 227     vftabscale       = _mm256_set1_ps(kernel_data->table_elec->scale);
 228     /* #elif 'Table' in KERNEL_VDW */
 229     vftab            = kernel_data->table_vdw->data;
 230     vftabscale       = _mm256_set1_ps(kernel_data->table_vdw->scale);
 231     /* #endif */
 232
 233     /* #if 'Ewald' in KERNEL_ELEC */
 234     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 235     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 236     beta2            = _mm256_mul_ps(beta,beta);
 237     beta3            = _mm256_mul_ps(beta,beta2);
 238
 239     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 240     ewtab            = fr->ic->tabq_coul_F;
 241     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 242     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 243     /*     #else */
 244     ewtab            = fr->ic->tabq_coul_FDV0;
 245     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 246     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 247      /*     #endif */
 248     /* #endif */
 249
 250     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 251     invsqrta         = fr->invsqrta;
 252     dvda             = fr->dvda;
 253     gbtabscale       = _mm256_set1_ps(fr->gbtab.scale);
 254     gbtab            = fr->gbtab.data;
 255     gbinvepsdiff     = _mm256_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 256     /* #endif */
 257
 258     /* #if 'Water' in GEOMETRY_I */
 259     /* Setup water-specific parameters */
 260     inr              = nlist->iinr[0];
 261     /*     #for I in PARTICLES_ELEC_I */
 262     iq{I}              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+{I}]));
 263     /*     #endfor */
 264     /*     #for I in PARTICLES_VDW_I */
 265     vdwioffsetptr{I}   = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
 266     /*         #if 'LJEwald' in KERNEL_VDW */
 267     vdwgridioffsetptr{I} = vdwgridparam+2*nvdwtype*vdwtype[inr+{I}];
 268     /*         #endif */
 269     /*     #endfor */
 270     /* #endif */
 271
 272     /* #if 'Water' in GEOMETRY_J */
 273     /*     #for J in PARTICLES_ELEC_J */
 274     jq{J}              = _mm256_set1_ps(charge[inr+{J}]);
 275     /*     #endfor */
 276     /*     #for J in PARTICLES_VDW_J */
 277     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 278     /*     #endfor */
 279     /*     #for I,J in PAIRS_IJ */
 280     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 281     qq{I}{J}             = _mm256_mul_ps(iq{I},jq{J});
 282     /*         #endif */
 283     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 284         /*         #if 'LJEwald' in KERNEL_VDW */
 285     c6_{I}{J}            = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A]);
 286     c12_{I}{J}           = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A+1]);
 287     c6grid_{I}{J}        = _mm256_set1_ps(vdwgridioffsetptr{I}[vdwjidx{J}A]);
 288         /*         #else */
 289     c6_{I}{J}            = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A]);
 290     c12_{I}{J}           = _mm256_set1_ps(vdwioffsetptr{I}[vdwjidx{J}A+1]);
 291         /*         #endif */
 292     /*         #endif */
 293     /*     #endfor */
 294     /* #endif */
 295
 296     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 297     /*     #if KERNEL_ELEC!='None' */
 298     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 299     rcutoff_scalar   = fr->rcoulomb;
 300     /*     #else */
 301     rcutoff_scalar   = fr->rvdw;
 302     /*     #endif */
 303     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 304     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 305     /* #endif */
 306
 307     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 308     sh_vdw_invrcut6  = _mm256_set1_ps(fr->ic->sh_invrc6);
 309     rvdw             = _mm256_set1_ps(fr->rvdw);
 310     /* #endif */
 311
 312     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 313     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 314     rswitch_scalar   = fr->rcoulomb_switch;
 315     rswitch          = _mm256_set1_ps(rswitch_scalar);
 316     /*     #else */
 317     rswitch_scalar   = fr->rvdw_switch;
 318     rswitch          = _mm256_set1_ps(rswitch_scalar);
 319     /*     #endif */
 320     /* Setup switch parameters */
 321     d_scalar         = rcutoff_scalar-rswitch_scalar;
 322     d                = _mm256_set1_ps(d_scalar);
 323     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 324     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 325     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 326     /*     #if 'Force' in KERNEL_VF */
 327     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 328     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 329     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 330     /*     #endif */
 331     /* #endif */
 332
 333     /* Avoid stupid compiler warnings */
 334     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 335     j_coord_offsetA = 0;
 336     j_coord_offsetB = 0;
 337     j_coord_offsetC = 0;
 338     j_coord_offsetD = 0;
 339     j_coord_offsetE = 0;
 340     j_coord_offsetF = 0;
 341     j_coord_offsetG = 0;
 342     j_coord_offsetH = 0;
 343
 344     /* ## Keep track of the floating point operations we issue for reporting! */
 345     /* #define OUTERFLOPS 0 */
 346     outeriter        = 0;
 347     inneriter        = 0;
 348
 349     for(iidx=0;iidx<4*DIM;iidx++)
 350     {
 351         scratch[iidx] = 0.0;
 352     }
 353
 354     /* Start outer loop over neighborlists */
 355     for(iidx=0; iidx<nri; iidx++)
 356     {
 357         /* Load shift vector for this list */
 358         i_shift_offset   = DIM*shiftidx[iidx];
 359
 360         /* Load limits for loop over neighbors */
 361         j_index_start    = jindex[iidx];
 362         j_index_end      = jindex[iidx+1];
 363
 364         /* Get outer coordinate index */
 365         inr              = iinr[iidx];
 366         i_coord_offset   = DIM*inr;
 367
 368         /* Load i particle coords and add shift vector */
 369         /* #if GEOMETRY_I == 'Particle' */
 370         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 371         /* #elif GEOMETRY_I == 'Water3' */
 372         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 373                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 374         /* #elif GEOMETRY_I == 'Water4' */
 375         /*     #if 0 in PARTICLES_I                 */
 376         gmx_mm256_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 377                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 378         /*     #else                                */
 379         gmx_mm256_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 380                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 381         /*     #endif                               */
 382         /* #endif                                   */
 383
 384         /* #if 'Force' in KERNEL_VF */
 385         /*     #for I in PARTICLES_I */
 386         fix{I}             = _mm256_setzero_ps();
 387         fiy{I}             = _mm256_setzero_ps();
 388         fiz{I}             = _mm256_setzero_ps();
 389         /*     #endfor */
 390         /* #endif */
 391
 392         /* ## For water we already preloaded parameters at the start of the kernel */
 393         /* #if not 'Water' in GEOMETRY_I */
 394         /* Load parameters for i particles */
 395         /*     #for I in PARTICLES_ELEC_I */
 396         iq{I}              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+{I}]));
 397         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 398         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 399         isai{I}            = _mm256_set1_ps(invsqrta[inr+{I}]);
 400         /*         #endif */
 401         /*     #endfor */
 402         /*     #for I in PARTICLES_VDW_I */
 403         vdwioffsetptr{I}   = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
 404         /*         #if 'LJEwald' in KERNEL_VDW */
 405         vdwgridioffsetptr{I} = vdwgridparam+2*nvdwtype*vdwtype[inr+{I}];
 406         /*         #endif */
 407         /*     #endfor */
 408         /* #endif */
 409
 410         /* #if 'Potential' in KERNEL_VF */
 411         /* Reset potential sums */
 412         /*     #if KERNEL_ELEC != 'None' */
 413         velecsum         = _mm256_setzero_ps();
 414         /*     #endif */
 415         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 416         vgbsum           = _mm256_setzero_ps();
 417         /*     #endif */
 418         /*     #if KERNEL_VDW != 'None' */
 419         vvdwsum          = _mm256_setzero_ps();
 420         /*     #endif */
 421         /* #endif */
 422         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 423         dvdasum          = _mm256_setzero_ps();
 424         /*     #endif */
 425
 426         /* #for ROUND in ['Loop','Epilogue'] */
 427
 428         /* #if ROUND =='Loop' */
 429         /* Start inner kernel loop */
 430         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 431         {
 432         /* ## First round is normal loop (next statement resets indentation) */
 433         /*     #if 0 */
 434         }
 435         /*     #endif */
 436         /* #else */
 437         if(jidx<j_index_end)
 438         {
 439         /* ## Second round is epilogue */
 440         /* #endif */
 441         /* #define INNERFLOPS 0 */
 442
 443             /* Get j neighbor index, and coordinate index */
 444             /* #if ROUND =='Loop' */
 445             jnrA             = jjnr[jidx];
 446             jnrB             = jjnr[jidx+1];
 447             jnrC             = jjnr[jidx+2];
 448             jnrD             = jjnr[jidx+3];
 449             jnrE             = jjnr[jidx+4];
 450             jnrF             = jjnr[jidx+5];
 451             jnrG             = jjnr[jidx+6];
 452             jnrH             = jjnr[jidx+7];
 453             /* #else */
 454             jnrlistA         = jjnr[jidx];
 455             jnrlistB         = jjnr[jidx+1];
 456             jnrlistC         = jjnr[jidx+2];
 457             jnrlistD         = jjnr[jidx+3];
 458             jnrlistE         = jjnr[jidx+4];
 459             jnrlistF         = jjnr[jidx+5];
 460             jnrlistG         = jjnr[jidx+6];
 461             jnrlistH         = jjnr[jidx+7];
 462             /* Sign of each element will be negative for non-real atoms.
 463              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 464              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 465              */
 466             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 467                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 468
 469             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 470             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 471             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 472             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 473             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 474             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 475             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 476             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 477             /* #endif */
 478             j_coord_offsetA  = DIM*jnrA;
 479             j_coord_offsetB  = DIM*jnrB;
 480             j_coord_offsetC  = DIM*jnrC;
 481             j_coord_offsetD  = DIM*jnrD;
 482             j_coord_offsetE  = DIM*jnrE;
 483             j_coord_offsetF  = DIM*jnrF;
 484             j_coord_offsetG  = DIM*jnrG;
 485             j_coord_offsetH  = DIM*jnrH;
 486
 487             /* load j atom coordinates */
 488             /* #if GEOMETRY_J == 'Particle'             */
 489             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 490                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 491                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 492                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 493                                                  &jx0,&jy0,&jz0);
 494             /* #elif GEOMETRY_J == 'Water3'             */
 495             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 496                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 497                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 498                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 499                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 500             /* #elif GEOMETRY_J == 'Water4'             */
 501             /*     #if 0 in PARTICLES_J                 */
 502             gmx_mm256_load_4rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 503                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 504                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 505                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 506                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 507                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 508             /*     #else                                */
 509             gmx_mm256_load_3rvec_8ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 510                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 511                                                  x+j_coord_offsetE+DIM,x+j_coord_offsetF+DIM,
 512                                                  x+j_coord_offsetG+DIM,x+j_coord_offsetH+DIM,
 513                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 514             /*     #endif                               */
 515             /* #endif                                   */
 516
 517             /* Calculate displacement vector */
 518             /* #for I,J in PAIRS_IJ */
 519             dx{I}{J}             = _mm256_sub_ps(ix{I},jx{J});
 520             dy{I}{J}             = _mm256_sub_ps(iy{I},jy{J});
 521             dz{I}{J}             = _mm256_sub_ps(iz{I},jz{J});
 522             /*     #define INNERFLOPS INNERFLOPS+3 */
 523             /* #endfor */
 524
 525             /* Calculate squared distance and things based on it */
 526             /* #for I,J in PAIRS_IJ */
 527             rsq{I}{J}            = gmx_mm256_calc_rsq_ps(dx{I}{J},dy{I}{J},dz{I}{J});
 528             /*     #define INNERFLOPS INNERFLOPS+5 */
 529             /* #endfor */
 530
 531             /* #for I,J in PAIRS_IJ */
 532             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 533             rinv{I}{J}           = gmx_mm256_invsqrt_ps(rsq{I}{J});
 534             /*         #define INNERFLOPS INNERFLOPS+5 */
 535             /*     #endif */
 536             /* #endfor */
 537
 538             /* #for I,J in PAIRS_IJ */
 539             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 540             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 541             rinvsq{I}{J}         = gmx_mm256_inv_ps(rsq{I}{J});
 542             /*             #define INNERFLOPS INNERFLOPS+4 */
 543             /*         #else */
 544             rinvsq{I}{J}         = _mm256_mul_ps(rinv{I}{J},rinv{I}{J});
 545             /*             #define INNERFLOPS INNERFLOPS+1 */
 546             /*         #endif */
 547             /*     #endif */
 548             /* #endfor */
 549
 550             /* #if not 'Water' in GEOMETRY_J */
 551             /* Load parameters for j particles */
 552             /*     #for J in PARTICLES_ELEC_J */
 553             jq{J}              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+{J},charge+jnrB+{J},
 554                                                                  charge+jnrC+{J},charge+jnrD+{J},
 555                                                                  charge+jnrE+{J},charge+jnrF+{J},
 556                                                                  charge+jnrG+{J},charge+jnrH+{J});
 557             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 558             isaj{J}            = gmx_mm256_load_8real_swizzle_ps(invsqrta+jnrA+{J},invsqrta+jnrB+{J},
 559                                                                  invsqrta+jnrC+{J},invsqrta+jnrD+{J},
 560                                                                  invsqrta+jnrE+{J},invsqrta+jnrF+{J},
 561                                                                  invsqrta+jnrG+{J},invsqrta+jnrH+{J});
 562             /*         #endif */
 563             /*     #endfor */
 564             /*     #for J in PARTICLES_VDW_J */
 565             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 566             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 567             vdwjidx{J}C        = 2*vdwtype[jnrC+{J}];
 568             vdwjidx{J}D        = 2*vdwtype[jnrD+{J}];
 569             vdwjidx{J}E        = 2*vdwtype[jnrE+{J}];
 570             vdwjidx{J}F        = 2*vdwtype[jnrF+{J}];
 571             vdwjidx{J}G        = 2*vdwtype[jnrG+{J}];
 572             vdwjidx{J}H        = 2*vdwtype[jnrH+{J}];
 573             /*     #endfor */
 574             /* #endif */
 575
 576             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 577             /*     #for J in PARTICLES_J */
 578             fjx{J}             = _mm256_setzero_ps();
 579             fjy{J}             = _mm256_setzero_ps();
 580             fjz{J}             = _mm256_setzero_ps();
 581             /*     #endfor */
 582             /* #endif */
 583
 584             /* #for I,J in PAIRS_IJ */
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 591             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 592             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 593             if (gmx_mm256_any_lt(rsq{I}{J},rcutoff2))
 594             {
 595                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 596             }
 597             /*         #endif */
 598             /*         #define INNERFLOPS INNERFLOPS+1 */
 599             /*     #endif */
 600
 601             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 602             r{I}{J}              = _mm256_mul_ps(rsq{I}{J},rinv{I}{J});
 603             /*         #if ROUND == 'Epilogue' */
 604             r{I}{J}              = _mm256_andnot_ps(dummy_mask,r{I}{J});
 605             /*             #define INNERFLOPS INNERFLOPS+1 */
 606             /*         #endif */
 607             /*         #define INNERFLOPS INNERFLOPS+1 */
 608             /*     #endif */
 609
 610             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 611             /*     #if not 'Water' in GEOMETRY_J */
 612             /* Compute parameters for interactions between i and j atoms */
 613             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 614             qq{I}{J}             = _mm256_mul_ps(iq{I},jq{J});
 615             /*             #define INNERFLOPS INNERFLOPS+1 */
 616             /*         #endif */
 617             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 618             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr{I}+vdwjidx{J}A,
 619                                             vdwioffsetptr{I}+vdwjidx{J}B,
 620                                             vdwioffsetptr{I}+vdwjidx{J}C,
 621                                             vdwioffsetptr{I}+vdwjidx{J}D,
 622                                             vdwioffsetptr{I}+vdwjidx{J}E,
 623                                             vdwioffsetptr{I}+vdwjidx{J}F,
 624                                             vdwioffsetptr{I}+vdwjidx{J}G,
 625                                             vdwioffsetptr{I}+vdwjidx{J}H,
 626                                             &c6_{I}{J},&c12_{I}{J});
 627
 628             /*         #if 'LJEwald' in KERNEL_VDW */
 629             c6grid_{I}{J}       = gmx_mm256_load_8real_swizzle_ps(vdwgridioffsetptr{I}+vdwjidx{J}A,
 630                                                                   vdwgridioffsetptr{I}+vdwjidx{J}B,
 631                                                                   vdwgridioffsetptr{I}+vdwjidx{J}C,
 632                                                                   vdwgridioffsetptr{I}+vdwjidx{J}D,
 633                                                                   vdwgridioffsetptr{I}+vdwjidx{J}E,
 634                                                                   vdwgridioffsetptr{I}+vdwjidx{J}F,
 635                                                                   vdwgridioffsetptr{I}+vdwjidx{J}G,
 636                                                                   vdwgridioffsetptr{I}+vdwjidx{J}H);
 637             /*          #endif */
 638             /*         #endif */
 639             /*     #endif */
 640
 641             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 642             /* Calculate table index by multiplying r with table scale and truncate to integer */
 643             rt               = _mm256_mul_ps(r{I}{J},vftabscale);
 644             vfitab           = _mm256_cvttps_epi32(rt);
 645             vfeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 646             /*         #define INNERFLOPS INNERFLOPS+4                          */
 647             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 648             vfitab_lo        = _mm256_extractf128_si256(vfitab,0x0);
 649             vfitab_hi        = _mm256_extractf128_si256(vfitab,0x1);
 650             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 651             /*             ## 3 tables, 4 bytes per point: multiply index by 12 */
 652             vfitab_lo        = _mm_slli_epi32(_mm_add_epi32(vfitab_lo,_mm_slli_epi32(vfitab_lo,1)),2);
 653             vfitab_hi        = _mm_slli_epi32(_mm_add_epi32(vfitab_hi,_mm_slli_epi32(vfitab_hi,1)),2);
 654             /*         #elif 'Table' in KERNEL_ELEC                             */
 655             /*             ## 1 table, 4 bytes per point: multiply index by 4   */
 656             vfitab_lo        = _mm_slli_epi32(vfitab_lo,2);
 657             vfitab_hi        = _mm_slli_epi32(vfitab_hi,2);
 658             /*         #elif 'Table' in KERNEL_VDW */
 659             /*             ## 2 tables, 4 bytes per point: multiply index by 8  */
 660             vfitab_lo        = _mm_slli_epi32(vfitab_lo,3);
 661             vfitab_hi        = _mm_slli_epi32(vfitab_hi,3);
 662             /*         #endif                                                   */
 663             /*     #endif */
 664
 665             /*     ## ELECTROSTATIC INTERACTIONS */
 666             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 667
 668             /*         #if KERNEL_ELEC=='Coulomb' */
 669
 670             /* COULOMB ELECTROSTATICS */
 671             velec            = _mm256_mul_ps(qq{I}{J},rinv{I}{J});
 672             /*             #define INNERFLOPS INNERFLOPS+1 */
 673             /*             #if 'Force' in KERNEL_VF */
 674             felec            = _mm256_mul_ps(velec,rinvsq{I}{J});
 675             /*                 #define INNERFLOPS INNERFLOPS+1 */
 676             /*             #endif */
 677
 678             /*         #elif KERNEL_ELEC=='ReactionField' */
 679
 680             /* REACTION-FIELD ELECTROSTATICS */
 681             /*             #if 'Potential' in KERNEL_VF */
 682             velec            = _mm256_mul_ps(qq{I}{J},_mm256_sub_ps(_mm256_add_ps(rinv{I}{J},_mm256_mul_ps(krf,rsq{I}{J})),crf));
 683             /*                 #define INNERFLOPS INNERFLOPS+4 */
 684             /*             #endif */
 685             /*             #if 'Force' in KERNEL_VF */
 686             felec            = _mm256_mul_ps(qq{I}{J},_mm256_sub_ps(_mm256_mul_ps(rinv{I}{J},rinvsq{I}{J}),krf2));
 687             /*                 #define INNERFLOPS INNERFLOPS+3 */
 688             /*             #endif */
 689
 690             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 691
 692             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 693             isaprod          = _mm256_mul_ps(isai{I},isaj{J});
 694             gbqqfactor       = _mm256_xor_ps(signbit,_mm256_mul_ps(qq{I}{J},_mm256_mul_ps(isaprod,gbinvepsdiff)));
 695             gbscale          = _mm256_mul_ps(isaprod,gbtabscale);
 696             /*             #define INNERFLOPS INNERFLOPS+5 */
 697
 698             /* Calculate generalized born table index - this is a separate table from the normal one,
 699              * but we use the same procedure by multiplying r with scale and truncating to integer.
 700              */
 701             rt               = _mm256_mul_ps(r{I}{J},gbscale);
 702             gbitab           = _mm256_cvttps_epi32(rt);
 703             gbeps            = _mm256_sub_ps(rt,_mm256_round_ps(rt, _MM_FROUND_FLOOR));
 704             /*         AVX1 does not support 256-bit integer operations, so now we go to 128-bit mode... */
 705             gbitab_lo        = _mm256_extractf128_si256(gbitab,0x0);
 706             gbitab_hi        = _mm256_extractf128_si256(gbitab,0x1);
 707             gbitab_lo        = _mm_slli_epi32(gbitab_lo,2);
 708             gbitab_hi        = _mm_slli_epi32(gbitab_hi,2);
 709             Y                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,0)),
 710                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,0)));
 711             F                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,1)),
 712                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,1)));
 713             G                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,2)),
 714                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,2)));
 715             H                = gmx_mm256_set_m128(_mm_load_ps(gbtab + _mm_extract_epi32(gbitab_hi,3)),
 716                                                   _mm_load_ps(gbtab + _mm_extract_epi32(gbitab_lo,3)));
 717             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 718             Heps             = _mm256_mul_ps(gbeps,H);
 719             Fp               = _mm256_add_ps(F,_mm256_mul_ps(gbeps,_mm256_add_ps(G,Heps)));
 720             VV               = _mm256_add_ps(Y,_mm256_mul_ps(gbeps,Fp));
 721             vgb              = _mm256_mul_ps(gbqqfactor,VV);
 722             /*             #define INNERFLOPS INNERFLOPS+10 */
 723
 724             /*             #if 'Force' in KERNEL_VF */
 725             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(gbeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 726             fgb              = _mm256_mul_ps(gbqqfactor,_mm256_mul_ps(FF,gbscale));
 727             dvdatmp          = _mm256_mul_ps(minushalf,_mm256_add_ps(vgb,_mm256_mul_ps(fgb,r{I}{J})));
 728             /*                 #if ROUND == 'Epilogue' */
 729             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 730             /*                 #endif */
 731             dvdasum          = _mm256_add_ps(dvdasum,dvdatmp);
 732             /*                 #if ROUND == 'Loop' */
 733             fjptrA           = dvda+jnrA;
 734             fjptrB           = dvda+jnrB;
 735             fjptrC           = dvda+jnrC;
 736             fjptrD           = dvda+jnrD;
 737             fjptrE           = dvda+jnrE;
 738             fjptrF           = dvda+jnrF;
 739             fjptrG           = dvda+jnrG;
 740             fjptrH           = dvda+jnrH;
 741             /*                 #else */
 742             /* 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. */
 743             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 744             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 745             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 746             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 747             fjptrE             = (jnrlistE>=0) ? dvda+jnrE : scratch;
 748             fjptrF             = (jnrlistF>=0) ? dvda+jnrF : scratch;
 749             fjptrG             = (jnrlistG>=0) ? dvda+jnrG : scratch;
 750             fjptrH             = (jnrlistH>=0) ? dvda+jnrH : scratch;
 751             /*                 #endif */
 752             gmx_mm256_increment_8real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
 753                                                  _mm256_mul_ps(dvdatmp,_mm256_mul_ps(isaj{J},isaj{J})));
 754             /*                 #define INNERFLOPS INNERFLOPS+12 */
 755             /*             #endif */
 756             velec            = _mm256_mul_ps(qq{I}{J},rinv{I}{J});
 757             /*                 #define INNERFLOPS INNERFLOPS+1 */
 758             /*             #if 'Force' in KERNEL_VF */
 759             felec            = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(velec,rinv{I}{J}),fgb),rinv{I}{J});
 760             /*                 #define INNERFLOPS INNERFLOPS+3 */
 761             /*             #endif */
 762
 763             /*         #elif KERNEL_ELEC=='Ewald' */
 764             /* EWALD ELECTROSTATICS */
 765
 766             /* Analytical PME correction */
 767             zeta2            = _mm256_mul_ps(beta2,rsq{I}{J});
 768             /*             #if 'Force' in KERNEL_VF */
 769             rinv3            = _mm256_mul_ps(rinvsq{I}{J},rinv{I}{J});
 770             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 771             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 772             felec            = _mm256_mul_ps(qq{I}{J},felec);
 773             /*                 #define INNERFLOPS INNERFLOPS+31 */
 774             /*             #endif */
 775             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 776             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 777             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 778             /*                 #define INNERFLOPS INNERFLOPS+27 */
 779             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 780             velec            = _mm256_sub_ps(_mm256_sub_ps(rinv{I}{J},sh_ewald),pmecorrV);
 781             /*                     #define INNERFLOPS INNERFLOPS+21 */
 782             /*                 #else */
 783             velec            = _mm256_sub_ps(rinv{I}{J},pmecorrV);
 784             /*                 #endif */
 785             velec            = _mm256_mul_ps(qq{I}{J},velec);
 786             /*             #endif */
 787
 788             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 789
 790             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 791             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 792                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 793             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 794                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 795             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 796                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 797             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 798                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 799             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 800             Heps             = _mm256_mul_ps(vfeps,H);
 801             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 802             /*             #define INNERFLOPS INNERFLOPS+4 */
 803             /*             #if 'Potential' in KERNEL_VF */
 804             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 805             velec            = _mm256_mul_ps(qq{I}{J},VV);
 806             /*                 #define INNERFLOPS INNERFLOPS+3 */
 807             /*             #endif */
 808             /*             #if 'Force' in KERNEL_VF */
 809             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 810             felec            = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_mul_ps(qq{I}{J},FF),_mm256_mul_ps(vftabscale,rinv{I}{J})));
 811             /*                 #define INNERFLOPS INNERFLOPS+7 */
 812             /*             #endif */
 813             /*         #endif */
 814             /*         ## End of check for electrostatics interaction forms */
 815             /*     #endif */
 816             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 817
 818             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 819
 820             /*         #if KERNEL_VDW=='LennardJones' */
 821
 822             /* LENNARD-JONES DISPERSION/REPULSION */
 823
 824             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 825             /*             #define INNERFLOPS INNERFLOPS+2 */
 826             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 827             vvdw6            = _mm256_mul_ps(c6_{I}{J},rinvsix);
 828             vvdw12           = _mm256_mul_ps(c12_{I}{J},_mm256_mul_ps(rinvsix,rinvsix));
 829             /*                 #define INNERFLOPS INNERFLOPS+3 */
 830             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 831             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) ,
 832                                           _mm256_mul_ps( _mm256_sub_ps(vvdw6,_mm256_mul_ps(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
 833             /*                     #define INNERFLOPS INNERFLOPS+8 */
 834             /*                 #else */
 835             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 836             /*                     #define INNERFLOPS INNERFLOPS+3 */
 837             /*                 #endif */
 838             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 839             /*                 #if 'Force' in KERNEL_VF */
 840             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq{I}{J});
 841             /*                     #define INNERFLOPS INNERFLOPS+2 */
 842             /*                 #endif */
 843             /*             #elif KERNEL_VF=='Force' */
 844             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 845             fvdw             = _mm256_mul_ps(_mm256_sub_ps(_mm256_mul_ps(c12_{I}{J},rinvsix),c6_{I}{J}),_mm256_mul_ps(rinvsix,rinvsq{I}{J}));
 846             /*                 #define INNERFLOPS INNERFLOPS+4 */
 847             /*             #endif */
 848
 849             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 850
 851             /* CUBIC SPLINE TABLE DISPERSION */
 852             /*             #if 'Table' in KERNEL_ELEC */
 853             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 854             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 855             /*             #endif                     */
 856             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 857                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 858             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 859                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 860             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 861                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 862             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 863                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 864             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 865             Heps             = _mm256_mul_ps(vfeps,H);
 866             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 867             /*             #define INNERFLOPS INNERFLOPS+4 */
 868             /*             #if 'Potential' in KERNEL_VF */
 869             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 870             vvdw6            = _mm256_mul_ps(c6_{I}{J},VV);
 871             /*                 #define INNERFLOPS INNERFLOPS+3 */
 872             /*             #endif */
 873             /*             #if 'Force' in KERNEL_VF */
 874             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 875             fvdw6            = _mm256_mul_ps(c6_{I}{J},FF);
 876             /*                 #define INNERFLOPS INNERFLOPS+4 */
 877             /*             #endif */
 878
 879             /* CUBIC SPLINE TABLE REPULSION */
 880             vfitab_lo        = _mm_add_epi32(vfitab_lo,ifour);
 881             vfitab_hi        = _mm_add_epi32(vfitab_hi,ifour);
 882             Y                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,0)),
 883                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,0)));
 884             F                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,1)),
 885                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,1)));
 886             G                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,2)),
 887                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,2)));
 888             H                = gmx_mm256_set_m128(_mm_load_ps(vftab + _mm_extract_epi32(vfitab_hi,3)),
 889                                                   _mm_load_ps(vftab + _mm_extract_epi32(vfitab_lo,3)));
 890             GMX_MM256_HALFTRANSPOSE4_PS(Y,F,G,H);
 891             Heps             = _mm256_mul_ps(vfeps,H);
 892             Fp               = _mm256_add_ps(F,_mm256_mul_ps(vfeps,_mm256_add_ps(G,Heps)));
 893             /*             #define INNERFLOPS INNERFLOPS+4 */
 894             /*             #if 'Potential' in KERNEL_VF */
 895             VV               = _mm256_add_ps(Y,_mm256_mul_ps(vfeps,Fp));
 896             vvdw12           = _mm256_mul_ps(c12_{I}{J},VV);
 897             /*                 #define INNERFLOPS INNERFLOPS+3 */
 898             /*             #endif */
 899             /*             #if 'Force' in KERNEL_VF */
 900             FF               = _mm256_add_ps(Fp,_mm256_mul_ps(vfeps,_mm256_add_ps(G,_mm256_add_ps(Heps,Heps))));
 901             fvdw12           = _mm256_mul_ps(c12_{I}{J},FF);
 902             /*                 #define INNERFLOPS INNERFLOPS+5 */
 903             /*             #endif */
 904             /*             #if 'Potential' in KERNEL_VF */
 905             vvdw             = _mm256_add_ps(vvdw12,vvdw6);
 906             /*                 #define INNERFLOPS INNERFLOPS+1 */
 907             /*             #endif */
 908             /*             #if 'Force' in KERNEL_VF */
 909             fvdw             = _mm256_xor_ps(signbit,_mm256_mul_ps(_mm256_add_ps(fvdw6,fvdw12),_mm256_mul_ps(vftabscale,rinv{I}{J})));
 910             /*                 #define INNERFLOPS INNERFLOPS+4 */
 911             /*             #endif */
 912
 913             /*         #elif KERNEL_VDW=='LJEwald' */
 914
 915             /* Analytical LJ-PME */
 916             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 917             ewcljrsq         = _mm256_mul_ps(ewclj2,rsq{I}{J});
 918             ewclj6           = _mm256_mul_ps(ewclj2,_mm256_mul_ps(ewclj2,ewclj2));
 919             exponent         = gmx_simd_exp_r(ewcljrsq);
 920             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 921             poly             = _mm256_mul_ps(exponent,_mm256_add_ps(_mm256_sub_ps(one,ewcljrsq),_mm256_mul_ps(_mm256_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 922             /*                 #define INNERFLOPS INNERFLOPS+11 */
 923             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 924             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 925             vvdw6            = _mm256_mul_ps(_mm256_sub_ps(c6_{I}{J},_mm256_mul_ps(c6grid_{I}{J},_mm256_sub_ps(one,poly))),rinvsix);
 926             vvdw12           = _mm256_mul_ps(c12_{I}{J},_mm256_mul_ps(rinvsix,rinvsix));
 927             /*                 #define INNERFLOPS INNERFLOPS+6 */
 928             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 929             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) ,
 930                                           _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));
 931             /*                     #define INNERFLOPS INNERFLOPS+10 */
 932             /*                 #else */
 933             vvdw             = _mm256_sub_ps(_mm256_mul_ps(vvdw12,one_twelfth),_mm256_mul_ps(vvdw6,one_sixth));
 934             /*                 #define INNERFLOPS INNERFLOPS+3 */
 935             /*                 #endif */
 936             /*                  ## Check for force inside potential check, i.e. this means we already did the potential part */
 937             /*                  #if 'Force' in KERNEL_VF */
 938             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 939             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});
 940             /*                 #define INNERFLOPS INNERFLOPS+6 */
 941             /*                  #endif */
 942             /*              #elif KERNEL_VF=='Force' */
 943             /* f6A = 6 * C6grid * (1 - poly) */
 944             f6A              = _mm256_mul_ps(c6grid_{I}{J},_mm256_sub_ps(one,poly));
 945             /* f6B = C6grid * exponent * beta^6 */
 946             f6B              = _mm256_mul_ps(_mm256_mul_ps(c6grid_{I}{J},one_sixth),_mm256_mul_ps(exponent,ewclj6));
 947             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 948             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});
 949             /*                 #define INNERFLOPS INNERFLOPS+11 */
 950             /*              #endif */
 951             /*         #endif */
 952             /*         ## End of check for vdw interaction forms */
 953             /*     #endif */
 954             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 955
 956             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 957             d                = _mm256_sub_ps(r{I}{J},rswitch);
 958             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 959             d2               = _mm256_mul_ps(d,d);
 960             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)))))));
 961             /*         #define INNERFLOPS INNERFLOPS+10 */
 962
 963             /*         #if 'Force' in KERNEL_VF */
 964             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 965             /*             #define INNERFLOPS INNERFLOPS+5 */
 966             /*         #endif */
 967
 968             /* Evaluate switch function */
 969             /*         #if 'Force' in KERNEL_VF */
 970             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 971             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 972             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv{I}{J},_mm256_mul_ps(velec,dsw)) );
 973             /*                 #define INNERFLOPS INNERFLOPS+4 */
 974             /*             #endif */
 975             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 976             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv{I}{J},_mm256_mul_ps(vvdw,dsw)) );
 977             /*                 #define INNERFLOPS INNERFLOPS+4 */
 978             /*             #endif */
 979             /*         #endif */
 980             /*         #if 'Potential' in KERNEL_VF */
 981             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 982             velec            = _mm256_mul_ps(velec,sw);
 983             /*                 #define INNERFLOPS INNERFLOPS+1 */
 984             /*             #endif */
 985             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 986             vvdw             = _mm256_mul_ps(vvdw,sw);
 987             /*                 #define INNERFLOPS INNERFLOPS+1 */
 988             /*             #endif */
 989             /*         #endif */
 990             /*     #endif */
 991             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 992             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 993             cutoff_mask      = _mm256_cmp_ps(rsq{I}{J},rcutoff2,_CMP_LT_OQ);
 994             /*         #define INNERFLOPS INNERFLOPS+1 */
 995             /*     #endif */
 996
 997             /*     #if 'Potential' in KERNEL_VF */
 998             /* Update potential sum for this i atom from the interaction with this j atom. */
 999             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
1000             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
1001             velec            = _mm256_and_ps(velec,cutoff_mask);
1002             /*                 #define INNERFLOPS INNERFLOPS+1 */
1003             /*             #endif                                       */
1004             /*             #if ROUND == 'Epilogue' */
1005             velec            = _mm256_andnot_ps(dummy_mask,velec);
1006             /*             #endif */
1007             velecsum         = _mm256_add_ps(velecsum,velec);
1008             /*             #define INNERFLOPS INNERFLOPS+1 */
1009             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
1010             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
1011             vgb              = _mm256_and_ps(vgb,cutoff_mask);
1012             /*                 #define INNERFLOPS INNERFLOPS+1 */
1013             /*             #endif                                       */
1014             /*             #if ROUND == 'Epilogue' */
1015             vgb              = _mm256_andnot_ps(dummy_mask,vgb);
1016             /*             #endif */
1017             vgbsum           = _mm256_add_ps(vgbsum,vgb);
1018             /*                 #define INNERFLOPS INNERFLOPS+1 */
1019             /*             #endif */
1020             /*         #endif */
1021             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
1022             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
1023             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
1024             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
1025             /*                 #define INNERFLOPS INNERFLOPS+1 */
1026             /*             #endif                                       */
1027             /*             #if ROUND == 'Epilogue' */
1028             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
1029             /*             #endif */
1030             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
1031             /*             #define INNERFLOPS INNERFLOPS+1 */
1032             /*         #endif */
1033             /*     #endif */
1034
1035             /*     #if 'Force' in KERNEL_VF */
1036
1037             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
1038             fscal            = _mm256_add_ps(felec,fvdw);
1039             /*             #define INNERFLOPS INNERFLOPS+1 */
1040             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
1041             fscal            = felec;
1042             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
1043             fscal            = fvdw;
1044             /*        #endif */
1045
1046             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
1047             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
1048             fscal            = _mm256_and_ps(fscal,cutoff_mask);
1049             /*                 #define INNERFLOPS INNERFLOPS+1 */
1050             /*             #endif                                       */
1051
1052             /*             #if ROUND == 'Epilogue' */
1053             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
1054             /*             #endif */
1055
1056             /* Calculate temporary vectorial force */
1057             tx               = _mm256_mul_ps(fscal,dx{I}{J});
1058             ty               = _mm256_mul_ps(fscal,dy{I}{J});
1059             tz               = _mm256_mul_ps(fscal,dz{I}{J});
1060
1061             /* Update vectorial force */
1062             fix{I}             = _mm256_add_ps(fix{I},tx);
1063             fiy{I}             = _mm256_add_ps(fiy{I},ty);
1064             fiz{I}             = _mm256_add_ps(fiz{I},tz);
1065             /*             #define INNERFLOPS INNERFLOPS+6 */
1066
1067             /* #if GEOMETRY_I == 'Particle'             */
1068             /*     #if ROUND == 'Loop' */
1069             fjptrA             = f+j_coord_offsetA;
1070             fjptrB             = f+j_coord_offsetB;
1071             fjptrC             = f+j_coord_offsetC;
1072             fjptrD             = f+j_coord_offsetD;
1073             fjptrE             = f+j_coord_offsetE;
1074             fjptrF             = f+j_coord_offsetF;
1075             fjptrG             = f+j_coord_offsetG;
1076             fjptrH             = f+j_coord_offsetH;
1077             /*     #else */
1078             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1079             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1080             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1081             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1082             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1083             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1084             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1085             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1086             /*     #endif */
1087             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
1088             /*     #define INNERFLOPS INNERFLOPS+3      */
1089             /* #else                                    */
1090             fjx{J}             = _mm256_add_ps(fjx{J},tx);
1091             fjy{J}             = _mm256_add_ps(fjy{J},ty);
1092             fjz{J}             = _mm256_add_ps(fjz{J},tz);
1093             /*     #define INNERFLOPS INNERFLOPS+3      */
1094             /* #endif                                   */
1095
1096             /*     #endif */
1097
1098             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
1099             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
1100             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
1101             {
1102                 /*     #endif */
1103             }
1104             /*     #endif */
1105             /*    ## End of check for the interaction being outside the cutoff */
1106
1107             /* #endfor */
1108             /* ## End of loop over i-j interaction pairs */
1109
1110             /* #if GEOMETRY_I != 'Particle' */
1111             /*     #if ROUND == 'Loop' */
1112             fjptrA             = f+j_coord_offsetA;
1113             fjptrB             = f+j_coord_offsetB;
1114             fjptrC             = f+j_coord_offsetC;
1115             fjptrD             = f+j_coord_offsetD;
1116             fjptrE             = f+j_coord_offsetE;
1117             fjptrF             = f+j_coord_offsetF;
1118             fjptrG             = f+j_coord_offsetG;
1119             fjptrH             = f+j_coord_offsetH;
1120             /*     #else */
1121             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1122             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1123             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1124             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1125             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
1126             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
1127             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
1128             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
1129             /*     #endif */
1130             /* #endif */
1131
1132             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
1133             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,fjx0,fjy0,fjz0);
1134             /*     #define INNERFLOPS INNERFLOPS+3      */
1135             /* #elif GEOMETRY_J == 'Water3'             */
1136             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1137                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1138             /*     #define INNERFLOPS INNERFLOPS+9      */
1139             /* #elif GEOMETRY_J == 'Water4'             */
1140             /*     #if 0 in PARTICLES_J                 */
1141             gmx_mm256_decrement_4rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,
1142                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1143                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1144             /*     #define INNERFLOPS INNERFLOPS+12     */
1145             /*     #else                                */
1146             gmx_mm256_decrement_3rvec_8ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
1147                                                       fjptrE+DIM,fjptrF+DIM,fjptrG+DIM,fjptrH+DIM,
1148                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1149             /*     #define INNERFLOPS INNERFLOPS+9      */
1150             /*     #endif                               */
1151             /* #endif                                   */
1152
1153             /* Inner loop uses {INNERFLOPS} flops */
1154         }
1155
1156         /* #endfor */
1157
1158         /* End of innermost loop */
1159
1160         /* #if 'Force' in KERNEL_VF */
1161         /*     #if GEOMETRY_I == 'Particle'            */
1162         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
1163                                                  f+i_coord_offset,fshift+i_shift_offset);
1164         /*         #define OUTERFLOPS OUTERFLOPS+6     */
1165         /*     #elif GEOMETRY_I == 'Water3'            */
1166         gmx_mm256_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1167                                                  f+i_coord_offset,fshift+i_shift_offset);
1168         /*         #define OUTERFLOPS OUTERFLOPS+18    */
1169         /*     #elif GEOMETRY_I == 'Water4'            */
1170         /*         #if 0 in PARTICLES_I                */
1171         gmx_mm256_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1172                                                  f+i_coord_offset,fshift+i_shift_offset);
1173         /*             #define OUTERFLOPS OUTERFLOPS+24    */
1174         /*         #else                               */
1175         gmx_mm256_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1176                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1177         /*             #define OUTERFLOPS OUTERFLOPS+18    */
1178         /*         #endif                              */
1179         /*     #endif                                  */
1180         /* #endif                                      */
1181
1182         /* #if 'Potential' in KERNEL_VF */
1183         ggid                        = gid[iidx];
1184         /* Update potential energies */
1185         /*     #if KERNEL_ELEC != 'None' */
1186         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1187         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1188         /*     #endif */
1189         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
1190         gmx_mm256_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
1191         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1192         /*     #endif */
1193         /*     #if KERNEL_VDW != 'None' */
1194         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1195         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1196         /*     #endif */
1197         /* #endif */
1198         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1199         dvdasum = _mm256_mul_ps(dvdasum, _mm256_mul_ps(isai{I},isai{I}));
1200         gmx_mm256_update_1pot_ps(dvdasum,dvda+inr);
1201         /*     #endif */
1202
1203         /* Increment number of inner iterations */
1204         inneriter                  += j_index_end - j_index_start;
1205
1206         /* Outer loop uses {OUTERFLOPS} flops */
1207     }
1208
1209     /* Increment number of outer iterations */
1210     outeriter        += nri;
1211
1212     /* Update outer/inner flops */
1213     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1214     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1215     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1216     /* #if GEOMETRY_I == 'Water3'            */
1217     /*     #define ISUFFIX '_W3'             */
1218     /* #elif GEOMETRY_I == 'Water4'          */
1219     /*     #define ISUFFIX '_W4'             */
1220     /* #else                                 */
1221     /*     #define ISUFFIX ''                */
1222     /* #endif                                */
1223     /* #if GEOMETRY_J == 'Water3'            */
1224     /*     #define JSUFFIX 'W3'              */
1225     /* #elif GEOMETRY_J == 'Water4'          */
1226     /*     #define JSUFFIX 'W4'              */
1227     /* #else                                 */
1228     /*     #define JSUFFIX ''                */
1229     /* #endif                                */
1230     /* #if 'PotentialAndForce' in KERNEL_VF  */
1231     /*     #define VFSUFFIX  '_VF'           */
1232     /* #elif 'Potential' in KERNEL_VF        */
1233     /*     #define VFSUFFIX '_V'             */
1234     /* #else                                 */
1235     /*     #define VFSUFFIX '_F'             */
1236     /* #endif                                */
1237
1238     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1239     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1240     /* #elif KERNEL_ELEC != 'None' */
1241     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1242     /* #else */
1243     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1244     /* #endif  */
1245 }