src/gmxlib/nonbonded/nb_kernel_avx_256_double/nb_kernel_template_avx_256_double.pre

   1 /* ## */
   2 /* ## This file is part of the GROMACS molecular simulation package. */
   3 /* ## */
   4 /* ## Copyright (c) 2012, by the GROMACS development team, led by */
   5 /* ## David van der Spoel, Berk Hess, Erik Lindahl, and including many */
   6 /* ## others, as listed in the AUTHORS file in the top-level source */
   7 /* ## directory and at http://www.gromacs.org. */
   8 /* ## */
   9 /* ## GROMACS is free software; you can redistribute it and/or */
  10 /* ## modify it under the terms of the GNU Lesser General Public License */
  11 /* ## as published by the Free Software Foundation; either version 2.1 */
  12 /* ## of the License, or (at your option) any later version. */
  13 /* ## */
  14 /* ## GROMACS is distributed in the hope that it will be useful, */
  15 /* ## but WITHOUT ANY WARRANTY; without even the implied warranty of */
  16 /* ## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU */
  17 /* ## Lesser General Public License for more details. */
  18 /* ## */
  19 /* ## You should have received a copy of the GNU Lesser General Public */
  20 /* ## License along with GROMACS; if not, see */
  21 /* ## http://www.gnu.org/licenses, or write to the Free Software Foundation, */
  22 /* ## Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA. */
  23 /* ## */
  24 /* ## If you want to redistribute modifications to GROMACS, please */
  25 /* ## consider that scientific software is very special. Version */
  26 /* ## control is crucial - bugs must be traceable. We will be happy to */
  27 /* ## consider code for inclusion in the official distribution, but */
  28 /* ## derived work must not be called official GROMACS. Details are found */
  29 /* ## in the README & COPYING files - if they are missing, get the */
  30 /* ## official version at http://www.gromacs.org. */
  31 /* ## */
  32 /* ## To help us fund GROMACS development, we humbly ask that you cite */
  33 /* ## the research papers on the package. Check out http://www.gromacs.org. */
  34 /* ## */
  35 /* #if 0 */
  36 #error This file must be processed with the Gromacs pre-preprocessor
  37 /* #endif */
  38 /* #if INCLUDE_HEADER */
  39 #ifdef HAVE_CONFIG_H
  40 #include <config.h>
  41 #endif
  42
  43 #include <math.h>
  44
  45 #include "../nb_kernel.h"
  46 #include "types/simple.h"
  47 #include "vec.h"
  48 #include "nrnb.h"
  49
  50 #include "gmx_math_x86_avx_256_double.h"
  51 #include "kernelutil_x86_avx_256_double.h"
  52 /* #endif */
  53
  54 /* ## List of variables set by the generating script:                                    */
  55 /* ##                                                                                    */
  56 /* ## Setttings that apply to the entire kernel:                                         */
  57 /* ## KERNEL_ELEC:           String, choice for electrostatic interactions               */
  58 /* ## KERNEL_VDW:            String, choice for van der Waals interactions               */
  59 /* ## KERNEL_NAME:           String, name of this kernel                                 */
  60 /* ## KERNEL_VF:             String telling if we calculate potential, force, or both    */
  61 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
  62 /* ##                                                                                    */
  63 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops:             */
  64 /* ## PARTICLES_I[]/         Arrays with lists of i/j particles to use in kernel. It is  */
  65 /* ## PARTICLES_J[]:         just [0] for particle geometry, but can be longer for water */
  66 /* ## PARTICLES_ELEC_I[]/    Arrays with lists of i/j particle that have electrostatics  */
  67 /* ## PARTICLES_ELEC_J[]:    interactions that should be calculated in this kernel.      */
  68 /* ## PARTICLES_VDW_I[]/     Arrays with the list of i/j particle that have VdW          */
  69 /* ## PARTICLES_VDW_J[]:     interactions that should be calculated in this kernel.      */
  70 /* ##                                                                                    */
  71 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle)   */
  72 /* ## PAIRS_IJ[]:            Array with (i,j) tuples of pairs for which interactions     */
  73 /* ##                        should be calculated in this kernel. Zero-charge particles  */
  74 /* ##                        do not have interactions with particles without vdw, and    */
  75 /* ##                        Vdw-only interactions are not evaluated in a no-vdw-kernel. */
  76 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
  77 /* ##                        For each i-j pair, the element [I][J] is a list of strings  */
  78 /* ##                        defining properties/flags of this interaction. Examples     */
  79 /* ##                        include 'electrostatics'/'vdw' if that type of interaction  */
  80 /* ##                        should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values  */
  81 /* ##                        are needed, and 'exactcutoff' or 'shift','switch' to        */
  82 /* ##                        decide if the force/potential should be modified. This way  */
  83 /* ##                        we only calculate values absolutely needed for each case.   */
  84
  85 /* ## Calculate the size and offset for (merged/interleaved) table data */
  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                      t_forcerec * gmx_restrict              fr,
 100                      t_mdatoms * gmx_restrict               mdatoms,
 101                      nb_kernel_data_t * 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 refer to j loop unrolling done with AVX, e.g. for the four 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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 115     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 116     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 117     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 118     real             rcutoff_scalar;
 119     real             *shiftvec,*fshift,*x,*f;
 120     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 121     real             scratch[4*DIM];
 122     __m256d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 123     /* #for I in PARTICLES_I */
 124     real *           vdwioffsetptr{I};
 125     __m256d          ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 126     /* #endfor */
 127     /* #for J in PARTICLES_J */
 128     int              vdwjidx{J}A,vdwjidx{J}B,vdwjidx{J}C,vdwjidx{J}D;
 129     __m256d          jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 130     /* #endfor */
 131     /* #for I,J in PAIRS_IJ */
 132     __m256d          dx{I}{J},dy{I}{J},dz{I}{J},rsq{I}{J},rinv{I}{J},rinvsq{I}{J},r{I}{J},qq{I}{J},c6_{I}{J},c12_{I}{J};
 133     /* #endfor */
 134     /* #if KERNEL_ELEC != 'None' */
 135     __m256d          velec,felec,velecsum,facel,crf,krf,krf2;
 136     real             *charge;
 137     /* #endif */
 138     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 139     __m128i          gbitab;
 140     __m256d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,dvdatmp;
 141     __m256d          minushalf = _mm256_set1_pd(-0.5);
 142     real             *invsqrta,*dvda,*gbtab;
 143     /* #endif */
 144     /* #if KERNEL_VDW != 'None' */
 145     int              nvdwtype;
 146     __m256d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 147     int              *vdwtype;
 148     real             *vdwparam;
 149     __m256d          one_sixth   = _mm256_set1_pd(1.0/6.0);
 150     __m256d          one_twelfth = _mm256_set1_pd(1.0/12.0);
 151     /* #endif */
 152     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 153     __m128i          vfitab;
 154     __m128i          ifour       = _mm_set1_epi32(4);
 155     __m256d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 156     real             *vftab;
 157     /* #endif */
 158     /* #if 'Ewald' in KERNEL_ELEC */
 159     __m128i          ewitab;
 160     __m256d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 161     __m256d          beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 162     real             *ewtab;
 163     /* #endif */
 164     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 165     __m256d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 166     real             rswitch_scalar,d_scalar;
 167     /* #endif */
 168     __m256d          dummy_mask,cutoff_mask;
 169     __m128           tmpmask0,tmpmask1;
 170     __m256d          signbit = _mm256_castsi256_pd( _mm256_set1_epi32(0x80000000) );
 171     __m256d          one     = _mm256_set1_pd(1.0);
 172     __m256d          two     = _mm256_set1_pd(2.0);
 173     x                = xx[0];
 174     f                = ff[0];
 175
 176     nri              = nlist->nri;
 177     iinr             = nlist->iinr;
 178     jindex           = nlist->jindex;
 179     jjnr             = nlist->jjnr;
 180     shiftidx         = nlist->shift;
 181     gid              = nlist->gid;
 182     shiftvec         = fr->shift_vec[0];
 183     fshift           = fr->fshift[0];
 184     /* #if KERNEL_ELEC != 'None' */
 185     facel            = _mm256_set1_pd(fr->epsfac);
 186     charge           = mdatoms->chargeA;
 187     /*     #if 'ReactionField' in KERNEL_ELEC */
 188     krf              = _mm256_set1_pd(fr->ic->k_rf);
 189     krf2             = _mm256_set1_pd(fr->ic->k_rf*2.0);
 190     crf              = _mm256_set1_pd(fr->ic->c_rf);
 191     /*     #endif */
 192     /* #endif */
 193     /* #if KERNEL_VDW != 'None' */
 194     nvdwtype         = fr->ntype;
 195     vdwparam         = fr->nbfp;
 196     vdwtype          = mdatoms->typeA;
 197     /* #endif */
 198
 199     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 200     vftab            = kernel_data->table_elec_vdw->data;
 201     vftabscale       = _mm256_set1_pd(kernel_data->table_elec_vdw->scale);
 202     /* #elif 'Table' in KERNEL_ELEC */
 203     vftab            = kernel_data->table_elec->data;
 204     vftabscale       = _mm256_set1_pd(kernel_data->table_elec->scale);
 205     /* #elif 'Table' in KERNEL_VDW */
 206     vftab            = kernel_data->table_vdw->data;
 207     vftabscale       = _mm256_set1_pd(kernel_data->table_vdw->scale);
 208     /* #endif */
 209
 210     /* #if 'Ewald' in KERNEL_ELEC */
 211     sh_ewald         = _mm256_set1_pd(fr->ic->sh_ewald);
 212     beta             = _mm256_set1_pd(fr->ic->ewaldcoeff);
 213     beta2            = _mm256_mul_pd(beta,beta);
 214     beta3            = _mm256_mul_pd(beta,beta2);
 215
 216     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 217     ewtab            = fr->ic->tabq_coul_F;
 218     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 219     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 220     /*     #else */
 221     ewtab            = fr->ic->tabq_coul_FDV0;
 222     ewtabscale       = _mm256_set1_pd(fr->ic->tabq_scale);
 223     ewtabhalfspace   = _mm256_set1_pd(0.5/fr->ic->tabq_scale);
 224      /*     #endif */
 225     /* #endif */
 226
 227     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 228     invsqrta         = fr->invsqrta;
 229     dvda             = fr->dvda;
 230     gbtabscale       = _mm256_set1_pd(fr->gbtab.scale);
 231     gbtab            = fr->gbtab.data;
 232     gbinvepsdiff     = _mm256_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 233     /* #endif */
 234
 235     /* #if 'Water' in GEOMETRY_I */
 236     /* Setup water-specific parameters */
 237     inr              = nlist->iinr[0];
 238     /*     #for I in PARTICLES_ELEC_I */
 239     iq{I}              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+{I}]));
 240     /*     #endfor */
 241     /*     #for I in PARTICLES_VDW_I */
 242     vdwioffsetptr{I}   = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
 243     /*     #endfor */
 244     /* #endif */
 245
 246     /* #if 'Water' in GEOMETRY_J */
 247     /*     #for J in PARTICLES_ELEC_J */
 248     jq{J}              = _mm256_set1_pd(charge[inr+{J}]);
 249     /*     #endfor */
 250     /*     #for J in PARTICLES_VDW_J */
 251     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 252     /*     #endfor */
 253     /*     #for I,J in PAIRS_IJ */
 254     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 255     qq{I}{J}             = _mm256_mul_pd(iq{I},jq{J});
 256     /*         #endif */
 257     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 258     c6_{I}{J}            = _mm256_set1_pd(vdwioffsetptr{I}[vdwjidx{J}A]);
 259     c12_{I}{J}           = _mm256_set1_pd(vdwioffsetptr{I}[vdwjidx{J}A+1]);
 260     /*         #endif */
 261     /*     #endfor */
 262     /* #endif */
 263
 264     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 265     /*     #if KERNEL_ELEC!='None' */
 266     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 267     rcutoff_scalar   = fr->rcoulomb;
 268     /*     #else */
 269     rcutoff_scalar   = fr->rvdw;
 270     /*     #endif */
 271     rcutoff          = _mm256_set1_pd(rcutoff_scalar);
 272     rcutoff2         = _mm256_mul_pd(rcutoff,rcutoff);
 273     /* #endif */
 274
 275     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 276     sh_vdw_invrcut6  = _mm256_set1_pd(fr->ic->sh_invrc6);
 277     rvdw             = _mm256_set1_pd(fr->rvdw);
 278     /* #endif */
 279
 280     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 281     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 282     rswitch_scalar   = fr->rcoulomb_switch;
 283     rswitch          = _mm256_set1_pd(rswitch_scalar);
 284     /*     #else */
 285     rswitch_scalar   = fr->rvdw_switch;
 286     rswitch          = _mm256_set1_pd(rswitch_scalar);
 287     /*     #endif */
 288     /* Setup switch parameters */
 289     d_scalar         = rcutoff_scalar-rswitch_scalar;
 290     d                = _mm256_set1_pd(d_scalar);
 291     swV3             = _mm256_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 292     swV4             = _mm256_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 293     swV5             = _mm256_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 294     /*     #if 'Force' in KERNEL_VF */
 295     swF2             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 296     swF3             = _mm256_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 297     swF4             = _mm256_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 298     /*     #endif */
 299     /* #endif */
 300
 301     /* Avoid stupid compiler warnings */
 302     jnrA = jnrB = jnrC = jnrD = 0;
 303     j_coord_offsetA = 0;
 304     j_coord_offsetB = 0;
 305     j_coord_offsetC = 0;
 306     j_coord_offsetD = 0;
 307
 308     /* ## Keep track of the floating point operations we issue for reporting! */
 309     /* #define OUTERFLOPS 0 */
 310     outeriter        = 0;
 311     inneriter        = 0;
 312
 313     for(iidx=0;iidx<4*DIM;iidx++)
 314     {
 315         scratch[iidx] = 0.0;
 316     }
 317
 318     /* Start outer loop over neighborlists */
 319     for(iidx=0; iidx<nri; iidx++)
 320     {
 321         /* Load shift vector for this list */
 322         i_shift_offset   = DIM*shiftidx[iidx];
 323
 324         /* Load limits for loop over neighbors */
 325         j_index_start    = jindex[iidx];
 326         j_index_end      = jindex[iidx+1];
 327
 328         /* Get outer coordinate index */
 329         inr              = iinr[iidx];
 330         i_coord_offset   = DIM*inr;
 331
 332         /* Load i particle coords and add shift vector */
 333         /* #if GEOMETRY_I == 'Particle' */
 334         gmx_mm256_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 335         /* #elif GEOMETRY_I == 'Water3' */
 336         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 337                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 338         /* #elif GEOMETRY_I == 'Water4' */
 339         /*     #if 0 in PARTICLES_I                 */
 340         gmx_mm256_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 341                                                     &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 342         /*     #else                                */
 343         gmx_mm256_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 344                                                     &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 345         /*     #endif                               */
 346         /* #endif                                   */
 347
 348         /* #if 'Force' in KERNEL_VF */
 349         /*     #for I in PARTICLES_I */
 350         fix{I}             = _mm256_setzero_pd();
 351         fiy{I}             = _mm256_setzero_pd();
 352         fiz{I}             = _mm256_setzero_pd();
 353         /*     #endfor */
 354         /* #endif */
 355
 356         /* ## For water we already preloaded parameters at the start of the kernel */
 357         /* #if not 'Water' in GEOMETRY_I */
 358         /* Load parameters for i particles */
 359         /*     #for I in PARTICLES_ELEC_I */
 360         iq{I}              = _mm256_mul_pd(facel,_mm256_set1_pd(charge[inr+{I}]));
 361         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 362         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 363         isai{I}            = _mm256_set1_pd(invsqrta[inr+{I}]);
 364         /*         #endif */
 365         /*     #endfor */
 366         /*     #for I in PARTICLES_VDW_I */
 367         vdwioffsetptr{I}   = vdwparam+2*nvdwtype*vdwtype[inr+{I}];
 368         /*     #endfor */
 369         /* #endif */
 370
 371         /* #if 'Potential' in KERNEL_VF */
 372         /* Reset potential sums */
 373         /*     #if KERNEL_ELEC != 'None' */
 374         velecsum         = _mm256_setzero_pd();
 375         /*     #endif */
 376         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 377         vgbsum           = _mm256_setzero_pd();
 378         /*     #endif */
 379         /*     #if KERNEL_VDW != 'None' */
 380         vvdwsum          = _mm256_setzero_pd();
 381         /*     #endif */
 382         /* #endif */
 383         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 384         dvdasum          = _mm256_setzero_pd();
 385         /*     #endif */
 386
 387         /* #for ROUND in ['Loop','Epilogue'] */
 388
 389         /* #if ROUND =='Loop' */
 390         /* Start inner kernel loop */
 391         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 392         {
 393         /* ## First round is normal loop (next statement resets indentation) */
 394         /*     #if 0 */
 395         }
 396         /*     #endif */
 397         /* #else */
 398         if(jidx<j_index_end)
 399         {
 400         /* ## Second round is epilogue */
 401         /* #endif */
 402         /* #define INNERFLOPS 0 */
 403
 404             /* Get j neighbor index, and coordinate index */
 405             /* #if ROUND =='Loop' */
 406             jnrA             = jjnr[jidx];
 407             jnrB             = jjnr[jidx+1];
 408             jnrC             = jjnr[jidx+2];
 409             jnrD             = jjnr[jidx+3];
 410             /* #else */
 411             jnrlistA         = jjnr[jidx];
 412             jnrlistB         = jjnr[jidx+1];
 413             jnrlistC         = jjnr[jidx+2];
 414             jnrlistD         = jjnr[jidx+3];
 415             /* Sign of each element will be negative for non-real atoms.
 416              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 417              * so use it as val = _mm_andnot_pd(mask,val) to clear dummy entries.
 418              */
 419             tmpmask0 = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 420
 421             tmpmask1 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(3,3,2,2));
 422             tmpmask0 = _mm_permute_ps(tmpmask0,_GMX_MM_PERMUTE(1,1,0,0));
 423             dummy_mask = _mm256_castps_pd(gmx_mm256_set_m128(tmpmask1,tmpmask0));
 424
 425             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 426             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 427             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 428             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 429             /* #endif */
 430             j_coord_offsetA  = DIM*jnrA;
 431             j_coord_offsetB  = DIM*jnrB;
 432             j_coord_offsetC  = DIM*jnrC;
 433             j_coord_offsetD  = DIM*jnrD;
 434
 435             /* load j atom coordinates */
 436             /* #if GEOMETRY_J == 'Particle'             */
 437             gmx_mm256_load_1rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 438                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 439                                                  &jx0,&jy0,&jz0);
 440             /* #elif GEOMETRY_J == 'Water3'             */
 441             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 442                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 443                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 444             /* #elif GEOMETRY_J == 'Water4'             */
 445             /*     #if 0 in PARTICLES_J                 */
 446             gmx_mm256_load_4rvec_4ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 447                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 448                                                  &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 449                                                  &jy2,&jz2,&jx3,&jy3,&jz3);
 450             /*     #else                                */
 451             gmx_mm256_load_3rvec_4ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 452                                                  x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 453                                                  &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 454             /*     #endif                               */
 455             /* #endif                                   */
 456
 457             /* Calculate displacement vector */
 458             /* #for I,J in PAIRS_IJ */
 459             dx{I}{J}             = _mm256_sub_pd(ix{I},jx{J});
 460             dy{I}{J}             = _mm256_sub_pd(iy{I},jy{J});
 461             dz{I}{J}             = _mm256_sub_pd(iz{I},jz{J});
 462             /*     #define INNERFLOPS INNERFLOPS+3 */
 463             /* #endfor */
 464
 465             /* Calculate squared distance and things based on it */
 466             /* #for I,J in PAIRS_IJ */
 467             rsq{I}{J}            = gmx_mm256_calc_rsq_pd(dx{I}{J},dy{I}{J},dz{I}{J});
 468             /*     #define INNERFLOPS INNERFLOPS+5 */
 469             /* #endfor */
 470
 471             /* #for I,J in PAIRS_IJ */
 472             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 473             rinv{I}{J}           = gmx_mm256_invsqrt_pd(rsq{I}{J});
 474             /*         #define INNERFLOPS INNERFLOPS+5 */
 475             /*     #endif */
 476             /* #endfor */
 477
 478             /* #for I,J in PAIRS_IJ */
 479             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 480             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 481             rinvsq{I}{J}         = gmx_mm256_inv_pd(rsq{I}{J});
 482             /*             #define INNERFLOPS INNERFLOPS+4 */
 483             /*         #else */
 484             rinvsq{I}{J}         = _mm256_mul_pd(rinv{I}{J},rinv{I}{J});
 485             /*             #define INNERFLOPS INNERFLOPS+1 */
 486             /*         #endif */
 487             /*     #endif */
 488             /* #endfor */
 489
 490             /* #if not 'Water' in GEOMETRY_J */
 491             /* Load parameters for j particles */
 492             /*     #for J in PARTICLES_ELEC_J */
 493             jq{J}              = gmx_mm256_load_4real_swizzle_pd(charge+jnrA+{J},charge+jnrB+{J},
 494                                                                  charge+jnrC+{J},charge+jnrD+{J});
 495             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 496             isaj{J}            = gmx_mm256_load_4real_swizzle_pd(invsqrta+jnrA+{J},invsqrta+jnrB+{J},
 497                                                                  invsqrta+jnrC+{J},invsqrta+jnrD+{J});
 498             /*         #endif */
 499             /*     #endfor */
 500             /*     #for J in PARTICLES_VDW_J */
 501             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 502             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 503             vdwjidx{J}C        = 2*vdwtype[jnrC+{J}];
 504             vdwjidx{J}D        = 2*vdwtype[jnrD+{J}];
 505             /*     #endfor */
 506             /* #endif */
 507
 508             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 509             /*     #for J in PARTICLES_J */
 510             fjx{J}             = _mm256_setzero_pd();
 511             fjy{J}             = _mm256_setzero_pd();
 512             fjz{J}             = _mm256_setzero_pd();
 513             /*     #endfor */
 514             /* #endif */
 515
 516             /* #for I,J in PAIRS_IJ */
 517
 518             /**************************
 519              * CALCULATE INTERACTIONS *
 520              **************************/
 521
 522             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 523             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 524             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 525             if (gmx_mm256_any_lt(rsq{I}{J},rcutoff2))
 526             {
 527                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 528             }
 529             /*         #endif */
 530             /*         #define INNERFLOPS INNERFLOPS+1 */
 531             /*     #endif */
 532
 533             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 534             r{I}{J}              = _mm256_mul_pd(rsq{I}{J},rinv{I}{J});
 535             /*         #if ROUND == 'Epilogue' */
 536             r{I}{J}              = _mm256_andnot_pd(dummy_mask,r{I}{J});
 537             /*             #define INNERFLOPS INNERFLOPS+1 */
 538             /*         #endif */
 539             /*         #define INNERFLOPS INNERFLOPS+1 */
 540             /*     #endif */
 541
 542             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 543             /*     #if not 'Water' in GEOMETRY_J */
 544             /* Compute parameters for interactions between i and j atoms */
 545             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 546             qq{I}{J}             = _mm256_mul_pd(iq{I},jq{J});
 547             /*             #define INNERFLOPS INNERFLOPS+1 */
 548             /*         #endif */
 549             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 550             gmx_mm256_load_4pair_swizzle_pd(vdwioffsetptr{I}+vdwjidx{J}A,
 551                                             vdwioffsetptr{I}+vdwjidx{J}B,
 552                                             vdwioffsetptr{I}+vdwjidx{J}C,
 553                                             vdwioffsetptr{I}+vdwjidx{J}D,
 554                                             &c6_{I}{J},&c12_{I}{J});
 555             /*         #endif */
 556             /*     #endif */
 557
 558             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 559             /* Calculate table index by multiplying r with table scale and truncate to integer */
 560             rt               = _mm256_mul_pd(r{I}{J},vftabscale);
 561             vfitab           = _mm256_cvttpd_epi32(rt);
 562             vfeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 563             /*         #define INNERFLOPS INNERFLOPS+4                          */
 564             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW     */
 565             /*             ## 3 tables, 4 bytes per point: multiply index by 12 */
 566             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 567             /*         #elif 'Table' in KERNEL_ELEC                             */
 568             /*             ## 1 table, 4 bytes per point: multiply index by 4   */
 569             vfitab           = _mm_slli_epi32(vfitab,2);
 570             /*         #elif 'Table' in KERNEL_VDW                              */
 571             /*             ## 2 tables, 4 bytes per point: multiply index by 8  */
 572             vfitab           = _mm_slli_epi32(vfitab,3);
 573             /*         #endif                                                   */
 574             /*     #endif */
 575
 576             /*     ## ELECTROSTATIC INTERACTIONS */
 577             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 578
 579             /*         #if KERNEL_ELEC=='Coulomb' */
 580
 581             /* COULOMB ELECTROSTATICS */
 582             velec            = _mm256_mul_pd(qq{I}{J},rinv{I}{J});
 583             /*             #define INNERFLOPS INNERFLOPS+1 */
 584             /*             #if 'Force' in KERNEL_VF */
 585             felec            = _mm256_mul_pd(velec,rinvsq{I}{J});
 586             /*                 #define INNERFLOPS INNERFLOPS+1 */
 587             /*             #endif */
 588
 589             /*         #elif KERNEL_ELEC=='ReactionField' */
 590
 591             /* REACTION-FIELD ELECTROSTATICS */
 592             /*             #if 'Potential' in KERNEL_VF */
 593             velec            = _mm256_mul_pd(qq{I}{J},_mm256_sub_pd(_mm256_add_pd(rinv{I}{J},_mm256_mul_pd(krf,rsq{I}{J})),crf));
 594             /*                 #define INNERFLOPS INNERFLOPS+4 */
 595             /*             #endif */
 596             /*             #if 'Force' in KERNEL_VF */
 597             felec            = _mm256_mul_pd(qq{I}{J},_mm256_sub_pd(_mm256_mul_pd(rinv{I}{J},rinvsq{I}{J}),krf2));
 598             /*                 #define INNERFLOPS INNERFLOPS+3 */
 599             /*             #endif */
 600
 601             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 602
 603             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 604             isaprod          = _mm256_mul_pd(isai{I},isaj{J});
 605             gbqqfactor       = _mm256_xor_pd(signbit,_mm256_mul_pd(qq{I}{J},_mm256_mul_pd(isaprod,gbinvepsdiff)));
 606             gbscale          = _mm256_mul_pd(isaprod,gbtabscale);
 607             /*             #define INNERFLOPS INNERFLOPS+5 */
 608
 609             /* Calculate generalized born table index - this is a separate table from the normal one,
 610              * but we use the same procedure by multiplying r with scale and truncating to integer.
 611              */
 612             rt               = _mm256_mul_pd(r{I}{J},gbscale);
 613             gbitab           = _mm256_cvttpd_epi32(rt);
 614             gbeps            = _mm256_sub_pd(rt,_mm256_round_pd(rt, _MM_FROUND_FLOOR));
 615             gbitab           = _mm_slli_epi32(gbitab,2);
 616             Y                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,0) );
 617             F                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,1) );
 618             G                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,2) );
 619             H                = _mm256_load_pd( gbtab + _mm_extract_epi32(gbitab,3) );
 620             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 621             Heps             = _mm256_mul_pd(gbeps,H);
 622             Fp               = _mm256_add_pd(F,_mm256_mul_pd(gbeps,_mm256_add_pd(G,Heps)));
 623             VV               = _mm256_add_pd(Y,_mm256_mul_pd(gbeps,Fp));
 624             vgb              = _mm256_mul_pd(gbqqfactor,VV);
 625             /*             #define INNERFLOPS INNERFLOPS+10 */
 626
 627             /*             #if 'Force' in KERNEL_VF */
 628             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(gbeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 629             fgb              = _mm256_mul_pd(gbqqfactor,_mm256_mul_pd(FF,gbscale));
 630             dvdatmp          = _mm256_mul_pd(minushalf,_mm256_add_pd(vgb,_mm256_mul_pd(fgb,r{I}{J})));
 631             /*                 #if ROUND == 'Epilogue' */
 632             dvdatmp          = _mm256_andnot_ps(dummy_mask,dvdatmp);
 633             /*                 #endif */
 634             dvdasum          = _mm256_add_pd(dvdasum,dvdatmp);
 635             /*                 #if ROUND == 'Loop' */
 636             fjptrA           = dvda+jnrA;
 637             fjptrB           = dvda+jnrB;
 638             fjptrC           = dvda+jnrC;
 639             fjptrD           = dvda+jnrD;
 640             /*                 #else */
 641             /* 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. */
 642             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 643             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 644             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 645             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 646             /*                 #endif */
 647             gmx_mm256_increment_4real_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 648                                                  _mm256_mul_pd(dvdatmp,_mm256_mul_pd(isaj{J},isaj{J})));
 649             /*                 #define INNERFLOPS INNERFLOPS+12 */
 650             /*             #endif */
 651             velec            = _mm256_mul_pd(qq{I}{J},rinv{I}{J});
 652             /*                 #define INNERFLOPS INNERFLOPS+1 */
 653             /*             #if 'Force' in KERNEL_VF */
 654             felec            = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(velec,rinv{I}{J}),fgb),rinv{I}{J});
 655             /*                 #define INNERFLOPS INNERFLOPS+3 */
 656             /*             #endif */
 657
 658             /*         #elif KERNEL_ELEC=='Ewald' */
 659             /* EWALD ELECTROSTATICS */
 660
 661             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 662             ewrt             = _mm256_mul_pd(r{I}{J},ewtabscale);
 663             ewitab           = _mm256_cvttpd_epi32(ewrt);
 664             eweps            = _mm256_sub_pd(ewrt,_mm256_round_pd(ewrt, _MM_FROUND_FLOOR));
 665             /*             #define INNERFLOPS INNERFLOPS+4 */
 666             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 667             ewitab           = _mm_slli_epi32(ewitab,2);
 668             ewtabF           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 669             ewtabD           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 670             ewtabV           = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,2) );
 671             ewtabFn          = _mm256_load_pd( ewtab + _mm_extract_epi32(ewitab,3) );
 672             GMX_MM256_FULLTRANSPOSE4_PD(ewtabF,ewtabD,ewtabV,ewtabFn);
 673             felec            = _mm256_add_pd(ewtabF,_mm256_mul_pd(eweps,ewtabD));
 674             /*                 #define INNERFLOPS INNERFLOPS+2 */
 675             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 676             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 677             velec            = _mm256_mul_pd(qq{I}{J},_mm256_sub_pd(_mm256_sub_pd(rinv{I}{J},sh_ewald),velec));
 678             /*                     #define INNERFLOPS INNERFLOPS+7 */
 679             /*                 #else */
 680             velec            = _mm256_sub_pd(ewtabV,_mm256_mul_pd(_mm256_mul_pd(ewtabhalfspace,eweps),_mm256_add_pd(ewtabF,felec)));
 681             velec            = _mm256_mul_pd(qq{I}{J},_mm256_sub_pd(rinv{I}{J},velec));
 682             /*                     #define INNERFLOPS INNERFLOPS+6 */
 683             /*                 #endif */
 684             /*                 #if 'Force' in KERNEL_VF */
 685             felec            = _mm256_mul_pd(_mm256_mul_pd(qq{I}{J},rinv{I}{J}),_mm256_sub_pd(rinvsq{I}{J},felec));
 686             /*                      #define INNERFLOPS INNERFLOPS+3 */
 687             /*                 #endif */
 688             /*             #elif KERNEL_VF=='Force' */
 689             gmx_mm256_load_4pair_swizzle_pd(ewtab + _mm_extract_epi32(ewitab,0),ewtab + _mm_extract_epi32(ewitab,1),
 690                                             ewtab + _mm_extract_epi32(ewitab,2),ewtab + _mm_extract_epi32(ewitab,3),
 691                                             &ewtabF,&ewtabFn);
 692             felec            = _mm256_add_pd(_mm256_mul_pd( _mm256_sub_pd(one,eweps),ewtabF),_mm256_mul_pd(eweps,ewtabFn));
 693             felec            = _mm256_mul_pd(_mm256_mul_pd(qq{I}{J},rinv{I}{J}),_mm256_sub_pd(rinvsq{I}{J},felec));
 694             /*                 #define INNERFLOPS INNERFLOPS+7 */
 695             /*             #endif */
 696
 697             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 698
 699             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 700             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 701             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 702             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 703             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 704             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 705             Heps             = _mm256_mul_pd(vfeps,H);
 706             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 707             /*             #define INNERFLOPS INNERFLOPS+4 */
 708             /*             #if 'Potential' in KERNEL_VF */
 709             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 710             velec            = _mm256_mul_pd(qq{I}{J},VV);
 711             /*                 #define INNERFLOPS INNERFLOPS+3 */
 712             /*             #endif */
 713             /*             #if 'Force' in KERNEL_VF */
 714             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 715             felec            = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_mul_pd(qq{I}{J},FF),_mm256_mul_pd(vftabscale,rinv{I}{J})));
 716             /*                 #define INNERFLOPS INNERFLOPS+7 */
 717             /*             #endif */
 718             /*         #endif */
 719             /*         ## End of check for electrostatics interaction forms */
 720             /*     #endif */
 721             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 722
 723             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 724
 725             /*         #if KERNEL_VDW=='LennardJones' */
 726
 727             /* LENNARD-JONES DISPERSION/REPULSION */
 728
 729             rinvsix          = _mm256_mul_pd(_mm256_mul_pd(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 730             /*             #define INNERFLOPS INNERFLOPS+2 */
 731             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 732             vvdw6            = _mm256_mul_pd(c6_{I}{J},rinvsix);
 733             vvdw12           = _mm256_mul_pd(c12_{I}{J},_mm256_mul_pd(rinvsix,rinvsix));
 734             /*                 #define INNERFLOPS INNERFLOPS+3 */
 735             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 736             vvdw             = _mm256_sub_pd(_mm256_mul_pd( _mm256_sub_pd(vvdw12 , _mm256_mul_pd(c12_{I}{J},_mm256_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 737                                           _mm256_mul_pd( _mm256_sub_pd(vvdw6,_mm256_mul_pd(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
 738             /*                     #define INNERFLOPS INNERFLOPS+8 */
 739             /*                 #else */
 740             vvdw             = _mm256_sub_pd( _mm256_mul_pd(vvdw12,one_twelfth) , _mm256_mul_pd(vvdw6,one_sixth) );
 741             /*                     #define INNERFLOPS INNERFLOPS+3 */
 742             /*                 #endif */
 743             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 744             /*                 #if 'Force' in KERNEL_VF */
 745             fvdw             = _mm256_mul_pd(_mm256_sub_pd(vvdw12,vvdw6),rinvsq{I}{J});
 746             /*                     #define INNERFLOPS INNERFLOPS+2 */
 747             /*                 #endif */
 748             /*             #elif KERNEL_VF=='Force' */
 749             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 750             fvdw             = _mm256_mul_pd(_mm256_sub_pd(_mm256_mul_pd(c12_{I}{J},rinvsix),c6_{I}{J}),_mm256_mul_pd(rinvsix,rinvsq{I}{J}));
 751             /*                 #define INNERFLOPS INNERFLOPS+4 */
 752             /*             #endif */
 753
 754             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 755
 756             /* CUBIC SPLINE TABLE DISPERSION */
 757             /*             #if 'Table' in KERNEL_ELEC */
 758             vfitab           = _mm_add_epi32(vfitab,ifour);
 759             /*             #endif                     */
 760             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 761             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 762             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 763             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 764             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 765             Heps             = _mm256_mul_pd(vfeps,H);
 766             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 767             /*             #define INNERFLOPS INNERFLOPS+4 */
 768             /*             #if 'Potential' in KERNEL_VF */
 769             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 770             vvdw6            = _mm256_mul_pd(c6_{I}{J},VV);
 771             /*                 #define INNERFLOPS INNERFLOPS+3 */
 772             /*             #endif */
 773             /*             #if 'Force' in KERNEL_VF */
 774             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 775             fvdw6            = _mm256_mul_pd(c6_{I}{J},FF);
 776             /*                 #define INNERFLOPS INNERFLOPS+4 */
 777             /*             #endif */
 778
 779             /* CUBIC SPLINE TABLE REPULSION */
 780             vfitab           = _mm_add_epi32(vfitab,ifour);
 781             Y                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 782             F                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 783             G                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,2) );
 784             H                = _mm256_load_pd( vftab + _mm_extract_epi32(vfitab,3) );
 785             GMX_MM256_FULLTRANSPOSE4_PD(Y,F,G,H);
 786             Heps             = _mm256_mul_pd(vfeps,H);
 787             Fp               = _mm256_add_pd(F,_mm256_mul_pd(vfeps,_mm256_add_pd(G,Heps)));
 788             /*             #define INNERFLOPS INNERFLOPS+4 */
 789             /*             #if 'Potential' in KERNEL_VF */
 790             VV               = _mm256_add_pd(Y,_mm256_mul_pd(vfeps,Fp));
 791             vvdw12           = _mm256_mul_pd(c12_{I}{J},VV);
 792             /*                 #define INNERFLOPS INNERFLOPS+3 */
 793             /*             #endif */
 794             /*             #if 'Force' in KERNEL_VF */
 795             FF               = _mm256_add_pd(Fp,_mm256_mul_pd(vfeps,_mm256_add_pd(G,_mm256_add_pd(Heps,Heps))));
 796             fvdw12           = _mm256_mul_pd(c12_{I}{J},FF);
 797             /*                 #define INNERFLOPS INNERFLOPS+5 */
 798             /*             #endif */
 799             /*             #if 'Potential' in KERNEL_VF */
 800             vvdw             = _mm256_add_pd(vvdw12,vvdw6);
 801             /*                 #define INNERFLOPS INNERFLOPS+1 */
 802             /*             #endif */
 803             /*             #if 'Force' in KERNEL_VF */
 804             fvdw             = _mm256_xor_pd(signbit,_mm256_mul_pd(_mm256_add_pd(fvdw6,fvdw12),_mm256_mul_pd(vftabscale,rinv{I}{J})));
 805             /*                 #define INNERFLOPS INNERFLOPS+4 */
 806             /*             #endif */
 807             /*         #endif */
 808             /*         ## End of check for vdw interaction forms */
 809             /*     #endif */
 810             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 811
 812             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 813             d                = _mm256_sub_pd(r{I}{J},rswitch);
 814             d                = _mm256_max_pd(d,_mm256_setzero_pd());
 815             d2               = _mm256_mul_pd(d,d);
 816             sw               = _mm256_add_pd(one,_mm256_mul_pd(d2,_mm256_mul_pd(d,_mm256_add_pd(swV3,_mm256_mul_pd(d,_mm256_add_pd(swV4,_mm256_mul_pd(d,swV5)))))));
 817             /*         #define INNERFLOPS INNERFLOPS+10 */
 818
 819             /*         #if 'Force' in KERNEL_VF */
 820             dsw              = _mm256_mul_pd(d2,_mm256_add_pd(swF2,_mm256_mul_pd(d,_mm256_add_pd(swF3,_mm256_mul_pd(d,swF4)))));
 821             /*             #define INNERFLOPS INNERFLOPS+5 */
 822             /*         #endif */
 823
 824             /* Evaluate switch function */
 825             /*         #if 'Force' in KERNEL_VF */
 826             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 827             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 828             felec            = _mm256_sub_pd( _mm256_mul_pd(felec,sw) , _mm256_mul_pd(rinv{I}{J},_mm256_mul_pd(velec,dsw)) );
 829             /*                 #define INNERFLOPS INNERFLOPS+4 */
 830             /*             #endif */
 831             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 832             fvdw             = _mm256_sub_pd( _mm256_mul_pd(fvdw,sw) , _mm256_mul_pd(rinv{I}{J},_mm256_mul_pd(vvdw,dsw)) );
 833             /*                 #define INNERFLOPS INNERFLOPS+4 */
 834             /*             #endif */
 835             /*         #endif */
 836             /*         #if 'Potential' in KERNEL_VF */
 837             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 838             velec            = _mm256_mul_pd(velec,sw);
 839             /*                 #define INNERFLOPS INNERFLOPS+1 */
 840             /*             #endif */
 841             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 842             vvdw             = _mm256_mul_pd(vvdw,sw);
 843             /*                 #define INNERFLOPS INNERFLOPS+1 */
 844             /*             #endif */
 845             /*         #endif */
 846             /*     #endif */
 847             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 848             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 849             cutoff_mask      = _mm256_cmp_pd(rsq{I}{J},rcutoff2,_CMP_LT_OQ);
 850             /*         #define INNERFLOPS INNERFLOPS+1 */
 851             /*     #endif */
 852
 853             /*     #if 'Potential' in KERNEL_VF */
 854             /* Update potential sum for this i atom from the interaction with this j atom. */
 855             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 856             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 857             velec            = _mm256_and_pd(velec,cutoff_mask);
 858             /*                 #define INNERFLOPS INNERFLOPS+1 */
 859             /*             #endif                                       */
 860             /*             #if ROUND == 'Epilogue' */
 861             velec            = _mm256_andnot_pd(dummy_mask,velec);
 862             /*             #endif */
 863             velecsum         = _mm256_add_pd(velecsum,velec);
 864             /*             #define INNERFLOPS INNERFLOPS+1 */
 865             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 866             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 867             vgb              = _mm256_and_pd(vgb,cutoff_mask);
 868             /*                 #define INNERFLOPS INNERFLOPS+1 */
 869             /*             #endif                                       */
 870             /*             #if ROUND == 'Epilogue' */
 871             vgb              = _mm256_andnot_pd(dummy_mask,vgb);
 872             /*             #endif */
 873             vgbsum           = _mm256_add_pd(vgbsum,vgb);
 874             /*                 #define INNERFLOPS INNERFLOPS+1 */
 875             /*             #endif */
 876             /*         #endif */
 877             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 878             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 879             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 880             vvdw             = _mm256_and_pd(vvdw,cutoff_mask);
 881             /*                 #define INNERFLOPS INNERFLOPS+1 */
 882             /*             #endif                                       */
 883             /*             #if ROUND == 'Epilogue' */
 884             vvdw             = _mm256_andnot_pd(dummy_mask,vvdw);
 885             /*             #endif */
 886             vvdwsum          = _mm256_add_pd(vvdwsum,vvdw);
 887             /*             #define INNERFLOPS INNERFLOPS+1 */
 888             /*         #endif */
 889             /*     #endif */
 890
 891             /*     #if 'Force' in KERNEL_VF */
 892
 893             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 894             fscal            = _mm256_add_pd(felec,fvdw);
 895             /*             #define INNERFLOPS INNERFLOPS+1 */
 896             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 897             fscal            = felec;
 898             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 899             fscal            = fvdw;
 900             /*        #endif */
 901
 902             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 903             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 904             fscal            = _mm256_and_pd(fscal,cutoff_mask);
 905             /*                 #define INNERFLOPS INNERFLOPS+1 */
 906             /*             #endif                                       */
 907
 908             /*             #if ROUND == 'Epilogue' */
 909             fscal            = _mm256_andnot_pd(dummy_mask,fscal);
 910             /*             #endif */
 911
 912             /* Calculate temporary vectorial force */
 913             tx               = _mm256_mul_pd(fscal,dx{I}{J});
 914             ty               = _mm256_mul_pd(fscal,dy{I}{J});
 915             tz               = _mm256_mul_pd(fscal,dz{I}{J});
 916
 917             /* Update vectorial force */
 918             fix{I}             = _mm256_add_pd(fix{I},tx);
 919             fiy{I}             = _mm256_add_pd(fiy{I},ty);
 920             fiz{I}             = _mm256_add_pd(fiz{I},tz);
 921             /*             #define INNERFLOPS INNERFLOPS+6 */
 922
 923             /* #if GEOMETRY_I == 'Particle'             */
 924             /*     #if ROUND == 'Loop' */
 925             fjptrA             = f+j_coord_offsetA;
 926             fjptrB             = f+j_coord_offsetB;
 927             fjptrC             = f+j_coord_offsetC;
 928             fjptrD             = f+j_coord_offsetD;
 929             /*     #else */
 930             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 931             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 932             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 933             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 934             /*     #endif */
 935             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 936             /*     #define INNERFLOPS INNERFLOPS+3      */
 937             /* #else                                    */
 938             fjx{J}             = _mm256_add_pd(fjx{J},tx);
 939             fjy{J}             = _mm256_add_pd(fjy{J},ty);
 940             fjz{J}             = _mm256_add_pd(fjz{J},tz);
 941             /*     #define INNERFLOPS INNERFLOPS+3      */
 942             /* #endif                                   */
 943
 944             /*     #endif */
 945
 946             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 947             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 948             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
 949             {
 950                 /*     #endif */
 951             }
 952             /*     #endif */
 953             /*    ## End of check for the interaction being outside the cutoff */
 954
 955             /* #endfor */
 956             /* ## End of loop over i-j interaction pairs */
 957
 958             /* #if GEOMETRY_I != 'Particle' */
 959             /*     #if ROUND == 'Loop' */
 960             fjptrA             = f+j_coord_offsetA;
 961             fjptrB             = f+j_coord_offsetB;
 962             fjptrC             = f+j_coord_offsetC;
 963             fjptrD             = f+j_coord_offsetD;
 964             /*     #else */
 965             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 966             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 967             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 968             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 969             /*     #endif */
 970             /* #endif */
 971
 972             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
 973             gmx_mm256_decrement_1rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 974             /*     #define INNERFLOPS INNERFLOPS+3      */
 975             /* #elif GEOMETRY_J == 'Water3'             */
 976             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 977                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 978             /*     #define INNERFLOPS INNERFLOPS+9      */
 979             /* #elif GEOMETRY_J == 'Water4'             */
 980             /*     #if 0 in PARTICLES_J                 */
 981             gmx_mm256_decrement_4rvec_4ptr_swizzle_pd(fjptrA,fjptrB,fjptrC,fjptrD,
 982                                                       fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 983                                                       fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 984             /*     #define INNERFLOPS INNERFLOPS+12     */
 985             /*     #else                                */
 986             gmx_mm256_decrement_3rvec_4ptr_swizzle_pd(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 987                                                       fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 988             /*     #define INNERFLOPS INNERFLOPS+9      */
 989             /*     #endif                               */
 990             /* #endif                                   */
 991
 992             /* Inner loop uses {INNERFLOPS} flops */
 993         }
 994
 995         /* #endfor */
 996
 997         /* End of innermost loop */
 998
 999         /* #if 'Force' in KERNEL_VF */
1000         /*     #if GEOMETRY_I == 'Particle'            */
1001         gmx_mm256_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
1002                                                  f+i_coord_offset,fshift+i_shift_offset);
1003         /*         #define OUTERFLOPS OUTERFLOPS+6     */
1004         /*     #elif GEOMETRY_I == 'Water3'            */
1005         gmx_mm256_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1006                                                  f+i_coord_offset,fshift+i_shift_offset);
1007         /*         #define OUTERFLOPS OUTERFLOPS+18    */
1008         /*     #elif GEOMETRY_I == 'Water4'            */
1009         /*         #if 0 in PARTICLES_I                */
1010         gmx_mm256_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1011                                                  f+i_coord_offset,fshift+i_shift_offset);
1012         /*             #define OUTERFLOPS OUTERFLOPS+24    */
1013         /*         #else                               */
1014         gmx_mm256_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1015                                                  f+i_coord_offset+DIM,fshift+i_shift_offset);
1016         /*             #define OUTERFLOPS OUTERFLOPS+18    */
1017         /*         #endif                              */
1018         /*     #endif                                  */
1019         /* #endif                                      */
1020
1021         /* #if 'Potential' in KERNEL_VF */
1022         ggid                        = gid[iidx];
1023         /* Update potential energies */
1024         /*     #if KERNEL_ELEC != 'None' */
1025         gmx_mm256_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1026         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1027         /*     #endif */
1028         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
1029         gmx_mm256_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
1030         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1031         /*     #endif */
1032         /*     #if KERNEL_VDW != 'None' */
1033         gmx_mm256_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1034         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1035         /*     #endif */
1036         /* #endif */
1037         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1038         dvdasum = _mm256_mul_pd(dvdasum, _mm256_mul_pd(isai{I},isai{I}));
1039         gmx_mm256_update_1pot_pd(dvdasum,dvda+inr);
1040         /*     #endif */
1041
1042         /* Increment number of inner iterations */
1043         inneriter                  += j_index_end - j_index_start;
1044
1045         /* Outer loop uses {OUTERFLOPS} flops */
1046     }
1047
1048     /* Increment number of outer iterations */
1049     outeriter        += nri;
1050
1051     /* Update outer/inner flops */
1052     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1053     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1054     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1055     /* #if GEOMETRY_I == 'Water3'            */
1056     /*     #define ISUFFIX '_W3'             */
1057     /* #elif GEOMETRY_I == 'Water4'          */
1058     /*     #define ISUFFIX '_W4'             */
1059     /* #else                                 */
1060     /*     #define ISUFFIX ''                */
1061     /* #endif                                */
1062     /* #if GEOMETRY_J == 'Water3'            */
1063     /*     #define JSUFFIX 'W3'              */
1064     /* #elif GEOMETRY_J == 'Water4'          */
1065     /*     #define JSUFFIX 'W4'              */
1066     /* #else                                 */
1067     /*     #define JSUFFIX ''                */
1068     /* #endif                                */
1069     /* #if 'PotentialAndForce' in KERNEL_VF  */
1070     /*     #define VFSUFFIX  '_VF'           */
1071     /* #elif 'Potential' in KERNEL_VF        */
1072     /*     #define VFSUFFIX '_V'             */
1073     /* #else                                 */
1074     /*     #define VFSUFFIX '_F'             */
1075     /* #endif                                */
1076
1077     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1078     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1079     /* #elif KERNEL_ELEC != 'None' */
1080     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1081     /* #else */
1082     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1083     /* #endif  */
1084 }