src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_template_avx_128_fma_single.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_128_fma_single.h"
  51 #include "kernelutil_x86_avx_128_fma_single.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_128, 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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 116     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 117     real             rcutoff_scalar;
 118     real             *shiftvec,*fshift,*x,*f;
 119     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 120     real             scratch[4*DIM];
 121     __m128           fscal,rcutoff,rcutoff2,jidxall;
 122     /* #for I in PARTICLES_I */
 123     int              vdwioffset{I};
 124     __m128           ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 125     /* #endfor */
 126     /* #for J in PARTICLES_J */
 127     int              vdwjidx{J}A,vdwjidx{J}B,vdwjidx{J}C,vdwjidx{J}D;
 128     __m128           jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 129     /* #endfor */
 130     /* #for I,J in PAIRS_IJ */
 131     __m128           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};
 132     /* #endfor */
 133     /* #if KERNEL_ELEC != 'None' */
 134     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 135     real             *charge;
 136     /* #endif */
 137     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 138     __m128i          gbitab;
 139     __m128           vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,gbeps,twogbeps,dvdatmp;
 140     __m128           minushalf = _mm_set1_ps(-0.5);
 141     real             *invsqrta,*dvda,*gbtab;
 142     /* #endif */
 143     /* #if KERNEL_VDW != 'None' */
 144     int              nvdwtype;
 145     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 146     int              *vdwtype;
 147     real             *vdwparam;
 148     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 149     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 150     /* #endif */
 151     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 152     __m128i          vfitab;
 153     __m128i          ifour       = _mm_set1_epi32(4);
 154     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 155     real             *vftab;
 156     /* #endif */
 157     /* #if 'Ewald' in KERNEL_ELEC */
 158     __m128i          ewitab;
 159     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 160     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 161     real             *ewtab;
 162     /* #endif */
 163     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 164     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 165     real             rswitch_scalar,d_scalar;
 166     /* #endif */
 167     __m128           dummy_mask,cutoff_mask;
 168     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 169     __m128           one     = _mm_set1_ps(1.0);
 170     __m128           two     = _mm_set1_ps(2.0);
 171     x                = xx[0];
 172     f                = ff[0];
 173
 174     nri              = nlist->nri;
 175     iinr             = nlist->iinr;
 176     jindex           = nlist->jindex;
 177     jjnr             = nlist->jjnr;
 178     shiftidx         = nlist->shift;
 179     gid              = nlist->gid;
 180     shiftvec         = fr->shift_vec[0];
 181     fshift           = fr->fshift[0];
 182     /* #if KERNEL_ELEC != 'None' */
 183     facel            = _mm_set1_ps(fr->epsfac);
 184     charge           = mdatoms->chargeA;
 185     /*     #if 'ReactionField' in KERNEL_ELEC */
 186     krf              = _mm_set1_ps(fr->ic->k_rf);
 187     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 188     crf              = _mm_set1_ps(fr->ic->c_rf);
 189     /*     #endif */
 190     /* #endif */
 191     /* #if KERNEL_VDW != 'None' */
 192     nvdwtype         = fr->ntype;
 193     vdwparam         = fr->nbfp;
 194     vdwtype          = mdatoms->typeA;
 195     /* #endif */
 196
 197     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 198     vftab            = kernel_data->table_elec_vdw->data;
 199     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 200     /* #elif 'Table' in KERNEL_ELEC */
 201     vftab            = kernel_data->table_elec->data;
 202     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 203     /* #elif 'Table' in KERNEL_VDW */
 204     vftab            = kernel_data->table_vdw->data;
 205     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 206     /* #endif */
 207
 208     /* #if 'Ewald' in KERNEL_ELEC */
 209     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 210     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 211     beta2            = _mm_mul_ps(beta,beta);
 212     beta3            = _mm_mul_ps(beta,beta2);
 213     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 214     ewtab            = fr->ic->tabq_coul_F;
 215     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 216     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 217     /*     #else */
 218     ewtab            = fr->ic->tabq_coul_FDV0;
 219     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 220     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 221      /*     #endif */
 222     /* #endif */
 223
 224     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 225     invsqrta         = fr->invsqrta;
 226     dvda             = fr->dvda;
 227     gbtabscale       = _mm_set1_ps(fr->gbtab.scale);
 228     gbtab            = fr->gbtab.data;
 229     gbinvepsdiff     = _mm_set1_ps((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 230     /* #endif */
 231
 232     /* #if 'Water' in GEOMETRY_I */
 233     /* Setup water-specific parameters */
 234     inr              = nlist->iinr[0];
 235     /*     #for I in PARTICLES_ELEC_I */
 236     iq{I}              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+{I}]));
 237     /*     #endfor */
 238     /*     #for I in PARTICLES_VDW_I */
 239     vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 240     /*     #endfor */
 241     /* #endif */
 242
 243     /* #if 'Water' in GEOMETRY_J */
 244     /*     #for J in PARTICLES_ELEC_J */
 245     jq{J}              = _mm_set1_ps(charge[inr+{J}]);
 246     /*     #endfor */
 247     /*     #for J in PARTICLES_VDW_J */
 248     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 249     /*     #endfor */
 250     /*     #for I,J in PAIRS_IJ */
 251     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 252     qq{I}{J}             = _mm_mul_ps(iq{I},jq{J});
 253     /*         #endif */
 254     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 255     c6_{I}{J}            = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 256     c12_{I}{J}           = _mm_set1_ps(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 257     /*         #endif */
 258     /*     #endfor */
 259     /* #endif */
 260
 261     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 262     /*     #if KERNEL_ELEC!='None' */
 263     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 264     rcutoff_scalar   = fr->rcoulomb;
 265     /*     #else */
 266     rcutoff_scalar   = fr->rvdw;
 267     /*     #endif */
 268     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 269     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 270     /* #endif */
 271
 272     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 273     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 274     rvdw             = _mm_set1_ps(fr->rvdw);
 275     /* #endif */
 276
 277     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 278     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 279     rswitch_scalar   = fr->rcoulomb_switch;
 280     rswitch          = _mm_set1_ps(rswitch_scalar);
 281     /*     #else */
 282     rswitch_scalar   = fr->rvdw_switch;
 283     rswitch          = _mm_set1_ps(rswitch_scalar);
 284     /*     #endif */
 285     /* Setup switch parameters */
 286     d_scalar         = rcutoff_scalar-rswitch_scalar;
 287     d                = _mm_set1_ps(d_scalar);
 288     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 289     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 290     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 291     /*     #if 'Force' in KERNEL_VF */
 292     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 293     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 294     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 295     /*     #endif */
 296     /* #endif */
 297
 298     /* Avoid stupid compiler warnings */
 299     jnrA = jnrB = jnrC = jnrD = 0;
 300     j_coord_offsetA = 0;
 301     j_coord_offsetB = 0;
 302     j_coord_offsetC = 0;
 303     j_coord_offsetD = 0;
 304
 305     /* ## Keep track of the floating point operations we issue for reporting! */
 306     /* #define OUTERFLOPS 0 */
 307     outeriter        = 0;
 308     inneriter        = 0;
 309
 310     for(iidx=0;iidx<4*DIM;iidx++)
 311     {
 312         scratch[iidx] = 0.0;
 313     }
 314
 315     /* Start outer loop over neighborlists */
 316     for(iidx=0; iidx<nri; iidx++)
 317     {
 318         /* Load shift vector for this list */
 319         i_shift_offset   = DIM*shiftidx[iidx];
 320
 321         /* Load limits for loop over neighbors */
 322         j_index_start    = jindex[iidx];
 323         j_index_end      = jindex[iidx+1];
 324
 325         /* Get outer coordinate index */
 326         inr              = iinr[iidx];
 327         i_coord_offset   = DIM*inr;
 328
 329         /* Load i particle coords and add shift vector */
 330         /* #if GEOMETRY_I == 'Particle' */
 331         gmx_mm_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 332         /* #elif GEOMETRY_I == 'Water3' */
 333         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 334                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 335         /* #elif GEOMETRY_I == 'Water4' */
 336         /*     #if 0 in PARTICLES_I                 */
 337         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 338                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 339         /*     #else                                */
 340         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 341                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 342         /*     #endif                               */
 343         /* #endif                                   */
 344
 345         /* #if 'Force' in KERNEL_VF */
 346         /*     #for I in PARTICLES_I */
 347         fix{I}             = _mm_setzero_ps();
 348         fiy{I}             = _mm_setzero_ps();
 349         fiz{I}             = _mm_setzero_ps();
 350         /*     #endfor */
 351         /* #endif */
 352
 353         /* ## For water we already preloaded parameters at the start of the kernel */
 354         /* #if not 'Water' in GEOMETRY_I */
 355         /* Load parameters for i particles */
 356         /*     #for I in PARTICLES_ELEC_I */
 357         iq{I}              = _mm_mul_ps(facel,_mm_load1_ps(charge+inr+{I}));
 358         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 359         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 360         isai{I}            = _mm_load1_ps(invsqrta+inr+{I});
 361         /*         #endif */
 362         /*     #endfor */
 363         /*     #for I in PARTICLES_VDW_I */
 364         vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 365         /*     #endfor */
 366         /* #endif */
 367
 368         /* #if 'Potential' in KERNEL_VF */
 369         /* Reset potential sums */
 370         /*     #if KERNEL_ELEC != 'None' */
 371         velecsum         = _mm_setzero_ps();
 372         /*     #endif */
 373         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 374         vgbsum           = _mm_setzero_ps();
 375         /*     #endif */
 376         /*     #if KERNEL_VDW != 'None' */
 377         vvdwsum          = _mm_setzero_ps();
 378         /*     #endif */
 379         /* #endif */
 380         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 381         dvdasum          = _mm_setzero_ps();
 382         /*     #endif */
 383
 384         /* #for ROUND in ['Loop','Epilogue'] */
 385
 386         /* #if ROUND =='Loop' */
 387         /* Start inner kernel loop */
 388         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 389         {
 390         /* ## First round is normal loop (next statement resets indentation) */
 391         /*     #if 0 */
 392         }
 393         /*     #endif */
 394         /* #else */
 395         if(jidx<j_index_end)
 396         {
 397         /* ## Second round is epilogue */
 398         /* #endif */
 399         /* #define INNERFLOPS 0 */
 400
 401             /* Get j neighbor index, and coordinate index */
 402             /* #if ROUND =='Loop' */
 403             jnrA             = jjnr[jidx];
 404             jnrB             = jjnr[jidx+1];
 405             jnrC             = jjnr[jidx+2];
 406             jnrD             = jjnr[jidx+3];
 407             /* #else */
 408             jnrlistA         = jjnr[jidx];
 409             jnrlistB         = jjnr[jidx+1];
 410             jnrlistC         = jjnr[jidx+2];
 411             jnrlistD         = jjnr[jidx+3];
 412             /* Sign of each element will be negative for non-real atoms.
 413              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 414              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 415              */
 416             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 417             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 418             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 419             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 420             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 421             /* #endif */
 422             j_coord_offsetA  = DIM*jnrA;
 423             j_coord_offsetB  = DIM*jnrB;
 424             j_coord_offsetC  = DIM*jnrC;
 425             j_coord_offsetD  = DIM*jnrD;
 426
 427             /* load j atom coordinates */
 428             /* #if GEOMETRY_J == 'Particle'             */
 429             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 430                                               x+j_coord_offsetC,x+j_coord_offsetD,
 431                                               &jx0,&jy0,&jz0);
 432             /* #elif GEOMETRY_J == 'Water3'             */
 433             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 434                                               x+j_coord_offsetC,x+j_coord_offsetD,
 435                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 436             /* #elif GEOMETRY_J == 'Water4'             */
 437             /*     #if 0 in PARTICLES_J                 */
 438             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 439                                               x+j_coord_offsetC,x+j_coord_offsetD,
 440                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 441                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 442             /*     #else                                */
 443             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 444                                               x+j_coord_offsetC+DIM,x+j_coord_offsetD+DIM,
 445                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 446             /*     #endif                               */
 447             /* #endif                                   */
 448
 449             /* Calculate displacement vector */
 450             /* #for I,J in PAIRS_IJ */
 451             dx{I}{J}             = _mm_sub_ps(ix{I},jx{J});
 452             dy{I}{J}             = _mm_sub_ps(iy{I},jy{J});
 453             dz{I}{J}             = _mm_sub_ps(iz{I},jz{J});
 454             /*     #define INNERFLOPS INNERFLOPS+3 */
 455             /* #endfor */
 456
 457             /* Calculate squared distance and things based on it */
 458             /* #for I,J in PAIRS_IJ */
 459             rsq{I}{J}            = gmx_mm_calc_rsq_ps(dx{I}{J},dy{I}{J},dz{I}{J});
 460             /*     #define INNERFLOPS INNERFLOPS+5 */
 461             /* #endfor */
 462
 463             /* #for I,J in PAIRS_IJ */
 464             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 465             rinv{I}{J}           = gmx_mm_invsqrt_ps(rsq{I}{J});
 466             /*         #define INNERFLOPS INNERFLOPS+5 */
 467             /*     #endif */
 468             /* #endfor */
 469
 470             /* #for I,J in PAIRS_IJ */
 471             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 472             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 473             rinvsq{I}{J}         = gmx_mm_inv_ps(rsq{I}{J});
 474             /*             #define INNERFLOPS INNERFLOPS+4 */
 475             /*         #else */
 476             rinvsq{I}{J}         = _mm_mul_ps(rinv{I}{J},rinv{I}{J});
 477             /*             #define INNERFLOPS INNERFLOPS+1 */
 478             /*         #endif */
 479             /*     #endif */
 480             /* #endfor */
 481
 482             /* #if not 'Water' in GEOMETRY_J */
 483             /* Load parameters for j particles */
 484             /*     #for J in PARTICLES_ELEC_J */
 485             jq{J}              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+{J},charge+jnrB+{J},
 486                                                               charge+jnrC+{J},charge+jnrD+{J});
 487             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 488             isaj{J}            = gmx_mm_load_4real_swizzle_ps(invsqrta+jnrA+{J},invsqrta+jnrB+{J},
 489                                                               invsqrta+jnrC+{J},invsqrta+jnrD+{J});
 490             /*         #endif */
 491             /*     #endfor */
 492             /*     #for J in PARTICLES_VDW_J */
 493             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 494             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 495             vdwjidx{J}C        = 2*vdwtype[jnrC+{J}];
 496             vdwjidx{J}D        = 2*vdwtype[jnrD+{J}];
 497             /*     #endfor */
 498             /* #endif */
 499
 500             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 501             /*     #for J in PARTICLES_J */
 502             fjx{J}             = _mm_setzero_ps();
 503             fjy{J}             = _mm_setzero_ps();
 504             fjz{J}             = _mm_setzero_ps();
 505             /*     #endfor */
 506             /* #endif */
 507
 508             /* #for I,J in PAIRS_IJ */
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 515             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 516             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 517             if (gmx_mm_any_lt(rsq{I}{J},rcutoff2))
 518             {
 519                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 520             }
 521             /*         #endif */
 522             /*         #define INNERFLOPS INNERFLOPS+1 */
 523             /*     #endif */
 524
 525             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 526             r{I}{J}              = _mm_mul_ps(rsq{I}{J},rinv{I}{J});
 527             /*         #if ROUND == 'Epilogue' */
 528             r{I}{J}              = _mm_andnot_ps(dummy_mask,r{I}{J});
 529             /*             #define INNERFLOPS INNERFLOPS+1 */
 530             /*         #endif */
 531             /*         #define INNERFLOPS INNERFLOPS+1 */
 532             /*     #endif */
 533
 534             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 535             /*     #if not 'Water' in GEOMETRY_J */
 536             /* Compute parameters for interactions between i and j atoms */
 537             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 538             qq{I}{J}             = _mm_mul_ps(iq{I},jq{J});
 539             /*             #define INNERFLOPS INNERFLOPS+1 */
 540             /*         #endif */
 541             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 542             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset{I}+vdwjidx{J}A,
 543                                          vdwparam+vdwioffset{I}+vdwjidx{J}B,
 544                                          vdwparam+vdwioffset{I}+vdwjidx{J}C,
 545                                          vdwparam+vdwioffset{I}+vdwjidx{J}D,
 546                                          &c6_{I}{J},&c12_{I}{J});
 547             /*         #endif */
 548             /*     #endif */
 549
 550             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 551             /* Calculate table index by multiplying r with table scale and truncate to integer */
 552             rt               = _mm_mul_ps(r{I}{J},vftabscale);
 553             vfitab           = _mm_cvttps_epi32(rt);
 554 #ifdef __XOP__
 555             vfeps            = _mm_frcz_ps(rt);
 556 #else
 557             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 558 #endif
 559             twovfeps         = _mm_add_ps(vfeps,vfeps);
 560             /*         #define INNERFLOPS INNERFLOPS+4                          */
 561             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW     */
 562             /*             ## 3 tables, 4 bytes per point: multiply index by 12 */
 563             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 564             /*         #elif 'Table' in KERNEL_ELEC                             */
 565             /*             ## 1 table, 4 bytes per point: multiply index by 4   */
 566             vfitab           = _mm_slli_epi32(vfitab,2);
 567             /*         #elif 'Table' in KERNEL_VDW                              */
 568             /*             ## 2 tables, 4 bytes per point: multiply index by 8  */
 569             vfitab           = _mm_slli_epi32(vfitab,3);
 570             /*         #endif                                                   */
 571             /*     #endif */
 572
 573             /*     ## ELECTROSTATIC INTERACTIONS */
 574             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 575
 576             /*         #if KERNEL_ELEC=='Coulomb' */
 577
 578             /* COULOMB ELECTROSTATICS */
 579             velec            = _mm_mul_ps(qq{I}{J},rinv{I}{J});
 580             /*             #define INNERFLOPS INNERFLOPS+1 */
 581             /*             #if 'Force' in KERNEL_VF */
 582             felec            = _mm_mul_ps(velec,rinvsq{I}{J});
 583             /*                 #define INNERFLOPS INNERFLOPS+2 */
 584             /*             #endif */
 585
 586             /*         #elif KERNEL_ELEC=='ReactionField' */
 587
 588             /* REACTION-FIELD ELECTROSTATICS */
 589             /*             #if 'Potential' in KERNEL_VF */
 590             velec            = _mm_mul_ps(qq{I}{J},_mm_sub_ps(_mm_macc_ps(krf,rsq{I}{J},rinv{I}{J}),crf));
 591             /*                 #define INNERFLOPS INNERFLOPS+4 */
 592             /*             #endif */
 593             /*             #if 'Force' in KERNEL_VF */
 594             felec            = _mm_mul_ps(qq{I}{J},_mm_msub_ps(rinv{I}{J},rinvsq{I}{J},krf2));
 595             /*                 #define INNERFLOPS INNERFLOPS+3 */
 596             /*             #endif */
 597
 598             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 599
 600             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 601             isaprod          = _mm_mul_ps(isai{I},isaj{J});
 602             gbqqfactor       = _mm_xor_ps(signbit,_mm_mul_ps(qq{I}{J},_mm_mul_ps(isaprod,gbinvepsdiff)));
 603             gbscale          = _mm_mul_ps(isaprod,gbtabscale);
 604             /*             #define INNERFLOPS INNERFLOPS+5 */
 605
 606             /* Calculate generalized born table index - this is a separate table from the normal one,
 607              * but we use the same procedure by multiplying r with scale and truncating to integer.
 608              */
 609             rt               = _mm_mul_ps(r{I}{J},gbscale);
 610             gbitab           = _mm_cvttps_epi32(rt);
 611 #ifdef __XOP__
 612             gbeps            = _mm_frcz_ps(rt);
 613 #else
 614             gbeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 615 #endif
 616             gbitab           = _mm_slli_epi32(gbitab,2);
 617
 618             Y                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,0) );
 619             F                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,1) );
 620             G                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,2) );
 621             H                = _mm_load_ps( gbtab + _mm_extract_epi32(gbitab,3) );
 622             _MM_TRANSPOSE4_PS(Y,F,G,H);
 623             Fp               = _mm_macc_ps(gbeps,_mm_macc_ps(gbeps,H,G),F);
 624             VV               = _mm_macc_ps(gbeps,Fp,Y);
 625             vgb              = _mm_mul_ps(gbqqfactor,VV);
 626             /*             #define INNERFLOPS INNERFLOPS+10 */
 627
 628             /*             #if 'Force' in KERNEL_VF */
 629             twogbeps         = _mm_add_ps(gbeps,gbeps);
 630             FF               = _mm_macc_ps(_mm_macc_ps(twogbeps,H,G),gbeps,Fp);
 631             fgb              = _mm_mul_ps(gbqqfactor,_mm_mul_ps(FF,gbscale));
 632             dvdatmp          = _mm_mul_ps(minushalf,_mm_macc_ps(fgb,r{I}{J},vgb));
 633             /*                 #if ROUND == 'Epilogue' */
 634             dvdatmp          = _mm_andnot_ps(dummy_mask,dvdatmp);
 635             /*                 #endif */
 636             dvdasum          = _mm_add_ps(dvdasum,dvdatmp);
 637             /*                 #if ROUND == 'Loop' */
 638             fjptrA           = dvda+jnrA;
 639             fjptrB           = dvda+jnrB;
 640             fjptrC           = dvda+jnrC;
 641             fjptrD           = dvda+jnrD;
 642             /*                 #else */
 643             /* 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. */
 644             fjptrA             = (jnrlistA>=0) ? dvda+jnrA : scratch;
 645             fjptrB             = (jnrlistB>=0) ? dvda+jnrB : scratch;
 646             fjptrC             = (jnrlistC>=0) ? dvda+jnrC : scratch;
 647             fjptrD             = (jnrlistD>=0) ? dvda+jnrD : scratch;
 648             /*                 #endif */
 649             gmx_mm_increment_4real_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,_mm_mul_ps(dvdatmp,_mm_mul_ps(isaj{J},isaj{J})));
 650             /*                 #define INNERFLOPS INNERFLOPS+13 */
 651             /*             #endif */
 652             velec            = _mm_mul_ps(qq{I}{J},rinv{I}{J});
 653             /*                 #define INNERFLOPS INNERFLOPS+1 */
 654             /*             #if 'Force' in KERNEL_VF */
 655             felec            = _mm_mul_ps(_mm_msub_ps(velec,rinv{I}{J},fgb),rinv{I}{J});
 656             /*                 #define INNERFLOPS INNERFLOPS+3 */
 657             /*             #endif */
 658
 659             /*         #elif KERNEL_ELEC=='Ewald' */
 660             /* EWALD ELECTROSTATICS */
 661
 662             /* Analytical PME correction */
 663             zeta2            = _mm_mul_ps(beta2,rsq{I}{J});
 664             /*             #if 'Force' in KERNEL_VF */
 665             rinv3            = _mm_mul_ps(rinvsq{I}{J},rinv{I}{J});
 666             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 667             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 668             felec            = _mm_mul_ps(qq{I}{J},felec);
 669             /*             #endif */
 670             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 671             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 672             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 673             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv{I}{J},sh_ewald));
 674             /*                 #else */
 675             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv{I}{J});
 676             /*                 #endif */
 677             velec            = _mm_mul_ps(qq{I}{J},velec);
 678             /*             #endif */
 679
 680             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 681
 682             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 683             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 684             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 685             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 686             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 687             _MM_TRANSPOSE4_PS(Y,F,G,H);
 688             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 689             /*             #define INNERFLOPS INNERFLOPS+4 */
 690             /*             #if 'Potential' in KERNEL_VF */
 691             VV               = _mm_macc_ps(vfeps,Fp,Y);
 692             velec            = _mm_mul_ps(qq{I}{J},VV);
 693             /*                 #define INNERFLOPS INNERFLOPS+3 */
 694             /*             #endif */
 695             /*             #if 'Force' in KERNEL_VF */
 696             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 697             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq{I}{J},FF),_mm_mul_ps(vftabscale,rinv{I}{J})));
 698             /*                 #define INNERFLOPS INNERFLOPS+7 */
 699             /*             #endif */
 700             /*         #endif */
 701             /*         ## End of check for electrostatics interaction forms */
 702             /*     #endif */
 703             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 704
 705             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 706
 707             /*         #if KERNEL_VDW=='LennardJones' */
 708
 709             /* LENNARD-JONES DISPERSION/REPULSION */
 710
 711             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 712             /*             #define INNERFLOPS INNERFLOPS+2 */
 713             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 714             vvdw6            = _mm_mul_ps(c6_{I}{J},rinvsix);
 715             vvdw12           = _mm_mul_ps(c12_{I}{J},_mm_mul_ps(rinvsix,rinvsix));
 716             /*                 #define INNERFLOPS INNERFLOPS+3 */
 717             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 718             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_{I}{J},_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 719                                           _mm_mul_ps( _mm_nmacc_ps(c6_{I}{J},sh_vdw_invrcut6,vvdw6),one_sixth));
 720             /*                     #define INNERFLOPS INNERFLOPS+8 */
 721             /*                 #else */
 722             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 723             /*                     #define INNERFLOPS INNERFLOPS+3 */
 724             /*                 #endif */
 725             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 726             /*                 #if 'Force' in KERNEL_VF */
 727             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq{I}{J});
 728             /*                     #define INNERFLOPS INNERFLOPS+2 */
 729             /*                 #endif */
 730             /*             #elif KERNEL_VF=='Force' */
 731             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 732             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_{I}{J},rinvsix,c6_{I}{J}),_mm_mul_ps(rinvsix,rinvsq{I}{J}));
 733             /*                 #define INNERFLOPS INNERFLOPS+4 */
 734             /*             #endif */
 735
 736             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 737
 738             /* CUBIC SPLINE TABLE DISPERSION */
 739             /*             #if 'Table' in KERNEL_ELEC */
 740             vfitab           = _mm_add_epi32(vfitab,ifour);
 741             /*             #endif                     */
 742             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 743             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 744             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 745             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 746             _MM_TRANSPOSE4_PS(Y,F,G,H);
 747             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 748             /*             #define INNERFLOPS INNERFLOPS+4 */
 749             /*             #if 'Potential' in KERNEL_VF */
 750             VV               = _mm_macc_ps(vfeps,Fp,Y);
 751             vvdw6            = _mm_mul_ps(c6_{I}{J},VV);
 752             /*                 #define INNERFLOPS INNERFLOPS+3 */
 753             /*             #endif */
 754             /*             #if 'Force' in KERNEL_VF */
 755             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 756             fvdw6            = _mm_mul_ps(c6_{I}{J},FF);
 757             /*                 #define INNERFLOPS INNERFLOPS+4 */
 758             /*             #endif */
 759
 760             /* CUBIC SPLINE TABLE REPULSION */
 761             vfitab           = _mm_add_epi32(vfitab,ifour);
 762             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 763             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 764             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 765             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 766             _MM_TRANSPOSE4_PS(Y,F,G,H);
 767             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 768             /*             #define INNERFLOPS INNERFLOPS+4 */
 769             /*             #if 'Potential' in KERNEL_VF */
 770             VV               = _mm_macc_ps(vfeps,Fp,Y);
 771             vvdw12           = _mm_mul_ps(c12_{I}{J},VV);
 772             /*                 #define INNERFLOPS INNERFLOPS+3 */
 773             /*             #endif */
 774             /*             #if 'Force' in KERNEL_VF */
 775             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 776             fvdw12           = _mm_mul_ps(c12_{I}{J},FF);
 777             /*                 #define INNERFLOPS INNERFLOPS+5 */
 778             /*             #endif */
 779             /*             #if 'Potential' in KERNEL_VF */
 780             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 781             /*                 #define INNERFLOPS INNERFLOPS+1 */
 782             /*             #endif */
 783             /*             #if 'Force' in KERNEL_VF */
 784             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv{I}{J})));
 785             /*                 #define INNERFLOPS INNERFLOPS+4 */
 786             /*             #endif */
 787             /*         #endif */
 788             /*         ## End of check for vdw interaction forms */
 789             /*     #endif */
 790             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 791
 792             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 793             d                = _mm_sub_ps(r{I}{J},rswitch);
 794             d                = _mm_max_ps(d,_mm_setzero_ps());
 795             d2               = _mm_mul_ps(d,d);
 796             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_macc_ps(d,_mm_macc_ps(d,swV5,swV4),swV3))));
 797             /*         #define INNERFLOPS INNERFLOPS+10 */
 798
 799             /*         #if 'Force' in KERNEL_VF */
 800             dsw              = _mm_mul_ps(d2,_mm_macc_ps(d,_mm_macc_ps(d,swF4,swF3),swF2));
 801             /*             #define INNERFLOPS INNERFLOPS+5 */
 802             /*         #endif */
 803
 804             /* Evaluate switch function */
 805             /*         #if 'Force' in KERNEL_VF */
 806             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 807             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 808             felec            = _mm_msub_ps( felec,sw , _mm_mul_ps(rinv{I}{J},_mm_mul_ps(velec,dsw)) );
 809             /*                 #define INNERFLOPS INNERFLOPS+4 */
 810             /*             #endif */
 811             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 812             fvdw             = _mm_msub_ps( fvdw,sw , _mm_mul_ps(rinv{I}{J},_mm_mul_ps(vvdw,dsw)) );
 813             /*                 #define INNERFLOPS INNERFLOPS+4 */
 814             /*             #endif */
 815             /*         #endif */
 816             /*         #if 'Potential' in KERNEL_VF */
 817             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 818             velec            = _mm_mul_ps(velec,sw);
 819             /*                 #define INNERFLOPS INNERFLOPS+1 */
 820             /*             #endif */
 821             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 822             vvdw             = _mm_mul_ps(vvdw,sw);
 823             /*                 #define INNERFLOPS INNERFLOPS+1 */
 824             /*             #endif */
 825             /*         #endif */
 826             /*     #endif */
 827             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 828             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 829             cutoff_mask      = _mm_cmplt_ps(rsq{I}{J},rcutoff2);
 830             /*         #define INNERFLOPS INNERFLOPS+1 */
 831             /*     #endif */
 832
 833             /*     #if 'Potential' in KERNEL_VF */
 834             /* Update potential sum for this i atom from the interaction with this j atom. */
 835             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 836             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 837             velec            = _mm_and_ps(velec,cutoff_mask);
 838             /*                 #define INNERFLOPS INNERFLOPS+1 */
 839             /*             #endif                                       */
 840             /*             #if ROUND == 'Epilogue' */
 841             velec            = _mm_andnot_ps(dummy_mask,velec);
 842             /*             #endif */
 843             velecsum         = _mm_add_ps(velecsum,velec);
 844             /*             #define INNERFLOPS INNERFLOPS+1 */
 845             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 846             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 847             vgb              = _mm_and_ps(vgb,cutoff_mask);
 848             /*                 #define INNERFLOPS INNERFLOPS+1 */
 849             /*             #endif                                       */
 850             /*             #if ROUND == 'Epilogue' */
 851             vgb              = _mm_andnot_ps(dummy_mask,vgb);
 852             /*             #endif */
 853             vgbsum           = _mm_add_ps(vgbsum,vgb);
 854             /*                 #define INNERFLOPS INNERFLOPS+1 */
 855             /*             #endif */
 856             /*         #endif */
 857             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 858             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 859             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 860             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 861             /*                 #define INNERFLOPS INNERFLOPS+1 */
 862             /*             #endif                                       */
 863             /*             #if ROUND == 'Epilogue' */
 864             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 865             /*             #endif */
 866             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 867             /*             #define INNERFLOPS INNERFLOPS+1 */
 868             /*         #endif */
 869             /*     #endif */
 870
 871             /*     #if 'Force' in KERNEL_VF */
 872
 873             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 874             fscal            = _mm_add_ps(felec,fvdw);
 875             /*             #define INNERFLOPS INNERFLOPS+1 */
 876             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 877             fscal            = felec;
 878             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 879             fscal            = fvdw;
 880             /*        #endif */
 881
 882             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 883             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 884             fscal            = _mm_and_ps(fscal,cutoff_mask);
 885             /*                 #define INNERFLOPS INNERFLOPS+1 */
 886             /*             #endif                                       */
 887
 888             /*             #if ROUND == 'Epilogue' */
 889             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 890             /*             #endif */
 891
 892             /* ## Construction of vectorial force built into FMA instructions now */
 893             /* #define INNERFLOPS INNERFLOPS+3      */
 894
 895              /* Update vectorial force */
 896             fix{I}             = _mm_macc_ps(dx{I}{J},fscal,fix{I});
 897             fiy{I}             = _mm_macc_ps(dy{I}{J},fscal,fiy{I});
 898             fiz{I}             = _mm_macc_ps(dz{I}{J},fscal,fiz{I});
 899             /*             #define INNERFLOPS INNERFLOPS+6 */
 900
 901             /* #if GEOMETRY_I == 'Particle'             */
 902             /*     #if ROUND == 'Loop' */
 903             fjptrA             = f+j_coord_offsetA;
 904             fjptrB             = f+j_coord_offsetB;
 905             fjptrC             = f+j_coord_offsetC;
 906             fjptrD             = f+j_coord_offsetD;
 907             /*     #else */
 908             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 909             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 910             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 911             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 912             /*     #endif */
 913             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 914                                                    _mm_mul_ps(dx{I}{J},fscal),
 915                                                    _mm_mul_ps(dy{I}{J},fscal),
 916                                                    _mm_mul_ps(dz{I}{J},fscal));
 917             /*     #define INNERFLOPS INNERFLOPS+3      */
 918             /* #else                                    */
 919             fjx{J}             = _mm_macc_ps(dx{I}{J},fscal,fjx{J});
 920             fjy{J}             = _mm_macc_ps(dy{I}{J},fscal,fjy{J});
 921             fjz{J}             = _mm_macc_ps(dz{I}{J},fscal,fjz{J});
 922             /*     #define INNERFLOPS INNERFLOPS+3      */
 923             /* #endif                                   */
 924
 925             /*     #endif */
 926
 927             /*     ## Note special check for TIP4P-TIP4P. Since we are cutting of all hydrogen interactions we also cut the LJ-only O-O interaction */
 928             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] or (GEOMETRY_I=='Water4' and GEOMETRY_J=='Water4' and 'exactcutoff' in INTERACTION_FLAGS[1][1]) */
 929             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
 930             {
 931                 /*     #endif */
 932             }
 933             /*     #endif */
 934             /*    ## End of check for the interaction being outside the cutoff */
 935
 936             /* #endfor */
 937             /* ## End of loop over i-j interaction pairs */
 938
 939             /* #if GEOMETRY_I != 'Particle' */
 940             /*     #if ROUND == 'Loop' */
 941             fjptrA             = f+j_coord_offsetA;
 942             fjptrB             = f+j_coord_offsetB;
 943             fjptrC             = f+j_coord_offsetC;
 944             fjptrD             = f+j_coord_offsetD;
 945             /*     #else */
 946             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 947             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 948             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 949             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 950             /*     #endif */
 951             /* #endif */
 952
 953             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
 954             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 955             /* #elif GEOMETRY_J == 'Water3'               */
 956             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 957                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 958             /*     #define INNERFLOPS INNERFLOPS+9      */
 959             /* #elif GEOMETRY_J == 'Water4'             */
 960             /*     #if 0 in PARTICLES_J                 */
 961             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 962                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 963                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 964             /*     #define INNERFLOPS INNERFLOPS+12     */
 965             /*     #else                                */
 966             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA+DIM,fjptrB+DIM,fjptrC+DIM,fjptrD+DIM,
 967                                                    fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 968             /*     #define INNERFLOPS INNERFLOPS+9      */
 969             /*     #endif                               */
 970             /* #endif                                   */
 971
 972             /* Inner loop uses {INNERFLOPS} flops */
 973         }
 974
 975         /* #endfor */
 976
 977         /* End of innermost loop */
 978
 979         /* #if 'Force' in KERNEL_VF */
 980         /*     #if GEOMETRY_I == 'Particle'            */
 981         gmx_mm_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 982                                               f+i_coord_offset,fshift+i_shift_offset);
 983         /*         #define OUTERFLOPS OUTERFLOPS+6     */
 984         /*     #elif GEOMETRY_I == 'Water3'            */
 985         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 986                                               f+i_coord_offset,fshift+i_shift_offset);
 987         /*         #define OUTERFLOPS OUTERFLOPS+18    */
 988         /*     #elif GEOMETRY_I == 'Water4'            */
 989         /*         #if 0 in PARTICLES_I                */
 990         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 991                                               f+i_coord_offset,fshift+i_shift_offset);
 992         /*             #define OUTERFLOPS OUTERFLOPS+24    */
 993         /*         #else                               */
 994         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 995                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 996         /*             #define OUTERFLOPS OUTERFLOPS+18    */
 997         /*         #endif                              */
 998         /*     #endif                                  */
 999         /* #endif                                      */
1000
1001         /* #if 'Potential' in KERNEL_VF */
1002         ggid                        = gid[iidx];
1003         /* Update potential energies */
1004         /*     #if KERNEL_ELEC != 'None' */
1005         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1006         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1007         /*     #endif */
1008         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
1009         gmx_mm_update_1pot_ps(vgbsum,kernel_data->energygrp_polarization+ggid);
1010         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1011         /*     #endif */
1012         /*     #if KERNEL_VDW != 'None' */
1013         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1014         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1015         /*     #endif */
1016         /* #endif */
1017         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1018         dvdasum = _mm_mul_ps(dvdasum, _mm_mul_ps(isai{I},isai{I}));
1019         gmx_mm_update_1pot_ps(dvdasum,dvda+inr);
1020         /*     #endif */
1021
1022         /* Increment number of inner iterations */
1023         inneriter                  += j_index_end - j_index_start;
1024
1025         /* Outer loop uses {OUTERFLOPS} flops */
1026     }
1027
1028     /* Increment number of outer iterations */
1029     outeriter        += nri;
1030
1031     /* Update outer/inner flops */
1032     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1033     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1034     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1035     /* #if GEOMETRY_I == 'Water3'            */
1036     /*     #define ISUFFIX '_W3'             */
1037     /* #elif GEOMETRY_I == 'Water4'          */
1038     /*     #define ISUFFIX '_W4'             */
1039     /* #else                                 */
1040     /*     #define ISUFFIX ''                */
1041     /* #endif                                */
1042     /* #if GEOMETRY_J == 'Water3'            */
1043     /*     #define JSUFFIX 'W3'              */
1044     /* #elif GEOMETRY_J == 'Water4'          */
1045     /*     #define JSUFFIX 'W4'              */
1046     /* #else                                 */
1047     /*     #define JSUFFIX ''                */
1048     /* #endif                                */
1049     /* #if 'PotentialAndForce' in KERNEL_VF  */
1050     /*     #define VFSUFFIX  '_VF'           */
1051     /* #elif 'Potential' in KERNEL_VF        */
1052     /*     #define VFSUFFIX '_V'             */
1053     /* #else                                 */
1054     /*     #define VFSUFFIX '_F'             */
1055     /* #endif                                */
1056
1057     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1058     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1059     /* #elif KERNEL_ELEC != 'None' */
1060     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1061     /* #else */
1062     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1063     /* #endif  */
1064 }