src/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_template_sse2_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_sse2_double.h"
  51 #include "kernelutil_x86_sse2_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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
 114     int              j_coord_offsetA,j_coord_offsetB;
 115     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 116     real             rcutoff_scalar;
 117     real             *shiftvec,*fshift,*x,*f;
 118     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 119     /* #for I in PARTICLES_I */
 120     int              vdwioffset{I};
 121     __m128d          ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
 122     /* #endfor */
 123     /* #for J in PARTICLES_J */
 124     int              vdwjidx{J}A,vdwjidx{J}B;
 125     __m128d          jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
 126     /* #endfor */
 127     /* #for I,J in PAIRS_IJ */
 128     __m128d          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};
 129     /* #endfor */
 130     /* #if KERNEL_ELEC != 'None' */
 131     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 132     real             *charge;
 133     /* #endif */
 134     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 135     __m128i          gbitab;
 136     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 137     __m128d          minushalf = _mm_set1_pd(-0.5);
 138     real             *invsqrta,*dvda,*gbtab;
 139     /* #endif */
 140     /* #if KERNEL_VDW != 'None' */
 141     int              nvdwtype;
 142     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 143     int              *vdwtype;
 144     real             *vdwparam;
 145     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 146     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 147     /* #endif */
 148     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 149     __m128i          vfitab;
 150     __m128i          ifour       = _mm_set1_epi32(4);
 151     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 152     real             *vftab;
 153     /* #endif */
 154     /* #if 'Ewald' in KERNEL_ELEC */
 155     __m128i          ewitab;
 156     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 157     real             *ewtab;
 158     /* #endif */
 159     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 160     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 161     real             rswitch_scalar,d_scalar;
 162     /* #endif */
 163     __m128d          dummy_mask,cutoff_mask;
 164     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 165     __m128d          one     = _mm_set1_pd(1.0);
 166     __m128d          two     = _mm_set1_pd(2.0);
 167     x                = xx[0];
 168     f                = ff[0];
 169
 170     nri              = nlist->nri;
 171     iinr             = nlist->iinr;
 172     jindex           = nlist->jindex;
 173     jjnr             = nlist->jjnr;
 174     shiftidx         = nlist->shift;
 175     gid              = nlist->gid;
 176     shiftvec         = fr->shift_vec[0];
 177     fshift           = fr->fshift[0];
 178     /* #if KERNEL_ELEC != 'None' */
 179     facel            = _mm_set1_pd(fr->epsfac);
 180     charge           = mdatoms->chargeA;
 181     /*     #if 'ReactionField' in KERNEL_ELEC */
 182     krf              = _mm_set1_pd(fr->ic->k_rf);
 183     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 184     crf              = _mm_set1_pd(fr->ic->c_rf);
 185     /*     #endif */
 186     /* #endif */
 187     /* #if KERNEL_VDW != 'None' */
 188     nvdwtype         = fr->ntype;
 189     vdwparam         = fr->nbfp;
 190     vdwtype          = mdatoms->typeA;
 191     /* #endif */
 192
 193     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 194     vftab            = kernel_data->table_elec_vdw->data;
 195     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 196     /* #elif 'Table' in KERNEL_ELEC */
 197     vftab            = kernel_data->table_elec->data;
 198     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 199     /* #elif 'Table' in KERNEL_VDW */
 200     vftab            = kernel_data->table_vdw->data;
 201     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 202     /* #endif */
 203
 204     /* #if 'Ewald' in KERNEL_ELEC */
 205     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 206     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 207     ewtab            = fr->ic->tabq_coul_F;
 208     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 209     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 210     /*     #else */
 211     ewtab            = fr->ic->tabq_coul_FDV0;
 212     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 213     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 214      /*     #endif */
 215     /* #endif */
 216
 217     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 218     invsqrta         = fr->invsqrta;
 219     dvda             = fr->dvda;
 220     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 221     gbtab            = fr->gbtab.data;
 222     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 223     /* #endif */
 224
 225     /* #if 'Water' in GEOMETRY_I */
 226     /* Setup water-specific parameters */
 227     inr              = nlist->iinr[0];
 228     /*     #for I in PARTICLES_ELEC_I */
 229     iq{I}              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+{I}]));
 230     /*     #endfor */
 231     /*     #for I in PARTICLES_VDW_I */
 232     vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 233     /*     #endfor */
 234     /* #endif */
 235
 236     /* #if 'Water' in GEOMETRY_J */
 237     /*     #for J in PARTICLES_ELEC_J */
 238     jq{J}              = _mm_set1_pd(charge[inr+{J}]);
 239     /*     #endfor */
 240     /*     #for J in PARTICLES_VDW_J */
 241     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 242     /*     #endfor */
 243     /*     #for I,J in PAIRS_IJ */
 244     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 245     qq{I}{J}             = _mm_mul_pd(iq{I},jq{J});
 246     /*         #endif */
 247     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 248     c6_{I}{J}            = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 249     c12_{I}{J}           = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 250     /*         #endif */
 251     /*     #endfor */
 252     /* #endif */
 253
 254     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 255     /*     #if KERNEL_ELEC!='None' */
 256     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 257     rcutoff_scalar   = fr->rcoulomb;
 258     /*     #else */
 259     rcutoff_scalar   = fr->rvdw;
 260     /*     #endif */
 261     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 262     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 263     /* #endif */
 264
 265     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 266     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 267     rvdw             = _mm_set1_pd(fr->rvdw);
 268     /* #endif */
 269
 270     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 271     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 272     rswitch_scalar   = fr->rcoulomb_switch;
 273     rswitch          = _mm_set1_pd(rswitch_scalar);
 274     /*     #else */
 275     rswitch_scalar   = fr->rvdw_switch;
 276     rswitch          = _mm_set1_pd(rswitch_scalar);
 277     /*     #endif */
 278     /* Setup switch parameters */
 279     d_scalar         = rcutoff_scalar-rswitch_scalar;
 280     d                = _mm_set1_pd(d_scalar);
 281     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 282     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 283     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 284     /*     #if 'Force' in KERNEL_VF */
 285     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 286     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 287     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 288     /*     #endif */
 289     /* #endif */
 290
 291     /* Avoid stupid compiler warnings */
 292     jnrA = jnrB = 0;
 293     j_coord_offsetA = 0;
 294     j_coord_offsetB = 0;
 295
 296     /* ## Keep track of the floating point operations we issue for reporting! */
 297     /* #define OUTERFLOPS 0 */
 298     outeriter        = 0;
 299     inneriter        = 0;
 300
 301     /* Start outer loop over neighborlists */
 302     for(iidx=0; iidx<nri; iidx++)
 303     {
 304         /* Load shift vector for this list */
 305         i_shift_offset   = DIM*shiftidx[iidx];
 306
 307         /* Load limits for loop over neighbors */
 308         j_index_start    = jindex[iidx];
 309         j_index_end      = jindex[iidx+1];
 310
 311         /* Get outer coordinate index */
 312         inr              = iinr[iidx];
 313         i_coord_offset   = DIM*inr;
 314
 315         /* Load i particle coords and add shift vector */
 316         /* #if GEOMETRY_I == 'Particle' */
 317         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 318         /* #elif GEOMETRY_I == 'Water3' */
 319         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 320                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 321         /* #elif GEOMETRY_I == 'Water4' */
 322         /*     #if 0 in PARTICLES_I                 */
 323         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 324                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 325         /*     #else                                */
 326         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 327                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 328         /*     #endif                               */
 329         /* #endif                                   */
 330
 331         /* #if 'Force' in KERNEL_VF */
 332         /*     #for I in PARTICLES_I */
 333         fix{I}             = _mm_setzero_pd();
 334         fiy{I}             = _mm_setzero_pd();
 335         fiz{I}             = _mm_setzero_pd();
 336         /*     #endfor */
 337         /* #endif */
 338
 339         /* ## For water we already preloaded parameters at the start of the kernel */
 340         /* #if not 'Water' in GEOMETRY_I */
 341         /* Load parameters for i particles */
 342         /*     #for I in PARTICLES_ELEC_I */
 343         iq{I}              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+{I}));
 344         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 345         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 346         isai{I}            = _mm_load1_pd(invsqrta+inr+{I});
 347         /*         #endif */
 348         /*     #endfor */
 349         /*     #for I in PARTICLES_VDW_I */
 350         vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 351         /*     #endfor */
 352         /* #endif */
 353
 354         /* #if 'Potential' in KERNEL_VF */
 355         /* Reset potential sums */
 356         /*     #if KERNEL_ELEC != 'None' */
 357         velecsum         = _mm_setzero_pd();
 358         /*     #endif */
 359         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 360         vgbsum           = _mm_setzero_pd();
 361         /*     #endif */
 362         /*     #if KERNEL_VDW != 'None' */
 363         vvdwsum          = _mm_setzero_pd();
 364         /*     #endif */
 365         /* #endif */
 366         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 367         dvdasum          = _mm_setzero_pd();
 368         /*     #endif */
 369
 370         /* #for ROUND in ['Loop','Epilogue'] */
 371
 372         /* #if ROUND =='Loop' */
 373         /* Start inner kernel loop */
 374         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 375         {
 376         /* ## First round is normal loop (next statement resets indentation) */
 377         /*     #if 0 */
 378         }
 379         /*     #endif */
 380         /* #else */
 381         if(jidx<j_index_end)
 382         {
 383         /* ## Second round is epilogue */
 384         /* #endif */
 385         /* #define INNERFLOPS 0 */
 386
 387             /* #if ROUND =='Loop' */
 388             /* Get j neighbor index, and coordinate index */
 389             jnrA             = jjnr[jidx];
 390             jnrB             = jjnr[jidx+1];
 391             j_coord_offsetA  = DIM*jnrA;
 392             j_coord_offsetB  = DIM*jnrB;
 393
 394             /* load j atom coordinates */
 395             /*     #if GEOMETRY_J == 'Particle'             */
 396             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 397                                               &jx0,&jy0,&jz0);
 398             /*     #elif GEOMETRY_J == 'Water3'             */
 399             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 400                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 401             /*     #elif GEOMETRY_J == 'Water4'             */
 402             /*         #if 0 in PARTICLES_J                 */
 403             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 404                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 405                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 406             /*         #else                                */
 407             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 408                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 409             /*         #endif                               */
 410             /*     #endif                                   */
 411             /* #else */
 412             jnrA             = jjnr[jidx];
 413             j_coord_offsetA  = DIM*jnrA;
 414
 415             /* load j atom coordinates */
 416             /*     #if GEOMETRY_J == 'Particle'             */
 417             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 418                                               &jx0,&jy0,&jz0);
 419             /*     #elif GEOMETRY_J == 'Water3'             */
 420             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 421                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 422             /*     #elif GEOMETRY_J == 'Water4'             */
 423             /*         #if 0 in PARTICLES_J                 */
 424             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 425                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 426                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 427             /*         #else                                */
 428             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
 429                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 430             /*         #endif                               */
 431             /*     #endif                                   */
 432             /* #endif */
 433
 434             /* Calculate displacement vector */
 435             /* #for I,J in PAIRS_IJ */
 436             dx{I}{J}             = _mm_sub_pd(ix{I},jx{J});
 437             dy{I}{J}             = _mm_sub_pd(iy{I},jy{J});
 438             dz{I}{J}             = _mm_sub_pd(iz{I},jz{J});
 439             /*     #define INNERFLOPS INNERFLOPS+3 */
 440             /* #endfor */
 441
 442             /* Calculate squared distance and things based on it */
 443             /* #for I,J in PAIRS_IJ */
 444             rsq{I}{J}            = gmx_mm_calc_rsq_pd(dx{I}{J},dy{I}{J},dz{I}{J});
 445             /*     #define INNERFLOPS INNERFLOPS+5 */
 446             /* #endfor */
 447
 448             /* #for I,J in PAIRS_IJ */
 449             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 450             rinv{I}{J}           = gmx_mm_invsqrt_pd(rsq{I}{J});
 451             /*         #define INNERFLOPS INNERFLOPS+5 */
 452             /*     #endif */
 453             /* #endfor */
 454
 455             /* #for I,J in PAIRS_IJ */
 456             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 457             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 458             rinvsq{I}{J}         = gmx_mm_inv_pd(rsq{I}{J});
 459             /*             #define INNERFLOPS INNERFLOPS+4 */
 460             /*         #else */
 461             rinvsq{I}{J}         = _mm_mul_pd(rinv{I}{J},rinv{I}{J});
 462             /*             #define INNERFLOPS INNERFLOPS+1 */
 463             /*         #endif */
 464             /*     #endif */
 465             /* #endfor */
 466
 467             /* #if not 'Water' in GEOMETRY_J */
 468             /* Load parameters for j particles */
 469             /*     #for J in PARTICLES_ELEC_J */
 470             /*         #if ROUND =='Loop' */
 471             jq{J}              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+{J},charge+jnrB+{J});
 472             /*         #else */
 473             jq{J}              = _mm_load_sd(charge+jnrA+{J});
 474             /*         #endif */
 475             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 476             /*             #if ROUND =='Loop' */
 477             isaj{J}            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+{J},invsqrta+jnrB+{J});
 478             /*             #else */
 479             isaj{J}            = _mm_load_sd(invsqrta+jnrA+{J});
 480             /*             #endif */
 481             /*         #endif */
 482             /*     #endfor */
 483             /*     #for J in PARTICLES_VDW_J */
 484             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 485             /*         #if ROUND =='Loop' */
 486             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 487             /*         #endif */
 488             /*     #endfor */
 489             /* #endif */
 490
 491             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 492             /*     #for J in PARTICLES_J */
 493             fjx{J}             = _mm_setzero_pd();
 494             fjy{J}             = _mm_setzero_pd();
 495             fjz{J}             = _mm_setzero_pd();
 496             /*     #endfor */
 497             /* #endif */
 498
 499             /* #for I,J in PAIRS_IJ */
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 506             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 507             if (gmx_mm_any_lt(rsq{I}{J},rcutoff2))
 508             {
 509                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 510             }
 511             /*         #endif */
 512             /*         #define INNERFLOPS INNERFLOPS+1 */
 513             /*     #endif */
 514
 515             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 516             r{I}{J}              = _mm_mul_pd(rsq{I}{J},rinv{I}{J});
 517              /*         #define INNERFLOPS INNERFLOPS+1 */
 518             /*     #endif */
 519
 520             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 521             /*     #if not 'Water' in GEOMETRY_J */
 522             /* Compute parameters for interactions between i and j atoms */
 523             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 524             qq{I}{J}             = _mm_mul_pd(iq{I},jq{J});
 525             /*             #define INNERFLOPS INNERFLOPS+1 */
 526             /*         #endif */
 527             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 528             /*             #if ROUND == 'Loop' */
 529             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset{I}+vdwjidx{J}A,
 530                                          vdwparam+vdwioffset{I}+vdwjidx{J}B,&c6_{I}{J},&c12_{I}{J});
 531             /*             #else */
 532             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset{I}+vdwjidx{J}A,&c6_{I}{J},&c12_{I}{J});
 533             /*             #endif */
 534             /*         #endif */
 535             /*     #endif */
 536
 537             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 538             /* Calculate table index by multiplying r with table scale and truncate to integer */
 539             rt               = _mm_mul_pd(r{I}{J},vftabscale);
 540             vfitab           = _mm_cvttpd_epi32(rt);
 541             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 542             /*         #define INNERFLOPS INNERFLOPS+4                          */
 543             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW     */
 544             /*             ## 3 tables, 4 data per point: multiply index by 12 */
 545             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 546             /*         #elif 'Table' in KERNEL_ELEC                             */
 547             /*             ## 1 table, 4 data per point: multiply index by 4   */
 548             vfitab           = _mm_slli_epi32(vfitab,2);
 549             /*         #elif 'Table' in KERNEL_VDW                              */
 550             /*             ## 2 tables, 4 data per point: multiply index by 8  */
 551             vfitab           = _mm_slli_epi32(vfitab,3);
 552             /*         #endif                                                   */
 553             /*     #endif */
 554
 555             /*     ## ELECTROSTATIC INTERACTIONS */
 556             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 557
 558             /*         #if KERNEL_ELEC=='Coulomb' */
 559
 560             /* COULOMB ELECTROSTATICS */
 561             velec            = _mm_mul_pd(qq{I}{J},rinv{I}{J});
 562             /*             #define INNERFLOPS INNERFLOPS+1 */
 563             /*             #if 'Force' in KERNEL_VF */
 564             felec            = _mm_mul_pd(velec,rinvsq{I}{J});
 565             /*                 #define INNERFLOPS INNERFLOPS+2 */
 566             /*             #endif */
 567
 568             /*         #elif KERNEL_ELEC=='ReactionField' */
 569
 570             /* REACTION-FIELD ELECTROSTATICS */
 571             /*             #if 'Potential' in KERNEL_VF */
 572             velec            = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_add_pd(rinv{I}{J},_mm_mul_pd(krf,rsq{I}{J})),crf));
 573             /*                 #define INNERFLOPS INNERFLOPS+4 */
 574             /*             #endif */
 575             /*             #if 'Force' in KERNEL_VF */
 576             felec            = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_mul_pd(rinv{I}{J},rinvsq{I}{J}),krf2));
 577             /*                 #define INNERFLOPS INNERFLOPS+3 */
 578             /*             #endif */
 579
 580             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 581
 582             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 583             isaprod          = _mm_mul_pd(isai{I},isaj{J});
 584             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq{I}{J},_mm_mul_pd(isaprod,gbinvepsdiff)));
 585             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 586             /*             #define INNERFLOPS INNERFLOPS+5 */
 587
 588             /* Calculate generalized born table index - this is a separate table from the normal one,
 589              * but we use the same procedure by multiplying r with scale and truncating to integer.
 590              */
 591             rt               = _mm_mul_pd(r{I}{J},gbscale);
 592             gbitab           = _mm_cvttpd_epi32(rt);
 593             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 594             gbitab           = _mm_slli_epi32(gbitab,2);
 595
 596             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 597             /*             #if ROUND == 'Loop' */
 598             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 599             /*             #else */
 600             F                = _mm_setzero_pd();
 601             /*             #endif */
 602             GMX_MM_TRANSPOSE2_PD(Y,F);
 603             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 604             /*             #if ROUND == 'Loop' */
 605             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 606             /*             #else */
 607             H                = _mm_setzero_pd();
 608             /*             #endif */
 609             GMX_MM_TRANSPOSE2_PD(G,H);
 610             Heps             = _mm_mul_pd(gbeps,H);
 611             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 612             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 613             vgb              = _mm_mul_pd(gbqqfactor,VV);
 614             /*             #define INNERFLOPS INNERFLOPS+10 */
 615
 616             /*             #if 'Force' in KERNEL_VF */
 617             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 618             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 619             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r{I}{J})));
 620             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 621             /*             #if ROUND == 'Loop' */
 622             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj{J},isaj{J})));
 623             /*             #else */
 624             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj{J},isaj{J})));
 625             /*             #endif */
 626             /*                 #define INNERFLOPS INNERFLOPS+13 */
 627             /*             #endif */
 628             velec            = _mm_mul_pd(qq{I}{J},rinv{I}{J});
 629             /*                 #define INNERFLOPS INNERFLOPS+1 */
 630             /*             #if 'Force' in KERNEL_VF */
 631             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv{I}{J}),fgb),rinv{I}{J});
 632             /*                 #define INNERFLOPS INNERFLOPS+3 */
 633             /*             #endif */
 634
 635             /*         #elif KERNEL_ELEC=='Ewald' */
 636             /* EWALD ELECTROSTATICS */
 637
 638             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 639             ewrt             = _mm_mul_pd(r{I}{J},ewtabscale);
 640             ewitab           = _mm_cvttpd_epi32(ewrt);
 641             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 642             /*             #define INNERFLOPS INNERFLOPS+4 */
 643             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 644             ewitab           = _mm_slli_epi32(ewitab,2);
 645             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 646             /*                 #if ROUND == 'Loop' */
 647             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 648             /*                 #else */
 649             ewtabD           = _mm_setzero_pd();
 650             /*                 #endif */
 651             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 652             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 653             /*                 #if ROUND == 'Loop' */
 654             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 655             /*                 #else */
 656             ewtabFn          = _mm_setzero_pd();
 657             /*                 #endif */
 658             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 659             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 660             /*                 #define INNERFLOPS INNERFLOPS+2 */
 661             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 662             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 663             velec            = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_sub_pd(rinv{I}{J},sh_ewald),velec));
 664             /*                     #define INNERFLOPS INNERFLOPS+7 */
 665             /*                 #else */
 666             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 667             velec            = _mm_mul_pd(qq{I}{J},_mm_sub_pd(rinv{I}{J},velec));
 668             /*                     #define INNERFLOPS INNERFLOPS+6 */
 669             /*                 #endif */
 670             /*                 #if 'Force' in KERNEL_VF */
 671             felec            = _mm_mul_pd(_mm_mul_pd(qq{I}{J},rinv{I}{J}),_mm_sub_pd(rinvsq{I}{J},felec));
 672             /*                      #define INNERFLOPS INNERFLOPS+3 */
 673             /*                 #endif */
 674             /*             #elif KERNEL_VF=='Force' */
 675             /*                 #if ROUND == 'Loop' */
 676             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 677                                          &ewtabF,&ewtabFn);
 678             /*                 #else */
 679             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 680             /*                 #endif */
 681             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 682             felec            = _mm_mul_pd(_mm_mul_pd(qq{I}{J},rinv{I}{J}),_mm_sub_pd(rinvsq{I}{J},felec));
 683             /*                 #define INNERFLOPS INNERFLOPS+7 */
 684             /*             #endif */
 685
 686             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 687
 688             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 689             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 690             /*             #if ROUND == 'Loop' */
 691             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 692             /*             #else */
 693             F                = _mm_setzero_pd();
 694             /*             #endif */
 695             GMX_MM_TRANSPOSE2_PD(Y,F);
 696             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 697             /*             #if ROUND == 'Loop' */
 698             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 699             /*             #else */
 700             H                = _mm_setzero_pd();
 701             /*             #endif */
 702             GMX_MM_TRANSPOSE2_PD(G,H);
 703             Heps             = _mm_mul_pd(vfeps,H);
 704             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 705             /*             #define INNERFLOPS INNERFLOPS+4 */
 706             /*             #if 'Potential' in KERNEL_VF */
 707             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 708             velec            = _mm_mul_pd(qq{I}{J},VV);
 709             /*                 #define INNERFLOPS INNERFLOPS+3 */
 710             /*             #endif */
 711             /*             #if 'Force' in KERNEL_VF */
 712             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 713             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq{I}{J},FF),_mm_mul_pd(vftabscale,rinv{I}{J})));
 714             /*                 #define INNERFLOPS INNERFLOPS+7 */
 715             /*             #endif */
 716             /*         #endif */
 717             /*         ## End of check for electrostatics interaction forms */
 718             /*     #endif */
 719             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 720
 721             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 722
 723             /*         #if KERNEL_VDW=='LennardJones' */
 724
 725             /* LENNARD-JONES DISPERSION/REPULSION */
 726
 727             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 728             /*             #define INNERFLOPS INNERFLOPS+2 */
 729             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 730             vvdw6            = _mm_mul_pd(c6_{I}{J},rinvsix);
 731             vvdw12           = _mm_mul_pd(c12_{I}{J},_mm_mul_pd(rinvsix,rinvsix));
 732             /*                 #define INNERFLOPS INNERFLOPS+3 */
 733             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 734             vvdw             = _mm_sub_pd(_mm_mul_pd( _mm_sub_pd(vvdw12 , _mm_mul_pd(c12_{I}{J},_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 735                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
 736             /*                     #define INNERFLOPS INNERFLOPS+8 */
 737             /*                 #else */
 738             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 739             /*                     #define INNERFLOPS INNERFLOPS+3 */
 740             /*                 #endif */
 741             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 742             /*                 #if 'Force' in KERNEL_VF */
 743             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq{I}{J});
 744             /*                     #define INNERFLOPS INNERFLOPS+2 */
 745             /*                 #endif */
 746             /*             #elif KERNEL_VF=='Force' */
 747             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 748             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_{I}{J},rinvsix),c6_{I}{J}),_mm_mul_pd(rinvsix,rinvsq{I}{J}));
 749             /*                 #define INNERFLOPS INNERFLOPS+4 */
 750             /*             #endif */
 751
 752             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 753
 754             /* CUBIC SPLINE TABLE DISPERSION */
 755             /*             #if 'Table' in KERNEL_ELEC */
 756             vfitab           = _mm_add_epi32(vfitab,ifour);
 757             /*             #endif                     */
 758             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 759             /*             #if ROUND == 'Loop' */
 760             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 761             /*             #else */
 762             F                = _mm_setzero_pd();
 763             /*             #endif */
 764             GMX_MM_TRANSPOSE2_PD(Y,F);
 765             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 766             /*             #if ROUND == 'Loop' */
 767             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 768             /*             #else */
 769             H                = _mm_setzero_pd();
 770             /*             #endif */
 771             GMX_MM_TRANSPOSE2_PD(G,H);
 772             Heps             = _mm_mul_pd(vfeps,H);
 773             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 774             /*             #define INNERFLOPS INNERFLOPS+4 */
 775             /*             #if 'Potential' in KERNEL_VF */
 776             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 777             vvdw6            = _mm_mul_pd(c6_{I}{J},VV);
 778             /*                 #define INNERFLOPS INNERFLOPS+3 */
 779             /*             #endif */
 780             /*             #if 'Force' in KERNEL_VF */
 781             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 782             fvdw6            = _mm_mul_pd(c6_{I}{J},FF);
 783             /*                 #define INNERFLOPS INNERFLOPS+4 */
 784             /*             #endif */
 785
 786             /* CUBIC SPLINE TABLE REPULSION */
 787             vfitab           = _mm_add_epi32(vfitab,ifour);
 788             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 789             /*             #if ROUND == 'Loop' */
 790             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 791             /*             #else */
 792             F                = _mm_setzero_pd();
 793             /*             #endif */
 794             GMX_MM_TRANSPOSE2_PD(Y,F);
 795             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 796             /*             #if ROUND == 'Loop' */
 797             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 798             /*             #else */
 799             H                = _mm_setzero_pd();
 800             /*             #endif */
 801             GMX_MM_TRANSPOSE2_PD(G,H);
 802             Heps             = _mm_mul_pd(vfeps,H);
 803             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 804             /*             #define INNERFLOPS INNERFLOPS+4 */
 805             /*             #if 'Potential' in KERNEL_VF */
 806             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 807             vvdw12           = _mm_mul_pd(c12_{I}{J},VV);
 808             /*                 #define INNERFLOPS INNERFLOPS+3 */
 809             /*             #endif */
 810             /*             #if 'Force' in KERNEL_VF */
 811             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 812             fvdw12           = _mm_mul_pd(c12_{I}{J},FF);
 813             /*                 #define INNERFLOPS INNERFLOPS+5 */
 814             /*             #endif */
 815             /*             #if 'Potential' in KERNEL_VF */
 816             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 817             /*                 #define INNERFLOPS INNERFLOPS+1 */
 818             /*             #endif */
 819             /*             #if 'Force' in KERNEL_VF */
 820             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv{I}{J})));
 821             /*                 #define INNERFLOPS INNERFLOPS+4 */
 822             /*             #endif */
 823             /*         #endif */
 824             /*         ## End of check for vdw interaction forms */
 825             /*     #endif */
 826             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 827
 828             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 829             d                = _mm_sub_pd(r{I}{J},rswitch);
 830             d                = _mm_max_pd(d,_mm_setzero_pd());
 831             d2               = _mm_mul_pd(d,d);
 832             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 833             /*         #define INNERFLOPS INNERFLOPS+10 */
 834
 835             /*         #if 'Force' in KERNEL_VF */
 836             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 837             /*             #define INNERFLOPS INNERFLOPS+5 */
 838             /*         #endif */
 839
 840             /* Evaluate switch function */
 841             /*         #if 'Force' in KERNEL_VF */
 842             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 843             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 844             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv{I}{J},_mm_mul_pd(velec,dsw)) );
 845             /*                 #define INNERFLOPS INNERFLOPS+4 */
 846             /*             #endif */
 847             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 848             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv{I}{J},_mm_mul_pd(vvdw,dsw)) );
 849             /*                 #define INNERFLOPS INNERFLOPS+4 */
 850             /*             #endif */
 851             /*         #endif */
 852             /*         #if 'Potential' in KERNEL_VF */
 853             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 854             velec            = _mm_mul_pd(velec,sw);
 855             /*                 #define INNERFLOPS INNERFLOPS+1 */
 856             /*             #endif */
 857             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 858             vvdw             = _mm_mul_pd(vvdw,sw);
 859             /*                 #define INNERFLOPS INNERFLOPS+1 */
 860             /*             #endif */
 861             /*         #endif */
 862             /*     #endif */
 863             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 864             cutoff_mask      = _mm_cmplt_pd(rsq{I}{J},rcutoff2);
 865             /*         #define INNERFLOPS INNERFLOPS+1 */
 866             /*     #endif */
 867
 868             /*     #if 'Potential' in KERNEL_VF */
 869             /* Update potential sum for this i atom from the interaction with this j atom. */
 870             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 871             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 872             velec            = _mm_and_pd(velec,cutoff_mask);
 873             /*                 #define INNERFLOPS INNERFLOPS+1 */
 874             /*             #endif                                       */
 875             /*             #if ROUND == 'Epilogue' */
 876             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 877             /*             #endif */
 878             velecsum         = _mm_add_pd(velecsum,velec);
 879             /*             #define INNERFLOPS INNERFLOPS+1 */
 880             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 881             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 882             vgb              = _mm_and_pd(vgb,cutoff_mask);
 883             /*                 #define INNERFLOPS INNERFLOPS+1 */
 884             /*             #endif                                       */
 885             /*             #if ROUND == 'Epilogue' */
 886             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 887             /*             #endif */
 888             vgbsum           = _mm_add_pd(vgbsum,vgb);
 889             /*                 #define INNERFLOPS INNERFLOPS+1 */
 890             /*             #endif */
 891             /*         #endif */
 892             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 893             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 894             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 895             /*                 #define INNERFLOPS INNERFLOPS+1 */
 896             /*             #endif                                       */
 897             /*             #if ROUND == 'Epilogue' */
 898             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 899             /*             #endif */
 900             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 901             /*             #define INNERFLOPS INNERFLOPS+1 */
 902             /*         #endif */
 903             /*     #endif */
 904
 905             /*     #if 'Force' in KERNEL_VF */
 906
 907             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] and 'vdw' in INTERACTION_FLAGS[I][J] */
 908             fscal            = _mm_add_pd(felec,fvdw);
 909             /*             #define INNERFLOPS INNERFLOPS+1 */
 910             /*         #elif 'electrostatics' in INTERACTION_FLAGS[I][J] */
 911             fscal            = felec;
 912             /*         #elif 'vdw' in INTERACTION_FLAGS[I][J] */
 913             fscal            = fvdw;
 914             /*        #endif */
 915
 916             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 917             fscal            = _mm_and_pd(fscal,cutoff_mask);
 918             /*                 #define INNERFLOPS INNERFLOPS+1 */
 919             /*             #endif                                       */
 920
 921             /*             #if ROUND == 'Epilogue' */
 922             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 923             /*             #endif */
 924
 925             /* Calculate temporary vectorial force */
 926             tx               = _mm_mul_pd(fscal,dx{I}{J});
 927             ty               = _mm_mul_pd(fscal,dy{I}{J});
 928             tz               = _mm_mul_pd(fscal,dz{I}{J});
 929
 930             /* Update vectorial force */
 931             fix{I}             = _mm_add_pd(fix{I},tx);
 932             fiy{I}             = _mm_add_pd(fiy{I},ty);
 933             fiz{I}             = _mm_add_pd(fiz{I},tz);
 934             /*             #define INNERFLOPS INNERFLOPS+6 */
 935
 936             /* #if GEOMETRY_I == 'Particle'             */
 937             /*     #if ROUND == 'Loop' */
 938             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 939             /*     #else */
 940             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 941             /*     #endif */
 942             /*     #define INNERFLOPS INNERFLOPS+3      */
 943             /* #else                                    */
 944             fjx{J}             = _mm_add_pd(fjx{J},tx);
 945             fjy{J}             = _mm_add_pd(fjy{J},ty);
 946             fjz{J}             = _mm_add_pd(fjz{J},tz);
 947             /*     #define INNERFLOPS INNERFLOPS+3      */
 948             /* #endif                                   */
 949
 950             /*     #endif */
 951
 952             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 953             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
 954             {
 955                 /*     #endif */
 956             }
 957             /*     #endif */
 958             /*    ## End of check for the interaction being outside the cutoff */
 959
 960             /* #endfor */
 961             /* ## End of loop over i-j interaction pairs */
 962
 963             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
 964             /*     #if ROUND == 'Loop' */
 965             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 966             /*     #else */
 967             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 968             /*     #endif */
 969             /*     #define INNERFLOPS INNERFLOPS+3      */
 970             /* #elif GEOMETRY_J == 'Water3'             */
 971             /*     #if ROUND == 'Loop' */
 972             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 973             /*     #else */
 974             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 975             /*     #endif */
 976             /*     #define INNERFLOPS INNERFLOPS+9      */
 977             /* #elif GEOMETRY_J == 'Water4'             */
 978             /*     #if 0 in PARTICLES_J                 */
 979             /*         #if ROUND == 'Loop' */
 980             gmx_mm_decrement_4rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 981             /*         #else */
 982             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 983             /*         #endif */
 984             /*         #define INNERFLOPS INNERFLOPS+12 */
 985             /*     #else                                */
 986             /*         #if ROUND == 'Loop' */
 987             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA+DIM,f+j_coord_offsetB+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 988             /*         #else */
 989             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 990             /*         #endif */
 991             /*         #define INNERFLOPS INNERFLOPS+9  */
 992             /*     #endif                               */
 993             /* #endif                                   */
 994
 995             /* Inner loop uses {INNERFLOPS} flops */
 996         }
 997
 998         /* #endfor */
 999
1000         /* End of innermost loop */
1001
1002         /* #if 'Force' in KERNEL_VF */
1003         /*     #if GEOMETRY_I == 'Particle'            */
1004         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
1005                                               f+i_coord_offset,fshift+i_shift_offset);
1006         /*         #define OUTERFLOPS OUTERFLOPS+6     */
1007         /*     #elif GEOMETRY_I == 'Water3'            */
1008         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1009                                               f+i_coord_offset,fshift+i_shift_offset);
1010         /*         #define OUTERFLOPS OUTERFLOPS+18    */
1011         /*     #elif GEOMETRY_I == 'Water4'            */
1012         /*         #if 0 in PARTICLES_I                */
1013         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1014                                               f+i_coord_offset,fshift+i_shift_offset);
1015         /*             #define OUTERFLOPS OUTERFLOPS+24    */
1016         /*         #else                               */
1017         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1018                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1019         /*             #define OUTERFLOPS OUTERFLOPS+18    */
1020         /*         #endif                              */
1021         /*     #endif                                  */
1022         /* #endif                                      */
1023
1024         /* #if 'Potential' in KERNEL_VF */
1025         ggid                        = gid[iidx];
1026         /* Update potential energies */
1027         /*     #if KERNEL_ELEC != 'None' */
1028         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
1029         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1030         /*     #endif */
1031         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
1032         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
1033         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1034         /*     #endif */
1035         /*     #if KERNEL_VDW != 'None' */
1036         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1037         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1038         /*     #endif */
1039         /* #endif */
1040         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1041         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai{I},isai{I}));
1042         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
1043         /*     #endif */
1044
1045         /* Increment number of inner iterations */
1046         inneriter                  += j_index_end - j_index_start;
1047
1048         /* Outer loop uses {OUTERFLOPS} flops */
1049     }
1050
1051     /* Increment number of outer iterations */
1052     outeriter        += nri;
1053
1054     /* Update outer/inner flops */
1055     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1056     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1057     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1058     /* #if GEOMETRY_I == 'Water3'            */
1059     /*     #define ISUFFIX '_W3'             */
1060     /* #elif GEOMETRY_I == 'Water4'          */
1061     /*     #define ISUFFIX '_W4'             */
1062     /* #else                                 */
1063     /*     #define ISUFFIX ''                */
1064     /* #endif                                */
1065     /* #if GEOMETRY_J == 'Water3'            */
1066     /*     #define JSUFFIX 'W3'              */
1067     /* #elif GEOMETRY_J == 'Water4'          */
1068     /*     #define JSUFFIX 'W4'              */
1069     /* #else                                 */
1070     /*     #define JSUFFIX ''                */
1071     /* #endif                                */
1072     /* #if 'PotentialAndForce' in KERNEL_VF  */
1073     /*     #define VFSUFFIX  '_VF'           */
1074     /* #elif 'Potential' in KERNEL_VF        */
1075     /*     #define VFSUFFIX '_V'             */
1076     /* #else                                 */
1077     /*     #define VFSUFFIX '_F'             */
1078     /* #endif                                */
1079
1080     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1081     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1082     /* #elif KERNEL_ELEC != 'None' */
1083     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1084     /* #else */
1085     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1086     /* #endif  */
1087 }