src/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_template_sse2_double.pre

   1 /* #if 0 */
   2 #error This file must be processed with the Gromacs pre-preprocessor
   3 /* #endif */
   4 /* #if INCLUDE_HEADER */
   5 #ifdef HAVE_CONFIG_H
   6 #include <config.h>
   7 #endif
   8
   9 #include <math.h>
  10
  11 #include "../nb_kernel.h"
  12 #include "types/simple.h"
  13 #include "vec.h"
  14 #include "nrnb.h"
  15
  16 #include "gmx_math_x86_sse2_double.h"
  17 #include "kernelutil_x86_sse2_double.h"
  18 /* #endif */
  19
  20 /* ## List of variables set by the generating script:                                    */
  21 /* ##                                                                                    */
  22 /* ## Setttings that apply to the entire kernel:                                         */
  23 /* ## KERNEL_ELEC:           String, choice for electrostatic interactions               */
  24 /* ## KERNEL_VDW:            String, choice for van der Waals interactions               */
  25 /* ## KERNEL_NAME:           String, name of this kernel                                 */
  26 /* ## KERNEL_VF:             String telling if we calculate potential, force, or both    */
  27 /* ## GEOMETRY_I/GEOMETRY_J: String, name of each geometry, e.g. 'Water3' or '1Particle' */
  28 /* ##                                                                                    */
  29 /* ## Setttings that apply to particles in the outer (I) or inner (J) loops:             */
  30 /* ## PARTICLES_I[]/         Arrays with lists of i/j particles to use in kernel. It is  */
  31 /* ## PARTICLES_J[]:         just [0] for particle geometry, but can be longer for water */
  32 /* ## PARTICLES_ELEC_I[]/    Arrays with lists of i/j particle that have electrostatics  */
  33 /* ## PARTICLES_ELEC_J[]:    interactions that should be calculated in this kernel.      */
  34 /* ## PARTICLES_VDW_I[]/     Arrays with the list of i/j particle that have VdW          */
  35 /* ## PARTICLES_VDW_J[]:     interactions that should be calculated in this kernel.      */
  36 /* ##                                                                                    */
  37 /* ## Setttings for pairs of interactions (e.g. 2nd i particle against 1st j particle)   */
  38 /* ## PAIRS_IJ[]:            Array with (i,j) tuples of pairs for which interactions     */
  39 /* ##                        should be calculated in this kernel. Zero-charge particles  */
  40 /* ##                        do not have interactions with particles without vdw, and    */
  41 /* ##                        Vdw-only interactions are not evaluated in a no-vdw-kernel. */
  42 /* ## INTERACTION_FLAGS[][]: 2D matrix, dimension e.g. 3*3 for water-water interactions. */
  43 /* ##                        For each i-j pair, the element [I][J] is a list of strings  */
  44 /* ##                        defining properties/flags of this interaction. Examples     */
  45 /* ##                        include 'electrostatics'/'vdw' if that type of interaction  */
  46 /* ##                        should be evaluated, 'rsq'/'rinv'/'rinvsq' if those values  */
  47 /* ##                        are needed, and 'exactcutoff' or 'shift','switch' to        */
  48 /* ##                        decide if the force/potential should be modified. This way  */
  49 /* ##                        we only calculate values absolutely needed for each case.   */
  50
  51 /* ## Calculate the size and offset for (merged/interleaved) table data */
  52
  53 /*
  54  * Gromacs nonbonded kernel:   {KERNEL_NAME}
  55  * Electrostatics interaction: {KERNEL_ELEC}
  56  * VdW interaction:            {KERNEL_VDW}
  57  * Geometry:                   {GEOMETRY_I}-{GEOMETRY_J}
  58  * Calculate force/pot:        {KERNEL_VF}
  59  */
  60 void
  61 {KERNEL_NAME}
  62                     (t_nblist * gmx_restrict                nlist,
  63                      rvec * gmx_restrict                    xx,
  64                      rvec * gmx_restrict                    ff,
  65                      t_forcerec * gmx_restrict              fr,
  66                      t_mdatoms * gmx_restrict               mdatoms,
  67                      nb_kernel_data_t * gmx_restrict        kernel_data,
  68                      t_nrnb * gmx_restrict                  nrnb)
  69 {
  70     /* ## Not all variables are used for all kernels, but any optimizing compiler fixes that, */
  71     /* ## so there is no point in going to extremes to exclude variables that are not needed. */
  72     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  73      * just 0 for non-waters.
  74      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  75      * jnr indices corresponding to data put in the four positions in the SIMD register.
  76      */
  77     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  78     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  79     int              jnrA,jnrB;
  80     int              j_coord_offsetA,j_coord_offsetB;
  81     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  82     real             rcutoff_scalar;
  83     real             *shiftvec,*fshift,*x,*f;
  84     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     /* #for I in PARTICLES_I */
  86     int              vdwioffset{I};
  87     __m128d          ix{I},iy{I},iz{I},fix{I},fiy{I},fiz{I},iq{I},isai{I};
  88     /* #endfor */
  89     /* #for J in PARTICLES_J */
  90     int              vdwjidx{J}A,vdwjidx{J}B;
  91     __m128d          jx{J},jy{J},jz{J},fjx{J},fjy{J},fjz{J},jq{J},isaj{J};
  92     /* #endfor */
  93     /* #for I,J in PAIRS_IJ */
  94     __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};
  95     /* #endfor */
  96     /* #if KERNEL_ELEC != 'None' */
  97     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  98     real             *charge;
  99     /* #endif */
 100     /* #if 'GeneralizedBorn' in KERNEL_ELEC */
 101     __m128i          gbitab;
 102     __m128d          vgb,fgb,vgbsum,dvdasum,gbscale,gbtabscale,isaprod,gbqqfactor,gbinvepsdiff,dvdaj,gbeps,dvdatmp;
 103     __m128d          minushalf = _mm_set1_pd(-0.5);
 104     real             *invsqrta,*dvda,*gbtab;
 105     /* #endif */
 106     /* #if KERNEL_VDW != 'None' */
 107     int              nvdwtype;
 108     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 109     int              *vdwtype;
 110     real             *vdwparam;
 111     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 112     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 113     /* #endif */
 114     /* #if 'Table' in KERNEL_ELEC or 'GeneralizedBorn' in KERNEL_ELEC or 'Table' in KERNEL_VDW */
 115     __m128i          vfitab;
 116     __m128i          ifour       = _mm_set1_epi32(4);
 117     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 118     real             *vftab;
 119     /* #endif */
 120     /* #if 'Ewald' in KERNEL_ELEC */
 121     __m128i          ewitab;
 122     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 123     real             *ewtab;
 124     /* #endif */
 125     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 126     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 127     real             rswitch_scalar,d_scalar;
 128     /* #endif */
 129     __m128d          dummy_mask,cutoff_mask;
 130     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 131     __m128d          one     = _mm_set1_pd(1.0);
 132     __m128d          two     = _mm_set1_pd(2.0);
 133     x                = xx[0];
 134     f                = ff[0];
 135
 136     nri              = nlist->nri;
 137     iinr             = nlist->iinr;
 138     jindex           = nlist->jindex;
 139     jjnr             = nlist->jjnr;
 140     shiftidx         = nlist->shift;
 141     gid              = nlist->gid;
 142     shiftvec         = fr->shift_vec[0];
 143     fshift           = fr->fshift[0];
 144     /* #if KERNEL_ELEC != 'None' */
 145     facel            = _mm_set1_pd(fr->epsfac);
 146     charge           = mdatoms->chargeA;
 147     /*     #if 'ReactionField' in KERNEL_ELEC */
 148     krf              = _mm_set1_pd(fr->ic->k_rf);
 149     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 150     crf              = _mm_set1_pd(fr->ic->c_rf);
 151     /*     #endif */
 152     /* #endif */
 153     /* #if KERNEL_VDW != 'None' */
 154     nvdwtype         = fr->ntype;
 155     vdwparam         = fr->nbfp;
 156     vdwtype          = mdatoms->typeA;
 157     /* #endif */
 158
 159     /* #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW */
 160     vftab            = kernel_data->table_elec_vdw->data;
 161     vftabscale       = _mm_set1_pd(kernel_data->table_elec_vdw->scale);
 162     /* #elif 'Table' in KERNEL_ELEC */
 163     vftab            = kernel_data->table_elec->data;
 164     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 165     /* #elif 'Table' in KERNEL_VDW */
 166     vftab            = kernel_data->table_vdw->data;
 167     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 168     /* #endif */
 169
 170     /* #if 'Ewald' in KERNEL_ELEC */
 171     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 172     /*     #if KERNEL_VF=='Force' and KERNEL_MOD_ELEC!='PotentialSwitch' */
 173     ewtab            = fr->ic->tabq_coul_F;
 174     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 175     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 176     /*     #else */
 177     ewtab            = fr->ic->tabq_coul_FDV0;
 178     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 179     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 180      /*     #endif */
 181     /* #endif */
 182
 183     /* #if KERNEL_ELEC=='GeneralizedBorn' */
 184     invsqrta         = fr->invsqrta;
 185     dvda             = fr->dvda;
 186     gbtabscale       = _mm_set1_pd(fr->gbtab.scale);
 187     gbtab            = fr->gbtab.data;
 188     gbinvepsdiff     = _mm_set1_pd((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
 189     /* #endif */
 190
 191     /* #if 'Water' in GEOMETRY_I */
 192     /* Setup water-specific parameters */
 193     inr              = nlist->iinr[0];
 194     /*     #for I in PARTICLES_ELEC_I */
 195     iq{I}              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+{I}]));
 196     /*     #endfor */
 197     /*     #for I in PARTICLES_VDW_I */
 198     vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 199     /*     #endfor */
 200     /* #endif */
 201
 202     /* #if 'Water' in GEOMETRY_J */
 203     /*     #for J in PARTICLES_ELEC_J */
 204     jq{J}              = _mm_set1_pd(charge[inr+{J}]);
 205     /*     #endfor */
 206     /*     #for J in PARTICLES_VDW_J */
 207     vdwjidx{J}A        = 2*vdwtype[inr+{J}];
 208     /*     #endfor */
 209     /*     #for I,J in PAIRS_IJ */
 210     /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 211     qq{I}{J}             = _mm_mul_pd(iq{I},jq{J});
 212     /*         #endif */
 213     /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 214     c6_{I}{J}            = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A]);
 215     c12_{I}{J}           = _mm_set1_pd(vdwparam[vdwioffset{I}+vdwjidx{J}A+1]);
 216     /*         #endif */
 217     /*     #endfor */
 218     /* #endif */
 219
 220     /* #if KERNEL_MOD_ELEC!='None' or KERNEL_MOD_VDW!='None' */
 221     /*     #if KERNEL_ELEC!='None' */
 222     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 223     rcutoff_scalar   = fr->rcoulomb;
 224     /*     #else */
 225     rcutoff_scalar   = fr->rvdw;
 226     /*     #endif */
 227     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 228     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 229     /* #endif */
 230
 231     /* #if KERNEL_MOD_VDW=='PotentialShift' */
 232     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 233     rvdw             = _mm_set1_pd(fr->rvdw);
 234     /* #endif */
 235
 236     /* #if 'PotentialSwitch' in [KERNEL_MOD_ELEC,KERNEL_MOD_VDW] */
 237     /*     #if KERNEL_MOD_ELEC=='PotentialSwitch'  */
 238     rswitch_scalar   = fr->rcoulomb_switch;
 239     rswitch          = _mm_set1_pd(rswitch_scalar);
 240     /*     #else */
 241     rswitch_scalar   = fr->rvdw_switch;
 242     rswitch          = _mm_set1_pd(rswitch_scalar);
 243     /*     #endif */
 244     /* Setup switch parameters */
 245     d_scalar         = rcutoff_scalar-rswitch_scalar;
 246     d                = _mm_set1_pd(d_scalar);
 247     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 248     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 249     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 250     /*     #if 'Force' in KERNEL_VF */
 251     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 252     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 253     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 254     /*     #endif */
 255     /* #endif */
 256
 257     /* Avoid stupid compiler warnings */
 258     jnrA = jnrB = 0;
 259     j_coord_offsetA = 0;
 260     j_coord_offsetB = 0;
 261
 262     /* ## Keep track of the floating point operations we issue for reporting! */
 263     /* #define OUTERFLOPS 0 */
 264     outeriter        = 0;
 265     inneriter        = 0;
 266
 267     /* Start outer loop over neighborlists */
 268     for(iidx=0; iidx<nri; iidx++)
 269     {
 270         /* Load shift vector for this list */
 271         i_shift_offset   = DIM*shiftidx[iidx];
 272
 273         /* Load limits for loop over neighbors */
 274         j_index_start    = jindex[iidx];
 275         j_index_end      = jindex[iidx+1];
 276
 277         /* Get outer coordinate index */
 278         inr              = iinr[iidx];
 279         i_coord_offset   = DIM*inr;
 280
 281         /* Load i particle coords and add shift vector */
 282         /* #if GEOMETRY_I == 'Particle' */
 283         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 284         /* #elif GEOMETRY_I == 'Water3' */
 285         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 286                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 287         /* #elif GEOMETRY_I == 'Water4' */
 288         /*     #if 0 in PARTICLES_I                 */
 289         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 290                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 291         /*     #else                                */
 292         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 293                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 294         /*     #endif                               */
 295         /* #endif                                   */
 296
 297         /* #if 'Force' in KERNEL_VF */
 298         /*     #for I in PARTICLES_I */
 299         fix{I}             = _mm_setzero_pd();
 300         fiy{I}             = _mm_setzero_pd();
 301         fiz{I}             = _mm_setzero_pd();
 302         /*     #endfor */
 303         /* #endif */
 304
 305         /* ## For water we already preloaded parameters at the start of the kernel */
 306         /* #if not 'Water' in GEOMETRY_I */
 307         /* Load parameters for i particles */
 308         /*     #for I in PARTICLES_ELEC_I */
 309         iq{I}              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+{I}));
 310         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 311         /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 312         isai{I}            = _mm_load1_pd(invsqrta+inr+{I});
 313         /*         #endif */
 314         /*     #endfor */
 315         /*     #for I in PARTICLES_VDW_I */
 316         vdwioffset{I}      = 2*nvdwtype*vdwtype[inr+{I}];
 317         /*     #endfor */
 318         /* #endif */
 319
 320         /* #if 'Potential' in KERNEL_VF */
 321         /* Reset potential sums */
 322         /*     #if KERNEL_ELEC != 'None' */
 323         velecsum         = _mm_setzero_pd();
 324         /*     #endif */
 325         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 326         vgbsum           = _mm_setzero_pd();
 327         /*     #endif */
 328         /*     #if KERNEL_VDW != 'None' */
 329         vvdwsum          = _mm_setzero_pd();
 330         /*     #endif */
 331         /* #endif */
 332         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
 333         dvdasum          = _mm_setzero_pd();
 334         /*     #endif */
 335
 336         /* #for ROUND in ['Loop','Epilogue'] */
 337
 338         /* #if ROUND =='Loop' */
 339         /* Start inner kernel loop */
 340         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 341         {
 342         /* ## First round is normal loop (next statement resets indentation) */
 343         /*     #if 0 */
 344         }
 345         /*     #endif */
 346         /* #else */
 347         if(jidx<j_index_end)
 348         {
 349         /* ## Second round is epilogue */
 350         /* #endif */
 351         /* #define INNERFLOPS 0 */
 352
 353             /* #if ROUND =='Loop' */
 354             /* Get j neighbor index, and coordinate index */
 355             jnrA             = jjnr[jidx];
 356             jnrB             = jjnr[jidx+1];
 357             j_coord_offsetA  = DIM*jnrA;
 358             j_coord_offsetB  = DIM*jnrB;
 359
 360             /* load j atom coordinates */
 361             /*     #if GEOMETRY_J == 'Particle'             */
 362             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 363                                               &jx0,&jy0,&jz0);
 364             /*     #elif GEOMETRY_J == 'Water3'             */
 365             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 366                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 367             /*     #elif GEOMETRY_J == 'Water4'             */
 368             /*         #if 0 in PARTICLES_J                 */
 369             gmx_mm_load_4rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 370                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 371                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 372             /*         #else                                */
 373             gmx_mm_load_3rvec_2ptr_swizzle_pd(x+j_coord_offsetA+DIM,x+j_coord_offsetB+DIM,
 374                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 375             /*         #endif                               */
 376             /*     #endif                                   */
 377             /* #else */
 378             jnrA             = jjnr[jidx];
 379             j_coord_offsetA  = DIM*jnrA;
 380
 381             /* load j atom coordinates */
 382             /*     #if GEOMETRY_J == 'Particle'             */
 383             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 384                                               &jx0,&jy0,&jz0);
 385             /*     #elif GEOMETRY_J == 'Water3'             */
 386             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 387                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 388             /*     #elif GEOMETRY_J == 'Water4'             */
 389             /*         #if 0 in PARTICLES_J                 */
 390             gmx_mm_load_4rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 391                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 392                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 393             /*         #else                                */
 394             gmx_mm_load_3rvec_1ptr_swizzle_pd(x+j_coord_offsetA+DIM,
 395                                               &jx1,&jy1,&jz1,&jx2,&jy2,&jz2,&jx3,&jy3,&jz3);
 396             /*         #endif                               */
 397             /*     #endif                                   */
 398             /* #endif */
 399
 400             /* Calculate displacement vector */
 401             /* #for I,J in PAIRS_IJ */
 402             dx{I}{J}             = _mm_sub_pd(ix{I},jx{J});
 403             dy{I}{J}             = _mm_sub_pd(iy{I},jy{J});
 404             dz{I}{J}             = _mm_sub_pd(iz{I},jz{J});
 405             /*     #define INNERFLOPS INNERFLOPS+3 */
 406             /* #endfor */
 407
 408             /* Calculate squared distance and things based on it */
 409             /* #for I,J in PAIRS_IJ */
 410             rsq{I}{J}            = gmx_mm_calc_rsq_pd(dx{I}{J},dy{I}{J},dz{I}{J});
 411             /*     #define INNERFLOPS INNERFLOPS+5 */
 412             /* #endfor */
 413
 414             /* #for I,J in PAIRS_IJ */
 415             /*     #if 'rinv' in INTERACTION_FLAGS[I][J] */
 416             rinv{I}{J}           = gmx_mm_invsqrt_pd(rsq{I}{J});
 417             /*         #define INNERFLOPS INNERFLOPS+5 */
 418             /*     #endif */
 419             /* #endfor */
 420
 421             /* #for I,J in PAIRS_IJ */
 422             /*     #if 'rinvsq' in INTERACTION_FLAGS[I][J] */
 423             /*         # if 'rinv' not in INTERACTION_FLAGS[I][J] */
 424             rinvsq{I}{J}         = gmx_mm_inv_pd(rsq{I}{J});
 425             /*             #define INNERFLOPS INNERFLOPS+4 */
 426             /*         #else */
 427             rinvsq{I}{J}         = _mm_mul_pd(rinv{I}{J},rinv{I}{J});
 428             /*             #define INNERFLOPS INNERFLOPS+1 */
 429             /*         #endif */
 430             /*     #endif */
 431             /* #endfor */
 432
 433             /* #if not 'Water' in GEOMETRY_J */
 434             /* Load parameters for j particles */
 435             /*     #for J in PARTICLES_ELEC_J */
 436             /*         #if ROUND =='Loop' */
 437             jq{J}              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+{J},charge+jnrB+{J});
 438             /*         #else */
 439             jq{J}              = _mm_load_sd(charge+jnrA+{J});
 440             /*         #endif */
 441             /*         #if KERNEL_ELEC=='GeneralizedBorn' */
 442             /*             #if ROUND =='Loop' */
 443             isaj{J}            = gmx_mm_load_2real_swizzle_pd(invsqrta+jnrA+{J},invsqrta+jnrB+{J});
 444             /*             #else */
 445             isaj{J}            = _mm_load_sd(invsqrta+jnrA+{J});
 446             /*             #endif */
 447             /*         #endif */
 448             /*     #endfor */
 449             /*     #for J in PARTICLES_VDW_J */
 450             vdwjidx{J}A        = 2*vdwtype[jnrA+{J}];
 451             /*         #if ROUND =='Loop' */
 452             vdwjidx{J}B        = 2*vdwtype[jnrB+{J}];
 453             /*         #endif */
 454             /*     #endfor */
 455             /* #endif */
 456
 457             /* #if 'Force' in KERNEL_VF and not 'Particle' in GEOMETRY_I */
 458             /*     #for J in PARTICLES_J */
 459             fjx{J}             = _mm_setzero_pd();
 460             fjy{J}             = _mm_setzero_pd();
 461             fjz{J}             = _mm_setzero_pd();
 462             /*     #endfor */
 463             /* #endif */
 464
 465             /* #for I,J in PAIRS_IJ */
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 472             /*         ## We always calculate rinv/rinvsq above to enable pipelineing in compilers (performance tested on x86) */
 473             if (gmx_mm_any_lt(rsq{I}{J},rcutoff2))
 474             {
 475                 /*     #if 0    ## this and the next two lines is a hack to maintain auto-indentation in template file */
 476             }
 477             /*         #endif */
 478             /*         #define INNERFLOPS INNERFLOPS+1 */
 479             /*     #endif */
 480
 481             /*     #if 'r' in INTERACTION_FLAGS[I][J] */
 482             r{I}{J}              = _mm_mul_pd(rsq{I}{J},rinv{I}{J});
 483              /*         #define INNERFLOPS INNERFLOPS+1 */
 484             /*     #endif */
 485
 486             /*     ## For water geometries we already loaded parameters at the start of the kernel */
 487             /*     #if not 'Water' in GEOMETRY_J */
 488             /* Compute parameters for interactions between i and j atoms */
 489             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 490             qq{I}{J}             = _mm_mul_pd(iq{I},jq{J});
 491             /*             #define INNERFLOPS INNERFLOPS+1 */
 492             /*         #endif */
 493             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 494             /*             #if ROUND == 'Loop' */
 495             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset{I}+vdwjidx{J}A,
 496                                          vdwparam+vdwioffset{I}+vdwjidx{J}B,&c6_{I}{J},&c12_{I}{J});
 497             /*             #else */
 498             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset{I}+vdwjidx{J}A,&c6_{I}{J},&c12_{I}{J});
 499             /*             #endif */
 500             /*         #endif */
 501             /*     #endif */
 502
 503             /*     #if 'table' in INTERACTION_FLAGS[I][J] */
 504             /* Calculate table index by multiplying r with table scale and truncate to integer */
 505             rt               = _mm_mul_pd(r{I}{J},vftabscale);
 506             vfitab           = _mm_cvttpd_epi32(rt);
 507             vfeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(vfitab));
 508             /*         #define INNERFLOPS INNERFLOPS+4                          */
 509             /*         #if 'Table' in KERNEL_ELEC and 'Table' in KERNEL_VDW     */
 510             /*             ## 3 tables, 4 data per point: multiply index by 12 */
 511             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 512             /*         #elif 'Table' in KERNEL_ELEC                             */
 513             /*             ## 1 table, 4 data per point: multiply index by 4   */
 514             vfitab           = _mm_slli_epi32(vfitab,2);
 515             /*         #elif 'Table' in KERNEL_VDW                              */
 516             /*             ## 2 tables, 4 data per point: multiply index by 8  */
 517             vfitab           = _mm_slli_epi32(vfitab,3);
 518             /*         #endif                                                   */
 519             /*     #endif */
 520
 521             /*     ## ELECTROSTATIC INTERACTIONS */
 522             /*     #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 523
 524             /*         #if KERNEL_ELEC=='Coulomb' */
 525
 526             /* COULOMB ELECTROSTATICS */
 527             velec            = _mm_mul_pd(qq{I}{J},rinv{I}{J});
 528             /*             #define INNERFLOPS INNERFLOPS+1 */
 529             /*             #if 'Force' in KERNEL_VF */
 530             felec            = _mm_mul_pd(velec,rinvsq{I}{J});
 531             /*                 #define INNERFLOPS INNERFLOPS+2 */
 532             /*             #endif */
 533
 534             /*         #elif KERNEL_ELEC=='ReactionField' */
 535
 536             /* REACTION-FIELD ELECTROSTATICS */
 537             /*             #if 'Potential' in KERNEL_VF */
 538             velec            = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_add_pd(rinv{I}{J},_mm_mul_pd(krf,rsq{I}{J})),crf));
 539             /*                 #define INNERFLOPS INNERFLOPS+4 */
 540             /*             #endif */
 541             /*             #if 'Force' in KERNEL_VF */
 542             felec            = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_mul_pd(rinv{I}{J},rinvsq{I}{J}),krf2));
 543             /*                 #define INNERFLOPS INNERFLOPS+3 */
 544             /*             #endif */
 545
 546             /*         #elif KERNEL_ELEC=='GeneralizedBorn' */
 547
 548             /* GENERALIZED BORN AND COULOMB ELECTROSTATICS */
 549             isaprod          = _mm_mul_pd(isai{I},isaj{J});
 550             gbqqfactor       = _mm_xor_pd(signbit,_mm_mul_pd(qq{I}{J},_mm_mul_pd(isaprod,gbinvepsdiff)));
 551             gbscale          = _mm_mul_pd(isaprod,gbtabscale);
 552             /*             #define INNERFLOPS INNERFLOPS+5 */
 553
 554             /* Calculate generalized born table index - this is a separate table from the normal one,
 555              * but we use the same procedure by multiplying r with scale and truncating to integer.
 556              */
 557             rt               = _mm_mul_pd(r{I}{J},gbscale);
 558             gbitab           = _mm_cvttpd_epi32(rt);
 559             gbeps            = _mm_sub_pd(rt,_mm_cvtepi32_pd(gbitab));
 560             gbitab           = _mm_slli_epi32(gbitab,2);
 561
 562             Y                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) );
 563             /*             #if ROUND == 'Loop' */
 564             F                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) );
 565             /*             #else */
 566             F                = _mm_setzero_pd();
 567             /*             #endif */
 568             GMX_MM_TRANSPOSE2_PD(Y,F);
 569             G                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,0) +2);
 570             /*             #if ROUND == 'Loop' */
 571             H                = _mm_load_pd( gbtab + gmx_mm_extract_epi32(gbitab,1) +2);
 572             /*             #else */
 573             H                = _mm_setzero_pd();
 574             /*             #endif */
 575             GMX_MM_TRANSPOSE2_PD(G,H);
 576             Heps             = _mm_mul_pd(gbeps,H);
 577             Fp               = _mm_add_pd(F,_mm_mul_pd(gbeps,_mm_add_pd(G,Heps)));
 578             VV               = _mm_add_pd(Y,_mm_mul_pd(gbeps,Fp));
 579             vgb              = _mm_mul_pd(gbqqfactor,VV);
 580             /*             #define INNERFLOPS INNERFLOPS+10 */
 581
 582             /*             #if 'Force' in KERNEL_VF */
 583             FF               = _mm_add_pd(Fp,_mm_mul_pd(gbeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 584             fgb              = _mm_mul_pd(gbqqfactor,_mm_mul_pd(FF,gbscale));
 585             dvdatmp          = _mm_mul_pd(minushalf,_mm_add_pd(vgb,_mm_mul_pd(fgb,r{I}{J})));
 586             dvdasum          = _mm_add_pd(dvdasum,dvdatmp);
 587             /*             #if ROUND == 'Loop' */
 588             gmx_mm_increment_2real_swizzle_pd(dvda+jnrA,dvda+jnrB,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj{J},isaj{J})));
 589             /*             #else */
 590             gmx_mm_increment_1real_pd(dvda+jnrA,_mm_mul_pd(dvdatmp,_mm_mul_pd(isaj{J},isaj{J})));
 591             /*             #endif */
 592             /*                 #define INNERFLOPS INNERFLOPS+13 */
 593             /*             #endif */
 594             velec            = _mm_mul_pd(qq{I}{J},rinv{I}{J});
 595             /*                 #define INNERFLOPS INNERFLOPS+1 */
 596             /*             #if 'Force' in KERNEL_VF */
 597             felec            = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(velec,rinv{I}{J}),fgb),rinv{I}{J});
 598             /*                 #define INNERFLOPS INNERFLOPS+3 */
 599             /*             #endif */
 600
 601             /*         #elif KERNEL_ELEC=='Ewald' */
 602             /* EWALD ELECTROSTATICS */
 603
 604             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 605             ewrt             = _mm_mul_pd(r{I}{J},ewtabscale);
 606             ewitab           = _mm_cvttpd_epi32(ewrt);
 607             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 608             /*             #define INNERFLOPS INNERFLOPS+4 */
 609             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_ELEC=='PotentialSwitch' */
 610             ewitab           = _mm_slli_epi32(ewitab,2);
 611             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 612             /*                 #if ROUND == 'Loop' */
 613             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 614             /*                 #else */
 615             ewtabD           = _mm_setzero_pd();
 616             /*                 #endif */
 617             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 618             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 619             /*                 #if ROUND == 'Loop' */
 620             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 621             /*                 #else */
 622             ewtabFn          = _mm_setzero_pd();
 623             /*                 #endif */
 624             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 625             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 626             /*                 #define INNERFLOPS INNERFLOPS+2 */
 627             /*                 #if KERNEL_MOD_ELEC=='PotentialShift' */
 628             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 629             velec            = _mm_mul_pd(qq{I}{J},_mm_sub_pd(_mm_sub_pd(rinv{I}{J},sh_ewald),velec));
 630             /*                     #define INNERFLOPS INNERFLOPS+7 */
 631             /*                 #else */
 632             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 633             velec            = _mm_mul_pd(qq{I}{J},_mm_sub_pd(rinv{I}{J},velec));
 634             /*                     #define INNERFLOPS INNERFLOPS+6 */
 635             /*                 #endif */
 636             /*                 #if 'Force' in KERNEL_VF */
 637             felec            = _mm_mul_pd(_mm_mul_pd(qq{I}{J},rinv{I}{J}),_mm_sub_pd(rinvsq{I}{J},felec));
 638             /*                      #define INNERFLOPS INNERFLOPS+3 */
 639             /*                 #endif */
 640             /*             #elif KERNEL_VF=='Force' */
 641             /*                 #if ROUND == 'Loop' */
 642             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 643                                          &ewtabF,&ewtabFn);
 644             /*                 #else */
 645             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 646             /*                 #endif */
 647             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 648             felec            = _mm_mul_pd(_mm_mul_pd(qq{I}{J},rinv{I}{J}),_mm_sub_pd(rinvsq{I}{J},felec));
 649             /*                 #define INNERFLOPS INNERFLOPS+7 */
 650             /*             #endif */
 651
 652             /*         #elif KERNEL_ELEC=='CubicSplineTable' */
 653
 654             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 655             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 656             /*             #if ROUND == 'Loop' */
 657             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 658             /*             #else */
 659             F                = _mm_setzero_pd();
 660             /*             #endif */
 661             GMX_MM_TRANSPOSE2_PD(Y,F);
 662             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 663             /*             #if ROUND == 'Loop' */
 664             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 665             /*             #else */
 666             H                = _mm_setzero_pd();
 667             /*             #endif */
 668             GMX_MM_TRANSPOSE2_PD(G,H);
 669             Heps             = _mm_mul_pd(vfeps,H);
 670             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 671             /*             #define INNERFLOPS INNERFLOPS+4 */
 672             /*             #if 'Potential' in KERNEL_VF */
 673             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 674             velec            = _mm_mul_pd(qq{I}{J},VV);
 675             /*                 #define INNERFLOPS INNERFLOPS+3 */
 676             /*             #endif */
 677             /*             #if 'Force' in KERNEL_VF */
 678             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 679             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq{I}{J},FF),_mm_mul_pd(vftabscale,rinv{I}{J})));
 680             /*                 #define INNERFLOPS INNERFLOPS+7 */
 681             /*             #endif */
 682             /*         #endif */
 683             /*         ## End of check for electrostatics interaction forms */
 684             /*     #endif */
 685             /*     ## END OF ELECTROSTATIC INTERACTION CHECK FOR PAIR I-J */
 686
 687             /*     #if 'vdw' in INTERACTION_FLAGS[I][J] */
 688
 689             /*         #if KERNEL_VDW=='LennardJones' */
 690
 691             /* LENNARD-JONES DISPERSION/REPULSION */
 692
 693             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq{I}{J},rinvsq{I}{J}),rinvsq{I}{J});
 694             /*             #define INNERFLOPS INNERFLOPS+2 */
 695             /*             #if 'Potential' in KERNEL_VF or KERNEL_MOD_VDW=='PotentialSwitch' */
 696             vvdw6            = _mm_mul_pd(c6_{I}{J},rinvsix);
 697             vvdw12           = _mm_mul_pd(c12_{I}{J},_mm_mul_pd(rinvsix,rinvsix));
 698             /*                 #define INNERFLOPS INNERFLOPS+3 */
 699             /*                 #if KERNEL_MOD_VDW=='PotentialShift' */
 700             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) ,
 701                                           _mm_mul_pd( _mm_sub_pd(vvdw6,_mm_mul_pd(c6_{I}{J},sh_vdw_invrcut6)),one_sixth));
 702             /*                     #define INNERFLOPS INNERFLOPS+8 */
 703             /*                 #else */
 704             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 705             /*                     #define INNERFLOPS INNERFLOPS+3 */
 706             /*                 #endif */
 707             /*                 ## Check for force inside potential check, i.e. this means we already did the potential part */
 708             /*                 #if 'Force' in KERNEL_VF */
 709             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq{I}{J});
 710             /*                     #define INNERFLOPS INNERFLOPS+2 */
 711             /*                 #endif */
 712             /*             #elif KERNEL_VF=='Force' */
 713             /*                 ## Force-only LennardJones makes it possible to save 1 flop (they do add up...) */
 714             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_{I}{J},rinvsix),c6_{I}{J}),_mm_mul_pd(rinvsix,rinvsq{I}{J}));
 715             /*                 #define INNERFLOPS INNERFLOPS+4 */
 716             /*             #endif */
 717
 718             /*         #elif KERNEL_VDW=='CubicSplineTable' */
 719
 720             /* CUBIC SPLINE TABLE DISPERSION */
 721             /*             #if 'Table' in KERNEL_ELEC */
 722             vfitab           = _mm_add_epi32(vfitab,ifour);
 723             /*             #endif                     */
 724             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 725             /*             #if ROUND == 'Loop' */
 726             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 727             /*             #else */
 728             F                = _mm_setzero_pd();
 729             /*             #endif */
 730             GMX_MM_TRANSPOSE2_PD(Y,F);
 731             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 732             /*             #if ROUND == 'Loop' */
 733             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 734             /*             #else */
 735             H                = _mm_setzero_pd();
 736             /*             #endif */
 737             GMX_MM_TRANSPOSE2_PD(G,H);
 738             Heps             = _mm_mul_pd(vfeps,H);
 739             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 740             /*             #define INNERFLOPS INNERFLOPS+4 */
 741             /*             #if 'Potential' in KERNEL_VF */
 742             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 743             vvdw6            = _mm_mul_pd(c6_{I}{J},VV);
 744             /*                 #define INNERFLOPS INNERFLOPS+3 */
 745             /*             #endif */
 746             /*             #if 'Force' in KERNEL_VF */
 747             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 748             fvdw6            = _mm_mul_pd(c6_{I}{J},FF);
 749             /*                 #define INNERFLOPS INNERFLOPS+4 */
 750             /*             #endif */
 751
 752             /* CUBIC SPLINE TABLE REPULSION */
 753             vfitab           = _mm_add_epi32(vfitab,ifour);
 754             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 755             /*             #if ROUND == 'Loop' */
 756             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 757             /*             #else */
 758             F                = _mm_setzero_pd();
 759             /*             #endif */
 760             GMX_MM_TRANSPOSE2_PD(Y,F);
 761             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 762             /*             #if ROUND == 'Loop' */
 763             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 764             /*             #else */
 765             H                = _mm_setzero_pd();
 766             /*             #endif */
 767             GMX_MM_TRANSPOSE2_PD(G,H);
 768             Heps             = _mm_mul_pd(vfeps,H);
 769             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 770             /*             #define INNERFLOPS INNERFLOPS+4 */
 771             /*             #if 'Potential' in KERNEL_VF */
 772             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 773             vvdw12           = _mm_mul_pd(c12_{I}{J},VV);
 774             /*                 #define INNERFLOPS INNERFLOPS+3 */
 775             /*             #endif */
 776             /*             #if 'Force' in KERNEL_VF */
 777             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 778             fvdw12           = _mm_mul_pd(c12_{I}{J},FF);
 779             /*                 #define INNERFLOPS INNERFLOPS+5 */
 780             /*             #endif */
 781             /*             #if 'Potential' in KERNEL_VF */
 782             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 783             /*                 #define INNERFLOPS INNERFLOPS+1 */
 784             /*             #endif */
 785             /*             #if 'Force' in KERNEL_VF */
 786             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv{I}{J})));
 787             /*                 #define INNERFLOPS INNERFLOPS+4 */
 788             /*             #endif */
 789             /*         #endif */
 790             /*         ## End of check for vdw interaction forms */
 791             /*     #endif */
 792             /*     ## END OF VDW INTERACTION CHECK FOR PAIR I-J */
 793
 794             /*     #if 'switch' in INTERACTION_FLAGS[I][J] */
 795             d                = _mm_sub_pd(r{I}{J},rswitch);
 796             d                = _mm_max_pd(d,_mm_setzero_pd());
 797             d2               = _mm_mul_pd(d,d);
 798             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)))))));
 799             /*         #define INNERFLOPS INNERFLOPS+10 */
 800
 801             /*         #if 'Force' in KERNEL_VF */
 802             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 803             /*             #define INNERFLOPS INNERFLOPS+5 */
 804             /*         #endif */
 805
 806             /* Evaluate switch function */
 807             /*         #if 'Force' in KERNEL_VF */
 808             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 809             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 810             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv{I}{J},_mm_mul_pd(velec,dsw)) );
 811             /*                 #define INNERFLOPS INNERFLOPS+4 */
 812             /*             #endif */
 813             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 814             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv{I}{J},_mm_mul_pd(vvdw,dsw)) );
 815             /*                 #define INNERFLOPS INNERFLOPS+4 */
 816             /*             #endif */
 817             /*         #endif */
 818             /*         #if 'Potential' in KERNEL_VF */
 819             /*             #if 'electrostatics' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_ELEC=='PotentialSwitch' */
 820             velec            = _mm_mul_pd(velec,sw);
 821             /*                 #define INNERFLOPS INNERFLOPS+1 */
 822             /*             #endif */
 823             /*             #if 'vdw' in INTERACTION_FLAGS[I][J] and KERNEL_MOD_VDW=='PotentialSwitch' */
 824             vvdw             = _mm_mul_pd(vvdw,sw);
 825             /*                 #define INNERFLOPS INNERFLOPS+1 */
 826             /*             #endif */
 827             /*         #endif */
 828             /*     #endif */
 829             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 830             cutoff_mask      = _mm_cmplt_pd(rsq{I}{J},rcutoff2);
 831             /*         #define INNERFLOPS INNERFLOPS+1 */
 832             /*     #endif */
 833
 834             /*     #if 'Potential' in KERNEL_VF */
 835             /* Update potential sum for this i atom from the interaction with this j atom. */
 836             /*         #if 'electrostatics' in INTERACTION_FLAGS[I][J] */
 837             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 838             velec            = _mm_and_pd(velec,cutoff_mask);
 839             /*                 #define INNERFLOPS INNERFLOPS+1 */
 840             /*             #endif                                       */
 841             /*             #if ROUND == 'Epilogue' */
 842             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 843             /*             #endif */
 844             velecsum         = _mm_add_pd(velecsum,velec);
 845             /*             #define INNERFLOPS INNERFLOPS+1 */
 846             /*             #if KERNEL_ELEC=='GeneralizedBorn' */
 847             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 848             vgb              = _mm_and_pd(vgb,cutoff_mask);
 849             /*                 #define INNERFLOPS INNERFLOPS+1 */
 850             /*             #endif                                       */
 851             /*             #if ROUND == 'Epilogue' */
 852             vgb              = _mm_unpacklo_pd(vgb,_mm_setzero_pd());
 853             /*             #endif */
 854             vgbsum           = _mm_add_pd(vgbsum,vgb);
 855             /*                 #define INNERFLOPS INNERFLOPS+1 */
 856             /*             #endif */
 857             /*         #endif */
 858             /*         #if 'vdw' in INTERACTION_FLAGS[I][J] */
 859             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 860             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 861             /*                 #define INNERFLOPS INNERFLOPS+1 */
 862             /*             #endif                                       */
 863             /*             #if ROUND == 'Epilogue' */
 864             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 865             /*             #endif */
 866             vvdwsum          = _mm_add_pd(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_pd(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             /*             #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 883             fscal            = _mm_and_pd(fscal,cutoff_mask);
 884             /*                 #define INNERFLOPS INNERFLOPS+1 */
 885             /*             #endif                                       */
 886
 887             /*             #if ROUND == 'Epilogue' */
 888             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 889             /*             #endif */
 890
 891             /* Calculate temporary vectorial force */
 892             tx               = _mm_mul_pd(fscal,dx{I}{J});
 893             ty               = _mm_mul_pd(fscal,dy{I}{J});
 894             tz               = _mm_mul_pd(fscal,dz{I}{J});
 895
 896             /* Update vectorial force */
 897             fix{I}             = _mm_add_pd(fix{I},tx);
 898             fiy{I}             = _mm_add_pd(fiy{I},ty);
 899             fiz{I}             = _mm_add_pd(fiz{I},tz);
 900             /*             #define INNERFLOPS INNERFLOPS+6 */
 901
 902             /* #if GEOMETRY_I == 'Particle'             */
 903             /*     #if ROUND == 'Loop' */
 904             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 905             /*     #else */
 906             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 907             /*     #endif */
 908             /*     #define INNERFLOPS INNERFLOPS+3      */
 909             /* #else                                    */
 910             fjx{J}             = _mm_add_pd(fjx{J},tx);
 911             fjy{J}             = _mm_add_pd(fjy{J},ty);
 912             fjz{J}             = _mm_add_pd(fjz{J},tz);
 913             /*     #define INNERFLOPS INNERFLOPS+3      */
 914             /* #endif                                   */
 915
 916             /*     #endif */
 917
 918             /*     #if 'exactcutoff' in INTERACTION_FLAGS[I][J] */
 919             /*         #if 0    ## This and next two lines is a hack to maintain indentation in template file */
 920             {
 921                 /*     #endif */
 922             }
 923             /*     #endif */
 924             /*    ## End of check for the interaction being outside the cutoff */
 925
 926             /* #endfor */
 927             /* ## End of loop over i-j interaction pairs */
 928
 929             /* #if 'Water' in GEOMETRY_I and GEOMETRY_J == 'Particle' */
 930             /*     #if ROUND == 'Loop' */
 931             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 932             /*     #else */
 933             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 934             /*     #endif */
 935             /*     #define INNERFLOPS INNERFLOPS+3      */
 936             /* #elif GEOMETRY_J == 'Water3'             */
 937             /*     #if ROUND == 'Loop' */
 938             gmx_mm_decrement_3rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 939             /*     #else */
 940             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 941             /*     #endif */
 942             /*     #define INNERFLOPS INNERFLOPS+9      */
 943             /* #elif GEOMETRY_J == 'Water4'             */
 944             /*     #if 0 in PARTICLES_J                 */
 945             /*         #if ROUND == 'Loop' */
 946             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);
 947             /*         #else */
 948             gmx_mm_decrement_4rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 949             /*         #endif */
 950             /*         #define INNERFLOPS INNERFLOPS+12 */
 951             /*     #else                                */
 952             /*         #if ROUND == 'Loop' */
 953             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);
 954             /*         #else */
 955             gmx_mm_decrement_3rvec_1ptr_swizzle_pd(f+j_coord_offsetA+DIM,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 956             /*         #endif */
 957             /*         #define INNERFLOPS INNERFLOPS+9  */
 958             /*     #endif                               */
 959             /* #endif                                   */
 960
 961             /* Inner loop uses {INNERFLOPS} flops */
 962         }
 963
 964         /* #endfor */
 965
 966         /* End of innermost loop */
 967
 968         /* #if 'Force' in KERNEL_VF */
 969         /*     #if GEOMETRY_I == 'Particle'            */
 970         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 971                                               f+i_coord_offset,fshift+i_shift_offset);
 972         /*         #define OUTERFLOPS OUTERFLOPS+6     */
 973         /*     #elif GEOMETRY_I == 'Water3'            */
 974         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 975                                               f+i_coord_offset,fshift+i_shift_offset);
 976         /*         #define OUTERFLOPS OUTERFLOPS+18    */
 977         /*     #elif GEOMETRY_I == 'Water4'            */
 978         /*         #if 0 in PARTICLES_I                */
 979         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 980                                               f+i_coord_offset,fshift+i_shift_offset);
 981         /*             #define OUTERFLOPS OUTERFLOPS+24    */
 982         /*         #else                               */
 983         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 984                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 985         /*             #define OUTERFLOPS OUTERFLOPS+18    */
 986         /*         #endif                              */
 987         /*     #endif                                  */
 988         /* #endif                                      */
 989
 990         /* #if 'Potential' in KERNEL_VF */
 991         ggid                        = gid[iidx];
 992         /* Update potential energies */
 993         /*     #if KERNEL_ELEC != 'None' */
 994         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 995         /*         #define OUTERFLOPS OUTERFLOPS+1 */
 996         /*     #endif */
 997         /*     #if 'GeneralizedBorn' in KERNEL_ELEC */
 998         gmx_mm_update_1pot_pd(vgbsum,kernel_data->energygrp_polarization+ggid);
 999         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1000         /*     #endif */
1001         /*     #if KERNEL_VDW != 'None' */
1002         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
1003         /*         #define OUTERFLOPS OUTERFLOPS+1 */
1004         /*     #endif */
1005         /* #endif */
1006         /*     #if 'GeneralizedBorn' in KERNEL_ELEC and 'Force' in KERNEL_VF */
1007         dvdasum = _mm_mul_pd(dvdasum, _mm_mul_pd(isai{I},isai{I}));
1008         gmx_mm_update_1pot_pd(dvdasum,dvda+inr);
1009         /*     #endif */
1010
1011         /* Increment number of inner iterations */
1012         inneriter                  += j_index_end - j_index_start;
1013
1014         /* Outer loop uses {OUTERFLOPS} flops */
1015     }
1016
1017     /* Increment number of outer iterations */
1018     outeriter        += nri;
1019
1020     /* Update outer/inner flops */
1021     /* ## NB: This is not important, it just affects the flopcount. However, since our preprocessor is */
1022     /* ## primitive and replaces aggressively even in strings inside these directives, we need to      */
1023     /* ## assemble the main part of the name (containing KERNEL/ELEC/VDW) directly in the source.      */
1024     /* #if GEOMETRY_I == 'Water3'            */
1025     /*     #define ISUFFIX '_W3'             */
1026     /* #elif GEOMETRY_I == 'Water4'          */
1027     /*     #define ISUFFIX '_W4'             */
1028     /* #else                                 */
1029     /*     #define ISUFFIX ''                */
1030     /* #endif                                */
1031     /* #if GEOMETRY_J == 'Water3'            */
1032     /*     #define JSUFFIX 'W3'              */
1033     /* #elif GEOMETRY_J == 'Water4'          */
1034     /*     #define JSUFFIX 'W4'              */
1035     /* #else                                 */
1036     /*     #define JSUFFIX ''                */
1037     /* #endif                                */
1038     /* #if 'PotentialAndForce' in KERNEL_VF  */
1039     /*     #define VFSUFFIX  '_VF'           */
1040     /* #elif 'Potential' in KERNEL_VF        */
1041     /*     #define VFSUFFIX '_V'             */
1042     /* #else                                 */
1043     /*     #define VFSUFFIX '_F'             */
1044     /* #endif                                */
1045
1046     /* #if KERNEL_ELEC != 'None' and KERNEL_VDW != 'None' */
1047     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1048     /* #elif KERNEL_ELEC != 'None' */
1049     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1050     /* #else */
1051     inc_nrnb(nrnb,eNR_NBKERNEL_VDW{ISUFFIX}{JSUFFIX}{VFSUFFIX},outeriter*{OUTERFLOPS} + inneriter*{INNERFLOPS});
1052     /* #endif  */
1053 }